diff options
| author | Dimitry Andric <dim@FreeBSD.org> | 2018-07-28 11:06:01 +0000 |
|---|---|---|
| committer | Dimitry Andric <dim@FreeBSD.org> | 2018-07-28 11:06:01 +0000 |
| commit | 486754660bb926339aefcf012a3f848592babb8b (patch) | |
| tree | ecdbc446c9876f4f120f701c243373cd3cb43db3 /test/OpenMP | |
| parent | 55e6d896ad333f07bb3b1ba487df214fc268a4ab (diff) | |
Notes
Diffstat (limited to 'test/OpenMP')
824 files changed, 18811 insertions, 3763 deletions
diff --git a/test/OpenMP/atomic_ast_print.cpp b/test/OpenMP/atomic_ast_print.cpp index f1fc7c8e559e6..c66249b708eda 100644 --- a/test/OpenMP/atomic_ast_print.cpp +++ b/test/OpenMP/atomic_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -44,7 +48,7 @@ T foo(T argc) { } // CHECK: T a = T(); -// CHECK-NEXT: #pragma omp atomic +// CHECK-NEXT: #pragma omp atomic{{$}} // CHECK-NEXT: a++; // CHECK-NEXT: #pragma omp atomic read // CHECK-NEXT: a = argc; diff --git a/test/OpenMP/atomic_capture_codegen.cpp b/test/OpenMP/atomic_capture_codegen.cpp index 306b83f62463f..51c65a09fb226 100644 --- a/test/OpenMP/atomic_capture_codegen.cpp +++ b/test/OpenMP/atomic_capture_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -target-cpu core2 -fopenmp -x c -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c -triple x86_64-apple-darwin10 -target-cpu core2 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c -triple x86_64-apple-darwin10 -target-cpu core2 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -target-cpu core2 -fopenmp-simd -x c -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c -triple x86_64-apple-darwin10 -target-cpu core2 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c -triple x86_64-apple-darwin10 -target-cpu core2 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/atomic_codegen.cpp b/test/OpenMP/atomic_codegen.cpp index 7f62a9bfa6fb9..c7411160604cd 100644 --- a/test/OpenMP/atomic_codegen.cpp +++ b/test/OpenMP/atomic_codegen.cpp @@ -2,6 +2,12 @@ // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -x c++ -emit-llvm -std=c++98 %s -o - | FileCheck %s // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -x c++ -emit-llvm -std=c++11 %s -o - | FileCheck %s // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -x c++ -emit-llvm -std=c++98 %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -x c++ -emit-llvm -std=c++11 %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics int a; diff --git a/test/OpenMP/atomic_messages.c b/test/OpenMP/atomic_messages.c index 7b3178bba49f7..1234e8580c02c 100644 --- a/test/OpenMP/atomic_messages.c +++ b/test/OpenMP/atomic_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + int foo() { L1: foo(); diff --git a/test/OpenMP/atomic_messages.cpp b/test/OpenMP/atomic_messages.cpp index efb368d1a7c0e..d7fa8c9e07587 100644 --- a/test/OpenMP/atomic_messages.cpp +++ b/test/OpenMP/atomic_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + int foo() { L1: foo(); diff --git a/test/OpenMP/atomic_read_codegen.c b/test/OpenMP/atomic_read_codegen.c index 0cfb2d26f2435..55fd03b440da8 100644 --- a/test/OpenMP/atomic_read_codegen.c +++ b/test/OpenMP/atomic_read_codegen.c @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -target-cpu core2 -fopenmp -x c -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c -triple x86_64-apple-darwin10 -target-cpu core2 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c -triple x86_64-apple-darwin10 -target-cpu core2 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -target-cpu core2 -fopenmp-simd -x c -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c -triple x86_64-apple-darwin10 -target-cpu core2 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c -triple x86_64-apple-darwin10 -target-cpu core2 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics // REQUIRES: x86-registered-target #ifndef HEADER diff --git a/test/OpenMP/atomic_update_codegen.cpp b/test/OpenMP/atomic_update_codegen.cpp index 1343cd8ad2cd0..d314c24a16242 100644 --- a/test/OpenMP/atomic_update_codegen.cpp +++ b/test/OpenMP/atomic_update_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -target-cpu core2 -fopenmp -x c -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c -triple x86_64-apple-darwin10 -target-cpu core2 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c -triple x86_64-apple-darwin10 -target-cpu core2 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -target-cpu core2 -fopenmp-simd -x c -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c -triple x86_64-apple-darwin10 -target-cpu core2 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c -triple x86_64-apple-darwin10 -target-cpu core2 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/atomic_write_codegen.c b/test/OpenMP/atomic_write_codegen.c index 0c85b6e88a3fc..242708cd7cf53 100644 --- a/test/OpenMP/atomic_write_codegen.c +++ b/test/OpenMP/atomic_write_codegen.c @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -target-cpu core2 -fopenmp -x c -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c -triple x86_64-apple-darwin10 -target-cpu core2 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c -triple x86_64-apple-darwin10 -target-cpu core2 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -target-cpu core2 -fopenmp-simd -x c -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c -triple x86_64-apple-darwin10 -target-cpu core2 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c -triple x86_64-apple-darwin10 -target-cpu core2 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics // REQUIRES: x86-registered-target #ifndef HEADER @@ -137,7 +142,7 @@ int main() { dx = dv; // CHECK: [[LD:%.+]] = load x86_fp80, x86_fp80* // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[LDTEMP:%.*]] to i8* -// CHECK: call void @llvm.memset.p0i8.i64(i8* [[BITCAST]], i8 0, i64 16, i32 16, i1 false) +// CHECK: call void @llvm.memset.p0i8.i64(i8* align 16 [[BITCAST]], i8 0, i64 16, i1 false) // CHECK: store x86_fp80 [[LD]], x86_fp80* [[LDTEMP]] // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[LDTEMP:%.*]] to i128* // CHECK: [[LD:%.+]] = load i128, i128* [[BITCAST]] @@ -226,7 +231,7 @@ int main() { // CHECK: load i64, i64* // CHECK: [[VAL:%.+]] = uitofp i64 %{{.+}} to x86_fp80 // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP:%.+]] to i8* -// CHECK: call void @llvm.memset.p0i8.i64(i8* [[BITCAST]], i8 0, i64 16, i32 16, i1 false) +// CHECK: call void @llvm.memset.p0i8.i64(i8* align 16 [[BITCAST]], i8 0, i64 16, i1 false) // CHECK: store x86_fp80 [[VAL]], x86_fp80* [[TEMP]] // CHECK: [[BITCAST:%.+]] = bitcast x86_fp80* [[TEMP]] to i128* // CHECK: [[VAL:%.+]] = load i128, i128* [[BITCAST]] diff --git a/test/OpenMP/barrier_ast_print.cpp b/test/OpenMP/barrier_ast_print.cpp index fd03a9ab0ffed..a8db2172c1928 100644 --- a/test/OpenMP/barrier_ast_print.cpp +++ b/test/OpenMP/barrier_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -24,7 +28,7 @@ T tmain(T argc) { return a + argc; } // CHECK: static T a; -// CHECK-NEXT: #pragma omp barrier +// CHECK-NEXT: #pragma omp barrier{{$}} // CHECK: static int a; // CHECK-NEXT: #pragma omp barrier // CHECK: static char a; diff --git a/test/OpenMP/barrier_codegen.cpp b/test/OpenMP/barrier_codegen.cpp index f335935050c6a..aa3358cede7ce 100644 --- a/test/OpenMP/barrier_codegen.cpp +++ b/test/OpenMP/barrier_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/barrier_messages.cpp b/test/OpenMP/barrier_messages.cpp index 7d79445d28577..137453a89326e 100644 --- a/test/OpenMP/barrier_messages.cpp +++ b/test/OpenMP/barrier_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + template <class T> T tmain(T argc) { #pragma omp barrier diff --git a/test/OpenMP/cancel_ast_print.cpp b/test/OpenMP/cancel_ast_print.cpp index be550843706c2..b376d4ec28072 100644 --- a/test/OpenMP/cancel_ast_print.cpp +++ b/test/OpenMP/cancel_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -22,7 +26,7 @@ int main (int argc, char **argv) { } // CHECK-NEXT: #pragma omp sections // CHECK: { -// CHECK: #pragma omp cancel sections +// CHECK: #pragma omp cancel sections{{$}} // CHECK: } #pragma omp for for (int i = 0; i < argc; ++i) { diff --git a/test/OpenMP/cancel_codegen.cpp b/test/OpenMP/cancel_codegen.cpp index 2ff9dcc186c99..75fd5e46618c2 100644 --- a/test/OpenMP/cancel_codegen.cpp +++ b/test/OpenMP/cancel_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -triple x86_64-apple-darwin13.4.0 -emit-llvm -o - %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple x86_64-apple-darwin13.4.0 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -triple x86_64-apple-darwin13.4.0 -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -triple x86_64-apple-darwin13.4.0 -emit-llvm -o - %s | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple x86_64-apple-darwin13.4.0 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -triple x86_64-apple-darwin13.4.0 -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -15,7 +20,7 @@ int main (int argc, char **argv) { #pragma omp barrier argv[0][0] += argc; } -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( #pragma omp sections { #pragma omp cancel sections @@ -23,7 +28,7 @@ int main (int argc, char **argv) { // CHECK: call void @__kmpc_for_static_init_4( // CHECK: call i32 @__kmpc_cancel( // CHECK: call void @__kmpc_for_static_fini( -// CHECK: call void @__kmpc_barrier(%ident_t* +// CHECK: call void @__kmpc_barrier(%struct.ident_t* #pragma omp sections { #pragma omp cancel sections @@ -33,14 +38,14 @@ int main (int argc, char **argv) { } } // CHECK: call void @__kmpc_for_static_init_4( -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%ident_t* {{[^,]+}}, i32 [[GTID]], i32 3) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%struct.ident_t* {{[^,]+}}, i32 [[GTID]], i32 3) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] // CHECK: [[EXIT]] // CHECK: br label // CHECK: [[CONTINUE]] // CHECK: br label -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%ident_t* {{[^,]+}}, i32 [[GTID]], i32 3) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%struct.ident_t* {{[^,]+}}, i32 [[GTID]], i32 3) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] // CHECK: [[EXIT]] @@ -57,7 +62,7 @@ for (int i = 0; i < argc; ++i) { // CHECK: [[BOOL:%.+]] = fcmp une float [[FLAG]], 0.000000e+00 // CHECK: br i1 [[BOOL]], label %[[THEN:[^,]+]], label %[[ELSE:[^,]+]] // CHECK: [[THEN]] -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%ident_t* {{[^,]+}}, i32 [[GTID]], i32 2) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%struct.ident_t* {{[^,]+}}, i32 [[GTID]], i32 2) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] // CHECK: [[EXIT]] @@ -67,7 +72,7 @@ for (int i = 0; i < argc; ++i) { // CHECK: [[ELSE]] // CHECK: br label // CHECK: call void @__kmpc_for_static_fini( -// CHECK: call void @__kmpc_barrier(%ident_t* +// CHECK: call void @__kmpc_barrier(%struct.ident_t* #pragma omp task { #pragma omp cancel taskgroup @@ -78,7 +83,7 @@ for (int i = 0; i < argc; ++i) { { #pragma omp cancel sections } -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( #pragma omp parallel sections { #pragma omp cancel sections @@ -87,14 +92,14 @@ for (int i = 0; i < argc; ++i) { #pragma omp cancel sections } } -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( int r = 0; #pragma omp parallel for reduction(+: r) for (int i = 0; i < argc; ++i) { #pragma omp cancel for r += i; } -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( return argc; } @@ -103,20 +108,20 @@ for (int i = 0; i < argc; ++i) { // CHECK: [[BOOL:%.+]] = fcmp une float [[FLAG]], 0.000000e+00 // CHECK: br i1 [[BOOL]], label %[[THEN:[^,]+]], label %[[ELSE:[^,]+]] // CHECK: [[THEN]] -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 1) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 1) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,]+]], // CHECK: [[EXIT]] // CHECK: br label %[[RETURN:.+]] // CHECK: [[ELSE]] // The barrier directive should now call __kmpc_cancel_barrier -// CHECK: call i32 @__kmpc_cancel_barrier(%ident_t* +// CHECK: call i32 @__kmpc_cancel_barrier(%struct.ident_t* // CHECK: br label // CHECK: [[RETURN]] // CHECK: ret void // CHECK: define internal i32 @{{[^(]+}}(i32 -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 4) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 4) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,]+]], // CHECK: [[EXIT]] @@ -132,14 +137,14 @@ for (int i = 0; i < argc; ++i) { // CHECK: define internal void @{{[^(]+}}(i32* {{[^,]+}}, i32* {{[^,]+}}) // CHECK: call void @__kmpc_for_static_init_4( -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%ident_t* {{[^,]+}}, i32 [[GTID:%.+]], i32 3) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%struct.ident_t* {{[^,]+}}, i32 [[GTID:%.+]], i32 3) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] // CHECK: [[EXIT]] // CHECK: br label // CHECK: [[CONTINUE]] // CHECK: br label -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%ident_t* {{[^,]+}}, i32 [[GTID]], i32 3) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%struct.ident_t* {{[^,]+}}, i32 [[GTID]], i32 3) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] // CHECK: [[EXIT]] @@ -151,7 +156,7 @@ for (int i = 0; i < argc; ++i) { // CHECK: define internal void @{{[^(]+}}(i32* {{[^,]+}}, i32* {{[^,]+}}, // CHECK: call void @__kmpc_for_static_init_4( -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%ident_t* {{[^,]+}}, i32 [[GTID:%.+]], i32 2) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancel(%struct.ident_t* {{[^,]+}}, i32 [[GTID:%.+]], i32 2) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] // CHECK: [[EXIT]] diff --git a/test/OpenMP/cancel_codegen_cleanup.cpp b/test/OpenMP/cancel_codegen_cleanup.cpp index 5a297bd377b1f..94d49068433a1 100644 --- a/test/OpenMP/cancel_codegen_cleanup.cpp +++ b/test/OpenMP/cancel_codegen_cleanup.cpp @@ -1,5 +1,8 @@ // RUN: %clang_cc1 -std=c++11 -fopenmp -fopenmp-version=45 -triple x86_64-apple-darwin13.4.0 -emit-llvm -o - %s | FileCheck %s +// RUN: %clang_cc1 -std=c++11 -fopenmp-simd -fopenmp-version=45 -triple x86_64-apple-darwin13.4.0 -emit-llvm -o - %s | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + //CHECK: call i32 @__kmpc_cancel //CHECK: br {{.*}}label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] //CHECK: [[EXIT]]: diff --git a/test/OpenMP/cancel_if_messages.cpp b/test/OpenMP/cancel_if_messages.cpp index 426ac49cfdfee..dc24fa4c99793 100644 --- a/test/OpenMP/cancel_if_messages.cpp +++ b/test/OpenMP/cancel_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/cancel_messages.cpp b/test/OpenMP/cancel_messages.cpp index f8bbe2c36bea3..28f22d17eaf5d 100644 --- a/test/OpenMP/cancel_messages.cpp +++ b/test/OpenMP/cancel_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + int main(int argc, char **argv) { #pragma omp cancellation // expected-error {{expected an OpenMP directive}} #pragma omp cancel // expected-error {{one of 'for', 'parallel', 'sections' or 'taskgroup' is expected}} diff --git a/test/OpenMP/cancellation_point_ast_print.cpp b/test/OpenMP/cancellation_point_ast_print.cpp index c8b133cbac68d..d27d2b75fd4c1 100644 --- a/test/OpenMP/cancellation_point_ast_print.cpp +++ b/test/OpenMP/cancellation_point_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -14,7 +18,7 @@ int main (int argc, char **argv) { } // CHECK: #pragma omp parallel // CHECK-NEXT: { -// CHECK-NEXT: #pragma omp cancellation point parallel +// CHECK-NEXT: #pragma omp cancellation point parallel{{$}} // CHECK-NEXT: } #pragma omp sections { diff --git a/test/OpenMP/cancellation_point_codegen.cpp b/test/OpenMP/cancellation_point_codegen.cpp index de65c114027ac..a67015fe43f53 100644 --- a/test/OpenMP/cancellation_point_codegen.cpp +++ b/test/OpenMP/cancellation_point_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -triple x86_64-apple-darwin13.4.0 -emit-llvm -o - %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-apple-darwin13.4.0 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -triple x86_64-apple-darwin13.4.0 -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -triple x86_64-apple-darwin13.4.0 -emit-llvm -o - %s | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin13.4.0 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -triple x86_64-apple-darwin13.4.0 -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -13,7 +18,7 @@ int main (int argc, char **argv) { #pragma omp cancel parallel argv[0][0] = argc; } -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( #pragma omp sections { { @@ -22,7 +27,7 @@ int main (int argc, char **argv) { } } // CHECK: call void @__kmpc_for_static_init_4( -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%ident_t* {{[^,]+}}, i32 [[GTID]], i32 3) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%struct.ident_t* {{[^,]+}}, i32 [[GTID]], i32 3) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] // CHECK: [[EXIT]] @@ -30,7 +35,7 @@ int main (int argc, char **argv) { // CHECK: [[CONTINUE]] // CHECK: br label // CHECK: call void @__kmpc_for_static_fini( -// CHECK: call void @__kmpc_barrier(%ident_t* +// CHECK: call void @__kmpc_barrier(%struct.ident_t* #pragma omp sections { #pragma omp cancellation point sections @@ -41,14 +46,14 @@ int main (int argc, char **argv) { } } // CHECK: call void @__kmpc_for_static_init_4( -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%ident_t* {{[^,]+}}, i32 [[GTID]], i32 3) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%struct.ident_t* {{[^,]+}}, i32 [[GTID]], i32 3) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] // CHECK: [[EXIT]] // CHECK: br label // CHECK: [[CONTINUE]] // CHECK: br label -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%ident_t* {{[^,]+}}, i32 [[GTID]], i32 3) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%struct.ident_t* {{[^,]+}}, i32 [[GTID]], i32 3) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] // CHECK: [[EXIT]] @@ -62,7 +67,7 @@ for (int i = 0; i < argc; ++i) { #pragma omp cancel for } // CHECK: call void @__kmpc_for_static_init_4( -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%ident_t* {{[^,]+}}, i32 [[GTID]], i32 2) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%struct.ident_t* {{[^,]+}}, i32 [[GTID]], i32 2) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] // CHECK: [[EXIT]] @@ -70,7 +75,7 @@ for (int i = 0; i < argc; ++i) { // CHECK: [[CONTINUE]] // CHECK: br label // CHECK: call void @__kmpc_for_static_fini( -// CHECK: call void @__kmpc_barrier(%ident_t* +// CHECK: call void @__kmpc_barrier(%struct.ident_t* #pragma omp task { #pragma omp cancellation point taskgroup @@ -91,7 +96,7 @@ for (int i = 0; i < argc; ++i) { #pragma omp cancel sections } } -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( #pragma omp parallel sections { { @@ -103,18 +108,18 @@ for (int i = 0; i < argc; ++i) { #pragma omp cancellation point sections } } -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( #pragma omp parallel for for (int i = 0; i < argc; ++i) { #pragma omp cancellation point for #pragma omp cancel for } -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( return argc; } // CHECK: define internal void @{{[^(]+}}(i32* {{[^,]+}}, i32* {{[^,]+}}, -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 1) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 1) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,]+]], // CHECK: [[EXIT]] @@ -123,7 +128,7 @@ for (int i = 0; i < argc; ++i) { // CHECK: ret void // CHECK: define internal i32 @{{[^(]+}}(i32 -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 4) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 4) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,]+]], // CHECK: [[EXIT]] @@ -132,7 +137,7 @@ for (int i = 0; i < argc; ++i) { // CHECK: ret i32 0 // CHECK: define internal i32 @{{[^(]+}}(i32 -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 4) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 4) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,]+]], // CHECK: [[EXIT]] @@ -142,7 +147,7 @@ for (int i = 0; i < argc; ++i) { // CHECK: define internal void @{{[^(]+}}(i32* {{[^,]+}}, i32* {{[^,]+}}) // CHECK: call void @__kmpc_for_static_init_4( -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%ident_t* {{[^,]+}}, i32 [[GTID:%.+]], i32 3) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%struct.ident_t* {{[^,]+}}, i32 [[GTID:%.+]], i32 3) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] // CHECK: [[EXIT]] @@ -154,14 +159,14 @@ for (int i = 0; i < argc; ++i) { // CHECK: define internal void @{{[^(]+}}(i32* {{[^,]+}}, i32* {{[^,]+}}) // CHECK: call void @__kmpc_for_static_init_4( -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%ident_t* {{[^,]+}}, i32 [[GTID:%.+]], i32 3) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%struct.ident_t* {{[^,]+}}, i32 [[GTID:%.+]], i32 3) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] // CHECK: [[EXIT]] // CHECK: br label // CHECK: [[CONTINUE]] // CHECK: br label -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%ident_t* {{[^,]+}}, i32 [[GTID]], i32 3) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%struct.ident_t* {{[^,]+}}, i32 [[GTID]], i32 3) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] // CHECK: [[EXIT]] @@ -173,7 +178,7 @@ for (int i = 0; i < argc; ++i) { // CHECK: define internal void @{{[^(]+}}(i32* {{[^,]+}}, i32* {{[^,]+}}, // CHECK: call void @__kmpc_for_static_init_4( -// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%ident_t* {{[^,]+}}, i32 [[GTID:%.+]], i32 2) +// CHECK: [[RES:%.+]] = call i32 @__kmpc_cancellationpoint(%struct.ident_t* {{[^,]+}}, i32 [[GTID:%.+]], i32 2) // CHECK: [[CMP:%.+]] = icmp ne i32 [[RES]], 0 // CHECK: br i1 [[CMP]], label %[[EXIT:[^,].+]], label %[[CONTINUE:.+]] // CHECK: [[EXIT]] diff --git a/test/OpenMP/cancellation_point_messages.cpp b/test/OpenMP/cancellation_point_messages.cpp index 21b0e9c2fa2db..8bf5b3d55fdd4 100644 --- a/test/OpenMP/cancellation_point_messages.cpp +++ b/test/OpenMP/cancellation_point_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + int main(int argc, char **argv) { #pragma omp cancellation // expected-error {{expected an OpenMP directive}} #pragma omp cancellation point // expected-error {{one of 'for', 'parallel', 'sections' or 'taskgroup' is expected}} diff --git a/test/OpenMP/capturing_in_templates.cpp b/test/OpenMP/capturing_in_templates.cpp index 9b2d9d304e0a3..97a935214bdb8 100644 --- a/test/OpenMP/capturing_in_templates.cpp +++ b/test/OpenMP/capturing_in_templates.cpp @@ -1,4 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-ibm-linux-gnu -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-ibm-linux-gnu -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics template <typename T1, typename T2> diff --git a/test/OpenMP/constexpr_codegen.cpp b/test/OpenMP/constexpr_codegen.cpp new file mode 100644 index 0000000000000..f1fa4ac2a9efb --- /dev/null +++ b/test/OpenMP/constexpr_codegen.cpp @@ -0,0 +1,29 @@ +// RUN: %clang_cc1 -verify -fopenmp -std=c++11 -triple x86_64-unknown-unknown -x c++ -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -triple x86_64-unknown-unknown -x c++ -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +// +#ifndef HEADER +#define HEADER + +// CHECK: @{{.*}}Foo{{.*}}bar{{.*}} = constant i32 1, + +// Section A - Define a class with a static constexpr data member. +struct Foo { + static constexpr int bar = 1; +}; + +// Section B - ODR-use the data member. +void F(const int &); +void Test() { F(Foo::bar); } + +// Section C - Define the data member. +constexpr int Foo::bar; +#endif + diff --git a/test/OpenMP/critical_ast_print.cpp b/test/OpenMP/critical_ast_print.cpp index 0579d030a8bca..f51145e5845b2 100644 --- a/test/OpenMP/critical_ast_print.cpp +++ b/test/OpenMP/critical_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -10,10 +14,10 @@ void foo() {} // CHECK: template <typename T, int N> int tmain(T argc, char **argv) // CHECK: static int a; -// CHECK-NEXT: #pragma omp critical +// CHECK-NEXT: #pragma omp critical{{$}} // CHECK-NEXT: a = 2; // CHECK-NEXT: ++a; -// CHECK-NEXT: #pragma omp critical (the_name) hint(N) +// CHECK-NEXT: #pragma omp critical (the_name) hint(N){{$}} // CHECK-NEXT: foo(); // CHECK-NEXT: return N; // CHECK: template<> int tmain<int, 4>(int argc, char **argv) diff --git a/test/OpenMP/critical_codegen.cpp b/test/OpenMP/critical_codegen.cpp index f4e449a22370f..ba540059dc897 100644 --- a/test/OpenMP/critical_codegen.cpp +++ b/test/OpenMP/critical_codegen.cpp @@ -2,6 +2,12 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/critical_messages.cpp b/test/OpenMP/critical_messages.cpp index bb5ec00fdbf1a..a04d2e5223312 100644 --- a/test/OpenMP/critical_messages.cpp +++ b/test/OpenMP/critical_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + int foo(); template<typename T, int N> diff --git a/test/OpenMP/debug-info-openmp-array.cpp b/test/OpenMP/debug-info-openmp-array.cpp index 19bf2a284ad60..e6de4763fbbc3 100644 --- a/test/OpenMP/debug-info-openmp-array.cpp +++ b/test/OpenMP/debug-info-openmp-array.cpp @@ -1,4 +1,7 @@ // RUN: %clang_cc1 -fopenmp -x c++ %s -verify -debug-info-kind=limited -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -fopenmp-simd -x c++ %s -verify -debug-info-kind=limited -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics void f(int m) { diff --git a/test/OpenMP/declare_reduction_ast_print.c b/test/OpenMP/declare_reduction_ast_print.c index 7b97a7c1da7c3..239b1cfd44326 100644 --- a/test/OpenMP/declare_reduction_ast_print.c +++ b/test/OpenMP/declare_reduction_ast_print.c @@ -1,13 +1,17 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER #define HEADER #pragma omp declare reduction(+ : int, char : omp_out *= omp_in) -// CHECK: #pragma omp declare reduction (+ : int : omp_out *= omp_in) +// CHECK: #pragma omp declare reduction (+ : int : omp_out *= omp_in){{$}} // CHECK-NEXT: #pragma omp declare reduction (+ : char : omp_out *= omp_in) #pragma omp declare reduction(fun : float : omp_out += omp_in) initializer(omp_priv = omp_orig + 15) diff --git a/test/OpenMP/declare_reduction_ast_print.cpp b/test/OpenMP/declare_reduction_ast_print.cpp index 816ff79376639..a8ca529c456b0 100644 --- a/test/OpenMP/declare_reduction_ast_print.cpp +++ b/test/OpenMP/declare_reduction_ast_print.cpp @@ -1,13 +1,17 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER #define HEADER #pragma omp declare reduction(+ : int, char : omp_out *= omp_in) -// CHECK: #pragma omp declare reduction (+ : int : omp_out *= omp_in) +// CHECK: #pragma omp declare reduction (+ : int : omp_out *= omp_in){{$}} // CHECK-NEXT: #pragma omp declare reduction (+ : char : omp_out *= omp_in) template <class T> diff --git a/test/OpenMP/declare_reduction_codegen.c b/test/OpenMP/declare_reduction_codegen.c index 579b664a512a6..8ccb8ccbaf406 100644 --- a/test/OpenMP/declare_reduction_codegen.c +++ b/test/OpenMP/declare_reduction_codegen.c @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -x c -emit-llvm %s -triple %itanium_abi_triple -o - -femit-all-decls -disable-llvm-passes | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c -triple %itanium_abi_triple -emit-pch -o %t %s -femit-all-decls -disable-llvm-passes // RUN: %clang_cc1 -fopenmp -x c -triple %itanium_abi_triple -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls -disable-llvm-passes | FileCheck --check-prefix=CHECK-LOAD %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c -emit-llvm %s -triple %itanium_abi_triple -o - -femit-all-decls -disable-llvm-passes | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c -triple %itanium_abi_triple -emit-pch -o %t %s -femit-all-decls -disable-llvm-passes +// RUN: %clang_cc1 -fopenmp-simd -x c -triple %itanium_abi_triple -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls -disable-llvm-passes | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER diff --git a/test/OpenMP/declare_reduction_codegen.cpp b/test/OpenMP/declare_reduction_codegen.cpp index ae6e047d94836..90dbd47214df0 100644 --- a/test/OpenMP/declare_reduction_codegen.cpp +++ b/test/OpenMP/declare_reduction_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -emit-llvm %s -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -o - -femit-all-decls -disable-llvm-passes | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -emit-pch -o %t %s -femit-all-decls -disable-llvm-passes // RUN: %clang_cc1 -fopenmp -x c++ -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls -disable-llvm-passes | FileCheck --check-prefix=CHECK-LOAD %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -emit-llvm %s -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -o - -femit-all-decls -disable-llvm-passes | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -emit-pch -o %t %s -femit-all-decls -disable-llvm-passes +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls -disable-llvm-passes | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER @@ -21,7 +26,7 @@ void add(short &out, short &in) {} // CHECK: foo_reduction_array void foo_reduction_array() { short y[1]; - // CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( + // CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( #pragma omp parallel for reduction(my_add : y) for (int i = 0; i < 1; i++) { } @@ -143,9 +148,9 @@ int main() { #pragma omp parallel reduction(fun : sss) { } - // CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( - // CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( - // CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call({{[^@]*}} @{{[^@]*}}[[REGION:@[^ ]+]] + // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( + // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( + // CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call({{[^@]*}} @{{[^@]*}}[[REGION:@[^ ]+]] // CHECK-LABEL: work work<A>(); // CHECK-LABEL: foo diff --git a/test/OpenMP/declare_reduction_messages.c b/test/OpenMP/declare_reduction_messages.c index fb9eacc90aa26..39387c795a3c9 100644 --- a/test/OpenMP/declare_reduction_messages.c +++ b/test/OpenMP/declare_reduction_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + int temp; // expected-note 6 {{'temp' declared here}} #pragma omp declare reduction // expected-error {{expected '(' after 'declare reduction'}} diff --git a/test/OpenMP/declare_reduction_messages.cpp b/test/OpenMP/declare_reduction_messages.cpp index 3e5a15ee3dd91..f22f92485a57e 100644 --- a/test/OpenMP/declare_reduction_messages.cpp +++ b/test/OpenMP/declare_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -std=c++11 %s + int temp; // expected-note 7 {{'temp' declared here}} #pragma omp declare reduction // expected-error {{expected '(' after 'declare reduction'}} diff --git a/test/OpenMP/declare_simd_ast_print.c b/test/OpenMP/declare_simd_ast_print.c index 04fd73f272d8b..414b01038793d 100644 --- a/test/OpenMP/declare_simd_ast_print.c +++ b/test/OpenMP/declare_simd_ast_print.c @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -12,10 +16,10 @@ #pragma omp declare simd notinbranch simdlen(2), uniform(s1, s2) linear(d: s1) void add_1(float *d, int s1, float *s2, double b[]) __attribute__((cold)); -// CHECK: #pragma omp declare simd notinbranch simdlen(2) uniform(s1, s2) linear(val(d): s1) -// CHECK-NEXT: #pragma omp declare simd inbranch uniform(d) linear(val(s1): 32) linear(val(s2): 32) -// CHECK-NEXT: #pragma omp declare simd simdlen(32) aligned(d) aligned(b) -// CHECK-NEXT: #pragma omp declare simd aligned(b: 64) +// CHECK: #pragma omp declare simd notinbranch simdlen(2) uniform(s1, s2) linear(val(d): s1){{$}} +// CHECK-NEXT: #pragma omp declare simd inbranch uniform(d) linear(val(s1): 32) linear(val(s2): 32){{$}} +// CHECK-NEXT: #pragma omp declare simd simdlen(32) aligned(d) aligned(b){{$}} +// CHECK-NEXT: #pragma omp declare simd aligned(b: 64){{$}} // CHECK-NEXT: void add_1(float *d, int s1, float *s2, double b[]) __attribute__((cold)) #endif diff --git a/test/OpenMP/declare_simd_ast_print.cpp b/test/OpenMP/declare_simd_ast_print.cpp index 632e038e221dd..c09f8b42e6d53 100644 --- a/test/OpenMP/declare_simd_ast_print.cpp +++ b/test/OpenMP/declare_simd_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -80,8 +84,8 @@ private: // CHECK-NEXT: float taddpf(float *a, T *&b) { // CHECK-NEXT: return *a + *b; // CHECK-NEXT: } -// CHECK: #pragma omp declare simd -// CHECK-NEXT: #pragma omp declare simd +// CHECK: #pragma omp declare simd uniform(this, b) +// CHECK-NEXT: #pragma omp declare simd{{$}} // CHECK-NEXT: int tadd(int b) { // CHECK-NEXT: return this->x[b] + b; // CHECK-NEXT: } diff --git a/test/OpenMP/declare_simd_codegen.cpp b/test/OpenMP/declare_simd_codegen.cpp index 9ae476677c42e..4ac0b7f03a0ec 100644 --- a/test/OpenMP/declare_simd_codegen.cpp +++ b/test/OpenMP/declare_simd_codegen.cpp @@ -1,10 +1,17 @@ // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -x c++ -emit-llvm %s -o - -femit-all-decls | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls | FileCheck %s + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -x c++ -emit-llvm %s -o - -femit-all-decls | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER +void add_1(float *d); + #pragma omp declare simd linear(d : 8) #pragma omp declare simd inbranch simdlen(32) #pragma omp declare simd notinbranch @@ -15,6 +22,8 @@ void add_1(float *d); #pragma omp declare simd notinbranch void add_1(float *d) {} +void add_1(float *d); + #pragma omp declare simd linear(d : 8) #pragma omp declare simd inbranch simdlen(32) #pragma omp declare simd notinbranch diff --git a/test/OpenMP/declare_simd_messages.cpp b/test/OpenMP/declare_simd_messages.cpp index af46283f9a57c..dab7d054e786c 100644 --- a/test/OpenMP/declare_simd_messages.cpp +++ b/test/OpenMP/declare_simd_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple=x86_64-pc-win32 -verify -fopenmp -x c++ -std=c++11 -fms-extensions -Wno-pragma-pack %s +// RUN: %clang_cc1 -triple=x86_64-pc-win32 -verify -fopenmp-simd -x c++ -std=c++11 -fms-extensions -Wno-pragma-pack %s + // expected-error@+1 {{expected an OpenMP directive}} #pragma omp declare diff --git a/test/OpenMP/declare_target_ast_print.cpp b/test/OpenMP/declare_target_ast_print.cpp index 6c59563d1d1e1..bd1acc28dc6c2 100644 --- a/test/OpenMP/declare_target_ast_print.cpp +++ b/test/OpenMP/declare_target_ast_print.cpp @@ -1,17 +1,21 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER #define HEADER #pragma omp declare target -// CHECK: #pragma omp declare target +// CHECK: #pragma omp declare target{{$}} void foo() {} // CHECK-NEXT: void foo() #pragma omp end declare target -// CHECK: #pragma omp end declare target +// CHECK: #pragma omp end declare target{{$}} extern "C" { #pragma omp declare target @@ -83,41 +87,41 @@ int a1; void f1() { } #pragma omp declare target (a1, f1) -// CHECK: #pragma omp declare target +// CHECK: #pragma omp declare target{{$}} // CHECK: int a1; -// CHECK: #pragma omp end declare target -// CHECK: #pragma omp declare target +// CHECK: #pragma omp end declare target{{$}} +// CHECK: #pragma omp declare target{{$}} // CHECK: void f1() -// CHECK: #pragma omp end declare target +// CHECK: #pragma omp end declare target{{$}} int b1, b2, b3; void f2() { } #pragma omp declare target to(b1) to(b2), to(b3, f2) -// CHECK: #pragma omp declare target +// CHECK: #pragma omp declare target{{$}} // CHECK: int b1; -// CHECK: #pragma omp end declare target -// CHECK: #pragma omp declare target +// CHECK: #pragma omp end declare target{{$}} +// CHECK: #pragma omp declare target{{$}} // CHECK: int b2; -// CHECK: #pragma omp end declare target -// CHECK: #pragma omp declare target +// CHECK: #pragma omp end declare target{{$}} +// CHECK: #pragma omp declare target{{$}} // CHECK: int b3; -// CHECK: #pragma omp end declare target -// CHECK: #pragma omp declare target +// CHECK: #pragma omp end declare target{{$}} +// CHECK: #pragma omp declare target{{$}} // CHECK: void f2() -// CHECK: #pragma omp end declare target +// CHECK: #pragma omp end declare target{{$}} int c1, c2, c3; #pragma omp declare target link(c1) link(c2), link(c3) -// CHECK: #pragma omp declare target link +// CHECK: #pragma omp declare target link{{$}} // CHECK: int c1; -// CHECK: #pragma omp end declare target -// CHECK: #pragma omp declare target link +// CHECK: #pragma omp end declare target{{$}} +// CHECK: #pragma omp declare target link{{$}} // CHECK: int c2; -// CHECK: #pragma omp end declare target -// CHECK: #pragma omp declare target link +// CHECK: #pragma omp end declare target{{$}} +// CHECK: #pragma omp declare target link{{$}} // CHECK: int c3; -// CHECK: #pragma omp end declare target +// CHECK: #pragma omp end declare target{{$}} struct SSSt { #pragma omp declare target diff --git a/test/OpenMP/declare_target_codegen.cpp b/test/OpenMP/declare_target_codegen.cpp new file mode 100644 index 0000000000000..a221387c73125 --- /dev/null +++ b/test/OpenMP/declare_target_codegen.cpp @@ -0,0 +1,94 @@ +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -emit-pch -o %t +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o -| FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -emit-pch -o %t +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify -o - | FileCheck %s --check-prefix SIMD-ONLY + +// expected-no-diagnostics + +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +// CHECK-NOT: define {{.*}}{{baz1|baz4|maini1}} +// CHECK-DAG: @{{.*}}maini1{{.*}}aaa = internal global i64 23, +// CHECK-DAG: @b = global i32 15, +// CHECK-DAG: @d = global i32 0, +// CHECK-DAG: @c = external global i32, +// CHECK-DAG: @globals = global %struct.S zeroinitializer, +// CHECK-DAG: @{{.+}}stat = weak_odr global %struct.S zeroinitializer, +// CHECK-DAG: @llvm.used = appending global [2 x i8*] [i8* bitcast (void ()* @__omp_offloading__{{.+}}_globals_l[[@LINE+42]]_ctor to i8*), i8* bitcast (void ()* @__omp_offloading__{{.+}}_stat_l[[@LINE+43]]_ctor to i8*)], section "llvm.metadata" + +// CHECK-DAG: define {{.*}}i32 @{{.*}}{{foo|bar|baz2|baz3|FA|f_method}}{{.*}}() +// CHECK-DAG: define {{.*}}void @{{.*}}TemplateClass{{.*}}(%class.TemplateClass* %{{.*}}) +// CHECK-DAG: define {{.*}}i32 @{{.*}}TemplateClass{{.*}}f_method{{.*}}(%class.TemplateClass* %{{.*}}) +// CHECK-DAG: define {{.*}}void @__omp_offloading__{{.*}}_globals_l[[@LINE+37]]_ctor() +// CHECK-DAG: define {{.*}}void @__omp_offloading__{{.*}}_stat_l[[@LINE+37]]_ctor() + +#ifndef HEADER +#define HEADER + +template <typename T> +class TemplateClass { + T a; +public: + TemplateClass() {} + T f_method() const { return a; } +}; + +int foo(); + +int baz1(); + +int baz2(); + +int baz4() { return 5; } + +template <typename T> +T FA() { + TemplateClass<T> s; + return s.f_method(); +} + +#pragma omp declare target +struct S { + int a; + S(int a) : a(a) {} +}; + +int foo() { return 0; } +int b = 15; +int d; +S globals(d); +static S stat(d); +#pragma omp end declare target +int c; + +int bar() { return 1 + foo() + bar() + baz1() + baz2(); } + +int maini1() { + int a; + static long aa = 32; +// CHECK-DAG: define weak void @__omp_offloading_{{.*}}maini1{{.*}}_l[[@LINE+1]](i32* dereferenceable{{.*}}, i64 {{.*}}, i64 {{.*}}) +#pragma omp target map(tofrom \ + : a, b) + { + S s(a); + static long aaa = 23; + a = foo() + bar() + b + c + d + aa + aaa + FA<int>(); + } + return baz4(); +} + +int baz3() { return 2 + baz2(); } +int baz2() { +// CHECK-DAG: define weak void @__omp_offloading_{{.*}}baz2{{.*}}_l[[@LINE+1]](i64 {{.*}}) +#pragma omp target parallel + ++c; + return 2 + baz3(); +} + +// CHECK-NOT: define {{.*}}{{baz1|baz4|maini1}} +#endif // HEADER diff --git a/test/OpenMP/declare_target_codegen_globalization.cpp b/test/OpenMP/declare_target_codegen_globalization.cpp new file mode 100644 index 0000000000000..7ef4f8af3deca --- /dev/null +++ b/test/OpenMP/declare_target_codegen_globalization.cpp @@ -0,0 +1,43 @@ +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-nvidia-cuda -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s +// expected-no-diagnostics + +int foo(int &a) { return a; } + +int bar() { + int a; + return foo(a); +} + +// CHECK: define weak void @__omp_offloading_{{.*}}maini1{{.*}}_l[[@LINE+5]](i32* dereferenceable{{.*}}) +// CHECK-NOT: @__kmpc_data_sharing_push_stack + +int maini1() { + int a; +#pragma omp target parallel map(from:a) + { + int b; + a = foo(b) + bar(); + } + return a; +} + +// parallel region +// CHECK: define {{.*}}void @{{.*}}(i32* noalias {{.*}}, i32* noalias {{.*}}, i32* dereferenceable{{.*}}) +// CHECK-NOT: call i8* @__kmpc_data_sharing_push_stack( +// CHECK: [[B_ADDR:%.+]] = alloca i32, +// CHECK: call {{.*}}[[FOO:@.*foo.*]](i32* dereferenceable{{.*}} [[B_ADDR]]) +// CHECK: call {{.*}}[[BAR:@.*bar.*]]() +// CHECK-NOT: call void @__kmpc_data_sharing_pop_stack( +// CHECK: ret void + +// CHECK: define {{.*}}[[FOO]](i32* dereferenceable{{.*}}) +// CHECK-NOT: @__kmpc_data_sharing_push_stack + +// CHECK: define {{.*}}[[BAR]]() +// CHECK: [[RES:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i64 4, i16 0) +// CHECK: [[GLOBALS:%.+]] = bitcast i8* [[RES]] to [[GLOBAL_ST:%struct[.].*]]* +// CHECK: [[A_ADDR:%.+]] = getelementptr inbounds [[GLOBAL_ST]], [[GLOBAL_ST]]* [[GLOBALS]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 +// CHECK: call {{.*}}[[FOO]](i32* dereferenceable{{.*}} [[A_ADDR]]) +// CHECK: call void @__kmpc_data_sharing_pop_stack(i8* [[RES]]) +// CHECK: ret i32 diff --git a/test/OpenMP/declare_target_link_codegen.cpp b/test/OpenMP/declare_target_link_codegen.cpp new file mode 100644 index 0000000000000..fdecd728ce684 --- /dev/null +++ b/test/OpenMP/declare_target_link_codegen.cpp @@ -0,0 +1,69 @@ +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix HOST --check-prefix CHECK +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix DEVICE --check-prefix CHECK +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -emit-pch -o %t +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -o - | FileCheck %s --check-prefix DEVICE --check-prefix CHECK + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o -| FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -emit-pch -o %t +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify -o - | FileCheck %s --check-prefix SIMD-ONLY + +// expected-no-diagnostics + +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +#ifndef HEADER +#define HEADER + +// HOST: @c = external global i32, +// DEVICE-NOT: @c = +// DEVICE: @c_decl_tgt_link_ptr = common global i32* null +// HOST: @c_decl_tgt_link_ptr = global i32* @c +// HOST: [[SIZES:@.+]] = private unnamed_addr constant [2 x i64] [i64 4, i64 4] +// HOST: [[MAPTYPES:@.+]] = private unnamed_addr constant [2 x i64] [i64 35, i64 531] +// HOST: @.omp_offloading.entry_name{{.*}} = internal unnamed_addr constant [{{[0-9]+}} x i8] c"c_decl_tgt_link_ptr\00" +// HOST: @.omp_offloading.entry.c_decl_tgt_link_ptr = weak constant %struct.__tgt_offload_entry { i8* bitcast (i32** @c_decl_tgt_link_ptr to i8*), i8* getelementptr inbounds ([{{[0-9]+}} x i8], [{{[0-9]+}} x i8]* @.omp_offloading.entry_name, i32 0, i32 0), i64 8, i32 1, i32 0 }, section ".omp_offloading.entries", align 1 +// DEVICE-NOT: internal unnamed_addr constant [{{[0-9]+}} x i8] c"c_decl_tgt_link_ptr\00" +// CHECK: @llvm.used = appending global [1 x i8*] [i8* bitcast (i32** @c_decl_tgt_link_ptr to i8*)] + +extern int c; +#pragma omp declare target link(c) + +int maini1() { + int a; +#pragma omp target map(tofrom : a) + { + a = c; + } + return 0; +} + +// DEVICE: define weak void @__omp_offloading_{{.*}}_{{.*}}maini1{{.*}}_l[[@LINE-7]](i32* dereferenceable{{[^,]*}} +// DEVICE: [[C_REF:%.+]] = load i32*, i32** @c_decl_tgt_link_ptr, +// DEVICE: [[C:%.+]] = load i32, i32* [[C_REF]], +// DEVICE: store i32 [[C]], i32* % + +// HOST: define {{.*}}i32 @{{.*}}maini1{{.*}}() +// HOST: [[BASEPTRS:%.+]] = alloca [2 x i8*], +// HOST: [[PTRS:%.+]] = alloca [2 x i8*], +// HOST: getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASEPTRS]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 +// HOST: getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTRS]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 +// HOST: [[BP1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASEPTRS]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 +// HOST: [[BP1_CAST:%.+]] = bitcast i8** [[BP1]] to i32*** +// HOST: store i32** @c_decl_tgt_link_ptr, i32*** [[BP1_CAST]], +// HOST: [[P1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTRS]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 +// HOST: [[P1_CAST:%.+]] = bitcast i8** [[P1]] to i32** +// HOST: store i32* @c, i32** [[P1_CAST]], +// HOST: [[BP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASEPTRS]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 +// HOST: [[P0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTRS]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 +// HOST: call i32 @__tgt_target(i64 -1, i8* @{{[^,]+}}, i32 2, i8** [[BP0]], i8** [[P0]], i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[SIZES]], i{{[0-9]+}} 0, i{{[0-9]+}} 0), i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPTYPES]], i{{[0-9]+}} 0, i{{[0-9]+}} 0)) +// HOST: call void @__omp_offloading_{{.*}}_{{.*}}maini1{{.*}}_l[[@LINE-26]](i32* %{{[^,]+}}) + +// HOST: define internal void @__omp_offloading_{{.*}}_{{.*}}maini1{{.*}}_l[[@LINE-28]](i32* dereferenceable{{.*}}) +// HOST: [[C:%.*]] = load i32, i32* @c, +// HOST: store i32 [[C]], i32* % + +// CHECK: !{i32 1, !"c_decl_tgt_link_ptr", i32 1, i32 {{[0-9]+}}} +#endif // HEADER diff --git a/test/OpenMP/declare_target_messages.cpp b/test/OpenMP/declare_target_messages.cpp index c1c03951a3c16..c8e73f6efbd71 100644 --- a/test/OpenMP/declare_target_messages.cpp +++ b/test/OpenMP/declare_target_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -fnoopenmp-use-tls -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -fnoopenmp-use-tls -ferror-limit 100 -o - %s + #pragma omp end declare target // expected-error {{unexpected OpenMP directive '#pragma omp end declare target'}} int a, b; // expected-warning {{declaration is not declared in any declare target region}} @@ -17,7 +19,7 @@ void f(); #pragma omp declare target link(foo2) // expected-error {{use of undeclared identifier 'foo2'}} -void c(); // expected-warning {{declaration is not declared in any declare target region}} +void c(); void func() {} // expected-note {{'func' defined here}} @@ -31,6 +33,33 @@ struct NonT { typedef int sint; +template <typename T> +T bla1() { return 0; } + +#pragma omp declare target +template <typename T> +T bla2() { return 0; } +#pragma omp end declare target + +template<> +float bla2() { return 1.0; } + +#pragma omp declare target +void blub2() { + bla2<float>(); + bla2<int>(); +} +#pragma omp end declare target + +void t2() { +#pragma omp target + { + bla2<float>(); + bla2<long>(); + } +} + + #pragma omp declare target // expected-note {{to match this '#pragma omp declare target'}} #pragma omp threadprivate(a) // expected-note {{defined as threadprivate or thread local}} extern int b; @@ -59,7 +88,7 @@ int C::method1() { return 0; } -void foo() { +void foo(int p) { a = 0; // expected-error {{threadprivate variables cannot be used in target constructs}} b = 0; // expected-note {{used here}} t = 1; // expected-error {{threadprivate variables cannot be used in target constructs}} @@ -67,9 +96,9 @@ void foo() { VC object1; g = object.method(); g += object.method1(); - g += object1.method(); + g += object1.method() + p; f(); - c(); // expected-note {{used here}} + c(); } #pragma omp declare target // expected-error {{expected '#pragma omp end declare target'}} void foo1() {} @@ -90,7 +119,7 @@ int main (int argc, char **argv) { #pragma omp declare target // expected-error {{unexpected OpenMP directive '#pragma omp declare target'}} int v; #pragma omp end declare target // expected-error {{unexpected OpenMP directive '#pragma omp end declare target'}} - foo(); + foo(v); return (0); } diff --git a/test/OpenMP/distribute_ast_print.cpp b/test/OpenMP/distribute_ast_print.cpp index d4a9df938f568..f675814e57a5a 100644 --- a/test/OpenMP/distribute_ast_print.cpp +++ b/test/OpenMP/distribute_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -83,7 +87,7 @@ T tmain(T argc) { static T a; // CHECK: static T a; #pragma omp distribute -// CHECK-NEXT: #pragma omp distribute +// CHECK-NEXT: #pragma omp distribute{{$}} for (int i=0; i < 2; ++i)a=2; // CHECK-NEXT: for (int i = 0; i < 2; ++i) // CHECK-NEXT: a = 2; diff --git a/test/OpenMP/distribute_codegen.cpp b/test/OpenMP/distribute_codegen.cpp index 7ea17b116c886..fe89ee0c1bdbe 100644 --- a/test/OpenMP/distribute_codegen.cpp +++ b/test/OpenMP/distribute_codegen.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix HCHECK +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target codegen - host bc file has to be created first. (no significant differences with host version of target region) // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s @@ -16,14 +24,24 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER -// CHECK-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CHECK-DAG: [[DEF_LOC_0:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } -// CHECK-DAG: [[DEF_LOC_DISTRIBUTE_0:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2050, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DAG: [[DEF_LOC_0:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DAG: [[DEF_LOC_DISTRIBUTE_0:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2050, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CHECK-LABEL: define {{.*void}} @{{.*}}without_schedule_clause{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}}) void without_schedule_clause(float *a, float *b, float *c, float *d) { @@ -49,7 +67,7 @@ void without_schedule_clause(float *a, float *b, float *c, float *d) { // CHECK-DAG: store i32 0, i32* [[LAST]] // CHECK-DAG: [[GBL_TID:%.+]] = load i32*, i32** [[TID_ADDR]] // CHECK-DAG: [[GBL_TIDV:%.+]] = load i32, i32* [[GBL_TID]] -// CHECK: call void @__kmpc_for_static_init_{{.+}}(%ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]], i32 92, i32* %.omp.is_last, i32* %.omp.lb, i32* %.omp.ub, i32* %.omp.stride, i32 1, i32 1) +// CHECK: call void @__kmpc_for_static_init_{{.+}}(%struct.ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]], i32 92, i32* %.omp.is_last, i32* %.omp.lb, i32* %.omp.ub, i32* %.omp.stride, i32 1, i32 1) // CHECK-DAG: [[UBV0:%.+]] = load i32, i32* [[UB]] // CHECK-DAG: [[USWITCH:%.+]] = icmp sgt i32 [[UBV0]], 4571423 // CHECK: br i1 [[USWITCH]], label %[[BBCT:.+]], label %[[BBCF:.+]] @@ -83,7 +101,7 @@ void without_schedule_clause(float *a, float *b, float *c, float *d) { // CHECK: [[BBINNEND]]: // CHECK: br label %[[LPEXIT:.+]] // CHECK: [[LPEXIT]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]]) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]]) // CHECK: ret void @@ -111,7 +129,7 @@ void static_not_chunked(float *a, float *b, float *c, float *d) { // CHECK-DAG: store i32 0, i32* [[LAST]] // CHECK-DAG: [[GBL_TID:%.+]] = load i32*, i32** [[TID_ADDR]] // CHECK-DAG: [[GBL_TIDV:%.+]] = load i32, i32* [[GBL_TID]] -// CHECK: call void @__kmpc_for_static_init_{{.+}}(%ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]], i32 92, i32* %.omp.is_last, i32* %.omp.lb, i32* %.omp.ub, i32* %.omp.stride, i32 1, i32 1) +// CHECK: call void @__kmpc_for_static_init_{{.+}}(%struct.ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]], i32 92, i32* %.omp.is_last, i32* %.omp.lb, i32* %.omp.ub, i32* %.omp.stride, i32 1, i32 1) // CHECK-DAG: [[UBV0:%.+]] = load i32, i32* [[UB]] // CHECK-DAG: [[USWITCH:%.+]] = icmp sgt i32 [[UBV0]], 4571423 // CHECK: br i1 [[USWITCH]], label %[[BBCT:.+]], label %[[BBCF:.+]] @@ -145,7 +163,7 @@ void static_not_chunked(float *a, float *b, float *c, float *d) { // CHECK: [[BBINNEND]]: // CHECK: br label %[[LPEXIT:.+]] // CHECK: [[LPEXIT]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]]) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]]) // CHECK: ret void @@ -173,7 +191,7 @@ void static_chunked(float *a, float *b, float *c, float *d) { // CHECK-DAG: store i32 0, i32* [[LAST]] // CHECK-DAG: [[GBL_TID:%.+]] = load i32*, i32** [[TID_ADDR]] // CHECK-DAG: [[GBL_TIDV:%.+]] = load i32, i32* [[GBL_TID]] -// CHECK: call void @__kmpc_for_static_init_{{.+}}(%ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]], i32 91, i32* %.omp.is_last, i32* %.omp.lb, i32* %.omp.ub, i32* %.omp.stride, i32 1, i32 5) +// CHECK: call void @__kmpc_for_static_init_{{.+}}(%struct.ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]], i32 91, i32* %.omp.is_last, i32* %.omp.lb, i32* %.omp.ub, i32* %.omp.stride, i32 1, i32 5) // CHECK-DAG: [[UBV0:%.+]] = load i32, i32* [[UB]] // CHECK-DAG: [[USWITCH:%.+]] = icmp ugt i32 [[UBV0]], 16908288 // CHECK: br i1 [[USWITCH]], label %[[BBCT:.+]], label %[[BBCF:.+]] @@ -207,7 +225,7 @@ void static_chunked(float *a, float *b, float *c, float *d) { // CHECK: [[BBINNEND]]: // CHECK: br label %[[LPEXIT:.+]] // CHECK: [[LPEXIT]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]]) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]]) // CHECK: ret void // CHECK-LABEL: test_precond diff --git a/test/OpenMP/distribute_collapse_messages.cpp b/test/OpenMP/distribute_collapse_messages.cpp index 4897e6931d274..b852fd2826b1b 100644 --- a/test/OpenMP/distribute_collapse_messages.cpp +++ b/test/OpenMP/distribute_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/distribute_dist_schedule_ast_print.cpp b/test/OpenMP/distribute_dist_schedule_ast_print.cpp index b06a56e3ee35e..de5e59aef73bf 100644 --- a/test/OpenMP/distribute_dist_schedule_ast_print.cpp +++ b/test/OpenMP/distribute_dist_schedule_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -14,7 +18,7 @@ T tmain(T argc) { static T a; // CHECK: static T a; #pragma omp distribute dist_schedule(static,10) -// CHECK-NEXT: #pragma omp distribute dist_schedule(static, 10) +// CHECK-NEXT: #pragma omp distribute dist_schedule(static, 10){{$}} for (int i=0; i < 2; ++i)a=2; // CHECK-NEXT: for (int i = 0; i < 2; ++i) // CHECK-NEXT: a = 2; diff --git a/test/OpenMP/distribute_dist_schedule_messages.cpp b/test/OpenMP/distribute_dist_schedule_messages.cpp index 98652a20034ea..9f91f41f86405 100644 --- a/test/OpenMP/distribute_dist_schedule_messages.cpp +++ b/test/OpenMP/distribute_dist_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/distribute_firstprivate_codegen.cpp b/test/OpenMP/distribute_firstprivate_codegen.cpp index 98c99d805ed5b..29d44fde92f24 100644 --- a/test/OpenMP/distribute_firstprivate_codegen.cpp +++ b/test/OpenMP/distribute_firstprivate_codegen.cpp @@ -5,12 +5,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -239,7 +255,7 @@ int main() { // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], // CHECK-DAG: [[VEC_ADDR_VAL_BCAST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i{{[0-9]+}}* // CHECK-DAG: [[VEC_PRIV_BCAST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i{{[0-9]+}}* -// CHECK-DAG: call void @llvm.memcpy.{{.*}}(i{{[0-9]+}}* [[VEC_PRIV_BCAST]], i{{[0-9]+}}* [[VEC_ADDR_VAL_BCAST]],{{.+}}) +// CHECK-DAG: call void @llvm.memcpy.{{.*}}(i{{[0-9]+}}* align {{[0-9]+}} [[VEC_PRIV_BCAST]], i{{[0-9]+}}* align {{[0-9]+}} [[VEC_ADDR_VAL_BCAST]],{{.+}}) // init s_arr // CHECK-DAG: [[S_ARR_ADDR_VAL:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[S_ARR_ADDR]], @@ -254,7 +270,7 @@ int main() { // CHECK-DAG: [[S_ARR_DST_PAST:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]],{{.+}} ], [ [[S_ARR_DST:%.+]],{{.+}} ] // CHECK-DAG: [[S_ARR_SRC_BCAST:%.+]] = bitcast{{.+}} [[S_ARR_SRC_PAST]] to{{.+}} // CHECK-DAG: [[S_ARR_DST_BCAST:%.+]] = bitcast{{.+}} [[S_ARR_DST_PAST]] to{{.+}} -// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* [[S_ARR_DST_BCAST]], {{.+}}* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[S_ARR_DST_BCAST]], {{.+}}* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK-DAG: [[S_ARR_SRC]] = getelementptr{{.+}} // CHECK-DAG: [[S_ARR_DST]] = getelementptr{{.+}} // CHECK-DAG: [[S_ARR_CPY_FIN:%.+]] = icmp{{.+}} [[S_ARR_DST]], [[S_ARR_PRIV_NEXT]] @@ -267,7 +283,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP]], // CHECK-DAG: [[VAR_PRIV_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_PRIV]] to{{.+}} // CHECK-DAG: [[TMP_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL]] to{{.+}} -// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* [[VAR_PRIV_BCAST]], {{.+}}* [[TMP_BCAST]],{{.+}}) +// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[VAR_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[TMP_BCAST]],{{.+}}) // CHECK-DAG: store [[S_FLOAT_TY]]* [[VAR_PRIV]], [[S_FLOAT_TY]]** [[TMP_PRIV]], // init svar @@ -334,7 +350,7 @@ int main() { // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], // CHECK-DAG: [[VEC_ADDR_VAL_BCAST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i{{[0-9]+}}* // CHECK-DAG: [[VEC_PRIV_BCAST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i{{[0-9]+}}* -// CHECK-DAG: call void @llvm.memcpy.{{.*}}(i{{[0-9]+}}* [[VEC_PRIV_BCAST]], i{{[0-9]+}}* [[VEC_ADDR_VAL_BCAST]],{{.+}}) +// CHECK-DAG: call void @llvm.memcpy.{{.*}}(i{{[0-9]+}}* align {{[0-9]+}} [[VEC_PRIV_BCAST]], i{{[0-9]+}}* align {{[0-9]+}} [[VEC_ADDR_VAL_BCAST]],{{.+}}) // init s_arr // CHECK-DAG: [[S_ARR_ADDR_VAL:%.+]] = load [2 x [[S_INT_TY]]]*, [2 x [[S_INT_TY]]]** [[S_ARR_ADDR]], @@ -349,7 +365,7 @@ int main() { // CHECK-DAG: [[S_ARR_DST_PAST:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]],{{.+}} ], [ [[S_ARR_DST:%.+]],{{.+}} ] // CHECK-DAG: [[S_ARR_SRC_BCAST:%.+]] = bitcast{{.+}} [[S_ARR_SRC_PAST]] to{{.+}} // CHECK-DAG: [[S_ARR_DST_BCAST:%.+]] = bitcast{{.+}} [[S_ARR_DST_PAST]] to{{.+}} -// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* [[S_ARR_DST_BCAST]], {{.+}}* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[S_ARR_DST_BCAST]], {{.+}}* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK-DAG: [[S_ARR_SRC]] = getelementptr{{.+}} // CHECK-DAG: [[S_ARR_DST]] = getelementptr{{.+}} // CHECK-DAG: [[S_ARR_CPY_FIN:%.+]] = icmp{{.+}} [[S_ARR_DST]], [[S_ARR_PRIV_NEXT]] @@ -362,7 +378,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP]], // CHECK-DAG: [[VAR_PRIV_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_PRIV]] to{{.+}} // CHECK-DAG: [[TMP_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL]] to{{.+}} -// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* [[VAR_PRIV_BCAST]], {{.+}}* [[TMP_BCAST]],{{.+}}) +// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[VAR_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[TMP_BCAST]],{{.+}}) // CHECK-DAG: store [[S_INT_TY]]* [[VAR_PRIV]], [[S_INT_TY]]** [[TMP_PRIV]], // CHECK-DAG: store i{{[0-9]+}} 0, i{{[0-9]+}}* %.omp{{.+}}, diff --git a/test/OpenMP/distribute_firstprivate_messages.cpp b/test/OpenMP/distribute_firstprivate_messages.cpp index a59c9528090fb..ea0e7b4a23145 100644 --- a/test/OpenMP/distribute_firstprivate_messages.cpp +++ b/test/OpenMP/distribute_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/distribute_lastprivate_codegen.cpp b/test/OpenMP/distribute_lastprivate_codegen.cpp index bd9dae956f1ce..66a860b7152e9 100644 --- a/test/OpenMP/distribute_lastprivate_codegen.cpp +++ b/test/OpenMP/distribute_lastprivate_codegen.cpp @@ -5,12 +5,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -240,7 +256,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK-DAG: [[S_ARR_PTR_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_PTR]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST]], i8* [[TMP_VAL_BCAST]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST]], // CHECK: call void @__kmpc_for_static_fini( // lastprivates @@ -253,7 +269,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -264,7 +280,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_FLOAT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -273,7 +289,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: [[SVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[S_VAR_PRIV]], // CHECK: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR_REF]], // CHECK: ret void @@ -342,7 +358,7 @@ int main() { // CHECK-DAG: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK-DAG: [[S_ARR_PTR_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_PTR1]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST1]], i8* [[TMP_VAL_BCAST1]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST1]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST1]], // CHECK: call void @__kmpc_for_static_fini( // lastprivates @@ -355,7 +371,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF1]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF1]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_PRIV1]] to [[S_INT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -366,7 +382,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_INT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -375,6 +391,6 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: ret void #endif diff --git a/test/OpenMP/distribute_parallel_for_ast_print.cpp b/test/OpenMP/distribute_parallel_for_ast_print.cpp index 351e15803c092..58e6eb8aa77a6 100644 --- a/test/OpenMP/distribute_parallel_for_ast_print.cpp +++ b/test/OpenMP/distribute_parallel_for_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -std=c++11 -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -std=c++11 -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -37,7 +41,7 @@ public: } }; -// CHECK: #pragma omp distribute parallel for private(this->a) private(this->a) private(T::a) +// CHECK: #pragma omp distribute parallel for private(this->a) private(this->a) private(T::a){{$}} // CHECK: #pragma omp distribute parallel for private(this->a) private(this->a) // CHECK: #pragma omp distribute parallel for private(this->a) private(this->a) private(this->S::a) diff --git a/test/OpenMP/distribute_parallel_for_codegen.cpp b/test/OpenMP/distribute_parallel_for_codegen.cpp index 4fbca7d2f29a6..a7700a3421acc 100644 --- a/test/OpenMP/distribute_parallel_for_codegen.cpp +++ b/test/OpenMP/distribute_parallel_for_codegen.cpp @@ -6,12 +6,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/distribute_parallel_for_collapse_messages.cpp b/test/OpenMP/distribute_parallel_for_collapse_messages.cpp index 41976a69ee2c9..f01dfeea5c273 100644 --- a/test/OpenMP/distribute_parallel_for_collapse_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_copyin_messages.cpp b/test/OpenMP/distribute_parallel_for_copyin_messages.cpp index 7d703413488df..5182f24bdb148 100644 --- a/test/OpenMP/distribute_parallel_for_copyin_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_copyin_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_default_messages.cpp b/test/OpenMP/distribute_parallel_for_default_messages.cpp index 3437bd55cf3fb..cb9cbcf6ba410 100644 --- a/test/OpenMP/distribute_parallel_for_default_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); template <class T, int N> diff --git a/test/OpenMP/distribute_parallel_for_dist_schedule_messages.cpp b/test/OpenMP/distribute_parallel_for_dist_schedule_messages.cpp index 0f5820ed4bbb8..adcef0bab92a0 100644 --- a/test/OpenMP/distribute_parallel_for_dist_schedule_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_dist_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_firstprivate_codegen.cpp b/test/OpenMP/distribute_parallel_for_firstprivate_codegen.cpp index de348272950c2..e4915acf69b68 100644 --- a/test/OpenMP/distribute_parallel_for_firstprivate_codegen.cpp +++ b/test/OpenMP/distribute_parallel_for_firstprivate_codegen.cpp @@ -5,12 +5,27 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -313,7 +328,7 @@ int main() { // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[VEC_ADDR]], // CHECK-DAG: [[VEC_PRIV_BCAST:%.+]] = bitcast {{.+}} [[VEC_PRIV]] to // CHECK-DAG: [[VEC_ADDR_BCAST:%.+]] = bitcast {{.+}} [[VEC_ADDR_VAL]] to -// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* [[VEC_PRIV_BCAST]], {{.+}}* [[VEC_ADDR_BCAST]], +// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* align {{[0-9]+}} [[VEC_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[VEC_ADDR_BCAST]], // s_arr // CHECK-DAG: [[S_ARR_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[S_ARR_ADDR]], @@ -330,7 +345,7 @@ int main() { // CHECK-DAG: [[TMP_REF:%.+]] = load {{.+}}*, {{.+}}* [[TMP]], // CHECK-DAG: [[VAR_PRIV_BCAST:%.+]] = bitcast {{.+}}* [[VAR_PRIV]] to // CHECK-DAG: [[TMP_REF_BCAST:%.+]] = bitcast {{.+}}* [[TMP_REF]] to -// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* [[VAR_PRIV_BCAST]], {{.+}}* [[TMP_REF_BCAST]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[VAR_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[TMP_REF_BCAST]], // CHECK-DAG: store {{.+}}* [[VAR_PRIV]], {{.+}}** [[TMP_PRIV]], // svar @@ -406,7 +421,7 @@ int main() { // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[VEC_ADDR]], // CHECK-DAG: [[VEC_PRIV_BCAST:%.+]] = bitcast {{.+}} [[VEC_PRIV]] to // CHECK-DAG: [[VEC_ADDR_BCAST:%.+]] = bitcast {{.+}} [[VEC_ADDR_VAL]] to -// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* [[VEC_PRIV_BCAST]], {{.+}}* [[VEC_ADDR_BCAST]], +// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* align {{[0-9]+}} [[VEC_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[VEC_ADDR_BCAST]], // s_arr // CHECK-DAG: [[S_ARR_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[S_ARR_ADDR]], @@ -423,7 +438,7 @@ int main() { // CHECK-DAG: [[VAR_ADDR_REF:%.+]] = load {{.+}}*, {{.+}}* [[VAR_ADDR]], // CHECK-DAG: [[VAR_PRIV_BCAST:%.+]] = bitcast {{.+}}* [[VAR_PRIV]] to // CHECK-DAG: [[VAR_ADDR_BCAST:%.+]] = bitcast {{.+}}* [[VAR_ADDR_REF]] to -// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* [[VAR_PRIV_BCAST]], {{.+}}* [[VAR_ADDR_BCAST]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[VAR_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[VAR_ADDR_BCAST]], // CHECK-DAG: store {{.+}}* [[VAR_PRIV]], {{.+}}** [[TMP_PRIV]], // CHECK: call void @__kmpc_for_static_init_4( @@ -495,7 +510,7 @@ int main() { // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[VEC_ADDR]], // CHECK-DAG: [[VEC_PRIV_BCAST:%.+]] = bitcast {{.+}} [[VEC_PRIV]] to // CHECK-DAG: [[VEC_ADDR_BCAST:%.+]] = bitcast {{.+}} [[VEC_ADDR_VAL]] to -// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* [[VEC_PRIV_BCAST]], {{.+}}* [[VEC_ADDR_BCAST]], +// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* align {{[0-9]+}} [[VEC_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[VEC_ADDR_BCAST]], // s_arr // CHECK-DAG: [[S_ARR_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[S_ARR_ADDR]], @@ -512,7 +527,7 @@ int main() { // CHECK-DAG: [[TMP_REF:%.+]] = load {{.+}}*, {{.+}}* [[TMP]], // CHECK-DAG: [[VAR_PRIV_BCAST:%.+]] = bitcast {{.+}}* [[VAR_PRIV]] to // CHECK-DAG: [[TMP_REF_BCAST:%.+]] = bitcast {{.+}}* [[TMP_REF]] to -// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* [[VAR_PRIV_BCAST]], {{.+}}* [[TMP_REF_BCAST]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[VAR_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[TMP_REF_BCAST]], // CHECK-DAG: store {{.+}}* [[VAR_PRIV]], {{.+}}** [[TMP_PRIV]], // CHECK: call void @__kmpc_for_static_init_4( @@ -576,7 +591,7 @@ int main() { // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[VEC_ADDR]], // CHECK-DAG: [[VEC_PRIV_BCAST:%.+]] = bitcast {{.+}} [[VEC_PRIV]] to // CHECK-DAG: [[VEC_ADDR_BCAST:%.+]] = bitcast {{.+}} [[VEC_ADDR_VAL]] to -// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* [[VEC_PRIV_BCAST]], {{.+}}* [[VEC_ADDR_BCAST]], +// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* align {{[0-9]+}} [[VEC_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[VEC_ADDR_BCAST]], // s_arr // CHECK-DAG: [[S_ARR_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[S_ARR_ADDR]], @@ -593,7 +608,7 @@ int main() { // CHECK-DAG: [[VAR_ADDR_REF:%.+]] = load {{.+}}*, {{.+}}* [[VAR_ADDR]], // CHECK-DAG: [[VAR_PRIV_BCAST:%.+]] = bitcast {{.+}}* [[VAR_PRIV]] to // CHECK-DAG: [[VAR_ADDR_BCAST:%.+]] = bitcast {{.+}}* [[VAR_ADDR_REF]] to -// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* [[VAR_PRIV_BCAST]], {{.+}}* [[VAR_ADDR_BCAST]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[VAR_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[VAR_ADDR_BCAST]], // CHECK-DAG: store {{.+}}* [[VAR_PRIV]], {{.+}}** [[TMP_PRIV]], // CHECK: call void @__kmpc_for_static_init_4( diff --git a/test/OpenMP/distribute_parallel_for_firstprivate_messages.cpp b/test/OpenMP/distribute_parallel_for_firstprivate_messages.cpp index 67075ca12939b..63f4dbec1786a 100644 --- a/test/OpenMP/distribute_parallel_for_firstprivate_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_if_codegen.cpp b/test/OpenMP/distribute_parallel_for_if_codegen.cpp index 9ee909e0139c2..7e7f24f3c9e8e 100644 --- a/test/OpenMP/distribute_parallel_for_if_codegen.cpp +++ b/test/OpenMP/distribute_parallel_for_if_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/distribute_parallel_for_if_messages.cpp b/test/OpenMP/distribute_parallel_for_if_messages.cpp index 5e9d9ad9e1f2b..85e78d0b92534 100644 --- a/test/OpenMP/distribute_parallel_for_if_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_lastprivate_codegen.cpp b/test/OpenMP/distribute_parallel_for_lastprivate_codegen.cpp index fccb0cfdb3d8d..dda8c932908ef 100644 --- a/test/OpenMP/distribute_parallel_for_lastprivate_codegen.cpp +++ b/test/OpenMP/distribute_parallel_for_lastprivate_codegen.cpp @@ -5,12 +5,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -333,7 +349,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -344,7 +360,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_FLOAT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -353,7 +369,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: [[SVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[S_VAR_PRIV]], // CHECK: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR_REF]], // CHECK: ret void @@ -425,7 +441,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK-DAG: [[S_ARR_PTR_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_PTR]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST]], i8* [[TMP_VAL_BCAST]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST]], // CHECK: call void @__kmpc_for_static_fini( @@ -439,7 +455,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -450,7 +466,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_FLOAT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -459,7 +475,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_ADDR_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: [[SVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[S_VAR_PRIV]], // CHECK: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR_REF]], // CHECK: ret void @@ -533,7 +549,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]] to [[S_INT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -544,7 +560,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_INT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -553,7 +569,7 @@ int main() { // CHECK: [[TMP_VAL:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_REF]] to i8* // CHECK: [[TMP_VAL_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST]],{{.+}}) // CHECK: ret void // outlined function for 'parallel for' @@ -619,7 +635,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV]], // CHECK-DAG: [[S_ARR_PTR_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_PTR]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL]] to i8* -// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST]], i8* [[TMP_VAL_BCAST]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST]], // CHECK: call void @__kmpc_for_static_fini( @@ -633,7 +649,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]] to [[S_INT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -644,7 +660,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_INT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -653,7 +669,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_ADDR_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: ret void #endif diff --git a/test/OpenMP/distribute_parallel_for_lastprivate_messages.cpp b/test/OpenMP/distribute_parallel_for_lastprivate_messages.cpp index b38d978cca034..c8eaad8cf1474 100644 --- a/test/OpenMP/distribute_parallel_for_lastprivate_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_messages.cpp b/test/OpenMP/distribute_parallel_for_messages.cpp index b1bb8d088e907..5ea73a88fc54d 100644 --- a/test/OpenMP/distribute_parallel_for_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -std=c++11 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -std=c++11 -o - %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_num_threads_codegen.cpp b/test/OpenMP/distribute_parallel_for_num_threads_codegen.cpp index 89e9905044399..21d0949454274 100644 --- a/test/OpenMP/distribute_parallel_for_num_threads_codegen.cpp +++ b/test/OpenMP/distribute_parallel_for_num_threads_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/distribute_parallel_for_num_threads_messages.cpp b/test/OpenMP/distribute_parallel_for_num_threads_messages.cpp index 7939514249a07..110f5d70f8d17 100644 --- a/test/OpenMP/distribute_parallel_for_num_threads_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_num_threads_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_private_codegen.cpp b/test/OpenMP/distribute_parallel_for_private_codegen.cpp index d8d8a9a1edec9..a731031db7341 100644 --- a/test/OpenMP/distribute_parallel_for_private_codegen.cpp +++ b/test/OpenMP/distribute_parallel_for_private_codegen.cpp @@ -5,12 +5,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/distribute_parallel_for_private_messages.cpp b/test/OpenMP/distribute_parallel_for_private_messages.cpp index 465357a433886..75fe47ae8204a 100644 --- a/test/OpenMP/distribute_parallel_for_private_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_proc_bind_codegen.cpp b/test/OpenMP/distribute_parallel_for_proc_bind_codegen.cpp index 958b4e782f109..fce005be80fb4 100644 --- a/test/OpenMP/distribute_parallel_for_proc_bind_codegen.cpp +++ b/test/OpenMP/distribute_parallel_for_proc_bind_codegen.cpp @@ -3,6 +3,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -60,7 +65,7 @@ int main() { // CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], // CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 4) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: ret void // CHECK: define{{.+}} [[OFFL2]]() @@ -72,7 +77,7 @@ int main() { // CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], // CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 3) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: ret void // CHECK: define{{.+}} [[TMAIN]]() @@ -88,6 +93,6 @@ int main() { // CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], // CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 2) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: ret void #endif diff --git a/test/OpenMP/distribute_parallel_for_proc_bind_messages.cpp b/test/OpenMP/distribute_parallel_for_proc_bind_messages.cpp index 9898f9dc65b69..febf4b05643a3 100644 --- a/test/OpenMP/distribute_parallel_for_proc_bind_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_proc_bind_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); template <class T, typename S, int N> diff --git a/test/OpenMP/distribute_parallel_for_reduction_codegen.cpp b/test/OpenMP/distribute_parallel_for_reduction_codegen.cpp new file mode 100644 index 0000000000000..dc2a3ca5350fd --- /dev/null +++ b/test/OpenMP/distribute_parallel_for_reduction_codegen.cpp @@ -0,0 +1,84 @@ +// Test host code gen + +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + + +template <typename T> +T tmain(T &r) { + int n = 1000; + // schedule: dynamic chunk + #pragma omp target map(tofrom:r) + #pragma omp teams + #pragma omp distribute parallel for reduction(+:r) + for (int i = 0; i < n; ++i) + r += (T)i; + + return r; +} + +int main() { + int n = 1000; + int r = 0; + #pragma omp target map(tofrom:r) + #pragma omp teams + #pragma omp distribute parallel for reduction(+:r) + for (int i = 0; i < n; ++i) + r += i; + + return tmain<int>(r); +} + +// CHECK-LABEL: main +// CHECK: call{{.+}} @__tgt_target_teams( +// CHECK: call void [[OFFL:@.+]]( +// CHECK: call{{.+}} [[TMAIN:@.+]](i{{32|64}} +// CHECK: ret + +// CHECK: define{{.+}} [[OFFL]]( +// CHECK: call{{.+}} @__kmpc_fork_teams({{.+}}, {{.+}}, {{.+}} [[TEOUTL:@.+]] to{{.+}} +// CHECK: ret void + +// CHECK: define{{.+}} [[TEOUTL]]( +// CHECK: call{{.+}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} [[PAROUTL:@.+]] to{{.+}} +// CHECK: ret void + +// CHECK: define{{.+}} [[PAROUTL]]( +// CHECK: call{{.+}} @__kmpc_reduce_nowait( +// CHECK: call{{.+}} @__kmpc_end_reduce_nowait( +// CHECK: ret void + +// CHECK: define{{.+}} [[TMAIN]](i{{32|64}} +// CHECK: call{{.+}} @__tgt_target_teams( +// CHECK: call void [[TOFFL:@.+]]( +// CHECK: ret + +// CHECK: define{{.+}} [[TOFFL]]( +// CHECK: call{{.+}} @__kmpc_fork_teams({{.+}}, {{.+}}, {{.+}} [[TEMPLTEOUTL:@.+]] to{{.+}} +// CHECK: ret void + +// CHECK: define{{.+}} [[TEMPLTEOUTL]]( +// CHECK: call{{.+}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} [[TPAROUTL:@.+]] to{{.+}} +// CHECK: ret void + +// CHECK: define{{.+}} [[TPAROUTL]]( +// CHECK: call{{.+}} @__kmpc_reduce_nowait( +// CHECK: call{{.+}} @__kmpc_end_reduce_nowait( +// CHECK: ret void + +#endif // HEADER diff --git a/test/OpenMP/distribute_parallel_for_reduction_messages.cpp b/test/OpenMP/distribute_parallel_for_reduction_messages.cpp index 0bf1bc0472eb6..21a1eb40608de 100644 --- a/test/OpenMP/distribute_parallel_for_reduction_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_schedule_messages.cpp b/test/OpenMP/distribute_parallel_for_schedule_messages.cpp index 6363cd7caef01..1f521d4f6fd6f 100644 --- a/test/OpenMP/distribute_parallel_for_schedule_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_shared_messages.cpp b/test/OpenMP/distribute_parallel_for_shared_messages.cpp index d5725e7f1b09d..01c582cdbd44b 100644 --- a/test/OpenMP/distribute_parallel_for_shared_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_shared_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + struct S1; // expected-note 2 {{declared here}} extern S1 a; diff --git a/test/OpenMP/distribute_parallel_for_simd_aligned_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_aligned_messages.cpp index 2d99d7d8f4bc8..0c94f5b12d0c1 100644 --- a/test/OpenMP/distribute_parallel_for_simd_aligned_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_aligned_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp %s +// RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp-simd %s + struct B { static int ib[20]; // expected-note 0 {{'B::ib' declared here}} static constexpr int bfoo() { return 8; } diff --git a/test/OpenMP/distribute_parallel_for_simd_ast_print.cpp b/test/OpenMP/distribute_parallel_for_simd_ast_print.cpp index 21010638be152..fff6f965b4c23 100644 --- a/test/OpenMP/distribute_parallel_for_simd_ast_print.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -38,7 +42,7 @@ public: } }; -// CHECK: #pragma omp distribute parallel for simd private(this->a) private(this->a) private(T::a) +// CHECK: #pragma omp distribute parallel for simd private(this->a) private(this->a) private(T::a){{$}} // CHECK: #pragma omp distribute parallel for simd private(this->a) private(this->a) linear(k) // CHECK: #pragma omp distribute parallel for simd private(this->a) private(this->a) private(this->S::a) diff --git a/test/OpenMP/distribute_parallel_for_simd_codegen.cpp b/test/OpenMP/distribute_parallel_for_simd_codegen.cpp index 59e6df90bbcb7..129d90b746c4f 100644 --- a/test/OpenMP/distribute_parallel_for_simd_codegen.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_codegen.cpp @@ -6,12 +6,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/distribute_parallel_for_simd_collapse_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_collapse_messages.cpp index 8690b4aac432a..b12dcc19d0205 100644 --- a/test/OpenMP/distribute_parallel_for_simd_collapse_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_simd_copyin_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_copyin_messages.cpp index 2d1a3dc5124b8..2d72925518d4c 100644 --- a/test/OpenMP/distribute_parallel_for_simd_copyin_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_copyin_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_simd_default_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_default_messages.cpp index 5c32306d96d16..3014d039d296c 100644 --- a/test/OpenMP/distribute_parallel_for_simd_default_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); template <class T, int N> diff --git a/test/OpenMP/distribute_parallel_for_simd_dist_schedule_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_dist_schedule_messages.cpp index 2e3ee2b1d50e4..f029b0e187fa8 100644 --- a/test/OpenMP/distribute_parallel_for_simd_dist_schedule_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_dist_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_simd_firstprivate_codegen.cpp b/test/OpenMP/distribute_parallel_for_simd_firstprivate_codegen.cpp index b8d0f4ad7540d..1447b5ff2ee21 100644 --- a/test/OpenMP/distribute_parallel_for_simd_firstprivate_codegen.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_firstprivate_codegen.cpp @@ -5,12 +5,27 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -311,7 +326,7 @@ int main() { // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[VEC_ADDR]], // CHECK-DAG: [[VEC_PRIV_BCAST:%.+]] = bitcast {{.+}} [[VEC_PRIV]] to // CHECK-DAG: [[VEC_ADDR_BCAST:%.+]] = bitcast {{.+}} [[VEC_ADDR_VAL]] to -// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* [[VEC_PRIV_BCAST]], {{.+}}* [[VEC_ADDR_BCAST]], +// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* align {{[0-9]+}} [[VEC_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[VEC_ADDR_BCAST]], // s_arr // CHECK-DAG: [[S_ARR_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[S_ARR_ADDR]], @@ -328,7 +343,7 @@ int main() { // CHECK-DAG: [[TMP_REF:%.+]] = load {{.+}}*, {{.+}}* [[TMP]], // CHECK-DAG: [[VAR_PRIV_BCAST:%.+]] = bitcast {{.+}}* [[VAR_PRIV]] to // CHECK-DAG: [[TMP_REF_BCAST:%.+]] = bitcast {{.+}}* [[TMP_REF]] to -// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* [[VAR_PRIV_BCAST]], {{.+}}* [[TMP_REF_BCAST]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[VAR_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[TMP_REF_BCAST]], // CHECK-DAG: store {{.+}}* [[VAR_PRIV]], {{.+}}** [[TMP_PRIV]], // svar @@ -404,7 +419,7 @@ int main() { // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[VEC_ADDR]], // CHECK-DAG: [[VEC_PRIV_BCAST:%.+]] = bitcast {{.+}} [[VEC_PRIV]] to // CHECK-DAG: [[VEC_ADDR_BCAST:%.+]] = bitcast {{.+}} [[VEC_ADDR_VAL]] to -// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* [[VEC_PRIV_BCAST]], {{.+}}* [[VEC_ADDR_BCAST]], +// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* align {{[0-9]+}} [[VEC_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[VEC_ADDR_BCAST]], // s_arr // CHECK-DAG: [[S_ARR_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[S_ARR_ADDR]], @@ -421,7 +436,7 @@ int main() { // CHECK-DAG: [[VAR_ADDR_REF:%.+]] = load {{.+}}*, {{.+}}* [[VAR_ADDR]], // CHECK-DAG: [[VAR_PRIV_BCAST:%.+]] = bitcast {{.+}}* [[VAR_PRIV]] to // CHECK-DAG: [[VAR_ADDR_BCAST:%.+]] = bitcast {{.+}}* [[VAR_ADDR_REF]] to -// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* [[VAR_PRIV_BCAST]], {{.+}}* [[VAR_ADDR_BCAST]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[VAR_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[VAR_ADDR_BCAST]], // CHECK-DAG: store {{.+}}* [[VAR_PRIV]], {{.+}}** [[TMP_PRIV]], // CHECK: call void @__kmpc_for_static_init_4( @@ -493,7 +508,7 @@ int main() { // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[VEC_ADDR]], // CHECK-DAG: [[VEC_PRIV_BCAST:%.+]] = bitcast {{.+}} [[VEC_PRIV]] to // CHECK-DAG: [[VEC_ADDR_BCAST:%.+]] = bitcast {{.+}} [[VEC_ADDR_VAL]] to -// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* [[VEC_PRIV_BCAST]], {{.+}}* [[VEC_ADDR_BCAST]], +// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* align {{[0-9]+}} [[VEC_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[VEC_ADDR_BCAST]], // s_arr // CHECK-DAG: [[S_ARR_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[S_ARR_ADDR]], @@ -510,7 +525,7 @@ int main() { // CHECK-DAG: [[TMP_REF:%.+]] = load {{.+}}*, {{.+}}* [[TMP]], // CHECK-DAG: [[VAR_PRIV_BCAST:%.+]] = bitcast {{.+}}* [[VAR_PRIV]] to // CHECK-DAG: [[TMP_REF_BCAST:%.+]] = bitcast {{.+}}* [[TMP_REF]] to -// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* [[VAR_PRIV_BCAST]], {{.+}}* [[TMP_REF_BCAST]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[VAR_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[TMP_REF_BCAST]], // CHECK-DAG: store {{.+}}* [[VAR_PRIV]], {{.+}}** [[TMP_PRIV]], // CHECK: call void @__kmpc_for_static_init_4( @@ -574,7 +589,7 @@ int main() { // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[VEC_ADDR]], // CHECK-DAG: [[VEC_PRIV_BCAST:%.+]] = bitcast {{.+}} [[VEC_PRIV]] to // CHECK-DAG: [[VEC_ADDR_BCAST:%.+]] = bitcast {{.+}} [[VEC_ADDR_VAL]] to -// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* [[VEC_PRIV_BCAST]], {{.+}}* [[VEC_ADDR_BCAST]], +// CHECK-DAG: call void @llvm.memcpy{{.+}}({{.+}}* align {{[0-9]+}} [[VEC_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[VEC_ADDR_BCAST]], // s_arr // CHECK-DAG: [[S_ARR_ADDR_VAL:%.+]] = load {{.+}}*, {{.+}}** [[S_ARR_ADDR]], @@ -591,7 +606,7 @@ int main() { // CHECK-DAG: [[VAR_ADDR_REF:%.+]] = load {{.+}}*, {{.+}}* [[VAR_ADDR]], // CHECK-DAG: [[VAR_PRIV_BCAST:%.+]] = bitcast {{.+}}* [[VAR_PRIV]] to // CHECK-DAG: [[VAR_ADDR_BCAST:%.+]] = bitcast {{.+}}* [[VAR_ADDR_REF]] to -// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* [[VAR_PRIV_BCAST]], {{.+}}* [[VAR_ADDR_BCAST]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[VAR_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[VAR_ADDR_BCAST]], // CHECK-DAG: store {{.+}}* [[VAR_PRIV]], {{.+}}** [[TMP_PRIV]], // CHECK: call void @__kmpc_for_static_init_4( diff --git a/test/OpenMP/distribute_parallel_for_simd_firstprivate_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_firstprivate_messages.cpp index 646407643b25b..d6f56d5b21698 100644 --- a/test/OpenMP/distribute_parallel_for_simd_firstprivate_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_simd_if_codegen.cpp b/test/OpenMP/distribute_parallel_for_simd_if_codegen.cpp index b550387201340..2beea79c42fe6 100644 --- a/test/OpenMP/distribute_parallel_for_simd_if_codegen.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_if_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/distribute_parallel_for_simd_if_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_if_messages.cpp index c6957b9a868bb..0f9e12381f0e5 100644 --- a/test/OpenMP/distribute_parallel_for_simd_if_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_simd_lastprivate_codegen.cpp b/test/OpenMP/distribute_parallel_for_simd_lastprivate_codegen.cpp index 89ef33abd64d7..af74ab7b72a56 100644 --- a/test/OpenMP/distribute_parallel_for_simd_lastprivate_codegen.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_lastprivate_codegen.cpp @@ -5,12 +5,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -343,7 +359,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -354,7 +370,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_FLOAT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -363,7 +379,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: [[SVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[S_VAR_PRIV]], // CHECK: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR_REF]], // CHECK: ret void @@ -435,7 +451,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK-DAG: [[S_ARR_PTR_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_PTR]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST]], i8* [[TMP_VAL_BCAST]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST]], // CHECK: call void @__kmpc_for_static_fini( @@ -449,7 +465,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -460,7 +476,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_FLOAT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -469,7 +485,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_ADDR_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: [[SVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[S_VAR_PRIV]], // CHECK: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR_REF]], // CHECK: ret void @@ -543,7 +559,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]] to [[S_INT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -554,7 +570,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_INT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -563,7 +579,7 @@ int main() { // CHECK: [[TMP_VAL:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_REF]] to i8* // CHECK: [[TMP_VAL_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST]],{{.+}}) // CHECK: ret void // outlined function for 'parallel for' @@ -629,7 +645,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV]], // CHECK-DAG: [[S_ARR_PTR_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_PTR]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL]] to i8* -// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST]], i8* [[TMP_VAL_BCAST]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST]], // CHECK: call void @__kmpc_for_static_fini( @@ -643,7 +659,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]] to [[S_INT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -654,7 +670,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_INT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -663,7 +679,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_ADDR_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: ret void // CHECK: !{!"llvm.loop.vectorize.enable", i1 true} diff --git a/test/OpenMP/distribute_parallel_for_simd_lastprivate_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_lastprivate_messages.cpp index 73450d4f1181f..45e0d44322a5a 100644 --- a/test/OpenMP/distribute_parallel_for_simd_lastprivate_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_simd_linear_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_linear_messages.cpp index 66fd94b9e0f2a..788bace434f9f 100644 --- a/test/OpenMP/distribute_parallel_for_simd_linear_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/distribute_parallel_for_simd_loop_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_loop_messages.cpp index d137d47ca2a4b..20ee49a6fcc4c 100644 --- a/test/OpenMP/distribute_parallel_for_simd_loop_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_loop_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s + class S { int a; S() : a(0) {} // expected-note {{implicitly declared private here}} diff --git a/test/OpenMP/distribute_parallel_for_simd_misc_messages.c b/test/OpenMP/distribute_parallel_for_simd_misc_messages.c index 4d071621fcb8c..c2bee7ffa70e0 100644 --- a/test/OpenMP/distribute_parallel_for_simd_misc_messages.c +++ b/test/OpenMP/distribute_parallel_for_simd_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp distribute parallel for simd'}} #pragma omp distribute parallel for simd diff --git a/test/OpenMP/distribute_parallel_for_simd_num_threads_codegen.cpp b/test/OpenMP/distribute_parallel_for_simd_num_threads_codegen.cpp index 5804a2a72b712..014fb9523fe59 100644 --- a/test/OpenMP/distribute_parallel_for_simd_num_threads_codegen.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_num_threads_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/distribute_parallel_for_simd_num_threads_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_num_threads_messages.cpp index 0d6376d76b7ba..5f5165d979c26 100644 --- a/test/OpenMP/distribute_parallel_for_simd_num_threads_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_num_threads_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_simd_private_codegen.cpp b/test/OpenMP/distribute_parallel_for_simd_private_codegen.cpp index 8ffe521bfcbc7..629e023b6597f 100644 --- a/test/OpenMP/distribute_parallel_for_simd_private_codegen.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_private_codegen.cpp @@ -5,12 +5,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/distribute_parallel_for_simd_private_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_private_messages.cpp index 9df368890c8d5..d23d9176ed2ee 100644 --- a/test/OpenMP/distribute_parallel_for_simd_private_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_simd_proc_bind_codegen.cpp b/test/OpenMP/distribute_parallel_for_simd_proc_bind_codegen.cpp index 7f0b631aee7f1..4fb1f5b0274d9 100644 --- a/test/OpenMP/distribute_parallel_for_simd_proc_bind_codegen.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_proc_bind_codegen.cpp @@ -3,6 +3,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -60,7 +65,7 @@ int main() { // CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], // CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 4) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: ret void // CHECK: define{{.+}} [[OFFL2]]() @@ -72,7 +77,7 @@ int main() { // CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], // CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 3) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: ret void // CHECK: define{{.+}} [[TMAIN]]() @@ -88,6 +93,6 @@ int main() { // CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], // CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 2) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: ret void #endif diff --git a/test/OpenMP/distribute_parallel_for_simd_proc_bind_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_proc_bind_messages.cpp index 6b64cc3879ea4..d81f7d9920717 100644 --- a/test/OpenMP/distribute_parallel_for_simd_proc_bind_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_proc_bind_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); template <class T, typename S, int N> diff --git a/test/OpenMP/distribute_parallel_for_simd_reduction_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_reduction_messages.cpp index 94e4541b901bf..348e664b782ae 100644 --- a/test/OpenMP/distribute_parallel_for_simd_reduction_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_simd_safelen_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_safelen_messages.cpp index a5fd1aeb8436a..93766f9a6a35e 100644 --- a/test/OpenMP/distribute_parallel_for_simd_safelen_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_safelen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_simd_schedule_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_schedule_messages.cpp index b3003dd9f2a1b..8a5ffd159425a 100644 --- a/test/OpenMP/distribute_parallel_for_simd_schedule_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/distribute_parallel_for_simd_shared_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_shared_messages.cpp index 134b852579333..fd694b7ba4880 100644 --- a/test/OpenMP/distribute_parallel_for_simd_shared_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_shared_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + struct S1; // expected-note 2 {{declared here}} extern S1 a; diff --git a/test/OpenMP/distribute_parallel_for_simd_simdlen_messages.cpp b/test/OpenMP/distribute_parallel_for_simd_simdlen_messages.cpp index 2d813ec232adc..8e40e3547f748 100644 --- a/test/OpenMP/distribute_parallel_for_simd_simdlen_messages.cpp +++ b/test/OpenMP/distribute_parallel_for_simd_simdlen_messages.cpp @@ -2,6 +2,9 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s + void foo() { } diff --git a/test/OpenMP/distribute_private_codegen.cpp b/test/OpenMP/distribute_private_codegen.cpp index c637fb4892419..c470e8f012227 100644 --- a/test/OpenMP/distribute_private_codegen.cpp +++ b/test/OpenMP/distribute_private_codegen.cpp @@ -5,12 +5,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/distribute_private_messages.cpp b/test/OpenMP/distribute_private_messages.cpp index 0aa1e4796975d..3cf2fdc228b54 100644 --- a/test/OpenMP/distribute_private_messages.cpp +++ b/test/OpenMP/distribute_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/distribute_simd_aligned_messages.cpp b/test/OpenMP/distribute_simd_aligned_messages.cpp index 21b1bc9da1fa0..51421d6cbdd22 100644 --- a/test/OpenMP/distribute_simd_aligned_messages.cpp +++ b/test/OpenMP/distribute_simd_aligned_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp %s +// RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp-simd %s + struct B { static int ib[20]; // expected-note 0 {{'B::ib' declared here}} static constexpr int bfoo() { return 8; } diff --git a/test/OpenMP/distribute_simd_ast_print.cpp b/test/OpenMP/distribute_simd_ast_print.cpp index 3246ccb46ae31..5133db695b3b6 100644 --- a/test/OpenMP/distribute_simd_ast_print.cpp +++ b/test/OpenMP/distribute_simd_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -37,7 +41,7 @@ public: } }; -// CHECK: #pragma omp distribute simd private(this->a) private(this->a) private(T::a) +// CHECK: #pragma omp distribute simd private(this->a) private(this->a) private(T::a){{$}} // CHECK: #pragma omp distribute simd private(this->a) private(this->a) // CHECK: #pragma omp distribute simd private(this->a) private(this->a) private(this->S::a) diff --git a/test/OpenMP/distribute_simd_codegen.cpp b/test/OpenMP/distribute_simd_codegen.cpp index d8f18c3e431b2..662d66af1107c 100644 --- a/test/OpenMP/distribute_simd_codegen.cpp +++ b/test/OpenMP/distribute_simd_codegen.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix HCHECK +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target codegen - host bc file has to be created first. (no significant differences with host version of target region) // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s @@ -16,14 +24,24 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER -// CHECK-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CHECK-DAG: [[DEF_LOC_0:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } -// CHECK-DAG: [[DEF_LOC_DISTRIBUTE_0:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2050, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DAG: [[DEF_LOC_0:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DAG: [[DEF_LOC_DISTRIBUTE_0:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2050, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CHECK-LABEL: define {{.*void}} @{{.*}}without_schedule_clause{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}}) void without_schedule_clause(float *a, float *b, float *c, float *d) { @@ -50,7 +68,7 @@ void without_schedule_clause(float *a, float *b, float *c, float *d) { // CHECK-DAG: store i32 0, i32* [[LAST]] // CHECK-DAG: [[GBL_TID:%.+]] = load i32*, i32** [[TID_ADDR]] // CHECK-DAG: [[GBL_TIDV:%.+]] = load i32, i32* [[GBL_TID]] -// CHECK: call void @__kmpc_for_static_init_{{.+}}(%ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]], i32 92, i32* %.omp.is_last, i32* %.omp.lb, i32* %.omp.ub, i32* %.omp.stride, i32 1, i32 1) +// CHECK: call void @__kmpc_for_static_init_{{.+}}(%struct.ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]], i32 92, i32* %.omp.is_last, i32* %.omp.lb, i32* %.omp.ub, i32* %.omp.stride, i32 1, i32 1) // CHECK-DAG: [[UBV0:%.+]] = load i32, i32* [[UB]] // CHECK-DAG: [[USWITCH:%.+]] = icmp sgt i32 [[UBV0]], 4571423 // CHECK: br i1 [[USWITCH]], label %[[BBCT:.+]], label %[[BBCF:.+]] @@ -84,7 +102,7 @@ void without_schedule_clause(float *a, float *b, float *c, float *d) { // CHECK: [[BBINNEND]]: // CHECK: br label %[[LPEXIT:.+]] // CHECK: [[LPEXIT]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]]) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]]) // CHECK: ret void @@ -112,7 +130,7 @@ void static_not_chunked(float *a, float *b, float *c, float *d) { // CHECK-DAG: store i32 0, i32* [[LAST]] // CHECK-DAG: [[GBL_TID:%.+]] = load i32*, i32** [[TID_ADDR]] // CHECK-DAG: [[GBL_TIDV:%.+]] = load i32, i32* [[GBL_TID]] -// CHECK: call void @__kmpc_for_static_init_{{.+}}(%ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]], i32 92, i32* %.omp.is_last, i32* %.omp.lb, i32* %.omp.ub, i32* %.omp.stride, i32 1, i32 1) +// CHECK: call void @__kmpc_for_static_init_{{.+}}(%struct.ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]], i32 92, i32* %.omp.is_last, i32* %.omp.lb, i32* %.omp.ub, i32* %.omp.stride, i32 1, i32 1) // CHECK-DAG: [[UBV0:%.+]] = load i32, i32* [[UB]] // CHECK-DAG: [[USWITCH:%.+]] = icmp sgt i32 [[UBV0]], 4571423 // CHECK: br i1 [[USWITCH]], label %[[BBCT:.+]], label %[[BBCF:.+]] @@ -146,7 +164,7 @@ void static_not_chunked(float *a, float *b, float *c, float *d) { // CHECK: [[BBINNEND]]: // CHECK: br label %[[LPEXIT:.+]] // CHECK: [[LPEXIT]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]]) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]]) // CHECK: ret void @@ -174,7 +192,7 @@ void static_chunked(float *a, float *b, float *c, float *d) { // CHECK-DAG: store i32 0, i32* [[LAST]] // CHECK-DAG: [[GBL_TID:%.+]] = load i32*, i32** [[TID_ADDR]] // CHECK-DAG: [[GBL_TIDV:%.+]] = load i32, i32* [[GBL_TID]] -// CHECK: call void @__kmpc_for_static_init_{{.+}}(%ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]], i32 91, i32* %.omp.is_last, i32* %.omp.lb, i32* %.omp.ub, i32* %.omp.stride, i32 1, i32 5) +// CHECK: call void @__kmpc_for_static_init_{{.+}}(%struct.ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]], i32 91, i32* %.omp.is_last, i32* %.omp.lb, i32* %.omp.ub, i32* %.omp.stride, i32 1, i32 5) // CHECK-DAG: [[UBV0:%.+]] = load i32, i32* [[UB]] // CHECK-DAG: [[USWITCH:%.+]] = icmp ugt i32 [[UBV0]], 16908288 // CHECK: br i1 [[USWITCH]], label %[[BBCT:.+]], label %[[BBCF:.+]] @@ -208,7 +226,7 @@ void static_chunked(float *a, float *b, float *c, float *d) { // CHECK: [[BBINNEND]]: // CHECK: br label %[[LPEXIT:.+]] // CHECK: [[LPEXIT]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]]) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* [[DEF_LOC_DISTRIBUTE_0]], i32 [[GBL_TIDV]]) // CHECK: ret void // CHECK-LABEL: test_precond diff --git a/test/OpenMP/distribute_simd_collapse_messages.cpp b/test/OpenMP/distribute_simd_collapse_messages.cpp index 182a09a92b6c8..f4d5b73634c09 100644 --- a/test/OpenMP/distribute_simd_collapse_messages.cpp +++ b/test/OpenMP/distribute_simd_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/distribute_simd_dist_schedule_messages.cpp b/test/OpenMP/distribute_simd_dist_schedule_messages.cpp index 6a8482d8be09b..c659abe71ba0f 100644 --- a/test/OpenMP/distribute_simd_dist_schedule_messages.cpp +++ b/test/OpenMP/distribute_simd_dist_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/distribute_simd_firstprivate_codegen.cpp b/test/OpenMP/distribute_simd_firstprivate_codegen.cpp index 5da9c5dc3b1a6..919a12284131b 100644 --- a/test/OpenMP/distribute_simd_firstprivate_codegen.cpp +++ b/test/OpenMP/distribute_simd_firstprivate_codegen.cpp @@ -5,12 +5,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -241,7 +257,7 @@ int main() { // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], // CHECK-DAG: [[VEC_ADDR_VAL_BCAST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i{{[0-9]+}}* // CHECK-DAG: [[VEC_PRIV_BCAST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i{{[0-9]+}}* -// CHECK-DAG: call void @llvm.memcpy.{{.*}}(i{{[0-9]+}}* [[VEC_PRIV_BCAST]], i{{[0-9]+}}* [[VEC_ADDR_VAL_BCAST]],{{.+}}) +// CHECK-DAG: call void @llvm.memcpy.{{.*}}(i{{[0-9]+}}* align {{[0-9]+}} [[VEC_PRIV_BCAST]], i{{[0-9]+}}* align {{[0-9]+}} [[VEC_ADDR_VAL_BCAST]],{{.+}}) // init s_arr // CHECK-DAG: [[S_ARR_ADDR_VAL:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[S_ARR_ADDR]], @@ -256,7 +272,7 @@ int main() { // CHECK-DAG: [[S_ARR_DST_PAST:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]],{{.+}} ], [ [[S_ARR_DST:%.+]],{{.+}} ] // CHECK-DAG: [[S_ARR_SRC_BCAST:%.+]] = bitcast{{.+}} [[S_ARR_SRC_PAST]] to{{.+}} // CHECK-DAG: [[S_ARR_DST_BCAST:%.+]] = bitcast{{.+}} [[S_ARR_DST_PAST]] to{{.+}} -// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* [[S_ARR_DST_BCAST]], {{.+}}* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[S_ARR_DST_BCAST]], {{.+}}* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK-DAG: [[S_ARR_SRC]] = getelementptr{{.+}} // CHECK-DAG: [[S_ARR_DST]] = getelementptr{{.+}} // CHECK-DAG: [[S_ARR_CPY_FIN:%.+]] = icmp{{.+}} [[S_ARR_DST]], [[S_ARR_PRIV_NEXT]] @@ -269,7 +285,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP]], // CHECK-DAG: [[VAR_PRIV_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_PRIV]] to{{.+}} // CHECK-DAG: [[TMP_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL]] to{{.+}} -// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* [[VAR_PRIV_BCAST]], {{.+}}* [[TMP_BCAST]],{{.+}}) +// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[VAR_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[TMP_BCAST]],{{.+}}) // CHECK-DAG: store [[S_FLOAT_TY]]* [[VAR_PRIV]], [[S_FLOAT_TY]]** [[TMP_PRIV]], // init svar @@ -336,7 +352,7 @@ int main() { // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], // CHECK-DAG: [[VEC_ADDR_VAL_BCAST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i{{[0-9]+}}* // CHECK-DAG: [[VEC_PRIV_BCAST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i{{[0-9]+}}* -// CHECK-DAG: call void @llvm.memcpy.{{.*}}(i{{[0-9]+}}* [[VEC_PRIV_BCAST]], i{{[0-9]+}}* [[VEC_ADDR_VAL_BCAST]],{{.+}}) +// CHECK-DAG: call void @llvm.memcpy.{{.*}}(i{{[0-9]+}}* align {{[0-9]+}} [[VEC_PRIV_BCAST]], i{{[0-9]+}}* align {{[0-9]+}} [[VEC_ADDR_VAL_BCAST]],{{.+}}) // init s_arr // CHECK-DAG: [[S_ARR_ADDR_VAL:%.+]] = load [2 x [[S_INT_TY]]]*, [2 x [[S_INT_TY]]]** [[S_ARR_ADDR]], @@ -351,7 +367,7 @@ int main() { // CHECK-DAG: [[S_ARR_DST_PAST:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]],{{.+}} ], [ [[S_ARR_DST:%.+]],{{.+}} ] // CHECK-DAG: [[S_ARR_SRC_BCAST:%.+]] = bitcast{{.+}} [[S_ARR_SRC_PAST]] to{{.+}} // CHECK-DAG: [[S_ARR_DST_BCAST:%.+]] = bitcast{{.+}} [[S_ARR_DST_PAST]] to{{.+}} -// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* [[S_ARR_DST_BCAST]], {{.+}}* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[S_ARR_DST_BCAST]], {{.+}}* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK-DAG: [[S_ARR_SRC]] = getelementptr{{.+}} // CHECK-DAG: [[S_ARR_DST]] = getelementptr{{.+}} // CHECK-DAG: [[S_ARR_CPY_FIN:%.+]] = icmp{{.+}} [[S_ARR_DST]], [[S_ARR_PRIV_NEXT]] @@ -364,7 +380,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP]], // CHECK-DAG: [[VAR_PRIV_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_PRIV]] to{{.+}} // CHECK-DAG: [[TMP_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL]] to{{.+}} -// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* [[VAR_PRIV_BCAST]], {{.+}}* [[TMP_BCAST]],{{.+}}) +// CHECK-DAG: call{{.+}} @llvm.memcpy.{{.+}}({{.+}}* align {{[0-9]+}} [[VAR_PRIV_BCAST]], {{.+}}* align {{[0-9]+}} [[TMP_BCAST]],{{.+}}) // CHECK-DAG: store [[S_INT_TY]]* [[VAR_PRIV]], [[S_INT_TY]]** [[TMP_PRIV]], // CHECK-DAG: store i{{[0-9]+}} 0, i{{[0-9]+}}* %.omp{{.+}}, diff --git a/test/OpenMP/distribute_simd_firstprivate_messages.cpp b/test/OpenMP/distribute_simd_firstprivate_messages.cpp index f0c293b78d800..e0f3ddeeeba1d 100644 --- a/test/OpenMP/distribute_simd_firstprivate_messages.cpp +++ b/test/OpenMP/distribute_simd_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/distribute_simd_lastprivate_codegen.cpp b/test/OpenMP/distribute_simd_lastprivate_codegen.cpp index df842a0098797..cde7c81d50f93 100644 --- a/test/OpenMP/distribute_simd_lastprivate_codegen.cpp +++ b/test/OpenMP/distribute_simd_lastprivate_codegen.cpp @@ -5,12 +5,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -251,7 +267,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK-DAG: [[S_ARR_PTR_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_PTR]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST]], i8* [[TMP_VAL_BCAST]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST]], // CHECK: call void @__kmpc_for_static_fini( // lastprivates @@ -264,7 +280,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -275,7 +291,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_FLOAT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -284,7 +300,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: [[SVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[S_VAR_PRIV]], // CHECK: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR_REF]], // CHECK: ret void @@ -353,7 +369,7 @@ int main() { // CHECK-DAG: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK-DAG: [[S_ARR_PTR_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_PTR1]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST1]], i8* [[TMP_VAL_BCAST1]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST1]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST1]], // CHECK: call void @__kmpc_for_static_fini( // lastprivates @@ -366,7 +382,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF1]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF1]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_PRIV1]] to [[S_INT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -377,7 +393,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_INT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -386,7 +402,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: ret void // CHECK: !{!"llvm.loop.vectorize.enable", i1 true} diff --git a/test/OpenMP/distribute_simd_lastprivate_messages.cpp b/test/OpenMP/distribute_simd_lastprivate_messages.cpp index 82ce15e550488..e47ad7ec1eef6 100644 --- a/test/OpenMP/distribute_simd_lastprivate_messages.cpp +++ b/test/OpenMP/distribute_simd_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/distribute_simd_linear_messages.cpp b/test/OpenMP/distribute_simd_linear_messages.cpp index e4d71fee82508..06d85d13501e8 100644 --- a/test/OpenMP/distribute_simd_linear_messages.cpp +++ b/test/OpenMP/distribute_simd_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/distribute_simd_loop_messages.cpp b/test/OpenMP/distribute_simd_loop_messages.cpp index 59fafda45f4be..1977ca789dcf9 100644 --- a/test/OpenMP/distribute_simd_loop_messages.cpp +++ b/test/OpenMP/distribute_simd_loop_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s + static int sii; // expected-note@+1 {{defined as threadprivate or thread local}} #pragma omp threadprivate(sii) @@ -322,9 +324,7 @@ int test_iteration_spaces() { #pragma omp target #pragma omp teams - // expected-error@+3 {{unexpected OpenMP clause 'shared' in directive '#pragma omp distribute simd'}} - // expected-note@+2 {{defined as shared}} - // expected-error@+2 {{loop iteration variable in the associated loop of 'omp distribute simd' directive may not be shared, predetermined as linear}} + // expected-error@+1 {{unexpected OpenMP clause 'shared' in directive '#pragma omp distribute simd'}} #pragma omp distribute simd shared(ii) for (ii = 0; ii < 10; ii++) c[ii] = a[ii]; diff --git a/test/OpenMP/distribute_simd_misc_messages.c b/test/OpenMP/distribute_simd_misc_messages.c index f643ed1d5b8e3..2cb5e97273ca4 100644 --- a/test/OpenMP/distribute_simd_misc_messages.c +++ b/test/OpenMP/distribute_simd_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp distribute simd'}} #pragma omp distribute simd diff --git a/test/OpenMP/distribute_simd_private_codegen.cpp b/test/OpenMP/distribute_simd_private_codegen.cpp index ede807a0fb63b..f675158be694f 100644 --- a/test/OpenMP/distribute_simd_private_codegen.cpp +++ b/test/OpenMP/distribute_simd_private_codegen.cpp @@ -5,12 +5,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/distribute_simd_private_messages.cpp b/test/OpenMP/distribute_simd_private_messages.cpp index c777c9971d4b5..8d82cc816649b 100644 --- a/test/OpenMP/distribute_simd_private_messages.cpp +++ b/test/OpenMP/distribute_simd_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/distribute_simd_reduction_codegen.cpp b/test/OpenMP/distribute_simd_reduction_codegen.cpp index 8485d8d27d523..85b0e80aade3a 100644 --- a/test/OpenMP/distribute_simd_reduction_codegen.cpp +++ b/test/OpenMP/distribute_simd_reduction_codegen.cpp @@ -5,10 +5,23 @@ // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/distribute_simd_reduction_messages.cpp b/test/OpenMP/distribute_simd_reduction_messages.cpp index ddedcb0892baf..7bb5bdea5b46c 100644 --- a/test/OpenMP/distribute_simd_reduction_messages.cpp +++ b/test/OpenMP/distribute_simd_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/distribute_simd_safelen_messages.cpp b/test/OpenMP/distribute_simd_safelen_messages.cpp index 4ae35fb2334da..c95d121ca2c9e 100644 --- a/test/OpenMP/distribute_simd_safelen_messages.cpp +++ b/test/OpenMP/distribute_simd_safelen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/distribute_simd_simdlen_messages.cpp b/test/OpenMP/distribute_simd_simdlen_messages.cpp index 4ae35fb2334da..c95d121ca2c9e 100644 --- a/test/OpenMP/distribute_simd_simdlen_messages.cpp +++ b/test/OpenMP/distribute_simd_simdlen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/driver.c b/test/OpenMP/driver.c index 74aaea5071966..4c8b1df538ea1 100644 --- a/test/OpenMP/driver.c +++ b/test/OpenMP/driver.c @@ -1,10 +1,13 @@ +// REQUIRES: x86-registered-target // Test that by default -fnoopenmp-use-tls is passed to frontend. // // RUN: %clang %s -### -o %t.o 2>&1 -fopenmp=libomp | FileCheck --check-prefix=CHECK-DEFAULT %s + // CHECK-DEFAULT: -cc1 // CHECK-DEFAULT-NOT: -fnoopenmp-use-tls // // RUN: %clang %s -### -o %t.o 2>&1 -fopenmp=libomp -fnoopenmp-use-tls | FileCheck --check-prefix=CHECK-NO-TLS %s + // CHECK-NO-TLS: -cc1 // CHECK-NO-TLS-SAME: -fnoopenmp-use-tls // @@ -13,17 +16,31 @@ // RUN: %clang %s -c -E -dM -fopenmp=libomp -fopenmp-version=0 | FileCheck --check-prefix=CHECK-DEFAULT-VERSION %s // RUN: %clang %s -c -E -dM -fopenmp=libomp -fopenmp-version=100 | FileCheck --check-prefix=CHECK-DEFAULT-VERSION %s // RUN: %clang %s -c -E -dM -fopenmp=libomp -fopenmp-version=31 | FileCheck --check-prefix=CHECK-DEFAULT-VERSION %s + // CHECK-DEFAULT-VERSION: #define _OPENMP 201107 // RUN: %clang %s -c -E -dM -fopenmp=libomp -fopenmp-version=40 | FileCheck --check-prefix=CHECK-40-VERSION %s // CHECK-40-VERSION: #define _OPENMP 201307 // RUN: %clang %s -c -E -dM -fopenmp=libomp -fopenmp-version=45 | FileCheck --check-prefix=CHECK-45-VERSION %s +// RUN: %clang %s -c -E -dM -fopenmp=libomp -fopenmp-simd | FileCheck --check-prefix=CHECK-45-VERSION %s +// RUN: %clang %s -c -E -dM -fopenmp=libomp -fopenmp-targets=x86_64-unknown-unknown -o - | FileCheck --check-prefix=CHECK-45-VERSION --check-prefix=CHECK-45-VERSION2 %s // CHECK-45-VERSION: #define _OPENMP 201511 +// CHECK-45-VERSION2: #define _OPENMP 201511 // RUN: %clang %s -c -E -dM -fopenmp-version=1 | FileCheck --check-prefix=CHECK-VERSION %s // RUN: %clang %s -c -E -dM -fopenmp-version=31 | FileCheck --check-prefix=CHECK-VERSION %s // RUN: %clang %s -c -E -dM -fopenmp-version=40 | FileCheck --check-prefix=CHECK-VERSION %s // RUN: %clang %s -c -E -dM -fopenmp-version=45 | FileCheck --check-prefix=CHECK-VERSION %s +// RUN: %clang %s -c -E -dM -fopenmp-version=45 | FileCheck --check-prefix=CHECK-VERSION %s +// RUN: %clang %s -c -E -dM -fopenmp-simd -fno-openmp-simd -fopenmp-version=45 | FileCheck --check-prefix=CHECK-VERSION %s +// RUN: %clang %s -c -E -dM -fopenmp-simd | FileCheck --check-prefix=CHECK-VERSION %s +// RUN: %clang %s -c -E -dM -fopenmp-simd -fopenmp-version=1 | FileCheck --check-prefix=CHECK-VERSION %s +// RUN: %clang %s -c -E -dM -fopenmp-simd -fopenmp-version=0 | FileCheck --check-prefix=CHECK-VERSION %s +// RUN: %clang %s -c -E -dM -fopenmp-simd -fopenmp-version=100 | FileCheck --check-prefix=CHECK-VERSION %s +// RUN: %clang %s -c -E -dM -fopenmp-simd -fopenmp-version=31 | FileCheck --check-prefix=CHECK-VERSION %s +// RUN: %clang %s -c -E -dM -fopenmp-simd -fopenmp-version=40 | FileCheck --check-prefix=CHECK-VERSION %s +// RUN: %clang %s -c -E -dM -fopenmp-simd -fopenmp-version=45 | FileCheck --check-prefix=CHECK-VERSION %s + // CHECK-VERSION-NOT: #define _OPENMP diff --git a/test/OpenMP/dump.cpp b/test/OpenMP/dump.cpp index 222c83d5ce082..812afb72f6801 100644 --- a/test/OpenMP/dump.cpp +++ b/test/OpenMP/dump.cpp @@ -1,4 +1,5 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-dump %s | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-dump %s | FileCheck %s // expected-no-diagnostics int ga, gb; @@ -12,7 +13,7 @@ int ga, gb; #pragma omp declare reduction(fun : float : omp_out += omp_in) initializer(omp_priv = omp_orig + 15) -// CHECK: |-OMPDeclareReductionDecl {{.+}} <line:11:35> col:35 operator+ 'int' combiner +// CHECK: |-OMPDeclareReductionDecl {{.+}} <line:12:35> col:35 operator+ 'int' combiner // CHECK-NEXT: | |-CompoundAssignOperator {{.+}} <col:47, col:58> 'int' lvalue '*=' ComputeLHSTy='int' ComputeResultTy='int' // CHECK-NEXT: | | |-DeclRefExpr {{.+}} <col:47> 'int' lvalue Var {{.+}} 'omp_out' 'int' // CHECK-NEXT: | | `-ImplicitCastExpr {{.+}} <col:58> 'int' <LValueToRValue> @@ -27,7 +28,7 @@ int ga, gb; // CHECK-NEXT: | | `-DeclRefExpr {{.+}} <col:58> 'char' lvalue Var {{.+}} 'omp_in' 'char' // CHECK-NEXT: | |-VarDecl {{.+}} <col:40> col:40 implicit used omp_in 'char' // CHECK-NEXT: | `-VarDecl {{.+}} <col:40> col:40 implicit used omp_out 'char' -// CHECK-NEXT: |-OMPDeclareReductionDecl {{.+}} <line:13:37> col:37 fun 'float' combiner initializer +// CHECK-NEXT: |-OMPDeclareReductionDecl {{.+}} <line:14:37> col:37 fun 'float' combiner initializer // CHECK-NEXT: | |-CompoundAssignOperator {{.+}} <col:45, col:56> 'float' lvalue '+=' ComputeLHSTy='float' ComputeResultTy='float' // CHECK-NEXT: | | |-DeclRefExpr {{.+}} <col:45> 'float' lvalue Var {{.+}} 'omp_out' 'float' // CHECK-NEXT: | | `-ImplicitCastExpr {{.+}} <col:56> 'float' <LValueToRValue> @@ -42,27 +43,42 @@ struct S { } }; -// CHECK: | `-OMPParallelForDirective {{.+}} <line:39:9, col:80> -// CHECK-NEXT: | |-OMPDefaultClause {{.+}} <col:26, col:40> -// CHECK-NEXT: | |-OMPPrivateClause {{.+}} <col:40, col:51> +// CHECK: | `-OMPParallelForDirective {{.+}} {{<line:40:9, col:80>|<col:9, col:80>}} +// CHECK-NEXT: | |-OMPDefaultClause {{.+}} <col:26, col:38> +// CHECK-NEXT: | |-OMPPrivateClause {{.+}} <col:40, col:49> // CHECK-NEXT: | | `-DeclRefExpr {{.+}} <col:48> 'int' lvalue OMPCapturedExpr {{.+}} 'a' 'int &' -// CHECK-NEXT: | |-OMPSharedClause {{.+}} <col:51, col:61> +// CHECK-NEXT: | |-OMPSharedClause {{.+}} <col:51, col:59> // CHECK-NEXT: | | `-MemberExpr {{.+}} <col:58> 'int' lvalue ->b // CHECK-NEXT: | | `-CXXThisExpr {{.+}} <col:58> 'S *' this // CHECK-NEXT: | |-OMPScheduleClause {{.+}} <col:61, col:79> // CHECK-NEXT: | | `-ImplicitCastExpr {{.+}} <col:78> 'int' <LValueToRValue> // CHECK-NEXT: | | `-DeclRefExpr {{.+}} <col:78> 'int' lvalue OMPCapturedExpr {{.+}} '.capture_expr.' 'int' -// CHECK-NEXT: | |-CapturedStmt {{.+}} <line:40:5, line:41:9> -// CHECK-NEXT: | | |-CapturedDecl {{.+}} <<invalid sloc>> <invalid sloc> -// CHECK-NEXT: | | | |-ForStmt {{.+}} <line:40:5, line:41:9> -// CHECK: | | | | `-UnaryOperator {{.+}} <line:41:7, col:9> 'int' lvalue prefix '++' -// CHECK-NEXT: | | | | `-DeclRefExpr {{.+}} <col:9> 'int' lvalue OMPCapturedExpr {{.+}} 'a' 'int &' +// CHECK-NEXT: | `-CapturedStmt {{.+}} <line:41:5, line:42:9> +// CHECK-NEXT: | |-CapturedDecl {{.+}} <<invalid sloc>> <invalid sloc> +// CHECK-NEXT: | | |-ForStmt {{.+}} <line:41:5, line:42:9> +// CHECK: | | | `-UnaryOperator {{.+}} <line:42:7, col:9> 'int' lvalue prefix '++' +// CHECK-NEXT: | | | `-DeclRefExpr {{.+}} <col:9> 'int' lvalue OMPCapturedExpr {{.+}} 'a' 'int &' #pragma omp declare simd #pragma omp declare simd inbranch void foo(); -// CHECK: `-FunctionDecl {{.+}} <line:63:1, col:10> col:6 foo 'void ()' -// CHECK-NEXT: |-OMPDeclareSimdDeclAttr {{.+}} <line:62:9, col:34> Implicit BS_Inbranch -// CHECK: `-OMPDeclareSimdDeclAttr {{.+}} <line:61:9, col:25> Implicit BS_Undefined +// CHECK: |-FunctionDecl {{.+}} <line:64:1, col:10> col:6 foo 'void ()' +// CHECK-NEXT: |-OMPDeclareSimdDeclAttr {{.+}} <line:63:9, col:34> Implicit BS_Inbranch +// CHECK: `-OMPDeclareSimdDeclAttr {{.+}} <line:62:9, col:25> Implicit BS_Undefined +#pragma omp declare target +int bar() { + int f; + return f; +} +#pragma omp end declare target + +// CHECK: `-FunctionDecl {{.+}} <line:71:1, line:74:1> line:71:5 bar 'int ()' +// CHECK-NEXT: |-CompoundStmt {{.+}} <col:11, line:74:1> +// CHECK-NEXT: | |-DeclStmt {{.+}} <line:72:3, col:8> +// CHECK-NEXT: | | `-VarDecl {{.+}} <col:3, col:7> col:7 used f 'int' +// CHECK-NEXT: | `-ReturnStmt {{.+}} <line:73:3, col:10> +// CHECK-NEXT: | `-ImplicitCastExpr {{.+}} <col:10> 'int' <LValueToRValue> +// CHECK-NEXT: | `-DeclRefExpr {{.+}} <col:10> 'int' lvalue Var {{.+}} 'f' 'int' +// CHECK-NEXT: `-OMPDeclareTargetDeclAttr {{.+}} <<invalid sloc>> Implicit MT_To diff --git a/test/OpenMP/flush_ast_print.cpp b/test/OpenMP/flush_ast_print.cpp index f3af47bce3214..8cde315a453cf 100644 --- a/test/OpenMP/flush_ast_print.cpp +++ b/test/OpenMP/flush_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -16,7 +20,7 @@ T tmain(T argc) { return a + argc; } // CHECK: static T a; -// CHECK-NEXT: #pragma omp flush +// CHECK-NEXT: #pragma omp flush{{$}} // CHECK-NEXT: #pragma omp flush (a) // CHECK: static int a; // CHECK-NEXT: #pragma omp flush diff --git a/test/OpenMP/flush_codegen.cpp b/test/OpenMP/flush_codegen.cpp index 89636a5bd87aa..543524c220bed 100644 --- a/test/OpenMP/flush_codegen.cpp +++ b/test/OpenMP/flush_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -30,4 +35,6 @@ int main() { // CHECK: call {{.*}}void @__kmpc_flush(%{{.+}}* {{(@|%).+}}) // CHECK: ret +// CHECK-NOT: line: 0, + #endif diff --git a/test/OpenMP/flush_messages.cpp b/test/OpenMP/flush_messages.cpp index 1c086a3f3fa4a..da1d0c2412427 100644 --- a/test/OpenMP/flush_messages.cpp +++ b/test/OpenMP/flush_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + struct S1 { // expected-note 2 {{declared here}} int a; }; diff --git a/test/OpenMP/for_ast_print.cpp b/test/OpenMP/for_ast_print.cpp index f489b217de964..82318ff800aa4 100644 --- a/test/OpenMP/for_ast_print.cpp +++ b/test/OpenMP/for_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -165,7 +169,7 @@ int main(int argc, char **argv) { for (q = &buf[0]; q <= buf + 7; q++) foo(); // CHECK: #pragma omp parallel - // CHECK-NEXT: #pragma omp for + // CHECK-NEXT: #pragma omp for{{$}} // CHECK-NEXT: for (p = buf; p < &buf[8]; p++) // CHECK-NEXT: for (q = &buf[0]; q <= buf + 7; q++) // CHECK-NEXT: foo(); diff --git a/test/OpenMP/for_codegen.cpp b/test/OpenMP/for_codegen.cpp index 1f5dd7ddd0dd7..ad8674fdde7ee 100644 --- a/test/OpenMP/for_codegen.cpp +++ b/test/OpenMP/for_codegen.cpp @@ -1,8 +1,15 @@ -// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fsanitize-address-use-after-scope | FileCheck -allow-deprecated-dag-overlap %s --check-prefix=CHECK --check-prefix=LIFETIME // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG -// RUN: %clang_cc1 -main-file-name for_codegen.cpp %s -o - -emit-llvm -fprofile-instrument=clang -fprofile-instrument-path=for_codegen-test.profraw | FileCheck %s --check-prefix=PROF-INSTR-PATH +// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix=TERM_DEBUG +// RUN: %clang_cc1 -main-file-name for_codegen.cpp %s -o - -emit-llvm -fprofile-instrument=clang -fprofile-instrument-path=for_codegen-test.profraw | FileCheck -allow-deprecated-dag-overlap %s --check-prefix=PROF-INSTR-PATH + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -main-file-name for_codegen.cpp %s -o - -emit-llvm -fprofile-instrument=clang -fprofile-instrument-path=for_codegen-test.profraw | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // // expected-no-diagnostics #ifndef HEADER @@ -18,8 +25,20 @@ // CHECK-LABEL: loop_with_counter_collapse void loop_with_counter_collapse() { - // CHECK: call void @__kmpc_for_static_init_8(%ident_t* @ - // CHECK: call void @__kmpc_for_static_fini(%ident_t* @ + // LIFETIME: call void @llvm.lifetime.end + // LIFETIME: call void @llvm.lifetime.end + // CHECK: call void @__kmpc_for_static_init_8(%struct.ident_t* @ + // CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* @ + // LIFETIME: call void @llvm.lifetime.end + // LIFETIME: call void @llvm.lifetime.end + // LIFETIME: call void @llvm.lifetime.end + // LIFETIME: call void @llvm.lifetime.end + // LIFETIME: call void @llvm.lifetime.end + // LIFETIME: call void @llvm.lifetime.end + // LIFETIME: call void @llvm.lifetime.end + // LIFETIME: call void @llvm.lifetime.end + // LIFETIME: call void @llvm.lifetime.end + // LIFETIME: call void @llvm.lifetime.end #pragma omp for collapse(2) for (int i = 0; i < 4; i++) { for (int j = i; j < 4; j++) { diff --git a/test/OpenMP/for_collapse_messages.cpp b/test/OpenMP/for_collapse_messages.cpp index a6fdf0066592c..230880b22db3c 100644 --- a/test/OpenMP/for_collapse_messages.cpp +++ b/test/OpenMP/for_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/for_firstprivate_codegen.cpp b/test/OpenMP/for_firstprivate_codegen.cpp index 88f54103ad69c..fc3caa9a26032 100644 --- a/test/OpenMP/for_firstprivate_codegen.cpp +++ b/test/OpenMP/for_firstprivate_codegen.cpp @@ -3,6 +3,13 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -234,7 +241,7 @@ int main() { // firstprivate vec(vec) // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* bitcast ([2 x i{{[0-9]+}}]* [[VEC]] to i8*), +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST]], i8* align {{[0-9]+}} bitcast ([2 x i{{[0-9]+}}]* [[VEC]] to i8*), // firstprivate s_arr(s_arr) // CHECK: [[S_ARR_PRIV_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 @@ -302,7 +309,7 @@ int main() { // firstprivate vec(vec) // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_REF]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* [[VEC_SRC]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST]], i8* align {{[0-9]+}} [[VEC_SRC]], // firstprivate s_arr(s_arr) // CHECK: [[S_ARR_PRIV_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 diff --git a/test/OpenMP/for_firstprivate_messages.cpp b/test/OpenMP/for_firstprivate_messages.cpp index 60be4f5f2a2b1..c34c918d120f9 100644 --- a/test/OpenMP/for_firstprivate_messages.cpp +++ b/test/OpenMP/for_firstprivate_messages.cpp @@ -1,4 +1,5 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s void foo() { } diff --git a/test/OpenMP/for_lastprivate_codegen.cpp b/test/OpenMP/for_lastprivate_codegen.cpp index 30a17eaf581a3..0061de632380f 100644 --- a/test/OpenMP/for_lastprivate_codegen.cpp +++ b/test/OpenMP/for_lastprivate_codegen.cpp @@ -3,6 +3,13 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -516,7 +523,7 @@ int main() { // original vec[]=private_vec[]; // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_REF]] to i8* // CHECK: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* [[VEC_SRC]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST]], i8* align {{[0-9]+}} [[VEC_SRC]], // original s_arr[]=private_s_arr[]; // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 @@ -745,7 +752,7 @@ int main() { // original vec[]=private_vec[]; // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_REF]] to i8* // CHECK: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* [[VEC_SRC]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST]], i8* align {{[0-9]+}} [[VEC_SRC]], // original s_arr[]=private_s_arr[]; // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 diff --git a/test/OpenMP/for_lastprivate_messages.cpp b/test/OpenMP/for_lastprivate_messages.cpp index 13ef35e925a74..7787ab12d481b 100644 --- a/test/OpenMP/for_lastprivate_messages.cpp +++ b/test/OpenMP/for_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/for_linear_codegen.cpp b/test/OpenMP/for_linear_codegen.cpp index 54cdf8a4cb33b..34e8c6a09aff6 100644 --- a/test/OpenMP/for_linear_codegen.cpp +++ b/test/OpenMP/for_linear_codegen.cpp @@ -3,6 +3,13 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/for_linear_messages.cpp b/test/OpenMP/for_linear_messages.cpp index ab8934979ac30..622cd4a8a3af6 100644 --- a/test/OpenMP/for_linear_messages.cpp +++ b/test/OpenMP/for_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/for_loop_messages.cpp b/test/OpenMP/for_loop_messages.cpp index 01cc6020b85fc..37ff0fc0ca2dd 100644 --- a/test/OpenMP/for_loop_messages.cpp +++ b/test/OpenMP/for_loop_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s + class S { int a; S() : a(0) {} diff --git a/test/OpenMP/for_misc_messages.c b/test/OpenMP/for_misc_messages.c index 0a7cdf8fc06b8..901d66dd688cf 100644 --- a/test/OpenMP/for_misc_messages.c +++ b/test/OpenMP/for_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -triple x86_64-unknown-unknown -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -triple x86_64-unknown-unknown -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp for'}} #pragma omp for @@ -52,6 +54,20 @@ void test_invalid_clause() { #pragma omp for foo bar for (i = 0; i < 16; ++i) ; +// At one time, this failed an assert. +// expected-error@+1 {{unexpected OpenMP clause 'num_teams' in directive '#pragma omp for'}} +#pragma omp for num_teams(3) + for (i = 0; i < 16; ++i) + ; +// At one time, this error was reported twice. +// expected-error@+1 {{unexpected OpenMP clause 'uniform' in directive '#pragma omp for'}} +#pragma omp for uniform + for (i = 0; i < 16; ++i) + ; +// expected-error@+1 {{unexpected OpenMP clause 'if' in directive '#pragma omp for'}} +#pragma omp for if(0) + for (i = 0; i < 16; ++i) + ; } void test_non_identifiers() { diff --git a/test/OpenMP/for_ordered_clause.cpp b/test/OpenMP/for_ordered_clause.cpp index 3335f4046f571..ec29a989d0b08 100644 --- a/test/OpenMP/for_ordered_clause.cpp +++ b/test/OpenMP/for_ordered_clause.cpp @@ -2,6 +2,11 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + void foo() { } diff --git a/test/OpenMP/for_private_codegen.cpp b/test/OpenMP/for_private_codegen.cpp index f1416d75b64ba..d914a62d3baeb 100644 --- a/test/OpenMP/for_private_codegen.cpp +++ b/test/OpenMP/for_private_codegen.cpp @@ -3,6 +3,13 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -40,7 +47,7 @@ T tmain() { int main() { static int svar; #ifdef LAMBDA - // LAMBDA: [[G:@.+]] = global double + // LAMBDA: [[G:@.+]] = {{(dso_local )?}}global double // LAMBDA-LABEL: @main // LAMBDA: call{{.*}} void [[OUTER_LAMBDA:@.+]]( [&]() { @@ -102,7 +109,7 @@ int main() { }(); return 0; #elif defined(BLOCKS) - // BLOCKS: [[G:@.+]] = global double + // BLOCKS: [[G:@.+]] = {{(dso_local )?}}global double // BLOCKS-LABEL: @main // BLOCKS: call {{.*}}void {{%.+}}(i8 ^{ diff --git a/test/OpenMP/for_private_messages.cpp b/test/OpenMP/for_private_messages.cpp index 4045c5b56ff5c..5b0b562869b22 100644 --- a/test/OpenMP/for_private_messages.cpp +++ b/test/OpenMP/for_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/for_reduction_codegen.cpp b/test/OpenMP/for_reduction_codegen.cpp index cef457333062d..55323122f3fc3 100644 --- a/test/OpenMP/for_reduction_codegen.cpp +++ b/test/OpenMP/for_reduction_codegen.cpp @@ -1,8 +1,15 @@ -// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s -// RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap -check-prefix=LAMBDA %s +// RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap -check-prefix=BLOCKS %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -345,7 +352,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // var1 = var1.operator &&(var1_reduction); // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]]) @@ -361,7 +368,7 @@ int main() { // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // t_var1 = min(t_var1, t_var1_reduction); // CHECK: [[T_VAR1_VAL:%.+]] = load float, float* [[T_VAR1_REF]], @@ -399,7 +406,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // CHECK: call void @__kmpc_end_critical( // var1 = var1.operator &&(var1_reduction); @@ -417,7 +424,7 @@ int main() { // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // CHECK: call void @__kmpc_end_critical( // t_var1 = min(t_var1, t_var1_reduction); @@ -503,7 +510,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_LHS]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_RHS]]) // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_LHS]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // var1_lhs = var1_lhs.operator &&(var1_rhs); // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_LHS]]) @@ -519,7 +526,7 @@ int main() { // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_LHS]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load float, float* [[T_VAR1_LHS]], @@ -622,7 +629,7 @@ int main() { // CHECK: phi [[S_FLOAT_TY]]* // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}) // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %{{.+}}, i8* align 4 [[BITCAST]], i64 4, i1 false) // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]] // CHECK: br i1 [[DONE]], @@ -652,7 +659,7 @@ int main() { // CHECK: call void @__kmpc_critical( // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}) // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %{{.+}}, i8* align 4 [[BITCAST]], i64 4, i1 false) // CHECK: call void @__kmpc_end_critical( // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]] // CHECK: br i1 [[DONE]], @@ -725,7 +732,7 @@ int main() { // CHECK: phi [[S_FLOAT_TY]]* // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}) // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %{{.+}}, i8* align 4 [[BITCAST]], i64 4, i1 false) // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]] // CHECK: br i1 [[DONE]], @@ -816,7 +823,7 @@ int main() { // CHECK: phi [[S_FLOAT_TY]]* // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}) // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %{{.+}}, i8* align 4 [[BITCAST]], i64 4, i1 false) // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]] // CHECK: br i1 [[DONE]], @@ -846,7 +853,7 @@ int main() { // CHECK: call void @__kmpc_critical( // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}) // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %{{.+}}, i8* align 4 [[BITCAST]], i64 4, i1 false) // CHECK: call void @__kmpc_end_critical( // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]] // CHECK: br i1 [[DONE]], @@ -915,7 +922,7 @@ int main() { // CHECK: phi [[S_FLOAT_TY]]* // CHECK: [[AND:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @_ZN1SIfEanERKS0_([[S_FLOAT_TY]]* %{{.+}}, [[S_FLOAT_TY]]* dereferenceable(4) %{{.+}}) // CHECK: [[BITCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[AND]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* [[BITCAST]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %{{.+}}, i8* align 4 [[BITCAST]], i64 4, i1 false) // CHECK: [[DONE:%.+]] = icmp eq [[S_FLOAT_TY]]* %{{.+}}, [[END]] // CHECK: br i1 [[DONE]], @@ -1285,7 +1292,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // var1 = var1.operator &&(var1_reduction); // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) @@ -1301,7 +1308,7 @@ int main() { // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // t_var1 = min(t_var1, t_var1_reduction); // CHECK: [[T_VAR1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_REF]], @@ -1327,7 +1334,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // CHECK: call void @__kmpc_end_critical( // var1 = var1.operator &&(var1_reduction); @@ -1345,7 +1352,7 @@ int main() { // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // CHECK: call void @__kmpc_end_critical( // t_var1 = min(t_var1, t_var1_reduction); @@ -1412,7 +1419,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_LHS]], [[S_INT_TY]]* dereferenceable(4) [[VAR_RHS]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_LHS]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // var1_lhs = var1_lhs.operator &&(var1_rhs); // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_LHS]]) @@ -1428,7 +1435,7 @@ int main() { // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_LHS]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_LHS]], diff --git a/test/OpenMP/for_reduction_codegen_UDR.cpp b/test/OpenMP/for_reduction_codegen_UDR.cpp index 05a46a6eb7ffc..0a6ba1d61008d 100644 --- a/test/OpenMP/for_reduction_codegen_UDR.cpp +++ b/test/OpenMP/for_reduction_codegen_UDR.cpp @@ -1,6 +1,11 @@ -// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/for_reduction_messages.cpp b/test/OpenMP/for_reduction_messages.cpp index d4f4a795fe2cc..a4f15aa6193e4 100644 --- a/test/OpenMP/for_reduction_messages.cpp +++ b/test/OpenMP/for_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/for_schedule_messages.cpp b/test/OpenMP/for_schedule_messages.cpp index 8946ce3aedfb5..f4b5b3a26d643 100644 --- a/test/OpenMP/for_schedule_messages.cpp +++ b/test/OpenMP/for_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/for_simd_aligned_messages.cpp b/test/OpenMP/for_simd_aligned_messages.cpp index 3ff25870e5a0d..fce9aca9989a5 100644 --- a/test/OpenMP/for_simd_aligned_messages.cpp +++ b/test/OpenMP/for_simd_aligned_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp %s +// RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp-simd %s + struct B { static int ib[20]; // expected-note 0 {{'B::ib' declared here}} static constexpr int bfoo() { return 8; } diff --git a/test/OpenMP/for_simd_ast_print.cpp b/test/OpenMP/for_simd_ast_print.cpp index befad16e0951e..13e82b46472d5 100644 --- a/test/OpenMP/for_simd_ast_print.cpp +++ b/test/OpenMP/for_simd_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -33,7 +37,7 @@ public: } }; -// CHECK: #pragma omp for simd private(this->a) private(this->a) private(T::a) +// CHECK: #pragma omp for simd private(this->a) private(this->a) private(T::a){{$}} // CHECK: #pragma omp for simd private(this->a) private(this->a) // CHECK: #pragma omp for simd private(this->a) private(this->a) private(this->S1::a) diff --git a/test/OpenMP/for_simd_codegen.cpp b/test/OpenMP/for_simd_codegen.cpp index 89f779db33cd7..7e540e263b144 100644 --- a/test/OpenMP/for_simd_codegen.cpp +++ b/test/OpenMP/for_simd_codegen.cpp @@ -2,6 +2,12 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t < %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify -emit-llvm -o - < %s | FileCheck %s // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm -o - < %s | FileCheck %s --check-prefix=TERM_DEBUG + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm -fexceptions -fcxx-exceptions -o - < %s | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t < %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify -emit-llvm -o - < %s | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm -o - < %s | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -12,7 +18,7 @@ double *g_ptr; // CHECK-LABEL: define {{.*void}} @{{.*}}simple{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}}) void simple(float *a, float *b, float *c, float *d) { #pragma omp for simd -// CHECK: call void @__kmpc_for_static_init_4(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) +// CHECK: call void @__kmpc_for_static_init_4(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) // CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]], // CHECK: [[CMP:%.+]] = icmp sgt i32 [[UB_VAL]], 5 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -48,8 +54,8 @@ void simple(float *a, float *b, float *c, float *d) { // br label %{{.+}}, !llvm.loop !{{.+}} } // CHECK: [[SIMPLE_LOOP1_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) -// CHECK: call void @__kmpc_barrier(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_barrier(%struct.ident_t* {{.+}}, i32 %{{.+}}) long long k = get_val(); @@ -59,8 +65,8 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_VAR]] // CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]] -// CHECK: call void @__kmpc_dispatch_init_4(%ident_t* {{.+}}, i32 %{{.+}}, i32 1073741859, i32 0, i32 8, i32 1, i32 1) -// CHECK: [[NEXT:%.+]] = call i32 @__kmpc_dispatch_next_4(%ident_t* {{.+}}, i32 %{{.+}}, i32* %{{.+}}, i32* [[LB:%.+]], i32* [[UB:%.+]], i32* %{{.+}}) +// CHECK: call void @__kmpc_dispatch_init_4(%struct.ident_t* {{.+}}, i32 %{{.+}}, i32 1073741859, i32 0, i32 8, i32 1, i32 1) +// CHECK: [[NEXT:%.+]] = call i32 @__kmpc_dispatch_next_4(%struct.ident_t* {{.+}}, i32 %{{.+}}, i32* %{{.+}}, i32* [[LB:%.+]], i32* [[UB:%.+]], i32* %{{.+}}) // CHECK: [[COND:%.+]] = icmp ne i32 [[NEXT]], 0 // CHECK: br i1 [[COND]], label %[[CONT:.+]], label %[[END:.+]] // CHECK: [[CONT]]: @@ -100,7 +106,7 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK: [[LIN0_2:%.+]] = load i64, i64* [[LIN0]] // CHECK-NEXT: [[LIN_ADD2:%.+]] = add nsw i64 [[LIN0_2]], 27 // CHECK-NEXT: store i64 [[LIN_ADD2]], i64* [[K_VAR]] -// CHECK: call void @__kmpc_barrier(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_barrier(%struct.ident_t* {{.+}}, i32 %{{.+}}) int lin = 12; #pragma omp for simd linear(lin : get_val()), linear(g_ptr) @@ -116,7 +122,7 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK: [[GLIN_LOAD:%.+]] = load double*, double** [[GLIN_VAR:@[^,]+]] // CHECK-NEXT: store double* [[GLIN_LOAD]], double** [[GLIN_START:%[^,]+]] -// CHECK: call void @__kmpc_for_static_init_8u(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) +// CHECK: call void @__kmpc_for_static_init_8u(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) // CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]], // CHECK: [[CMP:%.+]] = icmp ugt i64 [[UB_VAL]], 3 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -163,7 +169,7 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK-NEXT: store i64 [[ADD3_2]], i64* [[OMP_IV3]] } // CHECK: [[SIMPLE_LOOP3_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) // // Linear start and step are used to calculate final value of the linear variables. // CHECK: [[LINSTART:.+]] = load i32, i32* [[LIN_START]] @@ -171,10 +177,10 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK: store i32 {{.+}}, i32* [[LIN_VAR]], // CHECK: [[GLINSTART:.+]] = load double*, double** [[GLIN_START]] // CHECK: store double* {{.*}}[[GLIN_VAR]] -// CHECK: call void @__kmpc_barrier(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_barrier(%struct.ident_t* {{.+}}, i32 %{{.+}}) #pragma omp for simd -// CHECK: call void @__kmpc_for_static_init_4(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) +// CHECK: call void @__kmpc_for_static_init_4(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) // CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]], // CHECK: [[CMP:%.+]] = icmp sgt i32 [[UB_VAL]], 3 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -207,11 +213,11 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK-NEXT: store i32 [[ADD4_2]], i32* [[OMP_IV4]] } // CHECK: [[SIMPLE_LOOP4_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) -// CHECK: call void @__kmpc_barrier(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_barrier(%struct.ident_t* {{.+}}, i32 %{{.+}}) #pragma omp for simd -// CHECK: call void @__kmpc_for_static_init_4(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) +// CHECK: call void @__kmpc_for_static_init_4(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) // CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]], // CHECK: [[CMP:%.+]] = icmp sgt i32 [[UB_VAL]], 25 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -244,8 +250,8 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK-NEXT: store i32 [[ADD5_2]], i32* [[OMP_IV5]] } // CHECK: [[SIMPLE_LOOP5_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) -// CHECK: call void @__kmpc_barrier(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_barrier(%struct.ident_t* {{.+}}, i32 %{{.+}}) // CHECK-NOT: mul i32 %{{.+}}, 10 #pragma omp for simd @@ -258,7 +264,7 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK: store i32 -1, i32* [[A:%.+]], A = -1; #pragma omp for simd lastprivate(A) -// CHECK: call void @__kmpc_for_static_init_8(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) +// CHECK: call void @__kmpc_for_static_init_8(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) // CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]], // CHECK: [[CMP:%.+]] = icmp sgt i64 [[UB_VAL]], 6 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -295,14 +301,14 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK-NEXT: store i64 [[ADD7_2]], i64* [[OMP_IV7]] } // CHECK: [[SIMPLE_LOOP7_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) // CHECK: load i32, i32* // CHECK: icmp ne i32 %{{.+}}, 0 // CHECK: br i1 %{{.+}}, label // CHECK: [[A_PRIV_VAL:%.+]] = load i32, i32* [[A_PRIV]], // CHECK-NEXT: store i32 [[A_PRIV_VAL]], i32* %{{.+}}, // CHECK-NEXT: br label -// CHECK: call void @__kmpc_barrier(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_barrier(%struct.ident_t* {{.+}}, i32 %{{.+}}) } int R; #pragma omp parallel @@ -311,7 +317,7 @@ void simple(float *a, float *b, float *c, float *d) { R = -1; // CHECK: store i32 1, i32* [[R_PRIV:%[^,]+]], #pragma omp for simd reduction(*:R) -// CHECK: call void @__kmpc_for_static_init_8(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) +// CHECK: call void @__kmpc_for_static_init_8(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) // CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]], // CHECK: [[CMP:%.+]] = icmp sgt i64 [[UB_VAL]], 6 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -347,13 +353,13 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK-NEXT: store i64 [[ADD8_2]], i64* [[OMP_IV8]] } // CHECK: [[SIMPLE_LOOP8_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) // CHECK: call i32 @__kmpc_reduce( // CHECK: [[R_PRIV_VAL:%.+]] = load i32, i32* [[R_PRIV]], // CHECK: [[RED:%.+]] = mul nsw i32 %{{.+}}, [[R_PRIV_VAL]] // CHECK-NEXT: store i32 [[RED]], i32* %{{.+}}, // CHECK-NEXT: call void @__kmpc_end_reduce( -// CHECK: call void @__kmpc_barrier(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_barrier(%struct.ident_t* {{.+}}, i32 %{{.+}}) } } @@ -372,7 +378,7 @@ int templ1(T a, T *z) { // Instatiation templ1<float,2> // CHECK-LABEL: define {{.*i32}} @{{.*}}templ1{{.*}}(float {{.+}}, float* {{.+}}) -// CHECK: call void @__kmpc_for_static_init_8(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 45, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 2) +// CHECK: call void @__kmpc_for_static_init_8(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 45, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 2) // CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]], // CHECK: [[CMP:%.+]] = icmp sgt i64 [[UB_VAL]], 15 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -412,8 +418,8 @@ int templ1(T a, T *z) { // CHECK-NEXT: store i64 [[INC]], i64* [[T1_OMP_IV]] // CHECK-NEXT: br label {{%.+}} // CHECK: [[T1_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) -// CHECK: call void @__kmpc_barrier(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_barrier(%struct.ident_t* {{.+}}, i32 %{{.+}}) // CHECK: ret i32 0 // void inst_templ1() { @@ -463,7 +469,7 @@ void iter_simple(IterDouble ia, IterDouble ib, IterDouble ic) { // CHECK-NEXT: store i32 [[DIFF5]], i32* [[OMP_LAST_IT:%[^,]+]]{{.+}} #pragma omp for simd -// CHECK: call void @__kmpc_for_static_init_4(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) +// CHECK: call void @__kmpc_for_static_init_4(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) // CHECK-DAG: [[UB_VAL:%.+]] = load i32, i32* [[UB]], // CHECK-DAG: [[OMP_LAST_IT_VAL:%.+]] = load i32, i32* [[OMP_LAST_IT]], // CHECK: [[CMP:%.+]] = icmp sgt i32 [[UB_VAL]], [[OMP_LAST_IT_VAL]] @@ -504,8 +510,8 @@ void iter_simple(IterDouble ia, IterDouble ib, IterDouble ic) { // br label %{{.*}}, !llvm.loop ![[ITER_LOOP_ID]] } // CHECK: [[IT_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) -// CHECK: call void @__kmpc_barrier(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_barrier(%struct.ident_t* {{.+}}, i32 %{{.+}}) // CHECK: ret void } @@ -516,7 +522,7 @@ void collapsed(float *a, float *b, float *c, float *d) { unsigned j; // middle loop couter, leads to unsigned icmp in loop header. // k declared in the loop init below short l; // inner loop counter -// CHECK: call void @__kmpc_for_static_init_4u(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) +// CHECK: call void @__kmpc_for_static_init_4u(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) // CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]], // CHECK: [[CMP:%.+]] = icmp ugt i32 [[UB_VAL]], 119 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -580,13 +586,13 @@ void collapsed(float *a, float *b, float *c, float *d) { // CHECK: [[COLL1_END]]: } // i,j,l are updated; k is not updated. -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) // CHECK: br i1 // CHECK: store i32 3, i32* // CHECK-NEXT: store i32 5, // CHECK-NEXT: store i32 7, // CHECK-NEXT: store i16 9, i16* -// CHECK: call void @__kmpc_barrier(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_barrier(%struct.ident_t* {{.+}}, i32 %{{.+}}) // CHECK: ret void } @@ -604,7 +610,7 @@ void widened(float *a, float *b, float *c, float *d) { // CHECK: [[MUL:%.+]] = mul nsw i64 2, %{{.+}} // CHECK-NEXT: [[SUB:%.+]] = sub nsw i64 [[MUL]], 1 // CHECK-NEXT: store i64 [[SUB]], i64* [[OMP_LAST_IT:%[^,]+]], -// CHECK: call void @__kmpc_for_static_init_8(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) +// CHECK: call void @__kmpc_for_static_init_8(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) // CHECK-DAG: [[UB_VAL:%.+]] = load i64, i64* [[UB]], // CHECK-DAG: [[OMP_LAST_IT_VAL:%.+]] = load i64, i64* [[OMP_LAST_IT]], // CHECK: [[CMP:%.+]] = icmp sgt i64 [[UB_VAL]], [[OMP_LAST_IT_VAL]] @@ -684,4 +690,5 @@ void parallel_simd(float *a) { a[i] += bar(); } // TERM_DEBUG: !{{[0-9]+}} = !DILocation(line: [[@LINE-11]], +// TERM_DEBUG-NOT: line: 0, #endif // HEADER diff --git a/test/OpenMP/for_simd_collapse_messages.cpp b/test/OpenMP/for_simd_collapse_messages.cpp index 2efd49479ea9f..9e9b8ee6752ee 100644 --- a/test/OpenMP/for_simd_collapse_messages.cpp +++ b/test/OpenMP/for_simd_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/for_simd_firstprivate_messages.cpp b/test/OpenMP/for_simd_firstprivate_messages.cpp index 4e96866c1a2c2..ceacdb6a53c19 100644 --- a/test/OpenMP/for_simd_firstprivate_messages.cpp +++ b/test/OpenMP/for_simd_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/for_simd_lastprivate_messages.cpp b/test/OpenMP/for_simd_lastprivate_messages.cpp index a176a2d70eb9a..9ed2232d6a376 100644 --- a/test/OpenMP/for_simd_lastprivate_messages.cpp +++ b/test/OpenMP/for_simd_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/for_simd_linear_messages.cpp b/test/OpenMP/for_simd_linear_messages.cpp index 3f93125a38a29..ff522e7b7ece4 100644 --- a/test/OpenMP/for_simd_linear_messages.cpp +++ b/test/OpenMP/for_simd_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/for_simd_loop_messages.cpp b/test/OpenMP/for_simd_loop_messages.cpp index ea17a836cf80f..a3ae84efd4aec 100644 --- a/test/OpenMP/for_simd_loop_messages.cpp +++ b/test/OpenMP/for_simd_loop_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s + class S { int a; S() : a(0) {} diff --git a/test/OpenMP/for_simd_misc_messages.c b/test/OpenMP/for_simd_misc_messages.c index ca1e366ca092e..4da2e6bb6c15a 100644 --- a/test/OpenMP/for_simd_misc_messages.c +++ b/test/OpenMP/for_simd_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp for simd'}} #pragma omp for simd diff --git a/test/OpenMP/for_simd_private_messages.cpp b/test/OpenMP/for_simd_private_messages.cpp index ca4c3a33fcaff..8fe6d7b721eed 100644 --- a/test/OpenMP/for_simd_private_messages.cpp +++ b/test/OpenMP/for_simd_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/for_simd_reduction_messages.cpp b/test/OpenMP/for_simd_reduction_messages.cpp index a77ab754eff9f..1f5578e611eeb 100644 --- a/test/OpenMP/for_simd_reduction_messages.cpp +++ b/test/OpenMP/for_simd_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/for_simd_safelen_messages.cpp b/test/OpenMP/for_simd_safelen_messages.cpp index fa1b4442cee49..31b0f84cba42f 100644 --- a/test/OpenMP/for_simd_safelen_messages.cpp +++ b/test/OpenMP/for_simd_safelen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/for_simd_schedule_messages.cpp b/test/OpenMP/for_simd_schedule_messages.cpp index 76ba3a42eda2c..b330f8cfd0051 100644 --- a/test/OpenMP/for_simd_schedule_messages.cpp +++ b/test/OpenMP/for_simd_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/for_simd_simdlen_messages.cpp b/test/OpenMP/for_simd_simdlen_messages.cpp index 8fe197979cc27..09954283a11fb 100644 --- a/test/OpenMP/for_simd_simdlen_messages.cpp +++ b/test/OpenMP/for_simd_simdlen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/function-attr.cpp b/test/OpenMP/function-attr.cpp index 87aba40f2b7de..a370cc35f7e8b 100644 --- a/test/OpenMP/function-attr.cpp +++ b/test/OpenMP/function-attr.cpp @@ -1,5 +1,8 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -stack-protector 2 -emit-llvm -o - %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -stack-protector 2 -emit-llvm -o - %s | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Check that function attributes are added to the OpenMP runtime functions. template <class T> diff --git a/test/OpenMP/is_initial_device.c b/test/OpenMP/is_initial_device.c index 0521f0bd53939..2fe93a4eb6c46 100644 --- a/test/OpenMP/is_initial_device.c +++ b/test/OpenMP/is_initial_device.c @@ -7,6 +7,11 @@ // RUN: %clang_cc1 -verify -fopenmp -x c -triple powerpc64le-unknown-unknown -emit-llvm -fopenmp-is-device \ // RUN: %s -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefixes DEVICE,OUTLINED +// RUN: %clang_cc1 -verify -fopenmp-simd -x c -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-unknown-unknown -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x ir -triple powerpc64le-unknown-unknown -emit-llvm %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c -triple powerpc64le-unknown-unknown -emit-llvm -fopenmp-is-device %s -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics int check() { int host = omp_is_initial_device(); diff --git a/test/OpenMP/linking.c b/test/OpenMP/linking.c index 71e978f7fe488..802553c1be752 100644 --- a/test/OpenMP/linking.c +++ b/test/OpenMP/linking.c @@ -20,9 +20,19 @@ // RUN: %clang -no-canonical-prefixes %s -### -o %t.o 2>&1 \ // RUN: -fopenmp=libgomp -target i386-unknown-linux -rtlib=platform \ // RUN: | FileCheck --check-prefix=CHECK-GOMP-LD-32 %s + +// RUN: %clang -no-canonical-prefixes %s -### -o %t.o 2>&1 -fopenmp-simd -target i386-unknown-linux -rtlib=platform | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: lgomp +// SIMD-ONLY2-NOT: lomp +// SIMD-ONLY2-NOT: liomp // CHECK-GOMP-LD-32: "{{.*}}ld{{(.exe)?}}" // CHECK-GOMP-LD-32: "-lgomp" "-lrt" // CHECK-GOMP-LD-32: "-lpthread" "-lc" + +// RUN: %clang -no-canonical-prefixes %s -### -o %t.o 2>&1 -fopenmp-simd -target i386-unknown-linux -rtlib=platform | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: lgomp +// SIMD-ONLY2-NOT: lomp +// SIMD-ONLY2-NOT: liomp // // RUN: %clang -no-canonical-prefixes %s -### -o %t.o 2>&1 \ // RUN: -fopenmp=libgomp -target x86_64-unknown-linux -rtlib=platform \ @@ -79,10 +89,20 @@ // CHECK-MSVC-LINK-64-SAME: -nodefaultlib:vcompd.lib // CHECK-MSVC-LINK-64-SAME: -libpath:{{.+}}/../lib // CHECK-MSVC-LINK-64-SAME: -defaultlib:libomp.lib + +// RUN: %clang -no-canonical-prefixes %s -### -o %t.o 2>&1 -fopenmp-simd -target x86_64-msvc-win32 -rtlib=platform | FileCheck --check-prefix SIMD-ONLY11 %s +// SIMD-ONLY11-NOT: libiomp +// SIMD-ONLY11-NOT: libomp +// SIMD-ONLY11-NOT: libgomp // // RUN: %clang -no-canonical-prefixes %s -### -o %t.o 2>&1 \ // RUN: -fopenmp=libiomp5 -target x86_64-msvc-win32 -rtlib=platform \ // RUN: | FileCheck --check-prefix=CHECK-MSVC-ILINK-64 %s + +// RUN: %clang -no-canonical-prefixes %s -### -o %t.o 2>&1 -fopenmp-simd -target x86_64-msvc-win32 -rtlib=platform | FileCheck --check-prefix SIMD-ONLY11 %s +// SIMD-ONLY11-NOT: libiomp +// SIMD-ONLY11-NOT: libomp +// SIMD-ONLY11-NOT: libgomp // CHECK-MSVC-ILINK-64: link.exe // CHECK-MSVC-ILINK-64-SAME: -nodefaultlib:vcomp.lib // CHECK-MSVC-ILINK-64-SAME: -nodefaultlib:vcompd.lib diff --git a/test/OpenMP/loops_explicit_clauses_codegen.cpp b/test/OpenMP/loops_explicit_clauses_codegen.cpp index dc21fd11af6f9..43fbd56b592d0 100644 --- a/test/OpenMP/loops_explicit_clauses_codegen.cpp +++ b/test/OpenMP/loops_explicit_clauses_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics diff --git a/test/OpenMP/master_ast_print.cpp b/test/OpenMP/master_ast_print.cpp index a36dd02d557b2..40b14c5c50860 100644 --- a/test/OpenMP/master_ast_print.cpp +++ b/test/OpenMP/master_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -21,7 +25,7 @@ int main (int argc, char **argv) { } // CHECK-NEXT: #pragma omp parallel // CHECK-NEXT: { -// CHECK-NEXT: #pragma omp master +// CHECK-NEXT: #pragma omp master{{$}} // CHECK-NEXT: { // CHECK-NEXT: a = 2; // CHECK-NEXT: } diff --git a/test/OpenMP/master_codegen.cpp b/test/OpenMP/master_codegen.cpp index ad92a13e14cfe..d5b8036e8e824 100644 --- a/test/OpenMP/master_codegen.cpp +++ b/test/OpenMP/master_codegen.cpp @@ -2,6 +2,12 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/master_messages.cpp b/test/OpenMP/master_messages.cpp index 397f1394a3e91..26d6bc8e9f35a 100644 --- a/test/OpenMP/master_messages.cpp +++ b/test/OpenMP/master_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + int foo(); int main() { diff --git a/test/OpenMP/nesting_of_regions.cpp b/test/OpenMP/nesting_of_regions.cpp index 6282de4cc4787..0955ee215569e 100644 --- a/test/OpenMP/nesting_of_regions.cpp +++ b/test/OpenMP/nesting_of_regions.cpp @@ -1,5 +1,8 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + void bar(); template <class T> diff --git a/test/OpenMP/no_option.c b/test/OpenMP/no_option.c index 8a65b039fc9ff..b9bc2082b3f9a 100644 --- a/test/OpenMP/no_option.c +++ b/test/OpenMP/no_option.c @@ -1,4 +1,7 @@ // RUN: %clang_cc1 -verify -o - %s + +// RUN: %clang_cc1 -verify -o - %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics int a; diff --git a/test/OpenMP/no_option_no_warn.c b/test/OpenMP/no_option_no_warn.c index 10778c0b1e350..87556388ce0b7 100644 --- a/test/OpenMP/no_option_no_warn.c +++ b/test/OpenMP/no_option_no_warn.c @@ -1,4 +1,7 @@ // RUN: %clang_cc1 -verify -Wno-source-uses-openmp -o - %s + +// RUN: %clang_cc1 -verify -Wno-source-uses-openmp -o - %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics int a; diff --git a/test/OpenMP/nvptx_data_sharing.cpp b/test/OpenMP/nvptx_data_sharing.cpp index 53bac3ced96b3..0acb119915279 100644 --- a/test/OpenMP/nvptx_data_sharing.cpp +++ b/test/OpenMP/nvptx_data_sharing.cpp @@ -1,4 +1,4 @@ -// Test device data sharing codegen. +// Test device global memory data sharing codegen. ///==========================================================================/// // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc @@ -15,43 +15,92 @@ void test_ds(){ int a = 10; #pragma omp parallel { - a = 1000; + a = 1000; + } + int b = 100; + int c = 1000; + #pragma omp parallel private(c) + { + int *c1 = &c; + b = a + 10000; } } } /// ========= In the worker function ========= /// - -// CK1: define internal void @__omp_offloading_{{.*}}test_ds{{.*}}worker() [[ATTR1:#.*]] { -// CK1: [[SHAREDARGS:%.+]] = alloca i8** -// CK1: call i1 @__kmpc_kernel_parallel(i8** %work_fn, i8*** [[SHAREDARGS]], i16 1) -// CK1: [[SHARGSTMP:%.+]] = load i8**, i8*** [[SHAREDARGS]] -// CK1: call void @__omp_outlined___wrapper{{.*}}({{.*}}, i8** [[SHARGSTMP]]) +// CK1: {{.*}}define internal void @__omp_offloading{{.*}}test_ds{{.*}}_worker() +// CK1: call void @llvm.nvvm.barrier0() +// CK1-NOT: call void @__kmpc_data_sharing_init_stack /// ========= In the kernel function ========= /// -// CK1: {{.*}}define void @__omp_offloading{{.*}}test_ds{{.*}}() [[ATTR2:#.*]] { +// CK1: {{.*}}define weak void @__omp_offloading{{.*}}test_ds{{.*}}() // CK1: [[SHAREDARGS1:%.+]] = alloca i8** -// CK1: call void @__kmpc_kernel_prepare_parallel({{.*}}, i8*** [[SHAREDARGS1]], i32 1, i16 1) +// CK1: [[SHAREDARGS2:%.+]] = alloca i8** +// CK1: call void @__kmpc_kernel_init +// CK1: call void @__kmpc_data_sharing_init_stack +// CK1: [[GLOBALSTACK:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i64 8, i16 0) +// CK1: [[GLOBALSTACK2:%.+]] = bitcast i8* [[GLOBALSTACK]] to %struct._globalized_locals_ty* +// CK1: [[A:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[GLOBALSTACK2]], i32 0, i32 0 +// CK1: [[B:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[GLOBALSTACK2]], i32 0, i32 1 +// CK1: store i32 10, i32* [[A]] +// CK1: call void @__kmpc_kernel_prepare_parallel({{.*}}, i16 1) +// CK1: call void @__kmpc_begin_sharing_variables(i8*** [[SHAREDARGS1]], i64 1) // CK1: [[SHARGSTMP1:%.+]] = load i8**, i8*** [[SHAREDARGS1]] -// CK1: [[SHARGSTMP2:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP1]] -// CK1: [[SHAREDVAR:%.+]] = bitcast i32* {{.*}} to i8* +// CK1: [[SHARGSTMP2:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP1]], i64 0 +// CK1: [[SHAREDVAR:%.+]] = bitcast i32* [[A]] to i8* // CK1: store i8* [[SHAREDVAR]], i8** [[SHARGSTMP2]] +// CK1: call void @llvm.nvvm.barrier0() +// CK1: call void @llvm.nvvm.barrier0() +// CK1: call void @__kmpc_end_sharing_variables() +// CK1: store i32 100, i32* [[B]] +// CK1: call void @__kmpc_kernel_prepare_parallel({{.*}}, i16 1) +// CK1: call void @__kmpc_begin_sharing_variables(i8*** [[SHAREDARGS2]], i64 2) +// CK1: [[SHARGSTMP3:%.+]] = load i8**, i8*** [[SHAREDARGS2]] +// CK1: [[SHARGSTMP4:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP3]], i64 0 +// CK1: [[SHAREDVAR1:%.+]] = bitcast i32* [[B]] to i8* +// CK1: store i8* [[SHAREDVAR1]], i8** [[SHARGSTMP4]] +// CK1: [[SHARGSTMP12:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP3]], i64 1 +// CK1: [[SHAREDVAR2:%.+]] = bitcast i32* [[A]] to i8* +// CK1: store i8* [[SHAREDVAR2]], i8** [[SHARGSTMP12]] +// CK1: call void @llvm.nvvm.barrier0() +// CK1: call void @llvm.nvvm.barrier0() +// CK1: call void @__kmpc_end_sharing_variables() +// CK1: call void @__kmpc_data_sharing_pop_stack(i8* [[GLOBALSTACK]]) +// CK1: call void @__kmpc_kernel_deinit(i16 1) /// ========= In the data sharing wrapper function ========= /// -// CK1: {{.*}}define internal void @__omp_outlined___wrapper({{.*}}i8**) [[ATTR1]] { -// CK1: [[SHAREDARGS2:%.+]] = alloca i8** -// CK1: store i8** %2, i8*** [[SHAREDARGS2]] -// CK1: [[SHARGSTMP3:%.+]] = load i8**, i8*** [[SHAREDARGS2]] -// CK1: [[SHARGSTMP4:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP3]] -// CK1: [[SHARGSTMP5:%.+]] = bitcast i8** [[SHARGSTMP4]] to i32** -// CK1: [[SHARGSTMP6:%.+]] = load i32*, i32** [[SHARGSTMP5]] -// CK1: call void @__omp_outlined__({{.*}}, i32* [[SHARGSTMP6]]) +// CK1: {{.*}}define internal void @__omp_outlined{{.*}}wrapper({{.*}}) +// CK1: [[SHAREDARGS4:%.+]] = alloca i8** +// CK1: call void @__kmpc_get_shared_variables(i8*** [[SHAREDARGS4]]) +// CK1: [[SHARGSTMP13:%.+]] = load i8**, i8*** [[SHAREDARGS4]] +// CK1: [[SHARGSTMP14:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP13]], i64 0 +// CK1: [[SHARGSTMP15:%.+]] = bitcast i8** [[SHARGSTMP14]] to i32** +// CK1: [[SHARGSTMP16:%.+]] = load i32*, i32** [[SHARGSTMP15]] +// CK1: call void @__omp_outlined__{{.*}}({{.*}}, i32* [[SHARGSTMP16]]) + +/// outlined function for the second parallel region /// -/// ========= Attributes ========= /// +// CK1: define internal void @{{.+}}(i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable{{.+}}, i32* dereferenceable{{.+}}) +// CK1-NOT: call i8* @__kmpc_data_sharing_push_stack( +// CK1: [[C_ADDR:%.+]] = alloca i32, +// CK1: store i32* [[C_ADDR]], i32** % +// CK1i-NOT: call void @__kmpc_data_sharing_pop_stack( -// CK1-NOT: attributes [[ATTR1]] = { {{.*}}"has-nvptx-shared-depot"{{.*}} } -// CK1: attributes [[ATTR2]] = { {{.*}}"has-nvptx-shared-depot"{{.*}} } +/// ========= In the data sharing wrapper function ========= /// + +// CK1: {{.*}}define internal void @__omp_outlined{{.*}}wrapper({{.*}}) +// CK1: [[SHAREDARGS3:%.+]] = alloca i8** +// CK1: call void @__kmpc_get_shared_variables(i8*** [[SHAREDARGS3]]) +// CK1: [[SHARGSTMP5:%.+]] = load i8**, i8*** [[SHAREDARGS3]] +// CK1: [[SHARGSTMP6:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP5]], i64 0 +// CK1: [[SHARGSTMP7:%.+]] = bitcast i8** [[SHARGSTMP6]] to i32** +// CK1: [[SHARGSTMP8:%.+]] = load i32*, i32** [[SHARGSTMP7]] +// CK1: [[SHARGSTMP9:%.+]] = getelementptr inbounds i8*, i8** [[SHARGSTMP5]], i64 1 +// CK1: [[SHARGSTMP10:%.+]] = bitcast i8** [[SHARGSTMP9]] to i32** +// CK1: [[SHARGSTMP11:%.+]] = load i32*, i32** [[SHARGSTMP10]] +// CK1: call void @__omp_outlined__{{.*}}({{.*}}, i32* [[SHARGSTMP8]], i32* [[SHARGSTMP11]]) #endif + diff --git a/test/OpenMP/nvptx_declare_target_var_ctor_dtor_codegen.cpp b/test/OpenMP/nvptx_declare_target_var_ctor_dtor_codegen.cpp new file mode 100644 index 0000000000000..3cd85eefbc2c7 --- /dev/null +++ b/test/OpenMP/nvptx_declare_target_var_ctor_dtor_codegen.cpp @@ -0,0 +1,111 @@ +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -o - | FileCheck %s --check-prefix HOST --check-prefix CHECK +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -fopenmp -x c++ -triple nvptx64-nvidia-cuda -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix DEVICE --check-prefix CHECK +// RUN: %clang_cc1 -fopenmp -x c++ -triple nvptx64-nvidia-cuda -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -emit-pch -o %t +// RUN: %clang_cc1 -fopenmp -x c++ -triple nvptx64-nvidia-cuda -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -o - | FileCheck %s --check-prefix DEVICE --check-prefix CHECK + +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o -| FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -emit-pch -o %t +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -o - | FileCheck %s --check-prefix SIMD-ONLY + +#ifndef HEADER +#define HEADER + +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +// DEVICE-DAG: [[C_ADDR:.+]] = weak_odr global i32 0, +// DEVICE-DAG: [[CD_ADDR:@.+]] = global %struct.S zeroinitializer, +// HOST-DAG: @[[C_ADDR:.+]] = internal global i32 0, +// HOST-DAG: @[[CD_ADDR:.+]] = global %struct.S zeroinitializer, + +#pragma omp declare target +int foo() { return 0; } +#pragma omp end declare target +int bar() { return 0; } +#pragma omp declare target (bar) +int baz() { return 0; } + +#pragma omp declare target +int doo() { return 0; } +#pragma omp end declare target +int car() { return 0; } +#pragma omp declare target (bar) +int caz() { return 0; } + +// DEVICE-DAG: define i32 [[FOO:@.*foo.*]]() +// DEVICE-DAG: define i32 [[BAR:@.*bar.*]]() +// DEVICE-DAG: define i32 [[BAZ:@.*baz.*]]() +// DEVICE-DAG: define i32 [[DOO:@.*doo.*]]() +// DEVICE-DAG: define i32 [[CAR:@.*car.*]]() +// DEVICE-DAG: define i32 [[CAZ:@.*caz.*]]() + +static int c = foo() + bar() + baz(); +#pragma omp declare target (c) +// HOST-DAG: @[[C_CTOR:__omp_offloading__.+_c_l44_ctor]] = private constant i8 0 +// DEVICE-DAG: define internal void [[C_CTOR:@__omp_offloading__.+_c_l44_ctor]]() +// DEVICE-DAG: call i32 [[FOO]]() +// DEVICE-DAG: call i32 [[BAR]]() +// DEVICE-DAG: call i32 [[BAZ]]() +// DEVICE-DAG: ret void + +struct S { + int a; + S() = default; + S(int a) : a(a) {} + ~S() { a = 0; } +}; + +#pragma omp declare target +S cd = doo() + car() + caz() + baz(); +#pragma omp end declare target +// HOST-DAG: @[[CD_CTOR:__omp_offloading__.+_cd_l61_ctor]] = private constant i8 0 +// DEVICE-DAG: define internal void [[CD_CTOR:@__omp_offloading__.+_cd_l61_ctor]]() +// DEVICE-DAG: call i32 [[DOO]]() +// DEVICE-DAG: call i32 [[CAR]]() +// DEVICE-DAG: call i32 [[CAZ]]() +// DEVICE-DAG: ret void + +// HOST-DAG: @[[CD_DTOR:__omp_offloading__.+_cd_l61_dtor]] = private constant i8 0 +// DEVICE-DAG: define internal void [[CD_DTOR:@__omp_offloading__.+_cd_l61_dtor]]() +// DEVICE-DAG: call void +// DEVICE-DAG: ret void + +// HOST-DAG: @.omp_offloading.entry_name{{.*}} = internal unnamed_addr constant [{{[0-9]+}} x i8] c"[[C_ADDR]]\00" +// HOST-DAG: @.omp_offloading.entry.[[C_ADDR]] = weak constant %struct.__tgt_offload_entry { i8* bitcast (i32* @[[C_ADDR]] to i8*), i8* getelementptr inbounds ([{{[0-9]+}} x i8], [{{[0-9]+}} x i8]* @.omp_offloading.entry_name{{.*}}, i32 0, i32 0), i64 4, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// HOST-DAG: @.omp_offloading.entry_name{{.*}} = internal unnamed_addr constant [{{[0-9]+}} x i8] c"[[CD_ADDR]]\00" +// HOST-DAG: @.omp_offloading.entry.[[CD_ADDR]] = weak constant %struct.__tgt_offload_entry { i8* bitcast (%struct.S* @[[CD_ADDR]] to i8*), i8* getelementptr inbounds ([{{[0-9]+}} x i8], [{{[0-9]+}} x i8]* @.omp_offloading.entry_name{{.*}}, i32 0, i32 0), i64 4, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// HOST-DAG: @.omp_offloading.entry_name{{.*}} = internal unnamed_addr constant [{{[0-9]+}} x i8] c"[[C_CTOR]]\00" +// HOST-DAG: @.omp_offloading.entry.[[C_CTOR]] = weak constant %struct.__tgt_offload_entry { i8* @[[C_CTOR]], i8* getelementptr inbounds ([{{[0-9]+}} x i8], [{{[0-9]+}} x i8]* @.omp_offloading.entry_name{{.*}}, i32 0, i32 0), i64 0, i32 2, i32 0 }, section ".omp_offloading.entries", align 1 +// HOST-DAG: @.omp_offloading.entry_name{{.*}}= internal unnamed_addr constant [{{[0-9]+}} x i8] c"[[CD_CTOR]]\00" +// HOST-DAG: @.omp_offloading.entry.[[CD_CTOR]] = weak constant %struct.__tgt_offload_entry { i8* @[[CD_CTOR]], i8* getelementptr inbounds ([{{[0-9]+}} x i8], [{{[0-9]+}} x i8]* @.omp_offloading.entry_name{{.*}}, i32 0, i32 0), i64 0, i32 2, i32 0 }, section ".omp_offloading.entries", align 1 +// HOST-DAG: @.omp_offloading.entry_name{{.*}}= internal unnamed_addr constant [{{[0-9]+}} x i8] c"[[CD_DTOR]]\00" +// HOST-DAG: @.omp_offloading.entry.[[CD_DTOR]] = weak constant %struct.__tgt_offload_entry { i8* @[[CD_DTOR]], i8* getelementptr inbounds ([{{[0-9]+}} x i8], [{{[0-9]+}} x i8]* @.omp_offloading.entry_name{{.*}}, i32 0, i32 0), i64 0, i32 4, i32 0 }, section ".omp_offloading.entries", align 1 +int maini1() { + int a; +#pragma omp target map(tofrom : a) + { + a = c; + } + return 0; +} + +// DEVICE: define weak void @__omp_offloading_{{.*}}_{{.*}}maini1{{.*}}_l[[@LINE-7]](i32* dereferenceable{{[^,]*}} +// DEVICE: [[C:%.+]] = load i32, i32* [[C_ADDR]], +// DEVICE: store i32 [[C]], i32* % + +// HOST: define internal void @__omp_offloading_{{.*}}_{{.*}}maini1{{.*}}_l[[@LINE-11]](i32* dereferenceable{{.*}}) +// HOST: [[C:%.*]] = load i32, i32* @[[C_ADDR]], +// HOST: store i32 [[C]], i32* % + +// HOST-DAG: !{i32 1, !"[[CD_ADDR]]", i32 0, i32 {{[0-9]+}}} +// HOST-DAG: !{i32 1, !"[[C_ADDR]]", i32 0, i32 {{[0-9]+}}} + +// DEVICE: !nvvm.annotations +// DEVICE-DAG: !{void ()* [[C_CTOR]], !"kernel", i32 1} +// DEVICE-DAG: !{void ()* [[CD_CTOR]], !"kernel", i32 1} +// DEVICE-DAG: !{void ()* [[CD_DTOR]], !"kernel", i32 1} + +#endif // HEADER + diff --git a/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp b/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp new file mode 100644 index 0000000000000..2b18f6d3f9bdf --- /dev/null +++ b/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp @@ -0,0 +1,51 @@ +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +int a; + +int foo(int *a); + +int main(int argc, char **argv) { + int b[10], c[10], d[10]; +#pragma omp target teams map(tofrom:a) +#pragma omp distribute parallel for firstprivate(b) lastprivate(c) if(a) + for (int i= 0; i < argc; ++i) + a = foo(&i) + foo(&a) + foo(&b[i]) + foo(&c[i]) + foo(&d[i]); + return 0; +} + +// CHECK: @__omp_offloading_{{.*}}_main_l17_exec_mode = weak constant i8 1 + +// CHECK-LABEL: define internal void @__omp_offloading_{{.*}}_main_l17_worker( + +// CHECK: define weak void @__omp_offloading_{{.*}}_main_l17([10 x i32]* dereferenceable(40) %{{.+}}, [10 x i32]* dereferenceable(40) %{{.+}}, i32* dereferenceable(4) %{{.+}}, i{{64|32}} %{{.+}}, [10 x i32]* dereferenceable(40) %{{.+}}) +// CHECK: [[PTR:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i{{64|32}} 84, i16 0) +// CHECK: [[STACK:%.+]] = bitcast i8* [[PTR]] to %struct._globalized_locals_ty* +// CHECK: [[ARGC:%.+]] = load i32, i32* %{{.+}}, align +// CHECK: [[ARGC_ADDR:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[STACK]], i{{32|64}} 0, i{{32|64}} 0 +// CHECK: store i32 [[ARGC]], i32* [[ARGC_ADDR]], +// CHECK: getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[STACK]], i{{32|64}} 0, i{{32|64}} 1 +// CHECK: getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[STACK]], i{{32|64}} 0, i{{32|64}} 2 +// CHECK: call void @__kmpc_for_static_init_4( + +// CHECK: call void @__kmpc_serialized_parallel( +// CHECK: call void [[PARALLEL:@.+]]( +// CHECK: call void @__kmpc_end_serialized_parallel( + +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* @ + +// CHECK: call void @__kmpc_data_sharing_pop_stack(i8* [[PTR]]) + +// CHECK: define internal void [[PARALLEL]]( +// CHECK-NOT: call i8* @__kmpc_data_sharing_push_stack( + +// CHECK-NOT: call void @__kmpc_data_sharing_pop_stack( + +#endif diff --git a/test/OpenMP/nvptx_parallel_codegen.cpp b/test/OpenMP/nvptx_parallel_codegen.cpp index ba889bf4975f6..d1a3104407d09 100644 --- a/test/OpenMP/nvptx_parallel_codegen.cpp +++ b/test/OpenMP/nvptx_parallel_codegen.cpp @@ -51,6 +51,15 @@ tx ftemplate(int n) { b[2] += 1; } + #pragma omp target + { + #pragma omp parallel + { + #pragma omp critical + ++a; + } + ++a; + } return a; } @@ -62,256 +71,278 @@ int bar(int n){ return a; } - // CHECK-NOT: define {{.*}}void {{@__omp_offloading_.+template.+l17}}_worker() - - - - - +// CHECK-NOT: define {{.*}}void {{@__omp_offloading_.+template.+l17}}_worker() - // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l26}}_worker() - // CHECK-DAG: [[OMP_EXEC_STATUS:%.+]] = alloca i8, - // CHECK-DAG: [[OMP_WORK_FN:%.+]] = alloca i8*, - // CHECK: store i8* null, i8** [[OMP_WORK_FN]], - // CHECK: store i8 0, i8* [[OMP_EXEC_STATUS]], - // CHECK: br label {{%?}}[[AWAIT_WORK:.+]] - // - // CHECK: [[AWAIT_WORK]] - // CHECK: call void @llvm.nvvm.barrier0() - // CHECK: [[KPR:%.+]] = call i1 @__kmpc_kernel_parallel(i8** [[OMP_WORK_FN]], - // CHECK: [[KPRB:%.+]] = zext i1 [[KPR]] to i8 - // store i8 [[KPRB]], i8* [[OMP_EXEC_STATUS]], align 1 - // CHECK: [[WORK:%.+]] = load i8*, i8** [[OMP_WORK_FN]], - // CHECK: [[SHOULD_EXIT:%.+]] = icmp eq i8* [[WORK]], null - // CHECK: br i1 [[SHOULD_EXIT]], label {{%?}}[[EXIT:.+]], label {{%?}}[[SEL_WORKERS:.+]] - // - // CHECK: [[SEL_WORKERS]] - // CHECK: [[ST:%.+]] = load i8, i8* [[OMP_EXEC_STATUS]] - // CHECK: [[IS_ACTIVE:%.+]] = icmp ne i8 [[ST]], 0 - // CHECK: br i1 [[IS_ACTIVE]], label {{%?}}[[EXEC_PARALLEL:.+]], label {{%?}}[[BAR_PARALLEL:.+]] - // - // CHECK: [[EXEC_PARALLEL]] - // CHECK: [[WF1:%.+]] = load i8*, i8** [[OMP_WORK_FN]], - // CHECK: [[WM1:%.+]] = icmp eq i8* [[WF1]], bitcast (void (i16, i32, i8**)* [[PARALLEL_FN1:@.+]]_wrapper to i8*) - // CHECK: br i1 [[WM1]], label {{%?}}[[EXEC_PFN1:.+]], label {{%?}}[[CHECK_NEXT1:.+]] - // - // CHECK: [[EXEC_PFN1]] - // CHECK: call void [[PARALLEL_FN1]]_wrapper( - // CHECK: br label {{%?}}[[TERM_PARALLEL:.+]] - // - // CHECK: [[CHECK_NEXT1]] - // CHECK: [[WF2:%.+]] = load i8*, i8** [[OMP_WORK_FN]], - // CHECK: [[WM2:%.+]] = icmp eq i8* [[WF2]], bitcast (void (i16, i32, i8**)* [[PARALLEL_FN2:@.+]]_wrapper to i8*) - // CHECK: br i1 [[WM2]], label {{%?}}[[EXEC_PFN2:.+]], label {{%?}}[[CHECK_NEXT2:.+]] - // - // CHECK: [[EXEC_PFN2]] - // CHECK: call void [[PARALLEL_FN2]]_wrapper( - // CHECK: br label {{%?}}[[TERM_PARALLEL:.+]] - // - // CHECK: [[CHECK_NEXT2]] - // CHECK: br label {{%?}}[[TERM_PARALLEL:.+]] - // - // CHECK: [[TERM_PARALLEL]] - // CHECK: call void @__kmpc_kernel_end_parallel() - // CHECK: br label {{%?}}[[BAR_PARALLEL]] - // - // CHECK: [[BAR_PARALLEL]] - // CHECK: call void @llvm.nvvm.barrier0() - // CHECK: br label {{%?}}[[AWAIT_WORK]] - // - // CHECK: [[EXIT]] - // CHECK: ret void +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l26}}_worker() +// CHECK-DAG: [[OMP_EXEC_STATUS:%.+]] = alloca i8, +// CHECK-DAG: [[OMP_WORK_FN:%.+]] = alloca i8*, +// CHECK: store i8* null, i8** [[OMP_WORK_FN]], +// CHECK: store i8 0, i8* [[OMP_EXEC_STATUS]], +// CHECK: br label {{%?}}[[AWAIT_WORK:.+]] +// +// CHECK: [[AWAIT_WORK]] +// CHECK: call void @llvm.nvvm.barrier0() +// CHECK: [[KPR:%.+]] = call i1 @__kmpc_kernel_parallel(i8** [[OMP_WORK_FN]] +// CHECK: [[KPRB:%.+]] = zext i1 [[KPR]] to i8 +// store i8 [[KPRB]], i8* [[OMP_EXEC_STATUS]], align 1 +// CHECK: [[WORK:%.+]] = load i8*, i8** [[OMP_WORK_FN]], +// CHECK: [[SHOULD_EXIT:%.+]] = icmp eq i8* [[WORK]], null +// CHECK: br i1 [[SHOULD_EXIT]], label {{%?}}[[EXIT:.+]], label {{%?}}[[SEL_WORKERS:.+]] +// +// CHECK: [[SEL_WORKERS]] +// CHECK: [[ST:%.+]] = load i8, i8* [[OMP_EXEC_STATUS]] +// CHECK: [[IS_ACTIVE:%.+]] = icmp ne i8 [[ST]], 0 +// CHECK: br i1 [[IS_ACTIVE]], label {{%?}}[[EXEC_PARALLEL:.+]], label {{%?}}[[BAR_PARALLEL:.+]] +// +// CHECK: [[EXEC_PARALLEL]] +// CHECK: [[WF1:%.+]] = load i8*, i8** [[OMP_WORK_FN]], +// CHECK: [[WM1:%.+]] = icmp eq i8* [[WF1]], bitcast (void (i16, i32)* [[PARALLEL_FN1:@.+]]_wrapper to i8*) +// CHECK: br i1 [[WM1]], label {{%?}}[[EXEC_PFN1:.+]], label {{%?}}[[CHECK_NEXT1:.+]] +// +// CHECK: [[EXEC_PFN1]] +// CHECK: call void [[PARALLEL_FN1]]_wrapper( +// CHECK: br label {{%?}}[[TERM_PARALLEL:.+]] +// +// CHECK: [[CHECK_NEXT1]] +// CHECK: [[WF2:%.+]] = load i8*, i8** [[OMP_WORK_FN]], +// CHECK: [[WM2:%.+]] = icmp eq i8* [[WF2]], bitcast (void (i16, i32)* [[PARALLEL_FN2:@.+]]_wrapper to i8*) +// CHECK: br i1 [[WM2]], label {{%?}}[[EXEC_PFN2:.+]], label {{%?}}[[CHECK_NEXT2:.+]] +// +// CHECK: [[EXEC_PFN2]] +// CHECK: call void [[PARALLEL_FN2]]_wrapper( +// CHECK: br label {{%?}}[[TERM_PARALLEL:.+]] +// +// CHECK: [[CHECK_NEXT2]] +// CHECK: br label {{%?}}[[TERM_PARALLEL:.+]] +// +// CHECK: [[TERM_PARALLEL]] +// CHECK: call void @__kmpc_kernel_end_parallel() +// CHECK: br label {{%?}}[[BAR_PARALLEL]] +// +// CHECK: [[BAR_PARALLEL]] +// CHECK: call void @llvm.nvvm.barrier0() +// CHECK: br label {{%?}}[[AWAIT_WORK]] +// +// CHECK: [[EXIT]] +// CHECK: ret void - // CHECK: define {{.*}}void [[T6:@__omp_offloading_.+template.+l26]](i[[SZ:32|64]] - // Create local storage for each capture. - // CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]], - // CHECK-DAG: store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]] - // Store captures in the context. - // CHECK-64-DAG:[[REF_A:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32* - // - // CHECK-DAG: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() - // CHECK-DAG: [[NTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() - // CHECK-DAG: [[WS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK-DAG: [[TH_LIMIT:%.+]] = sub i32 [[NTH]], [[WS]] - // CHECK: [[IS_WORKER:%.+]] = icmp ult i32 [[TID]], [[TH_LIMIT]] - // CHECK: br i1 [[IS_WORKER]], label {{%?}}[[WORKER:.+]], label {{%?}}[[CHECK_MASTER:.+]] - // - // CHECK: [[WORKER]] - // CHECK: {{call|invoke}} void [[T6]]_worker() - // CHECK: br label {{%?}}[[EXIT:.+]] - // - // CHECK: [[CHECK_MASTER]] - // CHECK-DAG: [[CMTID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() - // CHECK-DAG: [[CMNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() - // CHECK-DAG: [[CMWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[IS_MASTER:%.+]] = icmp eq i32 [[CMTID]], - // CHECK: br i1 [[IS_MASTER]], label {{%?}}[[MASTER:.+]], label {{%?}}[[EXIT]] - // - // CHECK: [[MASTER]] - // CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() - // CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[MTMP1:%.+]] = sub i32 [[MNTH]], [[MWS]] - // CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] - // CHECK: call void @__kmpc_kernel_prepare_parallel(i8* bitcast (void (i16, i32, i8**)* [[PARALLEL_FN1]]_wrapper to i8*), - // CHECK: call void @llvm.nvvm.barrier0() - // CHECK: call void @llvm.nvvm.barrier0() - // CHECK: call void @__kmpc_serialized_parallel( - // CHECK: {{call|invoke}} void [[PARALLEL_FN3:@.+]]( - // CHECK: call void @__kmpc_end_serialized_parallel( - // CHECK: call void @__kmpc_kernel_prepare_parallel(i8* bitcast (void (i16, i32, i8**)* [[PARALLEL_FN2]]_wrapper to i8*), - // CHECK: call void @llvm.nvvm.barrier0() - // CHECK: call void @llvm.nvvm.barrier0() - // CHECK-64-DAG: load i32, i32* [[REF_A]] - // CHECK-32-DAG: load i32, i32* [[LOCAL_A]] - // CHECK: br label {{%?}}[[TERMINATE:.+]] - // - // CHECK: [[TERMINATE]] - // CHECK: call void @__kmpc_kernel_deinit( - // CHECK: call void @llvm.nvvm.barrier0() - // CHECK: br label {{%?}}[[EXIT]] - // - // CHECK: [[EXIT]] - // CHECK: ret void +// CHECK: define {{.*}}void [[T6:@__omp_offloading_.+template.+l26]](i[[SZ:32|64]] +// Create local storage for each capture. +// CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]], +// CHECK-DAG: store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]] +// Store captures in the context. +// CHECK-64-DAG:[[REF_A:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32* +// +// CHECK-DAG: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() +// CHECK-DAG: [[NTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK-DAG: [[WS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() +// CHECK-DAG: [[TH_LIMIT:%.+]] = sub nuw i32 [[NTH]], [[WS]] +// CHECK: [[IS_WORKER:%.+]] = icmp ult i32 [[TID]], [[TH_LIMIT]] +// CHECK: br i1 [[IS_WORKER]], label {{%?}}[[WORKER:.+]], label {{%?}}[[CHECK_MASTER:.+]] +// +// CHECK: [[WORKER]] +// CHECK: {{call|invoke}} void [[T6]]_worker() +// CHECK: br label {{%?}}[[EXIT:.+]] +// +// CHECK: [[CHECK_MASTER]] +// CHECK-DAG: [[CMTID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() +// CHECK-DAG: [[CMNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK-DAG: [[CMWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() +// CHECK: [[IS_MASTER:%.+]] = icmp eq i32 [[CMTID]], +// CHECK: br i1 [[IS_MASTER]], label {{%?}}[[MASTER:.+]], label {{%?}}[[EXIT]] +// +// CHECK: [[MASTER]] +// CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() +// CHECK: [[MTMP1:%.+]] = sub nuw i32 [[MNTH]], [[MWS]] +// CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] +// CHECK: call void @__kmpc_kernel_prepare_parallel(i8* bitcast (void (i16, i32)* [[PARALLEL_FN1]]_wrapper to i8*), +// CHECK: call void @llvm.nvvm.barrier0() +// CHECK: call void @llvm.nvvm.barrier0() +// CHECK: call void @__kmpc_serialized_parallel( +// CHECK: {{call|invoke}} void [[PARALLEL_FN3:@.+]]( +// CHECK: call void @__kmpc_end_serialized_parallel( +// CHECK: call void @__kmpc_kernel_prepare_parallel(i8* bitcast (void (i16, i32)* [[PARALLEL_FN2]]_wrapper to i8*), +// CHECK: call void @llvm.nvvm.barrier0() +// CHECK: call void @llvm.nvvm.barrier0() +// CHECK-64-DAG: load i32, i32* [[REF_A]] +// CHECK-32-DAG: load i32, i32* [[LOCAL_A]] +// CHECK: br label {{%?}}[[TERMINATE:.+]] +// +// CHECK: [[TERMINATE]] +// CHECK: call void @__kmpc_kernel_deinit( +// CHECK: call void @llvm.nvvm.barrier0() +// CHECK: br label {{%?}}[[EXIT]] +// +// CHECK: [[EXIT]] +// CHECK: ret void - // CHECK-DAG: define internal void [[PARALLEL_FN1]]( - // CHECK: [[A:%.+]] = alloca i[[SZ:32|64]], - // CHECK: store i[[SZ]] 42, i[[SZ]]* %a, - // CHECK: ret void +// CHECK-DAG: define internal void [[PARALLEL_FN1]]( +// CHECK: [[A:%.+]] = alloca i[[SZ:32|64]], +// CHECK: store i[[SZ]] 42, i[[SZ]]* %a, +// CHECK: ret void - // CHECK-DAG: define internal void [[PARALLEL_FN3]]( - // CHECK: [[A:%.+]] = alloca i[[SZ:32|64]], - // CHECK: store i[[SZ]] 43, i[[SZ]]* %a, - // CHECK: ret void +// CHECK-DAG: define internal void [[PARALLEL_FN3]]( +// CHECK: [[A:%.+]] = alloca i[[SZ:32|64]], +// CHECK: store i[[SZ]] 43, i[[SZ]]* %a, +// CHECK: ret void - // CHECK-DAG: define internal void [[PARALLEL_FN2]]( - // CHECK: [[A:%.+]] = alloca i[[SZ:32|64]], - // CHECK: store i[[SZ]] 44, i[[SZ]]* %a, - // CHECK: ret void +// CHECK-DAG: define internal void [[PARALLEL_FN2]]( +// CHECK: [[A:%.+]] = alloca i[[SZ:32|64]], +// CHECK: store i[[SZ]] 44, i[[SZ]]* %a, +// CHECK: ret void +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l43}}_worker() +// CHECK-DAG: [[OMP_EXEC_STATUS:%.+]] = alloca i8, +// CHECK-DAG: [[OMP_WORK_FN:%.+]] = alloca i8*, +// CHECK: store i8* null, i8** [[OMP_WORK_FN]], +// CHECK: store i8 0, i8* [[OMP_EXEC_STATUS]], +// CHECK: br label {{%?}}[[AWAIT_WORK:.+]] +// +// CHECK: [[AWAIT_WORK]] +// CHECK: call void @llvm.nvvm.barrier0() +// CHECK: [[KPR:%.+]] = call i1 @__kmpc_kernel_parallel(i8** [[OMP_WORK_FN]], +// CHECK: [[KPRB:%.+]] = zext i1 [[KPR]] to i8 +// store i8 [[KPRB]], i8* [[OMP_EXEC_STATUS]], align 1 +// CHECK: [[WORK:%.+]] = load i8*, i8** [[OMP_WORK_FN]], +// CHECK: [[SHOULD_EXIT:%.+]] = icmp eq i8* [[WORK]], null +// CHECK: br i1 [[SHOULD_EXIT]], label {{%?}}[[EXIT:.+]], label {{%?}}[[SEL_WORKERS:.+]] +// +// CHECK: [[SEL_WORKERS]] +// CHECK: [[ST:%.+]] = load i8, i8* [[OMP_EXEC_STATUS]] +// CHECK: [[IS_ACTIVE:%.+]] = icmp ne i8 [[ST]], 0 +// CHECK: br i1 [[IS_ACTIVE]], label {{%?}}[[EXEC_PARALLEL:.+]], label {{%?}}[[BAR_PARALLEL:.+]] +// +// CHECK: [[EXEC_PARALLEL]] +// CHECK: [[WF:%.+]] = load i8*, i8** [[OMP_WORK_FN]], +// CHECK: [[WM:%.+]] = icmp eq i8* [[WF]], bitcast (void (i16, i32)* [[PARALLEL_FN4:@.+]]_wrapper to i8*) +// CHECK: br i1 [[WM]], label {{%?}}[[EXEC_PFN:.+]], label {{%?}}[[CHECK_NEXT:.+]] +// +// CHECK: [[EXEC_PFN]] +// CHECK: call void [[PARALLEL_FN4]]_wrapper( +// CHECK: br label {{%?}}[[TERM_PARALLEL:.+]] +// +// CHECK: [[CHECK_NEXT]] +// CHECK: br label {{%?}}[[TERM_PARALLEL:.+]] +// +// CHECK: [[TERM_PARALLEL]] +// CHECK: call void @__kmpc_kernel_end_parallel() +// CHECK: br label {{%?}}[[BAR_PARALLEL]] +// +// CHECK: [[BAR_PARALLEL]] +// CHECK: call void @llvm.nvvm.barrier0() +// CHECK: br label {{%?}}[[AWAIT_WORK]] +// +// CHECK: [[EXIT]] +// CHECK: ret void +// CHECK: define {{.*}}void [[T6:@__omp_offloading_.+template.+l43]](i[[SZ:32|64]] +// Create local storage for each capture. +// CHECK: [[LOCAL_N:%.+]] = alloca i[[SZ]], +// CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]], +// CHECK: [[LOCAL_AA:%.+]] = alloca i[[SZ]], +// CHECK: [[LOCAL_B:%.+]] = alloca [10 x i32]* +// CHECK-DAG: store i[[SZ]] [[ARG_N:%.+]], i[[SZ]]* [[LOCAL_N]] +// CHECK-DAG: store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]] +// CHECK-DAG: store i[[SZ]] [[ARG_AA:%.+]], i[[SZ]]* [[LOCAL_AA]] +// CHECK-DAG: store [10 x i32]* [[ARG_B:%.+]], [10 x i32]** [[LOCAL_B]] +// Store captures in the context. +// CHECK-64-DAG:[[REF_N:%.+]] = bitcast i[[SZ]]* [[LOCAL_N]] to i32* +// CHECK-64-DAG:[[REF_A:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32* +// CHECK-DAG: [[REF_AA:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA]] to i16* +// CHECK-DAG: [[REF_B:%.+]] = load [10 x i32]*, [10 x i32]** [[LOCAL_B]], +// +// CHECK-DAG: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() +// CHECK-DAG: [[NTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK-DAG: [[WS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() +// CHECK-DAG: [[TH_LIMIT:%.+]] = sub nuw i32 [[NTH]], [[WS]] +// CHECK: [[IS_WORKER:%.+]] = icmp ult i32 [[TID]], [[TH_LIMIT]] +// CHECK: br i1 [[IS_WORKER]], label {{%?}}[[WORKER:.+]], label {{%?}}[[CHECK_MASTER:.+]] +// +// CHECK: [[WORKER]] +// CHECK: {{call|invoke}} void [[T6]]_worker() +// CHECK: br label {{%?}}[[EXIT:.+]] +// +// CHECK: [[CHECK_MASTER]] +// CHECK-DAG: [[CMTID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() +// CHECK-DAG: [[CMNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK-DAG: [[CMWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() +// CHECK: [[IS_MASTER:%.+]] = icmp eq i32 [[CMTID]], +// CHECK: br i1 [[IS_MASTER]], label {{%?}}[[MASTER:.+]], label {{%?}}[[EXIT]] +// +// CHECK: [[MASTER]] +// CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() +// CHECK: [[MTMP1:%.+]] = sub nuw i32 [[MNTH]], [[MWS]] +// CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] +// CHECK-64: [[N:%.+]] = load i32, i32* [[REF_N]], +// CHECK-32: [[N:%.+]] = load i32, i32* [[LOCAL_N]], +// CHECK: [[CMP:%.+]] = icmp sgt i32 [[N]], 1000 +// CHECK: br i1 [[CMP]], label {{%?}}[[IF_THEN:.+]], label {{%?}}[[IF_ELSE:.+]] +// +// CHECK: [[IF_THEN]] +// CHECK: call void @__kmpc_kernel_prepare_parallel(i8* bitcast (void (i16, i32)* [[PARALLEL_FN4]]_wrapper to i8*), +// CHECK: call void @llvm.nvvm.barrier0() +// CHECK: call void @llvm.nvvm.barrier0() +// CHECK: br label {{%?}}[[IF_END:.+]] +// +// CHECK: [[IF_ELSE]] +// CHECK: call void @__kmpc_serialized_parallel( +// CHECK: {{call|invoke}} void [[PARALLEL_FN4]]( +// CHECK: call void @__kmpc_end_serialized_parallel( +// br label [[IF_END]] +// +// CHECK: [[IF_END]] +// CHECK-64-DAG: load i32, i32* [[REF_A]] +// CHECK-32-DAG: load i32, i32* [[LOCAL_A]] +// CHECK-DAG: load i16, i16* [[REF_AA]] +// CHECK-DAG: getelementptr inbounds [10 x i32], [10 x i32]* [[REF_B]], i[[SZ]] 0, i[[SZ]] 2 +// +// CHECK: br label {{%?}}[[TERMINATE:.+]] +// +// CHECK: [[TERMINATE]] +// CHECK: call void @__kmpc_kernel_deinit( +// CHECK: call void @llvm.nvvm.barrier0() +// CHECK: br label {{%?}}[[EXIT]] +// +// CHECK: [[EXIT]] +// CHECK: ret void +// CHECK: define internal void [[PARALLEL_FN4]]( +// CHECK: [[A:%.+]] = alloca i[[SZ:32|64]], +// CHECK: store i[[SZ]] 45, i[[SZ]]* %a, +// CHECK: ret void +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l54}}_worker() +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l54}}( +// CHECK-32: [[A_ADDR:%.+]] = alloca i32, +// CHECK-64: [[A_ADDR:%.+]] = alloca i64, +// CHECK-64: [[CONV:%.+]] = bitcast i64* [[A_ADDR]] to i32* +// CHECK: [[STACK:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i{{64|32}} 4, i16 0) +// CHECK: [[BC:%.+]] = bitcast i8* [[STACK]] to %struct._globalized_locals_ty* +// CHECK-32: [[A:%.+]] = load i32, i32* [[A_ADDR]], +// CHECK-64: [[A:%.+]] = load i32, i32* [[CONV]], +// CHECK: [[GLOBAL_A_ADDR:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[BC]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 +// CHECK: store i32 [[A]], i32* [[GLOBAL_A_ADDR]], +// CHECK-LABEL: define internal void @{{.+}}(i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable{{.*}}) +// CHECK: [[CC:%.+]] = alloca i32, +// CHECK: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() +// CHECK: [[NUM_THREADS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK: store i32 0, i32* [[CC]], +// CHECK: br label +// CHECK: [[CC_VAL:%.+]] = load i32, i32* [[CC]], +// CHECK: [[RES:%.+]] = icmp slt i32 [[CC_VAL]], [[NUM_THREADS]] +// CHECK: br i1 [[RES]], label - // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l43}}_worker() - // CHECK-DAG: [[OMP_EXEC_STATUS:%.+]] = alloca i8, - // CHECK-DAG: [[OMP_WORK_FN:%.+]] = alloca i8*, - // CHECK: store i8* null, i8** [[OMP_WORK_FN]], - // CHECK: store i8 0, i8* [[OMP_EXEC_STATUS]], - // CHECK: br label {{%?}}[[AWAIT_WORK:.+]] - // - // CHECK: [[AWAIT_WORK]] - // CHECK: call void @llvm.nvvm.barrier0() - // CHECK: [[KPR:%.+]] = call i1 @__kmpc_kernel_parallel(i8** [[OMP_WORK_FN]], - // CHECK: [[KPRB:%.+]] = zext i1 [[KPR]] to i8 - // store i8 [[KPRB]], i8* [[OMP_EXEC_STATUS]], align 1 - // CHECK: [[WORK:%.+]] = load i8*, i8** [[OMP_WORK_FN]], - // CHECK: [[SHOULD_EXIT:%.+]] = icmp eq i8* [[WORK]], null - // CHECK: br i1 [[SHOULD_EXIT]], label {{%?}}[[EXIT:.+]], label {{%?}}[[SEL_WORKERS:.+]] - // - // CHECK: [[SEL_WORKERS]] - // CHECK: [[ST:%.+]] = load i8, i8* [[OMP_EXEC_STATUS]] - // CHECK: [[IS_ACTIVE:%.+]] = icmp ne i8 [[ST]], 0 - // CHECK: br i1 [[IS_ACTIVE]], label {{%?}}[[EXEC_PARALLEL:.+]], label {{%?}}[[BAR_PARALLEL:.+]] - // - // CHECK: [[EXEC_PARALLEL]] - // CHECK: [[WF:%.+]] = load i8*, i8** [[OMP_WORK_FN]], - // CHECK: [[WM:%.+]] = icmp eq i8* [[WF]], bitcast (void (i16, i32, i8**)* [[PARALLEL_FN4:@.+]]_wrapper to i8*) - // CHECK: br i1 [[WM]], label {{%?}}[[EXEC_PFN:.+]], label {{%?}}[[CHECK_NEXT:.+]] - // - // CHECK: [[EXEC_PFN]] - // CHECK: call void [[PARALLEL_FN4]]_wrapper( - // CHECK: br label {{%?}}[[TERM_PARALLEL:.+]] - // - // CHECK: [[CHECK_NEXT]] - // CHECK: br label {{%?}}[[TERM_PARALLEL:.+]] - // - // CHECK: [[TERM_PARALLEL]] - // CHECK: call void @__kmpc_kernel_end_parallel() - // CHECK: br label {{%?}}[[BAR_PARALLEL]] - // - // CHECK: [[BAR_PARALLEL]] - // CHECK: call void @llvm.nvvm.barrier0() - // CHECK: br label {{%?}}[[AWAIT_WORK]] - // - // CHECK: [[EXIT]] - // CHECK: ret void +// CHECK: [[CC_VAL:%.+]] = load i32, i32* [[CC]], +// CHECK: [[RES:%.+]] = icmp eq i32 [[TID]], [[CC_VAL]] +// CHECK: br i1 [[RES]], label - // CHECK: define {{.*}}void [[T6:@__omp_offloading_.+template.+l43]](i[[SZ:32|64]] - // Create local storage for each capture. - // CHECK: [[LOCAL_N:%.+]] = alloca i[[SZ]], - // CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]], - // CHECK: [[LOCAL_AA:%.+]] = alloca i[[SZ]], - // CHECK: [[LOCAL_B:%.+]] = alloca [10 x i32]* - // CHECK-DAG: store i[[SZ]] [[ARG_N:%.+]], i[[SZ]]* [[LOCAL_N]] - // CHECK-DAG: store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]] - // CHECK-DAG: store i[[SZ]] [[ARG_AA:%.+]], i[[SZ]]* [[LOCAL_AA]] - // CHECK-DAG: store [10 x i32]* [[ARG_B:%.+]], [10 x i32]** [[LOCAL_B]] - // Store captures in the context. - // CHECK-64-DAG:[[REF_N:%.+]] = bitcast i[[SZ]]* [[LOCAL_N]] to i32* - // CHECK-64-DAG:[[REF_A:%.+]] = bitcast i[[SZ]]* [[LOCAL_A]] to i32* - // CHECK-DAG: [[REF_AA:%.+]] = bitcast i[[SZ]]* [[LOCAL_AA]] to i16* - // CHECK-DAG: [[REF_B:%.+]] = load [10 x i32]*, [10 x i32]** [[LOCAL_B]], - // - // CHECK-DAG: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() - // CHECK-DAG: [[NTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() - // CHECK-DAG: [[WS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK-DAG: [[TH_LIMIT:%.+]] = sub i32 [[NTH]], [[WS]] - // CHECK: [[IS_WORKER:%.+]] = icmp ult i32 [[TID]], [[TH_LIMIT]] - // CHECK: br i1 [[IS_WORKER]], label {{%?}}[[WORKER:.+]], label {{%?}}[[CHECK_MASTER:.+]] - // - // CHECK: [[WORKER]] - // CHECK: {{call|invoke}} void [[T6]]_worker() - // CHECK: br label {{%?}}[[EXIT:.+]] - // - // CHECK: [[CHECK_MASTER]] - // CHECK-DAG: [[CMTID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() - // CHECK-DAG: [[CMNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() - // CHECK-DAG: [[CMWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[IS_MASTER:%.+]] = icmp eq i32 [[CMTID]], - // CHECK: br i1 [[IS_MASTER]], label {{%?}}[[MASTER:.+]], label {{%?}}[[EXIT]] - // - // CHECK: [[MASTER]] - // CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() - // CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[MTMP1:%.+]] = sub i32 [[MNTH]], [[MWS]] - // CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] - // CHECK-64: [[N:%.+]] = load i32, i32* [[REF_N]], - // CHECK-32: [[N:%.+]] = load i32, i32* [[LOCAL_N]], - // CHECK: [[CMP:%.+]] = icmp sgt i32 [[N]], 1000 - // CHECK: br i1 [[CMP]], label {{%?}}[[IF_THEN:.+]], label {{%?}}[[IF_ELSE:.+]] - // - // CHECK: [[IF_THEN]] - // CHECK: call void @__kmpc_kernel_prepare_parallel(i8* bitcast (void (i16, i32, i8**)* [[PARALLEL_FN4]]_wrapper to i8*), - // CHECK: call void @llvm.nvvm.barrier0() - // CHECK: call void @llvm.nvvm.barrier0() - // CHECK: br label {{%?}}[[IF_END:.+]] - // - // CHECK: [[IF_ELSE]] - // CHECK: call void @__kmpc_serialized_parallel( - // CHECK: {{call|invoke}} void [[PARALLEL_FN4]]( - // CHECK: call void @__kmpc_end_serialized_parallel( - // br label [[IF_END]] - // - // CHECK: [[IF_END]] - // CHECK-64-DAG: load i32, i32* [[REF_A]] - // CHECK-32-DAG: load i32, i32* [[LOCAL_A]] - // CHECK-DAG: load i16, i16* [[REF_AA]] - // CHECK-DAG: getelementptr inbounds [10 x i32], [10 x i32]* [[REF_B]], i[[SZ]] 0, i[[SZ]] 2 - // - // CHECK: br label {{%?}}[[TERMINATE:.+]] - // - // CHECK: [[TERMINATE]] - // CHECK: call void @__kmpc_kernel_deinit( - // CHECK: call void @llvm.nvvm.barrier0() - // CHECK: br label {{%?}}[[EXIT]] - // - // CHECK: [[EXIT]] - // CHECK: ret void +// CHECK: call void @llvm.nvvm.barrier0() +// CHECK: [[NEW_CC_VAL:%.+]] = add nsw i32 [[CC_VAL]], 1 +// CHECK: store i32 [[NEW_CC_VAL]], i32* [[CC]], +// CHECK: br label - // CHECK: define internal void [[PARALLEL_FN4]]( - // CHECK: [[A:%.+]] = alloca i[[SZ:32|64]], - // CHECK: store i[[SZ]] 45, i[[SZ]]* %a, - // CHECK: ret void #endif diff --git a/test/OpenMP/nvptx_target_codegen.cpp b/test/OpenMP/nvptx_target_codegen.cpp index 23b40e10c4c8e..5f9b3bd328087 100644 --- a/test/OpenMP/nvptx_target_codegen.cpp +++ b/test/OpenMP/nvptx_target_codegen.cpp @@ -9,19 +9,22 @@ #define HEADER // Check that the execution mode of all 6 target regions is set to Generic Mode. -// CHECK-DAG: {{@__omp_offloading_.+l100}}_exec_mode = weak constant i8 1 -// CHECK-DAG: {{@__omp_offloading_.+l177}}_exec_mode = weak constant i8 1 -// CHECK-DAG: {{@__omp_offloading_.+l287}}_exec_mode = weak constant i8 1 -// CHECK-DAG: {{@__omp_offloading_.+l324}}_exec_mode = weak constant i8 1 -// CHECK-DAG: {{@__omp_offloading_.+l342}}_exec_mode = weak constant i8 1 -// CHECK-DAG: {{@__omp_offloading_.+l307}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l103}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l180}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l290}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l328}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l346}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l311}}_exec_mode = weak constant i8 1 __thread int id; +int baz(int f, double &a); + template<typename tx, typename ty> struct TT{ tx X; ty Y; + tx &operator[](int i) { return X; } }; int foo(int n) { @@ -33,7 +36,7 @@ int foo(int n) { double cn[5][n]; TT<long long, char> d; - // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+foo.+l100}}_worker() + // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+foo.+l103}}_worker() // CHECK-DAG: [[OMP_EXEC_STATUS:%.+]] = alloca i8, // CHECK-DAG: [[OMP_WORK_FN:%.+]] = alloca i8*, // CHECK: store i8* null, i8** [[OMP_WORK_FN]], @@ -64,11 +67,11 @@ int foo(int n) { // CHECK: [[EXIT]] // CHECK: ret void - // CHECK: define {{.*}}void [[T1:@__omp_offloading_.+foo.+l100]]() + // CHECK: define {{.*}}void [[T1:@__omp_offloading_.+foo.+l103]]() // CHECK-DAG: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() // CHECK-DAG: [[NTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[WS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK-DAG: [[TH_LIMIT:%.+]] = sub i32 [[NTH]], [[WS]] + // CHECK-DAG: [[TH_LIMIT:%.+]] = sub nuw i32 [[NTH]], [[WS]] // CHECK: [[IS_WORKER:%.+]] = icmp ult i32 [[TID]], [[TH_LIMIT]] // CHECK: br i1 [[IS_WORKER]], label {{%?}}[[WORKER:.+]], label {{%?}}[[CHECK_MASTER:.+]] // @@ -86,7 +89,7 @@ int foo(int n) { // CHECK: [[MASTER]] // CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[MTMP1:%.+]] = sub i32 [[MNTH]], [[MWS]] + // CHECK: [[MTMP1:%.+]] = sub nuw i32 [[MNTH]], [[MWS]] // CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] // CHECK: br label {{%?}}[[TERMINATE:.+]] // @@ -106,7 +109,7 @@ int foo(int n) { { } - // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+foo.+l177}}_worker() + // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+foo.+l180}}_worker() // CHECK-DAG: [[OMP_EXEC_STATUS:%.+]] = alloca i8, // CHECK-DAG: [[OMP_WORK_FN:%.+]] = alloca i8*, // CHECK: store i8* null, i8** [[OMP_WORK_FN]], @@ -137,14 +140,14 @@ int foo(int n) { // CHECK: [[EXIT]] // CHECK: ret void - // CHECK: define {{.*}}void [[T2:@__omp_offloading_.+foo.+l177]](i[[SZ:32|64]] [[ARG1:%[a-zA-Z_]+]], i[[SZ:32|64]] [[ID:%[a-zA-Z_]+]]) + // CHECK: define {{.*}}void [[T2:@__omp_offloading_.+foo.+l180]](i[[SZ:32|64]] [[ARG1:%[a-zA-Z_]+]], i[[SZ:32|64]] [[ID:%[a-zA-Z_]+]]) // CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], // CHECK: store i[[SZ]] [[ARG1]], i[[SZ]]* [[AA_ADDR]], // CHECK: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16* // CHECK-DAG: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() // CHECK-DAG: [[NTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[WS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK-DAG: [[TH_LIMIT:%.+]] = sub i32 [[NTH]], [[WS]] + // CHECK-DAG: [[TH_LIMIT:%.+]] = sub nuw i32 [[NTH]], [[WS]] // CHECK: [[IS_WORKER:%.+]] = icmp ult i32 [[TID]], [[TH_LIMIT]] // CHECK: br i1 [[IS_WORKER]], label {{%?}}[[WORKER:.+]], label {{%?}}[[CHECK_MASTER:.+]] // @@ -162,7 +165,7 @@ int foo(int n) { // CHECK: [[MASTER]] // CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[MTMP1:%.+]] = sub i32 [[MNTH]], [[MWS]] + // CHECK: [[MTMP1:%.+]] = sub nuw i32 [[MNTH]], [[MWS]] // CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] // CHECK: load i16, i16* [[AA_CADDR]], // CHECK: br label {{%?}}[[TERMINATE:.+]] @@ -180,7 +183,7 @@ int foo(int n) { id = aa; } - // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+foo.+l287}}_worker() + // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+foo.+l290}}_worker() // CHECK-DAG: [[OMP_EXEC_STATUS:%.+]] = alloca i8, // CHECK-DAG: [[OMP_WORK_FN:%.+]] = alloca i8*, // CHECK: store i8* null, i8** [[OMP_WORK_FN]], @@ -211,7 +214,7 @@ int foo(int n) { // CHECK: [[EXIT]] // CHECK: ret void - // CHECK: define {{.*}}void [[T3:@__omp_offloading_.+foo.+l287]](i[[SZ]] + // CHECK: define {{.*}}void [[T3:@__omp_offloading_.+foo.+l290]](i[[SZ]] // Create local storage for each capture. // CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]] // CHECK: [[LOCAL_B:%.+]] = alloca [10 x float]* @@ -245,7 +248,7 @@ int foo(int n) { // CHECK-DAG: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() // CHECK-DAG: [[NTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[WS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK-DAG: [[TH_LIMIT:%.+]] = sub i32 [[NTH]], [[WS]] + // CHECK-DAG: [[TH_LIMIT:%.+]] = sub nuw i32 [[NTH]], [[WS]] // CHECK: [[IS_WORKER:%.+]] = icmp ult i32 [[TID]], [[TH_LIMIT]] // CHECK: br i1 [[IS_WORKER]], label {{%?}}[[WORKER:.+]], label {{%?}}[[CHECK_MASTER:.+]] // @@ -263,7 +266,7 @@ int foo(int n) { // CHECK: [[MASTER]] // CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[MTMP1:%.+]] = sub i32 [[MNTH]], [[MWS]] + // CHECK: [[MTMP1:%.+]] = sub nuw i32 [[MNTH]], [[MWS]] // CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] // // Use captures. @@ -293,6 +296,7 @@ int foo(int n) { cn[1][3] += 1.0; d.X += 1; d.Y += 1; + d[0] += 1; } return a; @@ -343,6 +347,7 @@ struct S1 { { this->a = (double)b + 1.5; c[1][1] = ++a; + baz(a, a); } return c[1][1] + (int)b; @@ -364,7 +369,13 @@ int bar(int n){ return a; } - // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+static.+324}}_worker() +int baz(int f, double &a) { +#pragma omp parallel + f = 2 + a; + return f; +} + + // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+static.+328}}_worker() // CHECK-DAG: [[OMP_EXEC_STATUS:%.+]] = alloca i8, // CHECK-DAG: [[OMP_WORK_FN:%.+]] = alloca i8*, // CHECK: store i8* null, i8** [[OMP_WORK_FN]], @@ -395,7 +406,7 @@ int bar(int n){ // CHECK: [[EXIT]] // CHECK: ret void - // CHECK: define {{.*}}void [[T4:@__omp_offloading_.+static.+l324]](i[[SZ]] + // CHECK: define {{.*}}void [[T4:@__omp_offloading_.+static.+l328]](i[[SZ]] // Create local storage for each capture. // CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]] // CHECK: [[LOCAL_AA:%.+]] = alloca i[[SZ]] @@ -414,7 +425,7 @@ int bar(int n){ // CHECK-DAG: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() // CHECK-DAG: [[NTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[WS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK-DAG: [[TH_LIMIT:%.+]] = sub i32 [[NTH]], [[WS]] + // CHECK-DAG: [[TH_LIMIT:%.+]] = sub nuw i32 [[NTH]], [[WS]] // CHECK: [[IS_WORKER:%.+]] = icmp ult i32 [[TID]], [[TH_LIMIT]] // CHECK: br i1 [[IS_WORKER]], label {{%?}}[[WORKER:.+]], label {{%?}}[[CHECK_MASTER:.+]] // @@ -432,7 +443,7 @@ int bar(int n){ // CHECK: [[MASTER]] // CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[MTMP1:%.+]] = sub i32 [[MNTH]], [[MWS]] + // CHECK: [[MTMP1:%.+]] = sub nuw i32 [[MNTH]], [[MWS]] // CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] // CHECK-64-DAG: load i32, i32* [[REF_A]] // CHECK-32-DAG: load i32, i32* [[LOCAL_A]] @@ -450,9 +461,10 @@ int bar(int n){ - // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+S1.+l342}}_worker() + // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+S1.+l346}}_worker() // CHECK-DAG: [[OMP_EXEC_STATUS:%.+]] = alloca i8, // CHECK-DAG: [[OMP_WORK_FN:%.+]] = alloca i8*, + // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* // CHECK: store i8* null, i8** [[OMP_WORK_FN]], // CHECK: store i8 0, i8* [[OMP_EXEC_STATUS]], // CHECK: br label {{%?}}[[AWAIT_WORK:.+]] @@ -469,6 +481,8 @@ int bar(int n){ // CHECK: br i1 [[IS_ACTIVE]], label {{%?}}[[EXEC_PARALLEL:.+]], label {{%?}}[[BAR_PARALLEL:.+]] // // CHECK: [[EXEC_PARALLEL]] + // CHECK: [[WORK_FN:%.+]] = bitcast i8* [[WORK]] to void (i16, i32)* + // CHECK: call void [[WORK_FN]](i16 0, i32 [[GTID]]) // CHECK: br label {{%?}}[[TERM_PARALLEL:.+]] // // CHECK: [[TERM_PARALLEL]] @@ -481,7 +495,7 @@ int bar(int n){ // CHECK: [[EXIT]] // CHECK: ret void - // CHECK: define {{.*}}void [[T5:@__omp_offloading_.+S1.+l342]]( + // CHECK: define {{.*}}void [[T5:@__omp_offloading_.+S1.+l346]]( // Create local storage for each capture. // CHECK: [[LOCAL_THIS:%.+]] = alloca [[S1:%struct.*]]* // CHECK: [[LOCAL_B:%.+]] = alloca i[[SZ]] @@ -503,7 +517,7 @@ int bar(int n){ // CHECK-DAG: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() // CHECK-DAG: [[NTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[WS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK-DAG: [[TH_LIMIT:%.+]] = sub i32 [[NTH]], [[WS]] + // CHECK-DAG: [[TH_LIMIT:%.+]] = sub nuw i32 [[NTH]], [[WS]] // CHECK: [[IS_WORKER:%.+]] = icmp ult i32 [[TID]], [[TH_LIMIT]] // CHECK: br i1 [[IS_WORKER]], label {{%?}}[[WORKER:.+]], label {{%?}}[[CHECK_MASTER:.+]] // @@ -521,13 +535,14 @@ int bar(int n){ // CHECK: [[MASTER]] // CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[MTMP1:%.+]] = sub i32 [[MNTH]], [[MWS]] + // CHECK: [[MTMP1:%.+]] = sub nuw i32 [[MNTH]], [[MWS]] // CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] // Use captures. // CHECK-DAG: getelementptr inbounds [[S1]], [[S1]]* [[REF_THIS]], i32 0, i32 0 // CHECK-64-DAG:load i32, i32* [[REF_B]] // CHECK-32-DAG:load i32, i32* [[LOCAL_B]] // CHECK-DAG: getelementptr inbounds i16, i16* [[REF_C]], i[[SZ]] %{{.+}} + // CHECK: call i32 [[BAZ:@.*baz.*]](i32 % // CHECK: br label {{%?}}[[TERMINATE:.+]] // // CHECK: [[TERMINATE]] @@ -538,9 +553,53 @@ int bar(int n){ // CHECK: [[EXIT]] // CHECK: ret void + // CHECK: define i32 [[BAZ]](i32 [[F:%.*]], double* dereferenceable{{.*}}) + // CHECK: [[ZERO_ADDR:%.+]] = alloca i32, + // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* + // CHECK: [[GTID_ADDR:%.+]] = alloca i32, + // CHECK: store i32 0, i32* [[ZERO_ADDR]] + // CHECK: [[PTR:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i{{64|32}} 4, i16 0) + // CHECK: [[REC_ADDR:%.+]] = bitcast i8* [[PTR]] to %struct._globalized_locals_ty* + // CHECK: [[F_PTR:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[REC_ADDR]], i32 0, i32 0 + // CHECK: store i32 %{{.+}}, i32* [[F_PTR]], + + // CHECK: [[RES:%.+]] = call i8 @__kmpc_is_spmd_exec_mode() + // CHECK: icmp ne i8 [[RES]], 0 + // CHECK: br i1 + + // CHECK: [[RES:%.+]] = call i16 @__kmpc_parallel_level(%struct.ident_t* @{{.+}}, i32 [[GTID]]) + // CHECK: icmp ne i16 [[RES]], 0 + // CHECK: br i1 + + // CHECK: call void @__kmpc_serialized_parallel(%struct.ident_t* @{{.+}}, i32 [[GTID]]) + // CHECK: call void [[OUTLINED:@.+]](i32* [[ZERO_ADDR]], i32* [[ZERO_ADDR]], i32* [[F_PTR]], double* %{{.+}}) + // CHECK: call void @__kmpc_end_serialized_parallel(%struct.ident_t* @{{.+}}, i32 [[GTID]]) + // CHECK: br label + + // CHECK: icmp eq i32 + // CHECK: br i1 + + // CHECK: call void @__kmpc_kernel_prepare_parallel(i8* bitcast (void (i16, i32)* @{{.+}} to i8*), i16 1) + // CHECK: call void @__kmpc_begin_sharing_variables(i8*** [[SHARED_PTR:%.+]], i{{64|32}} 2) + // CHECK: [[SHARED:%.+]] = load i8**, i8*** [[SHARED_PTR]], + // CHECK: [[REF:%.+]] = getelementptr inbounds i8*, i8** [[SHARED]], i{{64|32}} 0 + // CHECK: [[F_REF:%.+]] = bitcast i32* [[F_PTR]] to i8* + // CHECK: store i8* [[F_REF]], i8** [[REF]], + // CHECK: call void @llvm.nvvm.barrier0() + // CHECK: call void @llvm.nvvm.barrier0() + // CHECK: call void @__kmpc_end_sharing_variables() + // CHECK: br label + + // CHECK: store i32 [[GTID]], i32* [[GTID_ADDR]], + // CHECK: call void [[OUTLINED]](i32* [[GTID_ADDR]], i32* [[ZERO_ADDR]], i32* [[F_PTR]], double* %{{.+}}) + // CHECK: br label + + // CHECK: [[RES:%.+]] = load i32, i32* [[F_PTR]], + // CHECK: call void @__kmpc_data_sharing_pop_stack(i8* [[PTR]]) + // CHECK: ret i32 [[RES]] - // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l307}}_worker() + // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l311}}_worker() // CHECK-DAG: [[OMP_EXEC_STATUS:%.+]] = alloca i8, // CHECK-DAG: [[OMP_WORK_FN:%.+]] = alloca i8*, // CHECK: store i8* null, i8** [[OMP_WORK_FN]], @@ -571,7 +630,7 @@ int bar(int n){ // CHECK: [[EXIT]] // CHECK: ret void - // CHECK: define {{.*}}void [[T6:@__omp_offloading_.+template.+l307]](i[[SZ]] + // CHECK: define {{.*}}void [[T6:@__omp_offloading_.+template.+l311]](i[[SZ]] // Create local storage for each capture. // CHECK: [[LOCAL_A:%.+]] = alloca i[[SZ]] // CHECK: [[LOCAL_AA:%.+]] = alloca i[[SZ]] @@ -587,7 +646,7 @@ int bar(int n){ // CHECK-DAG: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() // CHECK-DAG: [[NTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[WS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK-DAG: [[TH_LIMIT:%.+]] = sub i32 [[NTH]], [[WS]] + // CHECK-DAG: [[TH_LIMIT:%.+]] = sub nuw i32 [[NTH]], [[WS]] // CHECK: [[IS_WORKER:%.+]] = icmp ult i32 [[TID]], [[TH_LIMIT]] // CHECK: br i1 [[IS_WORKER]], label {{%?}}[[WORKER:.+]], label {{%?}}[[CHECK_MASTER:.+]] // @@ -605,7 +664,7 @@ int bar(int n){ // CHECK: [[MASTER]] // CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[MTMP1:%.+]] = sub i32 [[MNTH]], [[MWS]] + // CHECK: [[MTMP1:%.+]] = sub nuw i32 [[MNTH]], [[MWS]] // CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] // // CHECK-64-DAG: load i32, i32* [[REF_A]] diff --git a/test/OpenMP/nvptx_target_cuda_mode_messages.cpp b/test/OpenMP/nvptx_target_cuda_mode_messages.cpp new file mode 100644 index 0000000000000..eecc26c6181a0 --- /dev/null +++ b/test/OpenMP/nvptx_target_cuda_mode_messages.cpp @@ -0,0 +1,108 @@ +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-cuda-mode -fopenmp-host-ir-file-path %t-ppc-host.bc -o - + +template <typename tx, typename ty> +struct TT { + tx X; + ty Y; +}; + +int foo(int n, double *ptr) { + int a = 0; + short aa = 0; + float b[10]; + double c[5][10]; + TT<long long, char> d; + +#pragma omp target firstprivate(a) map(tofrom: b) // expected-note 2 {{defined as threadprivate or thread local}} + { + int c; // expected-note {{defined as threadprivate or thread local}} +#pragma omp parallel shared(a, b, c, aa) // expected-error 3 {{threadprivate or thread local variable cannot be shared}} + b[a] = a; +#pragma omp parallel for + for (int i = 0; i < 10; ++i) // expected-note {{defined as threadprivate or thread local}} +#pragma omp parallel shared(i) // expected-error {{threadprivate or thread local variable cannot be shared}} + ++i; + } + +#pragma omp target map(aa, b, c, d) + { + int e; // expected-note {{defined as threadprivate or thread local}} +#pragma omp parallel private(b, e) // expected-error {{threadprivate or thread local variable cannot be private}} + { + aa += 1; + b[2] = 1.0; + c[1][2] = 1.0; + d.X = 1; + d.Y = 1; + } + } + +#pragma omp target private(ptr) + { + ptr[0]++; + } + + return a; +} + +template <typename tx> +tx ftemplate(int n) { + tx a = 0; + tx b[10]; + +#pragma omp target reduction(+ \ + : a, b) // expected-note {{defined as threadprivate or thread local}} + { + int e; // expected-note {{defined as threadprivate or thread local}} +#pragma omp parallel shared(a, e) // expected-error 2 {{threadprivate or thread local variable cannot be shared}} + a += 1; + b[2] += 1; + } + + return a; +} + +static int fstatic(int n) { + int a = 0; + char aaa = 0; + int b[10]; + +#pragma omp target firstprivate(a, aaa, b) + { + a += 1; + aaa += 1; + b[2] += 1; + } + + return a; +} + +struct S1 { + double a; + + int r1(int n) { + int b = n + 1; + +#pragma omp target firstprivate(b) // expected-note {{defined as threadprivate or thread local}} + { + int c; // expected-note {{defined as threadprivate or thread local}} +#pragma omp parallel shared(b, c) // expected-error 2 {{threadprivate or thread local variable cannot be shared}} + this->a = (double)b + 1.5; + } + + return (int)b; + } +}; + +int bar(int n, double *ptr) { + int a = 0; + a += foo(n, ptr); + S1 S; + a += S.r1(n); + a += fstatic(n); + a += ftemplate<int>(n); // expected-note {{in instantiation of function template specialization 'ftemplate<int>' requested here}} + + return a; +} + diff --git a/test/OpenMP/nvptx_target_exceptions_messages.cpp b/test/OpenMP/nvptx_target_exceptions_messages.cpp new file mode 100644 index 0000000000000..15c9522540fc6 --- /dev/null +++ b/test/OpenMP/nvptx_target_exceptions_messages.cpp @@ -0,0 +1,71 @@ +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc -fexceptions -fcxx-exceptions +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - -fexceptions -fcxx-exceptions -ferror-limit 100 + +#ifndef HEADER +#define HEADER + +template <typename T> +class TemplateClass { + T a; +public: + TemplateClass() { throw 1;} + T f_method() const { return a; } +}; + +int foo(); + +int baz1(); + +int baz2(); + +int baz4() { return 5; } + +template <typename T> +T FA() { + TemplateClass<T> s; + return s.f_method(); +} + +#pragma omp declare target +struct S { + int a; + S(int a) : a(a) { throw 1; } // expected-error {{cannot use 'throw' with exceptions disabled}} +}; + +int foo() { return 0; } +int b = 15; +int d; +#pragma omp end declare target +int c; + +int bar() { return 1 + foo() + bar() + baz1() + baz2(); } + +int maini1() { + int a; + static long aa = 32; + try { +#pragma omp target map(tofrom \ + : a, b) + { + S s(a); + static long aaa = 23; + a = foo() + bar() + b + c + d + aa + aaa + FA<int>(); + if (!a) + throw "Error"; // expected-error {{cannot use 'throw' with exceptions disabled}} + } + } catch(...) { + } + return baz4(); +} + +int baz3() { return 2 + baz2(); } +int baz2() { +#pragma omp target + try { // expected-error {{cannot use 'try' with exceptions disabled}} + ++c; + } catch (...) { + } + return 2 + baz3(); +} + +#endif // HEADER diff --git a/test/OpenMP/nvptx_target_firstprivate_codegen.cpp b/test/OpenMP/nvptx_target_firstprivate_codegen.cpp index f750be03e0523..53b0f758852c2 100644 --- a/test/OpenMP/nvptx_target_firstprivate_codegen.cpp +++ b/test/OpenMP/nvptx_target_firstprivate_codegen.cpp @@ -72,17 +72,17 @@ int foo(int n, double *ptr) { // firstprivate(b): memcpy(b_priv,b_in) // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x float]* [[B_PRIV]] to i8* // TCHECK: [[B_ADDR_REF_BCAST:%.+]] = bitcast [10 x float]* [[B_ADDR_REF]] to i8* - // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[B_PRIV_BCAST]], i8* [[B_ADDR_REF_BCAST]], {{.+}}) + // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_ADDR_REF_BCAST]], {{.+}}) // firstprivate(c) // TCHECK: [[C_PRIV_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C_PRIV]] to i8* // TCHECK: [[C_IN_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C_ADDR_REF]] to i8* - // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[C_PRIV_BCAST]], i8* [[C_IN_BCAST]],{{.+}}) + // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[C_PRIV_BCAST]], i8* align {{[0-9]+}} [[C_IN_BCAST]],{{.+}}) // firstprivate(d) // TCHECK: [[D_PRIV_BCAST:%.+]] = bitcast [[TT]]* [[D_PRIV]] to i8* // TCHECK: [[D_IN_BCAST:%.+]] = bitcast [[TT]]* [[D_ADDR_REF]] to i8* - // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[D_PRIV_BCAST]], i8* [[D_IN_BCAST]],{{.+}}) + // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[D_PRIV_BCAST]], i8* align {{[0-9]+}} [[D_IN_BCAST]],{{.+}}) // TCHECK: load i16, i16* [[A2_ADDR]], @@ -91,7 +91,7 @@ int foo(int n, double *ptr) { ptr[0]++; } - // TCHECK: define void @__omp_offloading_{{.+}}(double* [[PTR_IN:%.+]]) + // TCHECK: define weak void @__omp_offloading_{{.+}}(double* [[PTR_IN:%.+]]) // TCHECK: [[PTR_ADDR:%.+]] = alloca double*, // TCHECK-NOT: alloca double*, // TCHECK: store double* [[PTR_IN]], double** [[PTR_ADDR]], @@ -151,7 +151,7 @@ static int fstatic(int n) { // firstprivate(b) // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8* // TCHECK: [[B_IN_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_ADDR_REF]] to i8* -// TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[B_PRIV_BCAST]], i8* [[B_IN_BCAST]],{{.+}}) +// TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_IN_BCAST]],{{.+}}) // TCHECK: ret void @@ -214,8 +214,14 @@ int bar(int n, double *ptr) { // firstprivate(b) // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8* // TCHECK: [[B_IN_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_ADDR_REF]] to i8* -// TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[B_PRIV_BCAST]], i8* [[B_IN_BCAST]],{{.+}}) +// TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_IN_BCAST]],{{.+}}) // TCHECK: ret void #endif + +// TCHECK-DAG: distinct !DISubprogram(linkageName: "__omp_offloading_{{.+}}_worker", +// TCHECK-DAG: distinct !DISubprogram(linkageName: "__omp_offloading_{{.+}}_worker", +// TCHECK-DAG: distinct !DISubprogram(linkageName: "__omp_offloading_{{.+}}_worker", +// TCHECK-DAG: distinct !DISubprogram(linkageName: "__omp_offloading_{{.+}}_worker", +// TCHECK-DAG: distinct !DISubprogram(linkageName: "__omp_offloading_{{.+}}_worker", diff --git a/test/OpenMP/nvptx_target_parallel_codegen.cpp b/test/OpenMP/nvptx_target_parallel_codegen.cpp index 7d16624358424..15611f6475880 100644 --- a/test/OpenMP/nvptx_target_parallel_codegen.cpp +++ b/test/OpenMP/nvptx_target_parallel_codegen.cpp @@ -1,9 +1,9 @@ // Test target codegen - host bc file has to be created first. -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -55,14 +55,16 @@ int bar(int n){ // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l26}}( // CHECK: [[AA_ADDR:%.+]] = alloca i16*, align + // CHECK-NOT: call i8* @__kmpc_data_sharing_push_stack // CHECK: store i16* {{%.+}}, i16** [[AA_ADDR]], align // CHECK: [[AA:%.+]] = load i16*, i16** [[AA_ADDR]], align // CHECK: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], + // CHECK: call void @__kmpc_data_sharing_init_stack_spmd // CHECK: br label {{%?}}[[EXEC:.+]] // // CHECK: [[EXEC]] - // CHECK: {{call|invoke}} void [[OP1:@.+]](i32* null, i32* null, i16* [[AA]]) + // CHECK: {{call|invoke}} void [[OP1:@.+]]({{.+}}, {{.+}}, i16* [[AA]]) // CHECK: br label {{%?}}[[DONE:.+]] // // CHECK: [[DONE]] @@ -101,10 +103,11 @@ int bar(int n){ // CHECK: [[B:%.+]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align // CHECK: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], + // CHECK: call void @__kmpc_data_sharing_init_stack_spmd // CHECK: br label {{%?}}[[EXEC:.+]] // // CHECK: [[EXEC]] - // CHECK: {{call|invoke}} void [[OP2:@.+]](i32* null, i32* null, i32* [[A]], i16* [[AA]], [10 x i32]* [[B]]) + // CHECK: {{call|invoke}} void [[OP2:@.+]]({{.+}}, {{.+}}, i32* [[A]], i16* [[AA]], [10 x i32]* [[B]]) // CHECK: br label {{%?}}[[DONE:.+]] // // CHECK: [[DONE]] diff --git a/test/OpenMP/nvptx_target_parallel_num_threads_codegen.cpp b/test/OpenMP/nvptx_target_parallel_num_threads_codegen.cpp index bc423c1ee605f..13a7fb289dad4 100644 --- a/test/OpenMP/nvptx_target_parallel_num_threads_codegen.cpp +++ b/test/OpenMP/nvptx_target_parallel_num_threads_codegen.cpp @@ -1,16 +1,16 @@ // Test target codegen - host bc file has to be created first. -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // expected-no-diagnostics #ifndef HEADER #define HEADER -// Check that the execution mode of all 2 target regions on the gpu is set to SPMD Mode. -// CHECK-DAG: {{@__omp_offloading_.+l21}}_exec_mode = weak constant i8 0 -// CHECK-DAG: {{@__omp_offloading_.+l26}}_exec_mode = weak constant i8 0 +// Check that the execution mode of all 2 target regions on the gpu is set to non-SPMD Mode. +// CHECK-DAG: {{@__omp_offloading_.+l21}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l26}}_exec_mode = weak constant i8 1 template<typename tx> tx ftemplate(int n) { @@ -46,23 +46,13 @@ int bar(int n){ // CHECK: store i16* {{%.+}}, i16** [[AA_ADDR]], align // CHECK: [[AA:%.+]] = load i16*, i16** [[AA_ADDR]], align // CHECK: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() - // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], - // CHECK: br label {{%?}}[[EXEC:.+]] - // - // CHECK: [[EXEC]] - // CHECK-NOT: call void @__kmpc_push_num_threads - // CHECK: {{call|invoke}} void [[OP1:@.+]](i32* null, i32* null, i16* [[AA]]) - // CHECK: br label {{%?}}[[DONE:.+]] - // - // CHECK: [[DONE]] - // CHECK: call void @__kmpc_spmd_kernel_deinit() - // CHECK: br label {{%?}}[[EXIT:.+]] - // - // CHECK: [[EXIT]] + // CHECK: call void @__kmpc_kernel_init(i32 + // CHECK: call void @__kmpc_push_num_threads + // CHECK: call void @__kmpc_kernel_deinit(i16 1) // CHECK: ret void // CHECK: } - // CHECK: define internal void [[OP1]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i16* {{[^%]*}}[[ARG:%.+]]) + // CHECK: define internal void @{{.+}}(i32* noalias %{{.+}}, i32* noalias %{{.+}}, i16* {{[^%]*}}[[ARG:%.+]]) // CHECK: = alloca i32*, align // CHECK: = alloca i32*, align // CHECK: [[AA_ADDR:%.+]] = alloca i16*, align @@ -89,23 +79,13 @@ int bar(int n){ // CHECK: [[AA:%.+]] = load i16*, i16** [[AA_ADDR]], align // CHECK: [[B:%.+]] = load [10 x i32]*, [10 x i32]** [[B_ADDR]], align // CHECK: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() - // CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], - // CHECK: br label {{%?}}[[EXEC:.+]] - // - // CHECK: [[EXEC]] - // CHECK-NOT: call void @__kmpc_push_num_threads - // CHECK: {{call|invoke}} void [[OP2:@.+]](i32* null, i32* null, i32* [[A]], i16* [[AA]], [10 x i32]* [[B]]) - // CHECK: br label {{%?}}[[DONE:.+]] - // - // CHECK: [[DONE]] - // CHECK: call void @__kmpc_spmd_kernel_deinit() - // CHECK: br label {{%?}}[[EXIT:.+]] - // - // CHECK: [[EXIT]] + // CHECK: call void @__kmpc_kernel_init(i32 + // CHECK: call void @__kmpc_push_num_threads + // CHECK: call void @__kmpc_kernel_deinit(i16 1) // CHECK: ret void // CHECK: } - // CHECK: define internal void [[OP2]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i32* {{[^%]*}}[[ARG1:%.+]], i16* {{[^%]*}}[[ARG2:%.+]], [10 x i32]* {{[^%]*}}[[ARG3:%.+]]) + // CHECK: define internal void @{{.+}}(i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* {{[^%]*}}[[ARG1:%.+]], i16* {{[^%]*}}[[ARG2:%.+]], [10 x i32]* {{[^%]*}}[[ARG3:%.+]]) // CHECK: = alloca i32*, align // CHECK: = alloca i32*, align // CHECK: [[A_ADDR:%.+]] = alloca i32*, align diff --git a/test/OpenMP/nvptx_target_parallel_proc_bind_codegen.cpp b/test/OpenMP/nvptx_target_parallel_proc_bind_codegen.cpp index 91c6de1b85d34..1cdcb70417b82 100644 --- a/test/OpenMP/nvptx_target_parallel_proc_bind_codegen.cpp +++ b/test/OpenMP/nvptx_target_parallel_proc_bind_codegen.cpp @@ -1,9 +1,9 @@ // Test target codegen - host bc file has to be created first. -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -48,11 +48,12 @@ int bar(int n){ // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l22}}( // CHECK: call void @__kmpc_spmd_kernel_init( + // CHECK: call void @__kmpc_data_sharing_init_stack_spmd // CHECK: br label {{%?}}[[EXEC:.+]] // // CHECK: [[EXEC]] // CHECK-NOT: call void @__kmpc_push_proc_bind - // CHECK: {{call|invoke}} void [[OP1:@.+]](i32* null, i32* null + // CHECK: {{call|invoke}} void [[OP1:@.+]]( // CHECK: br label {{%?}}[[DONE:.+]] // // CHECK: [[DONE]] @@ -69,11 +70,12 @@ int bar(int n){ // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l26}}( // CHECK: call void @__kmpc_spmd_kernel_init( + // CHECK: call void @__kmpc_data_sharing_init_stack_spmd // CHECK: br label {{%?}}[[EXEC:.+]] // // CHECK: [[EXEC]] // CHECK-NOT: call void @__kmpc_push_proc_bind - // CHECK: {{call|invoke}} void [[OP1:@.+]](i32* null, i32* null + // CHECK: {{call|invoke}} void [[OP1:@.+]]( // CHECK: br label {{%?}}[[DONE:.+]] // // CHECK: [[DONE]] @@ -89,11 +91,12 @@ int bar(int n){ // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l31}}( // CHECK: call void @__kmpc_spmd_kernel_init( + // CHECK: call void @__kmpc_data_sharing_init_stack_spmd // CHECK: br label {{%?}}[[EXEC:.+]] // // CHECK: [[EXEC]] // CHECK-NOT: call void @__kmpc_push_proc_bind - // CHECK: {{call|invoke}} void [[OP1:@.+]](i32* null, i32* null + // CHECK: {{call|invoke}} void [[OP1:@.+]]( // CHECK: br label {{%?}}[[DONE:.+]] // // CHECK: [[DONE]] diff --git a/test/OpenMP/nvptx_target_parallel_reduction_codegen.cpp b/test/OpenMP/nvptx_target_parallel_reduction_codegen.cpp index c4c3e977b0e24..02676272b0fed 100644 --- a/test/OpenMP/nvptx_target_parallel_reduction_codegen.cpp +++ b/test/OpenMP/nvptx_target_parallel_reduction_codegen.cpp @@ -1,9 +1,9 @@ // Test target codegen - host bc file has to be created first. -// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc -// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc -// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -verify -fopenmp -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -24,18 +24,18 @@ tx ftemplate(int n) { float d; double e; - #pragma omp target parallel reduction(+: e) map(tofrom: e) + #pragma omp target parallel reduction(+: e) { e += 5; } - #pragma omp target parallel reduction(^: c) reduction(*: d) map(tofrom: c,d) + #pragma omp target parallel reduction(^: c) reduction(*: d) { c ^= 2; d *= 33; } - #pragma omp target parallel reduction(|: a) reduction(max: b) map(tofrom: a,b) + #pragma omp target parallel reduction(|: a) reduction(max: b) { a |= 1; b = 99 > b ? 99 : b; @@ -55,6 +55,7 @@ int bar(int n){ // CHECK: define {{.*}}void {{@__omp_offloading_.+template.+l27}}( // // CHECK: call void @__kmpc_spmd_kernel_init( + // CHECK: call void @__kmpc_data_sharing_init_stack_spmd // CHECK: br label {{%?}}[[EXECUTE:.+]] // // CHECK: [[EXECUTE]] @@ -117,15 +118,15 @@ int bar(int n){ // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[REMOTE_ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[REMOTE_RED_LIST:%.+]], i{{32|64}} 0, i{{32|64}} 0 // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to double* - // CHECK: [[ELT_VAL:%.+]] = load double, double* [[ELT]], align // - // CHECK: [[ELT_CAST:%.+]] = bitcast double [[ELT_VAL]] to i64 + // CHECK: [[ELT_CAST:%.+]] = bitcast double* [[ELT]] to i64* + // CHECK: [[REMOTE_ELT_CAST:%.+]] = bitcast double* [[REMOTE_ELT]] to i64* + // CHECK: [[ELT_VAL:%.+]] = load i64, i64* [[ELT_CAST]], align // CHECK: [[WS32:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() // CHECK: [[WS:%.+]] = trunc i32 [[WS32]] to i16 - // CHECK: [[REMOTE_ELT_VAL64:%.+]] = call i64 @__kmpc_shuffle_int64(i64 [[ELT_CAST]], i16 [[LANEOFFSET]], i16 [[WS]]) - // CHECK: [[REMOTE_ELT_VAL:%.+]] = bitcast i64 [[REMOTE_ELT_VAL64]] to double + // CHECK: [[REMOTE_ELT_VAL64:%.+]] = call i64 @__kmpc_shuffle_int64(i64 [[ELT_VAL]], i16 [[LANEOFFSET]], i16 [[WS]]) // - // CHECK: store double [[REMOTE_ELT_VAL]], double* [[REMOTE_ELT]], align + // CHECK: store i64 [[REMOTE_ELT_VAL64]], i64* [[REMOTE_ELT_CAST]], align // CHECK: [[REMOTE_ELT_VOID:%.+]] = bitcast double* [[REMOTE_ELT]] to i8* // CHECK: store i8* [[REMOTE_ELT_VOID]], i8** [[REMOTE_ELT_REF]], align // @@ -168,8 +169,8 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_VOID:%.+]] = load i8*, i8** [[REMOTE_ELT_REF]], // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i{{32|64}} 0, i{{32|64}} 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], - // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to double* // CHECK: [[REMOTE_ELT:%.+]] = bitcast i8* [[REMOTE_ELT_VOID]] to double* + // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to double* // CHECK: [[REMOTE_ELT_VAL:%.+]] = load double, double* [[REMOTE_ELT]], align // CHECK: store double [[REMOTE_ELT_VAL]], double* [[ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] @@ -193,10 +194,10 @@ int bar(int n){ // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i{{32|64}} 0, i{{32|64}} 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to double* - // CHECK: [[ELT_VAL:%.+]] = load double, double* [[ELT]], align // // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[WARPID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to double addrspace([[SHARED_ADDRSPACE]])* + // CHECK: [[ELT_VAL:%.+]] = load double, double* [[ELT]], align // CHECK: store double [[ELT_VAL]], double addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] // @@ -216,10 +217,10 @@ int bar(int n){ // CHECK: [[DO_READ]] // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[TID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to double addrspace([[SHARED_ADDRSPACE]])* - // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load double, double addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST:%.+]], i{{32|64}} 0, i{{32|64}} 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to double* + // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load double, double addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: store double [[MEDIUM_ELT_VAL]], double* [[ELT]], align // CHECK: br label {{%?}}[[READ_CONT:.+]] // @@ -242,6 +243,7 @@ int bar(int n){ // CHECK: define {{.*}}void {{@__omp_offloading_.+template.+l32}}( // // CHECK: call void @__kmpc_spmd_kernel_init( + // CHECK: call void @__kmpc_data_sharing_init_stack_spmd // CHECK: br label {{%?}}[[EXECUTE:.+]] // // CHECK: [[EXECUTE]] @@ -347,15 +349,15 @@ int bar(int n){ // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[REMOTE_ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[REMOTE_RED_LIST]], i{{32|64}} 0, i{{32|64}} 1 // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to float* - // CHECK: [[ELT_VAL:%.+]] = load float, float* [[ELT]], align // - // CHECK: [[ELT_CAST:%.+]] = bitcast float [[ELT_VAL]] to i32 + // CHECK: [[ELT_CAST:%.+]] = bitcast float* [[ELT]] to i32* + // CHECK: [[REMOTE_ELT2_CAST:%.+]] = bitcast float* [[REMOTE_ELT2]] to i32* + // CHECK: [[ELT_VAL:%.+]] = load i32, i32* [[ELT_CAST]], align // CHECK: [[WS32:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() // CHECK: [[WS:%.+]] = trunc i32 [[WS32]] to i16 - // CHECK: [[REMOTE_ELT2_VAL32:%.+]] = call i32 @__kmpc_shuffle_int32(i32 [[ELT_CAST]], i16 [[LANEOFFSET]], i16 [[WS]]) - // CHECK: [[REMOTE_ELT2_VAL:%.+]] = bitcast i32 [[REMOTE_ELT2_VAL32]] to float + // CHECK: [[REMOTE_ELT2_VAL32:%.+]] = call i32 @__kmpc_shuffle_int32(i32 [[ELT_VAL]], i16 [[LANEOFFSET]], i16 [[WS]]) // - // CHECK: store float [[REMOTE_ELT2_VAL]], float* [[REMOTE_ELT2]], align + // CHECK: store i32 [[REMOTE_ELT2_VAL32]], i32* [[REMOTE_ELT2_CAST]], align // CHECK: [[REMOTE_ELT2C:%.+]] = bitcast float* [[REMOTE_ELT2]] to i8* // CHECK: store i8* [[REMOTE_ELT2C]], i8** [[REMOTE_ELT_REF]], align // @@ -405,8 +407,8 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_VOID:%.+]] = load i8*, i8** [[REMOTE_ELT_REF]], // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i{{32|64}} 0, i{{32|64}} 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], - // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to float* // CHECK: [[REMOTE_ELT:%.+]] = bitcast i8* [[REMOTE_ELT_VOID]] to float* + // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to float* // CHECK: [[REMOTE_ELT_VAL:%.+]] = load float, float* [[REMOTE_ELT]], align // CHECK: store float [[REMOTE_ELT_VAL]], float* [[ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] @@ -429,10 +431,10 @@ int bar(int n){ // [[DO_COPY]] // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i{{32|64}} 0, i{{32|64}} 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], - // CHECK: [[ELT_VAL:%.+]] = load i8, i8* [[ELT_VOID]], align // // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[WARPID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to i8 addrspace([[SHARED_ADDRSPACE]])* + // CHECK: [[ELT_VAL:%.+]] = load i8, i8* [[ELT_VOID]], align // CHECK: store i8 [[ELT_VAL]], i8 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] // @@ -452,9 +454,9 @@ int bar(int n){ // CHECK: [[DO_READ]] // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[TID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to i8 addrspace([[SHARED_ADDRSPACE]])* - // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load i8, i8 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST:%.+]], i{{32|64}} 0, i{{32|64}} 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], + // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load i8, i8 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: store i8 [[MEDIUM_ELT_VAL]], i8* [[ELT_VOID]], align // CHECK: br label {{%?}}[[READ_CONT:.+]] // @@ -470,10 +472,10 @@ int bar(int n){ // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i{{32|64}} 0, i{{32|64}} 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to float* - // CHECK: [[ELT_VAL:%.+]] = load float, float* [[ELT]], align // // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[WARPID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to float addrspace([[SHARED_ADDRSPACE]])* + // CHECK: [[ELT_VAL:%.+]] = load float, float* [[ELT]], align // CHECK: store float [[ELT_VAL]], float addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] // @@ -493,10 +495,10 @@ int bar(int n){ // CHECK: [[DO_READ]] // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[TID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to float addrspace([[SHARED_ADDRSPACE]])* - // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load float, float addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST:%.+]], i{{32|64}} 0, i{{32|64}} 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to float* + // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load float, float addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: store float [[MEDIUM_ELT_VAL]], float* [[ELT]], align // CHECK: br label {{%?}}[[READ_CONT:.+]] // @@ -519,6 +521,7 @@ int bar(int n){ // CHECK: define {{.*}}void {{@__omp_offloading_.+template.+l38}}( // // CHECK: call void @__kmpc_spmd_kernel_init( + // CHECK: call void @__kmpc_data_sharing_init_stack_spmd // CHECK: br label {{%?}}[[EXECUTE:.+]] // // CHECK: [[EXECUTE]] @@ -714,8 +717,8 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_VOID:%.+]] = load i8*, i8** [[REMOTE_ELT_REF]], // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i{{32|64}} 0, i{{32|64}} 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], - // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i32* // CHECK: [[REMOTE_ELT:%.+]] = bitcast i8* [[REMOTE_ELT_VOID]] to i32* + // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i32* // CHECK: [[REMOTE_ELT_VAL:%.+]] = load i32, i32* [[REMOTE_ELT]], align // CHECK: store i32 [[REMOTE_ELT_VAL]], i32* [[ELT]], align // @@ -723,8 +726,8 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_VOID:%.+]] = load i8*, i8** [[REMOTE_ELT_REF]], // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i{{32|64}} 0, i{{32|64}} 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], - // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i16* // CHECK: [[REMOTE_ELT:%.+]] = bitcast i8* [[REMOTE_ELT_VOID]] to i16* + // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i16* // CHECK: [[REMOTE_ELT_VAL:%.+]] = load i16, i16* [[REMOTE_ELT]], align // CHECK: store i16 [[REMOTE_ELT_VAL]], i16* [[ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] @@ -748,10 +751,10 @@ int bar(int n){ // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i{{32|64}} 0, i{{32|64}} 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i32* - // CHECK: [[ELT_VAL:%.+]] = load i32, i32* [[ELT]], align // // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[WARPID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to i32 addrspace([[SHARED_ADDRSPACE]])* + // CHECK: [[ELT_VAL:%.+]] = load i32, i32* [[ELT]], align // CHECK: store i32 [[ELT_VAL]], i32 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] // @@ -771,10 +774,10 @@ int bar(int n){ // CHECK: [[DO_READ]] // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[TID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to i32 addrspace([[SHARED_ADDRSPACE]])* - // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load i32, i32 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST:%.+]], i{{32|64}} 0, i{{32|64}} 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i32* + // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load i32, i32 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: store i32 [[MEDIUM_ELT_VAL]], i32* [[ELT]], align // CHECK: br label {{%?}}[[READ_CONT:.+]] // @@ -790,10 +793,10 @@ int bar(int n){ // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i{{32|64}} 0, i{{32|64}} 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i16* - // CHECK: [[ELT_VAL:%.+]] = load i16, i16* [[ELT]], align // // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[WARPID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to i16 addrspace([[SHARED_ADDRSPACE]])* + // CHECK: [[ELT_VAL:%.+]] = load i16, i16* [[ELT]], align // CHECK: store i16 [[ELT_VAL]], i16 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] // @@ -813,10 +816,10 @@ int bar(int n){ // CHECK: [[DO_READ]] // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[TID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to i16 addrspace([[SHARED_ADDRSPACE]])* - // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load i16, i16 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST:%.+]], i{{32|64}} 0, i{{32|64}} 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i16* + // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load i16, i16 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: store i16 [[MEDIUM_ELT_VAL]], i16* [[ELT]], align // CHECK: br label {{%?}}[[READ_CONT:.+]] // diff --git a/test/OpenMP/nvptx_target_printf_codegen.c b/test/OpenMP/nvptx_target_printf_codegen.c index 0de6885a5a720..e7bfb874f4efe 100644 --- a/test/OpenMP/nvptx_target_printf_codegen.c +++ b/test/OpenMP/nvptx_target_printf_codegen.c @@ -3,12 +3,8 @@ // RUN: %clang_cc1 -verify -fopenmp -x c -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc // RUN: %clang_cc1 -verify -fopenmp -x c -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 - -#include <stdarg.h> - // expected-no-diagnostics extern int printf(const char *, ...); -extern int vprintf(const char *, va_list); // Check a simple call to printf end-to-end. // CHECK: [[SIMPLE_PRINTF_TY:%[a-zA-Z0-9_]+]] = type { i32, i64, double } @@ -33,7 +29,7 @@ int CheckSimple() { // CHECK: [[MASTER]] // CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[MTMP1:%.+]] = sub i32 [[MNTH]], [[MWS]] + // CHECK: [[MTMP1:%.+]] = sub nuw i32 [[MNTH]], [[MWS]] // CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] // printf in master-only basic block. @@ -73,7 +69,7 @@ void CheckNoArgs() { // CHECK: [[MASTER]] // CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[MTMP1:%.+]] = sub i32 [[MNTH]], [[MWS]] + // CHECK: [[MTMP1:%.+]] = sub nuw i32 [[MNTH]], [[MWS]] // CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] // printf in master-only basic block. @@ -106,7 +102,7 @@ void CheckAllocaIsInEntryBlock() { // CHECK: [[MASTER]] // CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[MTMP1:%.+]] = sub i32 [[MNTH]], [[MWS]] + // CHECK: [[MTMP1:%.+]] = sub nuw i32 [[MNTH]], [[MWS]] // CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] if (foo) { diff --git a/test/OpenMP/nvptx_target_simd_codegen.cpp b/test/OpenMP/nvptx_target_simd_codegen.cpp new file mode 100644 index 0000000000000..89ea173addcc9 --- /dev/null +++ b/test/OpenMP/nvptx_target_simd_codegen.cpp @@ -0,0 +1,87 @@ +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// Check that the execution mode of all 2 target regions on the gpu is set to NonSPMD Mode. +// CHECK-DAG: {{@__omp_offloading_.+l25}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l30}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l35}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l40}}_exec_mode = weak constant i8 1 + +#define N 1000 + +template<typename tx> +tx ftemplate(int n) { + tx a[N]; + short aa[N]; + tx b[10]; + + #pragma omp target simd + for(int i = 0; i < n; i++) { + a[i] = 1; + } + + #pragma omp target simd + for (int i = 0; i < n; i++) { + aa[i] += 1; + } + + #pragma omp target simd + for(int i = 0; i < 10; i++) { + b[i] += 1; + } + + #pragma omp target simd reduction(+:n) + for(int i = 0; i < 10; i++) { + b[i] += 1; + } + + return a[0]; +} + +int bar(int n){ + int a = 0; + + a += ftemplate<int>(n); + + return a; +} + +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+l25}}( +// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1) +// CHECK-NOT: call void @__kmpc_for_static_init +// CHECK-NOT: call void @__kmpc_for_static_fini +// CHECK: call void @__kmpc_kernel_deinit(i16 1) +// CHECK: ret void + +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+l30}}( +// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1) +// CHECK-NOT: call void @__kmpc_for_static_init +// CHECK-NOT: call void @__kmpc_for_static_fini +// CHECK: call void @__kmpc_kernel_deinit(i16 1) +// CHECK: ret void + +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+l35}}( +// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1) +// CHECK-NOT: call void @__kmpc_for_static_init +// CHECK-NOT: call void @__kmpc_for_static_fini +// CHECK: call void @__kmpc_kernel_deinit(i16 1) +// CHECK: ret void + +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+l40}}( +// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1) +// CHECK-NOT: call void @__kmpc_for_static_init +// CHECK-NOT: call void @__kmpc_for_static_fini +// CHECK-NOT: call i32 @__kmpc_nvptx_simd_reduce_nowait( +// CHECK-NOT: call void @__kmpc_nvptx_end_reduce_nowait( +// CHECK: call void @__kmpc_kernel_deinit(i16 1) +// CHECK: ret void + + +#endif diff --git a/test/OpenMP/nvptx_target_teams_codegen.cpp b/test/OpenMP/nvptx_target_teams_codegen.cpp index 8b016cc307d58..c7667d83b9aa9 100644 --- a/test/OpenMP/nvptx_target_teams_codegen.cpp +++ b/test/OpenMP/nvptx_target_teams_codegen.cpp @@ -9,8 +9,9 @@ #define HEADER // Check that the execution mode of all 2 target regions is set to Generic Mode. -// CHECK-DAG: {{@__omp_offloading_.+l26}}_exec_mode = weak constant i8 1 -// CHECK-DAG: {{@__omp_offloading_.+l31}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l27}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l32}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l37}}_exec_mode = weak constant i8 0 template<typename tx> tx ftemplate(int n) { @@ -33,6 +34,13 @@ tx ftemplate(int n) { aa = 1; } + #pragma omp target teams + { +#pragma omp parallel +#pragma omp parallel + aa = 1; + } + return a; } @@ -44,14 +52,14 @@ int bar(int n){ return a; } - // CHECK-NOT: define {{.*}}void {{@__omp_offloading_.+template.+l21}}_worker() + // CHECK-NOT: define {{.*}}void {{@__omp_offloading_.+template.+l22}}_worker() - // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l26}}_worker() + // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l27}}_worker() // CHECK-DAG: [[OMP_EXEC_STATUS:%.+]] = alloca i8, // CHECK-DAG: [[OMP_WORK_FN:%.+]] = alloca i8*, // CHECK: store i8* null, i8** [[OMP_WORK_FN]], @@ -60,7 +68,7 @@ int bar(int n){ // // CHECK: [[AWAIT_WORK]] // CHECK: call void @llvm.nvvm.barrier0() - // CHECK: [[KPR:%.+]] = call i1 @__kmpc_kernel_parallel(i8** [[OMP_WORK_FN]], i8*** %shared_args, i16 1) + // CHECK: [[KPR:%.+]] = call i1 @__kmpc_kernel_parallel(i8** [[OMP_WORK_FN]], i16 1) // CHECK: [[KPRB:%.+]] = zext i1 [[KPR]] to i8 // store i8 [[KPRB]], i8* [[OMP_EXEC_STATUS]], align 1 // CHECK: [[WORK:%.+]] = load i8*, i8** [[OMP_WORK_FN]], @@ -86,14 +94,14 @@ int bar(int n){ // CHECK: [[EXIT]] // CHECK: ret void - // CHECK: define {{.*}}void [[T1:@__omp_offloading_.+template.+l26]](i[[SZ:32|64]] [[A:%[^)]+]]) + // CHECK: define {{.*}}void [[T1:@__omp_offloading_.+template.+l27]](i[[SZ:32|64]] [[A:%[^)]+]]) // CHECK: store i[[SZ]] [[A]], i[[SZ]]* [[A_ADDR:%.+]], align // CHECK: [[CONV:%.+]] = bitcast i[[SZ]]* [[A_ADDR]] to i8* // CHECK-DAG: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() // CHECK-DAG: [[NTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[WS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK-DAG: [[TH_LIMIT:%.+]] = sub i32 [[NTH]], [[WS]] + // CHECK-DAG: [[TH_LIMIT:%.+]] = sub nuw i32 [[NTH]], [[WS]] // CHECK: [[IS_WORKER:%.+]] = icmp ult i32 [[TID]], [[TH_LIMIT]] // CHECK: br i1 [[IS_WORKER]], label {{%?}}[[WORKER:.+]], label {{%?}}[[CHECK_MASTER:.+]] // @@ -111,7 +119,7 @@ int bar(int n){ // CHECK: [[MASTER]] // CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[MTMP1:%.+]] = sub i32 [[MNTH]], [[MWS]] + // CHECK: [[MTMP1:%.+]] = sub nuw i32 [[MNTH]], [[MWS]] // CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] // // CHECK-NOT: kmpc_fork_teams @@ -121,9 +129,7 @@ int bar(int n){ // CHECK: [[ACV:%.+]] = load i[[SZ]], i[[SZ]]* [[AC]], align // CHECK: store i[[SZ]] [[ACV]], i[[SZ]]* [[A_ADDR_T:%.+]], align // CHECK: [[CONV2:%.+]] = bitcast i[[SZ]]* [[A_ADDR_T]] to i8* - // CHECK: [[LD_CONV2:%.+]] = load i8, i8* [[CONV2]], - // CHECK: store i8 [[LD_CONV2]], i8* [[A_PRIV:%[^,]+]], - // CHECK: store i8 49, i8* [[A_PRIV]], align + // CHECK: store i8 49, i8* [[CONV2]], align // CHECK: br label {{%?}}[[TERMINATE:.+]] // // CHECK: [[TERMINATE]] @@ -139,7 +145,7 @@ int bar(int n){ - // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l31}}_worker() + // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l32}}_worker() // CHECK-DAG: [[OMP_EXEC_STATUS:%.+]] = alloca i8, // CHECK-DAG: [[OMP_WORK_FN:%.+]] = alloca i8*, // CHECK: store i8* null, i8** [[OMP_WORK_FN]], @@ -148,7 +154,7 @@ int bar(int n){ // // CHECK: [[AWAIT_WORK]] // CHECK: call void @llvm.nvvm.barrier0() - // CHECK: [[KPR:%.+]] = call i1 @__kmpc_kernel_parallel(i8** [[OMP_WORK_FN]], i8*** %shared_args, i16 1) + // CHECK: [[KPR:%.+]] = call i1 @__kmpc_kernel_parallel(i8** [[OMP_WORK_FN]], i16 1) // CHECK: [[KPRB:%.+]] = zext i1 [[KPR]] to i8 // store i8 [[KPRB]], i8* [[OMP_EXEC_STATUS]], align 1 // CHECK: [[WORK:%.+]] = load i8*, i8** [[OMP_WORK_FN]], @@ -174,14 +180,14 @@ int bar(int n){ // CHECK: [[EXIT]] // CHECK: ret void - // CHECK: define {{.*}}void [[T2:@__omp_offloading_.+template.+l31]](i[[SZ:32|64]] [[AA:%[^)]+]]) + // CHECK: define {{.*}}void [[T2:@__omp_offloading_.+template.+l32]](i[[SZ:32|64]] [[AA:%[^)]+]]) // CHECK: store i[[SZ]] [[AA]], i[[SZ]]* [[AA_ADDR:%.+]], align // CHECK: [[CONV:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i16* // CHECK-DAG: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x() // CHECK-DAG: [[NTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[WS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK-DAG: [[TH_LIMIT:%.+]] = sub i32 [[NTH]], [[WS]] + // CHECK-DAG: [[TH_LIMIT:%.+]] = sub nuw i32 [[NTH]], [[WS]] // CHECK: [[IS_WORKER:%.+]] = icmp ult i32 [[TID]], [[TH_LIMIT]] // CHECK: br i1 [[IS_WORKER]], label {{%?}}[[WORKER:.+]], label {{%?}}[[CHECK_MASTER:.+]] // @@ -199,7 +205,7 @@ int bar(int n){ // CHECK: [[MASTER]] // CHECK-DAG: [[MNTH:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() // CHECK-DAG: [[MWS:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() - // CHECK: [[MTMP1:%.+]] = sub i32 [[MNTH]], [[MWS]] + // CHECK: [[MTMP1:%.+]] = sub nuw i32 [[MNTH]], [[MWS]] // CHECK: call void @__kmpc_kernel_init(i32 [[MTMP1]] // // CHECK-NOT: kmpc_fork_teams @@ -209,9 +215,7 @@ int bar(int n){ // CHECK: [[ACV:%.+]] = load i[[SZ]], i[[SZ]]* [[AC]], align // CHECK: store i[[SZ]] [[ACV]], i[[SZ]]* [[AA_ADDR_T:%.+]], align // CHECK: [[CONV2:%.+]] = bitcast i[[SZ]]* [[AA_ADDR_T]] to i16* - // CHECK: [[LD_CONV2:%.+]] = load i16, i16* [[CONV2]], - // CHECK: store i16 [[LD_CONV2]], i16* [[A_PRIV:%[^,]+]], - // CHECK: store i16 1, i16* [[A_PRIV]], align + // CHECK: store i16 1, i16* [[CONV2]], align // CHECK: br label {{%?}}[[TERMINATE:.+]] // // CHECK: [[TERMINATE]] @@ -222,5 +226,25 @@ int bar(int n){ // CHECK: [[EXIT]] // CHECK: ret void +// CHECK: define weak void @__omp_offloading_{{.*}}ftemplate{{.*}}_l37( +// CHECK: call void @__kmpc_spmd_kernel_init( +// CHECK: call void @__kmpc_data_sharing_init_stack_spmd +// CHECK: call i8* @__kmpc_data_sharing_push_stack( +// CHECK-NOT: call void @__kmpc_serialized_parallel( +// CHECK: call void [[L0:@.+]](i32* %{{.+}}, i32* %{{.+}}, i16* %{{.*}}) +// CHECK-NOT: call void @__kmpc_end_serialized_parallel( +// CHECK: call void @__kmpc_data_sharing_pop_stack( +// CHECK: call void @__kmpc_spmd_kernel_deinit() +// CHECK: ret + +// CHECK: define internal void [[L0]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i16* dereferenceable +// CHECK: call void @__kmpc_serialized_parallel( +// CHECK: call void [[L1:@.+]](i32* %{{.+}}, i32* %{{.+}}, i16* %{{.+}}) +// CHECK: call void @__kmpc_end_serialized_parallel( +// CHECK: ret void + +// CHECK: define internal void [[L1]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i16* dereferenceable +// CHECK: store i16 1, i16* % +// CHECK: ret void #endif diff --git a/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp b/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp new file mode 100644 index 0000000000000..f74919fc828d5 --- /dev/null +++ b/test/OpenMP/nvptx_target_teams_distribute_parallel_for_codegen.cpp @@ -0,0 +1,136 @@ +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// Check that the execution mode of all 2 target regions on the gpu is set to SPMD Mode. +// CHECK-DAG: {{@__omp_offloading_.+l32}}_exec_mode = weak constant i8 0 +// CHECK-DAG: {{@__omp_offloading_.+l38}}_exec_mode = weak constant i8 0 +// CHECK-DAG: {{@__omp_offloading_.+l43}}_exec_mode = weak constant i8 0 +// CHECK-DAG: {{@__omp_offloading_.+l48}}_exec_mode = weak constant i8 0 +// CHECK-DAG: {{@__omp_offloading_.+l56}}_exec_mode = weak constant i8 0 + +#define N 1000 +#define M 10 + +template<typename tx> +tx ftemplate(int n) { + tx a[N]; + short aa[N]; + tx b[10]; + tx c[M][M]; + tx f = n; + tx l; + int k; + tx *v; + +#pragma omp target teams distribute parallel for lastprivate(l) dist_schedule(static,128) schedule(static,32) + for(int i = 0; i < n; i++) { + a[i] = 1; + l = i; + } + + #pragma omp target teams distribute parallel for map(tofrom: aa) num_teams(M) thread_limit(64) + for(int i = 0; i < n; i++) { + aa[i] += 1; + } + +#pragma omp target teams distribute parallel for map(tofrom:a, aa, b) if(target: n>40) proc_bind(spread) + for(int i = 0; i < 10; i++) { + b[i] += 1; + } + +#pragma omp target teams distribute parallel for collapse(2) firstprivate(f) private(k) + for(int i = 0; i < M; i++) { + for(int j = 0; j < M; j++) { + k = M; + c[i][j] = i + j * f + k; + } + } + +#pragma omp target teams distribute parallel for map(a, v[:N]) + for(int i = 0; i < n; i++) + a[i] = v[i]; + return a[0]; +} + +int bar(int n){ + int a = 0; + + a += ftemplate<int>(n); + + return a; +} + +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}( +// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], +// CHECK: call void @__kmpc_data_sharing_init_stack_spmd +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 91, +// CHECK: {{call|invoke}} void [[OUTL1:@.+]]( +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: call void @__kmpc_spmd_kernel_deinit() +// CHECK: ret void + +// CHECK: define internal void [[OUTL1]]( +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 33, +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}( +// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], +// CHECK: call void @__kmpc_data_sharing_init_stack_spmd +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92, +// CHECK: {{call|invoke}} void [[OUTL2:@.+]]( +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: call void @__kmpc_spmd_kernel_deinit() +// CHECK: ret void + +// CHECK: define internal void [[OUTL2]]( +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 34, +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}( +// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], +// CHECK: call void @__kmpc_data_sharing_init_stack_spmd +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92, +// CHECK: {{call|invoke}} void [[OUTL3:@.+]]( +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: call void @__kmpc_spmd_kernel_deinit() +// CHECK: ret void + +// CHECK: define internal void [[OUTL3]]( +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 34, +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: define {{.*}}void {{@__omp_offloading_.+}}({{.+}}, i{{32|64}} [[F_IN:%.+]]) +// CHECK: store {{.+}} [[F_IN]], {{.+}}* {{.+}}, +// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], +// CHECK: call void @__kmpc_data_sharing_init_stack_spmd +// CHECK: store {{.+}} 99, {{.+}}* [[COMB_UB:%.+]], align +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92, {{.+}}, {{.+}}, {{.+}}* [[COMB_UB]], +// CHECK: {{call|invoke}} void [[OUTL4:@.+]]( +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: call void @__kmpc_spmd_kernel_deinit() +// CHECK: ret void + +// CHECK: define internal void [[OUTL4]]( +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 34, +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: define weak void @__omp_offloading_{{.*}}_l56(i[[SZ:64|32]] %{{[^,]+}}, [1000 x i32]* dereferenceable{{.*}}, i32* %{{[^)]+}}) +// CHECK: call void [[OUTLINED:@__omp_outlined.*]](i32* %{{.+}}, i32* %{{.+}}, i[[SZ]] %{{.*}}, i[[SZ]] %{{.*}}, i[[SZ]] %{{.*}}, [1000 x i32]* %{{.*}}, i32* %{{.*}}) +// CHECK: define internal void [[OUTLINED]](i32* noalias %{{.*}}, i32* noalias %{{.*}} i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [1000 x i32]* dereferenceable{{.*}}, i32* %{{.*}}) + +#endif diff --git a/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp b/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp new file mode 100644 index 0000000000000..d75a10210905b --- /dev/null +++ b/test/OpenMP/nvptx_target_teams_distribute_parallel_for_generic_mode_codegen.cpp @@ -0,0 +1,66 @@ +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +int a; + +int foo(int *a); + +int main(int argc, char **argv) { +#pragma omp target teams distribute parallel for map(tofrom:a) if(target:argc) schedule(static, a) + for (int i= 0; i < argc; ++i) + a = foo(&i) + foo(&a) + foo(&argc); + return 0; +} + +// CHECK: @__omp_offloading_{{.*}}_main_l16_exec_mode = weak constant i8 0 + +// CHECK: define weak void @__omp_offloading_{{.*}}_main_l16(i{{64|32}} %{{[^,].*}}, i32* dereferenceable{{[^,]*}}, i{{64|32}} %{{[^,)]*}}) +// CHECK: [[TID:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @ +// CHECK: call void @__kmpc_spmd_kernel_init( +// CHECK: call void @__kmpc_for_static_init_4( + +// CHECK: call void [[PARALLEL:@.+]](i32* %{{.*}}, i32* %{{.+}}, i{{64|32}} %{{.+}}, i{{64|32}} %{{.*}}, i{{64|32}} %{{.*}}, i32* %{{.*}}) +// CHECK: br label % + + +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* @ + +// CHECK: call void @__kmpc_spmd_kernel_deinit() + +// CHECK: define internal void [[PARALLEL]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i{{64|32}} %{{.+}}, i{{64|32}} %{{.+}}, i{{64|32}} [[ARGC:%.+]], i32* dereferenceable{{.*}}) +// CHECK-NOT: call i8* @__kmpc_data_sharing_push_stack( +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: [[ARGC_ADDR:%.+]] = alloca i{{32|64}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: [[I:%.+]] = alloca i32, +// CHECK-32: store i32 [[ARGC]], i32* [[ARGC_ADDR]], +// CHECK-64: store i{{64|32}} [[ARGC]], i{{64|32}}* [[ARGC_ADDR]], +// CHECK-64: [[ARGC:%.+]] = bitcast i64* [[ARGC_ADDR]] to i32* + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call i32 [[FOO:@.+foo.+]](i32* [[I]]) +// CHECK: call i32 [[FOO]](i32* %{{.+}}) +// CHECK-32: call i32 [[FOO]](i32* [[ARGC_ADDR]]) +// CHECK-64: call i32 [[FOO]](i32* [[ARGC]]) +// CHECK: call void @__kmpc_for_static_fini( + +// CHECK-NOT: call void @__kmpc_data_sharing_pop_stack( + +#endif diff --git a/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp b/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp new file mode 100644 index 0000000000000..0b945a69b054c --- /dev/null +++ b/test/OpenMP/nvptx_target_teams_distribute_parallel_for_simd_codegen.cpp @@ -0,0 +1,127 @@ +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// Check that the execution mode of all 2 target regions on the gpu is set to SPMD Mode. +// CHECK-DAG: {{@__omp_offloading_.+l30}}_exec_mode = weak constant i8 0 +// CHECK-DAG: {{@__omp_offloading_.+l36}}_exec_mode = weak constant i8 0 +// CHECK-DAG: {{@__omp_offloading_.+l41}}_exec_mode = weak constant i8 0 +// CHECK-DAG: {{@__omp_offloading_.+l46}}_exec_mode = weak constant i8 0 + +#define N 1000 +#define M 10 + +template<typename tx> +tx ftemplate(int n) { + tx a[N]; + short aa[N]; + tx b[10]; + tx c[M][M]; + tx f = n; + tx l; + int k; + +#pragma omp target teams distribute parallel for simd lastprivate(l) dist_schedule(static,128) schedule(static,32) + for(int i = 0; i < n; i++) { + a[i] = 1; + l = i; + } + + #pragma omp target teams distribute parallel for simd map(tofrom: aa) num_teams(M) thread_limit(64) + for(int i = 0; i < n; i++) { + aa[i] += 1; + } + +#pragma omp target teams distribute parallel for simd map(tofrom:a, aa, b) if(target: n>40) proc_bind(spread) + for(int i = 0; i < 10; i++) { + b[i] += 1; + } + +#pragma omp target teams distribute parallel for simd collapse(2) firstprivate(f) private(k) + for(int i = 0; i < M; i++) { + for(int j = 0; j < M; j++) { + k = M; + c[i][j] = i+j*f+k; + } + } + + return a[0]; +} + +int bar(int n){ + int a = 0; + + a += ftemplate<int>(n); + + return a; +} + +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}( +// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], +// CHECK: call void @__kmpc_data_sharing_init_stack_spmd +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 91, +// CHECK: {{call|invoke}} void [[OUTL1:@.+]]( +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: call void @__kmpc_spmd_kernel_deinit() +// CHECK: ret void + +// CHECK: define internal void [[OUTL1]]( +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 33, +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}( +// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], +// CHECK: call void @__kmpc_data_sharing_init_stack_spmd +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92, +// CHECK: {{call|invoke}} void [[OUTL2:@.+]]( +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: call void @__kmpc_spmd_kernel_deinit() +// CHECK: ret void + +// CHECK: define internal void [[OUTL2]]( +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 34, +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}( +// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], +// CHECK: call void @__kmpc_data_sharing_init_stack_spmd +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92, +// CHECK: {{call|invoke}} void [[OUTL3:@.+]]( +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: call void @__kmpc_spmd_kernel_deinit() +// CHECK: ret void + +// CHECK: define internal void [[OUTL3]]( +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 34, +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: define {{.*}}void {{@__omp_offloading_.+}}({{.+}}, i{{32|64}} [[F_IN:%.+]]) +// CHECK: store {{.+}} [[F_IN]], {{.+}}* {{.+}}, +// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() +// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]], +// CHECK: call void @__kmpc_data_sharing_init_stack_spmd +// CHECK: store {{.+}} 99, {{.+}}* [[COMB_UB:%.+]], align +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92, {{.+}}, {{.+}}, {{.+}}* [[COMB_UB]], +// CHECK: {{call|invoke}} void [[OUTL4:@.+]]( +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: call void @__kmpc_spmd_kernel_deinit() +// CHECK: ret void + +// CHECK: define internal void [[OUTL4]]( +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 34, +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +#endif diff --git a/test/OpenMP/nvptx_target_teams_distribute_simd_codegen.cpp b/test/OpenMP/nvptx_target_teams_distribute_simd_codegen.cpp new file mode 100644 index 0000000000000..48f314785c82c --- /dev/null +++ b/test/OpenMP/nvptx_target_teams_distribute_simd_codegen.cpp @@ -0,0 +1,95 @@ +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple i386-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-cuda-mode -fexceptions -fcxx-exceptions -x c++ -triple nvptx-unknown-unknown -fopenmp-targets=nvptx-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// Check that the execution mode of all 2 target regions on the gpu is set to NonSPMD Mode. +// CHECK-DAG: {{@__omp_offloading_.+l30}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l36}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l41}}_exec_mode = weak constant i8 1 +// CHECK-DAG: {{@__omp_offloading_.+l46}}_exec_mode = weak constant i8 1 + +#define N 1000 +#define M 10 + +template<typename tx> +tx ftemplate(int n) { + tx a[N]; + short aa[N]; + tx b[10]; + tx c[M][M]; + tx f = n; + tx l; + int k; + +#pragma omp target teams distribute simd lastprivate(l) dist_schedule(static,128) + for(int i = 0; i < n; i++) { + a[i] = 1; + l = i; + } + + #pragma omp target teams distribute simd map(tofrom: aa) num_teams(M) thread_limit(64) + for(int i = 0; i < n; i++) { + aa[i] += 1; + } + +#pragma omp target teams distribute simd map(tofrom:a, aa, b) if(target: n>40) + for(int i = 0; i < 10; i++) { + b[i] += 1; + } + +#pragma omp target teams distribute simd collapse(2) firstprivate(f) private(k) + for(int i = 0; i < M; i++) { + for(int j = 0; j < M; j++) { + k = M; + c[i][j] = i+j*f+k; + } + } + + return a[0]; +} + +int bar(int n){ + int a = 0; + + a += ftemplate<int>(n); + + return a; +} + +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}_l30( +// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1) +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 91, +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: call void @__kmpc_kernel_deinit(i16 1) +// CHECK: ret void + +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}_l36( +// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1) +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92, +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: call void @__kmpc_kernel_deinit(i16 1) +// CHECK: ret void + +// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}_l41( +// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1) +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92, +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: call void @__kmpc_kernel_deinit(i16 1) +// CHECK: ret void + +// CHECK: define {{.*}}void {{@__omp_offloading_.+}}_l46({{.+}}, i{{32|64}} [[F_IN:%.+]]) +// CHECK: store {{.+}} [[F_IN]], {{.+}}* {{.+}}, +// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1) +// CHECK: store {{.+}} 99, {{.+}}* [[COMB_UB:%.+]], align +// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92, {{.+}}, {{.+}}, {{.+}}* [[COMB_UB]], +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: call void @__kmpc_kernel_deinit(i16 1) +// CHECK: ret void + +#endif diff --git a/test/OpenMP/nvptx_teams_codegen.cpp b/test/OpenMP/nvptx_teams_codegen.cpp index b26d47c706aca..2c50f037b2280 100644 --- a/test/OpenMP/nvptx_teams_codegen.cpp +++ b/test/OpenMP/nvptx_teams_codegen.cpp @@ -36,11 +36,15 @@ int main (int argc, char **argv) { // CK1: store {{.+}} 0, {{.+}}, // CK1: store i{{[0-9]+}} [[ARGC]], i{{[0-9]+}}* [[ARGCADDR]], // CK1-64: [[CONV:%.+]] = bitcast i{{[0-9]+}}* [[ARGCADDR]] to i{{[0-9]+}}* -// CK1-64: store i{{[0-9]+}}* [[CONV]], i{{[0-9]+}}** [[ARGCADDR_PTR]], -// CK1-32: store i{{[0-9]+}}* [[ARGCADDR]], i{{[0-9]+}}** [[ARGCADDR_PTR]], +// CK1: call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} 4, i16 0) +// CK1-64: [[ARG:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[CONV]] +// CK1-32: [[ARG:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[ARGCADDR]] +// CK1: [[ARGCADDR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0 +// CK1: store i{{[0-9]+}} [[ARG]], i{{[0-9]+}}* [[ARGCADDR]], +// CK1: store i{{[0-9]+}}* [[ARGCADDR]], i{{[0-9]+}}** [[ARGCADDR_PTR]], // CK1: [[ARGCADDR_PTR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[ARGCADDR_PTR]], // CK1: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[ARGCADDR_PTR_REF]], -// CK1-NOT: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams( +// CK1-NOT: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams( // CK1: ret void // CK1-NEXT: } @@ -49,10 +53,14 @@ int main (int argc, char **argv) { // CK1: [[ARGCADDR_PTR:%.+]] = alloca i{{.+}}***, // CK1: [[ARGCADDR:%.+]] = alloca i{{.+}}**, // CK1: store i{{.+}}** [[ARGC]], i{{.+}}*** [[ARGCADDR]] +// CK1: call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} {{4|8}}, i16 0) +// CK1: [[ARG:%.+]] = load i{{[0-9]+}}**, i{{[0-9]+}}*** [[ARGCADDR]] +// CK1: [[ARGCADDR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0 +// CK1: store i{{[0-9]+}}** [[ARG]], i{{[0-9]+}}*** [[ARGCADDR]], // CK1: store i8*** [[ARGCADDR]], i8**** [[ARGCADDR_PTR]], // CK1: [[ARGCADDR_PTR_REF:%.+]] = load i{{.+}}**, i{{.+}}*** [[ARGCADDR_PTR]], // CK1: store i{{[0-9]+}}** null, i{{[0-9]+}}*** [[ARGCADDR_PTR_REF]], -// CK1-NOT: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams( +// CK1-NOT: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams( // CK1: ret void // CK1-NEXT: } @@ -97,19 +105,23 @@ int main (int argc, char **argv) { // CK2: [[AADDR:%.+]] = alloca i{{[0-9]+}}, // CK2: [[BADDR:%.+]] = alloca i{{[0-9]+}}, // CK2: [[ARGCADDR:%.+]] = alloca i{{[0-9]+}}, -// CK2-NOT: {{%.+}} = call i32 @__kmpc_global_thread_num( +// CK2: {{%.+}} = call i32 @__kmpc_global_thread_num( // CK2: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[AADDR]], // CK2: store i{{[0-9]+}} [[B_IN]], i{{[0-9]+}}* [[BADDR]], // CK2: store i{{[0-9]+}} [[ARGC_IN]], i{{[0-9]+}}* [[ARGCADDR]], // CK2-64: [[ACONV:%.+]] = bitcast i64* [[AADDR]] to i32* // CK2-64: [[BCONV:%.+]] = bitcast i64* [[BADDR]] to i32* // CK2-64: [[CONV:%.+]] = bitcast i64* [[ARGCADDR]] to i32* -// CK2-64: store i{{[0-9]+}}* [[CONV]], i{{[0-9]+}}** [[ARGCADDR_PTR]], -// CK2-32: store i{{[0-9]+}}* [[ARGCADDR]], i{{[0-9]+}}** [[ARGCADDR_PTR]], +// CK2: call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} 4, i16 0) +// CK2-64: [[ARG:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[CONV]] +// CK2-32: [[ARG:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[ARGCADDR]] +// CK2: [[ARGCADDR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0 +// CK2: store i{{[0-9]+}} [[ARG]], i{{[0-9]+}}* [[ARGCADDR]], +// CK2: store i{{[0-9]+}}* [[ARGCADDR]], i{{[0-9]+}}** [[ARGCADDR_PTR]], // CK2: [[ARGCADDR_PTR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[ARGCADDR_PTR]], // CK2: store i{{[0-9]+}} 0, i{{[0-9]+}}* [[ARGCADDR_PTR_REF]], // CK2-NOT: {{.+}} = call i32 @__kmpc_push_num_teams( -// CK2-NOT: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams( +// CK2-NOT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams( // CK2: ret // CK2: define {{.*}}void @{{[^,]+}}(i{{[0-9]+}} [[A_IN:%.+]], i{{[0-9]+}} [[BP:%.+]], i{{[0-9]+}}** [[ARGC:%.+]]) @@ -117,15 +129,19 @@ int main (int argc, char **argv) { // CK2: [[AADDR:%.+]] = alloca i{{[0-9]+}}, // CK2: [[BADDR:%.+]] = alloca i{{[0-9]+}}, // CK2: [[ARGCADDR:%.+]] = alloca i{{[0-9]+}}**, -// CK2-NOT: {{%.+}} = call i32 @__kmpc_global_thread_num( +// CK2: {{%.+}} = call i32 @__kmpc_global_thread_num( // CK2: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[AADDR]], // CK2: store i{{[0-9]+}} [[B_IN]], i{{[0-9]+}}* [[BADDR]], // CK2: store i{{[0-9]+}}** [[ARGC]], i{{[0-9]+}}*** [[ARGCADDR]], +// CK2: call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} {{4|8}}, i16 0) +// CK2: [[ARG:%.+]] = load i{{[0-9]+}}**, i{{[0-9]+}}*** [[ARGCADDR]] +// CK2: [[ARGCADDR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0 +// CK2: store i{{[0-9]+}}** [[ARG]], i{{[0-9]+}}*** [[ARGCADDR]], // CK2: store i{{[0-9]+}}*** [[ARGCADDR]], i{{[0-9]+}}**** [[ARGCADDR_PTR]], // CK2: [[ARGCADDR_PTR_REF:%.+]] = load i{{[0-9]+}}***, i{{[0-9]+}}**** [[ARGCADDR_PTR]], // CK2: store i{{[0-9]+}}** null, i{{[0-9]+}}*** [[ARGCADDR_PTR_REF]], // CK2-NOT: {{.+}} = call i32 @__kmpc_push_num_teams( -// CK2-NOT: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams( +// CK2-NOT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams( // CK2: ret void #endif // CK2 diff --git a/test/OpenMP/nvptx_teams_reduction_codegen.cpp b/test/OpenMP/nvptx_teams_reduction_codegen.cpp index d77231807fb43..6925b17163af8 100644 --- a/test/OpenMP/nvptx_teams_reduction_codegen.cpp +++ b/test/OpenMP/nvptx_teams_reduction_codegen.cpp @@ -117,15 +117,15 @@ int bar(int n){ // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[REMOTE_ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[REMOTE_RED_LIST:%.+]], i[[SZ]] 0, i[[SZ]] 0 // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to double* - // CHECK: [[ELT_VAL:%.+]] = load double, double* [[ELT]], align // - // CHECK: [[ELT_CAST:%.+]] = bitcast double [[ELT_VAL]] to i64 + // CHECK: [[ELT_CAST:%.+]] = bitcast double* [[ELT]] to i64* + // CHECK: [[REMOTE_ELT_CAST:%.+]] = bitcast double* [[REMOTE_ELT]] to i64* + // CHECK: [[ELT_VAL:%.+]] = load i64, i64* [[ELT_CAST]], align // CHECK: [[WS32:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() // CHECK: [[WS:%.+]] = trunc i32 [[WS32]] to i16 - // CHECK: [[REMOTE_ELT_VAL64:%.+]] = call i64 @__kmpc_shuffle_int64(i64 [[ELT_CAST]], i16 [[LANEOFFSET]], i16 [[WS]]) - // CHECK: [[REMOTE_ELT_VAL:%.+]] = bitcast i64 [[REMOTE_ELT_VAL64]] to double + // CHECK: [[REMOTE_ELT_VAL64:%.+]] = call i64 @__kmpc_shuffle_int64(i64 [[ELT_VAL]], i16 [[LANEOFFSET]], i16 [[WS]]) // - // CHECK: store double [[REMOTE_ELT_VAL]], double* [[REMOTE_ELT]], align + // CHECK: store i64 [[REMOTE_ELT_VAL64]], i64* [[REMOTE_ELT_CAST]], align // CHECK: [[REMOTE_ELT_VOID:%.+]] = bitcast double* [[REMOTE_ELT]] to i8* // CHECK: store i8* [[REMOTE_ELT_VOID]], i8** [[REMOTE_ELT_REF]], align // @@ -168,8 +168,8 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_VOID:%.+]] = load i8*, i8** [[REMOTE_ELT_REF]], // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], - // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to double* // CHECK: [[REMOTE_ELT:%.+]] = bitcast i8* [[REMOTE_ELT_VOID]] to double* + // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to double* // CHECK: [[REMOTE_ELT_VAL:%.+]] = load double, double* [[REMOTE_ELT]], align // CHECK: store double [[REMOTE_ELT_VAL]], double* [[ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] @@ -193,10 +193,10 @@ int bar(int n){ // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to double* - // CHECK: [[ELT_VAL:%.+]] = load double, double* [[ELT]], align // // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[WARPID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to double addrspace([[SHARED_ADDRSPACE]])* + // CHECK: [[ELT_VAL:%.+]] = load double, double* [[ELT]], align // CHECK: store double [[ELT_VAL]], double addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] // @@ -216,10 +216,10 @@ int bar(int n){ // CHECK: [[DO_READ]] // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[TID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to double addrspace([[SHARED_ADDRSPACE]])* - // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load double, double addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST:%.+]], i[[SZ]] 0, i[[SZ]] 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to double* + // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load double, double addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: store double [[MEDIUM_ELT_VAL]], double* [[ELT]], align // CHECK: br label {{%?}}[[READ_CONT:.+]] // @@ -246,11 +246,11 @@ int bar(int n){ // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // - // CHECK: [[P:%.+]] = mul i[[SZ]] 8, [[TEAM]] - // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add i[[SZ]] [[SCRATCHPAD]], [[P]] + // CHECK: [[P:%.+]] = mul nuw i[[SZ]] 8, [[TEAM]] + // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[P]] // CHECK: [[SCRATCHPAD_ELT_PTR_VOID:%.+]] = inttoptr i[[SZ]] [[SCRATCHPAD_ELT_PTR64]] to i8* - // CHECK: [[SCRATCHPAD_ELT_PTR:%.+]] = bitcast i8* [[SCRATCHPAD_ELT_PTR_VOID]] to double* // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to double* + // CHECK: [[SCRATCHPAD_ELT_PTR:%.+]] = bitcast i8* [[SCRATCHPAD_ELT_PTR_VOID]] to double* // CHECK: [[ELT_VAL:%.+]] = load double, double* [[ELT]], align // CHECK: store double [[ELT_VAL]], double* [[SCRATCHPAD_ELT_PTR]], align // @@ -272,8 +272,8 @@ int bar(int n){ // CHECK: [[SHOULD_REDUCE:%.+]] = load i32, i32* {{.+}}, align // CHECK: [[SCRATCHPAD:%.+]] = ptrtoint i8* [[SCRATCHPAD_PTR]] to i[[SZ]] // - // CHECK: [[P:%.+]] = mul i[[SZ]] 8, [[TEAM]] - // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add i[[SZ]] [[SCRATCHPAD]], [[P]] + // CHECK: [[P:%.+]] = mul nuw i[[SZ]] 8, [[TEAM]] + // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[P]] // CHECK: [[SCRATCHPAD_ELT_PTR_VOID:%.+]] = inttoptr i[[SZ]] [[SCRATCHPAD_ELT_PTR64]] to i8* // CHECK: [[REMOTE_ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[REMOTE_RED_LIST]], i[[SZ]] 0, i[[SZ]] 0 @@ -283,7 +283,7 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_PTR:%.+]] = bitcast double* [[REMOTE_ELT]] to i8* // CHECK: store i8* [[REMOTE_ELT_PTR]], i8** [[REMOTE_ELT_REF]], align // - // CHECK: [[REDUCE:%.+]] = icmp eq i32 [[SHOULD_REDUCE]], 1 + // CHECK: [[REDUCE:%.+]] = icmp ne i32 [[SHOULD_REDUCE]], 0 // CHECK: br i1 [[REDUCE]], label {{%?}}[[DO_REDUCE:.+]], label {{%?}}[[REDUCE_ELSE:.+]] // // CHECK: [[DO_REDUCE]] @@ -298,8 +298,8 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_VOID:%.+]] = load i8*, i8** [[REMOTE_ELT_REF]], // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST:%.+]], i[[SZ]] 0, i[[SZ]] 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], - // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to double* // CHECK: [[REMOTE_ELT:%.+]] = bitcast i8* [[REMOTE_ELT_VOID]] to double* + // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to double* // CHECK: [[REMOTE_ELT_VAL:%.+]] = load double, double* [[REMOTE_ELT]], align // CHECK: store double [[REMOTE_ELT_VAL]], double* [[ELT]], align // CHECK: br label {{%?}}[[REDUCE_CONT]] @@ -307,16 +307,6 @@ int bar(int n){ // CHECK: [[REDUCE_CONT]] // CHECK: ret - - - - - - - - - - // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l33}}_worker() // CHECK: define {{.*}}void [[T2:@__omp_offloading_.+template.+l33]]( @@ -422,15 +412,15 @@ int bar(int n){ // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[REMOTE_ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[REMOTE_RED_LIST]], i[[SZ]] 0, i[[SZ]] 1 // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to float* - // CHECK: [[ELT_VAL:%.+]] = load float, float* [[ELT]], align // - // CHECK: [[ELT_CAST:%.+]] = bitcast float [[ELT_VAL]] to i32 + // CHECK: [[ELT_CAST:%.+]] = bitcast float* [[ELT]] to i32* + // CHECK: [[REMOTE_ELT2_CAST:%.+]] = bitcast float* [[REMOTE_ELT2]] to i32* + // CHECK: [[ELT_VAL:%.+]] = load i32, i32* [[ELT_CAST]], align // CHECK: [[WS32:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.warpsize() // CHECK: [[WS:%.+]] = trunc i32 [[WS32]] to i16 - // CHECK: [[REMOTE_ELT2_VAL32:%.+]] = call i32 @__kmpc_shuffle_int32(i32 [[ELT_CAST]], i16 [[LANEOFFSET]], i16 [[WS]]) - // CHECK: [[REMOTE_ELT2_VAL:%.+]] = bitcast i32 [[REMOTE_ELT2_VAL32]] to float + // CHECK: [[REMOTE_ELT2_VAL32:%.+]] = call i32 @__kmpc_shuffle_int32(i32 [[ELT_VAL]], i16 [[LANEOFFSET]], i16 [[WS]]) // - // CHECK: store float [[REMOTE_ELT2_VAL]], float* [[REMOTE_ELT2]], align + // CHECK: store i32 [[REMOTE_ELT2_VAL32]], i32* [[REMOTE_ELT2_CAST]], align // CHECK: [[REMOTE_ELT2C:%.+]] = bitcast float* [[REMOTE_ELT2]] to i8* // CHECK: store i8* [[REMOTE_ELT2C]], i8** [[REMOTE_ELT_REF]], align // @@ -480,8 +470,8 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_VOID:%.+]] = load i8*, i8** [[REMOTE_ELT_REF]], // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], - // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to float* // CHECK: [[REMOTE_ELT:%.+]] = bitcast i8* [[REMOTE_ELT_VOID]] to float* + // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to float* // CHECK: [[REMOTE_ELT_VAL:%.+]] = load float, float* [[REMOTE_ELT]], align // CHECK: store float [[REMOTE_ELT_VAL]], float* [[ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] @@ -504,10 +494,10 @@ int bar(int n){ // [[DO_COPY]] // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], - // CHECK: [[ELT_VAL:%.+]] = load i8, i8* [[ELT_VOID]], align // // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[WARPID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to i8 addrspace([[SHARED_ADDRSPACE]])* + // CHECK: [[ELT_VAL:%.+]] = load i8, i8* [[ELT_VOID]], align // CHECK: store i8 [[ELT_VAL]], i8 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] // @@ -527,9 +517,9 @@ int bar(int n){ // CHECK: [[DO_READ]] // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[TID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to i8 addrspace([[SHARED_ADDRSPACE]])* - // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load i8, i8 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST:%.+]], i[[SZ]] 0, i[[SZ]] 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], + // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load i8, i8 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: store i8 [[MEDIUM_ELT_VAL]], i8* [[ELT_VOID]], align // CHECK: br label {{%?}}[[READ_CONT:.+]] // @@ -545,10 +535,10 @@ int bar(int n){ // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to float* - // CHECK: [[ELT_VAL:%.+]] = load float, float* [[ELT]], align // // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[WARPID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to float addrspace([[SHARED_ADDRSPACE]])* + // CHECK: [[ELT_VAL:%.+]] = load float, float* [[ELT]], align // CHECK: store float [[ELT_VAL]], float addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] // @@ -568,10 +558,10 @@ int bar(int n){ // CHECK: [[DO_READ]] // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[TID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to float addrspace([[SHARED_ADDRSPACE]])* - // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load float, float addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST:%.+]], i[[SZ]] 0, i[[SZ]] 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to float* + // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load float, float addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: store float [[MEDIUM_ELT_VAL]], float* [[ELT]], align // CHECK: br label {{%?}}[[READ_CONT:.+]] // @@ -598,27 +588,27 @@ int bar(int n){ // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // - // CHECK: [[P:%.+]] = mul i[[SZ]] 1, [[TEAM]] - // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add i[[SZ]] [[SCRATCHPAD]], [[P]] + // CHECK: [[P:%.+]] = mul nuw i[[SZ]] 1, [[TEAM]] + // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[P]] // CHECK: [[SCRATCHPAD_ELT_PTR:%.+]] = inttoptr i[[SZ]] [[SCRATCHPAD_ELT_PTR64]] to i8* // CHECK: [[ELT_VAL:%.+]] = load i8, i8* [[ELT_VOID]], align // CHECK: store i8 [[ELT_VAL]], i8* [[SCRATCHPAD_ELT_PTR]], align // - // CHECK: [[OF:%.+]] = mul i[[SZ]] [[NUM_TEAMS]], 1 - // CHECK: [[POS1:%.+]] = add i[[SZ]] [[SCRATCHPAD]], [[OF]] - // CHECK: [[POS2:%.+]] = sub i[[SZ]] [[POS1]], 1 - // CHECK: [[POS3:%.+]] = sdiv i[[SZ]] [[POS2]], 256 - // CHECK: [[POS4:%.+]] = add i[[SZ]] [[POS3]], 1 - // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul i[[SZ]] [[POS4]], 256 + // CHECK: [[OF:%.+]] = mul nuw i[[SZ]] [[NUM_TEAMS]], 1 + // CHECK: [[POS1:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[OF]] + // CHECK: [[POS2:%.+]] = sub nuw i[[SZ]] [[POS1]], 1 + // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 256 + // CHECK: [[POS4:%.+]] = add nuw i[[SZ]] [[POS3]], 1 + // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 256 // // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // - // CHECK: [[P:%.+]] = mul i[[SZ]] 4, [[TEAM]] - // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add i[[SZ]] [[SCRATCHPAD_NEXT]], [[P]] + // CHECK: [[P:%.+]] = mul nuw i[[SZ]] 4, [[TEAM]] + // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD_NEXT]], [[P]] // CHECK: [[SCRATCHPAD_ELT_PTR_VOID:%.+]] = inttoptr i[[SZ]] [[SCRATCHPAD_ELT_PTR64]] to i8* - // CHECK: [[SCRATCHPAD_ELT_PTR:%.+]] = bitcast i8* [[SCRATCHPAD_ELT_PTR_VOID]] to float* // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to float* + // CHECK: [[SCRATCHPAD_ELT_PTR:%.+]] = bitcast i8* [[SCRATCHPAD_ELT_PTR_VOID]] to float* // CHECK: [[ELT_VAL:%.+]] = load float, float* [[ELT]], align // CHECK: store float [[ELT_VAL]], float* [[SCRATCHPAD_ELT_PTR]], align // @@ -641,8 +631,8 @@ int bar(int n){ // CHECK: [[SHOULD_REDUCE:%.+]] = load i32, i32* {{.+}}, align // CHECK: [[SCRATCHPAD:%.+]] = ptrtoint i8* [[SCRATCHPAD_PTR]] to i[[SZ]] // - // CHECK: [[P:%.+]] = mul i[[SZ]] 1, [[TEAM]] - // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add i[[SZ]] [[SCRATCHPAD]], [[P]] + // CHECK: [[P:%.+]] = mul nuw i[[SZ]] 1, [[TEAM]] + // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[P]] // CHECK: [[SCRATCHPAD_ELT_PTR_VOID:%.+]] = inttoptr i[[SZ]] [[SCRATCHPAD_ELT_PTR64]] to i8* // CHECK: [[REMOTE_ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[REMOTE_RED_LIST]], i[[SZ]] 0, i[[SZ]] 0 @@ -650,15 +640,15 @@ int bar(int n){ // CHECK: store i8 [[REMOTE_ELT_VAL]], i8* [[REMOTE_ELT1]], align // CHECK: store i8* [[REMOTE_ELT1]], i8** [[REMOTE_ELT_REF]], align // - // CHECK: [[OF:%.+]] = mul i[[SZ]] [[NUM_TEAMS]], 1 - // CHECK: [[POS1:%.+]] = add i[[SZ]] [[SCRATCHPAD]], [[OF]] - // CHECK: [[POS2:%.+]] = sub i[[SZ]] [[POS1]], 1 - // CHECK: [[POS3:%.+]] = sdiv i[[SZ]] [[POS2]], 256 - // CHECK: [[POS4:%.+]] = add i[[SZ]] [[POS3]], 1 - // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul i[[SZ]] [[POS4]], 256 + // CHECK: [[OF:%.+]] = mul nuw i[[SZ]] [[NUM_TEAMS]], 1 + // CHECK: [[POS1:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[OF]] + // CHECK: [[POS2:%.+]] = sub nuw i[[SZ]] [[POS1]], 1 + // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 256 + // CHECK: [[POS4:%.+]] = add nuw i[[SZ]] [[POS3]], 1 + // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 256 // - // CHECK: [[P:%.+]] = mul i[[SZ]] 4, [[TEAM]] - // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add i[[SZ]] [[SCRATCHPAD_NEXT]], [[P]] + // CHECK: [[P:%.+]] = mul nuw i[[SZ]] 4, [[TEAM]] + // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD_NEXT]], [[P]] // CHECK: [[SCRATCHPAD_ELT_PTR_VOID:%.+]] = inttoptr i[[SZ]] [[SCRATCHPAD_ELT_PTR64]] to i8* // CHECK: [[REMOTE_ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[REMOTE_RED_LIST]], i[[SZ]] 0, i[[SZ]] 1 @@ -668,7 +658,7 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_PTR:%.+]] = bitcast float* [[REMOTE_ELT2]] to i8* // CHECK: store i8* [[REMOTE_ELT_PTR]], i8** [[REMOTE_ELT_REF]], align // - // CHECK: [[REDUCE:%.+]] = icmp eq i32 [[SHOULD_REDUCE]], 1 + // CHECK: [[REDUCE:%.+]] = icmp ne i32 [[SHOULD_REDUCE]], 0 // CHECK: br i1 [[REDUCE]], label {{%?}}[[DO_REDUCE:.+]], label {{%?}}[[REDUCE_ELSE:.+]] // // CHECK: [[DO_REDUCE]] @@ -690,8 +680,8 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_VOID:%.+]] = load i8*, i8** [[REMOTE_ELT_REF]], // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST:%.+]], i[[SZ]] 0, i[[SZ]] 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], - // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to float* // CHECK: [[REMOTE_ELT:%.+]] = bitcast i8* [[REMOTE_ELT_VOID]] to float* + // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to float* // CHECK: [[REMOTE_ELT_VAL:%.+]] = load float, float* [[REMOTE_ELT]], align // CHECK: store float [[REMOTE_ELT_VAL]], float* [[ELT]], align // CHECK: br label {{%?}}[[REDUCE_CONT]] @@ -699,15 +689,6 @@ int bar(int n){ // CHECK: [[REDUCE_CONT]] // CHECK: ret - - - - - - - - - // CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+template.+l40}}_worker() // CHECK: define {{.*}}void [[T3:@__omp_offloading_.+template.+l40]]( @@ -903,8 +884,8 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_VOID:%.+]] = load i8*, i8** [[REMOTE_ELT_REF]], // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], - // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i32* // CHECK: [[REMOTE_ELT:%.+]] = bitcast i8* [[REMOTE_ELT_VOID]] to i32* + // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i32* // CHECK: [[REMOTE_ELT_VAL:%.+]] = load i32, i32* [[REMOTE_ELT]], align // CHECK: store i32 [[REMOTE_ELT_VAL]], i32* [[ELT]], align // @@ -912,8 +893,8 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_VOID:%.+]] = load i8*, i8** [[REMOTE_ELT_REF]], // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], - // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i16* // CHECK: [[REMOTE_ELT:%.+]] = bitcast i8* [[REMOTE_ELT_VOID]] to i16* + // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i16* // CHECK: [[REMOTE_ELT_VAL:%.+]] = load i16, i16* [[REMOTE_ELT]], align // CHECK: store i16 [[REMOTE_ELT_VAL]], i16* [[ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] @@ -937,10 +918,10 @@ int bar(int n){ // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i32* - // CHECK: [[ELT_VAL:%.+]] = load i32, i32* [[ELT]], align // // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[WARPID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to i32 addrspace([[SHARED_ADDRSPACE]])* + // CHECK: [[ELT_VAL:%.+]] = load i32, i32* [[ELT]], align // CHECK: store i32 [[ELT_VAL]], i32 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] // @@ -960,10 +941,10 @@ int bar(int n){ // CHECK: [[DO_READ]] // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[TID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to i32 addrspace([[SHARED_ADDRSPACE]])* - // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load i32, i32 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST:%.+]], i[[SZ]] 0, i[[SZ]] 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i32* + // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load i32, i32 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: store i32 [[MEDIUM_ELT_VAL]], i32* [[ELT]], align // CHECK: br label {{%?}}[[READ_CONT:.+]] // @@ -979,10 +960,10 @@ int bar(int n){ // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i16* - // CHECK: [[ELT_VAL:%.+]] = load i16, i16* [[ELT]], align // // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[WARPID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to i16 addrspace([[SHARED_ADDRSPACE]])* + // CHECK: [[ELT_VAL:%.+]] = load i16, i16* [[ELT]], align // CHECK: store i16 [[ELT_VAL]], i16 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: br label {{%?}}[[COPY_CONT:.+]] // @@ -1002,10 +983,10 @@ int bar(int n){ // CHECK: [[DO_READ]] // CHECK: [[MEDIUM_ELT64:%.+]] = getelementptr inbounds [32 x i64], [32 x i64] addrspace([[SHARED_ADDRSPACE]])* [[TRANSFER_STORAGE]], i64 0, i32 [[TID]] // CHECK: [[MEDIUM_ELT:%.+]] = bitcast i64 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT64]] to i16 addrspace([[SHARED_ADDRSPACE]])* - // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load i16, i16 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST:%.+]], i[[SZ]] 0, i[[SZ]] 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i16* + // CHECK: [[MEDIUM_ELT_VAL:%.+]] = load i16, i16 addrspace([[SHARED_ADDRSPACE]])* [[MEDIUM_ELT]], align // CHECK: store i16 [[MEDIUM_ELT_VAL]], i16* [[ELT]], align // CHECK: br label {{%?}}[[READ_CONT:.+]] // @@ -1032,29 +1013,29 @@ int bar(int n){ // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // - // CHECK: [[P:%.+]] = mul i[[SZ]] 4, [[TEAM]] - // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add i[[SZ]] [[SCRATCHPAD]], [[P]] + // CHECK: [[P:%.+]] = mul nuw i[[SZ]] 4, [[TEAM]] + // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[P]] // CHECK: [[SCRATCHPAD_ELT_PTR_VOID:%.+]] = inttoptr i[[SZ]] [[SCRATCHPAD_ELT_PTR64]] to i8* - // CHECK: [[SCRATCHPAD_ELT_PTR:%.+]] = bitcast i8* [[SCRATCHPAD_ELT_PTR_VOID]] to i32* // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i32* + // CHECK: [[SCRATCHPAD_ELT_PTR:%.+]] = bitcast i8* [[SCRATCHPAD_ELT_PTR_VOID]] to i32* // CHECK: [[ELT_VAL:%.+]] = load i32, i32* [[ELT]], align // CHECK: store i32 [[ELT_VAL]], i32* [[SCRATCHPAD_ELT_PTR]], align // - // CHECK: [[OF:%.+]] = mul i[[SZ]] [[NUM_TEAMS]], 4 - // CHECK: [[POS1:%.+]] = add i[[SZ]] [[SCRATCHPAD]], [[OF]] - // CHECK: [[POS2:%.+]] = sub i[[SZ]] [[POS1]], 1 - // CHECK: [[POS3:%.+]] = sdiv i[[SZ]] [[POS2]], 256 - // CHECK: [[POS4:%.+]] = add i[[SZ]] [[POS3]], 1 - // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul i[[SZ]] [[POS4]], 256 + // CHECK: [[OF:%.+]] = mul nuw i[[SZ]] [[NUM_TEAMS]], 4 + // CHECK: [[POS1:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[OF]] + // CHECK: [[POS2:%.+]] = sub nuw i[[SZ]] [[POS1]], 1 + // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 256 + // CHECK: [[POS4:%.+]] = add nuw i[[SZ]] [[POS3]], 1 + // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 256 // // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], // - // CHECK: [[P:%.+]] = mul i[[SZ]] 2, [[TEAM]] - // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add i[[SZ]] [[SCRATCHPAD_NEXT]], [[P]] + // CHECK: [[P:%.+]] = mul nuw i[[SZ]] 2, [[TEAM]] + // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD_NEXT]], [[P]] // CHECK: [[SCRATCHPAD_ELT_PTR_VOID:%.+]] = inttoptr i[[SZ]] [[SCRATCHPAD_ELT_PTR64]] to i8* - // CHECK: [[SCRATCHPAD_ELT_PTR:%.+]] = bitcast i8* [[SCRATCHPAD_ELT_PTR_VOID]] to i16* // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i16* + // CHECK: [[SCRATCHPAD_ELT_PTR:%.+]] = bitcast i8* [[SCRATCHPAD_ELT_PTR_VOID]] to i16* // CHECK: [[ELT_VAL:%.+]] = load i16, i16* [[ELT]], align // CHECK: store i16 [[ELT_VAL]], i16* [[SCRATCHPAD_ELT_PTR]], align // @@ -1077,8 +1058,8 @@ int bar(int n){ // CHECK: [[SHOULD_REDUCE:%.+]] = load i32, i32* {{.+}}, align // CHECK: [[SCRATCHPAD:%.+]] = ptrtoint i8* [[SCRATCHPAD_PTR]] to i[[SZ]] // - // CHECK: [[P:%.+]] = mul i[[SZ]] 4, [[TEAM]] - // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add i[[SZ]] [[SCRATCHPAD]], [[P]] + // CHECK: [[P:%.+]] = mul nuw i[[SZ]] 4, [[TEAM]] + // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[P]] // CHECK: [[SCRATCHPAD_ELT_PTR_VOID:%.+]] = inttoptr i[[SZ]] [[SCRATCHPAD_ELT_PTR64]] to i8* // CHECK: [[REMOTE_ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[REMOTE_RED_LIST]], i[[SZ]] 0, i[[SZ]] 0 @@ -1088,15 +1069,15 @@ int bar(int n){ // CHECK: [[REMOTE_ELT1_PTR:%.+]] = bitcast i32* [[REMOTE_ELT1]] to i8* // CHECK: store i8* [[REMOTE_ELT1_PTR]], i8** [[REMOTE_ELT_REF]], align // - // CHECK: [[OF:%.+]] = mul i[[SZ]] [[NUM_TEAMS]], 4 - // CHECK: [[POS1:%.+]] = add i[[SZ]] [[SCRATCHPAD]], [[OF]] - // CHECK: [[POS2:%.+]] = sub i[[SZ]] [[POS1]], 1 - // CHECK: [[POS3:%.+]] = sdiv i[[SZ]] [[POS2]], 256 - // CHECK: [[POS4:%.+]] = add i[[SZ]] [[POS3]], 1 - // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul i[[SZ]] [[POS4]], 256 + // CHECK: [[OF:%.+]] = mul nuw i[[SZ]] [[NUM_TEAMS]], 4 + // CHECK: [[POS1:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[OF]] + // CHECK: [[POS2:%.+]] = sub nuw i[[SZ]] [[POS1]], 1 + // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 256 + // CHECK: [[POS4:%.+]] = add nuw i[[SZ]] [[POS3]], 1 + // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 256 // - // CHECK: [[P:%.+]] = mul i[[SZ]] 2, [[TEAM]] - // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add i[[SZ]] [[SCRATCHPAD_NEXT]], [[P]] + // CHECK: [[P:%.+]] = mul nuw i[[SZ]] 2, [[TEAM]] + // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD_NEXT]], [[P]] // CHECK: [[SCRATCHPAD_ELT_PTR_VOID:%.+]] = inttoptr i[[SZ]] [[SCRATCHPAD_ELT_PTR64]] to i8* // CHECK: [[REMOTE_ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[REMOTE_RED_LIST]], i[[SZ]] 0, i[[SZ]] 1 @@ -1106,7 +1087,7 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_PTR:%.+]] = bitcast i16* [[REMOTE_ELT2]] to i8* // CHECK: store i8* [[REMOTE_ELT_PTR]], i8** [[REMOTE_ELT_REF]], align // - // CHECK: [[REDUCE:%.+]] = icmp eq i32 [[SHOULD_REDUCE]], 1 + // CHECK: [[REDUCE:%.+]] = icmp ne i32 [[SHOULD_REDUCE]], 0 // CHECK: br i1 [[REDUCE]], label {{%?}}[[DO_REDUCE:.+]], label {{%?}}[[REDUCE_ELSE:.+]] // // CHECK: [[DO_REDUCE]] @@ -1121,8 +1102,8 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_VOID:%.+]] = load i8*, i8** [[REMOTE_ELT_REF]], // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST:%.+]], i[[SZ]] 0, i[[SZ]] 0 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], - // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i32* // CHECK: [[REMOTE_ELT:%.+]] = bitcast i8* [[REMOTE_ELT_VOID]] to i32* + // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i32* // CHECK: [[REMOTE_ELT_VAL:%.+]] = load i32, i32* [[REMOTE_ELT]], align // CHECK: store i32 [[REMOTE_ELT_VAL]], i32* [[ELT]], align // @@ -1130,8 +1111,8 @@ int bar(int n){ // CHECK: [[REMOTE_ELT_VOID:%.+]] = load i8*, i8** [[REMOTE_ELT_REF]], // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST:%.+]], i[[SZ]] 0, i[[SZ]] 1 // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]], - // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i16* // CHECK: [[REMOTE_ELT:%.+]] = bitcast i8* [[REMOTE_ELT_VOID]] to i16* + // CHECK: [[ELT:%.+]] = bitcast i8* [[ELT_VOID]] to i16* // CHECK: [[REMOTE_ELT_VAL:%.+]] = load i16, i16* [[REMOTE_ELT]], align // CHECK: store i16 [[REMOTE_ELT_VAL]], i16* [[ELT]], align // CHECK: br label {{%?}}[[REDUCE_CONT]] diff --git a/test/OpenMP/openmp_check.cpp b/test/OpenMP/openmp_check.cpp index d46213a90e9f3..cd4706b57e44d 100644 --- a/test/OpenMP/openmp_check.cpp +++ b/test/OpenMP/openmp_check.cpp @@ -2,7 +2,16 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -std=c++11 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +#define p _Pragma("omp parallel") + int nested(int a) { +#pragma omp parallel p // expected-error {{unexpected OpenMP directive}} + ++a; #pragma omp parallel ++a; @@ -11,8 +20,6 @@ int nested(int a) { // expected-warning@-2 {{'auto' type specifier is a C++11 extension}} // expected-error@-3 {{expected expression}} // expected-error@-4 {{expected ';' at end of declaration}} -#else - // expected-no-diagnostics #endif #pragma omp parallel diff --git a/test/OpenMP/openmp_common.c b/test/OpenMP/openmp_common.c index ed55796b33746..feeefc4820ce5 100644 --- a/test/OpenMP/openmp_common.c +++ b/test/OpenMP/openmp_common.c @@ -1,5 +1,8 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + #pragma omp // expected-error {{expected an OpenMP directive}} #pragma omp unknown_directive // expected-error {{expected an OpenMP directive}} diff --git a/test/OpenMP/openmp_offload_codegen.cpp b/test/OpenMP/openmp_offload_codegen.cpp index f89bccf3b5527..fff50133aa9c6 100644 --- a/test/OpenMP/openmp_offload_codegen.cpp +++ b/test/OpenMP/openmp_offload_codegen.cpp @@ -1,12 +1,18 @@ // Test device for mapping codegen. ///==========================================================================/// -// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -S -emit-llvm %s -o - 2>&1 \ -// RUN: | FileCheck -check-prefix=CK1 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -S -emit-llvm %s -o - 2>&1 | FileCheck -check-prefix=CK1 %s + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -S -emit-llvm %s -o - 2>&1 | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CK1-DEVICE +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifdef CK1 @@ -33,4 +39,4 @@ void target_maps_parallel_integer(int a){ // CK1: [[GEPOPARG:%.+]] = getelementptr inbounds {{.*}} // CK1: call {{.*}}tgt_target({{.*}}i8** [[GEPOBPARG]], i8** [[GEPOPARG]] -#endif
\ No newline at end of file +#endif diff --git a/test/OpenMP/openmp_seh.c b/test/OpenMP/openmp_seh.c index 74dccec125e3d..0bdd079478894 100644 --- a/test/OpenMP/openmp_seh.c +++ b/test/OpenMP/openmp_seh.c @@ -1,4 +1,7 @@ // RUN: %clang_cc1 -verify -triple x86_64-pc-windows-msvc19.0.0 -fopenmp -fms-compatibility -x c++ -emit-llvm %s -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -triple x86_64-pc-windows-msvc19.0.0 -fopenmp-simd -fms-compatibility -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics // REQUIRES: x86-registered-target extern "C" { diff --git a/test/OpenMP/openmp_win_codegen.cpp b/test/OpenMP/openmp_win_codegen.cpp index 45b2c185cb8fd..289f058966448 100644 --- a/test/OpenMP/openmp_win_codegen.cpp +++ b/test/OpenMP/openmp_win_codegen.cpp @@ -1,5 +1,8 @@ -// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-pc-windows-msvc18.0.0 -std=c++11 -fms-compatibility-version=18 -fms-extensions -emit-llvm %s -fexceptions -fcxx-exceptions -o - -O1 | FileCheck %s -// REQUIRES: x86-registered-target +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-pc-windows-msvc18.0.0 -std=c++11 -fms-compatibility-version=18 -fms-extensions -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-pc-windows-msvc18.0.0 -std=c++11 -fms-compatibility-version=18 -fms-extensions -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s + +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics void foo(); @@ -30,7 +33,7 @@ struct Test { int main() { // CHECK: call void @{{.+}}main Test::main(); - // CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* {{.*}}@0, i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*)) + // CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* {{.*}}@0, i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* [[OUTLINED:@.+]] to void (i32*, i32*, ...)*)) #pragma omp parallel { try { @@ -47,15 +50,21 @@ int main() { } // CHECK: define internal void [[OUTLINED]]( -// CHECK: [[GID:%.+]] = {{.*}}call i32 @__kmpc_global_thread_num(%ident_t* {{.*}}@0) +// CHECK: [[GID:%.+]] = {{.*}}call i32 @__kmpc_global_thread_num(%struct.ident_t* {{.*}}@0) // CHECK: invoke void @{{.+}}foo -// CHECK: catchswitch within -// CHECK: catchpad within -// CHECK: call void @__kmpc_critical(%ident_t* {{.*}}@0, i32 [[GID]], +// CHECK: [[CATCHSWITCH:%.+]] = catchswitch within none +// CHECK: [[CATCHPAD:%.+]] = catchpad within [[CATCHSWITCH]] +// CHECK: call void @__kmpc_critical(%struct.ident_t* {{.*}}@0, i32 [[GID]], // CHECK: invoke void @{{.+}}bar -// CHECK: call void @__kmpc_end_critical(%ident_t* {{.*}}@0, i32 [[GID]], -// CHECK: catchret from -// CHECK: cleanuppad within -// CHECK: call void @__kmpc_end_critical(%ident_t* {{.*}}@0, i32 [[GID]], -// CHECK: cleanupret from - +// CHECK: call void @__kmpc_end_critical(%struct.ident_t* {{.*}}@0, i32 [[GID]], +// CHECK: catchret from [[CATCHPAD]] to +// CHECK: cleanuppad within [[CATCHPAD]] [] +// CHECK-NEXT: call void @__kmpc_end_critical(%struct.ident_t* {{.*}}@0, i32 [[GID]], +// CHECK-NEXT: cleanupret from {{.*}} unwind label %[[CATCHTERM:[^ ]+]] +// CHECK: cleanuppad within none [] +// CHECK-NEXT: call void @"?terminate@@YAXXZ"() #5 [ "funclet"(token %{{.*}}) ] +// CHECK-NEXT: unreachable +// CHECK: [[CATCHTERM]] +// CHECK-NEXT: cleanuppad within [[CATCHPAD]] [] +// CHECK-NEXT: call void @"?terminate@@YAXXZ"() #5 [ "funclet"(token %{{.*}}) ] +// CHECK-NEXT: unreachable diff --git a/test/OpenMP/option_warn.c b/test/OpenMP/option_warn.c index ddec8e113e77b..d9be164aad771 100644 --- a/test/OpenMP/option_warn.c +++ b/test/OpenMP/option_warn.c @@ -1,5 +1,8 @@ // RUN: %clang_cc1 -verify -Wsource-uses-openmp -o - %s +// RUN: %clang_cc1 -verify -Wsource-uses-openmp -o - %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + int a; #pragma omp threadprivate(a,b) // expected-warning {{unexpected '#pragma omp ...' in program}} #pragma omp parallel diff --git a/test/OpenMP/ordered_ast_print.cpp b/test/OpenMP/ordered_ast_print.cpp index c7baa9ce04214..f36afab8b2075 100644 --- a/test/OpenMP/ordered_ast_print.cpp +++ b/test/OpenMP/ordered_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -54,7 +58,7 @@ T tmain (T argc) { // CHECK: static T a; // CHECK-NEXT: #pragma omp for ordered // CHECK-NEXT: for (int i = 0; i < argc; ++i) -// CHECK-NEXT: #pragma omp ordered +// CHECK-NEXT: #pragma omp ordered{{$}} // CHECK-NEXT: { // CHECK-NEXT: a = 2; // CHECK-NEXT: } diff --git a/test/OpenMP/ordered_codegen.cpp b/test/OpenMP/ordered_codegen.cpp index af6a8722438db..977aba9d4555d 100644 --- a/test/OpenMP/ordered_codegen.cpp +++ b/test/OpenMP/ordered_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/ordered_doacross_codegen.c b/test/OpenMP/ordered_doacross_codegen.c new file mode 100644 index 0000000000000..c12df2638f8fc --- /dev/null +++ b/test/OpenMP/ordered_doacross_codegen.c @@ -0,0 +1,61 @@ +// RUN: %clang_cc1 -verify -fopenmp -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s +// RUN: %clang_cc1 -fopenmp -triple x86_64-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -triple x86_64-unknown-unknown -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -triple x86_64-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -triple x86_64-unknown-unknown -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} +// expected-no-diagnostics + +#ifndef HEADER +#define HEADER + +// CHECK: [[KMP_DIM:%.+]] = type { i64, i64, i64 } +extern int n; +int a[10], b[10], c[10], d[10]; +void foo(); + +// CHECK-LABEL: @main() +int main() { + int i; +// CHECK: [[DIMS:%.+]] = alloca [[KMP_DIM]], +// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num([[IDENT:%.+]]) +// CHECK: icmp +// CHECK-NEXT: br i1 % +// CHECK: [[CAST:%.+]] = bitcast [[KMP_DIM]]* [[DIMS]] to i8* +// CHECK: call void @llvm.memset.p0i8.i64(i8* align 8 [[CAST]], i8 0, i64 24, i1 false) +// CHECK: getelementptr inbounds [[KMP_DIM]], [[KMP_DIM]]* [[DIMS]], i32 0, i32 1 +// CHECK: store i64 %{{.+}}, i64* % +// CHECK: getelementptr inbounds [[KMP_DIM]], [[KMP_DIM]]* [[DIMS]], i32 0, i32 2 +// CHECK: store i64 1, i64* % +// CHECK: [[CAST:%.+]] = bitcast [[KMP_DIM]]* [[DIMS]] to i8* +// CHECK: call void @__kmpc_doacross_init([[IDENT]], i32 [[GTID]], i32 1, i8* [[CAST]]) +// CHECK: call void @__kmpc_for_static_init_4( +#pragma omp for ordered(1) + for (i = 0; i < n; ++i) { + a[i] = b[i] + 1; + foo(); +// CHECK: call void [[FOO:.+]]( +// CHECK: load i32, i32* [[CNT:%.+]], +// CHECK-NEXT: sext i32 %{{.+}} to i64 +// CHECK-NEXT: store i64 %{{.+}}, i64* [[TMP:%.+]], +// CHECK-NEXT: call void @__kmpc_doacross_post([[IDENT]], i32 [[GTID]], i64* [[TMP]]) +#pragma omp ordered depend(source) + c[i] = c[i] + 1; + foo(); +// CHECK: call void [[FOO]] +// CHECK: load i32, i32* [[CNT]], +// CHECK-NEXT: sub nsw i32 %{{.+}}, 2 +// CHECK-NEXT: sext i32 %{{.+}} to i64 +// CHECK-NEXT: store i64 %{{.+}}, i64* [[TMP:%.+]], +// CHECK-NEXT: call void @__kmpc_doacross_wait([[IDENT]], i32 [[GTID]], i64* [[TMP]]) +#pragma omp ordered depend(sink : i - 2) + d[i] = a[i - 2]; + } + // CHECK: call void @__kmpc_for_static_fini( + // CHECK: call void @__kmpc_doacross_fini([[IDENT]], i32 [[GTID]]) + // CHECK: ret i32 0 + return 0; +} +#endif // HEADER diff --git a/test/OpenMP/ordered_doacross_codegen.cpp b/test/OpenMP/ordered_doacross_codegen.cpp index 28b02c35358be..60b0353b711e8 100644 --- a/test/OpenMP/ordered_doacross_codegen.cpp +++ b/test/OpenMP/ordered_doacross_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER @@ -19,7 +24,7 @@ int main() { // CHECK: icmp // CHECK-NEXT: br i1 % // CHECK: [[CAST:%.+]] = bitcast [[KMP_DIM]]* [[DIMS]] to i8* -// CHECK: call void @llvm.memset.p0i8.i64(i8* [[CAST]], i8 0, i64 24, i32 8, i1 false) +// CHECK: call void @llvm.memset.p0i8.i64(i8* align 8 [[CAST]], i8 0, i64 24, i1 false) // CHECK: getelementptr inbounds [[KMP_DIM]], [[KMP_DIM]]* [[DIMS]], i32 0, i32 1 // CHECK: store i64 %{{.+}}, i64* % // CHECK: getelementptr inbounds [[KMP_DIM]], [[KMP_DIM]]* [[DIMS]], i32 0, i32 2 @@ -76,7 +81,7 @@ struct TestStruct { // CHECK: icmp // CHECK-NEXT: br i1 % // CHECK: [[CAST:%.+]] = bitcast [[KMP_DIM]]* [[DIMS]] to i8* -// CHECK: call void @llvm.memset.p0i8.i64(i8* [[CAST]], i8 0, i64 24, i32 8, i1 false) +// CHECK: call void @llvm.memset.p0i8.i64(i8* align 8 [[CAST]], i8 0, i64 24, i1 false) // CHECK: getelementptr inbounds [[KMP_DIM]], [[KMP_DIM]]* [[DIMS]], i32 0, i32 1 // CHECK: store i64 %{{.+}}, i64* % // CHECK: getelementptr inbounds [[KMP_DIM]], [[KMP_DIM]]* [[DIMS]], i32 0, i32 2 diff --git a/test/OpenMP/ordered_messages.cpp b/test/OpenMP/ordered_messages.cpp index 2e3d55ca83703..43ac40efe6408 100644 --- a/test/OpenMP/ordered_messages.cpp +++ b/test/OpenMP/ordered_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -std=c++98 -o - %s // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -std=c++11 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -std=c++98 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -std=c++11 -o - %s + int foo(); template <class T> @@ -134,6 +138,8 @@ T foo() { #pragma omp ordered depend(source) depend(sink : i-0, j+sizeof(T)) // expected-error {{'depend(sink:vec)' clauses cannot be mixed with 'depend(source)' clause}} } } +#pragma omp ordered depend(source) // expected-error {{'ordered' directive with 'depend' clause cannot be closely nested inside ordered region without specified parameter}} +#pragma omp ordered depend(sink:k) // expected-error {{'ordered' directive with 'depend' clause cannot be closely nested inside ordered region without specified parameter}} return T(); } diff --git a/test/OpenMP/parallel_ast_print.cpp b/test/OpenMP/parallel_ast_print.cpp index 09141db5841f8..19ebb70483308 100644 --- a/test/OpenMP/parallel_ast_print.cpp +++ b/test/OpenMP/parallel_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -158,7 +162,7 @@ T tmain(T argc, T *argv) { // CHECK-NEXT: static T a; // CHECK-NEXT: S<T> s; // CHECK-NEXT: T arr[C][10], arr1[C]; -// CHECK-NEXT: #pragma omp parallel +// CHECK-NEXT: #pragma omp parallel{{$}} // CHECK-NEXT: a = 2; // CHECK-NEXT: #pragma omp parallel default(none) private(argc,b) firstprivate(argv) shared(d) if(parallel: argc > 0) num_threads(C) copyin(S<T>::TS,thrp) proc_bind(master) reduction(+: c,arr1[argc]) reduction(max: e,arr[:C][0:10]) // CHECK-NEXT: foo() diff --git a/test/OpenMP/parallel_codegen.cpp b/test/OpenMP/parallel_codegen.cpp index 77e8efded68b5..0ecca8de64fe9 100644 --- a/test/OpenMP/parallel_codegen.cpp +++ b/test/OpenMP/parallel_codegen.cpp @@ -1,15 +1,20 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -emit-llvm %s -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix=CHECK-DEBUG %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -emit-llvm %s -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER -// CHECK-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CHECK-DAG: [[DEF_LOC_2:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } -// CHECK-DEBUG-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: [[DEF_LOC_2:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DEBUG-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CHECK-DEBUG-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CHECK-DEBUG-DAG: [[DEF_LOC_2:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DEBUG-DAG: [[DEF_LOC_2:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CHECK-DEBUG-DAG: [[LOC1:@.+]] = private unnamed_addr constant [{{.+}} x i8] c";{{.*}}parallel_codegen.cpp;main;[[@LINE+15]];9;;\00" // CHECK-DEBUG-DAG: [[LOC2:@.+]] = private unnamed_addr constant [{{.+}} x i8] c";{{.*}}parallel_codegen.cpp;tmain;[[@LINE+7]];9;;\00" @@ -33,21 +38,21 @@ int main (int argc, char **argv) { // CHECK-LABEL: define {{[a-z\_\b]*[ ]?i32}} @main({{i32[ ]?[a-z]*}} %argc, i8** %argv) // CHECK: store i32 %argc, i32* [[ARGC_ADDR:%.+]], // CHECK: [[VLA:%.+]] = alloca i32, i{{[0-9]+}} [[VLA_SIZE:%[^,]+]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC_2]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i{{[0-9]+}}, i32*)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i{{[0-9]+}} [[VLA_SIZE]], i32* [[VLA]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC_2]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i{{[0-9]+}}, i32*)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i{{[0-9]+}} [[VLA_SIZE]], i32* [[VLA]]) // CHECK-NEXT: [[ARGV:%.+]] = load i8**, i8*** {{%[a-z0-9.]+}} // CHECK-NEXT: [[RET:%.+]] = call {{[a-z\_\b]*[ ]?i32}} [[TMAIN:@.+tmain.+]](i8** [[ARGV]]) // CHECK: ret i32 // CHECK-NEXT: } // CHECK-DEBUG-LABEL: define i32 @main(i32 %argc, i8** %argv) -// CHECK-DEBUG: [[LOC_2_ADDR:%.+]] = alloca %ident_t -// CHECK-DEBUG: [[KMPC_LOC_VOIDPTR:%.+]] = bitcast %ident_t* [[LOC_2_ADDR]] to i8* -// CHECK-DEBUG-NEXT: [[KMPC_DEFAULT_LOC_VOIDPTR:%.+]] = bitcast %ident_t* [[DEF_LOC_2]] to i8* -// CHECK-DEBUG-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[KMPC_LOC_VOIDPTR]], i8* [[KMPC_DEFAULT_LOC_VOIDPTR]], i64 24, i32 8, i1 false) +// CHECK-DEBUG: [[LOC_2_ADDR:%.+]] = alloca %struct.ident_t +// CHECK-DEBUG: [[KMPC_LOC_VOIDPTR:%.+]] = bitcast %struct.ident_t* [[LOC_2_ADDR]] to i8* +// CHECK-DEBUG-NEXT: [[KMPC_DEFAULT_LOC_VOIDPTR:%.+]] = bitcast %struct.ident_t* [[DEF_LOC_2]] to i8* +// CHECK-DEBUG-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[KMPC_LOC_VOIDPTR]], i8* align 8 [[KMPC_DEFAULT_LOC_VOIDPTR]], i64 24, i1 false) // CHECK-DEBUG: store i32 %argc, i32* [[ARGC_ADDR:%.+]], // CHECK-DEBUG: [[VLA:%.+]] = alloca i32, i64 [[VLA_SIZE:%[^,]+]], -// CHECK-DEBUG: [[KMPC_LOC_PSOURCE_REF:%.+]] = getelementptr inbounds %ident_t, %ident_t* [[LOC_2_ADDR]], i32 0, i32 4 +// CHECK-DEBUG: [[KMPC_LOC_PSOURCE_REF:%.+]] = getelementptr inbounds %struct.ident_t, %struct.ident_t* [[LOC_2_ADDR]], i32 0, i32 4 // CHECK-DEBUG-NEXT: store i8* getelementptr inbounds ([{{.+}} x i8], [{{.+}} x i8]* [[LOC1]], i32 0, i32 0), i8** [[KMPC_LOC_PSOURCE_REF]] -// CHECK-DEBUG: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[LOC_2_ADDR]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i32*)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i64 [[VLA_SIZE]], i32* [[VLA]]) +// CHECK-DEBUG: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[LOC_2_ADDR]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64, i32*)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i64 [[VLA_SIZE]], i32* [[VLA]]) // CHECK-DEBUG-NEXT: [[ARGV:%.+]] = load i8**, i8*** {{%[a-z0-9.]+}} // CHECK-DEBUG-NEXT: [[RET:%.+]] = call i32 [[TMAIN:@.+tmain.+]](i8** [[ARGV]]) // CHECK-DEBUG: ret i32 @@ -77,26 +82,26 @@ int main (int argc, char **argv) { // CHECK-DEBUG-NEXT: } // CHECK-DAG: define linkonce_odr {{.*}}void [[FOO]]({{i32[ ]?[a-z]*}} %argc) -// CHECK-DAG: declare {{.*}}void @__kmpc_fork_call(%ident_t*, i32, void (i32*, i32*, ...)*, ...) +// CHECK-DAG: declare {{.*}}void @__kmpc_fork_call(%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) // CHECK-DEBUG-DAG: define linkonce_odr void [[FOO]](i32 %argc) -// CHECK-DEBUG-DAG: declare void @__kmpc_fork_call(%ident_t*, i32, void (i32*, i32*, ...)*, ...) +// CHECK-DEBUG-DAG: declare void @__kmpc_fork_call(%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) // CHECK-DEBUG-DAG: define internal void [[OMP_OUTLINED]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i64 [[VLA_SIZE:%.+]], i32* {{.+}} [[VLA_ADDR:%[^)]+]]) // CHECK-DEBUG-DAG: call void [[OMP_OUTLINED_DEBUG]] // CHECK: define linkonce_odr {{[a-z\_\b]*[ ]?i32}} [[TMAIN]](i8** %argc) // CHECK: store i8** %argc, i8*** [[ARGC_ADDR:%.+]], -// CHECK-NEXT: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC_2]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i8***)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i8*** [[ARGC_ADDR]]) +// CHECK-NEXT: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC_2]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i8***)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i8*** [[ARGC_ADDR]]) // CHECK-NEXT: ret i32 0 // CHECK-NEXT: } // CHECK-DEBUG: define linkonce_odr i32 [[TMAIN]](i8** %argc) -// CHECK-DEBUG-DAG: [[LOC_2_ADDR:%.+]] = alloca %ident_t -// CHECK-DEBUG: [[KMPC_LOC_VOIDPTR:%.+]] = bitcast %ident_t* [[LOC_2_ADDR]] to i8* -// CHECK-DEBUG-NEXT: [[KMPC_DEFAULT_LOC_VOIDPTR:%.+]] = bitcast %ident_t* [[DEF_LOC_2]] to i8* -// CHECK-DEBUG-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[KMPC_LOC_VOIDPTR]], i8* [[KMPC_DEFAULT_LOC_VOIDPTR]], i64 24, i32 8, i1 false) +// CHECK-DEBUG-DAG: [[LOC_2_ADDR:%.+]] = alloca %struct.ident_t +// CHECK-DEBUG: [[KMPC_LOC_VOIDPTR:%.+]] = bitcast %struct.ident_t* [[LOC_2_ADDR]] to i8* +// CHECK-DEBUG-NEXT: [[KMPC_DEFAULT_LOC_VOIDPTR:%.+]] = bitcast %struct.ident_t* [[DEF_LOC_2]] to i8* +// CHECK-DEBUG-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[KMPC_LOC_VOIDPTR]], i8* align 8 [[KMPC_DEFAULT_LOC_VOIDPTR]], i64 24, i1 false) // CHECK-DEBUG-NEXT: store i8** %argc, i8*** [[ARGC_ADDR:%.+]], -// CHECK-DEBUG: [[KMPC_LOC_PSOURCE_REF:%.+]] = getelementptr inbounds %ident_t, %ident_t* [[LOC_2_ADDR]], i32 0, i32 4 +// CHECK-DEBUG: [[KMPC_LOC_PSOURCE_REF:%.+]] = getelementptr inbounds %struct.ident_t, %struct.ident_t* [[LOC_2_ADDR]], i32 0, i32 4 // CHECK-DEBUG-NEXT: store i8* getelementptr inbounds ([{{.+}} x i8], [{{.+}} x i8]* [[LOC2]], i32 0, i32 0), i8** [[KMPC_LOC_PSOURCE_REF]] -// CHECK-DEBUG-NEXT: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[LOC_2_ADDR]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i8***)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i8*** [[ARGC_ADDR]]) +// CHECK-DEBUG-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[LOC_2_ADDR]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i8***)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i8*** [[ARGC_ADDR]]) // CHECK-DEBUG-NEXT: ret i32 0 // CHECK-DEBUG-NEXT: } diff --git a/test/OpenMP/parallel_copyin_codegen.cpp b/test/OpenMP/parallel_copyin_codegen.cpp index 49e7b3fd618ba..3d33aa2cfe0a8 100644 --- a/test/OpenMP/parallel_copyin_codegen.cpp +++ b/test/OpenMP/parallel_copyin_codegen.cpp @@ -5,12 +5,28 @@ // RUN: %clang_cc1 -verify -fopenmp -fnoopenmp-use-tls -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s // RUN: %clang_cc1 -verify -fopenmp -fnoopenmp-use-tls -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fnoopenmp-use-tls -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fnoopenmp-use-tls -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fnoopenmp-use-tls -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fnoopenmp-use-tls -x c++ -std=c++11 -DLAMBDA -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fnoopenmp-use-tls -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fnoopenmp-use-tls -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck %s -check-prefix=TLS-CHECK // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TLS-CHECK // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=TLS-LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=TLS-BLOCKS %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DARRAY -triple x86_64-linux-gnu -emit-llvm %s -o - | FileCheck -check-prefix=TLS-ARRAY %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DARRAY -triple x86_64-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef ARRAY #ifndef HEADER @@ -40,18 +56,18 @@ struct S { // CHECK-DAG: [[VEC:@.+]] = internal global [2 x i{{[0-9]+}}] [i{{[0-9]+}} 1, i{{[0-9]+}} 2], // CHECK-DAG: [[S_ARR:@.+]] = internal global [2 x [[S_FLOAT_TY]]] zeroinitializer, // CHECK-DAG: [[VAR:@.+]] = internal global [[S_FLOAT_TY]] zeroinitializer, -// CHECK-DAG: [[TMAIN_T_VAR:@.+]] = linkonce_odr global i{{[0-9]+}} 333, -// CHECK-DAG: [[TMAIN_VEC:@.+]] = linkonce_odr global [2 x i{{[0-9]+}}] [i{{[0-9]+}} 3, i{{[0-9]+}} 3], -// CHECK-DAG: [[TMAIN_S_ARR:@.+]] = linkonce_odr global [2 x [[S_INT_TY]]] zeroinitializer, -// CHECK-DAG: [[TMAIN_VAR:@.+]] = linkonce_odr global [[S_INT_TY]] zeroinitializer, +// CHECK-DAG: [[TMAIN_T_VAR:@.+]] = linkonce_odr {{(dso_local )?}}global i{{[0-9]+}} 333, +// CHECK-DAG: [[TMAIN_VEC:@.+]] = linkonce_odr {{(dso_local )?}}global [2 x i{{[0-9]+}}] [i{{[0-9]+}} 3, i{{[0-9]+}} 3], +// CHECK-DAG: [[TMAIN_S_ARR:@.+]] = linkonce_odr {{(dso_local )?}}global [2 x [[S_INT_TY]]] zeroinitializer, +// CHECK-DAG: [[TMAIN_VAR:@.+]] = linkonce_odr {{(dso_local )?}}global [[S_INT_TY]] zeroinitializer, // TLS-CHECK-DAG: [[T_VAR:@.+]] = internal thread_local global i{{[0-9]+}} 1122, // TLS-CHECK-DAG: [[VEC:@.+]] = internal thread_local global [2 x i{{[0-9]+}}] [i{{[0-9]+}} 1, i{{[0-9]+}} 2], // TLS-CHECK-DAG: [[S_ARR:@.+]] = internal thread_local global [2 x [[S_FLOAT_TY]]] zeroinitializer, // TLS-CHECK-DAG: [[VAR:@.+]] = internal thread_local global [[S_FLOAT_TY]] zeroinitializer, -// TLS-CHECK-DAG: [[TMAIN_T_VAR:@.+]] = linkonce_odr thread_local global i{{[0-9]+}} 333, -// TLS-CHECK-DAG: [[TMAIN_VEC:@.+]] = linkonce_odr thread_local global [2 x i{{[0-9]+}}] [i{{[0-9]+}} 3, i{{[0-9]+}} 3], -// TLS-CHECK-DAG: [[TMAIN_S_ARR:@.+]] = linkonce_odr thread_local global [2 x [[S_INT_TY]]] zeroinitializer, -// TLS-CHECK-DAG: [[TMAIN_VAR:@.+]] = linkonce_odr thread_local global [[S_INT_TY]] zeroinitializer, +// TLS-CHECK-DAG: [[TMAIN_T_VAR:@.+]] = linkonce_odr {{(dso_local )?}}thread_local global i{{[0-9]+}} 333, +// TLS-CHECK-DAG: [[TMAIN_VEC:@.+]] = linkonce_odr {{(dso_local )?}}thread_local global [2 x i{{[0-9]+}}] [i{{[0-9]+}} 3, i{{[0-9]+}} 3], +// TLS-CHECK-DAG: [[TMAIN_S_ARR:@.+]] = linkonce_odr {{(dso_local )?}}thread_local global [2 x [[S_INT_TY]]] zeroinitializer, +// TLS-CHECK-DAG: [[TMAIN_VAR:@.+]] = linkonce_odr {{(dso_local )?}}thread_local global [[S_INT_TY]] zeroinitializer, template <typename T> T tmain() { @@ -74,7 +90,7 @@ T tmain() { int main() { #ifdef LAMBDA - // LAMBDA: [[G:@.+]] = global i{{[0-9]+}} 1212, + // LAMBDA: [[G:@.+]] = {{(dso_local )?}}global i{{[0-9]+}} 1212, // LAMBDA-LABEL: @main // LAMBDA: call{{.*}} void [[OUTER_LAMBDA:@.+]]( // TLS-LAMBDA: [[G:@.+]] = {{.*}}thread_local {{.*}}global i{{[0-9]+}} 1212, @@ -136,7 +152,7 @@ int main() { }(); return 0; #elif defined(BLOCKS) - // BLOCKS: [[G:@.+]] = global i{{[0-9]+}} 1212, + // BLOCKS: [[G:@.+]] = {{(dso_local )?}}global i{{[0-9]+}} 1212, // BLOCKS-LABEL: @main // BLOCKS: call {{.*}}void {{%.+}}(i8 @@ -278,10 +294,10 @@ int main() { // threadprivate_vec = vec; // CHECK: call {{.*}}i8* @__kmpc_threadprivate_cached({{.+}} [[VEC]] -// CHECK: call void @llvm.memcpy{{.*}}(i8* %{{.+}}, i8* bitcast ([2 x i{{[0-9]+}}]* [[VEC]] to i8*), +// CHECK: call void @llvm.memcpy{{.*}}(i8* align {{[0-9]+}} %{{.+}}, i8* align {{[0-9]+}} bitcast ([2 x i{{[0-9]+}}]* [[VEC]] to i8*), // TLS-CHECK: [[MASTER_CAST:%.+]] = bitcast [2 x i32]* [[MASTER_REF2]] to i8* -// TLS-CHECK: call void @llvm.memcpy{{.*}}(i8* bitcast ([2 x i{{[0-9]+}}]* [[VEC]] to i8*), i8* [[MASTER_CAST]] +// TLS-CHECK: call void @llvm.memcpy{{.*}}(i8* align {{[0-9]+}} bitcast ([2 x i{{[0-9]+}}]* [[VEC]] to i8*), i8* align {{[0-9]+}} [[MASTER_CAST]] // threadprivate_s_arr = s_arr; // CHECK: call {{.*}}i8* @__kmpc_threadprivate_cached({{.+}} [[S_ARR]] @@ -399,10 +415,10 @@ int main() { // threadprivate_vec = vec; // CHECK: call {{.*}}i8* @__kmpc_threadprivate_cached({{.+}} [[TMAIN_VEC]] -// CHECK: call {{.*}}void @llvm.memcpy{{.*}}(i8* %{{.+}}, i8* bitcast ([2 x i{{[0-9]+}}]* [[TMAIN_VEC]] to i8*), +// CHECK: call {{.*}}void @llvm.memcpy{{.*}}(i8* align {{[0-9]+}} %{{.+}}, i8* align {{[0-9]+}} bitcast ([2 x i{{[0-9]+}}]* [[TMAIN_VEC]] to i8*), // TLS-CHECK: [[MASTER_CAST:%.+]] = bitcast [2 x i32]* [[MASTER_REF1]] to i8* -// TLS-CHECK: call void @llvm.memcpy{{.*}}(i8* bitcast ([2 x i{{[0-9]+}}]* [[TMAIN_VEC]] to i8*), i8* [[MASTER_CAST]] +// TLS-CHECK: call void @llvm.memcpy{{.*}}(i8* align {{[0-9]+}} bitcast ([2 x i{{[0-9]+}}]* [[TMAIN_VEC]] to i8*), i8* align {{[0-9]+}} [[MASTER_CAST]] // threadprivate_s_arr = s_arr; // CHECK: call {{.*}}i8* @__kmpc_threadprivate_cached({{.+}} [[TMAIN_S_ARR]] @@ -492,13 +508,13 @@ void array_func() { static St s[2]; // ARRAY: @__kmpc_fork_call( // ARRAY: call i8* @__kmpc_threadprivate_cached( -// ARRAY: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* bitcast ([2 x i32]* @{{.+}} to i8*), i64 8, i32 4, i1 false) +// ARRAY: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %{{.+}}, i8* align 4 bitcast ([2 x i32]* @{{.+}} to i8*), i64 8, i1 false) // ARRAY: call dereferenceable(8) %struct.St* @{{.+}}(%struct.St* %{{.+}}, %struct.St* dereferenceable(8) %{{.+}}) // TLS-ARRAY: @__kmpc_fork_call( // TLS-ARRAY: [[REFT:%.+]] = load [2 x i32]*, [2 x i32]** [[ADDR:%.+]], // TLS-ARRAY: [[REF:%.+]] = bitcast [2 x i32]* [[REFT]] to i8* -// TLS-ARRAY: call void @llvm.memcpy.p0i8.p0i8.i64(i8* bitcast ([2 x i32]* @{{.+}} to i8*), i8* [[REF]], i64 8, i32 4, i1 false) +// TLS-ARRAY: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 bitcast ([2 x i32]* @{{.+}} to i8*), i8* align 4 [[REF]], i64 8, i1 false) // TLS-ARRAY: call dereferenceable(8) %struct.St* @{{.+}}(%struct.St* %{{.+}}, %struct.St* dereferenceable(8) %{{.+}}) #pragma omp threadprivate(a, s) diff --git a/test/OpenMP/parallel_copyin_messages.cpp b/test/OpenMP/parallel_copyin_messages.cpp index 2b54b43b87c66..81a507addf100 100644 --- a/test/OpenMP/parallel_copyin_messages.cpp +++ b/test/OpenMP/parallel_copyin_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/parallel_default_messages.cpp b/test/OpenMP/parallel_default_messages.cpp index cbc6a73fe35df..4b698bccb0fba 100644 --- a/test/OpenMP/parallel_default_messages.cpp +++ b/test/OpenMP/parallel_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/parallel_firstprivate_codegen.cpp b/test/OpenMP/parallel_firstprivate_codegen.cpp index d20f3f5135691..6772f61779ef2 100644 --- a/test/OpenMP/parallel_firstprivate_codegen.cpp +++ b/test/OpenMP/parallel_firstprivate_codegen.cpp @@ -4,13 +4,30 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA -check-prefix=LAMBDA-32 %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS -check-prefix=BLOCKS-32 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK -check-prefix=CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=CHECK -check-prefix=CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-pc-linux-gnu -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA -check-prefix=LAMBDA-64 %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-pc-linux-gnu -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS -check-prefix=BLOCKS-64 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef ARRAY #ifndef HEADER @@ -349,7 +366,7 @@ int main() { // CHECK-64: [[SIVAR7_CONV:%.+]] = bitcast i64* [[SIVAR7_PRIV]] to i32* // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_REF]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* [[VEC_SRC]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST]], i8* align {{[0-9]+}} [[VEC_SRC]], // CHECK: [[S_ARR_PRIV_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_BEGIN:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_REF]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_PRIV_END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_PRIV_BEGIN]], i{{[0-9]+}} 2 @@ -442,7 +459,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_PRIV]], align 128 // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_REF]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* [[VEC_SRC]], i{{[0-9]+}} {{[0-9]+}}, i{{[0-9]+}} 128, +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align 128 [[VEC_DEST]], i8* align 128 [[VEC_SRC]], i{{[0-9]+}} {{[0-9]+}}, i1 // CHECK: [[S_ARR_PRIV_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_BEGIN:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_REF]] to [[S_INT_TY]]* // CHECK: [[S_ARR_PRIV_END:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_PRIV_BEGIN]], i{{[0-9]+}} 2 @@ -490,7 +507,7 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) { // ARRAY-DAG: store double* %{{.+}}, double** [[PRIV_VLA2]], // ARRAY: call i8* @llvm.stacksave() // ARRAY: [[SIZE:%.+]] = mul nuw i64 %{{.+}}, 8 -// ARRAY: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* %{{.+}}, i64 [[SIZE]], i32 128, i1 false) +// ARRAY: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 128 %{{.+}}, i8* align 128 %{{.+}}, i64 [[SIZE]], i1 false) #pragma omp parallel firstprivate(a, s, vla1, vla2) s[0].St_func(s, n, vla1); ; @@ -506,7 +523,7 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) { // ARRAY-DAG: store double* %{{.+}}, double** [[PRIV_VLA2]], // ARRAY: call i8* @llvm.stacksave() // ARRAY: [[SIZE:%.+]] = mul nuw i64 %{{.+}}, 8 -// ARRAY: call void @llvm.memcpy.p0i8.p0i8.i64(i8* %{{.+}}, i8* %{{.+}}, i64 [[SIZE]], i32 128, i1 false) +// ARRAY: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 128 %{{.+}}, i8* align 128 %{{.+}}, i64 [[SIZE]], i1 false) #endif diff --git a/test/OpenMP/parallel_firstprivate_messages.cpp b/test/OpenMP/parallel_firstprivate_messages.cpp index b2cb4ff131ef1..7de268fb3b7e7 100644 --- a/test/OpenMP/parallel_firstprivate_messages.cpp +++ b/test/OpenMP/parallel_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/parallel_for_ast_print.cpp b/test/OpenMP/parallel_for_ast_print.cpp index 3c36bd80523ce..1e77bc855a22a 100644 --- a/test/OpenMP/parallel_for_ast_print.cpp +++ b/test/OpenMP/parallel_for_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -35,7 +39,7 @@ public: } }; -// CHECK: #pragma omp parallel for private(this->a) private(this->a) private(T::a) +// CHECK: #pragma omp parallel for private(this->a) private(this->a) private(T::a){{$}} // CHECK: #pragma omp parallel for private(this->a) private(this->a) // CHECK: #pragma omp parallel for private(this->a) private(this->a) private(this->S::a) diff --git a/test/OpenMP/parallel_for_codegen.cpp b/test/OpenMP/parallel_for_codegen.cpp index db3a2de580d80..e6199f00eabd4 100644 --- a/test/OpenMP/parallel_for_codegen.cpp +++ b/test/OpenMP/parallel_for_codegen.cpp @@ -3,6 +3,13 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -O1 -fopenmp -emit-llvm %s -o - | FileCheck %s --check-prefix=CLEANUP + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -O1 -fopenmp-simd -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -347,9 +354,9 @@ int foo() {return 0;}; // TERM_DEBUG-LABEL: parallel_for // CLEANUP: parallel_for -void parallel_for(float *a, int n) { +void parallel_for(float *a, const int n) { float arr[n]; -#pragma omp parallel for schedule(static, 5) private(arr) +#pragma omp parallel for schedule(static, 5) private(arr) default(none) firstprivate(n) shared(a) // TERM_DEBUG-NOT: __kmpc_global_thread_num // TERM_DEBUG: call void @__kmpc_for_static_init_4u({{.+}}), !dbg [[DBG_LOC_START:![0-9]+]] // TERM_DEBUG: invoke i32 {{.*}}foo{{.*}}() @@ -363,7 +370,7 @@ void parallel_for(float *a, int n) { // CLEANUP: call void @__kmpc_for_static_init_4u({{.+}}) // CLEANUP: call void @__kmpc_for_static_fini({{.+}}) for (unsigned i = 131071; i <= 2147483647; i += 127) - a[i] += foo() + arr[i]; + a[i] += foo() + arr[i] + n; } // Check source line corresponds to "#pragma omp parallel for schedule(static, 5)" above: // TERM_DEBUG-DAG: [[DBG_LOC_START]] = !DILocation(line: [[@LINE-4]], diff --git a/test/OpenMP/parallel_for_collapse_messages.cpp b/test/OpenMP/parallel_for_collapse_messages.cpp index 4461df8cccdba..9d99c25efa61c 100644 --- a/test/OpenMP/parallel_for_collapse_messages.cpp +++ b/test/OpenMP/parallel_for_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/parallel_for_copyin_messages.cpp b/test/OpenMP/parallel_for_copyin_messages.cpp index 5a5d1632c8eb2..265a1953e0686 100644 --- a/test/OpenMP/parallel_for_copyin_messages.cpp +++ b/test/OpenMP/parallel_for_copyin_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/parallel_for_default_messages.cpp b/test/OpenMP/parallel_for_default_messages.cpp index d2129a3459391..743383640f3b7 100644 --- a/test/OpenMP/parallel_for_default_messages.cpp +++ b/test/OpenMP/parallel_for_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/parallel_for_firstprivate_messages.cpp b/test/OpenMP/parallel_for_firstprivate_messages.cpp index c9a69dab2cb25..19b1ced757b3a 100644 --- a/test/OpenMP/parallel_for_firstprivate_messages.cpp +++ b/test/OpenMP/parallel_for_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/parallel_for_if_messages.cpp b/test/OpenMP/parallel_for_if_messages.cpp index b5a7fdcd1ab53..1d83d3585d608 100644 --- a/test/OpenMP/parallel_for_if_messages.cpp +++ b/test/OpenMP/parallel_for_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/parallel_for_lastprivate_messages.cpp b/test/OpenMP/parallel_for_lastprivate_messages.cpp index 134100bc17531..8d2f4b5886fcd 100644 --- a/test/OpenMP/parallel_for_lastprivate_messages.cpp +++ b/test/OpenMP/parallel_for_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/parallel_for_linear_codegen.cpp b/test/OpenMP/parallel_for_linear_codegen.cpp index 2c3cfddbd5712..c565097093da7 100644 --- a/test/OpenMP/parallel_for_linear_codegen.cpp +++ b/test/OpenMP/parallel_for_linear_codegen.cpp @@ -3,6 +3,13 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/parallel_for_linear_messages.cpp b/test/OpenMP/parallel_for_linear_messages.cpp index e5f5b61873c89..080add692cab1 100644 --- a/test/OpenMP/parallel_for_linear_messages.cpp +++ b/test/OpenMP/parallel_for_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/parallel_for_loop_messages.cpp b/test/OpenMP/parallel_for_loop_messages.cpp index cf1f6195cc861..193c84e27fa59 100644 --- a/test/OpenMP/parallel_for_loop_messages.cpp +++ b/test/OpenMP/parallel_for_loop_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s + class S { int a; S() : a(0) {} diff --git a/test/OpenMP/parallel_for_messages.cpp b/test/OpenMP/parallel_for_messages.cpp index 7c4f926d6d2b6..9ab4bccfc9849 100644 --- a/test/OpenMP/parallel_for_messages.cpp +++ b/test/OpenMP/parallel_for_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -std=c++11 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -std=c++11 -o - %s + void foo() { } diff --git a/test/OpenMP/parallel_for_misc_messages.c b/test/OpenMP/parallel_for_misc_messages.c index 1a773be617c7d..2d12ba4507276 100644 --- a/test/OpenMP/parallel_for_misc_messages.c +++ b/test/OpenMP/parallel_for_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp parallel for'}} #pragma omp parallel for diff --git a/test/OpenMP/parallel_for_num_threads_messages.cpp b/test/OpenMP/parallel_for_num_threads_messages.cpp index 10a4e1bfb1aa9..6f57d6e8f4594 100644 --- a/test/OpenMP/parallel_for_num_threads_messages.cpp +++ b/test/OpenMP/parallel_for_num_threads_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/parallel_for_ordered_messages.cpp b/test/OpenMP/parallel_for_ordered_messages.cpp index 055fe1bba4237..381bc6ae1bad1 100644 --- a/test/OpenMP/parallel_for_ordered_messages.cpp +++ b/test/OpenMP/parallel_for_ordered_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/parallel_for_private_messages.cpp b/test/OpenMP/parallel_for_private_messages.cpp index cc1b79f4111b0..7efc6eeac8f74 100644 --- a/test/OpenMP/parallel_for_private_messages.cpp +++ b/test/OpenMP/parallel_for_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/parallel_for_proc_bind_messages.cpp b/test/OpenMP/parallel_for_proc_bind_messages.cpp index 0f9c47f5f29e1..0ffff89a8c956 100644 --- a/test/OpenMP/parallel_for_proc_bind_messages.cpp +++ b/test/OpenMP/parallel_for_proc_bind_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/parallel_for_reduction_messages.cpp b/test/OpenMP/parallel_for_reduction_messages.cpp index 6499ef593a3d3..5cfc2b9041ab1 100644 --- a/test/OpenMP/parallel_for_reduction_messages.cpp +++ b/test/OpenMP/parallel_for_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/parallel_for_schedule_messages.cpp b/test/OpenMP/parallel_for_schedule_messages.cpp index f446609615da4..5664529956b10 100644 --- a/test/OpenMP/parallel_for_schedule_messages.cpp +++ b/test/OpenMP/parallel_for_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/parallel_for_simd_aligned_messages.cpp b/test/OpenMP/parallel_for_simd_aligned_messages.cpp index eed80170039e0..75f4adb00b205 100644 --- a/test/OpenMP/parallel_for_simd_aligned_messages.cpp +++ b/test/OpenMP/parallel_for_simd_aligned_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp %s +// RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp-simd %s + struct B { static int ib[20]; // expected-note 0 {{'B::ib' declared here}} static constexpr int bfoo() { return 8; } diff --git a/test/OpenMP/parallel_for_simd_ast_print.cpp b/test/OpenMP/parallel_for_simd_ast_print.cpp index 2668961892f34..fd4153752af32 100644 --- a/test/OpenMP/parallel_for_simd_ast_print.cpp +++ b/test/OpenMP/parallel_for_simd_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -35,7 +39,7 @@ public: } }; -// CHECK: #pragma omp parallel for simd private(this->a) private(this->a) private(T::a) +// CHECK: #pragma omp parallel for simd private(this->a) private(this->a) private(T::a){{$}} // CHECK: #pragma omp parallel for simd private(this->a) private(this->a) // CHECK: #pragma omp parallel for simd private(this->a) private(this->a) private(this->S1::a) diff --git a/test/OpenMP/parallel_for_simd_codegen.cpp b/test/OpenMP/parallel_for_simd_codegen.cpp index 369ea17844e9c..e8cb651e2bc5e 100644 --- a/test/OpenMP/parallel_for_simd_codegen.cpp +++ b/test/OpenMP/parallel_for_simd_codegen.cpp @@ -2,6 +2,12 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -11,21 +17,21 @@ double *g_ptr; // CHECK-LABEL: define {{.*void}} @{{.*}}simple{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}}) void simple(float *a, float *b, float *c, float *d) { -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: [[K0:%.+]] = call {{.*}}i64 @{{.*}}get_val // CHECK-NEXT: store i64 [[K0]], i64* [[K_VAR:%[^,]+]] -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: store i32 12, i32* [[LIN_VAR:%[^,]+]] -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: store i32 -1, i32* [[A:%.+]], -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: store i32 -1, i32* [[R:%[^,]+]], -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( #pragma omp parallel for simd -// CHECK: call void @__kmpc_for_static_init_4(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) +// CHECK: call void @__kmpc_for_static_init_4(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) // CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]], // CHECK: [[CMP:%.+]] = icmp sgt i32 [[UB_VAL]], 5 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -61,7 +67,7 @@ void simple(float *a, float *b, float *c, float *d) { // br label %{{.+}}, !llvm.loop !{{.+}} } // CHECK: [[SIMPLE_LOOP1_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) long long k = get_val(); @@ -69,8 +75,8 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_VAR:%[^,]+]] // CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]] -// CHECK: call void @__kmpc_dispatch_init_4(%ident_t* {{.+}}, i32 %{{.+}}, i32 35, i32 0, i32 8, i32 1, i32 1) -// CHECK: [[NEXT:%.+]] = call i32 @__kmpc_dispatch_next_4(%ident_t* {{.+}}, i32 %{{.+}}, i32* %{{.+}}, i32* [[LB:%.+]], i32* [[UB:%.+]], i32* %{{.+}}) +// CHECK: call void @__kmpc_dispatch_init_4(%struct.ident_t* {{.+}}, i32 %{{.+}}, i32 35, i32 0, i32 8, i32 1, i32 1) +// CHECK: [[NEXT:%.+]] = call i32 @__kmpc_dispatch_next_4(%struct.ident_t* {{.+}}, i32 %{{.+}}, i32* %{{.+}}, i32* [[LB:%.+]], i32* [[UB:%.+]], i32* %{{.+}}) // CHECK: [[COND:%.+]] = icmp ne i32 [[NEXT]], 0 // CHECK: br i1 [[COND]], label %[[CONT:.+]], label %[[END:.+]] // CHECK: [[CONT]]: @@ -126,7 +132,7 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK: [[GLIN_LOAD:%.+]] = load double*, double** [[GLIN_VAR:%.+]], // CHECK-NEXT: store double* [[GLIN_LOAD]], double** [[GLIN_START:%[^,]+]] -// CHECK: call void @__kmpc_for_static_init_8u(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) +// CHECK: call void @__kmpc_for_static_init_8u(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) // CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]], // CHECK: [[CMP:%.+]] = icmp ugt i64 [[UB_VAL]], 3 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -173,7 +179,7 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK-NEXT: store i64 [[ADD3_2]], i64* [[OMP_IV3]] } // CHECK: [[SIMPLE_LOOP3_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) // // Linear start and step are used to calculate final value of the linear variables. // CHECK: [[LINSTART:.+]] = load i32, i32* [[LIN_START]] @@ -183,7 +189,7 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK: store double* {{.*}}[[GLIN_VAR]] #pragma omp parallel for simd -// CHECK: call void @__kmpc_for_static_init_4(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) +// CHECK: call void @__kmpc_for_static_init_4(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) // CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]], // CHECK: [[CMP:%.+]] = icmp sgt i32 [[UB_VAL]], 3 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -216,10 +222,10 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK-NEXT: store i32 [[ADD4_2]], i32* [[OMP_IV4]] } // CHECK: [[SIMPLE_LOOP4_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) #pragma omp parallel for simd -// CHECK: call void @__kmpc_for_static_init_4(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) +// CHECK: call void @__kmpc_for_static_init_4(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) // CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]], // CHECK: [[CMP:%.+]] = icmp sgt i32 [[UB_VAL]], 25 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -252,7 +258,7 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK-NEXT: store i32 [[ADD5_2]], i32* [[OMP_IV5]] } // CHECK: [[SIMPLE_LOOP5_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) // CHECK-NOT: mul i32 %{{.+}}, 10 #pragma omp parallel for simd @@ -263,7 +269,7 @@ void simple(float *a, float *b, float *c, float *d) { { A = -1; #pragma omp parallel for simd lastprivate(A) -// CHECK: call void @__kmpc_for_static_init_8(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) +// CHECK: call void @__kmpc_for_static_init_8(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) // CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]], // CHECK: [[CMP:%.+]] = icmp sgt i64 [[UB_VAL]], 6 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -300,7 +306,7 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK-NEXT: store i64 [[ADD7_2]], i64* [[OMP_IV7]] } // CHECK: [[SIMPLE_LOOP7_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) // CHECK: load i32, i32* // CHECK: icmp ne i32 %{{.+}}, 0 // CHECK: br i1 %{{.+}}, label @@ -313,7 +319,7 @@ void simple(float *a, float *b, float *c, float *d) { R = -1; // CHECK: store i32 1, i32* [[R_PRIV:%[^,]+]], #pragma omp parallel for simd reduction(*:R) -// CHECK: call void @__kmpc_for_static_init_8(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) +// CHECK: call void @__kmpc_for_static_init_8(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) // CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]], // CHECK: [[CMP:%.+]] = icmp sgt i64 [[UB_VAL]], 6 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -349,7 +355,7 @@ void simple(float *a, float *b, float *c, float *d) { // CHECK-NEXT: store i64 [[ADD8_2]], i64* [[OMP_IV8]] } // CHECK: [[SIMPLE_LOOP8_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) // CHECK: call i32 @__kmpc_reduce_nowait( // CHECK: [[R_PRIV_VAL:%.+]] = load i32, i32* [[R_PRIV]], // CHECK: [[RED:%.+]] = mul nsw i32 %{{.+}}, [[R_PRIV_VAL]] @@ -373,7 +379,7 @@ int templ1(T a, T *z) { // Instatiation templ1<float,2> // CHECK-LABEL: define {{.*i32}} @{{.*}}templ1{{.*}}(float {{.+}}, float* {{.+}}) -// CHECK: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( void inst_templ1() { float a; float z[100]; @@ -421,7 +427,7 @@ void iter_simple(IterDouble ia, IterDouble ib, IterDouble ic) { // CHECK-NEXT: store i32 [[DIFF5]], i32* [[OMP_LAST_IT:%[^,]+]]{{.+}} #pragma omp parallel for simd -// CHECK: call void @__kmpc_for_static_init_4(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) +// CHECK: call void @__kmpc_for_static_init_4(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) // CHECK-DAG: [[UB_VAL:%.+]] = load i32, i32* [[UB]], // CHECK-DAG: [[OMP_LAST_IT_VAL:%.+]] = load i32, i32* [[OMP_LAST_IT]], // CHECK: [[CMP:%.+]] = icmp sgt i32 [[UB_VAL]], [[OMP_LAST_IT_VAL]] @@ -462,7 +468,7 @@ void iter_simple(IterDouble ia, IterDouble ib, IterDouble ic) { // br label %{{.*}}, !llvm.loop ![[ITER_LOOP_ID]] } // CHECK: [[IT_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) // CHECK: ret void } @@ -473,7 +479,7 @@ void collapsed(float *a, float *b, float *c, float *d) { unsigned j; // middle loop couter, leads to unsigned icmp in loop header. // k declared in the loop init below short l; // inner loop counter -// CHECK: call void @__kmpc_for_static_init_4u(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) +// CHECK: call void @__kmpc_for_static_init_4u(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i32* [[LB:%[^,]+]], i32* [[UB:%[^,]+]], i32* [[STRIDE:%[^,]+]], i32 1, i32 1) // CHECK: [[UB_VAL:%.+]] = load i32, i32* [[UB]], // CHECK: [[CMP:%.+]] = icmp ugt i32 [[UB_VAL]], 119 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -537,7 +543,7 @@ void collapsed(float *a, float *b, float *c, float *d) { // CHECK: [[COLL1_END]]: } // i,j,l are updated; k is not updated. -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) // CHECK: store i32 3, i32* [[I:%[^,]+]] // CHECK: store i32 5, i32* [[I:%[^,]+]] // CHECK: store i16 9, i16* [[I:%[^,]+]] @@ -558,7 +564,7 @@ void widened(float *a, float *b, float *c, float *d) { // CHECK: [[MUL:%.+]] = mul nsw i64 2, %{{.+}} // CHECK-NEXT: [[SUB:%.+]] = sub nsw i64 [[MUL]], 1 // CHECK-NEXT: store i64 [[SUB]], i64* [[OMP_LAST_IT:%[^,]+]], -// CHECK: call void @__kmpc_for_static_init_8(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) +// CHECK: call void @__kmpc_for_static_init_8(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) // CHECK-DAG: [[UB_VAL:%.+]] = load i64, i64* [[UB]], // CHECK-DAG: [[OMP_LAST_IT_VAL:%.+]] = load i64, i64* [[OMP_LAST_IT]], // CHECK: [[CMP:%.+]] = icmp sgt i64 [[UB_VAL]], [[OMP_LAST_IT_VAL]] @@ -620,7 +626,7 @@ void widened(float *a, float *b, float *c, float *d) { // CHECK: ret void } -// CHECK: call void @__kmpc_for_static_init_8(%ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) +// CHECK: call void @__kmpc_for_static_init_8(%struct.ident_t* {{[^,]+}}, i32 %{{[^,]+}}, i32 34, i32* %{{[^,]+}}, i64* [[LB:%[^,]+]], i64* [[UB:%[^,]+]], i64* [[STRIDE:%[^,]+]], i64 1, i64 1) // CHECK: [[UB_VAL:%.+]] = load i64, i64* [[UB]], // CHECK: [[CMP:%.+]] = icmp sgt i64 [[UB_VAL]], 15 // CHECK: br i1 [[CMP]], label %[[TRUE:.+]], label %[[FALSE:[^,]+]] @@ -659,7 +665,7 @@ void widened(float *a, float *b, float *c, float *d) { // CHECK-NEXT: store i64 [[INC]], i64* // CHECK-NEXT: br label {{%.+}} // CHECK: [[T1_END]]: -// CHECK: call void @__kmpc_for_static_fini(%ident_t* {{.+}}, i32 %{{.+}}) +// CHECK: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32 %{{.+}}) // CHECK: ret void // // TERM_DEBUG-LABEL: bar @@ -679,5 +685,6 @@ void parallel_simd(float *a) { a[i] += bar(); } // TERM_DEBUG: !{{[0-9]+}} = !DILocation(line: [[@LINE-11]], +// TERM_DEBUG-NOT: line: 0, #endif // HEADER diff --git a/test/OpenMP/parallel_for_simd_collapse_messages.cpp b/test/OpenMP/parallel_for_simd_collapse_messages.cpp index c7effbddeb28c..d23086b9e5398 100644 --- a/test/OpenMP/parallel_for_simd_collapse_messages.cpp +++ b/test/OpenMP/parallel_for_simd_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/parallel_for_simd_copyin_messages.cpp b/test/OpenMP/parallel_for_simd_copyin_messages.cpp index 1e6fdc9c31b1c..6b492b13f7b55 100644 --- a/test/OpenMP/parallel_for_simd_copyin_messages.cpp +++ b/test/OpenMP/parallel_for_simd_copyin_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -o - %s + void foo() { } diff --git a/test/OpenMP/parallel_for_simd_default_messages.cpp b/test/OpenMP/parallel_for_simd_default_messages.cpp index 9bd8113c8d556..2fccb5634b440 100644 --- a/test/OpenMP/parallel_for_simd_default_messages.cpp +++ b/test/OpenMP/parallel_for_simd_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/parallel_for_simd_firstprivate_messages.cpp b/test/OpenMP/parallel_for_simd_firstprivate_messages.cpp index 7a08572f7d1a9..e1e01aa661b6b 100644 --- a/test/OpenMP/parallel_for_simd_firstprivate_messages.cpp +++ b/test/OpenMP/parallel_for_simd_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/parallel_for_simd_if_messages.cpp b/test/OpenMP/parallel_for_simd_if_messages.cpp index 7842524fa40db..08f8b466ed35f 100644 --- a/test/OpenMP/parallel_for_simd_if_messages.cpp +++ b/test/OpenMP/parallel_for_simd_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 %s + void foo() { } diff --git a/test/OpenMP/parallel_for_simd_lastprivate_messages.cpp b/test/OpenMP/parallel_for_simd_lastprivate_messages.cpp index 8670e55c6e2bf..84e0896f1ff27 100644 --- a/test/OpenMP/parallel_for_simd_lastprivate_messages.cpp +++ b/test/OpenMP/parallel_for_simd_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/parallel_for_simd_linear_messages.cpp b/test/OpenMP/parallel_for_simd_linear_messages.cpp index fc1895a36854a..a6bcf64806f6b 100644 --- a/test/OpenMP/parallel_for_simd_linear_messages.cpp +++ b/test/OpenMP/parallel_for_simd_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/parallel_for_simd_loop_messages.cpp b/test/OpenMP/parallel_for_simd_loop_messages.cpp index 7feb4b0eb6eed..d9d05cccc54cd 100644 --- a/test/OpenMP/parallel_for_simd_loop_messages.cpp +++ b/test/OpenMP/parallel_for_simd_loop_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s + class S { int a; S() : a(0) {} diff --git a/test/OpenMP/parallel_for_simd_messages.cpp b/test/OpenMP/parallel_for_simd_messages.cpp index 18f25fa0f66ac..f1d4c5b935be7 100644 --- a/test/OpenMP/parallel_for_simd_messages.cpp +++ b/test/OpenMP/parallel_for_simd_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -o - %s + void foo() { } diff --git a/test/OpenMP/parallel_for_simd_misc_messages.c b/test/OpenMP/parallel_for_simd_misc_messages.c index 378c48ff8a4c4..d3565bfa4f195 100644 --- a/test/OpenMP/parallel_for_simd_misc_messages.c +++ b/test/OpenMP/parallel_for_simd_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp parallel for simd'}} #pragma omp parallel for simd diff --git a/test/OpenMP/parallel_for_simd_num_threads_messages.cpp b/test/OpenMP/parallel_for_simd_num_threads_messages.cpp index 940565cc10c33..06460c397de3d 100644 --- a/test/OpenMP/parallel_for_simd_num_threads_messages.cpp +++ b/test/OpenMP/parallel_for_simd_num_threads_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/parallel_for_simd_private_messages.cpp b/test/OpenMP/parallel_for_simd_private_messages.cpp index a33b35d57d1ef..5d263d6cc2765 100644 --- a/test/OpenMP/parallel_for_simd_private_messages.cpp +++ b/test/OpenMP/parallel_for_simd_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/parallel_for_simd_proc_bind_messages.cpp b/test/OpenMP/parallel_for_simd_proc_bind_messages.cpp index a05b150f56946..26cc729964dc7 100644 --- a/test/OpenMP/parallel_for_simd_proc_bind_messages.cpp +++ b/test/OpenMP/parallel_for_simd_proc_bind_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/parallel_for_simd_reduction_messages.cpp b/test/OpenMP/parallel_for_simd_reduction_messages.cpp index 8ff246e35f6ac..32222ac5a53f3 100644 --- a/test/OpenMP/parallel_for_simd_reduction_messages.cpp +++ b/test/OpenMP/parallel_for_simd_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -o - %s + void foo() { } diff --git a/test/OpenMP/parallel_for_simd_safelen_messages.cpp b/test/OpenMP/parallel_for_simd_safelen_messages.cpp index 3e643c619c6fc..49ebf428f3d02 100644 --- a/test/OpenMP/parallel_for_simd_safelen_messages.cpp +++ b/test/OpenMP/parallel_for_simd_safelen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/parallel_for_simd_schedule_messages.cpp b/test/OpenMP/parallel_for_simd_schedule_messages.cpp index 4135427ce10b1..29e46a7a0c794 100644 --- a/test/OpenMP/parallel_for_simd_schedule_messages.cpp +++ b/test/OpenMP/parallel_for_simd_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/parallel_for_simd_simdlen_messages.cpp b/test/OpenMP/parallel_for_simd_simdlen_messages.cpp index fa9e0d6802a9c..167644e3886b7 100644 --- a/test/OpenMP/parallel_for_simd_simdlen_messages.cpp +++ b/test/OpenMP/parallel_for_simd_simdlen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/parallel_if_codegen.cpp b/test/OpenMP/parallel_if_codegen.cpp index 87dcf1dde943c..49e4bd402ce87 100644 --- a/test/OpenMP/parallel_if_codegen.cpp +++ b/test/OpenMP/parallel_if_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -24,12 +29,12 @@ void gtid_test() { } // CHECK: define internal {{.*}}void [[GTID_TEST_REGION1]](i{{.+}}* noalias [[GTID_PARAM:%.+]], i32* noalias +// CHECK: store i32 0, i32* [[ZERO_ADDR:%.+]], // CHECK: store i{{[0-9]+}}* [[GTID_PARAM]], i{{[0-9]+}}** [[GTID_ADDR_REF:%.+]], // CHECK: [[GTID_ADDR:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_REF]] // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_ADDR]] // CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i{{.+}} [[GTID]]) -// CHECK: [[GTID_ADDR:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_REF]] -// CHECK: call void [[GTID_TEST_REGION2:@.+]](i{{[0-9]+}}* [[GTID_ADDR]] +// CHECK: call void [[GTID_TEST_REGION2:@.+]](i{{[0-9]+}}* [[ZERO_ADDR]] // CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i{{.+}} [[GTID]]) // CHECK: ret void @@ -51,12 +56,13 @@ int tmain(T Arg) { // CHECK-LABEL: define {{.*}}i{{[0-9]+}} @main() int main() { // CHECK: [[GTID:%.+]] = call {{.*}}i32 @__kmpc_global_thread_num( +// CHECK: store i32 0, i32* [[ZERO_ADDR2:%.+]], +// CHECK: store i32 0, i32* [[ZERO_ADDR1:%.+]], // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 0, void {{.+}}* [[CAP_FN4:@.+]] to void #pragma omp parallel if (true) fn4(); // CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) -// CHECK: store i32 [[GTID]], i32* [[GTID_ADDR:%.+]], -// CHECK: call void [[CAP_FN5:@.+]](i32* [[GTID_ADDR]], +// CHECK: call void [[CAP_FN5:@.+]](i32* [[ZERO_ADDR1]], i32* [[ZERO_ADDR1]]) // CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) #pragma omp parallel if (false) fn5(); @@ -67,8 +73,7 @@ int main() { // CHECK: br label %[[OMP_END:.+]] // CHECK: [[OMP_ELSE]] // CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) -// CHECK: store i32 [[GTID]], i32* [[GTID_ADDR:%.+]], -// CHECK: call void [[CAP_FN6]](i32* [[GTID_ADDR]], +// CHECK: call void [[CAP_FN6]](i32* [[ZERO_ADDR2]], i32* [[ZERO_ADDR2]]) // CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) // CHECK: br label %[[OMP_END]] // CHECK: [[OMP_END]] @@ -92,10 +97,11 @@ int main() { // CHECK-LABEL: define {{.+}} @{{.+}}tmain // CHECK: [[GTID:%.+]] = call {{.*}}i32 @__kmpc_global_thread_num( +// CHECK: store i32 0, i32* [[ZERO_ADDR2:%.+]], +// CHECK: store i32 0, i32* [[ZERO_ADDR1:%.+]], // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 0, void {{.+}}* [[CAP_FN1:@.+]] to void // CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) -// CHECK: store i32 [[GTID]], i32* [[GTID_ADDR:%.+]], -// CHECK: call void [[CAP_FN2:@.+]](i32* [[GTID_ADDR]], +// CHECK: call void [[CAP_FN2:@.+]](i32* [[ZERO_ADDR1]], i32* [[ZERO_ADDR1]]) // CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) // CHECK: br i1 %{{.+}}, label %[[OMP_THEN:.+]], label %[[OMP_ELSE:.+]] // CHECK: [[OMP_THEN]] @@ -103,8 +109,7 @@ int main() { // CHECK: br label %[[OMP_END:.+]] // CHECK: [[OMP_ELSE]] // CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) -// CHECK: store i32 [[GTID]], i32* [[GTID_ADDR:%.+]], -// CHECK: call void [[CAP_FN3]](i32* [[GTID_ADDR]], +// CHECK: call void [[CAP_FN3]](i32* [[ZERO_ADDR2]], i32* [[ZERO_ADDR2]]) // CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) // CHECK: br label %[[OMP_END]] // CHECK: [[OMP_END]] diff --git a/test/OpenMP/parallel_if_messages.cpp b/test/OpenMP/parallel_if_messages.cpp index 094e83b631451..d0a4c6d1108df 100644 --- a/test/OpenMP/parallel_if_messages.cpp +++ b/test/OpenMP/parallel_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/parallel_messages.cpp b/test/OpenMP/parallel_messages.cpp index 4db55a019549c..8b0c0353c20d6 100644 --- a/test/OpenMP/parallel_messages.cpp +++ b/test/OpenMP/parallel_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -std=c++11 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -std=c++11 -o - %s + void foo() { } @@ -29,6 +31,8 @@ int main(int argc, char **argv) { foo(); L1: foo(); + #pragma omp parallel ordered // expected-error {{unexpected OpenMP clause 'ordered' in directive '#pragma omp parallel'}} + ; #pragma omp parallel ; #pragma omp parallel diff --git a/test/OpenMP/parallel_num_threads_codegen.cpp b/test/OpenMP/parallel_num_threads_codegen.cpp index c5f11bde19aed..9ec712f83c530 100644 --- a/test/OpenMP/parallel_num_threads_codegen.cpp +++ b/test/OpenMP/parallel_num_threads_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple %itanium_abi_triple -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple %itanium_abi_triple -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/parallel_num_threads_messages.cpp b/test/OpenMP/parallel_num_threads_messages.cpp index adcf6ca4253fd..310e7ccfa4e8d 100644 --- a/test/OpenMP/parallel_num_threads_messages.cpp +++ b/test/OpenMP/parallel_num_threads_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/parallel_private_codegen.cpp b/test/OpenMP/parallel_private_codegen.cpp index fbb9b18d5df5e..33a5b4cba0e6c 100644 --- a/test/OpenMP/parallel_private_codegen.cpp +++ b/test/OpenMP/parallel_private_codegen.cpp @@ -3,6 +3,13 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -96,7 +103,7 @@ int main() { static int sivar; SS ss(sivar); #ifdef LAMBDA - // LAMBDA: [[G:@.+]] = global i{{[0-9]+}} 1212, + // LAMBDA: [[G:@.+]] = {{(dso_local )?}}global i{{[0-9]+}} 1212, // LAMBDA-LABEL: @main // LAMBDA: alloca [[SS_TY]], // LAMBDA: alloca [[CAP_TY:%.+]], @@ -168,7 +175,7 @@ int main() { }(); return 0; #elif defined(BLOCKS) - // BLOCKS: [[G:@.+]] = global i{{[0-9]+}} 1212, + // BLOCKS: [[G:@.+]] = {{(dso_local )?}}global i{{[0-9]+}} 1212, // BLOCKS-LABEL: @main // BLOCKS: call // BLOCKS: call{{.*}} void {{%.+}}(i8 diff --git a/test/OpenMP/parallel_private_messages.cpp b/test/OpenMP/parallel_private_messages.cpp index ab535b475265d..4adee55e341c4 100644 --- a/test/OpenMP/parallel_private_messages.cpp +++ b/test/OpenMP/parallel_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/parallel_proc_bind_codegen.cpp b/test/OpenMP/parallel_proc_bind_codegen.cpp index 34a64de7465da..8b9e09191b24b 100644 --- a/test/OpenMP/parallel_proc_bind_codegen.cpp +++ b/test/OpenMP/parallel_proc_bind_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple %itanium_abi_triple -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple %itanium_abi_triple -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple %itanium_abi_triple -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -38,14 +43,14 @@ int main() { // CHECK-LABEL: @main // CHECK: [[GTID:%.+]] = call {{.*}}i32 @__kmpc_global_thread_num([[IDENT_T_TY]]* [[DEF_LOC_2]]) // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID]], i32 4) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID]], i32 3) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK-LABEL: @{{.+}}tmain // CHECK: [[GTID:%.+]] = call {{.*}}i32 @__kmpc_global_thread_num([[IDENT_T_TY]]* [[DEF_LOC_2]]) // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID]], i32 2) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: ret i32 0 // CHECK-NEXT: } diff --git a/test/OpenMP/parallel_proc_bind_messages.cpp b/test/OpenMP/parallel_proc_bind_messages.cpp index 78ba297fdf5dd..91be0f1a8d83b 100644 --- a/test/OpenMP/parallel_proc_bind_messages.cpp +++ b/test/OpenMP/parallel_proc_bind_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/parallel_reduction_codegen.cpp b/test/OpenMP/parallel_reduction_codegen.cpp index c4c0ff473a59d..c064ed0712107 100644 --- a/test/OpenMP/parallel_reduction_codegen.cpp +++ b/test/OpenMP/parallel_reduction_codegen.cpp @@ -3,6 +3,13 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -407,7 +414,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // var1 = var1.operator &&(var1_reduction); // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_REF]]) @@ -423,7 +430,7 @@ int main() { // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // t_var1 = min(t_var1, t_var1_reduction); // CHECK: [[T_VAR1_VAL:%.+]] = load float, float* [[T_VAR1_REF]], @@ -461,7 +468,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_REF]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_PRIV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // CHECK: call void @__kmpc_end_critical( // var1 = var1.operator &&(var1_reduction); @@ -479,7 +486,7 @@ int main() { // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // CHECK: call void @__kmpc_end_critical( // t_var1 = min(t_var1, t_var1_reduction); @@ -561,7 +568,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_FLOAT_TY]]* @{{.+}}([[S_FLOAT_TY]]* [[VAR_LHS]], [[S_FLOAT_TY]]* dereferenceable(4) [[VAR_RHS]]) // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_LHS]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // var1_lhs = var1_lhs.operator &&(var1_rhs); // CHECK: [[TO_FLOAT:%.+]] = call float @{{.+}}([[S_FLOAT_TY]]* [[VAR1_LHS]]) @@ -577,7 +584,7 @@ int main() { // CHECK: call void @{{.+}}([[S_FLOAT_TY]]* [[COND_LVALUE:%.+]], float [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR1_LHS]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_FLOAT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load float, float* [[T_VAR1_LHS]], @@ -730,7 +737,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 128 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // var1 = var1.operator &&(var1_reduction); // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) @@ -745,7 +752,7 @@ int main() { // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 128 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // t_var1 = min(t_var1, t_var1_reduction); // CHECK: [[T_VAR1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_REF]], @@ -771,7 +778,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 128 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // CHECK: call void @__kmpc_end_critical( // var1 = var1.operator &&(var1_reduction); @@ -789,7 +796,7 @@ int main() { // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 128 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // CHECK: call void @__kmpc_end_critical( // t_var1 = min(t_var1, t_var1_reduction); @@ -857,7 +864,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_LHS]], [[S_INT_TY]]* dereferenceable(4) [[VAR_RHS]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_LHS]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 128 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // var1_lhs = var1_lhs.operator &&(var1_rhs); // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_LHS]]) @@ -873,7 +880,7 @@ int main() { // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_LHS]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 128 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_LHS]], diff --git a/test/OpenMP/parallel_reduction_messages.c b/test/OpenMP/parallel_reduction_messages.c new file mode 100644 index 0000000000000..f88f8e0564953 --- /dev/null +++ b/test/OpenMP/parallel_reduction_messages.c @@ -0,0 +1,11 @@ +// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 150 -o - %s + +int incomplete[]; + +void test() { +#pragma omp parallel reduction(+ : incomplete) // expected-error {{a reduction list item with incomplete type 'int []'}} + ; +} + +// complete to suppress an additional warning, but it's too late for pragmas +int incomplete[3]; diff --git a/test/OpenMP/parallel_reduction_messages.cpp b/test/OpenMP/parallel_reduction_messages.cpp index 96f4b58ec8a4d..b3619bb9ad4dd 100644 --- a/test/OpenMP/parallel_reduction_messages.cpp +++ b/test/OpenMP/parallel_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/parallel_sections_ast_print.cpp b/test/OpenMP/parallel_sections_ast_print.cpp index da3bd48201a00..8c0bd0bae5f8f 100644 --- a/test/OpenMP/parallel_sections_ast_print.cpp +++ b/test/OpenMP/parallel_sections_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -54,7 +58,7 @@ T tmain(T argc, T *argv) { // CHECK-NEXT: T b = argc, c, d, e, f, g; // CHECK-NEXT: static T a; // CHECK-NEXT: S<T> s; -// CHECK-NEXT: #pragma omp parallel sections +// CHECK-NEXT: #pragma omp parallel sections{{$}} // CHECK-NEXT: { // CHECK-NEXT: a = 2; // CHECK-NEXT: } diff --git a/test/OpenMP/parallel_sections_codegen.cpp b/test/OpenMP/parallel_sections_codegen.cpp index a26147303611d..eadc4937203a3 100644 --- a/test/OpenMP/parallel_sections_codegen.cpp +++ b/test/OpenMP/parallel_sections_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -emit-llvm -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -o - %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -triple x86_64-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -include-pch %t -fsyntax-only -verify %s -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -emit-llvm -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -o - %s | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -triple x86_64-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -include-pch %t -fsyntax-only -verify %s -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/parallel_sections_copyin_messages.cpp b/test/OpenMP/parallel_sections_copyin_messages.cpp index 62b5d05db0653..da0ef5cb1e041 100644 --- a/test/OpenMP/parallel_sections_copyin_messages.cpp +++ b/test/OpenMP/parallel_sections_copyin_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/parallel_sections_default_messages.cpp b/test/OpenMP/parallel_sections_default_messages.cpp index 9b2bdad627aac..a62b86ff34a8c 100644 --- a/test/OpenMP/parallel_sections_default_messages.cpp +++ b/test/OpenMP/parallel_sections_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/parallel_sections_firstprivate_messages.cpp b/test/OpenMP/parallel_sections_firstprivate_messages.cpp index b733aabb7f252..f8370f26b4d40 100644 --- a/test/OpenMP/parallel_sections_firstprivate_messages.cpp +++ b/test/OpenMP/parallel_sections_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/parallel_sections_if_messages.cpp b/test/OpenMP/parallel_sections_if_messages.cpp index ce067a254fea7..b03ed3551653d 100644 --- a/test/OpenMP/parallel_sections_if_messages.cpp +++ b/test/OpenMP/parallel_sections_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/parallel_sections_lastprivate_messages.cpp b/test/OpenMP/parallel_sections_lastprivate_messages.cpp index eccbfc5e7e02b..865dda01a330e 100644 --- a/test/OpenMP/parallel_sections_lastprivate_messages.cpp +++ b/test/OpenMP/parallel_sections_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/parallel_sections_messages.cpp b/test/OpenMP/parallel_sections_messages.cpp index 1e71a305069b5..459953acd9a2b 100644 --- a/test/OpenMP/parallel_sections_messages.cpp +++ b/test/OpenMP/parallel_sections_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -std=c++11 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -std=c++11 -o - %s + void foo() { } diff --git a/test/OpenMP/parallel_sections_misc_messages.c b/test/OpenMP/parallel_sections_misc_messages.c index 203d12c52aabe..780837a15f9c7 100644 --- a/test/OpenMP/parallel_sections_misc_messages.c +++ b/test/OpenMP/parallel_sections_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + void foo(); // expected-error@+1 {{unexpected OpenMP directive '#pragma omp parallel sections'}} diff --git a/test/OpenMP/parallel_sections_num_threads_messages.cpp b/test/OpenMP/parallel_sections_num_threads_messages.cpp index cda3841ace971..42bdee25e464a 100644 --- a/test/OpenMP/parallel_sections_num_threads_messages.cpp +++ b/test/OpenMP/parallel_sections_num_threads_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/parallel_sections_private_messages.cpp b/test/OpenMP/parallel_sections_private_messages.cpp index 40b0138b53967..75e3f173ae83e 100644 --- a/test/OpenMP/parallel_sections_private_messages.cpp +++ b/test/OpenMP/parallel_sections_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/parallel_sections_proc_bind_messages.cpp b/test/OpenMP/parallel_sections_proc_bind_messages.cpp index 79796daa8ce6b..bb173b61e5272 100644 --- a/test/OpenMP/parallel_sections_proc_bind_messages.cpp +++ b/test/OpenMP/parallel_sections_proc_bind_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/parallel_sections_reduction_messages.cpp b/test/OpenMP/parallel_sections_reduction_messages.cpp index 26b3a5355c19c..bbd3737de5c19 100644 --- a/test/OpenMP/parallel_sections_reduction_messages.cpp +++ b/test/OpenMP/parallel_sections_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/parallel_sections_shared_messages.cpp b/test/OpenMP/parallel_sections_shared_messages.cpp index 2b83597111192..1ab923c3308c6 100644 --- a/test/OpenMP/parallel_sections_shared_messages.cpp +++ b/test/OpenMP/parallel_sections_shared_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/parallel_shared_messages.cpp b/test/OpenMP/parallel_shared_messages.cpp index f6f4573a66938..915b6a2cfc626 100644 --- a/test/OpenMP/parallel_shared_messages.cpp +++ b/test/OpenMP/parallel_shared_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/predefined_macro.c b/test/OpenMP/predefined_macro.c index e18c3d26d4c22..b27b32b561fd5 100644 --- a/test/OpenMP/predefined_macro.c +++ b/test/OpenMP/predefined_macro.c @@ -1,5 +1,8 @@ // RUN: %clang_cc1 -fopenmp -verify -DFOPENMP -o - %s // RUN: %clang_cc1 -verify -o - %s + +// RUN: %clang_cc1 -fopenmp-simd -verify -o - %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -verify -o - %s // expected-no-diagnostics #ifdef FOPENMP // -fopenmp option is specified diff --git a/test/OpenMP/report_default_DSA.cpp b/test/OpenMP/report_default_DSA.cpp index d14cd5cbe9a47..ba620a4ce8ebf 100644 --- a/test/OpenMP/report_default_DSA.cpp +++ b/test/OpenMP/report_default_DSA.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 %s + void foo(int x, int n) { double vec[n]; for (int iter = 0; iter < x; iter++) { diff --git a/test/OpenMP/schedule_codegen.cpp b/test/OpenMP/schedule_codegen.cpp index bd5ec86b29175..394eae4a55e66 100644 --- a/test/OpenMP/schedule_codegen.cpp +++ b/test/OpenMP/schedule_codegen.cpp @@ -1,5 +1,8 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm -fexceptions -fcxx-exceptions -o - %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm -fexceptions -fcxx-exceptions -o - %s | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + int main() { // CHECK: @__kmpc_for_static_init // CHECK-NOT: !llvm.mem.parallel_loop_access diff --git a/test/OpenMP/sections_ast_print.cpp b/test/OpenMP/sections_ast_print.cpp index 7fecd5b777710..15c8c26e8ce6c 100644 --- a/test/OpenMP/sections_ast_print.cpp +++ b/test/OpenMP/sections_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -19,7 +23,7 @@ T tmain(T argc) { foo(); } // CHECK-NEXT: #pragma omp parallel - // CHECK-NEXT: #pragma omp sections private(argc,b) firstprivate(c,d) lastprivate(d,f) reduction(-: g) nowait + // CHECK-NEXT: #pragma omp sections private(argc,b) firstprivate(c,d) lastprivate(d,f) reduction(-: g) nowait{{$}} // CHECK-NEXT: { // CHECK-NEXT: foo(); // CHECK-NEXT: } @@ -42,7 +46,7 @@ int main(int argc, char **argv) { // CHECK-NEXT: #pragma omp parallel // CHECK-NEXT: #pragma omp sections private(argc,b) firstprivate(argv,c) lastprivate(d,f) reduction(+: g) nowait // CHECK-NEXT: { - // CHECK-NEXT: #pragma omp section + // CHECK-NEXT: #pragma omp section{{$}} // CHECK-NEXT: foo(); // CHECK-NEXT: #pragma omp section // CHECK-NEXT: foo(); diff --git a/test/OpenMP/sections_codegen.cpp b/test/OpenMP/sections_codegen.cpp index 94ded37db95d7..ba918c385fc30 100644 --- a/test/OpenMP/sections_codegen.cpp +++ b/test/OpenMP/sections_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -emit-llvm -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -o - %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -triple x86_64-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -include-pch %t -fsyntax-only -verify %s -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -emit-llvm -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -o - %s | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -triple x86_64-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -include-pch %t -fsyntax-only -verify %s -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/sections_firstprivate_codegen.cpp b/test/OpenMP/sections_firstprivate_codegen.cpp index 227bb076374f1..467c10f242c66 100644 --- a/test/OpenMP/sections_firstprivate_codegen.cpp +++ b/test/OpenMP/sections_firstprivate_codegen.cpp @@ -3,6 +3,13 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -218,7 +225,7 @@ int main() { // firstprivate vec(vec) // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* bitcast ([2 x i{{[0-9]+}}]* [[VEC]] to i8*), +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST]], i8* align {{[0-9]+}} bitcast ([2 x i{{[0-9]+}}]* [[VEC]] to i8*), // firstprivate s_arr(s_arr) // CHECK: [[S_ARR_PRIV_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 @@ -285,7 +292,7 @@ int main() { // firstprivate vec(vec) // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_REF]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* [[VEC_SRC]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST]], i8* align {{[0-9]+}} [[VEC_SRC]], // firstprivate s_arr(s_arr) // CHECK: [[S_ARR_PRIV_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 diff --git a/test/OpenMP/sections_firstprivate_messages.cpp b/test/OpenMP/sections_firstprivate_messages.cpp index cd2b4b883873a..425125c11a09f 100644 --- a/test/OpenMP/sections_firstprivate_messages.cpp +++ b/test/OpenMP/sections_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/sections_lastprivate_codegen.cpp b/test/OpenMP/sections_lastprivate_codegen.cpp index c9a224db23709..bb6a54f00aefc 100644 --- a/test/OpenMP/sections_lastprivate_codegen.cpp +++ b/test/OpenMP/sections_lastprivate_codegen.cpp @@ -3,6 +3,13 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -335,7 +342,7 @@ int main() { // original vec[]=private_vec[]; // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_REF]] to i8* // CHECK: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* [[VEC_SRC]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST]], i8* align {{[0-9]+}} [[VEC_SRC]], // original s_arr[]=private_s_arr[]; // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 diff --git a/test/OpenMP/sections_lastprivate_messages.cpp b/test/OpenMP/sections_lastprivate_messages.cpp index 01bb13de5510c..e24b58fa29bc9 100644 --- a/test/OpenMP/sections_lastprivate_messages.cpp +++ b/test/OpenMP/sections_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/sections_misc_messages.c b/test/OpenMP/sections_misc_messages.c index da20aa18aa122..ba030859e5756 100644 --- a/test/OpenMP/sections_misc_messages.c +++ b/test/OpenMP/sections_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + void foo(); // expected-error@+1 {{unexpected OpenMP directive '#pragma omp sections'}} diff --git a/test/OpenMP/sections_private_codegen.cpp b/test/OpenMP/sections_private_codegen.cpp index 337a0bfe3ec17..6a6d4fe6481dc 100644 --- a/test/OpenMP/sections_private_codegen.cpp +++ b/test/OpenMP/sections_private_codegen.cpp @@ -3,6 +3,13 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -40,7 +47,7 @@ T tmain() { int main() { static int sivar; #ifdef LAMBDA - // LAMBDA: [[G:@.+]] = global double + // LAMBDA: [[G:@.+]] = {{(dso_local )?}}global double // LAMBDA-LABEL: @main // LAMBDA: call{{.*}} void [[OUTER_LAMBDA:@.+]]( [&]() { @@ -85,7 +92,7 @@ int main() { }(); return 0; #elif defined(BLOCKS) - // BLOCKS: [[G:@.+]] = global double + // BLOCKS: [[G:@.+]] = {{(dso_local )?}}global double // BLOCKS-LABEL: @main // BLOCKS: call {{.*}}void {{%.+}}(i8 ^{ diff --git a/test/OpenMP/sections_private_messages.cpp b/test/OpenMP/sections_private_messages.cpp index 27bb3136e7201..81e3eb4779b9a 100644 --- a/test/OpenMP/sections_private_messages.cpp +++ b/test/OpenMP/sections_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/sections_reduction_codegen.cpp b/test/OpenMP/sections_reduction_codegen.cpp index 8c474d575e0aa..43b08b1160d3c 100644 --- a/test/OpenMP/sections_reduction_codegen.cpp +++ b/test/OpenMP/sections_reduction_codegen.cpp @@ -1,8 +1,15 @@ -// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s -// RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap -check-prefix=LAMBDA %s +// RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap -check-prefix=BLOCKS %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -296,7 +303,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // var1 = var1.operator &&(var1_reduction); // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_REF]]) @@ -312,7 +319,7 @@ int main() { // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // t_var1 = min(t_var1, t_var1_reduction); // CHECK: [[T_VAR1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_REF]], @@ -338,7 +345,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_REF]], [[S_INT_TY]]* dereferenceable(4) [[VAR_PRIV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // CHECK: call void @__kmpc_end_critical( // var1 = var1.operator &&(var1_reduction); @@ -356,7 +363,7 @@ int main() { // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_REF]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // CHECK: call void @__kmpc_end_critical( // t_var1 = min(t_var1, t_var1_reduction); @@ -423,7 +430,7 @@ int main() { // CHECK: [[UP:%.+]] = call dereferenceable(4) [[S_INT_TY]]* @{{.+}}([[S_INT_TY]]* [[VAR_LHS]], [[S_INT_TY]]* dereferenceable(4) [[VAR_RHS]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR_LHS]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[UP]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // var1_lhs = var1_lhs.operator &&(var1_rhs); // CHECK: [[TO_INT:%.+]] = call i{{[0-9]+}} @{{.+}}([[S_INT_TY]]* [[VAR1_LHS]]) @@ -439,7 +446,7 @@ int main() { // CHECK: call void @{{.+}}([[S_INT_TY]]* [[COND_LVALUE:%.+]], i32 [[CONV]]) // CHECK: [[BC1:%.+]] = bitcast [[S_INT_TY]]* [[VAR1_LHS]] to i8* // CHECK: [[BC2:%.+]] = bitcast [[S_INT_TY]]* [[COND_LVALUE]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[BC1]], i8* [[BC2]], i64 4, i32 4, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[BC1]], i8* align 4 [[BC2]], i64 4, i1 false) // t_var1_lhs = min(t_var1_lhs, t_var1_rhs); // CHECK: [[T_VAR1_LHS_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR1_LHS]], diff --git a/test/OpenMP/sections_reduction_messages.cpp b/test/OpenMP/sections_reduction_messages.cpp index 58b22a5a61d04..bf302cf7b335f 100644 --- a/test/OpenMP/sections_reduction_messages.cpp +++ b/test/OpenMP/sections_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/simd_aligned_messages.cpp b/test/OpenMP/simd_aligned_messages.cpp index d936fed014392..7621d7de096b2 100644 --- a/test/OpenMP/simd_aligned_messages.cpp +++ b/test/OpenMP/simd_aligned_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp %s +// RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp-simd %s + struct B { static int ib[20]; // expected-note 0 {{'B::ib' declared here}} static constexpr int bfoo() { return 8; } diff --git a/test/OpenMP/simd_ast_print.cpp b/test/OpenMP/simd_ast_print.cpp index 9fcd459f4956a..ad16fe2de9a42 100644 --- a/test/OpenMP/simd_ast_print.cpp +++ b/test/OpenMP/simd_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -34,7 +38,7 @@ public: } }; -// CHECK: #pragma omp simd aligned(this->a) +// CHECK: #pragma omp simd aligned(this->a){{$}} // CHECK: #pragma omp simd aligned(this->b: 8) // CHECK: #pragma omp simd aligned(this->a) @@ -149,7 +153,7 @@ int main (int argc, char **argv) { static int *a; // CHECK: static int *a; #pragma omp simd -// CHECK-NEXT: #pragma omp simd +// CHECK-NEXT: #pragma omp simd{{$}} for (int i=0; i < 2; ++i)*a=2; // CHECK-NEXT: for (int i = 0; i < 2; ++i) // CHECK-NEXT: *a = 2; diff --git a/test/OpenMP/simd_codegen.cpp b/test/OpenMP/simd_codegen.cpp index 47f2433f0bbc1..883273099ad83 100644 --- a/test/OpenMP/simd_codegen.cpp +++ b/test/OpenMP/simd_codegen.cpp @@ -2,6 +2,11 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck --check-prefix=TERM_DEBUG %s // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -641,25 +646,26 @@ void parallel_simd(float *a) { // TERM_DEBUG: !{{[0-9]+}} = !DILocation(line: [[@LINE-11]], // CHECK-LABEL: S8 -// CHECK: ptrtoint [[SS_TY]]* %{{.+}} to i64 -// CHECK-NEXT: and i64 %{{.+}}, 15 -// CHECK-NEXT: icmp eq i64 %{{.+}}, 0 -// CHECK-NEXT: call void @llvm.assume(i1 - -// CHECK: ptrtoint [[SS_TY]]* %{{.+}} to i64 -// CHECK-NEXT: and i64 %{{.+}}, 7 -// CHECK-NEXT: icmp eq i64 %{{.+}}, 0 -// CHECK-NEXT: call void @llvm.assume(i1 - -// CHECK: ptrtoint [[SS_TY]]* %{{.+}} to i64 -// CHECK-NEXT: and i64 %{{.+}}, 15 -// CHECK-NEXT: icmp eq i64 %{{.+}}, 0 -// CHECK-NEXT: call void @llvm.assume(i1 - -// CHECK: ptrtoint [[SS_TY]]* %{{.+}} to i64 -// CHECK-NEXT: and i64 %{{.+}}, 3 -// CHECK-NEXT: icmp eq i64 %{{.+}}, 0 -// CHECK-NEXT: call void @llvm.assume(i1 +// CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64 +// CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64 +// CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64 +// CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64 + +// CHECK-DAG: and i64 %{{.+}}, 15 +// CHECK-DAG: icmp eq i64 %{{.+}}, 0 +// CHECK-DAG: call void @llvm.assume(i1 + +// CHECK-DAG: and i64 %{{.+}}, 7 +// CHECK-DAG: icmp eq i64 %{{.+}}, 0 +// CHECK-DAG: call void @llvm.assume(i1 + +// CHECK-DAG: and i64 %{{.+}}, 15 +// CHECK-DAG: icmp eq i64 %{{.+}}, 0 +// CHECK-DAG: call void @llvm.assume(i1 + +// CHECK-DAG: and i64 %{{.+}}, 3 +// CHECK-DAG: icmp eq i64 %{{.+}}, 0 +// CHECK-DAG: call void @llvm.assume(i1 struct SS { SS(): a(0) {} SS(int v) : a(v) {} @@ -702,5 +708,8 @@ public: }; S8 s8(0); +// TERM_DEBUG-NOT: line: 0, +// TERM_DEBUG: distinct !DISubprogram(linkageName: "_GLOBAL__sub_I_simd_codegen.cpp", + #endif // HEADER diff --git a/test/OpenMP/simd_collapse_messages.cpp b/test/OpenMP/simd_collapse_messages.cpp index 5b8802468849b..b3100c097d66d 100644 --- a/test/OpenMP/simd_collapse_messages.cpp +++ b/test/OpenMP/simd_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/simd_lastprivate_messages.cpp b/test/OpenMP/simd_lastprivate_messages.cpp index 349d6a5b904d2..0af812889cf16 100644 --- a/test/OpenMP/simd_lastprivate_messages.cpp +++ b/test/OpenMP/simd_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/simd_linear_messages.cpp b/test/OpenMP/simd_linear_messages.cpp index 792f78aa3db6e..aad0d18d4281e 100644 --- a/test/OpenMP/simd_linear_messages.cpp +++ b/test/OpenMP/simd_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/simd_loop_messages.cpp b/test/OpenMP/simd_loop_messages.cpp index a47ccbc872196..b9d146c2150a6 100644 --- a/test/OpenMP/simd_loop_messages.cpp +++ b/test/OpenMP/simd_loop_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s + static int sii; // expected-note@+1 {{defined as threadprivate or thread local}} #pragma omp threadprivate(sii) @@ -234,9 +236,7 @@ int test_iteration_spaces() { for (ii = 0; ii < 10; ii++) c[ii] = a[ii]; - // expected-error@+3 {{unexpected OpenMP clause 'shared' in directive '#pragma omp simd'}} - // expected-note@+2 {{defined as shared}} - // expected-error@+2 {{loop iteration variable in the associated loop of 'omp simd' directive may not be shared, predetermined as linear}} + // expected-error@+1 {{unexpected OpenMP clause 'shared' in directive '#pragma omp simd'}} #pragma omp simd shared(ii) for (ii = 0; ii < 10; ii++) c[ii] = a[ii]; diff --git a/test/OpenMP/simd_metadata.c b/test/OpenMP/simd_metadata.c index 6ed660747dcb0..8fdc306930c5d 100644 --- a/test/OpenMP/simd_metadata.c +++ b/test/OpenMP/simd_metadata.c @@ -7,6 +7,15 @@ // RUN: %clang_cc1 -fopenmp -triple powerpc64-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK -check-prefix=PPC // RUN: %clang_cc1 -fopenmp -triple powerpc64-unknown-unknown -target-abi elfv1-qpx -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK -check-prefix=PPC-QPX +// RUN: %clang_cc1 -fopenmp-simd -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK -check-prefix=X86 +// RUN: %clang_cc1 -fopenmp-simd -triple x86_64-unknown-unknown -target-feature +avx -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK -check-prefix=X86-AVX +// RUN: %clang_cc1 -fopenmp-simd -triple x86_64-unknown-unknown -target-feature +avx512f -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK -check-prefix=X86-AVX512 +// RUN: %clang_cc1 -fopenmp-simd -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK -check-prefix=X86 +// RUN: %clang_cc1 -fopenmp-simd -triple i386-unknown-unknown -target-feature +avx -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK -check-prefix=X86-AVX +// RUN: %clang_cc1 -fopenmp-simd -triple i386-unknown-unknown -target-feature +avx512f -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK -check-prefix=X86-AVX512 +// RUN: %clang_cc1 -fopenmp-simd -triple powerpc64-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK -check-prefix=PPC +// RUN: %clang_cc1 -fopenmp-simd -triple powerpc64-unknown-unknown -target-abi elfv1-qpx -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK -check-prefix=PPC-QPX + void h1(float *c, float *a, double b[], int size) { // CHECK-LABEL: define void @h1 diff --git a/test/OpenMP/simd_misc_messages.c b/test/OpenMP/simd_misc_messages.c index 1e15482919473..75b8ce0596c7c 100644 --- a/test/OpenMP/simd_misc_messages.c +++ b/test/OpenMP/simd_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp simd'}} #pragma omp simd diff --git a/test/OpenMP/simd_private_messages.cpp b/test/OpenMP/simd_private_messages.cpp index 1850101891f7e..bfa26d164529a 100644 --- a/test/OpenMP/simd_private_messages.cpp +++ b/test/OpenMP/simd_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/simd_reduction_messages.cpp b/test/OpenMP/simd_reduction_messages.cpp index bdf429d1967c9..135a3ca71f6e1 100644 --- a/test/OpenMP/simd_reduction_messages.cpp +++ b/test/OpenMP/simd_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/simd_safelen_messages.cpp b/test/OpenMP/simd_safelen_messages.cpp index 56cb868337e0f..b75a923e7f9f7 100644 --- a/test/OpenMP/simd_safelen_messages.cpp +++ b/test/OpenMP/simd_safelen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/simd_simdlen_messages.cpp b/test/OpenMP/simd_simdlen_messages.cpp index 426d1878ba4ca..f34d628141973 100644 --- a/test/OpenMP/simd_simdlen_messages.cpp +++ b/test/OpenMP/simd_simdlen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/single_ast_print.cpp b/test/OpenMP/single_ast_print.cpp index e819ab9e14a57..a8eaeb0bff341 100644 --- a/test/OpenMP/single_ast_print.cpp +++ b/test/OpenMP/single_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -16,7 +20,7 @@ struct SS { #pragma omp parallel firstprivate(a, b, c) #pragma omp single copyprivate(a, this->b, (this)->c) // CHECK: #pragma omp parallel firstprivate(this->a,this->b,this->c) -// CHECK-NEXT: #pragma omp single copyprivate(this->a,this->b,this->c) +// CHECK-NEXT: #pragma omp single copyprivate(this->a,this->b,this->c){{$}} ++this->a, --b, (this)->c /= 1; } }; diff --git a/test/OpenMP/single_codegen.cpp b/test/OpenMP/single_codegen.cpp index 892e160324171..b6ecac72d4f5b 100644 --- a/test/OpenMP/single_codegen.cpp +++ b/test/OpenMP/single_codegen.cpp @@ -3,6 +3,13 @@ // RUN: %clang_cc1 -fopenmp -fnoopenmp-use-tls -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -std=c++11 -fopenmp -fnoopenmp-use-tls -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG // RUN: %clang_cc1 -verify -fopenmp -fnoopenmp-use-tls -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fnoopenmp-use-tls -x c++ -std=c++11 -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fnoopenmp-use-tls -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fnoopenmp-use-tls -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -std=c++11 -fopenmp-simd -fnoopenmp-use-tls -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fnoopenmp-use-tls -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef ARRAY #ifndef HEADER @@ -173,7 +180,7 @@ int main() { // CHECK: [[DST_A2_ADDR:%.+]] = load i8*, i8** [[DST_A2_ADDR_REF]], // CHECK: [[SRC_A2_ADDR_REF:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[SRC_ADDR]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 // CHECK: [[SRC_A2_ADDR:%.+]] = load i8*, i8** [[SRC_A2_ADDR_REF]], -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[DST_A2_ADDR]], i8* [[SRC_A2_ADDR]], i{{[0-9]+}} 2, i{{[0-9]+}} 1, i1 false) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align 1 [[DST_A2_ADDR]], i8* align 1 [[SRC_A2_ADDR]], i{{[0-9]+}} 2, i1 false) // CHECK: [[DST_TC2_ADDR_REF:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DST_ADDR]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 // CHECK: [[DST_TC2_ADDR_VOID_PTR:%.+]] = load i8*, i8** [[DST_TC2_ADDR_REF]], // CHECK: [[DST_TC2_ADDR:%.+]] = bitcast i8* [[DST_TC2_ADDR_VOID_PTR]] to [[TEST_CLASS_TY]]* @@ -214,7 +221,7 @@ struct St { }; void array_func(int n, int a[n], St s[2]) { -// ARRAY: call void @__kmpc_copyprivate(%ident_t* @{{.+}}, i32 %{{.+}}, i64 16, i8* %{{.+}}, void (i8*, i8*)* [[CPY:@.+]], i32 %{{.+}}) +// ARRAY: call void @__kmpc_copyprivate(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i64 16, i8* %{{.+}}, void (i8*, i8*)* [[CPY:@.+]], i32 %{{.+}}) #pragma omp single copyprivate(a, s) ; } diff --git a/test/OpenMP/single_copyprivate_messages.cpp b/test/OpenMP/single_copyprivate_messages.cpp index 4714753f494e0..bb18568d0f289 100644 --- a/test/OpenMP/single_copyprivate_messages.cpp +++ b/test/OpenMP/single_copyprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/single_firstprivate_codegen.cpp b/test/OpenMP/single_firstprivate_codegen.cpp index 4a91bc2d5bc50..23b543511b0f1 100644 --- a/test/OpenMP/single_firstprivate_codegen.cpp +++ b/test/OpenMP/single_firstprivate_codegen.cpp @@ -3,6 +3,13 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -185,7 +192,7 @@ int main() { // firstprivate vec(vec) // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* bitcast ([2 x i{{[0-9]+}}]* [[VEC]] to i8*), +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST]], i8* align {{[0-9]+}} bitcast ([2 x i{{[0-9]+}}]* [[VEC]] to i8*), // firstprivate s_arr(s_arr) // CHECK: [[S_ARR_PRIV_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 @@ -249,7 +256,7 @@ int main() { // firstprivate vec(vec) // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_REF]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* [[VEC_SRC]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST]], i8* align {{[0-9]+}} [[VEC_SRC]], // firstprivate s_arr(s_arr) // CHECK: [[S_ARR_PRIV_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 diff --git a/test/OpenMP/single_firstprivate_messages.cpp b/test/OpenMP/single_firstprivate_messages.cpp index 32de9fdc88067..4c714e392f971 100644 --- a/test/OpenMP/single_firstprivate_messages.cpp +++ b/test/OpenMP/single_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/single_misc_messages.c b/test/OpenMP/single_misc_messages.c index 2c922ddfd4f04..b8264198079ee 100644 --- a/test/OpenMP/single_misc_messages.c +++ b/test/OpenMP/single_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + void foo(); // expected-error@+1 {{unexpected OpenMP directive '#pragma omp single'}} diff --git a/test/OpenMP/single_private_codegen.cpp b/test/OpenMP/single_private_codegen.cpp index 6d47cb50dc590..4b4d103d7dd49 100644 --- a/test/OpenMP/single_private_codegen.cpp +++ b/test/OpenMP/single_private_codegen.cpp @@ -3,6 +3,13 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -39,7 +46,7 @@ T tmain() { int main() { static int sivar; #ifdef LAMBDA - // LAMBDA: [[G:@.+]] = global double + // LAMBDA: [[G:@.+]] = {{(dso_local )?}}global double // LAMBDA-LABEL: @main // LAMBDA: call{{.*}} void [[OUTER_LAMBDA:@.+]]( [&]() { @@ -79,7 +86,7 @@ int main() { }(); return 0; #elif defined(BLOCKS) - // BLOCKS: [[G:@.+]] = global double + // BLOCKS: [[G:@.+]] = {{(dso_local )?}}global double // BLOCKS-LABEL: @main // BLOCKS: call {{.*}}void {{%.+}}(i8 ^{ diff --git a/test/OpenMP/single_private_messages.cpp b/test/OpenMP/single_private_messages.cpp index 0ed0e6cfb0999..0a3ec2e6677f9 100644 --- a/test/OpenMP/single_private_messages.cpp +++ b/test/OpenMP/single_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_ast_print.cpp b/test/OpenMP/target_ast_print.cpp index d004f738fc0d3..827d6a4e3deb4 100644 --- a/test/OpenMP/target_ast_print.cpp +++ b/test/OpenMP/target_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -42,7 +46,7 @@ T tmain(T argc, T *argv) { // CHECK: template <typename T, int C> T tmain(T argc, T *argv) { // CHECK-NEXT: T i, j, a[20] -// CHECK-NEXT: #pragma omp target +// CHECK-NEXT: #pragma omp target{{$}} // CHECK-NEXT: foo(); // CHECK-NEXT: #pragma omp target if(target: argc > 0) // CHECK-NEXT: foo() diff --git a/test/OpenMP/target_codegen.cpp b/test/OpenMP/target_codegen.cpp index 20bc96b84cbfb..15eb469fa2804 100644 --- a/test/OpenMP/target_codegen.cpp +++ b/test/OpenMP/target_codegen.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 @@ -16,6 +24,16 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -31,38 +49,38 @@ // CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat // We have 8 target regions, but only 7 that actually will generate offloading -// code, only 6 will have mapped arguments, and only 4 have all-constant map -// sizes. +// code and have mapped arguments, and only 5 have all-constant map sizes. // CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 0, i[[SZ]] 4] -// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 32, i64 288] +// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 800] // CHECK-DAG: [[SIZET2:@.+]] = private unnamed_addr constant [1 x i{{32|64}}] [i[[SZ]] 2] -// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [1 x i64] [i64 288] +// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [1 x i64] [i64 800] // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2] -// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 288, i64 288] -// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [9 x i64] [i64 288, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547] +// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 800] +// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [9 x i64] [i64 800, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547] // CHECK-DAG: [[SIZET5:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 40] -// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 547] +// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 547] // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [4 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 1, i[[SZ]] 40] -// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [4 x i64] [i64 288, i64 288, i64 288, i64 547] -// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [5 x i64] [i64 547, i64 288, i64 288, i64 288, i64 547] -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 - -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]] -// TCHECK-NOT: @{{.+}} = constant [[ENTTY]] +// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [4 x i64] [i64 800, i64 800, i64 800, i64 547] +// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [6 x i64] [i64 32, i64 281474976711171, i64 800, i64 288, i64 288, i64 547] +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] // Check if offloading descriptor is created. // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] @@ -73,7 +91,7 @@ // CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) // Check target registration is registered as a Ctor. -// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] template<typename tx, typename ty> @@ -97,7 +115,9 @@ int foo(int n) { static long *plocal; // CHECK: [[ADD:%.+]] = add nsw i32 - // CHECK: [[DEVICE:%.+]] = sext i32 [[ADD]] to i64 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 // CHECK: [[RET:%.+]] = call i32 @__tgt_target(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 0, i8** null, i8** null, i[[SZ]]* null, i64* null) // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] @@ -110,7 +130,9 @@ int foo(int n) { } // CHECK-DAG: [[ADD:%.+]] = add nsw i32 - // CHECK-DAG: [[DEVICE:%.+]] = sext i32 [[ADD]] to i64 + // CHECK-DAG: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + // CHECK-DAG: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK-DAG: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_nowait(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 2, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* getelementptr inbounds ([2 x i[[SZ]]], [2 x i[[SZ]]]* [[SIZET]], i32 0, i32 0), i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT]], i32 0, i32 0) // CHECK-DAG: [[BPR]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 // CHECK-DAG: [[PR]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%[^,]+]], i32 0, i32 0 @@ -216,13 +238,13 @@ int foo(int n) { // CHECK-32: store i32 [[A_VAL]], i32* [[A_CADDR:%.+]], // CHECK-32: [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]], + // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 + // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[IFELSE:[^,]+]] + // CHECK: [[TRY]] // CHECK: [[BNSIZE:%.+]] = mul nuw i[[SZ]] [[VLA0:%.+]], 4 // CHECK: [[CNELEMSIZE2:%.+]] = mul nuw i[[SZ]] 5, [[VLA1:%.+]] // CHECK: [[CNSIZE:%.+]] = mul nuw i[[SZ]] [[CNELEMSIZE2]], 8 - // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 - // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[IFELSE:[^,]+]] - // CHECK: [[TRY]] // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target(i64 -1, i8* @{{[^,]+}}, i32 9, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([9 x i64], [9 x i64]* [[MAPT4]], i32 0, i32 0)) // CHECK-DAG: [[BPR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 // CHECK-DAG: [[PR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P:%[^,]+]], i32 0, i32 0 @@ -489,67 +511,77 @@ int bar(int n){ // CHECK-64: store i32 %{{.+}}, i32* [[B_ADDR]], // CHECK-64: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_CADDR]], +// CHECK-32: store i32 %{{.+}}, i32* %__vla_expr // CHECK-32: store i32 %{{.+}}, i32* [[B_ADDR:%.+]], // CHECK-32: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_ADDR]], +// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 +// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[IFELSE:[^,]+]] +// CHECK: [[TRY]] // We capture 2 VLA sizes in this target region // CHECK: [[CELEMSIZE2:%.+]] = mul nuw i[[SZ]] 2, [[VLA0:%.+]] // CHECK: [[CSIZE:%.+]] = mul nuw i[[SZ]] [[CELEMSIZE2]], 2 -// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 -// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[IFELSE:[^,]+]] -// CHECK: [[TRY]] -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target(i64 -1, i8* @{{[^,]+}}, i32 5, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([5 x i64], [5 x i64]* [[MAPT7]], i32 0, i32 0)) -// CHECK-DAG: [[BPR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP:%.+]], i32 0, i32 0 -// CHECK-DAG: [[PR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SR]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX0:0]] -// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 [[IDX0]] -// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 [[IDX0]] -// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX1:1]] -// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 [[IDX1]] -// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 [[IDX1]] -// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX2:2]] -// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 [[IDX2]] -// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 [[IDX2]] -// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX3:3]] -// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 [[IDX3]] -// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 [[IDX3]] -// CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX4:4]] -// CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 0, i32 [[IDX4]] -// CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 0, i32 [[IDX4]] +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target(i64 -1, i8* @{{[^,]+}}, i32 6, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([6 x i64], [6 x i64]* [[MAPT7]], i32 0, i32 0)) +// CHECK-DAG: [[BPR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP:%.+]], i32 0, i32 0 +// CHECK-DAG: [[PR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SR]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX0:0]] +// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 0, i32 [[IDX0]] +// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 0, i32 [[IDX0]] +// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX1:1]] +// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 0, i32 [[IDX1]] +// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 0, i32 [[IDX1]] +// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX2:2]] +// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 0, i32 [[IDX2]] +// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 0, i32 [[IDX2]] +// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX3:3]] +// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 0, i32 [[IDX3]] +// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 0, i32 [[IDX3]] +// CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX4:4]] +// CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 0, i32 [[IDX4]] +// CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 0, i32 [[IDX4]] +// CHECK-DAG: [[SADDR5:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX5:5]] +// CHECK-DAG: [[BPADDR5:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 0, i32 [[IDX5]] +// CHECK-DAG: [[PADDR5:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 0, i32 [[IDX5]] // The names below are not necessarily consistent with the names used for the // addresses above as some are repeated. +// CHECK-DAG: [[CBPADDR5:%.+]] = bitcast i8** [[BPADDR5]] to i16** +// CHECK-DAG: [[CPADDR5:%.+]] = bitcast i8** [[PADDR5]] to i16** +// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR5]] +// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR5]] +// CHECK-DAG: store i[[SZ]] [[CSIZE]], i[[SZ]]* [[SADDR5]] + +// CHECK-DAG: [[CBPADDR4:%.+]] = bitcast i8** [[BPADDR4]] to i[[SZ]]* +// CHECK-DAG: [[CPADDR4:%.+]] = bitcast i8** [[PADDR4]] to i[[SZ]]* +// CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CBPADDR4]] +// CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CPADDR4]] +// CHECK-DAG: store i[[SZ]] {{4|8}}, i[[SZ]]* [[SADDR4]] + // CHECK-DAG: [[CBPADDR3:%.+]] = bitcast i8** [[BPADDR3]] to i[[SZ]]* // CHECK-DAG: [[CPADDR3:%.+]] = bitcast i8** [[PADDR3]] to i[[SZ]]* -// CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CBPADDR3]] -// CHECK-DAG: store i[[SZ]] [[VLA0]], i[[SZ]]* [[CPADDR3]] +// CHECK-DAG: store i[[SZ]] 2, i[[SZ]]* [[CBPADDR3]] +// CHECK-DAG: store i[[SZ]] 2, i[[SZ]]* [[CPADDR3]] // CHECK-DAG: store i[[SZ]] {{4|8}}, i[[SZ]]* [[SADDR3]] // CHECK-DAG: [[CBPADDR2:%.+]] = bitcast i8** [[BPADDR2]] to i[[SZ]]* // CHECK-DAG: [[CPADDR2:%.+]] = bitcast i8** [[PADDR2]] to i[[SZ]]* -// CHECK-DAG: store i[[SZ]] 2, i[[SZ]]* [[CBPADDR2]] -// CHECK-DAG: store i[[SZ]] 2, i[[SZ]]* [[CPADDR2]] -// CHECK-DAG: store i[[SZ]] {{4|8}}, i[[SZ]]* [[SADDR2]] - -// CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* -// CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* -// CHECK-DAG: store i[[SZ]] [[B_CVAL]], i[[SZ]]* [[CBPADDR1]] -// CHECK-DAG: store i[[SZ]] [[B_CVAL]], i[[SZ]]* [[CPADDR1]] -// CHECK-DAG: store i[[SZ]] 4, i[[SZ]]* [[SADDR1]] +// CHECK-DAG: store i[[SZ]] [[B_CVAL]], i[[SZ]]* [[CBPADDR2]] +// CHECK-DAG: store i[[SZ]] [[B_CVAL]], i[[SZ]]* [[CPADDR2]] +// CHECK-DAG: store i[[SZ]] 4, i[[SZ]]* [[SADDR2]] // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to [[S1]]** // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to double** -// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR0]] -// CHECK-DAG: store double* %{{.+}}, double** [[CPADDR0]] -// CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* [[SADDR0]] +// CHECK-DAG: store [[S1]]* [[THIS:%.+]], [[S1]]** [[CBPADDR0]] +// CHECK-DAG: store double* [[A:%.+]], double** [[CPADDR0]] +// CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[SADDR0]] -// CHECK-DAG: [[CBPADDR4:%.+]] = bitcast i8** [[BPADDR4]] to i16** -// CHECK-DAG: [[CPADDR4:%.+]] = bitcast i8** [[PADDR4]] to i16** -// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR4]] -// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR4]] -// CHECK-DAG: store i[[SZ]] [[CSIZE]], i[[SZ]]* [[SADDR4]] +// CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to [[S1]]** +// CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to double** +// CHECK-DAG: store [[S1]]* [[THIS]], [[S1]]** [[CBPADDR1]] +// CHECK-DAG: store double* [[A]], double** [[CPADDR1]] +// CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* [[SADDR1]] // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] @@ -727,4 +759,10 @@ int bar(int n){ // CHECK-32-DAG: load i32, i32* [[LOCAL_A]] // CHECK-DAG: load i16, i16* [[REF_AA]] // CHECK-DAG: getelementptr inbounds [10 x i32], [10 x i32]* [[REF_B]], i[[SZ]] 0, i[[SZ]] 2 + +void bar () { +#define pragma_target _Pragma("omp target") +pragma_target +{} +} #endif diff --git a/test/OpenMP/target_codegen_global_capture.cpp b/test/OpenMP/target_codegen_global_capture.cpp index eb34082b2940b..27f8bcea020b9 100644 --- a/test/OpenMP/target_codegen_global_capture.cpp +++ b/test/OpenMP/target_codegen_global_capture.cpp @@ -4,6 +4,14 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -142,7 +150,7 @@ int foo(short a, short b, short c, short d){ // CHECK: define internal void [[OFFLOADF]]({{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}) // The parallel region only uses 3 captures. - // CHECK: call {{.*}}@__kmpc_fork_call(%ident_t* {{.+}}, i32 {{.+}}, void (i32*, i32*, ...)* bitcast ({{.*}}[[PARF:@.+]] to {{.*}}), {{.+}}* %{{.+}}, {{.+}}* %{{.+}}, {{.+}}* %{{.+}}) + // CHECK: call {{.*}}@__kmpc_fork_call(%struct.ident_t* {{.+}}, i32 {{.+}}, void (i32*, i32*, ...)* bitcast ({{.*}}[[PARF:@.+]] to {{.*}}), {{.+}}* %{{.+}}, {{.+}}* %{{.+}}, {{.+}}* %{{.+}}) // CHECK: call void @.omp_outlined.(i32* %{{.+}}, i32* %{{.+}}, {{.+}}* %{{.+}}, {{.+}}* %{{.+}}, {{.+}}* %{{.+}}) // Capture d, Gd, Sd, @@ -172,7 +180,7 @@ int bar(short a, short b, short c, short d){ static float Sc = 11.0; static float Sd = 12.0; - // CHECK: call void {{.*}}@__kmpc_fork_call(%ident_t* {{.+}}, i32 {{.+}}, void (i32*, i32*, ...)* bitcast ({{.*}}[[PARF:@.+]] to {{.*}}), i16* %{{.+}}, i16* %{{.+}}, i16* %{{.+}}, i16* %{{.+}}) + // CHECK: call void {{.*}}@__kmpc_fork_call(%struct.ident_t* {{.+}}, i32 {{.+}}, void (i32*, i32*, ...)* bitcast ({{.*}}[[PARF:@.+]] to {{.*}}), i16* %{{.+}}, i16* %{{.+}}, i16* %{{.+}}, i16* %{{.+}}) // CHECK: define internal void [[PARF]](i32* noalias %{{.*}}, i32* noalias %{{.*}}, i16* dereferenceable(2) [[A:%.+]], i16* dereferenceable(2) [[B:%.+]], i16* dereferenceable(2) [[C:%.+]], i16* dereferenceable(2) [[D:%.+]]) // Capture a, b, c, d // CHECK: [[ALLOCLA:%.+]] = alloca i16 @@ -280,7 +288,7 @@ int bar(short a, short b, short c, short d){ Sb += 1.0; // CHECK: define internal void [[OFFLOADF]]({{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}) - // CHECK: call void {{.*}}@__kmpc_fork_call(%ident_t* {{.+}}, i32 {{.+}}, void (i32*, i32*, ...)* bitcast ({{.*}}[[PARF:@.+]] to {{.*}}) + // CHECK: call void {{.*}}@__kmpc_fork_call(%struct.ident_t* {{.+}}, i32 {{.+}}, void (i32*, i32*, ...)* bitcast ({{.*}}[[PARF:@.+]] to {{.*}}) // CHECK: define internal void [[PARF]](i32* noalias %{{.*}}, i32* noalias %{{.*}}, {{.+}}* dereferenceable({{.+}}) %{{.+}}, {{.+}}* dereferenceable({{.+}}) %{{.+}}, {{.+}}* dereferenceable({{.+}}) %{{.+}}) // Capture d, Gd, Sd @@ -317,7 +325,7 @@ int tbar(T a, T b, T c, T d){ static float Sc = 19.0; static float Sd = 20.0; - // CHECK: call void {{.*}}@__kmpc_fork_call(%ident_t* {{.+}}, i32 {{.+}}, void (i32*, i32*, ...)* bitcast ({{.*}}[[PARF:@.+]] to {{.*}}), i16* %{{.+}}, i16* %{{.+}}, i16* %{{.+}}, i16* %{{.+}}) + // CHECK: call void {{.*}}@__kmpc_fork_call(%struct.ident_t* {{.+}}, i32 {{.+}}, void (i32*, i32*, ...)* bitcast ({{.*}}[[PARF:@.+]] to {{.*}}), i16* %{{.+}}, i16* %{{.+}}, i16* %{{.+}}, i16* %{{.+}}) // CHECK: define internal void [[PARF]](i32* noalias %{{.*}}, i32* noalias %{{.*}}, i16* dereferenceable(2) [[A:%.+]], i16* dereferenceable(2) [[B:%.+]], i16* dereferenceable(2) [[C:%.+]], i16* dereferenceable(2) [[D:%.+]]) // Capture a, b, c, d // CHECK: [[ALLOCLA:%.+]] = alloca i16 @@ -425,7 +433,7 @@ int tbar(T a, T b, T c, T d){ Sb += 1.0; // CHECK: define internal void [[OFFLOADF]]({{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}, {{.+}} {{.*}}%{{.+}}) - // CHECK: call void {{.*}}@__kmpc_fork_call(%ident_t* {{.+}}, i32 {{.+}}, void (i32*, i32*, ...)* bitcast ({{.*}}[[PARF:@.+]] to {{.*}}) + // CHECK: call void {{.*}}@__kmpc_fork_call(%struct.ident_t* {{.+}}, i32 {{.+}}, void (i32*, i32*, ...)* bitcast ({{.*}}[[PARF:@.+]] to {{.*}}) // CHECK: define internal void [[PARF]](i32* noalias %{{.*}}, i32* noalias %{{.*}}, {{.+}}* dereferenceable({{.+}}) %{{.+}}, {{.+}}* dereferenceable({{.+}}) %{{.+}}, {{.+}}* dereferenceable({{.+}}) %{{.+}}) // Capture d, Gd, Sd diff --git a/test/OpenMP/target_codegen_registration.cpp b/test/OpenMP/target_codegen_registration.cpp index 5684b2573b6e3..b98b33c9a4ddf 100644 --- a/test/OpenMP/target_codegen_registration.cpp +++ b/test/OpenMP/target_codegen_registration.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,9 +24,22 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK -// Check that no target code is emmitted if no omptests flag was provided. +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// Check that no target code is emitted if no omptests flag was provided. // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK-NTARGET +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -60,96 +81,96 @@ // We have 7 target regions -// CHECK-DAG: {{@.+}} = private constant i8 0 -// TCHECK-NOT: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// TCHECK-NOT: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] -// CHECK-NTARGET-NOT: private constant i8 0 +// CHECK-NTARGET-NOT: weak constant i8 0 // CHECK-NTARGET-NOT: private unnamed_addr constant [1 x i // CHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// CHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: @.omp_offloading.entry.[[NAME1]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// CHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: @.omp_offloading.entry.[[NAME2]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// CHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: @.omp_offloading.entry.[[NAME3]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// CHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: @.omp_offloading.entry.[[NAME4]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// CHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: @.omp_offloading.entry.[[NAME5]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// CHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: @.omp_offloading.entry.[[NAME6]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// CHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: @.omp_offloading.entry.[[NAME7]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// CHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: @.omp_offloading.entry.[[NAME8]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// CHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: @.omp_offloading.entry.[[NAME9]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// CHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: @.omp_offloading.entry.[[NAME10]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// CHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: @.omp_offloading.entry.[[NAME11]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// CHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: @.omp_offloading.entry.[[NAME12]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// TCHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: @.omp_offloading.entry.[[NAME1]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// TCHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: @.omp_offloading.entry.[[NAME2]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// TCHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: @.omp_offloading.entry.[[NAME3]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// TCHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: @.omp_offloading.entry.[[NAME4]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// TCHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: @.omp_offloading.entry.[[NAME5]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// TCHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: @.omp_offloading.entry.[[NAME6]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// TCHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: @.omp_offloading.entry.[[NAME7]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// TCHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: @.omp_offloading.entry.[[NAME8]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// TCHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: @.omp_offloading.entry.[[NAME9]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// TCHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: @.omp_offloading.entry.[[NAME10]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// TCHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: @.omp_offloading.entry.[[NAME11]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// TCHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: @.omp_offloading.entry.[[NAME12]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] // CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] @@ -163,7 +184,7 @@ // CHECK-SAME: { i32, void ()*, i8* } { i32 500, void ()* [[P500:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 501, void ()* [[P501:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 65535, void ()* [[PMAX:@[^,]+]], i8* null }, -// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] // CHECK-NTARGET: @llvm.global_ctors = appending global [3 x { i32, void ()*, i8* }] [ @@ -319,18 +340,18 @@ struct ST { //CHECK-DAG: define internal void @[[NAME12]]( //CHECK-DAG: call void @[[NAME12]]( -//TCHECK-DAG: define void @[[NAME1]]( -//TCHECK-DAG: define void @[[NAME2]]( -//TCHECK-DAG: define void @[[NAME3]]( -//TCHECK-DAG: define void @[[NAME4]]( -//TCHECK-DAG: define void @[[NAME5]]( -//TCHECK-DAG: define void @[[NAME6]]( -//TCHECK-DAG: define void @[[NAME7]]( -//TCHECK-DAG: define void @[[NAME8]]( -//TCHECK-DAG: define void @[[NAME9]]( -//TCHECK-DAG: define void @[[NAME10]]( -//TCHECK-DAG: define void @[[NAME11]]( -//TCHECK-DAG: define void @[[NAME12]]( +//TCHECK-DAG: define weak void @[[NAME1]]( +//TCHECK-DAG: define weak void @[[NAME2]]( +//TCHECK-DAG: define weak void @[[NAME3]]( +//TCHECK-DAG: define weak void @[[NAME4]]( +//TCHECK-DAG: define weak void @[[NAME5]]( +//TCHECK-DAG: define weak void @[[NAME6]]( +//TCHECK-DAG: define weak void @[[NAME7]]( +//TCHECK-DAG: define weak void @[[NAME8]]( +//TCHECK-DAG: define weak void @[[NAME9]]( +//TCHECK-DAG: define weak void @[[NAME10]]( +//TCHECK-DAG: define weak void @[[NAME11]]( +//TCHECK-DAG: define weak void @[[NAME12]]( // CHECK-NTARGET-NOT: __tgt_target // CHECK-NTARGET-NOT: __tgt_register_lib @@ -372,7 +393,7 @@ struct ST { //CHECK: ret void //CHECK: declare i32 @__tgt_unregister_lib([[DSCTY]]*) -//CHECK: define linkonce hidden void @[[REGFN]](i8*) +//CHECK: define linkonce hidden void @[[REGFN]]() //CHECK-SAME: comdat { //CHECK: call i32 @__tgt_register_lib([[DSCTY]]* [[DESC]]) //CHECK: call i32 @__cxa_atexit(void (i8*)* @[[UNREGFN]], i8* bitcast ([[DSCTY]]* [[DESC]] to i8*), @@ -411,31 +432,31 @@ int bar(int a){ // Check metadata is properly generated: // CHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 245, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 261, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 267, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 278, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 406, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 290, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 290, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 278, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 220, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 266, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 282, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 288, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 299, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 427, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 311, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 311, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 299, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 241, i32 {{[0-9]+}}} // TCHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 245, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 261, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 267, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 278, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 406, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 290, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 290, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 278, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 220, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 266, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 282, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 288, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 299, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 427, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 311, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 311, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 299, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 241, i32 {{[0-9]+}}} #endif diff --git a/test/OpenMP/target_codegen_registration_naming.cpp b/test/OpenMP/target_codegen_registration_naming.cpp index ce133eee03e80..0d584b87d251c 100644 --- a/test/OpenMP/target_codegen_registration_naming.cpp +++ b/test/OpenMP/target_codegen_registration_naming.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,6 +24,16 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/target_data_ast_print.cpp b/test/OpenMP/target_data_ast_print.cpp index bb55ea30d5325..17dace8bfac80 100644 --- a/test/OpenMP/target_data_ast_print.cpp +++ b/test/OpenMP/target_data_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -48,7 +52,7 @@ T tmain(T argc, T *argv) { // CHECK: template <typename T, int C> T tmain(T argc, T *argv) { // CHECK-NEXT: T i, j, b, c, d, e, x[20]; -// CHECK-NEXT: #pragma omp target data map(to: c) +// CHECK-NEXT: #pragma omp target data map(to: c){{$}} // CHECK-NEXT: i = argc; // CHECK-NEXT: #pragma omp target data map(to: c) if(target data: j > 0) // CHECK-NEXT: foo(); diff --git a/test/OpenMP/target_data_codegen.cpp b/test/OpenMP/target_data_codegen.cpp index 2c9526de73e11..80f7096a0b52e 100644 --- a/test/OpenMP/target_data_codegen.cpp +++ b/test/OpenMP/target_data_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 // CK1: [[ST:%.+]] = type { i32, double* } @@ -29,8 +37,9 @@ double gc[100]; // CK1: [[MTYPE03:@.+]] = {{.+}}constant [1 x i64] [i64 37] -// CK1: [[SIZE04:@.+]] = {{.+}}constant [2 x i[[sz]]] [i[[sz]] {{8|4}}, i[[sz]] 24] -// CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i64] [i64 33, i64 17] +// CK1-64: [[SIZE04:@.+]] = {{.+}}constant [2 x i64] [i64 sdiv exact (i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), i64 ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)), i64 24] +// CK1-32: [[SIZE04:@.+]] = {{.+}}constant [2 x i32] [i32 trunc (i64 sdiv exact (i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), i64 ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)) to i32), i32 24] +// CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i64] [i64 32, i64 281474976710673] // CK1-LABEL: _Z3fooi void foo(int arg) { @@ -57,7 +66,7 @@ void foo(int arg) { // CK1-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64 // CK1-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}}, // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP]] - // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P]] +// CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P]] #pragma omp target data if(1+3-5) device(arg) map(from: gc) {++arg;} @@ -137,7 +146,6 @@ void foo(int arg) { // CK1-DAG: store [[ST]]* @gb, [[ST]]** [[CBP0]] // CK1-DAG: store double** getelementptr inbounds ([[ST]], [[ST]]* @gb, i32 0, i32 1), double*** [[CP0]] - // CK1-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 // CK1-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 // CK1-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double*** @@ -163,6 +171,14 @@ void foo(int arg) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 // CK2: [[ST:%.+]] = type { i32, double* } @@ -179,8 +195,7 @@ struct ST { } }; -// CK2: [[SIZE00:@.+]] = {{.+}}constant [2 x i[[sz:64|32]]] [i{{64|32}} {{8|4}}, i{{64|32}} 24] -// CK2: [[MTYPE00:@.+]] = {{.+}}constant [2 x i64] [i64 37, i64 21] +// CK2: [[MTYPE00:@.+]] = {{.+}}constant [2 x i64] [i64 32, i64 281474976710677] // CK2-LABEL: _Z3bari int bar(int arg){ @@ -191,27 +206,31 @@ int bar(int arg){ // Region 00 // CK2: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]] // CK2: [[IFTHEN]] -// CK2-DAG: call void @__tgt_target_data_begin(i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE00]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}) +// CK2-DAG: call void @__tgt_target_data_begin(i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}) // CK2-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64 // CK2-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}}, -// CK2-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] -// CK2-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK2-DAG: [[GEPBP]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%[^,]+]] +// CK2-DAG: [[GEPP]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%[^,]+]] +// CK2-DAG: [[GEPS]] = getelementptr inbounds [2 x i[[sz]]], [2 x i[[sz]]]* [[S:%[^,]+]] // CK2-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK2-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK2-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK2-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]** // CK2-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double*** // CK2-DAG: store [[ST]]* [[VAR0:%.+]], [[ST]]** [[CBP0]] // CK2-DAG: store double** [[SEC0:%.+]], double*** [[CP0]] +// CK2-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK2-DAG: [[SEC0]] = getelementptr inbounds {{.*}}[[ST]]* [[VAR0]], i32 0, i32 1 - // CK2-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 // CK2-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 +// CK2-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 // CK2-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double*** // CK2-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double** // CK2-DAG: store double** [[SEC0]], double*** [[CBP1]] // CK2-DAG: store double* [[SEC1:%.+]], double** [[CP1]] +// CK2-DAG: store i[[sz]] 24, i[[sz]]* [[S1]] // CK2-DAG: [[SEC1]] = getelementptr inbounds {{.*}}double* [[SEC11:%[^,]+]], i{{.+}} 1 // CK2-DAG: [[SEC11]] = load double*, double** [[SEC111:%[^,]+]], // CK2-DAG: [[SEC111]] = getelementptr inbounds {{.*}}[[ST]]* [[VAR0]], i32 0, i32 1 @@ -225,11 +244,12 @@ int bar(int arg){ // CK2: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]] // CK2: [[IFTHEN]] -// CK2-DAG: call void @__tgt_target_data_end(i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE00]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}) +// CK2-DAG: call void @__tgt_target_data_end(i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}) // CK2-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64 // CK2-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}}, // CK2-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP]] // CK2-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P]] +// CK2-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S]] // CK2: br label %[[IFEND:[^,]+]] // CK2: [[IFELSE]] // CK2: br label %[[IFEND]] @@ -242,6 +262,14 @@ int bar(int arg){ // RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 + +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} #ifdef CK3 // CK3-LABEL: no_target_devices diff --git a/test/OpenMP/target_data_device_messages.cpp b/test/OpenMP/target_data_device_messages.cpp index 9ed7a54b4399a..51c809a8653e1 100644 --- a/test/OpenMP/target_data_device_messages.cpp +++ b/test/OpenMP/target_data_device_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_data_if_messages.cpp b/test/OpenMP/target_data_if_messages.cpp index 229d7f564738a..ae68d252a8e1f 100644 --- a/test/OpenMP/target_data_if_messages.cpp +++ b/test/OpenMP/target_data_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_data_messages.c b/test/OpenMP/target_data_messages.c index fd41e7df2a4cf..017a339cc0175 100644 --- a/test/OpenMP/target_data_messages.c +++ b/test/OpenMP/target_data_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } int main(int argc, char **argv) { diff --git a/test/OpenMP/target_data_use_device_ptr_ast_print.cpp b/test/OpenMP/target_data_use_device_ptr_ast_print.cpp index ce1ec1d2799af..ba429f50febed 100644 --- a/test/OpenMP/target_data_use_device_ptr_ast_print.cpp +++ b/test/OpenMP/target_data_use_device_ptr_ast_print.cpp @@ -1,6 +1,9 @@ // RxUN: %clang_cc1 -verify -fopenmp -std=c++11 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -23,7 +26,7 @@ struct SA { }; // CHECK: struct SA // CHECK: void func( -// CHECK: #pragma omp target data map(tofrom: this->i) use_device_ptr(this->k) +// CHECK: #pragma omp target data map(tofrom: this->i) use_device_ptr(this->k){{$}} // CHECK: #pragma omp target data map(tofrom: this->i) use_device_ptr(this->z) struct SB { unsigned A; diff --git a/test/OpenMP/target_data_use_device_ptr_codegen.cpp b/test/OpenMP/target_data_use_device_ptr_codegen.cpp index 36c42e32ea3f3..a3d8043b6b4e9 100644 --- a/test/OpenMP/target_data_use_device_ptr_codegen.cpp +++ b/test/OpenMP/target_data_use_device_ptr_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 double *g; @@ -327,13 +335,21 @@ void bar(float *&a, int *&b) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 // CK2: [[ST:%.+]] = type { double*, double** } -// CK2: [[MTYPE00:@.+]] = {{.*}}constant [2 x i64] [i64 35, i64 83] -// CK2: [[MTYPE01:@.+]] = {{.*}}constant [3 x i64] [i64 32, i64 19, i64 83] -// CK2: [[MTYPE02:@.+]] = {{.*}}constant [2 x i64] [i64 96, i64 35] -// CK2: [[MTYPE03:@.+]] = {{.*}}constant [4 x i64] [i64 96, i64 32, i64 19, i64 83] +// CK2: [[MTYPE00:@.+]] = {{.*}}constant [2 x i64] [i64 32, i64 281474976710739] +// CK2: [[MTYPE01:@.+]] = {{.*}}constant [2 x i64] [i64 32, i64 281474976710739] +// CK2: [[MTYPE02:@.+]] = {{.*}}constant [3 x i64] [i64 35, i64 32, i64 562949953421392] +// CK2: [[MTYPE03:@.+]] = {{.*}}constant [3 x i64] [i64 32, i64 281474976710739, i64 281474976710736] template <typename T> struct ST { @@ -366,7 +382,7 @@ struct ST { // CK2: getelementptr inbounds double, double* [[TTT]], i32 1 a++; - // CK2: [[BP:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* %{{.+}}, i32 0, i32 2 + // CK2: [[BP:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* %{{.+}}, i32 0, i32 1 // CK2: [[CBP:%.+]] = bitcast i8** [[BP]] to double*** // CK2: store double** [[RVAL:%.+]], double*** [[CBP]], // CK2: call void @__tgt_target_data_begin{{.+}}[[MTYPE01]] @@ -388,11 +404,12 @@ struct ST { // CK2: getelementptr inbounds double, double* [[TTTT]], i32 1 b++; - // CK2: [[BP:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* %{{.+}}, i32 0, i32 0 - // CK2: [[CBP:%.+]] = bitcast i8** [[BP]] to double** - // CK2: store double* [[RVAL:%.+]], double** [[CBP]], + // CK2: [[BP:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* %{{.+}}, i32 0, i32 2 + // CK2: [[CBP:%.+]] = bitcast i8** [[BP]] to double*** + // CK2: store double** [[RVAL:%.+]], double*** [[CBP]], // CK2: call void @__tgt_target_data_begin{{.+}}[[MTYPE02]] - // CK2: [[VAL:%.+]] = load double*, double** [[CBP]], + // CK2: [[CVAL:%.+]] = bitcast double*** [[CBP]] to double** + // CK2: [[VAL:%.+]] = load double*, double** [[CVAL]], // CK2: store double* [[VAL]], double** [[PVT:%.+]], // CK2: store double** [[PVT]], double*** [[PVT2:%.+]], // CK2: [[TT1:%.+]] = load double**, double*** [[PVT2]], @@ -410,26 +427,27 @@ struct ST { a++; la++; - // CK2: [[BP:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* %{{.+}}, i32 0, i32 0 - // CK2: [[CBP:%.+]] = bitcast i8** [[BP]] to double** - // CK2: store double* [[RVAL:%.+]], double** [[CBP]], - // CK2: [[_BP:%.+]] = getelementptr inbounds [4 x i8*], [4 x i8*]* %{{.+}}, i32 0, i32 3 - // CK2: [[_CBP:%.+]] = bitcast i8** [[_BP]] to double*** - // CK2: store double** [[_RVAL:%.+]], double*** [[_CBP]], + // CK2: [[BP1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* %{{.+}}, i32 0, i32 1 + // CK2: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double*** + // CK2: store double** [[RVAL1:%.+]], double*** [[CBP1]], + // CK2: [[BP2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* %{{.+}}, i32 0, i32 2 + // CK2: [[CBP2:%.+]] = bitcast i8** [[BP2]] to double*** + // CK2: store double** [[RVAL2:%.+]], double*** [[CBP2]], // CK2: call void @__tgt_target_data_begin{{.+}}[[MTYPE03]] - // CK2: [[VAL:%.+]] = load double*, double** [[CBP]], - // CK2: store double* [[VAL]], double** [[PVT:%.+]], - // CK2: store double** [[PVT]], double*** [[PVT2:%.+]], - // CK2: [[_CBP1:%.+]] = bitcast double*** [[_CBP]] to double** - // CK2: [[_VAL:%.+]] = load double*, double** [[_CBP1]], - // CK2: store double* [[_VAL]], double** [[_PVT:%.+]], - // CK2: store double** [[_PVT]], double*** [[_PVT2:%.+]], - // CK2: [[TT1:%.+]] = load double**, double*** [[PVT2]], - // CK2: [[TT2:%.+]] = load double*, double** [[TT1]], - // CK2: getelementptr inbounds double, double* [[TT2]], i32 1 - // CK2: [[_TT1:%.+]] = load double**, double*** [[_PVT2]], - // CK2: [[_TT2:%.+]] = load double*, double** [[_TT1]], + // CK2: [[_CBP2:%.+]] = bitcast double*** [[CBP2]] to double** + // CK2: [[VAL2:%.+]] = load double*, double** [[_CBP2]], + // CK2: store double* [[VAL2]], double** [[PVT2:%.+]], + // CK2: store double** [[PVT2]], double*** [[_PVT2:%.+]], + // CK2: [[_CBP1:%.+]] = bitcast double*** [[CBP1]] to double** + // CK2: [[VAL1:%.+]] = load double*, double** [[_CBP1]], + // CK2: store double* [[VAL1]], double** [[PVT1:%.+]], + // CK2: store double** [[PVT1]], double*** [[_PVT1:%.+]], + // CK2: [[TT2:%.+]] = load double**, double*** [[_PVT2]], + // CK2: [[_TT2:%.+]] = load double*, double** [[TT2]], // CK2: getelementptr inbounds double, double* [[_TT2]], i32 1 + // CK2: [[TT1:%.+]] = load double**, double*** [[_PVT1]], + // CK2: [[_TT1:%.+]] = load double*, double** [[TT1]], + // CK2: getelementptr inbounds double, double* [[_TT1]], i32 1 #pragma omp target data map(b[:10]) use_device_ptr(a, b) { a++; diff --git a/test/OpenMP/target_data_use_device_ptr_messages.cpp b/test/OpenMP/target_data_use_device_ptr_messages.cpp index 1d8002c1ff603..87f23e2de77b0 100644 --- a/test/OpenMP/target_data_use_device_ptr_messages.cpp +++ b/test/OpenMP/target_data_use_device_ptr_messages.cpp @@ -1,4 +1,6 @@ // RUN: %clang_cc1 -std=c++11 -verify -fopenmp -ferror-limit 200 %s + +// RUN: %clang_cc1 -std=c++11 -verify -fopenmp-simd -ferror-limit 200 %s struct ST { int *a; }; diff --git a/test/OpenMP/target_defaultmap_messages.cpp b/test/OpenMP/target_defaultmap_messages.cpp index 59348d42c5253..27079c3028110 100644 --- a/test/OpenMP/target_defaultmap_messages.cpp +++ b/test/OpenMP/target_defaultmap_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_depend_codegen.cpp b/test/OpenMP/target_depend_codegen.cpp new file mode 100644 index 0000000000000..bd709e1def7d6 --- /dev/null +++ b/test/OpenMP/target_depend_codegen.cpp @@ -0,0 +1,261 @@ +// Test host codegen. +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// CHECK-DAG: [[TT:%.+]] = type { i64, i8 } +// CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } +// CHECK-DAG: [[DEVTY:%.+]] = type { i8*, i8*, [[ENTTY]]*, [[ENTTY]]* } +// CHECK-DAG: [[DSCTY:%.+]] = type { i32, [[DEVTY]]*, [[ENTTY]]*, [[ENTTY]]* } + +// TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 } + +// CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat + +// CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 0, i[[SZ]] 4] +// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 800] +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] + +// Check if offloading descriptor is created. +// CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] +// CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] +// CHECK: [[DEVBEGIN:@.+]] = external constant i8 +// CHECK: [[DEVEND:@.+]] = external constant i8 +// CHECK: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x [[DEVTY]]] [{{.+}} { i8* [[DEVBEGIN]], i8* [[DEVEND]], [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }], comdat($[[REGFN]]) +// CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) + +// Check target registration is registered as a Ctor. +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] + + +template<typename tx, typename ty> +struct TT{ + tx X; + ty Y; +}; + +int global; +extern int global; + +// CHECK: define {{.*}}[[FOO:@.+]]( +int foo(int n) { + int a = 0; + short aa = 0; + float b[10]; + float bn[n]; + double c[5][10]; + double cn[5][n]; + TT<long long, char> d; + static long *plocal; + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID:%.+]], i32 1, i[[SZ]] {{20|40}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY0:%.+]]* + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 3 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* @0, i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + #pragma omp target device(global + a) depend(in: global) depend(out: a, b, cn[4]) + { + } + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + + // CHECK: [[BOOL:%.+]] = icmp ne i32 %{{.+}}, 0 + // CHECK: br i1 [[BOOL]], label %[[THEN:.+]], label %[[ELSE:.+]] + // CHECK: [[THEN]]: + // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%.+]], i32 0, i32 0 + // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%.+]], i32 0, i32 0 + // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]** + // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]** + // CHECK-DAG: store i[[SZ]]* [[BP0:%[^,]+]], i[[SZ]]** [[CBPADDR0]] + // CHECK-DAG: store i[[SZ]]* [[BP0]], i[[SZ]]** [[CPADDR0]] + + // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1 + // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1 + // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* + // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* + // CHECK-DAG: store i[[SZ]] [[BP1:%[^,]+]], i[[SZ]]* [[CBPADDR1]] + // CHECK-DAG: store i[[SZ]] [[BP1]], i[[SZ]]* [[CPADDR1]] + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID]], i32 1, i[[SZ]] {{104|52}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1_:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1_:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + + // CHECK: [[ELSE]]: + // CHECK-NOT: getelementptr inbounds [2 x i8*], [2 x i8*]* + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1__:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1__:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + // CHECK: [[EXIT]]: + + #pragma omp target device(global + a) nowait depend(inout: global, a, bn) if(a) + { + static int local1; + *plocal = global; + local1 = global; + } + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]* + // CHECK: getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* @0, i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + #pragma omp target if(0) firstprivate(global) depend(out:global) + { + global += 1; + } + + return a; +} + +// Check that the offloading functions are emitted and that the arguments are +// correct and loaded correctly for the target regions in foo(). + +// CHECK: define internal void [[HVT0:@.+]]() + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, [[TASK_TY0]]* noalias) +// CHECK: store void (i8*, ...)* null, void (i8*, ...)** % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 0, i8** null, i8** null, i[[SZ]]* null, i64* null) +// CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: call void [[HVT0]]() +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT1:@.+]](i[[SZ]]* %{{.+}}, i[[SZ]] %{{.+}}) + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1_]](i32{{.*}}, [[TASK_TY1_]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[SZT:%.+]] = getelementptr inbounds [2 x i[[SZ]]], [2 x i[[SZ]]]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_nowait(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 2, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SZT]], i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT]], i32 0, i32 0) + +// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1__]](i32{{.*}}, [[TASK_TY1__]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT2:@.+]](i[[SZ]] %{{.+}}) +// Create stack storage and store argument in there. +// CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align +// CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align +// CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* +// CHECK-64: load i32, i32* [[AA_CADDR]], align +// CHECK-32: load i32, i32* [[AA_ADDR]], align + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, [[TASK_TY2]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT2]](i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + + +#endif diff --git a/test/OpenMP/target_depend_messages.cpp b/test/OpenMP/target_depend_messages.cpp index 3eb3af132d8fd..5fdf37daf7dde 100644 --- a/test/OpenMP/target_depend_messages.cpp +++ b/test/OpenMP/target_depend_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_device_messages.cpp b/test/OpenMP/target_device_messages.cpp index 3befcd690810b..8cf19b25970d4 100644 --- a/test/OpenMP/target_device_messages.cpp +++ b/test/OpenMP/target_device_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_enter_data_ast_print.cpp b/test/OpenMP/target_enter_data_ast_print.cpp index 865cc7f1c17ef..1df1f8e1d6184 100644 --- a/test/OpenMP/target_enter_data_ast_print.cpp +++ b/test/OpenMP/target_enter_data_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -65,7 +69,7 @@ T tmain(T argc, T *argv) { // CHECK: template <typename T, int C> T tmain(T argc, T *argv) { // CHECK-NEXT: T i, j, b, c, d, e, x[20]; // CHECK-NEXT: i = argc; -// CHECK-NEXT: #pragma omp target enter data map(to: i) +// CHECK-NEXT: #pragma omp target enter data map(to: i){{$}} // CHECK-NEXT: #pragma omp target enter data map(to: i) if(target enter data: j > 0) // CHECK-NEXT: #pragma omp target enter data map(to: i) if(b) // CHECK-NEXT: #pragma omp target enter data map(to: c) diff --git a/test/OpenMP/target_enter_data_codegen.cpp b/test/OpenMP/target_enter_data_codegen.cpp index 08b6c7c2702e5..b337fccb1c4ce 100644 --- a/test/OpenMP/target_enter_data_codegen.cpp +++ b/test/OpenMP/target_enter_data_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 // CK1: [[ST:%.+]] = type { i32, double* } @@ -29,8 +37,9 @@ double gc[100]; // CK1: [[MTYPE03:@.+]] = {{.+}}constant [1 x i64] [i64 37] -// CK1: [[SIZE04:@.+]] = {{.+}}constant [2 x i[[sz]]] [i[[sz]] {{8|4}}, i[[sz]] 24] -// CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i64] [i64 33, i64 17] +// CK1-64: [[SIZE04:@.+]] = {{.+}}constant [2 x i64] [i64 sdiv exact (i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), i64 ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)), i64 24] +// CK1-32: [[SIZE04:@.+]] = {{.+}}constant [2 x i32] [i32 trunc (i64 sdiv exact (i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), i64 ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)) to i32), i32 24] +// CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i64] [i64 32, i64 281474976710673] // CK1-LABEL: _Z3fooi void foo(int arg) { @@ -145,6 +154,14 @@ void foo(int arg) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 // CK2: [[ST:%.+]] = type { i32, double* } @@ -161,8 +178,7 @@ struct ST { } }; -// CK2: [[SIZE00:@.+]] = {{.+}}constant [2 x i[[sz:64|32]]] [i{{64|32}} {{8|4}}, i{{64|32}} 24] -// CK2: [[MTYPE00:@.+]] = {{.+}}constant [2 x i64] [i64 37, i64 21] +// CK2: [[MTYPE00:@.+]] = {{.+}}constant [2 x i64] [i64 32, i64 281474976710677] // CK2-LABEL: _Z3bari int bar(int arg){ @@ -173,21 +189,23 @@ int bar(int arg){ // Region 00 // CK2: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]] // CK2: [[IFTHEN]] -// CK2-DAG: call void @__tgt_target_data_begin(i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE00]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}) +// CK2-DAG: call void @__tgt_target_data_begin(i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}) // CK2-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64 // CK2-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}}, // CK2-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK2-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK2-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK2-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK2-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK2-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK2-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]** // CK2-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double*** // CK2-DAG: store [[ST]]* [[VAR0:%.+]], [[ST]]** [[CBP0]] // CK2-DAG: store double** [[SEC0:%.+]], double*** [[CP0]] +// CK2-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK2-DAG: [[SEC0]] = getelementptr inbounds {{.*}}[[ST]]* [[VAR0]], i32 0, i32 1 - // CK2-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 // CK2-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 // CK2-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double*** @@ -213,6 +231,14 @@ int bar(int arg){ // RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 + +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} #ifdef CK3 // CK3-LABEL: no_target_devices @@ -233,6 +259,14 @@ void no_target_devices(int arg) { // RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-32 +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// SIMD-ONLY3-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-64 // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s @@ -241,6 +275,16 @@ void no_target_devices(int arg) { // RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-32 // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-32 + +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY4 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY4 %s +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY4 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY4 %s +// SIMD-ONLY4-NOT: {{__kmpc|__tgt}} #ifdef CK4 // CK4-LABEL: device_side_scan diff --git a/test/OpenMP/target_enter_data_depend_codegen.cpp b/test/OpenMP/target_enter_data_depend_codegen.cpp index 469a161eb341a..e17150d576777 100644 --- a/test/OpenMP/target_enter_data_depend_codegen.cpp +++ b/test/OpenMP/target_enter_data_depend_codegen.cpp @@ -5,6 +5,14 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics // CK1: [[ST:%.+]] = type { i32, double* } // CK1: %struct.kmp_depend_info = type { i[[sz:64|32]], @@ -29,8 +37,9 @@ double gc[100]; // CK1: [[MTYPE03:@.+]] = {{.+}}constant [1 x i64] [i64 32] -// CK1: [[SIZE04:@.+]] = {{.+}}constant [2 x i[[sz]]] [i[[sz]] {{8|4}}, i[[sz]] 24] -// CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i64] [i64 33, i64 17] +// CK1-64: [[SIZE04:@.+]] = {{.+}}constant [2 x i64] [i64 sdiv exact (i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), i64 ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)), i64 24] +// CK1-32: [[SIZE04:@.+]] = {{.+}}constant [2 x i32] [i32 trunc (i64 sdiv exact (i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), i64 ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)) to i32), i32 24] +// CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i64] [i64 32, i64 281474976710673] // CK1-LABEL: _Z3fooi void foo(int arg) { @@ -54,25 +63,25 @@ void foo(int arg) { // CK1: [[CAP_DEVICE:%.+]] = getelementptr inbounds %struct.anon, %struct.anon* [[CAPTURES:%.+]], i32 0, i32 0 // CK1: [[DEVICE:%.+]] = load i32, i32* %{{.+}} // CK1: store i32 [[DEVICE]], i32* [[CAP_DEVICE]], - // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) + // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) // CK1: [[BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates* // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, %struct.kmp_task_t_with_privates* [[BC]], i32 0, i32 0 // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, %struct.kmp_task_t* [[TASK_T]], i32 0, i32 0 // CK1: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS]], // CK1: [[BC1:%.+]] = bitcast %struct.anon* [[CAPTURES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[SHAREDS_REF]], i8* [[BC1]], i[[sz]] 4, i32 4, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 4 [[SHAREDS_REF]], i8* align 4 [[BC1]], i[[sz]] 4, i1 false) // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, %struct.kmp_task_t_with_privates* [[BC]], i32 0, i32 1 // CK1: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 0 // CK1: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8* // CK1: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_BASEPTRS]], i8* [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 1 // CK1: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8* // CK1: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_PTRS]], i8* [[BC_PTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 2 // CK1: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_SIZES]], i8* bitcast ([1 x i[[sz]]]* [[SIZE00]] to i8*), i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i[[sz]]]* [[SIZE00]] to i8*), i[[sz]] {{8|4}}, i1 false) // CK1: [[DEP:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* [[MAIN_DEP:%.+]], i[[sz]] 0, i[[sz]] 0 // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0 // CK1: [[BC_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]] @@ -83,7 +92,7 @@ void foo(int arg) { // CK1: store i8 1, i8* [[DEP_ATTRS]] // CK1: [[DEP:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* [[MAIN_DEP]], i32 0, i32 0 // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[DEP]] to i8* - // CK1: = call i32 @__kmpc_omp_task_with_deps(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i8* [[BC]], i32 0, i8* null) + // CK1: = call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i8* [[BC]], i32 0, i8* null) // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1 #pragma omp target enter data if(1+3-5) device(arg) map(alloc:gc) nowait depend(in: arg) @@ -110,25 +119,25 @@ void foo(int arg) { // CK1: [[IF_BOOL:%.+]] = trunc i8 [[IF]] to i1 // CK1: [[IF:%.+]] = zext i1 [[IF_BOOL]] to i8 // CK1: store i8 [[IF]], i8* [[IF_DEVICE]], - // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) + // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}* // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0 // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, %struct.kmp_task_t* [[TASK_T]], i32 0, i32 0 // CK1: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS]], // CK1: [[BC1:%.+]] = bitcast %struct.anon{{.+}}* [[CAPTURES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[SHAREDS_REF]], i8* [[BC1]], i[[sz]] 1, i32 1, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 1 [[SHAREDS_REF]], i8* align 1 [[BC1]], i[[sz]] 1, i1 false) // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1 // CK1: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0 // CK1: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8* // CK1: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_BASEPTRS]], i8* [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1 // CK1: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8* // CK1: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_PTRS]], i8* [[BC_PTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2 // CK1: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_SIZES]], i8* bitcast ([1 x i[[sz]]]* [[SIZE02]] to i8*), i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i[[sz]]]* [[SIZE02]] to i8*), i[[sz]] {{8|4}}, i1 false) // CK1: [[DEP:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* [[MAIN_DEP:%.+]], i[[sz]] 0, i[[sz]] 0 // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0 // CK1: [[BC_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]] @@ -154,10 +163,10 @@ void foo(int arg) { // CK1: store i8 3, i8* [[DEP_ATTRS]] // CK1: [[DEP:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* [[MAIN_DEP]], i32 0, i32 0 // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[DEP]] to i8* - // CK1: call void @__kmpc_omp_wait_deps(%ident_t* @{{.+}}, i32 %{{.+}}, i32 3, i8* [[BC]], i32 0, i8* null) - // CK1: call void @__kmpc_omp_task_begin_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 3, i8* [[BC]], i32 0, i8* null) + // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) // CK1: = call i32 [[TASK_ENTRY2]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]]) - // CK1: call void @__kmpc_omp_task_complete_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) // CK1: br label %[[IFEND:[^,]+]] @@ -183,22 +192,22 @@ void foo(int arg) { // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0 // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0 // CK1: [[GEPS0:%.+]] = getelementptr inbounds [1 x i[[sz]]], [1 x i[[sz]]]* [[S]], i32 0, i32 0 - // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY3:@.+]] to i32 (i32, i8*)*)) + // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY3:@.+]] to i32 (i32, i8*)*)) // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}* // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0 // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1 // CK1: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0 // CK1: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8* // CK1: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_BASEPTRS]], i8* [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1 // CK1: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8* // CK1: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_PTRS]], i8* [[BC_PTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2 // CK1: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8* // CK1: [[BC_SIZES:%.+]] = bitcast i[[sz]]* [[GEPS0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_SIZES]], i8* [[BC_SIZES]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{8|4}}, i1 false) // CK1: [[DEP:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* [[MAIN_DEP:%.+]], i[[sz]] 0, i[[sz]] 0 // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0 // CK1: [[BC_ADR:%.+]] = ptrtoint float* %{{.+}} to i[[sz]] @@ -232,10 +241,10 @@ void foo(int arg) { // CK1: store i8 3, i8* [[DEP_ATTRS]] // CK1: [[DEP:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* [[MAIN_DEP]], i32 0, i32 0 // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[DEP]] to i8* - // CK1: call void @__kmpc_omp_wait_deps(%ident_t* @{{.+}}, i32 %{{.+}}, i32 4, i8* [[BC]], i32 0, i8* null) - // CK1: call void @__kmpc_omp_task_begin_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 4, i8* [[BC]], i32 0, i8* null) + // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) // CK1: = call i32 [[TASK_ENTRY3]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]]) - // CK1: call void @__kmpc_omp_task_complete_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) #pragma omp target enter data map(alloc:lb) depend(out: lb, arg, la, gc) {++arg;} @@ -257,21 +266,21 @@ void foo(int arg) { // CK1: store double* %{{.+}}, double** [[P1_BC]], // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 // CK1: [[GEPP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 - // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{88|44}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY4:@.+]] to i32 (i32, i8*)*)) + // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{88|44}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY4:@.+]] to i32 (i32, i8*)*)) // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}* // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0 // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1 // CK1: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0 // CK1: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_BASEPTRS]] to i8* // CK1: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_BASEPTRS]], i8* [[BC_BASEPTRS]], i[[sz]] {{16|8}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{16|8}}, i1 false) // CK1: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1 // CK1: [[BC_PRIVS_PTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_PTRS]] to i8* // CK1: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_PTRS]], i8* [[BC_PTRS]], i[[sz]] {{16|8}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{16|8}}, i1 false) // CK1: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2 // CK1: [[BC_PRIVS_SIZES:%.+]] = bitcast [2 x i[[sz]]]* [[PRIVS_SIZES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_SIZES]], i8* bitcast ([2 x i[[sz]]]* [[SIZE04]] to i8*), i[[sz]] {{16|8}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([2 x i[[sz]]]* [[SIZE04]] to i8*), i[[sz]] {{16|8}}, i1 false) // CK1: [[DEP:%.+]] = getelementptr inbounds [5 x %struct.kmp_depend_info], [5 x %struct.kmp_depend_info]* [[MAIN_DEP:%.+]], i[[sz]] 0, i[[sz]] 0 // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0 // CK1: [[BC_ADR:%.+]] = ptrtoint double* %{{.+}} to i[[sz]] @@ -313,10 +322,10 @@ void foo(int arg) { // CK1: store i8 1, i8* [[DEP_ATTRS]] // CK1: [[DEP:%.+]] = getelementptr inbounds [5 x %struct.kmp_depend_info], [5 x %struct.kmp_depend_info]* [[MAIN_DEP]], i32 0, i32 0 // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[DEP]] to i8* - // CK1: call void @__kmpc_omp_wait_deps(%ident_t* @{{.+}}, i32 %{{.+}}, i32 5, i8* [[BC]], i32 0, i8* null) - // CK1: call void @__kmpc_omp_task_begin_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 5, i8* [[BC]], i32 0, i8* null) + // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) // CK1: = call i32 [[TASK_ENTRY4]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]]) - // CK1: call void @__kmpc_omp_task_complete_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) #pragma omp target enter data map(to:gb.b[:3]) depend(in: gb.b[:3], la, lb, gc, arg) {++arg;} } @@ -328,9 +337,9 @@ void foo(int arg) { // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] -// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%.+]], -// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%.+]], -// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%.+]], +// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]], +// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]], +// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%[^,]+]], // CK1-DAG: call void (i8*, ...) %{{.+}}(i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i[[sz]]]** [[S_PRIV]]) // CK1: ret i32 0 // CK1: } @@ -340,9 +349,9 @@ void foo(int arg) { // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] -// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%.+]], -// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%.+]], -// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%.+]], +// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]], +// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]], +// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%[^,]+]], // CK1-DAG: call void (i8*, ...) %{{.+}}(i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i[[sz]]]** [[S_PRIV]]) // CK1: ret i32 0 // CK1: } @@ -353,9 +362,9 @@ void foo(int arg) { // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] -// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%.+]], -// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%.+]], -// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%.+]], +// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]], +// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]], +// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%[^,]+]], // CK1-DAG: call void (i8*, ...) %{{.+}}(i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i[[sz]]]** [[S_PRIV]]) // CK1-NOT: __tgt_target_data_end // CK1: ret i32 0 @@ -367,9 +376,9 @@ void foo(int arg) { // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] -// CK1-DAG: [[BP]] = load [2 x i8*]*, [2 x i8*]** [[BP_PRIV:%.+]], -// CK1-DAG: [[P]] = load [2 x i8*]*, [2 x i8*]** [[P_PRIV:%.+]], -// CK1-DAG: [[S]] = load [2 x i[[sz]]]*, [2 x i[[sz]]]** [[S_PRIV:%.+]], +// CK1-DAG: [[BP]] = load [2 x i8*]*, [2 x i8*]** [[BP_PRIV:%[^,]+]], +// CK1-DAG: [[P]] = load [2 x i8*]*, [2 x i8*]** [[P_PRIV:%[^,]+]], +// CK1-DAG: [[S]] = load [2 x i[[sz]]]*, [2 x i[[sz]]]** [[S_PRIV:%[^,]+]], // CK1-DAG: call void (i8*, ...) %{{.+}}(i8* %{{[^,]+}}, [2 x i8*]** [[BP_PRIV]], [2 x i8*]** [[P_PRIV]], [2 x i[[sz]]]** [[S_PRIV]]) // CK1-NOT: __tgt_target_data_end // CK1: ret i32 0 diff --git a/test/OpenMP/target_enter_data_depend_messages.cpp b/test/OpenMP/target_enter_data_depend_messages.cpp index 77e7039e8e205..ccc29ea49a266 100644 --- a/test/OpenMP/target_enter_data_depend_messages.cpp +++ b/test/OpenMP/target_enter_data_depend_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_enter_data_device_messages.cpp b/test/OpenMP/target_enter_data_device_messages.cpp index d954eca319dd2..de8405b4a62fc 100644 --- a/test/OpenMP/target_enter_data_device_messages.cpp +++ b/test/OpenMP/target_enter_data_device_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_enter_data_if_messages.cpp b/test/OpenMP/target_enter_data_if_messages.cpp index 137ff97516171..26404a5d97a77 100644 --- a/test/OpenMP/target_enter_data_if_messages.cpp +++ b/test/OpenMP/target_enter_data_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_enter_data_map_messages.c b/test/OpenMP/target_enter_data_map_messages.c index 8a7de3b96effb..381c94bf3986a 100644 --- a/test/OpenMP/target_enter_data_map_messages.c +++ b/test/OpenMP/target_enter_data_map_messages.c @@ -1,6 +1,9 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - -x c++ %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - -x c++ %s + int main(int argc, char **argv) { int r; diff --git a/test/OpenMP/target_enter_data_nowait_messages.cpp b/test/OpenMP/target_enter_data_nowait_messages.cpp index 508b1b254906e..6f59c5ceb7b7b 100644 --- a/test/OpenMP/target_enter_data_nowait_messages.cpp +++ b/test/OpenMP/target_enter_data_nowait_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + int main(int argc, char **argv) { int i; diff --git a/test/OpenMP/target_exit_data_ast_print.cpp b/test/OpenMP/target_exit_data_ast_print.cpp index a2f5d641f3cc6..1408023b06376 100644 --- a/test/OpenMP/target_exit_data_ast_print.cpp +++ b/test/OpenMP/target_exit_data_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -69,7 +73,7 @@ T tmain(T argc, T *argv) { // CHECK: template <typename T, int C> T tmain(T argc, T *argv) { // CHECK-NEXT: T i, j, b, c, d, e, x[20]; // CHECK-NEXT: i = argc; -// CHECK-NEXT: #pragma omp target exit data map(from: i) +// CHECK-NEXT: #pragma omp target exit data map(from: i){{$}} // CHECK-NEXT: #pragma omp target exit data map(from: i) if(target exit data: j > 0) // CHECK-NEXT: #pragma omp target exit data map(from: i) if(b) // CHECK-NEXT: #pragma omp target exit data map(from: c) diff --git a/test/OpenMP/target_exit_data_codegen.cpp b/test/OpenMP/target_exit_data_codegen.cpp index 9359e3be451b6..ff84303bb0ee8 100644 --- a/test/OpenMP/target_exit_data_codegen.cpp +++ b/test/OpenMP/target_exit_data_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 // CK1: [[ST:%.+]] = type { i32, double* } @@ -29,8 +37,9 @@ double gc[100]; // CK1: [[MTYPE03:@.+]] = {{.+}}constant [1 x i64] [i64 38] -// CK1: [[SIZE04:@.+]] = {{.+}}constant [2 x i[[sz]]] [i[[sz]] {{8|4}}, i[[sz]] 24] -// CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i64] [i64 32, i64 16] +// CK1-64: [[SIZE04:@.+]] = {{.+}}constant [2 x i64] [i64 sdiv exact (i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), i64 ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)), i64 24] +// CK1-32: [[SIZE04:@.+]] = {{.+}}constant [2 x i32] [i32 trunc (i64 sdiv exact (i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), i64 ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)) to i32), i32 24] +// CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i64] [i64 32, i64 281474976710672] // CK1-LABEL: _Z3fooi void foo(int arg) { @@ -145,6 +154,14 @@ void foo(int arg) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 // CK2: [[ST:%.+]] = type { i32, double* } @@ -161,8 +178,7 @@ struct ST { } }; -// CK2: [[SIZE00:@.+]] = {{.+}}constant [2 x i[[sz:64|32]]] [i{{64|32}} {{8|4}}, i{{64|32}} 24] -// CK2: [[MTYPE00:@.+]] = {{.+}}constant [2 x i64] [i64 36, i64 20] +// CK2: [[MTYPE00:@.+]] = {{.+}}constant [2 x i64] [i64 32, i64 281474976710676] // CK2-LABEL: _Z3bari int bar(int arg){ @@ -174,21 +190,23 @@ int bar(int arg){ // CK2-NOT: __tgt_target_data_begin // CK2: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]] // CK2: [[IFTHEN]] -// CK2-DAG: call void @__tgt_target_data_end(i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE00]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}) +// CK2-DAG: call void @__tgt_target_data_end(i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:.+]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}) // CK2-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64 // CK2-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}}, // CK2-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK2-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK2-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK2-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK2-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK2-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK2-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]** // CK2-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double*** // CK2-DAG: store [[ST]]* [[VAR0:%[^,]+]], [[ST]]** [[CBP0]] // CK2-DAG: store double** [[SEC0:%[^,]+]], double*** [[CP0]] +// CK2-DAG: store i[[sz]] [[CSVAL0:%[^,]+]], i[[sz]]* [[S0]] // CK2-DAG: [[SEC0]] = getelementptr inbounds {{.*}}[[ST]]* [[VAR0]], i32 0, i32 1 - // CK2-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 // CK2-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 // CK2-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double*** @@ -213,6 +231,14 @@ int bar(int arg){ // RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 + +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} #ifdef CK3 // CK3-LABEL: no_target_devices diff --git a/test/OpenMP/target_exit_data_depend_codegen.cpp b/test/OpenMP/target_exit_data_depend_codegen.cpp index 6f7fba23db003..fec750dde2377 100644 --- a/test/OpenMP/target_exit_data_depend_codegen.cpp +++ b/test/OpenMP/target_exit_data_depend_codegen.cpp @@ -5,6 +5,14 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics // CK1: [[ST:%.+]] = type { i32, double* } // CK1: %struct.kmp_depend_info = type { i[[sz:64|32]], @@ -29,8 +37,9 @@ double gc[100]; // CK1: [[MTYPE03:@.+]] = {{.+}}constant [1 x i64] [i64 34] -// CK1: [[SIZE04:@.+]] = {{.+}}constant [2 x i[[sz]]] [i[[sz]] {{8|4}}, i[[sz]] 24] -// CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i64] [i64 34, i64 18] +// CK1-64: [[SIZE04:@.+]] = {{.+}}constant [2 x i64] [i64 sdiv exact (i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), i64 ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)), i64 24] +// CK1-32: [[SIZE04:@.+]] = {{.+}}constant [2 x i32] [i32 trunc (i64 sdiv exact (i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), i64 ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)) to i32), i32 24] +// CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i64] [i64 32, i64 281474976710674] // CK1-LABEL: _Z3fooi void foo(int arg) { @@ -54,25 +63,25 @@ void foo(int arg) { // CK1: [[CAP_DEVICE:%.+]] = getelementptr inbounds %struct.anon, %struct.anon* [[CAPTURES:%.+]], i32 0, i32 0 // CK1: [[DEVICE:%.+]] = load i32, i32* %{{.+}} // CK1: store i32 [[DEVICE]], i32* [[CAP_DEVICE]], - // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) + // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) // CK1: [[BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates* // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, %struct.kmp_task_t_with_privates* [[BC]], i32 0, i32 0 // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, %struct.kmp_task_t* [[TASK_T]], i32 0, i32 0 // CK1: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS]], // CK1: [[BC1:%.+]] = bitcast %struct.anon* [[CAPTURES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[SHAREDS_REF]], i8* [[BC1]], i[[sz]] 4, i32 4, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 4 [[SHAREDS_REF]], i8* align 4 [[BC1]], i[[sz]] 4, i1 false) // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, %struct.kmp_task_t_with_privates* [[BC]], i32 0, i32 1 // CK1: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 0 // CK1: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8* // CK1: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_BASEPTRS]], i8* [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 1 // CK1: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8* // CK1: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_PTRS]], i8* [[BC_PTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 2 // CK1: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_SIZES]], i8* bitcast ([1 x i[[sz]]]* [[SIZE00]] to i8*), i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i[[sz]]]* [[SIZE00]] to i8*), i[[sz]] {{8|4}}, i1 false) // CK1: [[DEP:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* [[MAIN_DEP:%.+]], i[[sz]] 0, i[[sz]] 0 // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0 // CK1: [[BC_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]] @@ -83,7 +92,7 @@ void foo(int arg) { // CK1: store i8 1, i8* [[DEP_ATTRS]] // CK1: [[DEP:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* [[MAIN_DEP]], i32 0, i32 0 // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[DEP]] to i8* - // CK1: = call i32 @__kmpc_omp_task_with_deps(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i8* [[BC]], i32 0, i8* null) + // CK1: = call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i8* [[BC]], i32 0, i8* null) // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1 #pragma omp target exit data if(1+3-5) device(arg) map(from:gc) nowait depend(in: arg) @@ -110,25 +119,25 @@ void foo(int arg) { // CK1: [[IF_BOOL:%.+]] = trunc i8 [[IF]] to i1 // CK1: [[IF:%.+]] = zext i1 [[IF_BOOL]] to i8 // CK1: store i8 [[IF]], i8* [[IF_DEVICE]], - // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) + // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}* // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0 // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, %struct.kmp_task_t* [[TASK_T]], i32 0, i32 0 // CK1: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS]], // CK1: [[BC1:%.+]] = bitcast %struct.anon{{.+}}* [[CAPTURES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[SHAREDS_REF]], i8* [[BC1]], i[[sz]] 1, i32 1, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 1 [[SHAREDS_REF]], i8* align 1 [[BC1]], i[[sz]] 1, i1 false) // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1 // CK1: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0 // CK1: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8* // CK1: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_BASEPTRS]], i8* [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1 // CK1: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8* // CK1: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_PTRS]], i8* [[BC_PTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2 // CK1: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_SIZES]], i8* bitcast ([1 x i[[sz]]]* [[SIZE02]] to i8*), i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i[[sz]]]* [[SIZE02]] to i8*), i[[sz]] {{8|4}}, i1 false) // CK1: [[DEP:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* [[MAIN_DEP:%.+]], i[[sz]] 0, i[[sz]] 0 // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0 // CK1: [[BC_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]] @@ -154,10 +163,10 @@ void foo(int arg) { // CK1: store i8 3, i8* [[DEP_ATTRS]] // CK1: [[DEP:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* [[MAIN_DEP]], i32 0, i32 0 // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[DEP]] to i8* - // CK1: call void @__kmpc_omp_wait_deps(%ident_t* @{{.+}}, i32 %{{.+}}, i32 3, i8* [[BC]], i32 0, i8* null) - // CK1: call void @__kmpc_omp_task_begin_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 3, i8* [[BC]], i32 0, i8* null) + // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) // CK1: = call i32 [[TASK_ENTRY2]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]]) - // CK1: call void @__kmpc_omp_task_complete_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) // CK1: br label %[[IFEND:[^,]+]] @@ -183,22 +192,22 @@ void foo(int arg) { // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0 // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0 // CK1: [[GEPS0:%.+]] = getelementptr inbounds [1 x i[[sz]]], [1 x i[[sz]]]* [[S]], i32 0, i32 0 - // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY3:@.+]] to i32 (i32, i8*)*)) + // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY3:@.+]] to i32 (i32, i8*)*)) // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}* // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0 // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1 // CK1: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0 // CK1: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8* // CK1: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_BASEPTRS]], i8* [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1 // CK1: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8* // CK1: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_PTRS]], i8* [[BC_PTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2 // CK1: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8* // CK1: [[BC_SIZES:%.+]] = bitcast i[[sz]]* [[GEPS0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_SIZES]], i8* [[BC_SIZES]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{8|4}}, i1 false) // CK1: [[DEP:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* [[MAIN_DEP:%.+]], i[[sz]] 0, i[[sz]] 0 // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0 // CK1: [[BC_ADR:%.+]] = ptrtoint float* %{{.+}} to i[[sz]] @@ -232,10 +241,10 @@ void foo(int arg) { // CK1: store i8 3, i8* [[DEP_ATTRS]] // CK1: [[DEP:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* [[MAIN_DEP]], i32 0, i32 0 // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[DEP]] to i8* - // CK1: call void @__kmpc_omp_wait_deps(%ident_t* @{{.+}}, i32 %{{.+}}, i32 4, i8* [[BC]], i32 0, i8* null) - // CK1: call void @__kmpc_omp_task_begin_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 4, i8* [[BC]], i32 0, i8* null) + // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) // CK1: = call i32 [[TASK_ENTRY3]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]]) - // CK1: call void @__kmpc_omp_task_complete_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) #pragma omp target exit data map(from:lb) depend(out: lb, arg, la, gc) {++arg;} @@ -257,21 +266,21 @@ void foo(int arg) { // CK1: store double* %{{.+}}, double** [[P1_BC]], // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 // CK1: [[GEPP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 - // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{88|44}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY4:@.+]] to i32 (i32, i8*)*)) + // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{88|44}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY4:@.+]] to i32 (i32, i8*)*)) // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}* // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0 // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1 // CK1: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0 // CK1: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_BASEPTRS]] to i8* // CK1: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_BASEPTRS]], i8* [[BC_BASEPTRS]], i[[sz]] {{16|8}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{16|8}}, i1 false) // CK1: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1 // CK1: [[BC_PRIVS_PTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_PTRS]] to i8* // CK1: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_PTRS]], i8* [[BC_PTRS]], i[[sz]] {{16|8}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{16|8}}, i1 false) // CK1: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2 // CK1: [[BC_PRIVS_SIZES:%.+]] = bitcast [2 x i[[sz]]]* [[PRIVS_SIZES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_SIZES]], i8* bitcast ([2 x i[[sz]]]* [[SIZE04]] to i8*), i[[sz]] {{16|8}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([2 x i[[sz]]]* [[SIZE04]] to i8*), i[[sz]] {{16|8}}, i1 false) // CK1: [[DEP:%.+]] = getelementptr inbounds [5 x %struct.kmp_depend_info], [5 x %struct.kmp_depend_info]* [[MAIN_DEP:%.+]], i[[sz]] 0, i[[sz]] 0 // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0 // CK1: [[BC_ADR:%.+]] = ptrtoint double* %{{.+}} to i[[sz]] @@ -313,10 +322,10 @@ void foo(int arg) { // CK1: store i8 1, i8* [[DEP_ATTRS]] // CK1: [[DEP:%.+]] = getelementptr inbounds [5 x %struct.kmp_depend_info], [5 x %struct.kmp_depend_info]* [[MAIN_DEP]], i32 0, i32 0 // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[DEP]] to i8* - // CK1: call void @__kmpc_omp_wait_deps(%ident_t* @{{.+}}, i32 %{{.+}}, i32 5, i8* [[BC]], i32 0, i8* null) - // CK1: call void @__kmpc_omp_task_begin_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 5, i8* [[BC]], i32 0, i8* null) + // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) // CK1: = call i32 [[TASK_ENTRY4]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]]) - // CK1: call void @__kmpc_omp_task_complete_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) #pragma omp target exit data map(from:gb.b[:3]) depend(in: gb.b[:3], la, lb, gc, arg) {++arg;} } @@ -328,9 +337,9 @@ void foo(int arg) { // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] -// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%.+]], -// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%.+]], -// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%.+]], +// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]], +// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]], +// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%[^,]+]], // CK1-DAG: call void (i8*, ...) %{{.+}}(i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i[[sz]]]** [[S_PRIV]]) // CK1: ret i32 0 // CK1: } @@ -340,9 +349,9 @@ void foo(int arg) { // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] -// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%.+]], -// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%.+]], -// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%.+]], +// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]], +// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]], +// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%[^,]+]], // CK1-DAG: call void (i8*, ...) %{{.+}}(i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i[[sz]]]** [[S_PRIV]]) // CK1: ret i32 0 // CK1: } @@ -353,9 +362,9 @@ void foo(int arg) { // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] -// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%.+]], -// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%.+]], -// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%.+]], +// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]], +// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]], +// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%[^,]+]], // CK1-DAG: call void (i8*, ...) %{{.+}}(i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i[[sz]]]** [[S_PRIV]]) // CK1-NOT: __tgt_target_data_end // CK1: ret i32 0 @@ -367,9 +376,9 @@ void foo(int arg) { // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] -// CK1-DAG: [[BP]] = load [2 x i8*]*, [2 x i8*]** [[BP_PRIV:%.+]], -// CK1-DAG: [[P]] = load [2 x i8*]*, [2 x i8*]** [[P_PRIV:%.+]], -// CK1-DAG: [[S]] = load [2 x i[[sz]]]*, [2 x i[[sz]]]** [[S_PRIV:%.+]], +// CK1-DAG: [[BP]] = load [2 x i8*]*, [2 x i8*]** [[BP_PRIV:%[^,]+]], +// CK1-DAG: [[P]] = load [2 x i8*]*, [2 x i8*]** [[P_PRIV:%[^,]+]], +// CK1-DAG: [[S]] = load [2 x i[[sz]]]*, [2 x i[[sz]]]** [[S_PRIV:%[^,]+]], // CK1-DAG: call void (i8*, ...) %{{.+}}(i8* %{{[^,]+}}, [2 x i8*]** [[BP_PRIV]], [2 x i8*]** [[P_PRIV]], [2 x i[[sz]]]** [[S_PRIV]]) // CK1-NOT: __tgt_target_data_end // CK1: ret i32 0 diff --git a/test/OpenMP/target_exit_data_depend_messages.cpp b/test/OpenMP/target_exit_data_depend_messages.cpp index ce8d237f75e9e..7334cd751f14d 100644 --- a/test/OpenMP/target_exit_data_depend_messages.cpp +++ b/test/OpenMP/target_exit_data_depend_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_exit_data_device_messages.cpp b/test/OpenMP/target_exit_data_device_messages.cpp index d9ce0ddd7ea00..ad27e8a50dd8c 100644 --- a/test/OpenMP/target_exit_data_device_messages.cpp +++ b/test/OpenMP/target_exit_data_device_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_exit_data_if_messages.cpp b/test/OpenMP/target_exit_data_if_messages.cpp index 6d5ada14e9da1..9341fe81794b8 100644 --- a/test/OpenMP/target_exit_data_if_messages.cpp +++ b/test/OpenMP/target_exit_data_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_exit_data_map_messages.c b/test/OpenMP/target_exit_data_map_messages.c index 84bd7e79fd03b..1c15aac2e5d9a 100644 --- a/test/OpenMP/target_exit_data_map_messages.c +++ b/test/OpenMP/target_exit_data_map_messages.c @@ -1,6 +1,9 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - -x c++ %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - -x c++ %s + int main(int argc, char **argv) { int r; diff --git a/test/OpenMP/target_exit_data_nowait_messages.cpp b/test/OpenMP/target_exit_data_nowait_messages.cpp index d0286031f3624..02eded5a66df7 100644 --- a/test/OpenMP/target_exit_data_nowait_messages.cpp +++ b/test/OpenMP/target_exit_data_nowait_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + int main(int argc, char **argv) { int i; diff --git a/test/OpenMP/target_firstprivate_codegen.cpp b/test/OpenMP/target_firstprivate_codegen.cpp index c2909c6f18b8f..4a2837b3c83cc 100644 --- a/test/OpenMP/target_firstprivate_codegen.cpp +++ b/test/OpenMP/target_firstprivate_codegen.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 @@ -16,6 +24,16 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -33,15 +51,15 @@ struct TT{ // TCHECK: [[S1:%.+]] = type { double } // CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [1 x i{{32|64}}] [i[[SZ:32|64]] 4] -// CHECK: [[MAPT:@.+]] = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [9 x i64] [i64 288, i64 161, i64 288, i64 161, i64 161, i64 288, i64 288, i64 161, i64 161] +// CHECK: [[MAPT:@.+]] = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [9 x i64] [i64 800, i64 673, i64 288, i64 673, i64 673, i64 288, i64 288, i64 673, i64 673] // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [1 x i{{32|64}}] zeroinitializer -// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [1 x i64] [i64 32] -// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [5 x i64] [i64 547, i64 288, i64 288, i64 288, i64 161] +// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [1 x i64] [i64 544] +// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [6 x i64] [i64 32, i64 281474976711171, i64 800, i64 288, i64 288, i64 673] // CHECK-DAG: [[SIZET5:@.+]] = private unnamed_addr constant [3 x i{{32|64}}] [i[[SZ]] 4, i[[SZ]] 1, i[[SZ]] 40] -// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 161] +// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 673] // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [2 x i{{32|64}}] [i[[SZ]] 4, i[[SZ]] 40] -// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [2 x i64] [i64 288, i64 161] +// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 673] // CHECK: define {{.*}}[[FOO:@.+]]( @@ -102,7 +120,7 @@ int foo(int n, double *ptr) { // CHECK: [[PTR_GEP_ARG:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: {{.+}} = call i32 @__tgt_target(i64 -1, {{.+}}, i32 1, i8** [[BASE_PTR_GEP_ARG]], i8** [[PTR_GEP_ARG]], i[[SZ]]* getelementptr inbounds ([1 x i[[SZ]]], [1 x i[[SZ]]]* [[SIZET]], i32 0, i32 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* [[MAPT]], i32 0, i32 0)) - // TCHECK: define void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]]) + // TCHECK: define weak void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]]) // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}}, // TCHECK-NOT: alloca i{{[0-9]+}}, // TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]], @@ -263,7 +281,7 @@ int foo(int n, double *ptr) { // firstprivate(b): memcpy(b_priv,b_in) // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x float]* [[B_PRIV]] to i8* // TCHECK: [[B_ADDR_REF_BCAST:%.+]] = bitcast [10 x float]* [[B_ADDR_REF]] to i8* - // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[B_PRIV_BCAST]], i8* [[B_ADDR_REF_BCAST]], {{.+}}) + // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_ADDR_REF_BCAST]], {{.+}}) // TCHECK: [[RET_STACK:%.+]] = call i8* @llvm.stacksave() // TCHECK: store i8* [[RET_STACK]], i8** [[SSTACK]], @@ -273,12 +291,12 @@ int foo(int n, double *ptr) { // TCHECK: [[BN_COPY_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[BN_SZ_VAL]], 4 // TCHECK: [[BN_PRIV__BCAST:%.+]] = bitcast float* [[BN_PRIV]] to i8* // TCHECK: [[BN_REF_IN_BCAST:%.+]] = bitcast float* [[BN_ADDR_REF]] to i8* - // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[BN_PRIV__BCAST]], i8* [[BN_REF_IN_BCAST]], i{{[0-9]+}} [[BN_COPY_SZ]],{{.+}}) + // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BN_PRIV__BCAST]], i8* align {{[0-9]+}} [[BN_REF_IN_BCAST]], i{{[0-9]+}} [[BN_COPY_SZ]],{{.+}}) // firstprivate(c) // TCHECK: [[C_PRIV_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C_PRIV]] to i8* // TCHECK: [[C_IN_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C_ADDR_REF]] to i8* - // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[C_PRIV_BCAST]], i8* [[C_IN_BCAST]],{{.+}}) + // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[C_PRIV_BCAST]], i8* align {{[0-9]+}} [[C_IN_BCAST]],{{.+}}) // firstprivate(cn) // TCHECK: [[CN_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ1_VAL]], [[CN_SZ2_VAL]] @@ -287,12 +305,12 @@ int foo(int n, double *ptr) { // TCHECK: [[CN_SZ2_CPY:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ2]], 8 // TCHECK: [[CN_PRIV_BCAST:%.+]] = bitcast double* [[CN_PRIV]] to i8* // TCHECK: [[CN_IN_BCAST:%.+]] = bitcast double* [[CN_ADDR_REF]] to i8* - // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[CN_PRIV_BCAST]], i8* [[CN_IN_BCAST]], i{{[0-9]+}} [[CN_SZ2_CPY]],{{.+}}) + // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[CN_PRIV_BCAST]], i8* align {{[0-9]+}} [[CN_IN_BCAST]], i{{[0-9]+}} [[CN_SZ2_CPY]],{{.+}}) // firstprivate(d) // TCHECK: [[D_PRIV_BCAST:%.+]] = bitcast [[TT]]* [[D_PRIV]] to i8* // TCHECK: [[D_IN_BCAST:%.+]] = bitcast [[TT]]* [[D_ADDR_REF]] to i8* - // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[D_PRIV_BCAST]], i8* [[D_IN_BCAST]],{{.+}}) + // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[D_PRIV_BCAST]], i8* align {{[0-9]+}} [[D_IN_BCAST]],{{.+}}) #pragma omp target firstprivate(ptr) @@ -312,7 +330,7 @@ int foo(int n, double *ptr) { // CHECK: [[PTR_GEP_ARG3:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: {{.+}} = call i32 @__tgt_target(i64 -1, {{.+}}, i32 1, i8** [[BASE_PTR_GEP_ARG3]], i8** [[PTR_GEP_ARG3]], i[[SZ]]* getelementptr inbounds ([1 x i[[SZ]]], [1 x i[[SZ]]]* [[SIZET3]], i32 0, i32 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* [[MAPT3]], i32 0, i32 0)) - // TCHECK: define void @__omp_offloading_{{.+}}(double* [[PTR_IN:%.+]]) + // TCHECK: define weak void @__omp_offloading_{{.+}}(double* [[PTR_IN:%.+]]) // TCHECK: [[PTR_ADDR:%.+]] = alloca double*, // TCHECK-NOT: alloca double*, // TCHECK: store double* [[PTR_IN]], double** [[PTR_ADDR]], @@ -352,7 +370,7 @@ int fstatic(int n) { return a; } -// TCHECK: define void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], i{{[0-9]+}} [[A3_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]]) +// TCHECK: define weak void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], i{{[0-9]+}} [[A3_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]]) // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}}, // TCHECK: [[A3_ADDR:%.+]] = alloca i{{[0-9]+}}, // TCHECK: [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*, @@ -373,7 +391,7 @@ int fstatic(int n) { // firstprivate(b) // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8* // TCHECK: [[B_IN_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_ADDR_REF]] to i8* -// TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[B_PRIV_BCAST]], i8* [[B_IN_BCAST]],{{.+}}) +// TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_IN_BCAST]],{{.+}}) // TCHECK: ret void @@ -395,60 +413,71 @@ struct S1 { // on the host side, we first generate r1, then the static function and the template above // CHECK: define{{.+}} i32 {{.+}}([[S1]]* {{.+}}, i{{[0-9]+}} {{.+}}) - // CHECK: [[BASE_PTRS4:%.+]] = alloca [5 x i8*], - // CHECK: [[PTRS4:%.+]] = alloca [5 x i8*], - // CHECK: [[SIZET4:%.+]] = alloca [5 x i{{[0-9]+}}], + // CHECK: [[BASE_PTRS4:%.+]] = alloca [6 x i8*], + // CHECK: [[PTRS4:%.+]] = alloca [6 x i8*], + // CHECK: [[SIZET4:%.+]] = alloca [6 x i{{[0-9]+}}], // map(this: this ptr is implicitly captured (not firstprivate matter) - // CHECK: {{.+}} = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 - // CHECK: store {{.+}}, {{.+}}, - // CHECK: {{.+}} = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 - // CHECK: store {{.+}}, {{.+}}, - // CHECK: {{.+}} getelementptr inbounds [5 x i{{[0-9]+}}], [5 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 - // CHECK: store {{.+}}, {{.+}} + // CHECK: [[BP0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 + // CHECK: [[CBP0:%.+]] = bitcast i8** [[BP0]] to %struct.S1** + // CHECK: store %struct.S1* [[THIS:%.+]], %struct.S1** [[CBP0]], + // CHECK: [[P0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 + // CHECK: [[CP0:%.+]] = bitcast i8** [[P0]] to double** + // CHECK: store double* [[A:%.+]], double** [[CP0]], + // CHECK: [[SZ0:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 + // CHECK: store i{{[0-9]+}} %{{.+}}, i{{[0-9]+}}* [[SZ0]], + + // CHECK: [[BP1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 + // CHECK: [[CBP1:%.+]] = bitcast i8** [[BP1]] to %struct.S1** + // CHECK: store %struct.S1* [[THIS]], %struct.S1** [[CBP1]], + // CHECK: [[P1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 + // CHECK: [[CP1:%.+]] = bitcast i8** [[P1]] to double** + // CHECK: store double* [[A]], double** [[CP1]], + // CHECK: [[SZ1:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 + // CHECK: store i{{[0-9]+}} 8, i{{[0-9]+}}* [[SZ1]], // firstprivate(b): base_ptr = b, ptr = b, size = 4 (pass by-value) - // CHECK: [[BASE_PTRS_GEP4_1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 + // CHECK: [[BASE_PTRS_GEP4_1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_1]] to i{{[0-9]+}}* // CHECK: store i{{[0-9]+}} [[B_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]], - // CHECK: [[PTRS_GEP4_1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 + // CHECK: [[PTRS_GEP4_1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_1]] to i{{[0-9]+}}* // CHECK: store i{{[0-9]+}} [[B_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]], - // CHECK: [[SIZES_GEP4_1:%.+]] = getelementptr inbounds [5 x i{{[0-9]+}}], [5 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 + // CHECK: [[SIZES_GEP4_1:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 // CHECK: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_1]], // firstprivate(c), 3 entries: 2, n, c - // CHECK: [[BASE_PTRS_GEP4_2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 + // CHECK: [[BASE_PTRS_GEP4_2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_2]] to i{{[0-9]+}}* // CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[BCAST_TOPTR]], - // CHECK: [[PTRS_GEP4_2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 + // CHECK: [[PTRS_GEP4_2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_2]] to i{{[0-9]+}}* // CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[BCAST_TOPTR]], - // CHECK: [[SIZES_GEP4_2:%.+]] = getelementptr inbounds [5 x i{{[0-9]+}}], [5 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 + // CHECK: [[SIZES_GEP4_2:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 // CHECK-64: store i{{[0-9]+}} 8, i{{[0-9]+}}* [[SIZES_GEP4_2]], // CHECK-32: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_2]], - // CHECK: [[BASE_PTRS_GEP4_3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 + // CHECK: [[BASE_PTRS_GEP4_3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_3]] to i{{[0-9]+}}* // CHECK: store i{{[0-9]+}} [[N:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]], - // CHECK: [[PTRS_GEP4_3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 + // CHECK: [[PTRS_GEP4_3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_3]] to i{{[0-9]+}}* // CHECK: store i{{[0-9]+}} [[N]], i{{[0-9]+}}* [[BCAST_TOPTR]], - // CHECK: [[SIZES_GEP4_3:%.+]] = getelementptr inbounds [5 x i{{[0-9]+}}], [5 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 + // CHECK: [[SIZES_GEP4_3:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 // CHECK-64: store i{{[0-9]+}} 8, i{{[0-9]+}}* [[SIZES_GEP4_3]], // CHECK-32: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_3]], - // CHECK: [[BASE_PTRS_GEP4_4:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 + // CHECK: [[BASE_PTRS_GEP4_4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_4]] to i{{[0-9]+}}** // CHECK: store i{{[0-9]+}}* [[B:%.+]], i{{[0-9]+}}** [[BCAST_TOPTR]], - // CHECK: [[PTRS_GEP4_4:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 + // CHECK: [[PTRS_GEP4_4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 // CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_4]] to i{{[0-9]+}}** // CHECK: store i{{[0-9]+}}* [[B]], i{{[0-9]+}}** [[BCAST_TOPTR]], - // CHECK: [[SIZES_GEP4_4:%.+]] = getelementptr inbounds [5 x i{{[0-9]+}}], [5 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4 + // CHECK: [[SIZES_GEP4_4:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 5 // CHECK: store i{{[0-9]+}} [[B_SIZE:%.+]], i{{[0-9]+}}* [[SIZES_GEP4_4]], // only check that we use the map types stored in the global variable - // CHECK: call i32 @__tgt_target(i64 -1, {{.+}}, i32 5, i8** {{.+}}, i8** {{.+}}, i{{[0-9]+}}* {{.+}}, i64* getelementptr inbounds ([5 x i64], [5 x i64]* [[MAPT4]], i32 0, i32 0)) - - // TCHECK: define void @__omp_offloading_{{.+}}([[S1]]* [[TH:%.+]], i{{[0-9]+}} [[B_IN:%.+]], i{{[0-9]+}} [[VLA:%.+]], i{{[0-9]+}} [[VLA1:%.+]], i{{[0-9]+}}{{.+}} [[C_IN:%.+]]) + // CHECK: call i32 @__tgt_target(i64 -1, {{.+}}, i32 6, i8** {{.+}}, i8** {{.+}}, i{{[0-9]+}}* {{.+}}, i64* getelementptr inbounds ([6 x i64], [6 x i64]* [[MAPT4]], i32 0, i32 0)) + + // TCHECK: define weak void @__omp_offloading_{{.+}}([[S1]]* [[TH:%.+]], i{{[0-9]+}} [[B_IN:%.+]], i{{[0-9]+}} [[VLA:%.+]], i{{[0-9]+}} [[VLA1:%.+]], i{{[0-9]+}}{{.+}} [[C_IN:%.+]]) // TCHECK: [[TH_ADDR:%.+]] = alloca [[S1]]*, // TCHECK: [[B_ADDR:%.+]] = alloca i{{[0-9]+}}, // TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}}, @@ -481,7 +510,7 @@ struct S1 { // TCHECK: [[C_SZ_CPY:%.+]] = mul{{.+}} i{{[0-9]+}} [[C_SZ2]], 2 // TCHECK: [[C_PRIV_BCAST:%.+]] = bitcast i{{[0-9]+}}* [[C_PRIV]] to i8* // TCHECK: [[C_IN_BCAST:%.+]] = bitcast i{{[0-9]+}}* [[C_ADDR_REF]] to i8* - // TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[C_PRIV_BCAST]], i8* [[C_IN_BCAST]],{{.+}}) + // TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[C_PRIV_BCAST]], i8* align {{[0-9]+}} [[C_IN_BCAST]],{{.+}}) // finish // TCHECK: [[RELOAD_SSTACK:%.+]] = load i8*, i8** [[SSTACK]], @@ -559,8 +588,7 @@ int bar(int n, double *ptr){ // CHECK: call i32 @__tgt_target(i64 -1, {{.+}}, i32 2, i8** {{.+}}, i8** {{.+}}, i[[SZ]]* getelementptr inbounds ([2 x i[[SZ]]], [2 x i[[SZ]]]* [[SIZET6]], i32 0, i32 0), i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT6]], i32 0, i32 0)) - -// TCHECK: define void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]]) +// TCHECK: define weak void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]]) // TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}}, // TCHECK: [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*, // TCHECK-NOT: alloca i{{[0-9]+}}, @@ -576,7 +604,7 @@ int bar(int n, double *ptr){ // firstprivate(b) // TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8* // TCHECK: [[B_IN_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_ADDR_REF]] to i8* -// TCHECK: call void @llvm.memcpy.{{.+}}(i8* [[B_PRIV_BCAST]], i8* [[B_IN_BCAST]],{{.+}}) +// TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_IN_BCAST]],{{.+}}) // TCHECK: ret void diff --git a/test/OpenMP/target_firstprivate_messages.cpp b/test/OpenMP/target_firstprivate_messages.cpp index 6dbad671de7c6..4f82c96e564ce 100644 --- a/test/OpenMP/target_firstprivate_messages.cpp +++ b/test/OpenMP/target_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + struct S1; // expected-note 2 {{declared here}} expected-note 2 {{forward declaration of 'S1'}} extern S1 a; class S2 { @@ -36,7 +38,7 @@ public: S5(int v) : a(v) {} S5 &operator=(S5 &s) { #pragma omp target firstprivate(a) firstprivate(this->a) firstprivate(s.a) // expected-error {{expected variable name or data member of current class}} - for (int k = 0; k < s.a; ++k) + for (int k = 0; k < s.a; ++k) // expected-warning {{Non-trivial type 'S5' is mapped, only trivial types are guaranteed to be mapped correctly}} ++s.a; return *this; } diff --git a/test/OpenMP/target_if_messages.cpp b/test/OpenMP/target_if_messages.cpp index 14b6a5fbf938b..982fd4b0c028f 100644 --- a/test/OpenMP/target_if_messages.cpp +++ b/test/OpenMP/target_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 %s + void foo() { } diff --git a/test/OpenMP/target_is_device_ptr_ast_print.cpp b/test/OpenMP/target_is_device_ptr_ast_print.cpp index 8e56aa24d9bdd..4e8bf81435421 100644 --- a/test/OpenMP/target_is_device_ptr_ast_print.cpp +++ b/test/OpenMP/target_is_device_ptr_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -50,7 +54,7 @@ struct SA { // CHECK-NEXT: int aa[10]; // CHECK-NEXT: arr &raa = this->aa; // CHECK-NEXT: func( -// CHECK-NEXT: #pragma omp target is_device_ptr(this->k) +// CHECK-NEXT: #pragma omp target is_device_ptr(this->k){{$}} // CHECK-NEXT: { // CHECK-NEXT: } // CHECK-NEXT: #pragma omp target is_device_ptr(this->z) diff --git a/test/OpenMP/target_is_device_ptr_codegen.cpp b/test/OpenMP/target_is_device_ptr_codegen.cpp index 1491f41a787fd..90514acadf815 100644 --- a/test/OpenMP/target_is_device_ptr_codegen.cpp +++ b/test/OpenMP/target_is_device_ptr_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 double *g; @@ -35,7 +43,7 @@ double *g; // CK1: [[SIZES06:@.+]] = {{.+}}constant [2 x i[[sz]]] [i[[sz]] {{8|4}}, i[[sz]] {{8|4}}] // CK1: [[TYPES06:@.+]] = {{.+}}constant [2 x i64] [i64 288, i64 288] -// CK1-LABEL: @_Z3foo +// CK1-LABEL: @_Z3foo{{.*}}( template<typename T> void foo(float *&lr, T *&tr) { float *l; @@ -186,18 +194,32 @@ void bar(float *&a, int *&b) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 // CK2: [[ST:%.+]] = type { double*, double** } +// CK2-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l244.region_id = weak constant i8 0 + // CK2: [[SIZE00:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} {{8|4}}] -// CK2: [[MTYPE00:@.+]] = {{.+}}constant [1 x i64] [i64 33] +// CK2: [[MTYPE00:@.+]] = {{.+}}constant [1 x i64] [i64 288] + +// CK2-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l259.region_id = weak constant i8 0 -// CK2: [[SIZE01:@.+]] = {{.+}}constant [2 x i[[sz]]] [i[[sz]] {{8|4}}, i[[sz]] {{8|4}}] -// CK2: [[MTYPE01:@.+]] = {{.+}}constant [2 x i64] [i64 32, i64 17] +// CK2: [[SIZE01:@.+]] = {{.+}}constant [1 x i[[sz]]] [i[[sz]] {{8|4}}] +// CK2: [[MTYPE01:@.+]] = {{.+}}constant [1 x i64] [i64 288] -// CK2: [[SIZE02:@.+]] = {{.+}}constant [3 x i[[sz]]] [i[[sz]] {{8|4}}, i[[sz]] {{8|4}}, i[[sz]] {{8|4}}] -// CK2: [[MTYPE02:@.+]] = {{.+}}constant [3 x i64] [i64 33, i64 0, i64 17] +// CK2-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l274.region_id = weak constant i8 0 + +// CK2: [[SIZE02:@.+]] = {{.+}}constant [1 x i[[sz]]] [i[[sz]] {{8|4}}] +// CK2: [[MTYPE02:@.+]] = {{.+}}constant [1 x i64] [i64 288] template <typename T> struct ST { @@ -216,66 +238,39 @@ struct ST { // CK2-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK2-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 // CK2-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]** - // CK2-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double*** + // CK2-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[ST]]** // CK2-DAG: store [[ST]]* [[VAR0:%.+]], [[ST]]** [[CBP0]] - // CK2-DAG: store double** [[SEC0:%.+]], double*** [[CP0]] - // CK2-DAG: [[SEC0]] = getelementptr {{.*}}[[ST]]* [[VAR0]], i{{.+}} 0, i{{.+}} 0 + // CK2-DAG: store [[ST]]* [[VAR0]], [[ST]]** [[CP0]] #pragma omp target is_device_ptr(a) { a++; } - // CK2-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE01]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE01]]{{.+}}) + // CK2-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE01]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE01]]{{.+}}) // CK2-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK2-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] // CK2-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK2-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 // CK2-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]** - // CK2-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double**** + // CK2-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[ST]]** // CK2-DAG: store [[ST]]* [[VAR0:%.+]], [[ST]]** [[CBP0]] - // CK2-DAG: store double*** [[SEC0:%.+]], double**** [[CP0]] - // CK2-DAG: [[SEC0]] = getelementptr {{.*}}[[ST]]* [[VAR0]], i{{.+}} 0, i{{.+}} 1 - - // CK2-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 - // CK2-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 - // CK2-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double**** - // CK2-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double*** - // CK2-DAG: store double*** [[SEC0]], double**** [[CBP1]] - // CK2-DAG: store double** [[SEC1:%.+]], double*** [[CP1]] - // CK2-DAG: [[SEC1]] = load double**, double*** [[SEC0]] + // CK2-DAG: store [[ST]]* [[VAR0]], [[ST]]** [[CP0]] #pragma omp target is_device_ptr(b) { b++; } - // CK2-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 3, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[3 x i{{.+}}]* [[SIZE02]], {{.+}}getelementptr {{.+}}[3 x i{{.+}}]* [[MTYPE02]]{{.+}}) + // CK2-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE02]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE02]]{{.+}}) // CK2-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK2-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] - // CK2-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 - // CK2-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 + // CK2-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 + // CK2-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 // CK2-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]** - // CK2-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double**** + // CK2-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[ST]]** // CK2-DAG: store [[ST]]* [[VAR0:%.+]], [[ST]]** [[CBP0]] - // CK2-DAG: store double*** [[SEC0:%.+]], double**** [[CP0]] - // CK2-DAG: [[SEC0]] = getelementptr {{.*}}[[ST]]* [[VAR0]], i{{.+}} 0, i{{.+}} 1 - - // CK2-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 2 - // CK2-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 2 - // CK2-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to double**** - // CK2-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double*** - // CK2-DAG: store double*** [[SEC0]], double**** [[CBP1]] - // CK2-DAG: store double** [[SEC1:%.+]], double*** [[CP1]] - // CK2-DAG: [[SEC1]] = load double**, double*** [[SEC0]] - - // CK2-DAG: [[BP2:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 - // CK2-DAG: [[P2:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 - // CK2-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to [[ST]]** - // CK2-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to double*** - // CK2-DAG: store [[ST]]* [[VAR2:%.+]], [[ST]]** [[CBP2]] - // CK2-DAG: store double** [[SEC2:%.+]], double*** [[CP2]] - // CK2-DAG: [[SEC2]] = getelementptr {{.*}}[[ST]]* [[VAR2]], i{{.+}} 0, i{{.+}} 0 + // CK2-DAG: store [[ST]]* [[VAR0]], [[ST]]** [[CP0]] #pragma omp target is_device_ptr(a, b) { a++; diff --git a/test/OpenMP/target_is_device_ptr_messages.cpp b/test/OpenMP/target_is_device_ptr_messages.cpp index 8cd1426081abe..6f218e20f98cb 100644 --- a/test/OpenMP/target_is_device_ptr_messages.cpp +++ b/test/OpenMP/target_is_device_ptr_messages.cpp @@ -1,4 +1,6 @@ // RUN: %clang_cc1 -std=c++11 -verify -fopenmp -ferror-limit 200 %s + +// RUN: %clang_cc1 -std=c++11 -verify -fopenmp-simd -ferror-limit 200 %s struct ST { int *a; }; diff --git a/test/OpenMP/target_map_codegen.cpp b/test/OpenMP/target_map_codegen.cpp index 34b2395044624..ec99e8ed396dc 100644 --- a/test/OpenMP/target_map_codegen.cpp +++ b/test/OpenMP/target_map_codegen.cpp @@ -7,12 +7,20 @@ /// ///==========================================================================/// -// RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK1 --check-prefix CK1-64 // RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 -// RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK1 --check-prefix CK1-64 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 class B { @@ -31,11 +39,13 @@ public: }; double B::VAR = 1.0; +// CK1-LABEL: @.__omp_offloading_{{.*}}implicit_maps_integer{{.*}}_l68.region_id = weak constant i8 0 + // CK1-DAG: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} 4] -// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_IS_FIRST = 288 -// CK1-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 288] +// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 800 +// CK1-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 800] -// CK1-LABEL: implicit_maps_integer +// CK1-LABEL: implicit_maps_integer{{.*}}( void implicit_maps_integer (int a){ // CK1: call void{{.*}}modify B::modify(a); @@ -69,23 +79,36 @@ void implicit_maps_integer (int a){ // CK1-32: {{.+}} = load i32, i32* [[ADDR]], #endif + + ///==========================================================================/// -// RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64 +// RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK2 --check-prefix CK2-64 // RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64 -// RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK2 --check-prefix CK2-64 +// RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 +// CK2-LABEL: @.__omp_offloading_{{.*}}implicit_maps_reference{{.*}}_l128.region_id = weak constant i8 0 + // CK2: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} 4] -// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_IS_FIRST = 288 -// CK2: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 288] +// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 800 +// CK2: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 800] +// CK2-LABEL: @.__omp_offloading_{{.*}}implicit_maps_reference{{.*}}_l147.region_id = weak constant i8 0 // CK2: [[SIZES2:@.+]] = {{.+}}constant [1 x i[[sz]]] zeroinitializer -// Map types: OMP_MAP_IS_PTR = 32 -// CK2: [[TYPES2:@.+]] = {{.+}}constant [1 x i64] [i64 32] +// Map types: OMP_MAP_IS_PTR | OMP_MAP_IMPLICIT = 544 +// CK2: [[TYPES2:@.+]] = {{.+}}constant [1 x i64] [i64 544] -// CK2-LABEL: implicit_maps_reference +// CK2-LABEL: implicit_maps_reference{{.*}}( void implicit_maps_reference (int a, int *b){ int &i = a; // CK2-DAG: call i32 @__tgt_target(i64 {{.+}}, i8* {{.+}}, i32 1, i8** [[BPGEP:%[0-9]+]], i8** [[PGEP:%[0-9]+]], {{.+}}[[SIZES]]{{.+}}, {{.+}}[[TYPES]]{{.+}}) @@ -149,19 +172,29 @@ void implicit_maps_reference (int a, int *b){ // CK2: getelementptr inbounds i32, i32* [[TT]], i32 1 #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK3 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-64 +// RUN: %clang_cc1 -DCK3 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK3 --check-prefix CK3-64 // RUN: %clang_cc1 -DCK3 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-64 -// RUN: %clang_cc1 -DCK3 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK3 --check-prefix CK3-64 +// RUN: %clang_cc1 -DCK3 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK3 --check-prefix CK3-32 // RUN: %clang_cc1 -DCK3 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK3 --check-prefix CK3-32 + +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} #ifdef CK3 +// CK3-LABEL: @.__omp_offloading_{{.*}}implicit_maps_parameter{{.*}}_l214.region_id = weak constant i8 0 + // CK3-DAG: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} 4] -// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_IS_FIRST = 288 -// CK3-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 288] +// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 800 +// CK3-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 800] -// CK3-LABEL: implicit_maps_parameter +// CK3-LABEL: implicit_maps_parameter{{.*}}( void implicit_maps_parameter (int a){ // CK3-DAG: call i32 @__tgt_target(i64 {{.+}}, i8* {{.+}}, i32 1, i8** [[BPGEP:%[0-9]+]], i8** [[PGEP:%[0-9]+]], {{.+}}[[SIZES]]{{.+}}, {{.+}}[[TYPES]]{{.+}}) @@ -193,19 +226,29 @@ void implicit_maps_parameter (int a){ #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK4 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-64 +// RUN: %clang_cc1 -DCK4 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK4 --check-prefix CK4-64 // RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-64 -// RUN: %clang_cc1 -DCK4 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK4 --check-prefix CK4-64 +// RUN: %clang_cc1 -DCK4 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK4 --check-prefix CK4-32 // RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK4 --check-prefix CK4-32 + +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY3 %s +// SIMD-ONLY3-NOT: {{__kmpc|__tgt}} #ifdef CK4 +// CK4-LABEL: @.__omp_offloading_{{.*}}implicit_maps_nested_integer{{.*}}_l276.region_id = weak constant i8 0 + // CK4-DAG: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} 4] -// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_IS_FIRST = 288 -// CK4-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 288] +// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 800 +// CK4-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 800] -// CK4-LABEL: implicit_maps_nested_integer +// CK4-LABEL: implicit_maps_nested_integer{{.*}}( void implicit_maps_nested_integer (int a){ int i = a; @@ -249,19 +292,29 @@ void implicit_maps_nested_integer (int a){ // CK4: define internal void [[KERNELP2]](i32* {{[^,]+}}, i32* {{[^,]+}}, i32* {{[^,]+}}) #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK5 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK5 --check-prefix CK5-64 +// RUN: %clang_cc1 -DCK5 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK5 --check-prefix CK5-64 // RUN: %clang_cc1 -DCK5 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK5 --check-prefix CK5-64 -// RUN: %clang_cc1 -DCK5 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK5 --check-prefix CK5-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK5 --check-prefix CK5-64 +// RUN: %clang_cc1 -DCK5 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK5 --check-prefix CK5-32 // RUN: %clang_cc1 -DCK5 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK5 --check-prefix CK5-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK5 --check-prefix CK5-32 + +// RUN: %clang_cc1 -DCK5 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY4 %s +// RUN: %clang_cc1 -DCK5 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY4 %s +// RUN: %clang_cc1 -DCK5 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY4 %s +// RUN: %clang_cc1 -DCK5 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY4 %s +// SIMD-ONLY4-NOT: {{__kmpc|__tgt}} #ifdef CK5 +// CK5-LABEL: @.__omp_offloading_{{.*}}implicit_maps_nested_integer_and_enum{{.*}}_l340.region_id = weak constant i8 0 + // CK5-DAG: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} 4] -// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_IS_FIRST = 288 -// CK5-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 288] +// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 800 +// CK5-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 800] -// CK5-LABEL: implicit_maps_nested_integer_and_enum +// CK5-LABEL: implicit_maps_nested_integer_and_enum{{.*}}( void implicit_maps_nested_integer_and_enum (int a){ enum Bla { SomeEnum = 0x09 @@ -300,19 +353,29 @@ void implicit_maps_nested_integer_and_enum (int a){ #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK6 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK6 --check-prefix CK6-64 +// RUN: %clang_cc1 -DCK6 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK6 --check-prefix CK6-64 // RUN: %clang_cc1 -DCK6 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK6 --check-prefix CK6-64 -// RUN: %clang_cc1 -DCK6 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK6 --check-prefix CK6-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK6 --check-prefix CK6-64 +// RUN: %clang_cc1 -DCK6 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK6 --check-prefix CK6-32 // RUN: %clang_cc1 -DCK6 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK6 --check-prefix CK6-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK6 --check-prefix CK6-32 + +// RUN: %clang_cc1 -DCK6 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY5 %s +// RUN: %clang_cc1 -DCK6 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY5 %s +// RUN: %clang_cc1 -DCK6 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY5 %s +// RUN: %clang_cc1 -DCK6 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY5 %s +// SIMD-ONLY5-NOT: {{__kmpc|__tgt}} #ifdef CK6 +// CK6-LABEL: @.__omp_offloading_{{.*}}implicit_maps_host_global{{.*}}_l397.region_id = weak constant i8 0 + // CK6-DAG: [[GBL:@Gi]] = global i32 0 // CK6-DAG: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} 4] -// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_IS_FIRST = 288 -// CK6-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 288] +// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 800 +// CK6-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 800] -// CK6-LABEL: implicit_maps_host_global +// CK6-LABEL: implicit_maps_host_global{{.*}}( int Gi; void implicit_maps_host_global (int a){ // CK6-DAG: call i32 @__tgt_target(i64 {{.+}}, i8* {{.+}}, i32 1, i8** [[BPGEP:%[0-9]+]], i8** [[PGEP:%[0-9]+]], {{.+}}[[SIZES]]{{.+}}, {{.+}}[[TYPES]]{{.+}}) @@ -346,24 +409,34 @@ void implicit_maps_host_global (int a){ #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK7 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK7 --check-prefix CK7-64 +// RUN: %clang_cc1 -DCK7 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK7 --check-prefix CK7-64 // RUN: %clang_cc1 -DCK7 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK7 --check-prefix CK7-64 -// RUN: %clang_cc1 -DCK7 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK7 --check-prefix CK7-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK7 --check-prefix CK7-64 +// RUN: %clang_cc1 -DCK7 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK7 --check-prefix CK7-32 // RUN: %clang_cc1 -DCK7 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK7 --check-prefix CK7-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK7 --check-prefix CK7-32 + +// RUN: %clang_cc1 -DCK7 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY6 %s +// RUN: %clang_cc1 -DCK7 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY6 %s +// RUN: %clang_cc1 -DCK7 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY6 %s +// RUN: %clang_cc1 -DCK7 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY6 %s +// SIMD-ONLY6-NOT: {{__kmpc|__tgt}} #ifdef CK7 // For a 32-bit targets, the value doesn't fit the size of the pointer, // therefore it is passed by reference with a map 'to' specification. +// CK7-LABEL: @.__omp_offloading_{{.*}}implicit_maps_double{{.*}}_l464.region_id = weak constant i8 0 + // CK7-DAG: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} 8] -// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_IS_FIRST = 288 -// CK7-64-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 288] -// Map types: OMP_MAP_TO | OMP_MAP_PRIVATE_PTR | OMP_MAP_FIRST_REF = 161 -// CK7-32-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 161] +// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 800 +// CK7-64-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 800] +// Map types: OMP_MAP_TO | OMP_MAP_PRIVATE | OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 673 +// CK7-32-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 673] -// CK7-LABEL: implicit_maps_double +// CK7-LABEL: implicit_maps_double{{.*}}( void implicit_maps_double (int a){ double d = (double)a; @@ -408,19 +481,29 @@ void implicit_maps_double (int a){ #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK8 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK8 +// RUN: %clang_cc1 -DCK8 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK8 // RUN: %clang_cc1 -DCK8 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK8 -// RUN: %clang_cc1 -DCK8 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK8 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK8 +// RUN: %clang_cc1 -DCK8 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK8 // RUN: %clang_cc1 -DCK8 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK8 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK8 + +// RUN: %clang_cc1 -DCK8 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY7 %s +// RUN: %clang_cc1 -DCK8 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY7 %s +// RUN: %clang_cc1 -DCK8 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY7 %s +// RUN: %clang_cc1 -DCK8 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY7 %s +// SIMD-ONLY7-NOT: {{__kmpc|__tgt}} #ifdef CK8 +// CK8-LABEL: @.__omp_offloading_{{.*}}implicit_maps_float{{.*}}_l524.region_id = weak constant i8 0 + // CK8-DAG: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} 4] -// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_IS_FIRST = 288 -// CK8-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 288] +// Map types: OMP_MAP_PRIVATE_VAL | OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 800 +// CK8-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 800] -// CK8-LABEL: implicit_maps_float +// CK8-LABEL: implicit_maps_float{{.*}}( void implicit_maps_float (int a){ float f = (float)a; @@ -452,19 +535,29 @@ void implicit_maps_float (int a){ #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK9 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK9 +// RUN: %clang_cc1 -DCK9 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK9 // RUN: %clang_cc1 -DCK9 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK9 -// RUN: %clang_cc1 -DCK9 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK9 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK9 +// RUN: %clang_cc1 -DCK9 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK9 // RUN: %clang_cc1 -DCK9 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK9 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK9 + +// RUN: %clang_cc1 -DCK9 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY8 %s +// RUN: %clang_cc1 -DCK9 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY8 %s +// RUN: %clang_cc1 -DCK9 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY8 %s +// RUN: %clang_cc1 -DCK9 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY8 %s +// SIMD-ONLY8-NOT: {{__kmpc|__tgt}} #ifdef CK9 +// CK9-LABEL: @.__omp_offloading_{{.*}}implicit_maps_array{{.*}}_l575.region_id = weak constant i8 0 + // CK9-DAG: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} 16] -// Map types: OMP_MAP_TO + OMP_MAP_FROM + OMP_MAP_IS_FIRST + OMP_MAP_IMPLICIT = 547 +// Map types: OMP_MAP_TO + OMP_MAP_FROM + OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 547 // CK9-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 547] -// CK9-LABEL: implicit_maps_array +// CK9-LABEL: implicit_maps_array{{.*}}( void implicit_maps_array (int a){ double darr[2] = {(double)a, (double)a}; @@ -493,19 +586,29 @@ void implicit_maps_array (int a){ // CK9: {{.+}} = getelementptr inbounds [2 x double], [2 x double]* [[REF]], i[[sz]] 0, i[[sz]] 0 #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK10 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK10 +// RUN: %clang_cc1 -DCK10 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK10 // RUN: %clang_cc1 -DCK10 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK10 -// RUN: %clang_cc1 -DCK10 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK10 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK10 +// RUN: %clang_cc1 -DCK10 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK10 // RUN: %clang_cc1 -DCK10 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK10 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK10 + +// RUN: %clang_cc1 -DCK10 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY9 %s +// RUN: %clang_cc1 -DCK10 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY9 %s +// RUN: %clang_cc1 -DCK10 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY9 %s +// RUN: %clang_cc1 -DCK10 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY9 %s +// SIMD-ONLY9-NOT: {{__kmpc|__tgt}} #ifdef CK10 +// CK10-LABEL: @.__omp_offloading_{{.*}}implicit_maps_pointer{{.*}}_l626.region_id = weak constant i8 0 + // CK10-DAG: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] zeroinitializer -// Map types: OMP_MAP_IS_FIRST = 32 -// CK10-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 32] +// Map types: OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 544 +// CK10-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 544] -// CK10-LABEL: implicit_maps_pointer +// CK10-LABEL: implicit_maps_pointer{{.*}}( void implicit_maps_pointer (){ double *ddyn; @@ -535,19 +638,29 @@ void implicit_maps_pointer (){ #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK11 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK11 +// RUN: %clang_cc1 -DCK11 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK11 // RUN: %clang_cc1 -DCK11 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK11 -// RUN: %clang_cc1 -DCK11 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK11 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK11 +// RUN: %clang_cc1 -DCK11 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK11 // RUN: %clang_cc1 -DCK11 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK11 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK11 + +// RUN: %clang_cc1 -DCK11 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY10 %s +// RUN: %clang_cc1 -DCK11 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY10 %s +// RUN: %clang_cc1 -DCK11 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY10 %s +// RUN: %clang_cc1 -DCK11 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY10 %s +// SIMD-ONLY10-NOT: {{__kmpc|__tgt}} #ifdef CK11 +// CK11-LABEL: @.__omp_offloading_{{.*}}implicit_maps_double_complex{{.*}}_l678.region_id = weak constant i8 0 + // CK11-DAG: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} 16] -// Map types: OMP_MAP_TO + OMP_MAP_FROM + OMP_MAP_IS_FIRST + OMP_MAP_IMPLICIT = 547 +// Map types: OMP_MAP_TO | OMP_MAP_FROM | OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 547 // CK11-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 547] -// CK11-LABEL: implicit_maps_double_complex +// CK11-LABEL: implicit_maps_double_complex{{.*}}( void implicit_maps_double_complex (int a){ double _Complex dc = (double)a; @@ -575,24 +688,34 @@ void implicit_maps_double_complex (int a){ // CK11: {{.+}} = getelementptr inbounds { double, double }, { double, double }* [[REF]], i32 0, i32 0 #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK12 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK12 --check-prefix CK12-64 +// RUN: %clang_cc1 -DCK12 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK12 --check-prefix CK12-64 // RUN: %clang_cc1 -DCK12 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK12 --check-prefix CK12-64 -// RUN: %clang_cc1 -DCK12 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK12 --check-prefix CK12-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK12 --check-prefix CK12-64 +// RUN: %clang_cc1 -DCK12 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK12 --check-prefix CK12-32 // RUN: %clang_cc1 -DCK12 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK12 --check-prefix CK12-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK12 --check-prefix CK12-32 + +// RUN: %clang_cc1 -DCK12 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY11 %s +// RUN: %clang_cc1 -DCK12 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY11 %s +// RUN: %clang_cc1 -DCK12 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY11 %s +// RUN: %clang_cc1 -DCK12 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY11 %s +// SIMD-ONLY11-NOT: {{__kmpc|__tgt}} #ifdef CK12 // For a 32-bit targets, the value doesn't fit the size of the pointer, // therefore it is passed by reference with a map 'to' specification. +// CK12-LABEL: @.__omp_offloading_{{.*}}implicit_maps_float_complex{{.*}}_l743.region_id = weak constant i8 0 + // CK12-DAG: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} 8] -// Map types: OMP_MAP_PRIVATE_VAL + OMP_MAP_IS_FIRST = 288 -// CK12-64-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 288] -// Map types: OMP_MAP_TO | OMP_MAP_PRIVATE_PTR | OMP_MAP_FIRST_REF = 161 -// CK12-32-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 161] +// Map types: OMP_MAP_PRIVATE_VAL + OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 800 +// CK12-64-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 800] +// Map types: OMP_MAP_TO | OMP_MAP_PRIVATE | OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 673 +// CK12-32-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 673] -// CK12-LABEL: implicit_maps_float_complex +// CK12-LABEL: implicit_maps_float_complex{{.*}}( void implicit_maps_float_complex (int a){ float _Complex fc = (float)a; @@ -636,22 +759,32 @@ void implicit_maps_float_complex (int a){ // CK12-32: {{.+}} = getelementptr inbounds { float, float }, { float, float }* [[REF]], i32 0, i32 0 #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK13 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK13 +// RUN: %clang_cc1 -DCK13 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK13 // RUN: %clang_cc1 -DCK13 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK13 -// RUN: %clang_cc1 -DCK13 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK13 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK13 +// RUN: %clang_cc1 -DCK13 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK13 // RUN: %clang_cc1 -DCK13 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK13 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK13 + +// RUN: %clang_cc1 -DCK13 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY12 %s +// RUN: %clang_cc1 -DCK13 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY12 %s +// RUN: %clang_cc1 -DCK13 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY12 %s +// RUN: %clang_cc1 -DCK13 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY12 %s +// SIMD-ONLY12-NOT: {{__kmpc|__tgt}} #ifdef CK13 +// CK13-LABEL: @.__omp_offloading_{{.*}}implicit_maps_variable_length_array{{.*}}_l825.region_id = weak constant i8 0 + // We don't have a constant map size for VLAs. // Map types: -// - OMP_MAP_PRIVATE_VAL + OMP_MAP_IS_FIRST = 288 (vla size) -// - OMP_MAP_PRIVATE_VAL + OMP_MAP_IS_FIRST = 288 (vla size) -// - OMP_MAP_TO + OMP_MAP_FROM + OMP_MAP_IS_FIRST + OMP_IMPICIT_MAP = 547 +// - OMP_MAP_PRIVATE_VAL + OMP_MAP_TARGET_PARAM = 288 (vla size) +// - OMP_MAP_PRIVATE_VAL + OMP_MAP_TARGET_PARAM = 288 (vla size) +// - OMP_MAP_TO + OMP_MAP_FROM + OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 547 // CK13-DAG: [[TYPES:@.+]] = {{.+}}constant [3 x i64] [i64 288, i64 288, i64 547] -// CK13-LABEL: implicit_maps_variable_length_array +// CK13-LABEL: implicit_maps_variable_length_array{{.*}}( void implicit_maps_variable_length_array (int a){ double vla[2][a]; @@ -708,21 +841,32 @@ void implicit_maps_variable_length_array (int a){ // CK13: {{.+}} = getelementptr inbounds double, double* [[REF]], i[[sz]] %{{.+}} #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK14 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK14 --check-prefix CK14-64 +// RUN: %clang_cc1 -DCK14 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK14 --check-prefix CK14-64 // RUN: %clang_cc1 -DCK14 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK14 --check-prefix CK14-64 -// RUN: %clang_cc1 -DCK14 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK14 --check-prefix CK14-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK14 --check-prefix CK14-64 +// RUN: %clang_cc1 -DCK14 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK14 --check-prefix CK14-32 // RUN: %clang_cc1 -DCK14 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK14 --check-prefix CK14-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK14 --check-prefix CK14-32 + +// RUN: %clang_cc1 -DCK14 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY13 %s +// RUN: %clang_cc1 -DCK14 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY13 %s +// RUN: %clang_cc1 -DCK14 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY13 %s +// RUN: %clang_cc1 -DCK14 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY13 %s +// SIMD-ONLY13-NOT: {{__kmpc|__tgt}} #ifdef CK14 // CK14-DAG: [[ST:%.+]] = type { i32, double } -// CK14-DAG: [[SIZES:@.+]] = {{.+}}constant [3 x i[[sz:64|32]]] [i{{64|32}} 4, i{{64|32}} 8, i{{64|32}} 4] + +// CK14-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l877.region_id = weak constant i8 0 + + // Map types: -// - OMP_MAP_TO + OMP_MAP_FROM + OMP_MAP_IS_FIRST + OMP_IMPLICIT_MAP = 547 -// - OMP_MAP_TO + OMP_MAP_FROM + OMP_IMPLICIT_MAP = 515 -// - OMP_MAP_PRIVATE_VAL + OMP_MAP_IS_FIRST = 288 -// CK14-DAG: [[TYPES:@.+]] = {{.+}}constant [3 x i64] [i64 547, i64 515, i64 288] +// - OMP_MAP_TARGET_PARAM = 32 +// - OMP_MAP_TO + OMP_MAP_FROM | OMP_MAP_IMPLICIT | OMP_MAP_MEMBER_OF = 281474976711171 +// - OMP_MAP_PRIVATE_VAL + OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 800 +// CK14-DAG: [[TYPES:@.+]] = {{.+}}constant [4 x i64] [i64 32, i64 281474976711171, i64 281474976711171, i64 800] class SSS { public: @@ -740,35 +884,51 @@ public: SSS(int a, double b) : a(a), b(b) {} }; -// CK14-LABEL: implicit_maps_class +// CK14-LABEL: implicit_maps_class{{.*}}( void implicit_maps_class (int a){ SSS sss(a, (double)a); // CK14: define {{.*}}void @{{.+}}foo{{.+}}([[ST]]* {{[^,]+}}, i32 {{[^,]+}}) - // CK14-DAG: call i32 @__tgt_target(i64 {{.+}}, i8* {{.+}}, i32 3, i8** [[BPGEP:%[0-9]+]], i8** [[PGEP:%[0-9]+]], {{.+}}[[SIZES]]{{.+}}, {{.+}}[[TYPES]]{{.+}}) + // CK14-DAG: call i32 @__tgt_target(i64 {{.+}}, i8* {{.+}}, i32 4, i8** [[BPGEP:%[0-9]+]], i8** [[PGEP:%[0-9]+]], i[[sz:64|32]]* [[SIZES:%[^,]+]], {{.+}}[[TYPES]]{{.+}}) // CK14-DAG: [[BPGEP]] = getelementptr inbounds {{.+}}[[BPS:%[^,]+]], i32 0, i32 0 // CK14-DAG: [[PGEP]] = getelementptr inbounds {{.+}}[[PS:%[^,]+]], i32 0, i32 0 + // CK14-DAG: [[SIZES]] = getelementptr inbounds {{.+}}[[S:%[^,]+]], i32 0, i32 0 // CK14-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 0 // CK14-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 0 + // CK14-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i32 0, i32 0 // CK14-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]** // CK14-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32** // CK14-DAG: store [[ST]]* [[DECL:%.+]], [[ST]]** [[CBP0]] - // CK14-DAG: store i32* %{{.+}}, i32** [[CP0]] + // CK14-DAG: store i32* [[A:%.+]], i32** [[CP0]] + // CK14-DAG: store i[[sz]] %{{.+}}, i[[sz]]* [[S0]] // CK14-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 1 // CK14-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 1 + // CK14-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i32 0, i32 1 // CK14-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[ST]]** - // CK14-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double** + // CK14-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** // CK14-DAG: store [[ST]]* [[DECL]], [[ST]]** [[CBP1]] - // CK14-DAG: store double* %{{.+}}, double** [[CP1]] - - // CK14-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 2 - // CK14-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 2 - // CK14-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to i[[sz]]* - // CK14-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i[[sz]]* - // CK14-DAG: store i[[sz]] [[VAL:%.+]], i[[sz]]* [[CBP1]] - // CK14-DAG: store i[[sz]] [[VAL]], i[[sz]]* [[CP1]] + // CK14-DAG: store i32* [[A]], i32** [[CP1]] + // CK14-DAG: store i[[sz]] 4, i[[sz]]* [[S1]] + + // CK14-DAG: [[BP2:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 2 + // CK14-DAG: [[P2:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 2 + // CK14-DAG: [[S2:%.+]] = getelementptr inbounds {{.+}}[[S]], i32 0, i32 2 + // CK14-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to [[ST]]** + // CK14-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to double** + // CK14-DAG: store [[ST]]* [[DECL]], [[ST]]** [[CBP2]] + // CK14-DAG: store double* %{{.+}}, double** [[CP2]] + // CK14-DAG: store i[[sz]] 8, i[[sz]]* [[S2]] + + // CK14-DAG: [[BP3:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 3 + // CK14-DAG: [[P3:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 3 + // CK14-DAG: [[S3:%.+]] = getelementptr inbounds {{.+}}[[S]], i32 0, i32 3 + // CK14-DAG: [[CBP3:%.+]] = bitcast i8** [[BP3]] to i[[sz]]* + // CK14-DAG: [[CP3:%.+]] = bitcast i8** [[P3]] to i[[sz]]* + // CK14-DAG: store i[[sz]] [[VAL:%.+]], i[[sz]]* [[CBP3]] + // CK14-DAG: store i[[sz]] [[VAL]], i[[sz]]* [[CP3]] + // CK14-DAG: store i[[sz]] 4, i[[sz]]* [[S3]] // CK14-DAG: [[VAL]] = load i[[sz]], i[[sz]]* [[ADDR:%.+]], // CK14-64-DAG: [[CADDR:%.+]] = bitcast i[[sz]]* [[ADDR]] to i32* // CK14-64-DAG: store i32 {{.+}}, i32* [[CADDR]], @@ -790,40 +950,46 @@ void implicit_maps_class (int a){ #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK15 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK15 --check-prefix CK15-64 +// RUN: %clang_cc1 -DCK15 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK15 --check-prefix CK15-64 // RUN: %clang_cc1 -DCK15 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK15 --check-prefix CK15-64 -// RUN: %clang_cc1 -DCK15 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK15 --check-prefix CK15-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK15 --check-prefix CK15-64 +// RUN: %clang_cc1 -DCK15 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK15 --check-prefix CK15-32 // RUN: %clang_cc1 -DCK15 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK15 --check-prefix CK15-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK15 --check-prefix CK15-32 + +// RUN: %clang_cc1 -DCK15 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY14 %s +// RUN: %clang_cc1 -DCK15 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY14 %s +// RUN: %clang_cc1 -DCK15 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY14 %s +// RUN: %clang_cc1 -DCK15 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY14 %s +// SIMD-ONLY14-NOT: {{__kmpc|__tgt}} #ifdef CK15 -// CK15: [[ST:%.+]] = type { i32, double, i32* } -// CK15: [[SIZES:@.+]] = {{.+}}constant [5 x i[[sz:64|32]]] [i{{64|32}} 4, i{{64|32}} 8, i{{64|32}} {{8|4}}, i{{64|32}} 4, i{{64|32}} 4] +// CK15: [[ST:%.+]] = type { i32, double } // Map types: -// - OMP_MAP_TO + OMP_MAP_FROM + OMP_MAP_IS_FIRST + OMP_MAP_IMPLICIT = 547 -// - OMP_MAP_PRIVATE_VAL + OMP_MAP_IS_FIRST = 288 -// CK15: [[TYPES:@.+]] = {{.+}}constant [5 x i64] [i64 547, i64 515, i64 512, i64 531, i64 288] +// - OMP_MAP_TARGET_PARAM = 32 +// - OMP_MAP_TO + OMP_MAP_FROM | OMP_MAP_IMPLICIT | OMP_MAP_MEMBER_OF = 281474976711171 +// - OMP_MAP_PRIVATE_VAL + OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 800 +// CK15: [[TYPES:@.+]] = {{.+}}constant [4 x i64] [i64 32, i64 281474976711171, i64 281474976711171, i64 800] -// CK15: [[SIZES2:@.+]] = {{.+}}constant [5 x i[[sz]]] [i{{64|32}} 4, i{{64|32}} 8, i{{64|32}} {{8|4}}, i{{64|32}} 4, i{{64|32}} 4] // Map types: -// - OMP_MAP_TO + OMP_MAP_FROM + OMP_MAP_IS_FIRST = 35 -// - OMP_MAP_PRIVATE_VAL + OMP_MAP_IS_FIRST = 288 -// CK15: [[TYPES2:@.+]] = {{.+}}constant [5 x i64] [i64 547, i64 515, i64 512, i64 531, i64 288] +// - OMP_MAP_TARGET_PARAM = 32 +// - OMP_MAP_TO + OMP_MAP_FROM | OMP_MAP_IMPLICIT | OMP_MAP_MEMBER_OF = 281474976711171 +// - OMP_MAP_PRIVATE_VAL + OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 800 +// CK15: [[TYPES2:@.+]] = {{.+}}constant [4 x i64] [i64 32, i64 281474976711171, i64 281474976711171, i64 800] template<int x> class SSST { public: int a; double b; - int &r; void foo(int c) { #pragma omp target { a += c + x; b += (double)(c + x); - r += x; } } template<int y> @@ -832,56 +998,57 @@ public: { a += c + x + y; b += (double)(c + x + y); - r += x + y; } } - SSST(int a, double b, int &r) : a(a), b(b), r(r) {} + SSST(int a, double b) : a(a), b(b) {} }; -// CK15-LABEL: implicit_maps_templated_class +// CK15-LABEL: implicit_maps_templated_class{{.*}}( void implicit_maps_templated_class (int a){ - SSST<123> ssst(a, (double)a, a); + SSST<123> ssst(a, (double)a); // CK15: define {{.*}}void @{{.+}}foo{{.+}}([[ST]]* {{[^,]+}}, i32 {{[^,]+}}) - // CK15-DAG: call i32 @__tgt_target(i64 {{.+}}, i8* {{.+}}, i32 5, i8** [[BPGEP:%[0-9]+]], i8** [[PGEP:%[0-9]+]], {{.+}}[[SIZES]]{{.+}}, {{.+}}[[TYPES]]{{.+}}) + // CK15-DAG: call i32 @__tgt_target(i64 {{.+}}, i8* {{.+}}, i32 4, i8** [[BPGEP:%[0-9]+]], i8** [[PGEP:%[0-9]+]], i[[sz:64|32]]* [[SIZES:%[^,]+]], {{.+}}[[TYPES]]{{.+}}) // CK15-DAG: [[BPGEP]] = getelementptr inbounds {{.+}}[[BPS:%[^,]+]], i32 0, i32 0 // CK15-DAG: [[PGEP]] = getelementptr inbounds {{.+}}[[PS:%[^,]+]], i32 0, i32 0 + // CK15-DAG: [[SIZES]] = getelementptr inbounds {{.+}}[[S:%[^,]+]], i32 0, i32 0 // CK15-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 0 // CK15-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 0 + // CK15-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i32 0, i32 0 // CK15-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]** // CK15-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32** // CK15-DAG: store [[ST]]* [[DECL:%.+]], [[ST]]** [[CBP0]] - // CK15-DAG: store i32* %{{.+}}, i32** [[CP0]] + // CK15-DAG: store i32* [[A:%.+]], i32** [[CP0]] + // CK15-DAG: store i[[sz]] %{{.+}}, i[[sz]]* [[S0]] // CK15-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 1 // CK15-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 1 + // CK15-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i32 0, i32 1 // CK15-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[ST]]** - // CK15-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double** + // CK15-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** // CK15-DAG: store [[ST]]* [[DECL]], [[ST]]** [[CBP1]] - // CK15-DAG: store double* %{{.+}}, double** [[CP1]] + // CK15-DAG: store i32* [[A]], i32** [[CP1]] + // CK15-DAG: store i[[sz]] 4, i[[sz]]* [[S1]] // CK15-DAG: [[BP2:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 2 // CK15-DAG: [[P2:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 2 + // CK15-DAG: [[S2:%.+]] = getelementptr inbounds {{.+}}[[S]], i32 0, i32 2 // CK15-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to [[ST]]** - // CK15-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to i32*** + // CK15-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to double** // CK15-DAG: store [[ST]]* [[DECL]], [[ST]]** [[CBP2]] - // CK15-DAG: store i32** %{{.+}}, i32*** [[CP2]] + // CK15-DAG: store double* %{{.+}}, double** [[CP2]] + // CK15-DAG: store i[[sz]] 8, i[[sz]]* [[S2]] // CK15-DAG: [[BP3:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 3 // CK15-DAG: [[P3:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 3 - // CK15-DAG: [[CBP3:%.+]] = bitcast i8** [[BP3]] to i32*** - // CK15-DAG: [[CP3:%.+]] = bitcast i8** [[P3]] to i32** - // CK15-DAG: store i32** %{{.+}}, i32*** [[CBP3]] - // CK15-DAG: store i32* %{{.+}}, i32** [[CP3]] - - // CK15-DAG: [[BP4:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 4 - // CK15-DAG: [[P4:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 4 - // CK15-DAG: [[CBP4:%.+]] = bitcast i8** [[BP4]] to i[[sz]]* - // CK15-DAG: [[CP4:%.+]] = bitcast i8** [[P4]] to i[[sz]]* - // CK15-DAG: store i[[sz]] [[VAL:%.+]], i[[sz]]* [[CBP4]] - // CK15-DAG: store i[[sz]] [[VAL]], i[[sz]]* [[CP4]] + // CK15-DAG: [[S3:%.+]] = getelementptr inbounds {{.+}}[[S]], i32 0, i32 3 + // CK15-DAG: [[CBP3:%.+]] = bitcast i8** [[BP3]] to i[[sz]]* + // CK15-DAG: [[CP3:%.+]] = bitcast i8** [[P3]] to i[[sz]]* + // CK15-DAG: store i[[sz]] [[VAL:%.+]], i[[sz]]* [[CBP3]] + // CK15-DAG: store i[[sz]] [[VAL]], i[[sz]]* [[CP3]] + // CK15-DAG: store i[[sz]] 4, i[[sz]]* [[S3]] // CK15-DAG: [[VAL]] = load i[[sz]], i[[sz]]* [[ADDR:%.+]], // CK15-64-DAG: [[CADDR:%.+]] = bitcast i[[sz]]* [[ADDR]] to i32* // CK15-64-DAG: store i32 {{.+}}, i32* [[CADDR]], @@ -890,44 +1057,46 @@ void implicit_maps_templated_class (int a){ ssst.foo(456); // CK15: define {{.*}}void @{{.+}}bar{{.+}}([[ST]]* {{[^,]+}}, i32 {{[^,]+}}) - // CK15-DAG: call i32 @__tgt_target(i64 {{.+}}, i8* {{.+}}, i32 5, i8** [[BPGEP:%[0-9]+]], i8** [[PGEP:%[0-9]+]], {{.+}}[[SIZES2]]{{.+}}, {{.+}}[[TYPES2]]{{.+}}) + // CK15-DAG: call i32 @__tgt_target(i64 {{.+}}, i8* {{.+}}, i32 4, i8** [[BPGEP:%[0-9]+]], i8** [[PGEP:%[0-9]+]], i[[sz]]* [[SIZES:[^,]+]], {{.+}}[[TYPES2]]{{.+}}) // CK15-DAG: [[BPGEP]] = getelementptr inbounds {{.+}}[[BPS:%[^,]+]], i32 0, i32 0 // CK15-DAG: [[PGEP]] = getelementptr inbounds {{.+}}[[PS:%[^,]+]], i32 0, i32 0 + // CK15-DAG: [[SIZES]] = getelementptr inbounds {{.+}}[[S:%[^,]+]], i32 0, i32 0 // CK15-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 0 // CK15-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 0 + // CK15-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i32 0, i32 0 // CK15-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]** // CK15-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32** - // CK15-DAG: store [[ST]]* [[DECL]], [[ST]]** [[CBP0]] - // CK15-DAG: store i32* %{{.+}}, i32** [[CP0]] + // CK15-DAG: store [[ST]]* [[DECL:%.+]], [[ST]]** [[CBP0]] + // CK15-DAG: store i32* [[A:%.+]], i32** [[CP0]] + // CK15-DAG: store i[[sz]] %{{.+}}, i[[sz]]* [[S0]] // CK15-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 1 // CK15-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 1 + // CK15-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i32 0, i32 1 // CK15-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[ST]]** - // CK15-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double** + // CK15-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** // CK15-DAG: store [[ST]]* [[DECL]], [[ST]]** [[CBP1]] - // CK15-DAG: store double* %{{.+}}, double** [[CP1]] + // CK15-DAG: store i32* [[A]], i32** [[CP1]] + // CK15-DAG: store i[[sz]] 4, i[[sz]]* [[S1]] // CK15-DAG: [[BP2:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 2 // CK15-DAG: [[P2:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 2 + // CK15-DAG: [[S2:%.+]] = getelementptr inbounds {{.+}}[[S]], i32 0, i32 2 // CK15-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to [[ST]]** - // CK15-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to i32*** + // CK15-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to double** // CK15-DAG: store [[ST]]* [[DECL]], [[ST]]** [[CBP2]] - // CK15-DAG: store i32** %{{.+}}, i32*** [[CP2]] + // CK15-DAG: store double* %{{.+}}, double** [[CP2]] + // CK15-DAG: store i[[sz]] 8, i[[sz]]* [[S2]] // CK15-DAG: [[BP3:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 3 // CK15-DAG: [[P3:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 3 - // CK15-DAG: [[CBP3:%.+]] = bitcast i8** [[BP3]] to i32*** - // CK15-DAG: [[CP3:%.+]] = bitcast i8** [[P3]] to i32** - // CK15-DAG: store i32** %{{.+}}, i32*** [[CBP3]] - // CK15-DAG: store i32* %{{.+}}, i32** [[CP3]] - - // CK15-DAG: [[BP4:%.+]] = getelementptr inbounds {{.+}}[[BPS]], i32 0, i32 4 - // CK15-DAG: [[P4:%.+]] = getelementptr inbounds {{.+}}[[PS]], i32 0, i32 4 - // CK15-DAG: [[CBP4:%.+]] = bitcast i8** [[BP4]] to i[[sz]]* - // CK15-DAG: [[CP4:%.+]] = bitcast i8** [[P4]] to i[[sz]]* - // CK15-DAG: store i[[sz]] [[VAL:%.+]], i[[sz]]* [[CBP4]] - // CK15-DAG: store i[[sz]] [[VAL]], i[[sz]]* [[CP4]] + // CK15-DAG: [[S3:%.+]] = getelementptr inbounds {{.+}}[[S]], i32 0, i32 3 + // CK15-DAG: [[CBP3:%.+]] = bitcast i8** [[BP3]] to i[[sz]]* + // CK15-DAG: [[CP3:%.+]] = bitcast i8** [[P3]] to i[[sz]]* + // CK15-DAG: store i[[sz]] [[VAL:%.+]], i[[sz]]* [[CBP3]] + // CK15-DAG: store i[[sz]] [[VAL]], i[[sz]]* [[CP3]] + // CK15-DAG: store i[[sz]] 4, i[[sz]]* [[S3]] // CK15-DAG: [[VAL]] = load i[[sz]], i[[sz]]* [[ADDR:%.+]], // CK15-64-DAG: [[CADDR:%.+]] = bitcast i[[sz]]* [[ADDR]] to i32* // CK15-64-DAG: store i32 {{.+}}, i32* [[CADDR]], @@ -960,18 +1129,26 @@ void implicit_maps_templated_class (int a){ #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK16 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK16 --check-prefix CK16-64 +// RUN: %clang_cc1 -DCK16 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK16 --check-prefix CK16-64 // RUN: %clang_cc1 -DCK16 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK16 --check-prefix CK16-64 -// RUN: %clang_cc1 -DCK16 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK16 --check-prefix CK16-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK16 --check-prefix CK16-64 +// RUN: %clang_cc1 -DCK16 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK16 --check-prefix CK16-32 // RUN: %clang_cc1 -DCK16 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK16 --check-prefix CK16-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK16 --check-prefix CK16-32 + +// RUN: %clang_cc1 -DCK16 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY15 %s +// RUN: %clang_cc1 -DCK16 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY15 %s +// RUN: %clang_cc1 -DCK16 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY15 %s +// RUN: %clang_cc1 -DCK16 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY15 %s +// SIMD-ONLY15-NOT: {{__kmpc|__tgt}} #ifdef CK16 // CK16-DAG: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} 4] // Map types: -// - OMP_MAP_PRIVATE_VAL + OMP_MAP_IS_FIRST = 288 -// CK16-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 288] +// - OMP_MAP_PRIVATE_VAL + OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 800 +// CK16-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 800] template<int y> int foo(int d) { @@ -982,7 +1159,7 @@ int foo(int d) { } return res; } -// CK16-LABEL: implicit_maps_templated_function +// CK16-LABEL: implicit_maps_templated_function{{.*}}( void implicit_maps_templated_function (int a){ int i = a; @@ -1013,17 +1190,26 @@ void implicit_maps_templated_function (int a){ #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK17 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK17 +// RUN: %clang_cc1 -DCK17 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK17 // RUN: %clang_cc1 -DCK17 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK17 -// RUN: %clang_cc1 -DCK17 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK17 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK17 +// RUN: %clang_cc1 -DCK17 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK17 // RUN: %clang_cc1 -DCK17 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK17 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK17 + +// RUN: %clang_cc1 -DCK17 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY16 %s +// RUN: %clang_cc1 -DCK17 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY16 %s +// RUN: %clang_cc1 -DCK17 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY16 %s +// RUN: %clang_cc1 -DCK17 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY16 %s +// SIMD-ONLY16-NOT: {{__kmpc|__tgt}} #ifdef CK17 // CK17-DAG: [[ST:%.+]] = type { i32, double } +// CK17-LABEL: @.__omp_offloading_{{.*}}implicit_maps_struct{{.*}}_l1236.region_id = weak constant i8 0 // CK17-DAG: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} {{16|12}}] -// Map types: OMP_MAP_TO + OMP_MAP_FROM + OMP_MAP_IS_FIRST + OMP_MAP_IMPLICIT = 547 +// Map types: OMP_MAP_TO + OMP_MAP_FROM + OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 547 // CK17-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 547] class SSS { @@ -1032,7 +1218,7 @@ public: double b; }; -// CK17-LABEL: implicit_maps_struct +// CK17-LABEL: implicit_maps_struct{{.*}}( void implicit_maps_struct (int a){ SSS s = {a, (double)a}; @@ -1061,18 +1247,26 @@ void implicit_maps_struct (int a){ // CK17: {{.+}} = getelementptr inbounds [[ST]], [[ST]]* [[REF]], i32 0, i32 0 #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK18 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK18 --check-prefix CK18-64 +// RUN: %clang_cc1 -DCK18 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK18 --check-prefix CK18-64 // RUN: %clang_cc1 -DCK18 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK18 --check-prefix CK18-64 -// RUN: %clang_cc1 -DCK18 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK18 --check-prefix CK18-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK18 --check-prefix CK18-64 +// RUN: %clang_cc1 -DCK18 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK18 --check-prefix CK18-32 // RUN: %clang_cc1 -DCK18 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK18 --check-prefix CK18-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK18 --check-prefix CK18-32 + +// RUN: %clang_cc1 -DCK18 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY17 %s +// RUN: %clang_cc1 -DCK18 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY17 %s +// RUN: %clang_cc1 -DCK18 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY17 %s +// RUN: %clang_cc1 -DCK18 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY17 %s +// SIMD-ONLY17-NOT: {{__kmpc|__tgt}} #ifdef CK18 // CK18-DAG: [[SIZES:@.+]] = {{.+}}constant [1 x i[[sz:64|32]]] [i{{64|32}} 4] // Map types: -// - OMP_MAP_PRIVATE_VAL + OMP_MAP_IS_FIRST = 288 -// CK18-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 288] +// - OMP_MAP_PRIVATE_VAL + OMP_MAP_TARGET_PARAM | OMP_MAP_IMPLICIT = 800 +// CK18-DAG: [[TYPES:@.+]] = {{.+}}constant [1 x i64] [i64 800] template<typename T> int foo(T d) { @@ -1082,7 +1276,7 @@ int foo(T d) { } return d; } -// CK18-LABEL: implicit_maps_template_type_capture +// CK18-LABEL: implicit_maps_template_type_capture{{.*}}( void implicit_maps_template_type_capture (int a){ int i = a; @@ -1113,136 +1307,189 @@ void implicit_maps_template_type_capture (int a){ #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK19 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK19 --check-prefix CK19-64 +// RUN: %clang_cc1 -DCK19 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK19 --check-prefix CK19-64 // RUN: %clang_cc1 -DCK19 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK19 --check-prefix CK19-64 -// RUN: %clang_cc1 -DCK19 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK19 --check-prefix CK19-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK19 --check-prefix CK19-64 +// RUN: %clang_cc1 -DCK19 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK19 --check-prefix CK19-32 // RUN: %clang_cc1 -DCK19 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK19 --check-prefix CK19-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK19 --check-prefix CK19-32 + +// RUN: %clang_cc1 -DCK19 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY18 %s +// RUN: %clang_cc1 -DCK19 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY18 %s +// RUN: %clang_cc1 -DCK19 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY18 %s +// RUN: %clang_cc1 -DCK19 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY18 %s +// SIMD-ONLY18-NOT: {{__kmpc|__tgt}} #ifdef CK19 +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1510.region_id = weak constant i8 0 // CK19: [[SIZE00:@.+]] = private {{.*}}constant [1 x i[[Z:64|32]]] [i[[Z:64|32]] 4] // CK19: [[MTYPE00:@.+]] = private {{.*}}constant [1 x i64] [i64 32] -// CK19: [[SIZE00n:@.+]] = private {{.*}}constant [1 x i[[Z:64|32]]] [i[[Z:64|32]] 4] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1531.region_id = weak constant i8 0 +// CK19: [[SIZE00n:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] // CK19: [[MTYPE00n:@.+]] = private {{.*}}constant [1 x i64] [i64 32] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1553.region_id = weak constant i8 0 // CK19: [[SIZE01:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 400] // CK19: [[MTYPE01:@.+]] = private {{.*}}constant [1 x i64] [i64 33] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1572.region_id = weak constant i8 0 // CK19: [[SIZE02:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 240] // CK19: [[MTYPE02:@.+]] = private {{.*}}constant [1 x i64] [i64 34] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1591.region_id = weak constant i8 0 // CK19: [[SIZE03:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 240] // CK19: [[MTYPE03:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1610.region_id = weak constant i8 0 // CK19: [[SIZE04:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 400] // CK19: [[MTYPE04:@.+]] = private {{.*}}constant [1 x i64] [i64 32] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1629.region_id = weak constant i8 0 // CK19: [[SIZE05:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] // CK19: [[MTYPE05:@.+]] = private {{.*}}constant [1 x i64] [i64 33] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1652.region_id = weak constant i8 0 // CK19: [[MTYPE06:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1675.region_id = weak constant i8 0 // CK19: [[MTYPE07:@.+]] = private {{.*}}constant [1 x i64] [i64 32] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1694.region_id = weak constant i8 0 // CK19: [[SIZE08:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] // CK19: [[MTYPE08:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1715.region_id = weak constant i8 0 // CK19: [[SIZE09:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] {{8|4}}] // CK19: [[MTYPE09:@.+]] = private {{.*}}constant [1 x i64] [i64 34] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1736.region_id = weak constant i8 0 // CK19: [[SIZE10:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 240] // CK19: [[MTYPE10:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1757.region_id = weak constant i8 0 // CK19: [[SIZE11:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 240] // CK19: [[MTYPE11:@.+]] = private {{.*}}constant [1 x i64] [i64 32] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1778.region_id = weak constant i8 0 // CK19: [[SIZE12:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] // CK19: [[MTYPE12:@.+]] = private {{.*}}constant [1 x i64] [i64 33] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1803.region_id = weak constant i8 0 // CK19: [[MTYPE13:@.+]] = private {{.*}}constant [1 x i64] [i64 32] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1828.region_id = weak constant i8 0 // CK19: [[MTYPE14:@.+]] = private {{.*}}constant [1 x i64] [i64 33] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1849.region_id = weak constant i8 0 // CK19: [[SIZE15:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] // CK19: [[MTYPE15:@.+]] = private {{.*}}constant [1 x i64] [i64 34] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1883.region_id = weak constant i8 0 // CK19: [[MTYPE16:@.+]] = private {{.*}}constant [2 x i64] [i64 288, i64 33] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1909.region_id = weak constant i8 0 // CK19: [[SIZE17:@.+]] = private {{.*}}constant [2 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] 240] // CK19: [[MTYPE17:@.+]] = private {{.*}}constant [2 x i64] [i64 288, i64 34] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1935.region_id = weak constant i8 0 // CK19: [[SIZE18:@.+]] = private {{.*}}constant [2 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] 240] // CK19: [[MTYPE18:@.+]] = private {{.*}}constant [2 x i64] [i64 288, i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1967.region_id = weak constant i8 0 // CK19: [[MTYPE19:@.+]] = private {{.*}}constant [2 x i64] [i64 288, i64 32] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l1993.region_id = weak constant i8 0 // CK19: [[SIZE20:@.+]] = private {{.*}}constant [2 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] 4] // CK19: [[MTYPE20:@.+]] = private {{.*}}constant [2 x i64] [i64 288, i64 33] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2025.region_id = weak constant i8 0 // CK19: [[MTYPE21:@.+]] = private {{.*}}constant [2 x i64] [i64 288, i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2051.region_id = weak constant i8 0 // CK19: [[SIZE22:@.+]] = private {{.*}}constant [2 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] 4] // CK19: [[MTYPE22:@.+]] = private {{.*}}constant [2 x i64] [i64 288, i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2070.region_id = weak constant i8 0 // CK19: [[SIZE23:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] // CK19: [[MTYPE23:@.+]] = private {{.*}}constant [1 x i64] [i64 39] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2092.region_id = weak constant i8 0 // CK19: [[SIZE24:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 480] // CK19: [[MTYPE24:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2113.region_id = weak constant i8 0 // CK19: [[SIZE25:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 16] // CK19: [[MTYPE25:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2134.region_id = weak constant i8 0 // CK19: [[SIZE26:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 24] // CK19: [[MTYPE26:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2155.region_id = weak constant i8 0 // CK19: [[SIZE27:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] // CK19: [[MTYPE27:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2200.region_id = weak constant i8 0 // CK19: [[SIZE28:@.+]] = private {{.*}}constant [3 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] {{8|4}}, i[[Z]] 16] -// CK19: [[MTYPE28:@.+]] = private {{.*}}constant [3 x i64] [i64 35, i64 19, i64 19] +// CK19: [[MTYPE28:@.+]] = private {{.*}}constant [3 x i64] [i64 35, i64 16, i64 19] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2245.region_id = weak constant i8 0 // CK19: [[SIZE29:@.+]] = private {{.*}}constant [3 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] {{8|4}}, i[[Z]] 4] -// CK19: [[MTYPE29:@.+]] = private {{.*}}constant [3 x i64] [i64 35, i64 19, i64 19] +// CK19: [[MTYPE29:@.+]] = private {{.*}}constant [3 x i64] [i64 35, i64 16, i64 19] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2301.region_id = weak constant i8 0 // CK19: [[MTYPE30:@.+]] = private {{.*}}constant [4 x i64] [i64 288, i64 288, i64 288, i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2345.region_id = weak constant i8 0 // CK19: [[SIZE31:@.+]] = private {{.*}}constant [4 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] {{8|4}}, i[[Z]] {{8|4}}, i[[Z]] 40] // CK19: [[MTYPE31:@.+]] = private {{.*}}constant [4 x i64] [i64 288, i64 288, i64 288, i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2368.region_id = weak constant i8 0 // CK19: [[SIZE32:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 13728] // CK19: [[MTYPE32:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2387.region_id = weak constant i8 0 // CK19: [[SIZE33:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 13728] // CK19: [[MTYPE33:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2406.region_id = weak constant i8 0 // CK19: [[SIZE34:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 13728] // CK19: [[MTYPE34:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2431.region_id = weak constant i8 0 // CK19: [[MTYPE35:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2452.region_id = weak constant i8 0 // CK19: [[SIZE36:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 208] // CK19: [[MTYPE36:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2492.region_id = weak constant i8 0 // CK19: [[MTYPE37:@.+]] = private {{.*}}constant [3 x i64] [i64 288, i64 288, i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2534.region_id = weak constant i8 0 // CK19: [[MTYPE38:@.+]] = private {{.*}}constant [3 x i64] [i64 288, i64 288, i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2576.region_id = weak constant i8 0 // CK19: [[MTYPE39:@.+]] = private {{.*}}constant [3 x i64] [i64 288, i64 288, i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2618.region_id = weak constant i8 0 // CK19: [[MTYPE40:@.+]] = private {{.*}}constant [3 x i64] [i64 288, i64 288, i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2653.region_id = weak constant i8 0 // CK19: [[SIZE41:@.+]] = private {{.*}}constant [3 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] {{8|4}}, i[[Z]] 208] // CK19: [[MTYPE41:@.+]] = private {{.*}}constant [3 x i64] [i64 288, i64 288, i64 35] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2698.region_id = weak constant i8 0 // CK19: [[SIZE42:@.+]] = private {{.*}}constant [3 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] {{8|4}}, i[[Z]] 104] -// CK19: [[MTYPE42:@.+]] = private {{.*}}constant [3 x i64] [i64 35, i64 19, i64 19] +// CK19: [[MTYPE42:@.+]] = private {{.*}}constant [3 x i64] [i64 35, i64 16, i64 19] +// CK19-LABEL: @.__omp_offloading_{{.*}}explicit_maps_single{{.*}}_l2723.region_id = weak constant i8 0 // CK19: [[MTYPE43:@.+]] = private {{.*}}constant [1 x i64] [i64 35] -// CK19-LABEL: explicit_maps_single +// CK19-LABEL: explicit_maps_single{{.*}}( void explicit_maps_single (int ii){ // Map of a scalar. int a = ii; @@ -2481,7 +2728,6 @@ void explicit_maps_single (int ii){ } // CK19: define {{.+}}[[CALL00]] -// CK19: define {{.+}}[[CALL00n]] // CK19: define {{.+}}[[CALL01]] // CK19: define {{.+}}[[CALL02]] // CK19: define {{.+}}[[CALL03]] @@ -2528,27 +2774,39 @@ void explicit_maps_single (int ii){ #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK20 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK20 --check-prefix CK20-64 +// RUN: %clang_cc1 -DCK20 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK20 --check-prefix CK20-64 // RUN: %clang_cc1 -DCK20 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK20 --check-prefix CK20-64 -// RUN: %clang_cc1 -DCK20 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK20 --check-prefix CK20-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK20 --check-prefix CK20-64 +// RUN: %clang_cc1 -DCK20 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK20 --check-prefix CK20-32 // RUN: %clang_cc1 -DCK20 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK20 --check-prefix CK20-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK20 --check-prefix CK20-32 + +// RUN: %clang_cc1 -DCK20 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY19 %s +// RUN: %clang_cc1 -DCK20 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY19 %s +// RUN: %clang_cc1 -DCK20 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY19 %s +// RUN: %clang_cc1 -DCK20 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY19 %s +// SIMD-ONLY19-NOT: {{__kmpc|__tgt}} #ifdef CK20 +// CK20-LABEL: @.__omp_offloading_{{.*}}explicit_maps_references_and_function_args{{.*}}_l2832.region_id = weak constant i8 0 // CK20: [[SIZE00:@.+]] = private {{.*}}constant [1 x i[[Z:64|32]]] [i[[Z:64|32]] 4] // CK20: [[MTYPE00:@.+]] = private {{.*}}constant [1 x i64] [i64 33] +// CK20-LABEL: @.__omp_offloading_{{.*}}explicit_maps_references_and_function_args{{.*}}_l2853.region_id = weak constant i8 0 // CK20: [[SIZE01:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 20] // CK20: [[MTYPE01:@.+]] = private {{.*}}constant [1 x i64] [i64 33] +// CK20-LABEL: @.__omp_offloading_{{.*}}explicit_maps_references_and_function_args{{.*}}_l2871.region_id = weak constant i8 0 // CK20: [[SIZE02:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] // CK20: [[MTYPE02:@.+]] = private {{.*}}constant [1 x i64] [i64 34] +// CK20-LABEL: @.__omp_offloading_{{.*}}explicit_maps_references_and_function_args{{.*}}_l2892.region_id = weak constant i8 0 // CK20: [[SIZE03:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 12] // CK20: [[MTYPE03:@.+]] = private {{.*}}constant [1 x i64] [i64 34] -// CK20-LABEL: explicit_maps_references_and_function_args +// CK20-LABEL: explicit_maps_references_and_function_args{{.*}}( void explicit_maps_references_and_function_args (int a, float b, int (&c)[10], float *d){ int &aa = a; @@ -2644,34 +2902,45 @@ void explicit_maps_references_and_function_args (int a, float b, int (&c)[10], f #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK21 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK21 --check-prefix CK21-64 +// RUN: %clang_cc1 -DCK21 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK21 --check-prefix CK21-64 // RUN: %clang_cc1 -DCK21 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK21 --check-prefix CK21-64 -// RUN: %clang_cc1 -DCK21 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK21 --check-prefix CK21-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK21 --check-prefix CK21-64 +// RUN: %clang_cc1 -DCK21 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK21 --check-prefix CK21-32 // RUN: %clang_cc1 -DCK21 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK21 --check-prefix CK21-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK21 --check-prefix CK21-32 + +// RUN: %clang_cc1 -DCK21 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY20 %s +// RUN: %clang_cc1 -DCK21 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY20 %s +// RUN: %clang_cc1 -DCK21 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY20 %s +// RUN: %clang_cc1 -DCK21 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY20 %s +// SIMD-ONLY20-NOT: {{__kmpc|__tgt}} #ifdef CK21 // CK21: [[ST:%.+]] = type { i32, i32, float* } -// CK21: [[SIZE00:@.+]] = private {{.*}}constant [1 x i[[Z:64|32]]] [i[[Z:64|32]] 4] -// CK21: [[MTYPE00:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK21-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK21: [[MTYPE00:@.+]] = private {{.*}}constant [2 x i64] [i64 32, i64 281474976710659] -// CK21: [[SIZE01:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 492] +// CK21-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK21: [[SIZE01:@.+]] = private {{.*}}constant [1 x i[[Z:64|32]]] [i[[Z:64|32]] 492] // CK21: [[MTYPE01:@.+]] = private {{.*}}constant [1 x i64] [i64 35] -// CK21: [[SIZE02:@.+]] = private {{.*}}constant [2 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] 500] -// CK21: [[MTYPE02:@.+]] = private {{.*}}constant [2 x i64] [i64 34, i64 18] +// CK21-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK21: [[MTYPE02:@.+]] = private {{.*}}constant [2 x i64] [i64 32, i64 281474976710674] +// CK21-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK21: [[SIZE03:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 492] // CK21: [[MTYPE03:@.+]] = private {{.*}}constant [1 x i64] [i64 34] +// CK21-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK21: [[SIZE04:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] // CK21: [[MTYPE04:@.+]] = private {{.*}}constant [1 x i64] [i64 34] -// CK21: [[SIZE05:@.+]] = private {{.*}}constant [2 x i[[Z]]] [i[[Z]] 4, i[[Z]] 4] -// CK21: [[MTYPE05:@.+]] = private {{.*}}constant [2 x i64] [i64 35, i64 3] +// CK21-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK21: [[MTYPE05:@.+]] = private {{.*}}constant [3 x i64] [i64 32, i64 281474976710659, i64 281474976710659] -// CK21-LABEL: explicit_maps_template_args_and_members +// CK21-LABEL: explicit_maps_template_args_and_members{{.*}}( template <int X, typename T> struct CC { @@ -2684,18 +2953,31 @@ struct CC { T *lb; // Region 00 - // CK21-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE00]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE00]]{{.+}}) + // CK21-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[Z]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}) // CK21-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK21-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] + // CK21-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK21-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK21-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 + // CK21-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK21-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]** // CK21-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32** // CK21-DAG: store [[ST]]* [[VAR0:%.+]], [[ST]]** [[CBP0]] // CK21-DAG: store i32* [[SEC0:%.+]], i32** [[CP0]] + // CK21-DAG: store i[[Z]] {{%.+}}, i[[Z]]* [[S0]] // CK21-DAG: [[SEC0]] = getelementptr {{.*}}[[ST]]* [[VAR0:%.+]], i{{.+}} 0, i{{.+}} 0 + // CK21-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 + // CK21-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 + // CK21-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 + // CK21-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[ST]]** + // CK21-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** + // CK21-DAG: store [[ST]]* [[VAR1:%.+]], [[ST]]** [[CBP1]] + // CK21-DAG: store i32* [[SEC1:%.+]], i32** [[CP1]] + // CK21-DAG: store i[[Z]] {{.+}}, i[[Z]]* [[S1]] + // CK21-DAG: [[SEC1]] = getelementptr {{.*}}[[ST]]* [[VAR1:%.+]], i{{.+}} 0, i{{.+}} 0 + // CK21: call void [[CALL00:@.+]]([[ST]]* {{[^,]+}}) #pragma omp target map(A) { @@ -2724,24 +3006,29 @@ struct CC { } // Region 02 - // CK21-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE02]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE02]]{{.+}}) + // CK21-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[Z]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE02]]{{.+}}) // CK21-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK21-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] + // CK21-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK21-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK21-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 + // CK21-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK21-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]** // CK21-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to float*** // CK21-DAG: store [[ST]]* [[VAR0:%.+]], [[ST]]** [[CBP0]] // CK21-DAG: store float** [[SEC0:%.+]], float*** [[CP0]] + // CK21-DAG: store i[[Z]] {{%.+}}, i[[Z]]* [[S0]] // CK21-DAG: [[SEC0]] = getelementptr {{.*}}[[ST]]* [[VAR0]], i{{.+}} 0, i{{.+}} 2 // CK21-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 // CK21-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 + // CK21-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 // CK21-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to float*** // CK21-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to float** // CK21-DAG: store float** [[SEC0]], float*** [[CBP1]] // CK21-DAG: store float* [[SEC1:%.+]], float** [[CP1]] + // CK21-DAG: store i[[Z]] {{.+}}, i[[Z]]* [[S1]] // CK21-DAG: [[SEC1]] = getelementptr {{.*}}float* [[RVAR1:%[^,]+]], i{{.+}} 123 // CK21-DAG: [[RVAR1]] = load float*, float** [[SEC1_:%[^,]+]] // CK21-DAG: [[SEC1_]] = getelementptr {{.*}}[[ST]]* [[VAR0]], i{{.+}} 0, i{{.+}} 2 @@ -2790,25 +3077,39 @@ struct CC { // Make sure the extra flag is passed to the second map. // Region 05 - // CK21-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE05]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE05]]{{.+}}) + // CK21-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 3, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[Z]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[3 x i{{.+}}]* [[MTYPE05]]{{.+}}) // CK21-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK21-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] + // CK21-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK21-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK21-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 + // CK21-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK21-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]** // CK21-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32** // CK21-DAG: store [[ST]]* [[VAR0:%.+]], [[ST]]** [[CBP0]] // CK21-DAG: store i32* [[SEC0:%.+]], i32** [[CP0]] + // CK21-DAG: store i[[Z]] {{%.+}}, i[[Z]]* [[S0]] // CK21-DAG: [[SEC0]] = getelementptr {{.*}}[[ST]]* [[VAR0]], i{{.+}} 0, i{{.+}} 0 // CK21-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 // CK21-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 + // CK21-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 // CK21-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[ST]]** // CK21-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** - // CK21-DAG: store [[ST]]* [[VAR1:%.+]], [[ST]]** [[CBP1]] - // CK21-DAG: store i32* [[SEC1:%.+]], i32** [[CP1]] - // CK21-DAG: [[SEC1]] = getelementptr {{.*}}[[ST]]* [[VAR0]], i{{.+}} 0, i{{.+}} 1 + // CK21-DAG: store [[ST]]* [[VAR0]], [[ST]]** [[CBP1]] + // CK21-DAG: store i32* [[SEC0]], i32** [[CP1]] + // CK21-DAG: store i[[Z]] {{.+}}, i[[Z]]* [[S1]] + + // CK21-DAG: [[BP2:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 2 + // CK21-DAG: [[P2:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 2 + // CK21-DAG: [[S2:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 2 + // CK21-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to [[ST]]** + // CK21-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to i32** + // CK21-DAG: store [[ST]]* [[VAR2:%.+]], [[ST]]** [[CBP2]] + // CK21-DAG: store i32* [[SEC2:%.+]], i32** [[CP2]] + // CK21-DAG: store i[[Z]] {{.+}}, i[[Z]]* [[S2]] + // CK21-DAG: [[SEC2]] = getelementptr {{.*}}[[ST]]* [[VAR2]], i{{.+}} 0, i{{.+}} 1 // CK21: call void [[CALL05:@.+]]([[ST]]* {{[^,]+}}) #pragma omp target map(A, A2) @@ -2833,59 +3134,82 @@ int explicit_maps_template_args_and_members(int a){ // CK21: define {{.+}}[[CALL05]] #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK22 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK22 --check-prefix CK22-64 +// RUN: %clang_cc1 -DCK22 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK22 --check-prefix CK22-64 // RUN: %clang_cc1 -DCK22 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK22 --check-prefix CK22-64 -// RUN: %clang_cc1 -DCK22 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK22 --check-prefix CK22-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK22 --check-prefix CK22-64 +// RUN: %clang_cc1 -DCK22 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK22 --check-prefix CK22-32 // RUN: %clang_cc1 -DCK22 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK22 --check-prefix CK22-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK22 --check-prefix CK22-32 + +// RUN: %clang_cc1 -DCK22 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY21 %s +// RUN: %clang_cc1 -DCK22 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY21 %s +// RUN: %clang_cc1 -DCK22 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY21 %s +// RUN: %clang_cc1 -DCK22 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY21 %s +// SIMD-ONLY21-NOT: {{__kmpc|__tgt}} #ifdef CK22 // CK22-DAG: [[ST:%.+]] = type { float } // CK22-DAG: [[STT:%.+]] = type { i32 } +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE00:@.+]] = private {{.*}}constant [1 x i[[Z:64|32]]] [i[[Z:64|32]] 4] // CK22: [[MTYPE00:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE01:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 400] // CK22: [[MTYPE01:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE02:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] {{8|4}}] // CK22: [[MTYPE02:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE03:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 16] // CK22: [[MTYPE03:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE04:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 20] // CK22: [[MTYPE04:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE05:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] // CK22: [[MTYPE05:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE06:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 400] // CK22: [[MTYPE06:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE07:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] {{8|4}}] // CK22: [[MTYPE07:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE08:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 16] // CK22: [[MTYPE08:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE09:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 20] // CK22: [[MTYPE09:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE10:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] // CK22: [[MTYPE10:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE11:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 400] // CK22: [[MTYPE11:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE12:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] {{8|4}}] // CK22: [[MTYPE12:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE13:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 16] // CK22: [[MTYPE13:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK22-LABEL: @.__omp_offloading_{{.*}}explicit_maps_globals{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK22: [[SIZE14:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 20] // CK22: [[MTYPE14:@.+]] = private {{.*}}constant [1 x i64] [i64 35] @@ -2910,7 +3234,7 @@ STT<int> sta ; STT<int> stc[100]; STT<int> *std; -// CK22-LABEL: explicit_maps_globals +// CK22-LABEL: explicit_maps_globals{{.*}}( int explicit_maps_globals(void){ // Region 00 // CK22-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE00]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE00]]{{.+}}) @@ -3180,33 +3504,47 @@ int explicit_maps_globals(void){ // CK22: define {{.+}}[[CALL14]] #endif ///==========================================================================/// -// RUN: %clang_cc1 -std=c++11 -DCK23 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK23 --check-prefix CK23-64 +// RUN: %clang_cc1 -std=c++11 -DCK23 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK23 --check-prefix CK23-64 // RUN: %clang_cc1 -std=c++11 -DCK23 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -std=c++11 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK23 --check-prefix CK23-64 -// RUN: %clang_cc1 -std=c++11 -DCK23 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK23 --check-prefix CK23-32 +// RUN: %clang_cc1 -std=c++11 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK23 --check-prefix CK23-64 +// RUN: %clang_cc1 -std=c++11 -DCK23 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK23 --check-prefix CK23-32 // RUN: %clang_cc1 -std=c++11 -DCK23 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -std=c++11 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK23 --check-prefix CK23-32 +// RUN: %clang_cc1 -std=c++11 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK23 --check-prefix CK23-32 + +// RUN: %clang_cc1 -std=c++11 -DCK23 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY22 %s +// RUN: %clang_cc1 -std=c++11 -DCK23 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -std=c++11 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY22 %s +// RUN: %clang_cc1 -std=c++11 -DCK23 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY22 %s +// RUN: %clang_cc1 -std=c++11 -DCK23 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -std=c++11 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY22 %s +// SIMD-ONLY22-NOT: {{__kmpc|__tgt}} #ifdef CK23 +// CK23-LABEL: @.__omp_offloading_{{.*}}explicit_maps_inside_captured{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK23: [[SIZE00:@.+]] = private {{.*}}constant [1 x i[[Z:64|32]]] [i[[Z:64|32]] 4] // CK23: [[MTYPE00:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK23-LABEL: @.__omp_offloading_{{.*}}explicit_maps_inside_captured{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK23: [[SIZE01:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] // CK23: [[MTYPE01:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK23-LABEL: @.__omp_offloading_{{.*}}explicit_maps_inside_captured{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK23: [[SIZE02:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 400] // CK23: [[MTYPE02:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK23-LABEL: @.__omp_offloading_{{.*}}explicit_maps_inside_captured{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK23: [[SIZE03:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] {{8|4}}] // CK23: [[MTYPE03:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK23-LABEL: @.__omp_offloading_{{.*}}explicit_maps_inside_captured{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK23: [[SIZE04:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 16] // CK23: [[MTYPE04:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK23-LABEL: @.__omp_offloading_{{.*}}explicit_maps_inside_captured{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK23: [[SIZE05:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 16] // CK23: [[MTYPE05:@.+]] = private {{.*}}constant [1 x i64] [i64 35] -// CK23-LABEL: explicit_maps_inside_captured +// CK23-LABEL: explicit_maps_inside_captured{{.*}}( int explicit_maps_inside_captured(int a){ float b; float c[100]; @@ -3347,12 +3685,20 @@ int explicit_maps_inside_captured(int a){ // CK23: define {{.+}}[[CALL05]] #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK24 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK24 --check-prefix CK24-64 +// RUN: %clang_cc1 -DCK24 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK24 --check-prefix CK24-64 // RUN: %clang_cc1 -DCK24 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK24 --check-prefix CK24-64 -// RUN: %clang_cc1 -DCK24 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK24 --check-prefix CK24-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK24 --check-prefix CK24-64 +// RUN: %clang_cc1 -DCK24 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK24 --check-prefix CK24-32 // RUN: %clang_cc1 -DCK24 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK24 --check-prefix CK24-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK24 --check-prefix CK24-32 + +// RUN: %clang_cc1 -DCK24 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY23 %s +// RUN: %clang_cc1 -DCK24 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY23 %s +// RUN: %clang_cc1 -DCK24 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY23 %s +// RUN: %clang_cc1 -DCK24 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY23 %s +// SIMD-ONLY23-NOT: {{__kmpc|__tgt}} #ifdef CK24 // CK24-DAG: [[SC:%.+]] = type { i32, [[SB:%.+]], [[SB:%.+]]*, [10 x i32] } @@ -3378,44 +3724,44 @@ struct SC{ int b[10]; }; -// CK24: [[SIZE01:@.+]] = private {{.*}}constant [1 x i[[Z:64|32]]] [i[[Z:64|32]] 4] -// CK24: [[MTYPE01:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK24-LABEL: @.__omp_offloading_{{.*}}explicit_maps_struct_fields{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK24: [[MTYPE01:@.+]] = private {{.*}}constant [2 x i64] [i64 32, i64 281474976710659] -// CK24: [[SIZE13:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] -// CK24: [[MTYPE13:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK24-LABEL: @.__omp_offloading_{{.*}}explicit_maps_struct_fields{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK24: [[MTYPE13:@.+]] = private {{.*}}constant [2 x i64] [i64 32, i64 281474976710659] -// CK24: [[SIZE14:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] {{56|48}}] -// CK24: [[MTYPE14:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK24-LABEL: @.__omp_offloading_{{.*}}explicit_maps_struct_fields{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK24: [[MTYPE14:@.+]] = private {{.*}}constant [2 x i64] [i64 32, i64 281474976710659] -// CK24: [[SIZE15:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] -// CK24: [[MTYPE15:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK24-LABEL: @.__omp_offloading_{{.*}}explicit_maps_struct_fields{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK24: [[MTYPE15:@.+]] = private {{.*}}constant [2 x i64] [i64 32, i64 281474976710659] -// CK24: [[SIZE16:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 20] -// CK24: [[MTYPE16:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK24-LABEL: @.__omp_offloading_{{.*}}explicit_maps_struct_fields{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK24: [[MTYPE16:@.+]] = private {{.*}}constant [2 x i64] [i64 32, i64 281474976710659] -// CK24: [[SIZE17:@.+]] = private {{.*}}constant [2 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] {{3560|2880}}] -// CK24: [[MTYPE17:@.+]] = private {{.*}}constant [2 x i64] [i64 35, i64 19] +// CK24-LABEL: @.__omp_offloading_{{.*}}explicit_maps_struct_fields{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK24: [[MTYPE17:@.+]] = private {{.*}}constant [2 x i64] [i64 32, i64 281474976710675] -// CK24: [[SIZE18:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] -// CK24: [[MTYPE18:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK24-LABEL: @.__omp_offloading_{{.*}}explicit_maps_struct_fields{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK24: [[MTYPE18:@.+]] = private {{.*}}constant [2 x i64] [i64 32, i64 281474976710659] -// CK24: [[SIZE19:@.+]] = private {{.*}}constant [2 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] 4] -// CK24: [[MTYPE19:@.+]] = private {{.*}}constant [2 x i64] [i64 35, i64 19] +// CK24-LABEL: @.__omp_offloading_{{.*}}explicit_maps_struct_fields{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK24: [[MTYPE19:@.+]] = private {{.*}}constant [3 x i64] [i64 32, i64 281474976710659, i64 281474976710675] -// CK24: [[SIZE20:@.+]] = private {{.*}}constant [2 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] 4] -// CK24: [[MTYPE20:@.+]] = private {{.*}}constant [2 x i64] [i64 35, i64 19] +// CK24-LABEL: @.__omp_offloading_{{.*}}explicit_maps_struct_fields{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK24: [[MTYPE20:@.+]] = private {{.*}}constant [2 x i64] [i64 32, i64 281474976710675] -// CK24: [[SIZE21:@.+]] = private {{.*}}constant [2 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] 4] -// CK24: [[MTYPE21:@.+]] = private {{.*}}constant [2 x i64] [i64 35, i64 19] +// CK24-LABEL: @.__omp_offloading_{{.*}}explicit_maps_struct_fields{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK24: [[MTYPE21:@.+]] = private {{.*}}constant [3 x i64] [i64 32, i64 281474976710659, i64 281474976710675] -// CK24: [[SIZE22:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] {{8|4}}] -// CK24: [[MTYPE22:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK24-LABEL: @.__omp_offloading_{{.*}}explicit_maps_struct_fields{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK24: [[MTYPE22:@.+]] = private {{.*}}constant [2 x i64] [i64 32, i64 281474976710659] -// CK24: [[SIZE23:@.+]] = private {{.*}}constant [2 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] {{8|4}}] -// CK24: [[MTYPE23:@.+]] = private {{.*}}constant [2 x i64] [i64 35, i64 19] +// CK24-LABEL: @.__omp_offloading_{{.*}}explicit_maps_struct_fields{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK24: [[MTYPE23:@.+]] = private {{.*}}constant [3 x i64] [i64 32, i64 281474976710659, i64 281474976710675] -// CK24: [[SIZE24:@.+]] = private {{.*}}constant [4 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] {{8|4}}, i[[Z]] {{8|4}}, i[[Z]] 4] -// CK24: [[MTYPE24:@.+]] = private {{.*}}constant [4 x i64] [i64 35, i64 19, i64 19, i64 19] +// CK24-LABEL: @.__omp_offloading_{{.*}}explicit_maps_struct_fields{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK24: [[MTYPE24:@.+]] = private {{.*}}constant [4 x i64] [i64 32, i64 281474976710672, i64 16, i64 19] // CK24-LABEL: explicit_maps_struct_fields int explicit_maps_struct_fields(int a){ @@ -3423,81 +3769,129 @@ int explicit_maps_struct_fields(int a){ SC *p; // Region 01 -// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE01]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE01]]{{.+}}) +// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE01]]{{.+}}) // CK24-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK24-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK24-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK24-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SC]]** // CK24-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32** // CK24-DAG: store [[SC]]* [[VAR0:%.+]], [[SC]]** [[CBP0]] // CK24-DAG: store i32* [[SEC0:%.+]], i32** [[CP0]] +// CK24-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK24-DAG: [[SEC0]] = getelementptr {{.*}}[[SC]]* [[VAR0]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SC]]** +// CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** +// CK24-DAG: store [[SC]]* [[VAR0]], [[SC]]** [[CBP1]] +// CK24-DAG: store i32* [[SEC0]], i32** [[CP1]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S1]] + // CK24: call void [[CALL01:@.+]]([[SC]]* {{[^,]+}}) #pragma omp target map(s.a) { s.a++; } - + // // Same thing but starting from a pointer. // // Region 13 -// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE13]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE13]]{{.+}}) +// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE13]]{{.+}}) // CK24-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK24-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK24-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK24-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SC]]** // CK24-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32** // CK24-DAG: store [[SC]]* [[VAR0:%.+]], [[SC]]** [[CBP0]] // CK24-DAG: store i32* [[SEC0:%.+]], i32** [[CP0]] +// CK24-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK24-DAG: [[SEC0]] = getelementptr {{.*}}[[SC]]* [[VAR00:%.+]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SC]]** +// CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** +// CK24-DAG: store [[SC]]* [[VAR0]], [[SC]]** [[CBP1]] +// CK24-DAG: store i32* [[SEC0]], i32** [[CP1]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S1]] + // CK24-DAG: [[VAR0]] = load [[SC]]*, [[SC]]** %{{.+}} // CK24-DAG: [[VAR00]] = load [[SC]]*, [[SC]]** %{{.+}} // CK24: call void [[CALL13:@.+]]([[SC]]* {{[^,]+}}) #pragma omp target map(p->a) { p->a++; } - + // Region 14 -// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE14]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE14]]{{.+}}) +// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE14]]{{.+}}) // CK24-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK24-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK24-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK24-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SC]]** // CK24-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[SA]]** // CK24-DAG: store [[SC]]* [[VAR0:%.+]], [[SC]]** [[CBP0]] // CK24-DAG: store [[SA]]* [[SEC0:%.+]], [[SA]]** [[CP0]] +// CK24-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK24-DAG: [[SEC0]] = getelementptr {{.*}}[[SB]]* [[SEC00:%[^,]+]], i{{.+}} 0, i{{.+}} 1 // CK24-DAG: [[SEC00]] = getelementptr {{.*}}[[SC]]* [[VAR00:%.+]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SC]]** +// CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to [[SA]]** +// CK24-DAG: store [[SC]]* [[VAR0]], [[SC]]** [[CBP1]] +// CK24-DAG: store [[SA]]* [[SEC0]], [[SA]]** [[CP1]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S1]] + // CK24-DAG: [[VAR0]] = load [[SC]]*, [[SC]]** %{{.+}} // CK24-DAG: [[VAR00]] = load [[SC]]*, [[SC]]** %{{.+}} // CK24: call void [[CALL14:@.+]]([[SC]]* {{[^,]+}}) #pragma omp target map(p->s.s) { p->a++; } - + // Region 15 -// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE15]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE15]]{{.+}}) +// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE15]]{{.+}}) // CK24-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK24-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK24-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK24-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SC]]** // CK24-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32** // CK24-DAG: store [[SC]]* [[VAR0:%.+]], [[SC]]** [[CBP0]] // CK24-DAG: store i32* [[SEC0:%.+]], i32** [[CP0]] +// CK24-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK24-DAG: [[SEC0]] = getelementptr {{.*}}[[SA]]* [[SEC00:%[^,]+]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[SEC00]] = getelementptr {{.*}}[[SB]]* [[SEC000:%[^,]+]], i{{.+}} 0, i{{.+}} 1 // CK24-DAG: [[SEC000]] = getelementptr {{.*}}[[SC]]* [[VAR00:%.+]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SC]]** +// CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** +// CK24-DAG: store [[SC]]* [[VAR0]], [[SC]]** [[CBP1]] +// CK24-DAG: store i32* [[SEC0]], i32** [[CP1]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S1]] + // CK24-DAG: [[VAR0]] = load [[SC]]*, [[SC]]** %{{.+}} // CK24-DAG: [[VAR00]] = load [[SC]]*, [[SC]]** %{{.+}} @@ -3506,19 +3900,31 @@ int explicit_maps_struct_fields(int a){ { p->a++; } // Region 16 -// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE16]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE16]]{{.+}}) +// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE16]]{{.+}}) // CK24-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK24-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK24-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK24-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SC]]** // CK24-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32** // CK24-DAG: store [[SC]]* [[VAR0:%.+]], [[SC]]** [[CBP0]] // CK24-DAG: store i32* [[SEC0:%.+]], i32** [[CP0]] +// CK24-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK24-DAG: [[SEC0]] = getelementptr {{.*}}[10 x i32]* [[SEC00:%[^,]+]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[SEC00]] = getelementptr {{.*}}[[SC]]* [[VAR00:%.+]], i{{.+}} 0, i{{.+}} 3 +// CK24-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SC]]** +// CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** +// CK24-DAG: store [[SC]]* [[VAR0]], [[SC]]** [[CBP1]] +// CK24-DAG: store i32* [[SEC0]], i32** [[CP1]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S1]] + // CK24-DAG: [[VAR0]] = load [[SC]]*, [[SC]]** %{{.+}} // CK24-DAG: [[VAR00]] = load [[SC]]*, [[SC]]** %{{.+}} @@ -3527,24 +3933,29 @@ int explicit_maps_struct_fields(int a){ { p->a++; } // Region 17 -// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE17]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE17]]{{.+}}) +// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE17]]{{.+}}) // CK24-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK24-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK24-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK24-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SC]]** // CK24-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[SB]]*** // CK24-DAG: store [[SC]]* [[VAR0:%.+]], [[SC]]** [[CBP0]] // CK24-DAG: store [[SB]]** [[SEC0:%.+]], [[SB]]*** [[CP0]] +// CK24-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK24-DAG: [[SEC0]] = getelementptr {{.*}}[[SC]]* [[VAR00:%.+]], i{{.+}} 0, i{{.+}} 2 // CK24-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 // CK24-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 // CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SB]]*** // CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to [[SB]]** // CK24-DAG: store [[SB]]** [[SEC0]], [[SB]]*** [[CBP1]] // CK24-DAG: store [[SB]]* [[SEC1:%.+]], [[SB]]** [[CP1]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S1]] // CK24-DAG: [[SEC1]] = getelementptr {{.*}}[[SB]]* [[SEC11:%[^,]+]], i{{.+}} 0 // CK24-DAG: [[SEC11]] = load [[SB]]*, [[SB]]** [[SEC111:%[^,]+]], // CK24-DAG: [[SEC111]] = getelementptr {{.*}}[[SC]]* [[VAR000:%.+]], i{{.+}} 0, i{{.+}} 2 @@ -3558,21 +3969,33 @@ int explicit_maps_struct_fields(int a){ { p->a++; } // Region 18 -// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE18]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE18]]{{.+}}) +// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE18]]{{.+}}) // CK24-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK24-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK24-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK24-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SC]]** // CK24-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32** // CK24-DAG: store [[SC]]* [[VAR0:%.+]], [[SC]]** [[CBP0]] // CK24-DAG: store i32* [[SEC0:%.+]], i32** [[CP0]] +// CK24-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK24-DAG: [[SEC0]] = getelementptr {{.*}}[[SA]]* [[SEC00:%[^,]+]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[SEC00]] = getelementptr {{.*}}[10 x [[SA]]]* [[SEC000:%[^,]+]], i{{.+}} 0, i{{.+}} 3 // CK24-DAG: [[SEC000]] = getelementptr {{.*}}[[SB]]* [[SEC0000:%[^,]+]], i{{.+}} 0, i{{.+}} 2 // CK24-DAG: [[SEC0000]] = getelementptr {{.*}}[[SC]]* [[VAR00:%.+]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SC]]** +// CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** +// CK24-DAG: store [[SC]]* [[VAR0]], [[SC]]** [[CBP1]] +// CK24-DAG: store i32* [[SEC0]], i32** [[CP1]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S1]] + // CK24-DAG: [[VAR0]] = load [[SC]]*, [[SC]]** %{{.+}} // CK24-DAG: [[VAR00]] = load [[SC]]*, [[SC]]** %{{.+}} @@ -3581,26 +4004,38 @@ int explicit_maps_struct_fields(int a){ { p->a++; } // Region 19 -// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE19]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE19]]{{.+}}) +// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 3, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[3 x i{{.+}}]* [[MTYPE19]]{{.+}}) // CK24-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK24-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK24-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK24-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SC]]** // CK24-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[SA]]*** // CK24-DAG: store [[SC]]* [[VAR0:%.+]], [[SC]]** [[CBP0]] // CK24-DAG: store [[SA]]** [[SEC0:%.+]], [[SA]]*** [[CP0]] +// CK24-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK24-DAG: [[SEC0]] = getelementptr {{.*}}[10 x [[SA]]*]* [[SEC00:%[^,]+]], i{{.+}} 0, i{{.+}} 3 // CK24-DAG: [[SEC00]] = getelementptr {{.*}}[[SB]]* [[SEC000:%[^,]+]], i{{.+}} 0, i{{.+}} 3 // CK24-DAG: [[SEC000]] = getelementptr {{.*}}[[SC]]* [[VAR00:%.+]], i{{.+}} 0, i{{.+}} 1 // CK24-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 // CK24-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 -// CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SA]]*** -// CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** -// CK24-DAG: store [[SA]]** [[SEC0]], [[SA]]*** [[CBP1]] -// CK24-DAG: store i32* [[SEC1:%.+]], i32** [[CP1]] +// CK24-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SC]]** +// CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to [[SA]]*** +// CK24-DAG: store [[SC]]* [[VAR0]], [[SC]]** [[CBP1]] +// CK24-DAG: store [[SA]]** [[SEC0]], [[SA]]*** [[CP1]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S1]] + +// CK24-DAG: [[BP2:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 2 +// CK24-DAG: [[P2:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 2 +// CK24-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to [[SA]]*** +// CK24-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to i32** +// CK24-DAG: store [[SA]]** [[SEC0]], [[SA]]*** [[CBP2]] +// CK24-DAG: store i32* [[SEC1:%.+]], i32** [[CP2]] // CK24-DAG: [[SEC1]] = getelementptr {{.*}}[[SA]]* [[SEC11:%[^,]+]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[SEC11]] = load [[SA]]*, [[SA]]** [[SEC111:%[^,]+]], // CK24-DAG: [[SEC111]] = getelementptr {{.*}}[10 x [[SA]]*]* [[SEC1111:%[^,]+]], i{{.+}} 0, i{{.+}} 3 @@ -3616,16 +4051,19 @@ int explicit_maps_struct_fields(int a){ { p->a++; } // Region 20 -// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE20]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE20]]{{.+}}) +// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE20]]{{.+}}) // CK24-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK24-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK24-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK24-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SC]]** // CK24-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[SB]]*** // CK24-DAG: store [[SC]]* [[VAR0:%.+]], [[SC]]** [[CBP0]] // CK24-DAG: store [[SB]]** [[SEC0:%.+]], [[SB]]*** [[CP0]] +// CK24-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK24-DAG: [[SEC0]] = getelementptr {{.*}}[[SC]]* [[VAR00:%.+]], i{{.+}} 0, i{{.+}} 2 // CK24-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 @@ -3647,25 +4085,39 @@ int explicit_maps_struct_fields(int a){ { p->a++; } // Region 21 -// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE21]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE21]]{{.+}}) +// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 3, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[3 x i{{.+}}]* [[MTYPE21]]{{.+}}) // CK24-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK24-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK24-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK24-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SC]]** // CK24-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[SA]]*** // CK24-DAG: store [[SC]]* [[VAR0:%.+]], [[SC]]** [[CBP0]] // CK24-DAG: store [[SA]]** [[SEC0:%.+]], [[SA]]*** [[CP0]] +// CK24-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK24-DAG: [[SEC0]] = getelementptr {{.*}}[[SB]]* [[SEC00:[^,]+]], i{{.+}} 0, i{{.+}} 4 // CK24-DAG: [[SEC00]] = getelementptr {{.*}}[[SC]]* [[VAR00:%.+]], i{{.+}} 0, i{{.+}} 1 // CK24-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 // CK24-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 -// CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SA]]*** -// CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** -// CK24-DAG: store [[SA]]** [[SEC0]], [[SA]]*** [[CBP1]] -// CK24-DAG: store i32* [[SEC1:%.+]], i32** [[CP1]] +// CK24-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SC]]** +// CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to [[SA]]*** +// CK24-DAG: store [[SC]]* [[VAR0]], [[SC]]** [[CBP1]] +// CK24-DAG: store [[SA]]** [[SEC0]], [[SA]]*** [[CP1]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S1]] + +// CK24-DAG: [[BP2:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 2 +// CK24-DAG: [[P2:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 2 +// CK24-DAG: [[S2:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 2 +// CK24-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to [[SA]]*** +// CK24-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to i32** +// CK24-DAG: store [[SA]]** [[SEC0]], [[SA]]*** [[CBP2]] +// CK24-DAG: store i32* [[SEC1:%.+]], i32** [[CP2]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S2]] // CK24-DAG: [[SEC1]] = getelementptr {{.*}}[[SA]]* [[SEC11:%[^,]+]], i{{.+}} 0 // CK24-DAG: [[SEC11]] = load [[SA]]*, [[SA]]** [[SEC111:%[^,]+]], // CK24-DAG: [[SEC111]] = getelementptr {{.*}}[[SB]]* [[SEC1111:[^,]+]], i{{.+}} 0, i{{.+}} 4 @@ -3680,21 +4132,33 @@ int explicit_maps_struct_fields(int a){ { p->a++; } // Region 22 -// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE22]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE22]]{{.+}}) +// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE22]]{{.+}}) // CK24-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK24-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK24-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK24-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SC]]** // CK24-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32** // CK24-DAG: store [[SC]]* [[VAR0:%.+]], [[SC]]** [[CBP0]] // CK24-DAG: store i32* [[SEC0:%.+]], i32** [[CP0]] +// CK24-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK24-DAG: [[SEC0]] = getelementptr {{.*}}[10 x i32]* [[SEC00:%[^,]+]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[SEC00]] = getelementptr {{.*}}[[SA]]* [[SEC000:%[^,]+]], i{{.+}} 0, i{{.+}} 2 // CK24-DAG: [[SEC000]] = getelementptr {{.*}}[[SB]]* [[SEC0000:%[^,]+]], i{{.+}} 0, i{{.+}} 1 // CK24-DAG: [[SEC0000]] = getelementptr {{.*}}[[SC]]* [[VAR00:%.+]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SC]]** +// CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** +// CK24-DAG: store [[SC]]* [[VAR0]], [[SC]]** [[CBP1]] +// CK24-DAG: store i32* [[SEC0]], i32** [[CP1]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S1]] + // CK24-DAG: [[VAR0]] = load [[SC]]*, [[SC]]** %{{.+}} // CK24-DAG: [[VAR00]] = load [[SC]]*, [[SC]]** %{{.+}} @@ -3703,25 +4167,39 @@ int explicit_maps_struct_fields(int a){ { p->a++; } // Region 23 -// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE23]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE23]]{{.+}}) +// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 3, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[3 x i{{.+}}]* [[MTYPE23]]{{.+}}) // CK24-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK24-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK24-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK24-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SC]]** // CK24-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[SA]]*** // CK24-DAG: store [[SC]]* [[VAR0:%.+]], [[SC]]** [[CBP0]] // CK24-DAG: store [[SA]]** [[SEC0:%.+]], [[SA]]*** [[CP0]] +// CK24-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK24-DAG: [[SEC0]] = getelementptr {{.*}}[[SB]]* [[SEC00:%[^,]+]], i{{.+}} 0, i{{.+}} 4 // CK24-DAG: [[SEC00]] = getelementptr {{.*}}[[SC]]* [[VAR00:%.+]], i{{.+}} 0, i{{.+}} 1 // CK24-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 // CK24-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 -// CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SA]]*** -// CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** -// CK24-DAG: store [[SA]]** [[SEC0]], [[SA]]*** [[CBP1]] -// CK24-DAG: store i32* [[SEC1:%.+]], i32** [[CP1]] +// CK24-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SC]]** +// CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to [[SA]]*** +// CK24-DAG: store [[SC]]* [[VAR0]], [[SC]]** [[CBP1]] +// CK24-DAG: store [[SA]]** [[SEC0]], [[SA]]*** [[CP1]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S1]] + +// CK24-DAG: [[BP2:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 2 +// CK24-DAG: [[P2:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 2 +// CK24-DAG: [[S2:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 2 +// CK24-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to [[SA]]*** +// CK24-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to i32** +// CK24-DAG: store [[SA]]** [[SEC0]], [[SA]]*** [[CBP2]] +// CK24-DAG: store i32* [[SEC1:%.+]], i32** [[CP2]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S2]] // CK24-DAG: [[SEC1]] = getelementptr {{.*}}[10 x i32]* [[SEC11:%[^,]+]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[SEC11]] = getelementptr {{.*}}[[SA]]* [[SEC111:%[^,]+]], i{{.+}} 0, i{{.+}} 2 // CK24-DAG: [[SEC111]] = load [[SA]]*, [[SA]]** [[SEC1111:%[^,]+]], @@ -3737,34 +4215,41 @@ int explicit_maps_struct_fields(int a){ { p->a++; } // Region 24 -// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 4, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[4 x i{{.+}}]* [[SIZE24]], {{.+}}getelementptr {{.+}}[4 x i{{.+}}]* [[MTYPE24]]{{.+}}) +// CK24-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 4, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[4 x i{{.+}}]* [[MTYPE24]]{{.+}}) // CK24-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK24-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK24-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK24-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK24-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SC]]** // CK24-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[SB]]*** // CK24-DAG: store [[SC]]* [[VAR0:%.+]], [[SC]]** [[CBP0]] // CK24-DAG: store [[SB]]** [[SEC0:%.+]], [[SB]]*** [[CP0]] +// CK24-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK24-DAG: [[SEC0]] = getelementptr {{.*}}[[SC]]* [[VAR00:%.+]], i{{.+}} 0, i{{.+}} 2 // CK24-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 // CK24-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 +// CK24-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 // CK24-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SB]]*** // CK24-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to [[SA]]*** // CK24-DAG: store [[SB]]** [[SEC0]], [[SB]]*** [[CBP1]] // CK24-DAG: store [[SA]]** [[SEC1:%.+]], [[SA]]*** [[CP1]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S1]] // CK24-DAG: [[SEC1]] = getelementptr {{.*}}[[SB]]* [[SEC11:%[^,]+]], i{{.+}} 0, i{{.+}} 4 // CK24-DAG: [[SEC11]] = load [[SB]]*, [[SB]]** [[SEC111:%[^,]+]], // CK24-DAG: [[SEC111]] = getelementptr {{.*}}[[SC]]* [[VAR000:%.+]], i{{.+}} 0, i{{.+}} 2 // CK24-DAG: [[BP2:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 2 // CK24-DAG: [[P2:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 2 +// CK24-DAG: [[S2:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 2 // CK24-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to [[SA]]*** // CK24-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to [[SA]]*** // CK24-DAG: store [[SA]]** [[SEC1]], [[SA]]*** [[CBP2]] // CK24-DAG: store [[SA]]** [[SEC2:%.+]], [[SA]]*** [[CP2]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S2]] // CK24-DAG: [[SEC2]] = getelementptr {{.*}}[[SA]]* [[SEC22:%[^,]+]], i{{.+}} 0, i{{.+}} 1 // CK24-DAG: [[SEC22]] = load [[SA]]*, [[SA]]** [[SEC222:%[^,]+]], // CK24-DAG: [[SEC222]] = getelementptr {{.*}}[[SB]]* [[SEC2222:%[^,]+]], i{{.+}} 0, i{{.+}} 4 @@ -3773,10 +4258,12 @@ int explicit_maps_struct_fields(int a){ // CK24-DAG: [[BP3:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 3 // CK24-DAG: [[P3:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 3 +// CK24-DAG: [[S3:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 3 // CK24-DAG: [[CBP3:%.+]] = bitcast i8** [[BP3]] to [[SA]]*** // CK24-DAG: [[CP3:%.+]] = bitcast i8** [[P3]] to i32** // CK24-DAG: store [[SA]]** [[SEC2]], [[SA]]*** [[CBP3]] // CK24-DAG: store i32* [[SEC3:%.+]], i32** [[CP3]] +// CK24-DAG: store i[[sz]] {{.+}}, i[[sz]]* [[S3]] // CK24-DAG: [[SEC3]] = getelementptr {{.*}}[[SA]]* [[SEC33:%[^,]+]], i{{.+}} 0, i{{.+}} 0 // CK24-DAG: [[SEC33]] = load [[SA]]*, [[SA]]** [[SEC333:%[^,]+]], // CK24-DAG: [[SEC333]] = getelementptr {{.*}}[[SA]]* [[SEC3333:%[^,]+]], i{{.+}} 0, i{{.+}} 1 @@ -3813,24 +4300,33 @@ int explicit_maps_struct_fields(int a){ // CK24: define {{.+}}[[CALL24]] #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK25 -std=c++11 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK25 --check-prefix CK25-64 +// RUN: %clang_cc1 -DCK25 -std=c++11 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK25 --check-prefix CK25-64 // RUN: %clang_cc1 -DCK25 -std=c++11 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -std=c++11 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK25 --check-prefix CK25-64 -// RUN: %clang_cc1 -DCK25 -std=c++11 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK25 --check-prefix CK25-32 +// RUN: %clang_cc1 -fopenmp -std=c++11 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK25 --check-prefix CK25-64 +// RUN: %clang_cc1 -DCK25 -std=c++11 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK25 --check-prefix CK25-32 // RUN: %clang_cc1 -DCK25 -std=c++11 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -std=c++11 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK25 --check-prefix CK25-32 +// RUN: %clang_cc1 -fopenmp -std=c++11 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK25 --check-prefix CK25-32 + +// RUN: %clang_cc1 -DCK25 -std=c++11 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY24 %s +// RUN: %clang_cc1 -DCK25 -std=c++11 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY24 %s +// RUN: %clang_cc1 -DCK25 -std=c++11 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY24 %s +// RUN: %clang_cc1 -DCK25 -std=c++11 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY24 %s +// SIMD-ONLY24-NOT: {{__kmpc|__tgt}} #ifdef CK25 // CK25: [[ST:%.+]] = type { i32, float } // CK25: [[CA00:%.+]] = type { [[ST]]* } // CK25: [[CA01:%.+]] = type { i32* } -// CK25: [[SIZE00:@.+]] = private {{.*}}constant [1 x i[[Z:64|32]]] [i[[Z:64|32]] 4] -// CK25: [[MTYPE00:@.+]] = private {{.*}}constant [1 x i64] [i64 33] +// CK25-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK25: [[MTYPE00:@.+]] = private {{.*}}constant [2 x i64] [i64 32, i64 281474976710657] -// CK25: [[SIZE01:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] +// CK25-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK25: [[SIZE01:@.+]] = private {{.*}}constant [1 x i[[Z:64|32]]] [i[[Z:64|32]] 4] // CK25: [[MTYPE01:@.+]] = private {{.*}}constant [1 x i64] [i64 33] -// CK25-LABEL: explicit_maps_with_inner_lambda +// CK25-LABEL: explicit_maps_with_inner_lambda{{.*}}( template <int X, typename T> struct CC { @@ -3839,18 +4335,30 @@ struct CC { int foo(T arg) { // Region 00 - // CK25-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 1, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[SIZE00]], {{.+}}getelementptr {{.+}}[1 x i{{.+}}]* [[MTYPE00]]{{.+}}) + // CK25-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[Z]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}) // CK25-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK25-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] + // CK25-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK25-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK25-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 + // CK25-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK25-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]** // CK25-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to i32** // CK25-DAG: store [[ST]]* [[VAR0:%.+]], [[ST]]** [[CBP0]] // CK25-DAG: store i32* [[SEC0:%.+]], i32** [[CP0]] + // CK25-DAG: store i[[Z]] {{%.+}}, i[[Z]]* [[S0]] // CK25-DAG: [[SEC0]] = getelementptr {{.*}}[[ST]]* [[VAR0:%.+]], i{{.+}} 0, i{{.+}} 0 + // CK25-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 + // CK25-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 + // CK25-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 + // CK25-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[ST]]** + // CK25-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to i32** + // CK25-DAG: store [[ST]]* [[VAR0]], [[ST]]** [[CBP1]] + // CK25-DAG: store i32* [[SEC0]], i32** [[CP1]] + // CK25-DAG: store i[[Z]] {{.+}}, i[[Z]]* [[S1]] + // CK25: call void [[CALL00:@.+]]([[ST]]* {{[^,]+}}) #pragma omp target map(to:A) { @@ -3907,28 +4415,40 @@ int explicit_maps_with_inner_lambda(int a){ // CK25: define {{.+}}[[LAMBDA2]] #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK26 -std=c++11 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK26 --check-prefix CK26-64 +// RUN: %clang_cc1 -DCK26 -std=c++11 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK26 --check-prefix CK26-64 // RUN: %clang_cc1 -DCK26 -std=c++11 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -std=c++11 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK26 --check-prefix CK26-64 -// RUN: %clang_cc1 -DCK26 -std=c++11 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK26 --check-prefix CK26-32 +// RUN: %clang_cc1 -fopenmp -std=c++11 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK26 --check-prefix CK26-64 +// RUN: %clang_cc1 -DCK26 -std=c++11 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK26 --check-prefix CK26-32 // RUN: %clang_cc1 -DCK26 -std=c++11 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -std=c++11 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK26 --check-prefix CK26-32 +// RUN: %clang_cc1 -fopenmp -std=c++11 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK26 --check-prefix CK26-32 + +// RUN: %clang_cc1 -DCK26 -std=c++11 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY25 %s +// RUN: %clang_cc1 -DCK26 -std=c++11 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY25 %s +// RUN: %clang_cc1 -DCK26 -std=c++11 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY25 %s +// RUN: %clang_cc1 -DCK26 -std=c++11 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY25 %s +// SIMD-ONLY25-NOT: {{__kmpc|__tgt}} #ifdef CK26 // CK26: [[ST:%.+]] = type { i32, float*, i32, float* } +// CK26-LABEL: @.__omp_offloading_{{.*}}CC{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK26: [[SIZE00:@.+]] = private {{.*}}constant [2 x i[[Z:64|32]]] [i[[Z:64|32]] {{32|16}}, i[[Z:64|32]] 4] -// CK26: [[MTYPE00:@.+]] = private {{.*}}constant [2 x i64] [i64 35, i64 35] +// CK26: [[MTYPE00:@.+]] = private {{.*}}constant [2 x i64] [i64 547, i64 35] +// CK26-LABEL: @.__omp_offloading_{{.*}}CC{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK26: [[SIZE01:@.+]] = private {{.*}}constant [2 x i[[Z]]] [i[[Z]] {{32|16}}, i[[Z]] 4] -// CK26: [[MTYPE01:@.+]] = private {{.*}}constant [2 x i64] [i64 35, i64 35] +// CK26: [[MTYPE01:@.+]] = private {{.*}}constant [2 x i64] [i64 547, i64 35] +// CK26-LABEL: @.__omp_offloading_{{.*}}CC{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK26: [[SIZE02:@.+]] = private {{.*}}constant [2 x i[[Z]]] [i[[Z]] {{32|16}}, i[[Z]] 4] -// CK26: [[MTYPE02:@.+]] = private {{.*}}constant [2 x i64] [i64 35, i64 35] +// CK26: [[MTYPE02:@.+]] = private {{.*}}constant [2 x i64] [i64 547, i64 35] +// CK26-LABEL: @.__omp_offloading_{{.*}}CC{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK26: [[SIZE03:@.+]] = private {{.*}}constant [2 x i[[Z]]] [i[[Z]] {{32|16}}, i[[Z]] 4] -// CK26: [[MTYPE03:@.+]] = private {{.*}}constant [2 x i64] [i64 35, i64 35] +// CK26: [[MTYPE03:@.+]] = private {{.*}}constant [2 x i64] [i64 547, i64 35] -// CK26-LABEL: explicit_maps_with_private_class_members +// CK26-LABEL: explicit_maps_with_private_class_members{{.*}}( struct CC { int fA; @@ -4091,36 +4611,54 @@ int explicit_maps_with_private_class_members(){ } #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK27 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK27 --check-prefix CK27-64 +// RUN: %clang_cc1 -DCK27 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK27 --check-prefix CK27-64 // RUN: %clang_cc1 -DCK27 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK27 --check-prefix CK27-64 -// RUN: %clang_cc1 -DCK27 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK27 --check-prefix CK27-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK27 --check-prefix CK27-64 +// RUN: %clang_cc1 -DCK27 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK27 --check-prefix CK27-32 // RUN: %clang_cc1 -DCK27 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK27 --check-prefix CK27-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK27 --check-prefix CK27-32 + +// RUN: %clang_cc1 -DCK27 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY26 %s +// RUN: %clang_cc1 -DCK27 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY26 %s +// RUN: %clang_cc1 -DCK27 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY26 %s +// RUN: %clang_cc1 -DCK27 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY26 %s +// SIMD-ONLY26-NOT: {{__kmpc|__tgt}} #ifdef CK27 +// CK27-LABEL: @.__omp_offloading_{{.*}}zero_size_section_and_private_maps{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK27: [[SIZE00:@.+]] = private {{.*}}constant [1 x i[[Z:64|32]]] zeroinitializer -// CK27: [[MTYPE00:@.+]] = private {{.*}}constant [1 x i64] [i64 32] +// CK27: [[MTYPE00:@.+]] = private {{.*}}constant [1 x i64] [i64 544] +// CK27-LABEL: @.__omp_offloading_{{.*}}zero_size_section_and_private_maps{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK27: [[SIZE01:@.+]] = private {{.*}}constant [1 x i[[Z]]] zeroinitializer // CK27: [[MTYPE01:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK27-LABEL: @.__omp_offloading_{{.*}}zero_size_section_and_private_maps{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK27: [[SIZE02:@.+]] = private {{.*}}constant [1 x i[[Z]]] zeroinitializer // CK27: [[MTYPE02:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK27-LABEL: @.__omp_offloading_{{.*}}zero_size_section_and_private_maps{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK27: [[SIZE03:@.+]] = private {{.*}}constant [1 x i[[Z]]] zeroinitializer // CK27: [[MTYPE03:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK27-LABEL: @.__omp_offloading_{{.*}}zero_size_section_and_private_maps{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK27-LABEL: @.__omp_offloading_{{.*}}zero_size_section_and_private_maps{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK27: [[SIZE05:@.+]] = private {{.*}}constant [1 x i[[Z]]] zeroinitializer -// CK27: [[MTYPE05:@.+]] = private {{.*}}constant [1 x i64] [i64 32] +// CK27: [[MTYPE05:@.+]] = private {{.*}}constant [1 x i64] [i64 544] +// CK27-LABEL: @.__omp_offloading_{{.*}}zero_size_section_and_private_maps{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK27-LABEL: @.__omp_offloading_{{.*}}zero_size_section_and_private_maps{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK27: [[SIZE07:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 4] -// CK27: [[MTYPE07:@.+]] = private {{.*}}constant [1 x i64] [i64 288] +// CK27: [[MTYPE07:@.+]] = private {{.*}}constant [1 x i64] [i64 800] +// CK27-LABEL: @.__omp_offloading_{{.*}}zero_size_section_and_private_maps{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK27-LABEL: @.__omp_offloading_{{.*}}zero_size_section_and_private_maps{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK27: [[SIZE09:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 40] -// CK27: [[MTYPE09:@.+]] = private {{.*}}constant [1 x i64] [i64 161] +// CK27: [[MTYPE09:@.+]] = private {{.*}}constant [1 x i64] [i64 673] -// CK27-LABEL: zero_size_section_and_private_maps +// CK27-LABEL: zero_size_section_and_private_maps{{.*}}( void zero_size_section_and_private_maps (int ii){ // Map of a pointer. @@ -4302,21 +4840,31 @@ void zero_size_section_and_private_maps (int ii){ // CK27: define {{.+}}[[CALL07]] #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK28 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK28 --check-prefix CK28-64 +// RUN: %clang_cc1 -DCK28 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK28 --check-prefix CK28-64 // RUN: %clang_cc1 -DCK28 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK28 --check-prefix CK28-64 -// RUN: %clang_cc1 -DCK28 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK28 --check-prefix CK28-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK28 --check-prefix CK28-64 +// RUN: %clang_cc1 -DCK28 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK28 --check-prefix CK28-32 // RUN: %clang_cc1 -DCK28 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK28 --check-prefix CK28-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK28 --check-prefix CK28-32 + +// RUN: %clang_cc1 -DCK28 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY27 %s +// RUN: %clang_cc1 -DCK28 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY27 %s +// RUN: %clang_cc1 -DCK28 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY27 %s +// RUN: %clang_cc1 -DCK28 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY27 %s +// SIMD-ONLY27-NOT: {{__kmpc|__tgt}} #ifdef CK28 +// CK28-LABEL: @.__omp_offloading_{{.*}}explicit_maps_pointer_references{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK28: [[SIZE00:@.+]] = private {{.*}}constant [1 x i[[Z:64|32]]] [i[[Z:64|32]] {{8|4}}] // CK28: [[MTYPE00:@.+]] = private {{.*}}constant [1 x i64] [i64 35] +// CK28-LABEL: @.__omp_offloading_{{.*}}explicit_maps_pointer_references{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 // CK28: [[SIZE01:@.+]] = private {{.*}}constant [1 x i[[Z]]] [i[[Z]] 400] // CK28: [[MTYPE01:@.+]] = private {{.*}}constant [1 x i64] [i64 35] -// CK28-LABEL: explicit_maps_pointer_references +// CK28-LABEL: explicit_maps_pointer_references{{.*}}( void explicit_maps_pointer_references (int *p){ int *&a = p; @@ -4365,25 +4913,33 @@ void explicit_maps_pointer_references (int *p){ } #endif ///==========================================================================/// -// RUN: %clang_cc1 -DCK29 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK29 --check-prefix CK29-64 +// RUN: %clang_cc1 -DCK29 -verify -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK29 --check-prefix CK29-64 // RUN: %clang_cc1 -DCK29 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK29 --check-prefix CK29-64 -// RUN: %clang_cc1 -DCK29 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK29 --check-prefix CK29-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK29 --check-prefix CK29-64 +// RUN: %clang_cc1 -DCK29 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK29 --check-prefix CK29-32 // RUN: %clang_cc1 -DCK29 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK29 --check-prefix CK29-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CK29 --check-prefix CK29-32 + +// RUN: %clang_cc1 -DCK29 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY28 %s +// RUN: %clang_cc1 -DCK29 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY28 %s +// RUN: %clang_cc1 -DCK29 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY28 %s +// RUN: %clang_cc1 -DCK29 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY28 %s +// SIMD-ONLY28-NOT: {{__kmpc|__tgt}} #ifdef CK29 // CK29: [[SSA:%.+]] = type { double*, double** } // CK29: [[SSB:%.+]] = type { [[SSA]]*, [[SSA]]** } -// CK29: [[SIZE00:@.+]] = private {{.*}}constant [4 x i[[Z:64|32]]] [i[[Z:64|32]] {{8|4}}, i[[Z:64|32]] {{8|4}}, i[[Z:64|32]] {{8|4}}, i[[Z:64|32]] 80] -// CK29: [[MTYPE00:@.+]] = private {{.*}}constant [4 x i64] [i64 35, i64 16, i64 19, i64 19] +// CK29-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK29: [[MTYPE00:@.+]] = private {{.*}}constant [3 x i64] [i64 32, i64 281474976710672, i64 19] -// CK29: [[SIZE01:@.+]] = private {{.*}}constant [4 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] {{8|4}}, i[[Z]] {{8|4}}, i[[Z]] 80] -// CK29: [[MTYPE01:@.+]] = private {{.*}}constant [4 x i64] [i64 32, i64 19, i64 19, i64 19] +// CK29-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK29: [[MTYPE01:@.+]] = private {{.*}}constant [3 x i64] [i64 32, i64 281474976710672, i64 19] -// CK29: [[SIZE02:@.+]] = private {{.*}}constant [5 x i[[Z]]] [i[[Z]] {{8|4}}, i[[Z]] {{8|4}}, i[[Z]] {{8|4}}, i[[Z]] {{8|4}}, i[[Z]] 80] -// CK29: [[MTYPE02:@.+]] = private {{.*}}constant [5 x i64] [i64 32, i64 19, i64 16, i64 19, i64 19] +// CK29-LABEL: @.__omp_offloading_{{.*}}foo{{.*}}_l{{[0-9]+}}.region_id = weak constant i8 0 +// CK29: [[MTYPE02:@.+]] = private {{.*}}constant [3 x i64] [i64 32, i64 281474976710672, i64 19] struct SSA{ double *p; @@ -4400,45 +4956,44 @@ struct SSB{ void foo() { // Region 00 - // CK29-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 4, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[4 x i{{.+}}]* [[SIZE00]], {{.+}}getelementptr {{.+}}[4 x i{{.+}}]* [[MTYPE00]]{{.+}}) + // CK29-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 3, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[Z:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[3 x i{{.+}}]* [[MTYPE00]]{{.+}}) // CK29-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK29-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] + // CK29-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK29-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK29-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 + // CK29-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK29-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SSB]]** // CK29-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[SSA]]** // CK29-DAG: store [[SSB]]* [[VAR0:%.+]], [[SSB]]** [[CBP0]] // CK29-DAG: store [[SSA]]** [[VAR00:%.+]], [[SSA]]*** [[CP0]] + // CK29-DAG: store i[[Z]] %{{.+}}, i[[Z]]* [[S0]] // CK29-DAG: [[VAR0]] = load [[SSB]]*, [[SSB]]** % // CK29-DAG: [[VAR00]] = getelementptr inbounds [[SSB]], [[SSB]]* [[VAR0]], i32 0, i32 0 // CK29-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 // CK29-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 + // CK29-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 // CK29-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SSA]]*** - // CK29-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double**** + // CK29-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double*** // CK29-DAG: store [[SSA]]** [[VAR00]], [[SSA]]*** [[CBP1]] - // CK29-DAG: store double*** [[VAR1:%.+]], double**** [[CP1]] - // CK29-DAG: [[VAR1]] = getelementptr inbounds [[SSA]], [[SSA]]* %{{.+}}, i32 0, i32 1 + // CK29-DAG: store double** [[VAR1:%.+]], double*** [[CP1]] + // CK29-DAG: store i[[Z]] {{8|4}}, i[[Z]]* [[S1]] + // CK29-DAG: [[VAR1]] = load double**, double*** [[VAR1_REF:%.+]], + // CK29-DAG: [[VAR1_REF]] = getelementptr inbounds [[SSA]], [[SSA]]* %{{.+}}, i32 0, i32 1 // CK29-DAG: [[BP2:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 2 // CK29-DAG: [[P2:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 2 - // CK29-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to double**** - // CK29-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to double*** - // CK29-DAG: store double*** [[VAR1]], double**** [[CBP2]] - // CK29-DAG: store double** [[VAR2:%.+]], double*** [[CP2]] - // CK29-DAG: [[VAR2]] = load double**, double*** [[VAR22:%.+]], - // CK29-DAG: [[VAR22]] = getelementptr inbounds [[SSA]], [[SSA]]* %{{.+}}, i32 0, i32 1 - - // CK29-DAG: [[BP3:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 3 - // CK29-DAG: [[P3:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 3 - // CK29-DAG: [[CBP3:%.+]] = bitcast i8** [[BP3]] to double*** - // CK29-DAG: [[CP3:%.+]] = bitcast i8** [[P3]] to double** - // CK29-DAG: store double** [[VAR2]], double*** [[CBP3]] - // CK29-DAG: store double* [[VAR3:%.+]], double** [[CP3]] - // CK29-DAG: [[VAR3]] = getelementptr inbounds double, double* [[VAR33:%.+]], i{{.+}} 0 - // CK29-DAG: [[VAR33]] = load double*, double** %{{.+}}, + // CK29-DAG: [[S2:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 2 + // CK29-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to double*** + // CK29-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to double** + // CK29-DAG: store double** [[VAR1]], double*** [[CBP2]] + // CK29-DAG: store double* [[VAR2:%.+]], double** [[CP2]] + // CK29-DAG: store i[[Z]] 80, i[[Z]]* [[S2]] + // CK29-DAG: [[VAR2]] = getelementptr inbounds double, double* [[VAR22:%.+]], i{{.+}} 0 + // CK29-DAG: [[VAR22]] = load double*, double** %{{.+}}, // CK29: call void [[CALL00:@.+]]([[SSB]]* {{[^,]+}}) #pragma omp target map(p->pr[:10]) @@ -4447,43 +5002,43 @@ struct SSB{ } // Region 01 - // CK29-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 4, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[4 x i{{.+}}]* [[SIZE01]], {{.+}}getelementptr {{.+}}[4 x i{{.+}}]* [[MTYPE01]]{{.+}}) + // CK29-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 3, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[Z]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[3 x i{{.+}}]* [[MTYPE01]]{{.+}}) // CK29-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK29-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] + // CK29-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK29-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK29-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 + // CK29-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK29-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SSB]]** - // CK29-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[SSA]]**** - // CK29-DAG: store [[SSB]]* [[VAR0:%.+]], [[SSB]]** [[CBP0]] - // CK29-DAG: store [[SSA]]*** [[VAR00:%.+]], [[SSA]]**** [[CP0]] - // CK29-DAG: [[VAR00]] = getelementptr inbounds [[SSB]], [[SSB]]* [[VAR0]], i32 0, i32 1 + // CK29-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[SSA]]** + // CK29-DAG: store [[SSB]]* [[VAR0]], [[SSB]]** [[CBP0]] + // CK29-DAG: store [[SSA]]** [[VAR00:%.+]], [[SSA]]*** [[CP0]] + // CK29-DAG: store i[[Z]] %{{.+}}, i[[Z]]* [[S0]] + // CK29-DAG: [[VAR00]] = load [[SSA]]**, [[SSA]]*** [[VAR000:%.+]], + // CK29-DAG: [[VAR000]] = getelementptr inbounds [[SSB]], [[SSB]]* [[VAR0]], i32 0, i32 1 // CK29-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 // CK29-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 - // CK29-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SSA]]**** - // CK29-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to [[SSA]]*** - // CK29-DAG: store [[SSA]]*** [[VAR00]], [[SSA]]**** [[CBP1]] - // CK29-DAG: store [[SSA]]** [[VAR1:%.+]], [[SSA]]*** [[CP1]] - // CK29-DAG: [[VAR1]] = load [[SSA]]**, [[SSA]]*** [[VAR00]], + // CK29-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 + // CK29-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SSA]]*** + // CK29-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double*** + // CK29-DAG: store [[SSA]]** [[VAR00]], [[SSA]]*** [[CBP1]] + // CK29-DAG: store double** [[VAR1:%.+]], double*** [[CP1]] + // CK29-DAG: store i[[Z]] {{8|4}}, i[[Z]]* [[S1]] + // CK29-DAG: [[VAR1]] = getelementptr inbounds [[SSA]], [[SSA]]* %{{.+}}, i32 0, i32 0 // CK29-DAG: [[BP2:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 2 // CK29-DAG: [[P2:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 2 - // CK29-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to [[SSA]]*** - // CK29-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to double*** - // CK29-DAG: store [[SSA]]** [[VAR1]], [[SSA]]*** [[CBP2]] - // CK29-DAG: store double** [[VAR2:%.+]], double*** [[CP2]] - // CK29-DAG: [[VAR2]] = getelementptr inbounds [[SSA]], [[SSA]]* %{{.+}}, i32 0, i32 0 - - // CK29-DAG: [[BP3:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 3 - // CK29-DAG: [[P3:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 3 - // CK29-DAG: [[CBP3:%.+]] = bitcast i8** [[BP3]] to double*** - // CK29-DAG: [[CP3:%.+]] = bitcast i8** [[P3]] to double** - // CK29-DAG: store double** [[VAR2]], double*** [[CBP3]] - // CK29-DAG: store double* [[VAR3:%.+]], double** [[CP3]] - // CK29-DAG: [[VAR3]] = getelementptr inbounds double, double* [[VAR33:%.+]], i{{.+}} 0 - // CK29-DAG: [[VAR33]] = load double*, double** %{{.+}}, + // CK29-DAG: [[S2:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 2 + // CK29-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to double*** + // CK29-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to double** + // CK29-DAG: store double** [[VAR1]], double*** [[CBP2]] + // CK29-DAG: store double* [[VAR2:%.+]], double** [[CP2]] + // CK29-DAG: store i[[Z]] 80, i[[Z]]* [[S2]] + // CK29-DAG: [[VAR2]] = getelementptr inbounds double, double* [[VAR22:%.+]], i{{.+}} 0 + // CK29-DAG: [[VAR22]] = load double*, double** %{{.+}}, // CK29: call void [[CALL00:@.+]]([[SSB]]* {{[^,]+}}) #pragma omp target map(pr->p[:10]) @@ -4492,51 +5047,44 @@ struct SSB{ } // Region 02 - // CK29-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 5, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[5 x i{{.+}}]* [[SIZE02]], {{.+}}getelementptr {{.+}}[5 x i{{.+}}]* [[MTYPE02]]{{.+}}) + // CK29-DAG: call i32 @__tgt_target(i64 {{[^,]+}}, i8* {{[^,]+}}, i32 3, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[Z]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[3 x i{{.+}}]* [[MTYPE02]]{{.+}}) // CK29-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK29-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] + // CK29-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK29-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK29-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 + // CK29-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK29-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[SSB]]** - // CK29-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[SSA]]**** - // CK29-DAG: store [[SSB]]* [[VAR0:%.+]], [[SSB]]** [[CBP0]] - // CK29-DAG: store [[SSA]]*** [[VAR00:%.+]], [[SSA]]**** [[CP0]] - // CK29-DAG: [[VAR00]] = getelementptr inbounds [[SSB]], [[SSB]]* [[VAR0]], i32 0, i32 1 + // CK29-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to [[SSA]]** + // CK29-DAG: store [[SSB]]* [[VAR0]], [[SSB]]** [[CBP0]] + // CK29-DAG: store [[SSA]]** [[VAR00:%.+]], [[SSA]]*** [[CP0]] + // CK29-DAG: store i[[Z]] %{{.+}}, i[[Z]]* [[S0]] + // CK29-DAG: [[VAR00]] = load [[SSA]]**, [[SSA]]*** [[VAR000:%.+]], + // CK29-DAG: [[VAR000]] = getelementptr inbounds [[SSB]], [[SSB]]* [[VAR0]], i32 0, i32 1 // CK29-DAG: [[BP1:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 1 // CK29-DAG: [[P1:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 1 - // CK29-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SSA]]**** - // CK29-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to [[SSA]]*** - // CK29-DAG: store [[SSA]]*** [[VAR00]], [[SSA]]**** [[CBP1]] - // CK29-DAG: store [[SSA]]** [[VAR1:%.+]], [[SSA]]*** [[CP1]] - // CK29-DAG: [[VAR1]] = load [[SSA]]**, [[SSA]]*** [[VAR00]], + // CK29-DAG: [[S1:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 1 + // CK29-DAG: [[CBP1:%.+]] = bitcast i8** [[BP1]] to [[SSA]]*** + // CK29-DAG: [[CP1:%.+]] = bitcast i8** [[P1]] to double*** + // CK29-DAG: store [[SSA]]** [[VAR00]], [[SSA]]*** [[CBP1]] + // CK29-DAG: store double** [[VAR1:%.+]], double*** [[CP1]] + // CK29-DAG: store i[[Z]] {{8|4}}, i[[Z]]* [[S1]] + // CK29-DAG: [[VAR1]] = load double**, double*** [[VAR1_REF:%.+]], + // CK29-DAG: [[VAR1_REF]] = getelementptr inbounds [[SSA]], [[SSA]]* %{{.+}}, i32 0, i32 1 // CK29-DAG: [[BP2:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 2 // CK29-DAG: [[P2:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 2 - // CK29-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to [[SSA]]*** - // CK29-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to double**** - // CK29-DAG: store [[SSA]]** [[VAR1]], [[SSA]]*** [[CBP2]] - // CK29-DAG: store double*** [[VAR2:%.+]], double**** [[CP2]] - // CK29-DAG: [[VAR2]] = getelementptr inbounds [[SSA]], [[SSA]]* %{{.+}}, i32 0, i32 1 - - // CK29-DAG: [[BP3:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 3 - // CK29-DAG: [[P3:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 3 - // CK29-DAG: [[CBP3:%.+]] = bitcast i8** [[BP3]] to double**** - // CK29-DAG: [[CP3:%.+]] = bitcast i8** [[P3]] to double*** - // CK29-DAG: store double*** [[VAR2]], double**** [[CBP3]] - // CK29-DAG: store double** [[VAR3:%.+]], double*** [[CP3]] - // CK29-DAG: [[VAR3]] = load double**, double*** [[VAR2]], - - // CK29-DAG: [[BP4:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 4 - // CK29-DAG: [[P4:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 4 - // CK29-DAG: [[CBP4:%.+]] = bitcast i8** [[BP4]] to double*** - // CK29-DAG: [[CP4:%.+]] = bitcast i8** [[P4]] to double** - // CK29-DAG: store double** [[VAR3]], double*** [[CBP4]] - // CK29-DAG: store double* [[VAR4:%.+]], double** [[CP4]] - // CK29-DAG: [[VAR4]] = getelementptr inbounds double, double* [[VAR44:%.+]], i{{.+}} 0 - // CK29-DAG: [[VAR44]] = load double*, double** + // CK29-DAG: [[S2:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 2 + // CK29-DAG: [[CBP2:%.+]] = bitcast i8** [[BP2]] to double*** + // CK29-DAG: [[CP2:%.+]] = bitcast i8** [[P2]] to double** + // CK29-DAG: store double** [[VAR1]], double*** [[CBP2]] + // CK29-DAG: store double* [[VAR2:%.+]], double** [[CP2]] + // CK29-DAG: store i[[Z]] 80, i[[Z]]* [[S2]] + // CK29-DAG: [[VAR2]] = getelementptr inbounds double, double* [[VAR22:%.+]], i{{.+}} 0 + // CK29-DAG: [[VAR22]] = load double*, double** %{{.+}}, // CK29: call void [[CALL00:@.+]]([[SSB]]* {{[^,]+}}) #pragma omp target map(pr->pr[:10]) diff --git a/test/OpenMP/target_map_messages.cpp b/test/OpenMP/target_map_messages.cpp index 9ffbec4725728..03a81ee575b6f 100644 --- a/test/OpenMP/target_map_messages.cpp +++ b/test/OpenMP/target_map_messages.cpp @@ -1,5 +1,8 @@ -// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 200 %s -// RUN: %clang_cc1 -DCCODE -verify -fopenmp -ferror-limit 200 -x c %s +// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 200 %s -Wno-openmp-target +// RUN: %clang_cc1 -DCCODE -verify -fopenmp -ferror-limit 200 -x c %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 200 %s -Wno-openmp-target +// RUN: %clang_cc1 -DCCODE -verify -fopenmp-simd -ferror-limit 200 -x c %s -Wno-openmp-target #ifdef CCODE void foo(int arg) { const int n = 0; @@ -16,6 +19,13 @@ void foo(int arg) { {} } #else + +struct SREF { + int &a; + int b; + SREF(int &a) : a(a) {} +}; + template <typename T, int I> struct SA { static int ss; @@ -28,13 +38,14 @@ struct SA { T *f; int bf : 20; void func(int arg) { + SREF sref(arg); #pragma omp target { a = 0.0; func(arg); bf = 20; } - #pragma omp target map(arg,a,d) + #pragma omp target map(arg,a,d,sref.b) {} #pragma omp target map(arg[2:2],a,d) // expected-error {{subscripted value is not an array or pointer}} {} @@ -232,9 +243,17 @@ void SAclient(int arg) { {} #pragma omp target map(r.ArrS[0].A, t.ArrS[1].A) {} - #pragma omp target map(r.PtrS[0], r.PtrS->B) // expected-error {{same pointer derreferenced in multiple different ways in map clause expressions}} expected-note {{used here}} + #pragma omp target map(r.PtrS[0], r.PtrS->B) // expected-error {{same pointer dereferenced in multiple different ways in map clause expressions}} expected-note {{used here}} + {} + #pragma omp target map(r.PtrS, r.PtrS->B) // expected-error {{pointer cannot be mapped along with a section derived from itself}} expected-note {{used here}} + {} + #pragma omp target map(r.PtrS->A, r.PtrS->B) + {} + #pragma omp target map(r.RPtrS[0], r.RPtrS->B) // expected-error {{same pointer dereferenced in multiple different ways in map clause expressions}} expected-note {{used here}} + {} + #pragma omp target map(r.RPtrS, r.RPtrS->B) // expected-error {{pointer cannot be mapped along with a section derived from itself}} expected-note {{used here}} {} - #pragma omp target map(r.RPtrS[0], r.RPtrS->B) // expected-error {{same pointer derreferenced in multiple different ways in map clause expressions}} expected-note {{used here}} + #pragma omp target map(r.RPtrS->A, r.RPtrS->B) {} #pragma omp target map(r.S.Arr[:12]) {} diff --git a/test/OpenMP/target_messages.cpp b/test/OpenMP/target_messages.cpp index 77d808493789b..c9dc80aca44c8 100644 --- a/test/OpenMP/target_messages.cpp +++ b/test/OpenMP/target_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -o - %s // RUN: not %clang_cc1 -fopenmp -std=c++11 -fopenmp-targets=aaa-bbb-ccc-ddd -o - %s 2>&1 | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -o - %s // CHECK: error: OpenMP target is invalid: 'aaa-bbb-ccc-ddd' // RUN: not %clang_cc1 -fopenmp -std=c++11 -triple nvptx64-nvidia-cuda -o - %s 2>&1 | FileCheck --check-prefix CHECK-UNSUPPORTED-HOST-TARGET %s // RUN: not %clang_cc1 -fopenmp -std=c++11 -triple nvptx-nvidia-cuda -o - %s 2>&1 | FileCheck --check-prefix CHECK-UNSUPPORTED-HOST-TARGET %s @@ -7,6 +9,25 @@ // RUN: not %clang_cc1 -fopenmp -std=c++11 -fopenmp-targets=hexagon-linux-gnu -o - %s 2>&1 | FileCheck --check-prefix CHECK-UNSUPPORTED-DEVICE-TARGET %s // CHECK-UNSUPPORTED-DEVICE-TARGET: OpenMP target is invalid: 'hexagon-linux-gnu' +// RUN: not %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path 1111.bc -o - 2>&1 | FileCheck --check-prefix NO-HOST-BC %s +// NO-HOST-BC: The provided host compiler IR file '1111.bc' is required to generate code for OpenMP target regions but cannot be found. + +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc -DREGION_HOST +// RUN: not %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - -DREGION_DEVICE 2>&1 | FileCheck %s --check-prefix NO-REGION +// NO-REGION: Offloading entry for target region is incorrect: either the address or the ID is invalid. + +#if defined(REGION_HOST) || defined(REGION_DEVICE) +void foo() { +#ifdef REGION_HOST +#pragma omp target + ; +#endif +#ifdef REGION_DEVICE +#pragma omp target + ; +#endif +} +#else void foo() { } @@ -69,4 +90,4 @@ int main(int argc, char **argv) { return 0; } - +#endif diff --git a/test/OpenMP/target_nowait_messages.cpp b/test/OpenMP/target_nowait_messages.cpp index 7531c81c0f6ae..1f982dd674031 100644 --- a/test/OpenMP/target_nowait_messages.cpp +++ b/test/OpenMP/target_nowait_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_parallel_ast_print.cpp b/test/OpenMP/target_parallel_ast_print.cpp index 82c9203f030f0..1b61f15330443 100644 --- a/test/OpenMP/target_parallel_ast_print.cpp +++ b/test/OpenMP/target_parallel_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -70,7 +74,7 @@ T tmain(T argc, T *argv) { // CHECK-NEXT: S<T> s; // CHECK-NEXT: T arr[C][10], arr1[C]; // CHECK-NEXT: T i, j, a[20] -// CHECK-NEXT: #pragma omp target parallel +// CHECK-NEXT: #pragma omp target parallel{{$}} // CHECK-NEXT: h = 2; // CHECK-NEXT: #pragma omp target parallel default(none) private(argc,b) firstprivate(argv) shared(d) if(parallel: argc > 0) num_threads(C) proc_bind(master) reduction(+: c,arr1[argc]) reduction(max: e,arr[:C][0:10]) // CHECK-NEXT: foo() diff --git a/test/OpenMP/target_parallel_codegen.cpp b/test/OpenMP/target_parallel_codegen.cpp index b9d00da02f452..f1a511a6bb454 100644 --- a/test/OpenMP/target_parallel_codegen.cpp +++ b/test/OpenMP/target_parallel_codegen.cpp @@ -1,28 +1,46 @@ // Test host codegen. -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER -// CHECK-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CHECK-DAG: [[TT:%.+]] = type { i64, i8 } // CHECK-DAG: [[S1:%.+]] = type { double } @@ -34,36 +52,35 @@ // CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat -// We have 8 target regions, but only 7 that actually will generate offloading -// code, only 6 will have mapped arguments, and only 4 have all-constant map -// sizes. +// We have 8 target regions, but only 6 that actually will generate offloading +// code and have mapped arguments, and only 4 have all-constant map sizes. -// CHECK-DAG: [[SIZET2:@.+]] = private unnamed_addr constant [1 x i{{32|64}}] [i[[SZ:32|64]] 2] -// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [1 x i64] [i64 288] +// CHECK-DAG: [[SIZET2:@.+]] = private unnamed_addr constant [1 x i{{32|64}}] [i[[SZ]] 2] +// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [1 x i64] [i64 800] // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2] -// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 288, i64 288] -// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [9 x i64] [i64 288, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547] +// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 800] +// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [9 x i64] [i64 800, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547] // CHECK-DAG: [[SIZET5:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 40] -// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 547] +// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 547] // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [4 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 1, i[[SZ]] 40] -// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [4 x i64] [i64 288, i64 288, i64 288, i64 547] -// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [5 x i64] [i64 547, i64 288, i64 288, i64 288, i64 547] -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 - -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK-NOT: @{{.+}} = constant [[ENTTY]] +// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [4 x i64] [i64 800, i64 800, i64 800, i64 547] +// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [6 x i64] [i64 32, i64 281474976711171, i64 800, i64 288, i64 288, i64 547] +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] // Check if offloading descriptor is created. // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] @@ -74,7 +91,7 @@ // CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) // Check target registration is registered as a Ctor. -// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] template<typename tx, typename ty> @@ -179,13 +196,13 @@ int foo(int n) { // CHECK-32: store i32 [[A_VAL]], i32* [[A_CADDR:%.+]], // CHECK-32: [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]], + // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 + // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] + // CHECK: [[TRY]] // CHECK: [[BNSIZE:%.+]] = mul nuw i[[SZ]] [[VLA0:%.+]], 4 // CHECK: [[CNELEMSIZE2:%.+]] = mul nuw i[[SZ]] 5, [[VLA1:%.+]] // CHECK: [[CNSIZE:%.+]] = mul nuw i[[SZ]] [[CNELEMSIZE2]], 8 - // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 - // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] - // CHECK: [[TRY]] // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 9, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([9 x i64], [9 x i64]* [[MAPT4]], i32 0, i32 0), i32 1, i32 0) // CHECK-DAG: [[BPR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 // CHECK-DAG: [[PR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P:%[^,]+]], i32 0, i32 0 @@ -300,7 +317,7 @@ int foo(int n) { // correct and loaded correctly for the target regions in foo(). // CHECK: define internal void [[HVT0]]() -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*)) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*)) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED]](i32* noalias %.global_tid., i32* noalias %.bound_tid.) @@ -320,7 +337,7 @@ int foo(int n) { // CHECK-64: store i32 [[AA]], i32* [[AA_C]], align // CHECK-32: store i32 [[AA]], i32* [[AA_CASTED]], align // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED1]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) @@ -342,7 +359,7 @@ int foo(int n) { // CHECK: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i16* // CHECK: store i16 [[AA]], i16* [[AA_C]], align // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED2]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) @@ -373,7 +390,7 @@ int foo(int n) { // CHECK-DAG: store i16 [[AA]], i16* [[AA_C]], align // CHECK-DAG: [[PARAM1:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CASTED]], align // CHECK-DAG: [[PARAM2:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK-DAG: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]]) +// CHECK-DAG: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED3]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}) @@ -425,7 +442,7 @@ int foo(int n) { // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 9, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], [10 x float]*, i[[SZ]], float*, [5 x [10 x double]]*, i[[SZ]], i[[SZ]], double*, [[TT]]*)* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], [10 x float]* [[REF_B]], i[[SZ]] [[VAL_VLA1]], float* [[REF_BN]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] [[VAL_VLA2]], i[[SZ]] [[VAL_VLA3]], double* [[REF_CN]], [[TT]]* [[REF_D]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 9, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], [10 x float]*, i[[SZ]], float*, [5 x [10 x double]]*, i[[SZ]], i[[SZ]], double*, [[TT]]*)* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], [10 x float]* [[REF_B]], i[[SZ]] [[VAL_VLA1]], float* [[REF_BN]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] [[VAL_VLA2]], i[[SZ]] [[VAL_VLA3]], double* [[REF_CN]], [[TT]]* [[REF_D]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED4]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, [10 x float]* {{.+}}, i[[SZ]] %{{.+}}, float* {{.+}}, [5 x [10 x double]]* {{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, double* {{.+}}, [[TT]]* {{.+}}) @@ -510,32 +527,36 @@ int bar(int n){ // CHECK-64: store i32 %{{.+}}, i32* [[B_ADDR]], // CHECK-64: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_CADDR]], +// CHECK-32: store i32 %{{.+}}, i32* %__vla_expr // CHECK-32: store i32 %{{.+}}, i32* [[B_ADDR:%.+]], // CHECK-32: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_ADDR]], +// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 +// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] +// CHECK: [[TRY]] // We capture 2 VLA sizes in this target region // CHECK: [[CELEMSIZE2:%.+]] = mul nuw i[[SZ]] 2, [[VLA0:%.+]] // CHECK: [[CSIZE:%.+]] = mul nuw i[[SZ]] [[CELEMSIZE2]], 2 -// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 -// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] -// CHECK: [[TRY]] -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 5, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([5 x i64], [5 x i64]* [[MAPT7]], i32 0, i32 0), i32 1, i32 0) -// CHECK-DAG: [[BPR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP:%.+]], i32 0, i32 0 -// CHECK-DAG: [[PR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SR]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] -// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX0]] -// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX0]] -// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] -// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX1]] -// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX1]] -// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] -// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX2]] -// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX2]] -// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] -// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX3]] -// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 6, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([6 x i64], [6 x i64]* [[MAPT7]], i32 0, i32 0), i32 1, i32 0) +// CHECK-DAG: [[BPR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP:%.+]], i32 0, i32 0 +// CHECK-DAG: [[PR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SR]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] +// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX0]] +// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX0]] +// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] +// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX1]] +// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX1]] +// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] +// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX2]] +// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX2]] +// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] +// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] +// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX4:[0-9]+]] +// CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX4]] +// CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX4]] // The names below are not necessarily consistent with the names used for the // addresses above as some are repeated. @@ -561,12 +582,18 @@ int bar(int n){ // CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CPADDR3:%.+]], // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** // CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** +// CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* {{%[^,]+}} + +// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR4:%.+]], +// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CPADDR4:%.+]], +// CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to [[S1]]** +// CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to [[S1]]** // CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* {{%[^,]+}} -// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR4:%.+]], -// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR4:%.+]], -// CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to i16** -// CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to i16** +// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR5:%.+]], +// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR5:%.+]], +// CHECK-DAG: [[CBPADDR5]] = bitcast i8** {{%[^,]+}} to i16** +// CHECK-DAG: [[CPADDR5]] = bitcast i8** {{%[^,]+}} to i16** // CHECK-DAG: store i[[SZ]] [[CSIZE]], i[[SZ]]* {{%[^,]+}} // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 @@ -700,7 +727,7 @@ int bar(int n){ // CHECK-32-DAG:store i32 [[LOCAL_BV]], i32* [[LOCAL_B_CASTED]], align // CHECK-DAG: [[REF_B:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_B_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]], i[[SZ]], i[[SZ]], i16*)* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), [[S1]]* [[REF_THIS]], i[[SZ]] [[REF_B]], i[[SZ]] [[VAL_VLA1]], i[[SZ]] [[VAL_VLA2]], i16* [[REF_C]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]], i[[SZ]], i[[SZ]], i16*)* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), [[S1]]* [[REF_THIS]], i[[SZ]] [[REF_B]], i[[SZ]] [[VAL_VLA1]], i[[SZ]] [[VAL_VLA2]], i16* [[REF_C]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED5]](i32* noalias %.global_tid., i32* noalias %.bound_tid., [[S1]]* %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i16* {{.+}}) @@ -743,7 +770,7 @@ int bar(int n){ // CHECK-DAG: store i8 [[CONV_AAA]], i8* [[CONV]], align // CHECK-DAG: [[REF_AAA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AAA_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED6:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], i[[SZ]] [[REF_AAA]], [10 x i32]* [[REF_B]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED6:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], i[[SZ]] [[REF_AAA]], [10 x i32]* [[REF_B]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED6]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) @@ -776,7 +803,7 @@ int bar(int n){ // CHECK-DAG: store i16 [[CONV_AA]], i16* [[CONV]], align // CHECK-DAG: [[REF_AA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AA_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED7:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], [10 x i32]* [[REF_B]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED7:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], [10 x i32]* [[REF_B]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED7]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) diff --git a/test/OpenMP/target_parallel_codegen_registration.cpp b/test/OpenMP/target_parallel_codegen_registration.cpp index d2cb1ec6f7783..c9892eae36d63 100644 --- a/test/OpenMP/target_parallel_codegen_registration.cpp +++ b/test/OpenMP/target_parallel_codegen_registration.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target parallel codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,9 +24,22 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK -// Check that no target code is emmitted if no omptests flag was provided. +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// Check that no target code is emitted if no omptests flag was provided. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK-NTARGET +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -60,96 +81,96 @@ // We have 7 target regions -// CHECK-DAG: {{@.+}} = private constant i8 0 -// TCHECK-NOT: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// TCHECK-NOT: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] -// CHECK-NTARGET-NOT: private constant i8 0 +// CHECK-NTARGET-NOT: weak constant i8 0 // CHECK-NTARGET-NOT: private unnamed_addr constant [1 x i // CHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// CHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// CHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// CHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// CHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// CHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// CHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// CHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// CHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// CHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// CHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// CHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// CHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// TCHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// TCHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// TCHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// TCHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// TCHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// TCHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// TCHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// TCHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// TCHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// TCHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// TCHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// TCHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] // CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] @@ -163,7 +184,7 @@ // CHECK-SAME: { i32, void ()*, i8* } { i32 500, void ()* [[P500:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 501, void ()* [[P501:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 65535, void ()* [[PMAX:@[^,]+]], i8* null }, -// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] // CHECK-NTARGET: @llvm.global_ctors = appending global [3 x { i32, void ()*, i8* }] [ @@ -319,18 +340,18 @@ struct ST { //CHECK-DAG: define internal void @[[NAME12]]( //CHECK-DAG: call void @[[NAME12]]( -//TCHECK-DAG: define void @[[NAME1]]( -//TCHECK-DAG: define void @[[NAME2]]( -//TCHECK-DAG: define void @[[NAME3]]( -//TCHECK-DAG: define void @[[NAME4]]( -//TCHECK-DAG: define void @[[NAME5]]( -//TCHECK-DAG: define void @[[NAME6]]( -//TCHECK-DAG: define void @[[NAME7]]( -//TCHECK-DAG: define void @[[NAME8]]( -//TCHECK-DAG: define void @[[NAME9]]( -//TCHECK-DAG: define void @[[NAME10]]( -//TCHECK-DAG: define void @[[NAME11]]( -//TCHECK-DAG: define void @[[NAME12]]( +//TCHECK-DAG: define weak void @[[NAME1]]( +//TCHECK-DAG: define weak void @[[NAME2]]( +//TCHECK-DAG: define weak void @[[NAME3]]( +//TCHECK-DAG: define weak void @[[NAME4]]( +//TCHECK-DAG: define weak void @[[NAME5]]( +//TCHECK-DAG: define weak void @[[NAME6]]( +//TCHECK-DAG: define weak void @[[NAME7]]( +//TCHECK-DAG: define weak void @[[NAME8]]( +//TCHECK-DAG: define weak void @[[NAME9]]( +//TCHECK-DAG: define weak void @[[NAME10]]( +//TCHECK-DAG: define weak void @[[NAME11]]( +//TCHECK-DAG: define weak void @[[NAME12]]( // CHECK-NTARGET-NOT: __tgt_target // CHECK-NTARGET-NOT: __tgt_register_lib @@ -372,7 +393,7 @@ struct ST { //CHECK: ret void //CHECK: declare i32 @__tgt_unregister_lib([[DSCTY]]*) -//CHECK: define linkonce hidden void @[[REGFN]](i8*) +//CHECK: define linkonce hidden void @[[REGFN]]() //CHECK-SAME: comdat { //CHECK: call i32 @__tgt_register_lib([[DSCTY]]* [[DESC]]) //CHECK: call i32 @__cxa_atexit(void (i8*)* @[[UNREGFN]], i8* bitcast ([[DSCTY]]* [[DESC]] to i8*), @@ -411,31 +432,31 @@ int bar(int a){ // Check metadata is properly generated: // CHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 245, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 261, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 267, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 278, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 406, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 290, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 290, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 278, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 220, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 266, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 282, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 288, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 299, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 427, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 311, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 311, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 299, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 241, i32 {{[0-9]+}}} // TCHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 245, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 261, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 267, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 278, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 406, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 290, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 290, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 278, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 220, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 266, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 282, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 288, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 299, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 427, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 311, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 311, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 299, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 241, i32 {{[0-9]+}}} #endif diff --git a/test/OpenMP/target_parallel_codegen_registration_naming.cpp b/test/OpenMP/target_parallel_codegen_registration_naming.cpp index 4b7940290d07e..d2b88dc0729eb 100644 --- a/test/OpenMP/target_parallel_codegen_registration_naming.cpp +++ b/test/OpenMP/target_parallel_codegen_registration_naming.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target parallel codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,6 +24,16 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/target_parallel_debug_codegen.cpp b/test/OpenMP/target_parallel_debug_codegen.cpp index 385f7a29cb8fd..ebd761fb9699f 100644 --- a/test/OpenMP/target_parallel_debug_codegen.cpp +++ b/test/OpenMP/target_parallel_debug_codegen.cpp @@ -1,5 +1,5 @@ -// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc -// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - -debug-info-kind=limited | FileCheck %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -fopenmp-cuda-mode -emit-llvm-bc %s -o %t-ppc-host.bc -fopenmp-version=45 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -fopenmp-cuda-mode -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - -debug-info-kind=limited -fopenmp-version=45 | FileCheck %s // expected-no-diagnostics int main() { @@ -11,7 +11,7 @@ int main() { int c[10][10][10]; #pragma omp target parallel firstprivate(a, b) map(tofrom \ : c) map(tofrom \ - : bb) if (a) + : bb) if (target:a) { int &f = c[1][1][1]; int &g = a; @@ -54,7 +54,7 @@ int main() { return 0; } -// CHECK: define internal void @__omp_offloading{{[^(]+}}([10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 {{[^,]+}}, [10 x [10 x i32]]* {{[^,]+}}, i8 addrspace(1)* noalias{{[^,]+}}, i1 {{[^)]+}}) +// CHECK: define internal void @__omp_offloading{{[^(]+}}([10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 {{[^,]+}}, [10 x [10 x i32]]* {{[^,]+}}, i8 addrspace(1)* noalias{{[^,]+}}) // CHECK: addrspacecast [10 x [10 x [10 x i32]]] addrspace(1)* %{{.+}} to [10 x [10 x [10 x i32]]]* // CHECK: call void [[NONDEBUG_WRAPPER:.+]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]]* {{[^,]+}}, i64 {{[^,]+}}, [10 x [10 x i32]]* {{[^,]+}}, i8* {{[^)]+}}) @@ -65,7 +65,7 @@ int main() { // CHECK: addrspacecast [10 x [10 x [10 x i32]]]* %{{.+}} to [10 x [10 x [10 x i32]]] addrspace(1)* // CHECK: call void [[DEBUG_PARALLEL]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 {{[^,]+}}, [10 x [10 x i32]]* {{[^,]+}}, i8 addrspace(1)* {{[^)]+}}) -// CHECK: define void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i64 {{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) +// CHECK: define weak void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i64 {{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) // CHECK: addrspacecast [10 x [10 x [10 x i32]]]* %{{.+}} to [10 x [10 x [10 x i32]]] addrspace(1)* // CHECK: call void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 {{[^,]+}}, [10 x [10 x i32]]* {{[^,]+}}, i8 addrspace(1)* {{[^)]+}}) @@ -83,7 +83,7 @@ int main() { // CHECK: addrspacecast [10 x [10 x i32]]* %{{.+}} to [10 x [10 x i32]] addrspace(1)* // CHECK: call void [[DEBUG_PARALLEL]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 {{[^,]+}}, [10 x [10 x i32]] addrspace(1)* {{[^,]+}}, i8 addrspace(1)* {{[^)]+}}) -// CHECK: define void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i64 {{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) +// CHECK: define weak void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i64 {{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) // CHECK: addrspacecast [10 x [10 x [10 x i32]]]* %{{.+}} to [10 x [10 x [10 x i32]]] addrspace(1)* // CHECK: addrspacecast [10 x [10 x i32]]* %{{.+}} to [10 x [10 x i32]] addrspace(1)* // CHECK: call void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 {{[^,]+}}, [10 x [10 x i32]] addrspace(1)* {{[^,]+}}, i8 addrspace(1)* {{[^)]+}}) @@ -92,20 +92,20 @@ int main() { // CHECK: addrspacecast [10 x [10 x [10 x i32]]] addrspace(1)* %{{.+}} to [10 x [10 x [10 x i32]]]* // CHECK: addrspacecast i32 addrspace(1)* %{{.+}} to i32* // CHECK: addrspacecast [10 x [10 x i32]] addrspace(1)* %{{.+}} to [10 x [10 x i32]]* -// CHECK: call void [[NONDEBUG_WRAPPER:@.+]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]]* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x i32]]* {{[^,]+}}, i8* {{[^)]+}}) +// CHECK: call void @[[NONDEBUG_WRAPPER:.+]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]]* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x i32]]* {{[^,]+}}, i8* {{[^)]+}}) -// CHECK: define internal void [[DEBUG_PARALLEL:@.+]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]] addrspace(1)* noalias{{[^,]+}}, i32 addrspace(1)* noalias{{[^,]+}}, [10 x [10 x i32]] addrspace(1)* noalias{{[^,]+}}, i8 addrspace(1)* noalias{{[^)]+}}) +// CHECK: define internal void @[[DEBUG_PARALLEL:.+]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]] addrspace(1)* noalias{{[^,]+}}, i32 addrspace(1)* noalias{{[^,]+}}, [10 x [10 x i32]] addrspace(1)* noalias{{[^,]+}}, i8 addrspace(1)* noalias{{[^)]+}}) // CHECK: addrspacecast [10 x [10 x [10 x i32]]] addrspace(1)* %{{.+}} to [10 x [10 x [10 x i32]]]* // CHECK: addrspacecast i32 addrspace(1)* %{{.+}} to i32* // CHECK: addrspacecast [10 x [10 x i32]] addrspace(1)* %{{.+}} to [10 x [10 x i32]]* -// CHECK: define internal void [[NONDEBUG_WRAPPER]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i32* dereferenceable{{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) +// CHECK: define internal void @[[NONDEBUG_WRAPPER]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i32* dereferenceable{{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) // CHECK: addrspacecast [10 x [10 x [10 x i32]]]* %{{.+}} to [10 x [10 x [10 x i32]]] addrspace(1)* // CHECK: addrspacecast i32* %{{.+}} to i32 addrspace(1)* // CHECK: addrspacecast [10 x [10 x i32]]* %{{.+}} to [10 x [10 x i32]] addrspace(1)* -// CHECK: call void [[DEBUG_PARALLEL]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 addrspace(1)* {{[^,]+}}, [10 x [10 x i32]] addrspace(1)* {{[^,]+}}, i8 addrspace(1)* {{[^)]+}}) +// CHECK: call void @[[DEBUG_PARALLEL]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 addrspace(1)* {{[^,]+}}, [10 x [10 x i32]] addrspace(1)* {{[^,]+}}, i8 addrspace(1)* {{[^)]+}}) -// CHECK: define void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i32* dereferenceable{{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) +// CHECK: define weak void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i32* dereferenceable{{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) // CHECK: addrspacecast [10 x [10 x [10 x i32]]]* %{{.+}} to [10 x [10 x [10 x i32]]] addrspace(1)* // CHECK: addrspacecast i32* %{{.+}} to i32 addrspace(1)* // CHECK: addrspacecast [10 x [10 x i32]]* %{{.+}} to [10 x [10 x i32]] addrspace(1)* @@ -116,8 +116,11 @@ int main() { // CHECK: !DILocalVariable(name: ".bound_tid.", // CHECK-SAME: DIFlagArtificial // CHECK: !DILocalVariable(name: "c", -// CHECK-SAMEi-NOT: DIFlagArtificial +// CHECK-SAME: line: 11 // CHECK: !DILocalVariable(name: "a", -// CHECK-SAMEi-NOT: DIFlagArtificial +// CHECK-SAME: line: 9 // CHECK: !DILocalVariable(name: "b", -// CHECK-SAMEi-NOT: DIFlagArtificial +// CHECK-SAME: line: 10 + +// CHECK-DAG: distinct !DISubprogram(name: "[[NONDEBUG_WRAPPER]]", +// CHECK-DAG: distinct !DISubprogram(name: "[[DEBUG_PARALLEL]]", diff --git a/test/OpenMP/target_parallel_default_messages.cpp b/test/OpenMP/target_parallel_default_messages.cpp index cf497937f3e0b..143a94c553268 100644 --- a/test/OpenMP/target_parallel_default_messages.cpp +++ b/test/OpenMP/target_parallel_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/target_parallel_defaultmap_messages.cpp b/test/OpenMP/target_parallel_defaultmap_messages.cpp index 49e7c30a7d15f..33078a2ea7e8f 100644 --- a/test/OpenMP/target_parallel_defaultmap_messages.cpp +++ b/test/OpenMP/target_parallel_defaultmap_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_parallel_depend_codegen.cpp b/test/OpenMP/target_parallel_depend_codegen.cpp new file mode 100644 index 0000000000000..52d13a34d85af --- /dev/null +++ b/test/OpenMP/target_parallel_depend_codegen.cpp @@ -0,0 +1,261 @@ +// Test host codegen. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// CHECK-DAG: [[TT:%.+]] = type { i64, i8 } +// CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } +// CHECK-DAG: [[DEVTY:%.+]] = type { i8*, i8*, [[ENTTY]]*, [[ENTTY]]* } +// CHECK-DAG: [[DSCTY:%.+]] = type { i32, [[DEVTY]]*, [[ENTTY]]*, [[ENTTY]]* } + +// TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 } + +// CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat + +// CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 0, i[[SZ]] 4] +// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 800] +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] + +// Check if offloading descriptor is created. +// CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] +// CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] +// CHECK: [[DEVBEGIN:@.+]] = external constant i8 +// CHECK: [[DEVEND:@.+]] = external constant i8 +// CHECK: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x [[DEVTY]]] [{{.+}} { i8* [[DEVBEGIN]], i8* [[DEVEND]], [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }], comdat($[[REGFN]]) +// CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) + +// Check target registration is registered as a Ctor. +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] + + +template<typename tx, typename ty> +struct TT{ + tx X; + ty Y; +}; + +int global; +extern int global; + +// CHECK: define {{.*}}[[FOO:@.+]]( +int foo(int n) { + int a = 0; + short aa = 0; + float b[10]; + float bn[n]; + double c[5][10]; + double cn[5][n]; + TT<long long, char> d; + static long *plocal; + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID:%.+]], i32 1, i[[SZ]] {{20|40}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY0:%.+]]* + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 3 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* @0, i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + #pragma omp target parallel device(global + a) depend(in: global) depend(out: a, b, cn[4]) + { + } + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + + // CHECK: [[BOOL:%.+]] = icmp ne i32 %{{.+}}, 0 + // CHECK: br i1 [[BOOL]], label %[[THEN:.+]], label %[[ELSE:.+]] + // CHECK: [[THEN]]: + // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%.+]], i32 0, i32 0 + // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%.+]], i32 0, i32 0 + // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]** + // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]** + // CHECK-DAG: store i[[SZ]]* [[BP0:%[^,]+]], i[[SZ]]** [[CBPADDR0]] + // CHECK-DAG: store i[[SZ]]* [[BP0]], i[[SZ]]** [[CPADDR0]] + + // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1 + // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1 + // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* + // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* + // CHECK-DAG: store i[[SZ]] [[BP1:%[^,]+]], i[[SZ]]* [[CBPADDR1]] + // CHECK-DAG: store i[[SZ]] [[BP1]], i[[SZ]]* [[CPADDR1]] + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID]], i32 1, i[[SZ]] {{104|52}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1_:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1_:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + + // CHECK: [[ELSE]]: + // CHECK-NOT: getelementptr inbounds [2 x i8*], [2 x i8*]* + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1__:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1__:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + // CHECK: [[EXIT]]: + + #pragma omp target parallel device(global + a) nowait depend(inout: global, a, bn) if(target:a) + { + static int local1; + *plocal = global; + local1 = global; + } + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]* + // CHECK: getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* @0, i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + #pragma omp target parallel if(0) firstprivate(global) depend(out:global) + { + global += 1; + } + + return a; +} + +// Check that the offloading functions are emitted and that the arguments are +// correct and loaded correctly for the target regions in foo(). + +// CHECK: define internal void [[HVT0:@.+]]() + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, [[TASK_TY0]]* noalias) +// CHECK: store void (i8*, ...)* null, void (i8*, ...)** % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 0, i8** null, i8** null, i[[SZ]]* null, i64* null, i32 1, i32 0) +// CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: call void [[HVT0]]() +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT1:@.+]](i[[SZ]]* %{{.+}}, i[[SZ]] %{{.+}}) + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1_]](i32{{.*}}, [[TASK_TY1_]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[SZT:%.+]] = getelementptr inbounds [2 x i[[SZ]]], [2 x i[[SZ]]]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams_nowait(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 2, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SZT]], i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT]], i32 0, i32 0), i32 1, i32 0) + +// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1__]](i32{{.*}}, [[TASK_TY1__]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT2:@.+]](i[[SZ]] %{{.+}}) +// Create stack storage and store argument in there. +// CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align +// CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align +// CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* +// CHECK-64: load i32, i32* [[AA_CADDR]], align +// CHECK-32: load i32, i32* [[AA_ADDR]], align + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, [[TASK_TY2]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT2]](i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + + +#endif diff --git a/test/OpenMP/target_parallel_depend_messages.cpp b/test/OpenMP/target_parallel_depend_messages.cpp index 24c69e95cc0a8..746c2237066e4 100644 --- a/test/OpenMP/target_parallel_depend_messages.cpp +++ b/test/OpenMP/target_parallel_depend_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_device_messages.cpp b/test/OpenMP/target_parallel_device_messages.cpp index 6c8d4c2e43ab2..cc2525375316e 100644 --- a/test/OpenMP/target_parallel_device_messages.cpp +++ b/test/OpenMP/target_parallel_device_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_parallel_firstprivate_messages.cpp b/test/OpenMP/target_parallel_firstprivate_messages.cpp index dd6825aa251f9..02186e737b756 100644 --- a/test/OpenMP/target_parallel_firstprivate_messages.cpp +++ b/test/OpenMP/target_parallel_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_ast_print.cpp b/test/OpenMP/target_parallel_for_ast_print.cpp index d5aa055842a7b..c6c46c0031b13 100644 --- a/test/OpenMP/target_parallel_for_ast_print.cpp +++ b/test/OpenMP/target_parallel_for_ast_print.cpp @@ -1,6 +1,10 @@ -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s -Wno-openmp-target | FileCheck %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print -Wno-openmp-target | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s -Wno-openmp-target | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print -Wno-openmp-target | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -35,7 +39,7 @@ public: } }; -// CHECK: #pragma omp target parallel for private(this->a) private(this->a) private(T::a) +// CHECK: #pragma omp target parallel for private(this->a) private(this->a) private(T::a){{$}} // CHECK: #pragma omp target parallel for private(this->a) private(this->a) // CHECK: #pragma omp target parallel for private(this->a) private(this->a) private(this->S::a) diff --git a/test/OpenMP/target_parallel_for_codegen.cpp b/test/OpenMP/target_parallel_for_codegen.cpp index 58ac10bf9b2ce..d541d14670a5c 100644 --- a/test/OpenMP/target_parallel_for_codegen.cpp +++ b/test/OpenMP/target_parallel_for_codegen.cpp @@ -1,28 +1,46 @@ // Test host codegen. -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER -// CHECK-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CHECK-DAG: [[TT:%.+]] = type { i64, i8 } // CHECK-DAG: [[S1:%.+]] = type { double } @@ -39,31 +57,31 @@ // sizes. // CHECK-DAG: [[SIZET2:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 2, i[[SZ]] 4, i[[SZ]] 4] -// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 288] +// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 800] // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2] -// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 288, i64 288] -// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [10 x i64] [i64 288, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547, i64 288] +// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 800] +// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [10 x i64] [i64 800, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547, i64 800] // CHECK-DAG: [[SIZET5:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 40] -// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 547] +// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 547] // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [4 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 1, i[[SZ]] 40] -// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [4 x i64] [i64 288, i64 288, i64 288, i64 547] -// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [5 x i64] [i64 547, i64 288, i64 288, i64 288, i64 547] -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 - -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK-NOT: @{{.+}} = constant [[ENTTY]] +// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [4 x i64] [i64 800, i64 800, i64 800, i64 547] +// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [6 x i64] [i64 32, i64 281474976711171, i64 800, i64 288, i64 288, i64 547] +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] // Check if offloading descriptor is created. // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] @@ -74,7 +92,7 @@ // CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) // Check target registration is registered as a Ctor. -// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] template<typename tx, typename ty> @@ -200,13 +218,13 @@ int foo(int n) { // CHECK-32: store i32 [[A_VAL]], i32* [[A_CADDR:%.+]], // CHECK-32: [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]], + // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 + // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] + // CHECK: [[TRY]] // CHECK: [[BNSIZE:%.+]] = mul nuw i[[SZ]] [[VLA0:%.+]], 4 // CHECK: [[CNELEMSIZE2:%.+]] = mul nuw i[[SZ]] 5, [[VLA1:%.+]] // CHECK: [[CNSIZE:%.+]] = mul nuw i[[SZ]] [[CNELEMSIZE2]], 8 - // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 - // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] - // CHECK: [[TRY]] // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 10, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([10 x i64], [10 x i64]* [[MAPT4]], i32 0, i32 0), i32 1, i32 0) // CHECK-DAG: [[BPR]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 // CHECK-DAG: [[PR]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P:%[^,]+]], i32 0, i32 0 @@ -321,12 +339,12 @@ int foo(int n) { // correct and loaded correctly for the target regions in foo(). // CHECK: define internal void [[HVT0]]() -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*)) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*)) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED]](i32* noalias %.global_tid., i32* noalias %.bound_tid.) -// CHECK: call i32 @__kmpc_cancel(%ident_t* @ -// CHECK: call i32 @__kmpc_cancellationpoint(%ident_t* @ +// CHECK: call i32 @__kmpc_cancel(%struct.ident_t* @ +// CHECK: call i32 @__kmpc_cancellationpoint(%struct.ident_t* @ // CHECK: ret void // CHECK: } @@ -344,7 +362,7 @@ int foo(int n) { // CHECK-64: store i32 [[AA]], i32* [[AA_C]], align // CHECK-32: store i32 [[AA]], i32* [[AA_CASTED]], align // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i{{32|64}}{{[*]*}})* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]], i{{32|64}}{{[*]*}} %{{.+}}) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i{{32|64}}{{[*]*}})* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]], i{{32|64}}{{[*]*}} %{{.+}}) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED1]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) @@ -368,7 +386,7 @@ int foo(int n) { // CHECK: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i16* // CHECK: store i16 [[AA]], i16* [[AA_C]], align // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]], i[[SZ]] {{.+}}, i[[SZ]] {{.+}}) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]], i[[SZ]] {{.+}}, i[[SZ]] {{.+}}) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED2]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}) @@ -399,7 +417,7 @@ int foo(int n) { // CHECK-DAG: store i16 [[AA]], i16* [[AA_C]], align // CHECK-DAG: [[PARAM1:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CASTED]], align // CHECK-DAG: [[PARAM2:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK-DAG: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]]) +// CHECK-DAG: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED3]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}) @@ -452,7 +470,7 @@ int foo(int n) { // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 10, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], [10 x float]*, i[[SZ]], float*, [5 x [10 x double]]*, i[[SZ]], i[[SZ]], double*, [[TT]]*, i[[SZ]])* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], [10 x float]* [[REF_B]], i[[SZ]] [[VAL_VLA1]], float* [[REF_BN]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] [[VAL_VLA2]], i[[SZ]] [[VAL_VLA3]], double* [[REF_CN]], [[TT]]* [[REF_D]], i[[SZ]] %{{.+}}) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 10, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], [10 x float]*, i[[SZ]], float*, [5 x [10 x double]]*, i[[SZ]], i[[SZ]], double*, [[TT]]*, i[[SZ]])* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], [10 x float]* [[REF_B]], i[[SZ]] [[VAL_VLA1]], float* [[REF_BN]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] [[VAL_VLA2]], i[[SZ]] [[VAL_VLA3]], double* [[REF_CN]], [[TT]]* [[REF_D]], i[[SZ]] %{{.+}}) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED4]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, [10 x float]* {{.+}}, i[[SZ]] %{{.+}}, float* {{.+}}, [5 x [10 x double]]* {{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, double* {{.+}}, [[TT]]* {{.+}}) @@ -537,32 +555,39 @@ int bar(int n){ // CHECK-64: store i32 %{{.+}}, i32* [[B_ADDR]], // CHECK-64: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_CADDR]], +// CHECK-32: store i32 %{{.+}}, i32* %__vla_expr // CHECK-32: store i32 %{{.+}}, i32* [[B_ADDR:%.+]], // CHECK-32: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_ADDR]], +// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 +// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] +// CHECK: [[TRY]] // We capture 2 VLA sizes in this target region // CHECK: [[CELEMSIZE2:%.+]] = mul nuw i[[SZ]] 2, [[VLA0:%.+]] // CHECK: [[CSIZE:%.+]] = mul nuw i[[SZ]] [[CELEMSIZE2]], 2 -// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 -// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] -// CHECK: [[TRY]] -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 5, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([5 x i64], [5 x i64]* [[MAPT7]], i32 0, i32 0), i32 1, i32 0) -// CHECK-DAG: [[BPR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP:%.+]], i32 0, i32 0 -// CHECK-DAG: [[PR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SR]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] -// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX0]] -// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX0]] -// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] -// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX1]] -// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX1]] -// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] -// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX2]] -// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX2]] -// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] -// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX3]] -// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 6, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([6 x i64], [6 x i64]* [[MAPT7]], i32 0, i32 0), i32 1, i32 0) +// CHECK-DAG: [[BPR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP:%.+]], i32 0, i32 0 +// CHECK-DAG: [[PR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SR]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] +// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX0]] +// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX0]] +// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] +// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX1]] +// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX1]] +// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] +// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX2]] +// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX2]] +// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] +// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] +// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX4:[0-9]+]] +// CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX4]] +// CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX4]] +// CHECK-DAG: [[SADDR5:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX5:[0-9]+]] +// CHECK-DAG: [[BPADDR5:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX5]] +// CHECK-DAG: [[PADDR5:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX5]] // The names below are not necessarily consistent with the names used for the // addresses above as some are repeated. @@ -585,15 +610,21 @@ int bar(int n){ // CHECK-DAG: store i[[SZ]] 4, i[[SZ]]* {{%[^,]+}} // CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR3:%.+]], -// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CPADDR3:%.+]], +// CHECK-DAG: store double* %{{.+}}, double** [[CPADDR3:%.+]], // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** -// CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** +// CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to double** +// CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* {{%[^,]+}} + +// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR4:%.+]], +// CHECK-DAG: store double* %{{.+}}, double** [[CPADDR4:%.+]], +// CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to [[S1]]** +// CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to double** // CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* {{%[^,]+}} -// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR4:%.+]], -// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR4:%.+]], -// CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to i16** -// CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to i16** +// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR5:%.+]], +// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR5:%.+]], +// CHECK-DAG: [[CBPADDR5]] = bitcast i8** {{%[^,]+}} to i16** +// CHECK-DAG: [[CPADDR5]] = bitcast i8** {{%[^,]+}} to i16** // CHECK-DAG: store i[[SZ]] [[CSIZE]], i[[SZ]]* {{%[^,]+}} // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 @@ -727,7 +758,7 @@ int bar(int n){ // CHECK-32-DAG:store i32 [[LOCAL_BV]], i32* [[LOCAL_B_CASTED]], align // CHECK-DAG: [[REF_B:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_B_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]], i[[SZ]], i[[SZ]], i16*)* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), [[S1]]* [[REF_THIS]], i[[SZ]] [[REF_B]], i[[SZ]] [[VAL_VLA1]], i[[SZ]] [[VAL_VLA2]], i16* [[REF_C]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]], i[[SZ]], i[[SZ]], i16*)* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), [[S1]]* [[REF_THIS]], i[[SZ]] [[REF_B]], i[[SZ]] [[VAL_VLA1]], i[[SZ]] [[VAL_VLA2]], i16* [[REF_C]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED5]](i32* noalias %.global_tid., i32* noalias %.bound_tid., [[S1]]* %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i16* {{.+}}) @@ -770,7 +801,7 @@ int bar(int n){ // CHECK-DAG: store i8 [[CONV_AAA]], i8* [[CONV]], align // CHECK-DAG: [[REF_AAA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AAA_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED6:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], i[[SZ]] [[REF_AAA]], [10 x i32]* [[REF_B]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED6:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], i[[SZ]] [[REF_AAA]], [10 x i32]* [[REF_B]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED6]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) @@ -803,7 +834,7 @@ int bar(int n){ // CHECK-DAG: store i16 [[CONV_AA]], i16* [[CONV]], align // CHECK-DAG: [[REF_AA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AA_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED7:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], [10 x i32]* [[REF_B]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED7:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], [10 x i32]* [[REF_B]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED7]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) diff --git a/test/OpenMP/target_parallel_for_codegen_registration.cpp b/test/OpenMP/target_parallel_for_codegen_registration.cpp index 24f0c06771ece..9a69ea994efc9 100644 --- a/test/OpenMP/target_parallel_for_codegen_registration.cpp +++ b/test/OpenMP/target_parallel_for_codegen_registration.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target parallel for codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,9 +24,22 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK -// Check that no target code is emmitted if no omptests flag was provided. +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// Check that no target code is emitted if no omptests flag was provided. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK-NTARGET +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -60,96 +81,96 @@ // We have 7 target regions -// CHECK-DAG: {{@.+}} = private constant i8 0 -// TCHECK-NOT: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// TCHECK-NOT: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] -// CHECK-NTARGET-NOT: private constant i8 0 +// CHECK-NTARGET-NOT: weak constant i8 0 // CHECK-NTARGET-NOT: private unnamed_addr constant [1 x i // CHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// CHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// CHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// CHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// CHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// CHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// CHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// CHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// CHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// CHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// CHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// CHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// CHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// TCHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// TCHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// TCHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// TCHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// TCHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// TCHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// TCHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// TCHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// TCHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// TCHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// TCHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// TCHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] // CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] @@ -163,7 +184,7 @@ // CHECK-SAME: { i32, void ()*, i8* } { i32 500, void ()* [[P500:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 501, void ()* [[P501:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 65535, void ()* [[PMAX:@[^,]+]], i8* null }, -// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] // CHECK-NTARGET: @llvm.global_ctors = appending global [3 x { i32, void ()*, i8* }] [ @@ -328,18 +349,18 @@ struct ST { //CHECK-DAG: define internal void @[[NAME12]]( //CHECK-DAG: call void @[[NAME12]]( -//TCHECK-DAG: define void @[[NAME1]]( -//TCHECK-DAG: define void @[[NAME2]]( -//TCHECK-DAG: define void @[[NAME3]]( -//TCHECK-DAG: define void @[[NAME4]]( -//TCHECK-DAG: define void @[[NAME5]]( -//TCHECK-DAG: define void @[[NAME6]]( -//TCHECK-DAG: define void @[[NAME7]]( -//TCHECK-DAG: define void @[[NAME8]]( -//TCHECK-DAG: define void @[[NAME9]]( -//TCHECK-DAG: define void @[[NAME10]]( -//TCHECK-DAG: define void @[[NAME11]]( -//TCHECK-DAG: define void @[[NAME12]]( +//TCHECK-DAG: define weak void @[[NAME1]]( +//TCHECK-DAG: define weak void @[[NAME2]]( +//TCHECK-DAG: define weak void @[[NAME3]]( +//TCHECK-DAG: define weak void @[[NAME4]]( +//TCHECK-DAG: define weak void @[[NAME5]]( +//TCHECK-DAG: define weak void @[[NAME6]]( +//TCHECK-DAG: define weak void @[[NAME7]]( +//TCHECK-DAG: define weak void @[[NAME8]]( +//TCHECK-DAG: define weak void @[[NAME9]]( +//TCHECK-DAG: define weak void @[[NAME10]]( +//TCHECK-DAG: define weak void @[[NAME11]]( +//TCHECK-DAG: define weak void @[[NAME12]]( // CHECK-NTARGET-NOT: __tgt_target // CHECK-NTARGET-NOT: __tgt_register_lib @@ -381,7 +402,7 @@ struct ST { //CHECK: ret void //CHECK: declare i32 @__tgt_unregister_lib([[DSCTY]]*) -//CHECK: define linkonce hidden void @[[REGFN]](i8*) +//CHECK: define linkonce hidden void @[[REGFN]]() //CHECK-SAME: comdat { //CHECK: call i32 @__tgt_register_lib([[DSCTY]]* [[DESC]]) //CHECK: call i32 @__cxa_atexit(void (i8*)* @[[UNREGFN]], i8* bitcast ([[DSCTY]]* [[DESC]] to i8*), @@ -421,31 +442,31 @@ int bar(int a){ // Check metadata is properly generated: // CHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 247, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 265, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 272, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 291, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 415, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 298, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 291, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 298, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 221, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 268, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 286, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 293, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 312, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 436, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 319, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 312, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 319, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 242, i32 {{[0-9]+}}} // TCHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 247, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 265, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 272, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 291, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 415, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 298, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 291, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 298, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 221, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 268, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 286, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 293, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 312, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 436, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 319, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 312, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 319, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 242, i32 {{[0-9]+}}} #endif diff --git a/test/OpenMP/target_parallel_for_codegen_registration_naming.cpp b/test/OpenMP/target_parallel_for_codegen_registration_naming.cpp index b42b5b55a5051..db7f9ec392ff0 100644 --- a/test/OpenMP/target_parallel_for_codegen_registration_naming.cpp +++ b/test/OpenMP/target_parallel_for_codegen_registration_naming.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target parallel for codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,6 +24,16 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/target_parallel_for_collapse_messages.cpp b/test/OpenMP/target_parallel_for_collapse_messages.cpp index 6163b52fd21cf..d8ebdda81aa6a 100644 --- a/test/OpenMP/target_parallel_for_collapse_messages.cpp +++ b/test/OpenMP/target_parallel_for_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_debug_codegen.cpp b/test/OpenMP/target_parallel_for_debug_codegen.cpp index 0e0bb7e853061..55671f4d0d03e 100644 --- a/test/OpenMP/target_parallel_for_debug_codegen.cpp +++ b/test/OpenMP/target_parallel_for_debug_codegen.cpp @@ -1,5 +1,5 @@ -// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc -// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - -debug-info-kind=limited | FileCheck %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -fopenmp-cuda-mode -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -fopenmp-cuda-mode -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - -debug-info-kind=limited | FileCheck %s // expected-no-diagnostics int main() { @@ -58,14 +58,14 @@ int main() { // CHECK: addrspacecast [10 x [10 x [10 x i32]]] addrspace(1)* %{{.+}} to [10 x [10 x [10 x i32]]]* // CHECK: call void [[NONDEBUG_WRAPPER:.+]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]]* {{[^,]+}}, i64 {{[^,]+}}, [10 x [10 x i32]]* {{[^,]+}}, i8* {{[^)]+}}) -// CHECK: define internal void [[DEBUG_PARALLEL:@.+]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]] addrspace(1)* noalias{{[^,]+}}, i32 {{[^,]+}}, [10 x [10 x i32]]* noalias{{[^,]+}}, i8 addrspace(1)* noalias{{[^)]+}}) +// CHECK: define internal void [[DEBUG_PARALLEL:.+]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]] addrspace(1)* noalias{{[^,]+}}, i32 {{[^,]+}}, [10 x [10 x i32]]* noalias{{[^,]+}}, i8 addrspace(1)* noalias{{[^)]+}}) // CHECK: addrspacecast [10 x [10 x [10 x i32]]] addrspace(1)* %{{.+}} to [10 x [10 x [10 x i32]]]* // CHECK: define internal void [[NONDEBUG_WRAPPER]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i64 {{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) // CHECK: addrspacecast [10 x [10 x [10 x i32]]]* %{{.+}} to [10 x [10 x [10 x i32]]] addrspace(1)* // CHECK: call void [[DEBUG_PARALLEL]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 {{[^,]+}}, [10 x [10 x i32]]* {{[^,]+}}, i8 addrspace(1)* {{[^)]+}}) -// CHECK: define void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i64 {{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) +// CHECK: define weak void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i64 {{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) // CHECK: addrspacecast [10 x [10 x [10 x i32]]]* %{{.+}} to [10 x [10 x [10 x i32]]] addrspace(1)* // CHECK: call void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 {{[^,]+}}, [10 x [10 x i32]]* {{[^,]+}}, i8 addrspace(1)* {{[^)]+}}) @@ -83,7 +83,7 @@ int main() { // CHECK: addrspacecast [10 x [10 x i32]]* %{{.+}} to [10 x [10 x i32]] addrspace(1)* // CHECK: call void [[DEBUG_PARALLEL]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 {{[^,]+}}, [10 x [10 x i32]] addrspace(1)* {{[^,]+}}, i8 addrspace(1)* {{[^)]+}}) -// CHECK: define void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i64 {{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) +// CHECK: define weak void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i64 {{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) // CHECK: addrspacecast [10 x [10 x [10 x i32]]]* %{{.+}} to [10 x [10 x [10 x i32]]] addrspace(1)* // CHECK: addrspacecast [10 x [10 x i32]]* %{{.+}} to [10 x [10 x i32]] addrspace(1)* // CHECK: call void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 {{[^,]+}}, [10 x [10 x i32]] addrspace(1)* {{[^,]+}}, i8 addrspace(1)* {{[^)]+}}) @@ -92,22 +92,24 @@ int main() { // CHECK: addrspacecast [10 x [10 x [10 x i32]]] addrspace(1)* %{{.+}} to [10 x [10 x [10 x i32]]]* // CHECK: addrspacecast i32 addrspace(1)* %{{.+}} to i32* // CHECK: addrspacecast [10 x [10 x i32]] addrspace(1)* %{{.+}} to [10 x [10 x i32]]* -// CHECK: call void [[NONDEBUG_WRAPPER:@.+]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]]* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x i32]]* {{[^,]+}}, i8* {{[^)]+}}) +// CHECK: call void @[[NONDEBUG_WRAPPER:.+]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]]* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x i32]]* {{[^,]+}}, i8* {{[^)]+}}) -// CHECK: define internal void [[DEBUG_PARALLEL:@.+]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]] addrspace(1)* noalias{{[^,]+}}, i32 addrspace(1)* noalias{{[^,]+}}, [10 x [10 x i32]] addrspace(1)* noalias{{[^,]+}}, i8 addrspace(1)* noalias{{[^)]+}}) +// CHECK: define internal void @[[DEBUG_PARALLEL:.+]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]] addrspace(1)* noalias{{[^,]+}}, i32 addrspace(1)* noalias{{[^,]+}}, [10 x [10 x i32]] addrspace(1)* noalias{{[^,]+}}, i8 addrspace(1)* noalias{{[^)]+}}) // CHECK: addrspacecast [10 x [10 x [10 x i32]]] addrspace(1)* %{{.+}} to [10 x [10 x [10 x i32]]]* // CHECK: addrspacecast i32 addrspace(1)* %{{.+}} to i32* // CHECK: addrspacecast [10 x [10 x i32]] addrspace(1)* %{{.+}} to [10 x [10 x i32]]* -// CHECK: define internal void [[NONDEBUG_WRAPPER]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i32* dereferenceable{{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) +// CHECK: define internal void @[[NONDEBUG_WRAPPER]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i32* dereferenceable{{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) // CHECK: addrspacecast [10 x [10 x [10 x i32]]]* %{{.+}} to [10 x [10 x [10 x i32]]] addrspace(1)* // CHECK: addrspacecast i32* %{{.+}} to i32 addrspace(1)* // CHECK: addrspacecast [10 x [10 x i32]]* %{{.+}} to [10 x [10 x i32]] addrspace(1)* -// CHECK: call void [[DEBUG_PARALLEL]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 addrspace(1)* {{[^,]+}}, [10 x [10 x i32]] addrspace(1)* {{[^,]+}}, i8 addrspace(1)* {{[^)]+}}) +// CHECK: call void @[[DEBUG_PARALLEL]](i32* {{[^,]+}}, i32* {{[^,]+}}, [10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 addrspace(1)* {{[^,]+}}, [10 x [10 x i32]] addrspace(1)* {{[^,]+}}, i8 addrspace(1)* {{[^)]+}}) -// CHECK: define void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i32* dereferenceable{{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) +// CHECK: define weak void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]]* dereferenceable{{[^,]+}}, i32* dereferenceable{{[^,]+}}, [10 x [10 x i32]]* dereferenceable{{[^,]+}}, i8* dereferenceable{{[^)]+}}) // CHECK: addrspacecast [10 x [10 x [10 x i32]]]* %{{.+}} to [10 x [10 x [10 x i32]]] addrspace(1)* // CHECK: addrspacecast i32* %{{.+}} to i32 addrspace(1)* // CHECK: addrspacecast [10 x [10 x i32]]* %{{.+}} to [10 x [10 x i32]] addrspace(1)* // CHECK: call void @__omp_offloading_{{[^(]+}}([10 x [10 x [10 x i32]]] addrspace(1)* {{[^,]+}}, i32 addrspace(1)* {{[^,]+}}, [10 x [10 x i32]] addrspace(1)* {{[^,]+}}, i8 addrspace(1)* {{[^)]+}}) +// CHECK-DAG: distinct !DISubprogram(name: "[[NONDEBUG_WRAPPER]]", +// CHECK-DAG: distinct !DISubprogram(name: "[[DEBUG_PARALLEL]]", diff --git a/test/OpenMP/target_parallel_for_default_messages.cpp b/test/OpenMP/target_parallel_for_default_messages.cpp index c1f04f4f6e409..94049cd37bafc 100644 --- a/test/OpenMP/target_parallel_for_default_messages.cpp +++ b/test/OpenMP/target_parallel_for_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/target_parallel_for_defaultmap_messages.cpp b/test/OpenMP/target_parallel_for_defaultmap_messages.cpp index 24973edd2bda8..9f97b60ff41cf 100644 --- a/test/OpenMP/target_parallel_for_defaultmap_messages.cpp +++ b/test/OpenMP/target_parallel_for_defaultmap_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_depend_codegen.cpp b/test/OpenMP/target_parallel_for_depend_codegen.cpp new file mode 100644 index 0000000000000..c6e71375f18c6 --- /dev/null +++ b/test/OpenMP/target_parallel_for_depend_codegen.cpp @@ -0,0 +1,261 @@ +// Test host codegen. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// CHECK-DAG: [[TT:%.+]] = type { i64, i8 } +// CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } +// CHECK-DAG: [[DEVTY:%.+]] = type { i8*, i8*, [[ENTTY]]*, [[ENTTY]]* } +// CHECK-DAG: [[DSCTY:%.+]] = type { i32, [[DEVTY]]*, [[ENTTY]]*, [[ENTTY]]* } + +// TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 } + +// CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat + +// CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 0, i[[SZ]] 4] +// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 800] +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] + +// Check if offloading descriptor is created. +// CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] +// CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] +// CHECK: [[DEVBEGIN:@.+]] = external constant i8 +// CHECK: [[DEVEND:@.+]] = external constant i8 +// CHECK: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x [[DEVTY]]] [{{.+}} { i8* [[DEVBEGIN]], i8* [[DEVEND]], [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }], comdat($[[REGFN]]) +// CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) + +// Check target registration is registered as a Ctor. +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] + + +template<typename tx, typename ty> +struct TT{ + tx X; + ty Y; +}; + +int global; +extern int global; + +// CHECK: define {{.*}}[[FOO:@.+]]( +int foo(int n) { + int a = 0; + short aa = 0; + float b[10]; + float bn[n]; + double c[5][10]; + double cn[5][n]; + TT<long long, char> d; + static long *plocal; + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[IN:@.+]], i32 [[GTID:%.+]], i32 1, i[[SZ]] {{20|40}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY0:%.+]]* + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 3 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* [[IN]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]]) + #pragma omp target parallel for device(global + a) depend(in: global) depend(out: a, b, cn[4]) + for (int i = 0; i < 10; ++i) { + } + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + + // CHECK: [[BOOL:%.+]] = icmp ne i32 %{{.+}}, 0 + // CHECK: br i1 [[BOOL]], label %[[THEN:.+]], label %[[ELSE:.+]] + // CHECK: [[THEN]]: + // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%.+]], i32 0, i32 0 + // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%.+]], i32 0, i32 0 + // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]** + // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]** + // CHECK-DAG: store i[[SZ]]* [[BP0:%[^,]+]], i[[SZ]]** [[CBPADDR0]] + // CHECK-DAG: store i[[SZ]]* [[BP0]], i[[SZ]]** [[CPADDR0]] + + // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1 + // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1 + // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* + // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* + // CHECK-DAG: store i[[SZ]] [[BP1:%[^,]+]], i[[SZ]]* [[CBPADDR1]] + // CHECK-DAG: store i[[SZ]] [[BP1]], i[[SZ]]* [[CPADDR1]] + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[IN]], i32 [[GTID]], i32 1, i[[SZ]] {{104|52}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1_:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1_:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + + // CHECK: [[ELSE]]: + // CHECK-NOT: getelementptr inbounds [2 x i8*], [2 x i8*]* + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[IN]], i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1__:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1__:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + // CHECK: [[EXIT]]: + + #pragma omp target parallel for device(global + a) nowait depend(inout: global, a, bn) if(target:a) + for (int i = 0; i < *plocal; ++i) { + static int local1; + *plocal = global; + local1 = global; + } + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[IN]], i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]* + // CHECK: getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* [[IN]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]]) + #pragma omp target parallel for if(0) firstprivate(global) depend(out:global) + for (int i = 0; i < global; ++i) { + global += 1; + } + + return a; +} + +// Check that the offloading functions are emitted and that the arguments are +// correct and loaded correctly for the target regions in foo(). + +// CHECK: define internal void [[HVT0:@.+]]() + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, [[TASK_TY0]]* noalias) +// CHECK: store void (i8*, ...)* null, void (i8*, ...)** % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 0, i8** null, i8** null, i[[SZ]]* null, i64* null, i32 1, i32 0) +// CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: call void [[HVT0]]() +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT1:@.+]](i[[SZ]]* %{{.+}}, i[[SZ]] %{{.+}}) + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1_]](i32{{.*}}, [[TASK_TY1_]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[SZT:%.+]] = getelementptr inbounds [2 x i[[SZ]]], [2 x i[[SZ]]]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams_nowait(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 2, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SZT]], i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT]], i32 0, i32 0), i32 1, i32 0) + +// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1__]](i32{{.*}}, [[TASK_TY1__]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT2:@.+]](i[[SZ]] %{{.+}}) +// Create stack storage and store argument in there. +// CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align +// CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align +// CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* +// CHECK-64: load i32, i32* [[AA_CADDR]], align +// CHECK-32: load i32, i32* [[AA_ADDR]], align + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, [[TASK_TY2]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT2]](i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + + +#endif diff --git a/test/OpenMP/target_parallel_for_depend_messages.cpp b/test/OpenMP/target_parallel_for_depend_messages.cpp index 19872566161a3..d644711b5e272 100644 --- a/test/OpenMP/target_parallel_for_depend_messages.cpp +++ b/test/OpenMP/target_parallel_for_depend_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_device_messages.cpp b/test/OpenMP/target_parallel_for_device_messages.cpp index 16a21ba4d662d..bacf57d99cccb 100644 --- a/test/OpenMP/target_parallel_for_device_messages.cpp +++ b/test/OpenMP/target_parallel_for_device_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_firstprivate_messages.cpp b/test/OpenMP/target_parallel_for_firstprivate_messages.cpp index 36bfe25e94682..350bb89108b88 100644 --- a/test/OpenMP/target_parallel_for_firstprivate_messages.cpp +++ b/test/OpenMP/target_parallel_for_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_if_messages.cpp b/test/OpenMP/target_parallel_for_if_messages.cpp index 4acf4b8d4a777..2bc433cf421db 100644 --- a/test/OpenMP/target_parallel_for_if_messages.cpp +++ b/test/OpenMP/target_parallel_for_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_is_device_ptr_ast_print.cpp b/test/OpenMP/target_parallel_for_is_device_ptr_ast_print.cpp index eaa0941a388e8..fe600101b8ff8 100644 --- a/test/OpenMP/target_parallel_for_is_device_ptr_ast_print.cpp +++ b/test/OpenMP/target_parallel_for_is_device_ptr_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -59,7 +63,7 @@ struct SA { // CHECK-NEXT: int aa[10]; // CHECK-NEXT: arr &raa = this->aa; // CHECK-NEXT: func( -// CHECK-NEXT: #pragma omp target parallel for is_device_ptr(this->k) +// CHECK-NEXT: #pragma omp target parallel for is_device_ptr(this->k){{$}} // CHECK-NEXT: for (int i = 0; i < 100; i++) // CHECK-NEXT: foo(); // CHECK-NEXT: #pragma omp target parallel for is_device_ptr(this->z) diff --git a/test/OpenMP/target_parallel_for_is_device_ptr_messages.cpp b/test/OpenMP/target_parallel_for_is_device_ptr_messages.cpp index b3b33bf91ed0f..3e675ead2acc6 100644 --- a/test/OpenMP/target_parallel_for_is_device_ptr_messages.cpp +++ b/test/OpenMP/target_parallel_for_is_device_ptr_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -std=c++11 -verify -fopenmp %s +// RUN: %clang_cc1 -std=c++11 -verify -fopenmp-simd %s + struct ST { int *a; }; diff --git a/test/OpenMP/target_parallel_for_lastprivate_messages.cpp b/test/OpenMP/target_parallel_for_lastprivate_messages.cpp index ee703a1ce32e5..0ff2c722f7de8 100644 --- a/test/OpenMP/target_parallel_for_lastprivate_messages.cpp +++ b/test/OpenMP/target_parallel_for_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_linear_messages.cpp b/test/OpenMP/target_parallel_for_linear_messages.cpp index 36e897d5946e6..b7224b538f4cb 100644 --- a/test/OpenMP/target_parallel_for_linear_messages.cpp +++ b/test/OpenMP/target_parallel_for_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/target_parallel_for_loop_messages.cpp b/test/OpenMP/target_parallel_for_loop_messages.cpp index d7c189116bf01..416cbfa1f3c3f 100644 --- a/test/OpenMP/target_parallel_for_loop_messages.cpp +++ b/test/OpenMP/target_parallel_for_loop_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s -Wno-openmp-target + +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s -Wno-openmp-target class S { int a; diff --git a/test/OpenMP/target_parallel_for_map_messages.cpp b/test/OpenMP/target_parallel_for_map_messages.cpp index f3e1c616b4062..f4f98dfc4915d 100644 --- a/test/OpenMP/target_parallel_for_map_messages.cpp +++ b/test/OpenMP/target_parallel_for_map_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/target_parallel_for_messages.cpp b/test/OpenMP/target_parallel_for_messages.cpp index 173025c01ce76..6bf4ac2807644 100644 --- a/test/OpenMP/target_parallel_for_messages.cpp +++ b/test/OpenMP/target_parallel_for_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -std=c++11 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -std=c++11 -o - %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_misc_messages.c b/test/OpenMP/target_parallel_for_misc_messages.c index cfe83f1301075..44629e10c0c0d 100644 --- a/test/OpenMP/target_parallel_for_misc_messages.c +++ b/test/OpenMP/target_parallel_for_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp target parallel for'}} #pragma omp target parallel for diff --git a/test/OpenMP/target_parallel_for_nowait_messages.cpp b/test/OpenMP/target_parallel_for_nowait_messages.cpp index 06d22969c4a4c..2fb875b9de9f5 100644 --- a/test/OpenMP/target_parallel_for_nowait_messages.cpp +++ b/test/OpenMP/target_parallel_for_nowait_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_num_threads_messages.cpp b/test/OpenMP/target_parallel_for_num_threads_messages.cpp index 915cfb1dfc9e0..b47c96b82b55b 100644 --- a/test/OpenMP/target_parallel_for_num_threads_messages.cpp +++ b/test/OpenMP/target_parallel_for_num_threads_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_ordered_messages.cpp b/test/OpenMP/target_parallel_for_ordered_messages.cpp index 72bb6e34912d2..002e8e9a5c747 100644 --- a/test/OpenMP/target_parallel_for_ordered_messages.cpp +++ b/test/OpenMP/target_parallel_for_ordered_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_private_messages.cpp b/test/OpenMP/target_parallel_for_private_messages.cpp index 1d6381a1d5d3e..dae4f17cccb88 100644 --- a/test/OpenMP/target_parallel_for_private_messages.cpp +++ b/test/OpenMP/target_parallel_for_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } @@ -43,7 +45,7 @@ public: S5(int v) : a(v) {} S5 &operator=(S5 &s) { #pragma omp target parallel for private(a) private(this->a) private(s.a) // expected-error {{expected variable name or data member of current class}} - for (int k = 0; k < s.a; ++k) + for (int k = 0; k < s.a; ++k) // expected-warning {{Non-trivial type 'S5' is mapped, only trivial types are guaranteed to be mapped correctly}} ++s.a; return *this; } diff --git a/test/OpenMP/target_parallel_for_proc_bind_messages.cpp b/test/OpenMP/target_parallel_for_proc_bind_messages.cpp index eeb232aeca811..60d5b74fdc008 100644 --- a/test/OpenMP/target_parallel_for_proc_bind_messages.cpp +++ b/test/OpenMP/target_parallel_for_proc_bind_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/target_parallel_for_reduction_messages.cpp b/test/OpenMP/target_parallel_for_reduction_messages.cpp index 9f1f68f25f830..5cc40e9b9924b 100644 --- a/test/OpenMP/target_parallel_for_reduction_messages.cpp +++ b/test/OpenMP/target_parallel_for_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_schedule_messages.cpp b/test/OpenMP/target_parallel_for_schedule_messages.cpp index 075e1df7827c5..544ace25e408c 100644 --- a/test/OpenMP/target_parallel_for_schedule_messages.cpp +++ b/test/OpenMP/target_parallel_for_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_aligned_messages.cpp b/test/OpenMP/target_parallel_for_simd_aligned_messages.cpp index 655f90642f5bc..54a6a9e8de26a 100644 --- a/test/OpenMP/target_parallel_for_simd_aligned_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_aligned_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp %s +// RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp-simd %s + struct B { static int ib[20]; // expected-note 0 {{'B::ib' declared here}} static constexpr int bfoo() { return 8; } diff --git a/test/OpenMP/target_parallel_for_simd_ast_print.cpp b/test/OpenMP/target_parallel_for_simd_ast_print.cpp index 8f0ed97d5cd94..de7fb12ab9cc8 100644 --- a/test/OpenMP/target_parallel_for_simd_ast_print.cpp +++ b/test/OpenMP/target_parallel_for_simd_ast_print.cpp @@ -1,6 +1,10 @@ -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s -Wno-openmp-target | FileCheck %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print -Wno-openmp-target | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s -Wno-openmp-target | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print -Wno-openmp-target | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -35,7 +39,7 @@ public: } }; -// CHECK: #pragma omp target parallel for simd private(this->a) private(this->a) private(T::a) +// CHECK: #pragma omp target parallel for simd private(this->a) private(this->a) private(T::a){{$}} // CHECK: #pragma omp target parallel for simd private(this->a) private(this->a) // CHECK: #pragma omp target parallel for simd private(this->a) private(this->a) private(this->S::a) diff --git a/test/OpenMP/target_parallel_for_simd_codegen.cpp b/test/OpenMP/target_parallel_for_simd_codegen.cpp index d168214df35fe..012b2f86be480 100644 --- a/test/OpenMP/target_parallel_for_simd_codegen.cpp +++ b/test/OpenMP/target_parallel_for_simd_codegen.cpp @@ -1,27 +1,45 @@ // Test host codegen. -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER -// CHECK-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CHECK-DAG: [[TT:%.+]] = type { i64, i8 } // CHECK-DAG: [[S1:%.+]] = type { double } @@ -38,31 +56,31 @@ // sizes. // CHECK-DAG: [[SIZET2:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 2, i[[SZ]] 4, i[[SZ]] 4] -// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 288] +// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 800] // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2] -// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 288, i64 288] -// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [10 x i64] [i64 288, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547, i64 288] +// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 800] +// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [10 x i64] [i64 800, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547, i64 800] // CHECK-DAG: [[SIZET5:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 40] -// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 547] +// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 547] // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [4 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 1, i[[SZ]] 40] -// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [4 x i64] [i64 288, i64 288, i64 288, i64 547] -// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [5 x i64] [i64 547, i64 288, i64 288, i64 288, i64 547] -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 - -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK-NOT: @{{.+}} = constant [[ENTTY]] +// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [4 x i64] [i64 800, i64 800, i64 800, i64 547] +// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [6 x i64] [i64 32, i64 281474976711171, i64 800, i64 288, i64 288, i64 547] +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] // Check if offloading descriptor is created. // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] @@ -73,7 +91,7 @@ // CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) // Check target registration is registered as a Ctor. -// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] template<typename tx, typename ty> @@ -197,13 +215,13 @@ int foo(int n) { // CHECK-32: store i32 [[A_VAL]], i32* [[A_CADDR:%.+]], // CHECK-32: [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]], + // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 + // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] + // CHECK: [[TRY]] // CHECK: [[BNSIZE:%.+]] = mul nuw i[[SZ]] [[VLA0:%.+]], 4 // CHECK: [[CNELEMSIZE2:%.+]] = mul nuw i[[SZ]] 5, [[VLA1:%.+]] // CHECK: [[CNSIZE:%.+]] = mul nuw i[[SZ]] [[CNELEMSIZE2]], 8 - // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 - // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] - // CHECK: [[TRY]] // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 10, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([10 x i64], [10 x i64]* [[MAPT4]], i32 0, i32 0), i32 1, i32 0) // CHECK-DAG: [[BPR]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 // CHECK-DAG: [[PR]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P:%[^,]+]], i32 0, i32 0 @@ -318,7 +336,7 @@ int foo(int n) { // correct and loaded correctly for the target regions in foo(). // CHECK: define internal void [[HVT0]]() -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*)) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*)) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED]](i32* noalias %.global_tid., i32* noalias %.bound_tid.) @@ -340,7 +358,7 @@ int foo(int n) { // CHECK-64: store i32 [[AA]], i32* [[AA_C]], align // CHECK-32: store i32 [[AA]], i32* [[AA_CASTED]], align // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i{{32|64}}{{[*]*}})* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]], i{{32|64}}{{[*]*}} %{{.+}}) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i{{32|64}}{{[*]*}})* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]], i{{32|64}}{{[*]*}} %{{.+}}) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED1]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) @@ -366,7 +384,7 @@ int foo(int n) { // CHECK: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i16* // CHECK: store i16 [[AA]], i16* [[AA_C]], align // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]], i[[SZ]] {{.+}}, i[[SZ]] {{.+}}) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]], i[[SZ]] {{.+}}, i[[SZ]] {{.+}}) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED2]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}) @@ -398,7 +416,7 @@ int foo(int n) { // CHECK-DAG: store i16 [[AA]], i16* [[AA_C]], align // CHECK-DAG: [[PARAM1:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CASTED]], align // CHECK-DAG: [[PARAM2:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK-DAG: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]]) +// CHECK-DAG: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED3]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}) @@ -452,7 +470,7 @@ int foo(int n) { // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 10, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], [10 x float]*, i[[SZ]], float*, [5 x [10 x double]]*, i[[SZ]], i[[SZ]], double*, [[TT]]*, i[[SZ]])* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], [10 x float]* [[REF_B]], i[[SZ]] [[VAL_VLA1]], float* [[REF_BN]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] [[VAL_VLA2]], i[[SZ]] [[VAL_VLA3]], double* [[REF_CN]], [[TT]]* [[REF_D]], i[[SZ]] %{{.+}}) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 10, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], [10 x float]*, i[[SZ]], float*, [5 x [10 x double]]*, i[[SZ]], i[[SZ]], double*, [[TT]]*, i[[SZ]])* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], [10 x float]* [[REF_B]], i[[SZ]] [[VAL_VLA1]], float* [[REF_BN]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] [[VAL_VLA2]], i[[SZ]] [[VAL_VLA3]], double* [[REF_CN]], [[TT]]* [[REF_D]], i[[SZ]] %{{.+}}) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED4]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, [10 x float]* {{.+}}, i[[SZ]] %{{.+}}, float* {{.+}}, [5 x [10 x double]]* {{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, double* {{.+}}, [[TT]]* {{.+}}) @@ -537,32 +555,39 @@ int bar(int n){ // CHECK-64: store i32 %{{.+}}, i32* [[B_ADDR]], // CHECK-64: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_CADDR]], +// CHECK-32: store i32 %{{.+}}, i32* %__vla_expr // CHECK-32: store i32 %{{.+}}, i32* [[B_ADDR:%.+]], // CHECK-32: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_ADDR]], +// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 +// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] +// CHECK: [[TRY]] // We capture 2 VLA sizes in this target region // CHECK: [[CELEMSIZE2:%.+]] = mul nuw i[[SZ]] 2, [[VLA0:%.+]] // CHECK: [[CSIZE:%.+]] = mul nuw i[[SZ]] [[CELEMSIZE2]], 2 -// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 -// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] -// CHECK: [[TRY]] -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 5, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([5 x i64], [5 x i64]* [[MAPT7]], i32 0, i32 0), i32 1, i32 0) -// CHECK-DAG: [[BPR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP:%.+]], i32 0, i32 0 -// CHECK-DAG: [[PR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SR]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] -// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX0]] -// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX0]] -// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] -// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX1]] -// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX1]] -// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] -// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX2]] -// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX2]] -// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] -// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX3]] -// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 6, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([6 x i64], [6 x i64]* [[MAPT7]], i32 0, i32 0), i32 1, i32 0) +// CHECK-DAG: [[BPR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP:%.+]], i32 0, i32 0 +// CHECK-DAG: [[PR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SR]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] +// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX0]] +// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX0]] +// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] +// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX1]] +// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX1]] +// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] +// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX2]] +// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX2]] +// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] +// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] +// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX4:[0-9]+]] +// CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX4]] +// CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX4]] +// CHECK-DAG: [[SADDR5:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX5:[0-9]+]] +// CHECK-DAG: [[BPADDR5:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX5]] +// CHECK-DAG: [[PADDR5:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX5]] // The names below are not necessarily consistent with the names used for the // addresses above as some are repeated. @@ -585,15 +610,21 @@ int bar(int n){ // CHECK-DAG: store i[[SZ]] 4, i[[SZ]]* {{%[^,]+}} // CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR3:%.+]], -// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CPADDR3:%.+]], +// CHECK-DAG: store double* %{{.+}}, double** [[CPADDR3:%.+]], // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** -// CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** +// CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to double** +// CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* {{%[^,]+}} + +// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR4:%.+]], +// CHECK-DAG: store double* %{{.+}}, double** [[CPADDR4:%.+]], +// CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to [[S1]]** +// CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to double** // CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* {{%[^,]+}} -// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR4:%.+]], -// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR4:%.+]], -// CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to i16** -// CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to i16** +// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR5:%.+]], +// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR5:%.+]], +// CHECK-DAG: [[CBPADDR5]] = bitcast i8** {{%[^,]+}} to i16** +// CHECK-DAG: [[CPADDR5]] = bitcast i8** {{%[^,]+}} to i16** // CHECK-DAG: store i[[SZ]] [[CSIZE]], i[[SZ]]* {{%[^,]+}} // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 @@ -727,7 +758,7 @@ int bar(int n){ // CHECK-32-DAG:store i32 [[LOCAL_BV]], i32* [[LOCAL_B_CASTED]], align // CHECK-DAG: [[REF_B:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_B_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]], i[[SZ]], i[[SZ]], i16*)* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), [[S1]]* [[REF_THIS]], i[[SZ]] [[REF_B]], i[[SZ]] [[VAL_VLA1]], i[[SZ]] [[VAL_VLA2]], i16* [[REF_C]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]], i[[SZ]], i[[SZ]], i16*)* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), [[S1]]* [[REF_THIS]], i[[SZ]] [[REF_B]], i[[SZ]] [[VAL_VLA1]], i[[SZ]] [[VAL_VLA2]], i16* [[REF_C]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED5]](i32* noalias %.global_tid., i32* noalias %.bound_tid., [[S1]]* %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i16* {{.+}}) @@ -770,7 +801,7 @@ int bar(int n){ // CHECK-DAG: store i8 [[CONV_AAA]], i8* [[CONV]], align // CHECK-DAG: [[REF_AAA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AAA_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED6:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], i[[SZ]] [[REF_AAA]], [10 x i32]* [[REF_B]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED6:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], i[[SZ]] [[REF_AAA]], [10 x i32]* [[REF_B]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED6]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) @@ -803,7 +834,7 @@ int bar(int n){ // CHECK-DAG: store i16 [[CONV_AA]], i16* [[CONV]], align // CHECK-DAG: [[REF_AA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AA_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED7:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], [10 x i32]* [[REF_B]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED7:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], [10 x i32]* [[REF_B]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED7]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) diff --git a/test/OpenMP/target_parallel_for_simd_codegen_registration.cpp b/test/OpenMP/target_parallel_for_simd_codegen_registration.cpp index 5cc9d7015cbda..5973c8d182f1a 100644 --- a/test/OpenMP/target_parallel_for_simd_codegen_registration.cpp +++ b/test/OpenMP/target_parallel_for_simd_codegen_registration.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target parallel for simd codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,9 +24,22 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK -// Check that no target code is emmitted if no omptests flag was provided. +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// Check that no target code is emitted if no omptests flag was provided. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK-NTARGET +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -60,96 +81,96 @@ // We have 7 target regions -// CHECK-DAG: {{@.+}} = private constant i8 0 -// TCHECK-NOT: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// TCHECK-NOT: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] -// CHECK-NTARGET-NOT: private constant i8 0 +// CHECK-NTARGET-NOT: weak constant i8 0 // CHECK-NTARGET-NOT: private unnamed_addr constant [1 x i // CHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// CHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// CHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// CHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// CHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// CHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// CHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// CHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// CHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// CHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// CHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// CHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// CHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// TCHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// TCHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// TCHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// TCHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// TCHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// TCHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// TCHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// TCHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// TCHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// TCHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// TCHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// TCHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] // CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] @@ -163,7 +184,7 @@ // CHECK-SAME: { i32, void ()*, i8* } { i32 500, void ()* [[P500:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 501, void ()* [[P501:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 65535, void ()* [[PMAX:@[^,]+]], i8* null }, -// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] // CHECK-NTARGET: @llvm.global_ctors = appending global [3 x { i32, void ()*, i8* }] [ @@ -328,18 +349,18 @@ struct ST { //CHECK-DAG: define internal void @[[NAME12]]( //CHECK-DAG: call void @[[NAME12]]( -//TCHECK-DAG: define void @[[NAME1]]( -//TCHECK-DAG: define void @[[NAME2]]( -//TCHECK-DAG: define void @[[NAME3]]( -//TCHECK-DAG: define void @[[NAME4]]( -//TCHECK-DAG: define void @[[NAME5]]( -//TCHECK-DAG: define void @[[NAME6]]( -//TCHECK-DAG: define void @[[NAME7]]( -//TCHECK-DAG: define void @[[NAME8]]( -//TCHECK-DAG: define void @[[NAME9]]( -//TCHECK-DAG: define void @[[NAME10]]( -//TCHECK-DAG: define void @[[NAME11]]( -//TCHECK-DAG: define void @[[NAME12]]( +//TCHECK-DAG: define weak void @[[NAME1]]( +//TCHECK-DAG: define weak void @[[NAME2]]( +//TCHECK-DAG: define weak void @[[NAME3]]( +//TCHECK-DAG: define weak void @[[NAME4]]( +//TCHECK-DAG: define weak void @[[NAME5]]( +//TCHECK-DAG: define weak void @[[NAME6]]( +//TCHECK-DAG: define weak void @[[NAME7]]( +//TCHECK-DAG: define weak void @[[NAME8]]( +//TCHECK-DAG: define weak void @[[NAME9]]( +//TCHECK-DAG: define weak void @[[NAME10]]( +//TCHECK-DAG: define weak void @[[NAME11]]( +//TCHECK-DAG: define weak void @[[NAME12]]( // CHECK-NTARGET-NOT: __tgt_target // CHECK-NTARGET-NOT: __tgt_register_lib @@ -381,7 +402,7 @@ struct ST { //CHECK: ret void //CHECK: declare i32 @__tgt_unregister_lib([[DSCTY]]*) -//CHECK: define linkonce hidden void @[[REGFN]](i8*) +//CHECK: define linkonce hidden void @[[REGFN]]() //CHECK-SAME: comdat { //CHECK: call i32 @__tgt_register_lib([[DSCTY]]* [[DESC]]) //CHECK: call i32 @__cxa_atexit(void (i8*)* @[[UNREGFN]], i8* bitcast ([[DSCTY]]* [[DESC]] to i8*), @@ -421,31 +442,31 @@ int bar(int a){ // Check metadata is properly generated: // CHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 247, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 265, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 272, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 291, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 415, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 298, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 291, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 298, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 221, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 268, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 286, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 293, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 312, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 436, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 319, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 312, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 319, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 242, i32 {{[0-9]+}}} // TCHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 247, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 265, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 272, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 291, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 415, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 298, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 291, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 298, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 221, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 268, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 286, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 293, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 312, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 436, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 319, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 312, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 319, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 242, i32 {{[0-9]+}}} #endif diff --git a/test/OpenMP/target_parallel_for_simd_codegen_registration_naming.cpp b/test/OpenMP/target_parallel_for_simd_codegen_registration_naming.cpp index bdfa38669859d..6c8e06a455109 100644 --- a/test/OpenMP/target_parallel_for_simd_codegen_registration_naming.cpp +++ b/test/OpenMP/target_parallel_for_simd_codegen_registration_naming.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target parallel for simd codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,6 +24,16 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/target_parallel_for_simd_collapse_messages.cpp b/test/OpenMP/target_parallel_for_simd_collapse_messages.cpp index ecf2d6e7889e9..f3f2aa9ace56b 100644 --- a/test/OpenMP/target_parallel_for_simd_collapse_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_default_messages.cpp b/test/OpenMP/target_parallel_for_simd_default_messages.cpp index 5d41cbcb457e1..b67bfc52e75b0 100644 --- a/test/OpenMP/target_parallel_for_simd_default_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/target_parallel_for_simd_defaultmap_messages.cpp b/test/OpenMP/target_parallel_for_simd_defaultmap_messages.cpp index e922e0a000b45..c019c37c39d38 100644 --- a/test/OpenMP/target_parallel_for_simd_defaultmap_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_defaultmap_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_depend_codegen.cpp b/test/OpenMP/target_parallel_for_simd_depend_codegen.cpp new file mode 100644 index 0000000000000..6fc71a7875592 --- /dev/null +++ b/test/OpenMP/target_parallel_for_simd_depend_codegen.cpp @@ -0,0 +1,261 @@ +// Test host codegen. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// CHECK-DAG: [[TT:%.+]] = type { i64, i8 } +// CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } +// CHECK-DAG: [[DEVTY:%.+]] = type { i8*, i8*, [[ENTTY]]*, [[ENTTY]]* } +// CHECK-DAG: [[DSCTY:%.+]] = type { i32, [[DEVTY]]*, [[ENTTY]]*, [[ENTTY]]* } + +// TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 } + +// CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat + +// CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 0, i[[SZ]] 4] +// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 800] +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] + +// Check if offloading descriptor is created. +// CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] +// CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] +// CHECK: [[DEVBEGIN:@.+]] = external constant i8 +// CHECK: [[DEVEND:@.+]] = external constant i8 +// CHECK: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x [[DEVTY]]] [{{.+}} { i8* [[DEVBEGIN]], i8* [[DEVEND]], [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }], comdat($[[REGFN]]) +// CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) + +// Check target registration is registered as a Ctor. +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] + + +template<typename tx, typename ty> +struct TT{ + tx X; + ty Y; +}; + +int global; +extern int global; + +// CHECK: define {{.*}}[[FOO:@.+]]( +int foo(int n) { + int a = 0; + short aa = 0; + float b[10]; + float bn[n]; + double c[5][10]; + double cn[5][n]; + TT<long long, char> d; + static long *plocal; + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[IN:@.+]], i32 [[GTID:%.+]], i32 1, i[[SZ]] {{20|40}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY0:%.+]]* + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 3 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* [[IN]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]]) + #pragma omp target parallel for simd device(global + a) depend(in: global) depend(out: a, b, cn[4]) + for (int i = 0; i < 10; ++i) { + } + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + + // CHECK: [[BOOL:%.+]] = icmp ne i32 %{{.+}}, 0 + // CHECK: br i1 [[BOOL]], label %[[THEN:.+]], label %[[ELSE:.+]] + // CHECK: [[THEN]]: + // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%.+]], i32 0, i32 0 + // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%.+]], i32 0, i32 0 + // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]** + // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]** + // CHECK-DAG: store i[[SZ]]* [[BP0:%[^,]+]], i[[SZ]]** [[CBPADDR0]] + // CHECK-DAG: store i[[SZ]]* [[BP0]], i[[SZ]]** [[CPADDR0]] + + // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1 + // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1 + // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* + // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* + // CHECK-DAG: store i[[SZ]] [[BP1:%[^,]+]], i[[SZ]]* [[CBPADDR1]] + // CHECK-DAG: store i[[SZ]] [[BP1]], i[[SZ]]* [[CPADDR1]] + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[IN]], i32 [[GTID]], i32 1, i[[SZ]] {{104|52}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1_:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1_:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + + // CHECK: [[ELSE]]: + // CHECK-NOT: getelementptr inbounds [2 x i8*], [2 x i8*]* + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[IN]], i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1__:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1__:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + // CHECK: [[EXIT]]: + + #pragma omp target parallel for simd device(global + a) nowait depend(inout: global, a, bn) if(target:a) + for (int i = 0; i < *plocal; ++i) { + static int local1; + *plocal = global; + local1 = global; + } + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[IN]], i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]* + // CHECK: getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* [[IN]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[IN]], i32 [[GTID]], i8* [[TASK]]) + #pragma omp target parallel for simd if(0) firstprivate(global) depend(out:global) + for (int i = 0; i < global; ++i) { + global += 1; + } + + return a; +} + +// Check that the offloading functions are emitted and that the arguments are +// correct and loaded correctly for the target regions in foo(). + +// CHECK: define internal void [[HVT0:@.+]]() + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, [[TASK_TY0]]* noalias) +// CHECK: store void (i8*, ...)* null, void (i8*, ...)** % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 0, i8** null, i8** null, i[[SZ]]* null, i64* null, i32 1, i32 0) +// CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: call void [[HVT0]]() +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT1:@.+]](i[[SZ]]* %{{.+}}, i[[SZ]] %{{.+}}) + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1_]](i32{{.*}}, [[TASK_TY1_]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[SZT:%.+]] = getelementptr inbounds [2 x i[[SZ]]], [2 x i[[SZ]]]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams_nowait(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 2, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SZT]], i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT]], i32 0, i32 0), i32 1, i32 0) + +// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1__]](i32{{.*}}, [[TASK_TY1__]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT2:@.+]](i[[SZ]] %{{.+}}) +// Create stack storage and store argument in there. +// CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align +// CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align +// CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* +// CHECK-64: load i32, i32* [[AA_CADDR]], align +// CHECK-32: load i32, i32* [[AA_ADDR]], align + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, [[TASK_TY2]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT2]](i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + + +#endif diff --git a/test/OpenMP/target_parallel_for_simd_depend_messages.cpp b/test/OpenMP/target_parallel_for_simd_depend_messages.cpp index 4375941036cf9..5a735fc77afea 100644 --- a/test/OpenMP/target_parallel_for_simd_depend_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_depend_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_device_messages.cpp b/test/OpenMP/target_parallel_for_simd_device_messages.cpp index 2c9d43f07b257..8d2719571aba1 100644 --- a/test/OpenMP/target_parallel_for_simd_device_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_device_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_firstprivate_messages.cpp b/test/OpenMP/target_parallel_for_simd_firstprivate_messages.cpp index 9ad255f033e43..8a32c6ffc28aa 100644 --- a/test/OpenMP/target_parallel_for_simd_firstprivate_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_if_messages.cpp b/test/OpenMP/target_parallel_for_simd_if_messages.cpp index 640e24c1eae49..fea1ad9778f54 100644 --- a/test/OpenMP/target_parallel_for_simd_if_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_is_device_ptr_ast_print.cpp b/test/OpenMP/target_parallel_for_simd_is_device_ptr_ast_print.cpp index af98f99ada959..feeaa1ef72acb 100644 --- a/test/OpenMP/target_parallel_for_simd_is_device_ptr_ast_print.cpp +++ b/test/OpenMP/target_parallel_for_simd_is_device_ptr_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -56,7 +60,7 @@ struct SA { // CHECK-NEXT: int aa[10]; // CHECK-NEXT: arr &raa = this->aa; // CHECK-NEXT: func( -// CHECK-NEXT: #pragma omp target parallel for simd is_device_ptr(this->k) +// CHECK-NEXT: #pragma omp target parallel for simd is_device_ptr(this->k){{$}} // CHECK-NEXT: for (int i = 0; i < 100; i++) // CHECK-NEXT: ; // CHECK-NEXT: #pragma omp target parallel for simd is_device_ptr(this->z) diff --git a/test/OpenMP/target_parallel_for_simd_is_device_ptr_messages.cpp b/test/OpenMP/target_parallel_for_simd_is_device_ptr_messages.cpp index 435b9a95e9e61..1079bf3cbb847 100644 --- a/test/OpenMP/target_parallel_for_simd_is_device_ptr_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_is_device_ptr_messages.cpp @@ -1,4 +1,6 @@ // RUN: %clang_cc1 -std=c++11 -verify -fopenmp %s + +// RUN: %clang_cc1 -std=c++11 -verify -fopenmp-simd %s struct ST { int *a; }; diff --git a/test/OpenMP/target_parallel_for_simd_lastprivate_messages.cpp b/test/OpenMP/target_parallel_for_simd_lastprivate_messages.cpp index 313192dbb5f66..7c2c23f9d70d8 100644 --- a/test/OpenMP/target_parallel_for_simd_lastprivate_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_linear_messages.cpp b/test/OpenMP/target_parallel_for_simd_linear_messages.cpp index e17f15550e653..40fd052df2e3d 100644 --- a/test/OpenMP/target_parallel_for_simd_linear_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/target_parallel_for_simd_loop_messages.cpp b/test/OpenMP/target_parallel_for_simd_loop_messages.cpp index f07f19e2a6903..fabf389150d09 100644 --- a/test/OpenMP/target_parallel_for_simd_loop_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_loop_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s -Wno-openmp-target + +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s -Wno-openmp-target class S { int a; diff --git a/test/OpenMP/target_parallel_for_simd_map_messages.cpp b/test/OpenMP/target_parallel_for_simd_map_messages.cpp index 7e95e10ae5caa..1cab5988396f0 100644 --- a/test/OpenMP/target_parallel_for_simd_map_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_map_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_messages.cpp b/test/OpenMP/target_parallel_for_simd_messages.cpp index 0e1a0fea7e619..c0e03289d7503 100644 --- a/test/OpenMP/target_parallel_for_simd_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_misc_messages.c b/test/OpenMP/target_parallel_for_simd_misc_messages.c index f2574a35472d9..8886ddd4a1e0b 100644 --- a/test/OpenMP/target_parallel_for_simd_misc_messages.c +++ b/test/OpenMP/target_parallel_for_simd_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp target parallel for simd'}} #pragma omp target parallel for simd diff --git a/test/OpenMP/target_parallel_for_simd_nowait_messages.cpp b/test/OpenMP/target_parallel_for_simd_nowait_messages.cpp index 3c4b512048177..5e1e13168d5b4 100644 --- a/test/OpenMP/target_parallel_for_simd_nowait_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_nowait_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_num_threads_messages.cpp b/test/OpenMP/target_parallel_for_simd_num_threads_messages.cpp index 1076b86ea5e67..b3323f67728fb 100644 --- a/test/OpenMP/target_parallel_for_simd_num_threads_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_num_threads_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_ordered_messages.cpp b/test/OpenMP/target_parallel_for_simd_ordered_messages.cpp index 065d81517f837..3e3f6c26c3430 100644 --- a/test/OpenMP/target_parallel_for_simd_ordered_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_ordered_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_private_messages.cpp b/test/OpenMP/target_parallel_for_simd_private_messages.cpp index 57262a58598c5..51f1d9b8be9e3 100644 --- a/test/OpenMP/target_parallel_for_simd_private_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } @@ -43,7 +45,7 @@ public: S5(int v) : a(v) {} S5 &operator=(S5 &s) { #pragma omp target parallel for simd private(a) private(this->a) private(s.a) // expected-error {{expected variable name or data member of current class}} - for (int k = 0; k < s.a; ++k) + for (int k = 0; k < s.a; ++k) // expected-warning {{Non-trivial type 'S5' is mapped, only trivial types are guaranteed to be mapped correctly}} ++s.a; return *this; } diff --git a/test/OpenMP/target_parallel_for_simd_proc_bind_messages.cpp b/test/OpenMP/target_parallel_for_simd_proc_bind_messages.cpp index 5bb6d92cd7e9e..83b4efad15bd0 100644 --- a/test/OpenMP/target_parallel_for_simd_proc_bind_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_proc_bind_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/target_parallel_for_simd_reduction_messages.cpp b/test/OpenMP/target_parallel_for_simd_reduction_messages.cpp index 72e2130f3df4c..30aaf832d6fe3 100644 --- a/test/OpenMP/target_parallel_for_simd_reduction_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_safelen_messages.cpp b/test/OpenMP/target_parallel_for_simd_safelen_messages.cpp index f99010179dc30..74b1a7374f0f9 100644 --- a/test/OpenMP/target_parallel_for_simd_safelen_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_safelen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_schedule_messages.cpp b/test/OpenMP/target_parallel_for_simd_schedule_messages.cpp index f0d86e952de6a..6f26e7aac4336 100644 --- a/test/OpenMP/target_parallel_for_simd_schedule_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_parallel_for_simd_simdlen_messages.cpp b/test/OpenMP/target_parallel_for_simd_simdlen_messages.cpp index e51e67b6b45f2..f8b375ef1c6c4 100644 --- a/test/OpenMP/target_parallel_for_simd_simdlen_messages.cpp +++ b/test/OpenMP/target_parallel_for_simd_simdlen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_if_codegen.cpp b/test/OpenMP/target_parallel_if_codegen.cpp index c155bf7c051aa..aab1a14a5eb5f 100644 --- a/test/OpenMP/target_parallel_if_codegen.cpp +++ b/test/OpenMP/target_parallel_if_codegen.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 @@ -16,13 +24,23 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER -// CHECK-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CHECK-DAG: [[S1:%.+]] = type { double } // CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } @@ -35,19 +53,19 @@ // We have 6 target regions -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] // Check if offloading descriptor is created. // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] @@ -58,7 +76,7 @@ // CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) // Check target registration is registered as a Ctor. -// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] template<typename tx> @@ -145,7 +163,7 @@ int bar(int n){ // CHECK: store i8 [[FB]], i8* [[CONV]], align // CHECK: [[ARG:%.+]] = load i[[SZ]], i[[SZ]]* [[CAPEC_ADDR]], align // -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 3, {{.*}}, i32 1, i32 0) +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 4, {{.*}}, i32 1, i32 0) // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 // CHECK: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] // @@ -171,7 +189,7 @@ int bar(int n){ // CHECK: br i1 [[CMP]], label {{%?}}[[IF_THEN:.+]], label {{%?}}[[IF_ELSE:.+]] // // CHECK: [[IF_THEN]] -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 2, {{.*}}, i32 1, i32 0) +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 3, {{.*}}, i32 1, i32 0) // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] // CHECK: [[FAIL]] @@ -291,7 +309,7 @@ int bar(int n){ // CHECK: br i1 [[TB]], label {{%?}}[[IF_THEN:.+]], label {{%?}}[[IF_ELSE:.+]] // // CHECK: [[IF_THEN]] -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), [[S1]]* {{.+}}, i[[SZ]] [[ARG]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), [[S1]]* {{.+}}, i[[SZ]] [[ARG]]) // CHECK: br label {{%?}}[[END:.+]] // // CHECK: [[IF_ELSE]] @@ -313,7 +331,7 @@ int bar(int n){ // CHECK: br i1 [[TB]], label {{%?}}[[IF_THEN:.+]], label {{%?}}[[IF_ELSE:.+]] // // CHECK: [[IF_THEN]] -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*)* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), [[S1]]* {{.+}}) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*)* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), [[S1]]* {{.+}}) // CHECK: br label {{%?}}[[END:.+]] // // CHECK: [[IF_ELSE]] @@ -341,7 +359,7 @@ int bar(int n){ // CHECK: br i1 [[TB]], label {{%?}}[[IF_THEN:.+]], label {{%?}}[[IF_ELSE:.+]] // // CHECK: [[IF_THEN]] -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*)) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*)) // CHECK: br label {{%?}}[[END:.+]] // // CHECK: [[IF_ELSE]] @@ -354,7 +372,7 @@ int bar(int n){ // // // CHECK: define internal void [[HVT4]]() -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*)) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*)) // CHECK-NEXT: ret // // @@ -377,7 +395,7 @@ int bar(int n){ // CHECK-NOT: call void @__kmpc_serialized_parallel( // CHECK-NOT: call void [[OMP_OUTLINED5:@.+]](i32* {{%.+}}, i32* {{%.+}}, i[[SZ]] {{.+}}) // CHECK-NOT: call void @__kmpc_end_serialized_parallel( -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), // CHECK: ret // // diff --git a/test/OpenMP/target_parallel_if_messages.cpp b/test/OpenMP/target_parallel_if_messages.cpp index 06b7a59b7e59c..d59b4ab5074b2 100644 --- a/test/OpenMP/target_parallel_if_messages.cpp +++ b/test/OpenMP/target_parallel_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_is_device_ptr_ast_print.cpp b/test/OpenMP/target_parallel_is_device_ptr_ast_print.cpp index 7fbddd8622ae0..16cc188e872e5 100644 --- a/test/OpenMP/target_parallel_is_device_ptr_ast_print.cpp +++ b/test/OpenMP/target_parallel_is_device_ptr_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -50,7 +54,7 @@ struct SA { // CHECK-NEXT: int aa[10]; // CHECK-NEXT: arr &raa = this->aa; // CHECK-NEXT: func( -// CHECK-NEXT: #pragma omp target parallel is_device_ptr(this->k) +// CHECK-NEXT: #pragma omp target parallel is_device_ptr(this->k){{$}} // CHECK-NEXT: { // CHECK-NEXT: } // CHECK-NEXT: #pragma omp target parallel is_device_ptr(this->z) diff --git a/test/OpenMP/target_parallel_is_device_ptr_messages.cpp b/test/OpenMP/target_parallel_is_device_ptr_messages.cpp index d9c3ee907b615..b46fd713ff638 100644 --- a/test/OpenMP/target_parallel_is_device_ptr_messages.cpp +++ b/test/OpenMP/target_parallel_is_device_ptr_messages.cpp @@ -1,4 +1,6 @@ // RUN: %clang_cc1 -std=c++11 -verify -fopenmp -ferror-limit 200 %s + +// RUN: %clang_cc1 -std=c++11 -verify -fopenmp-simd -ferror-limit 200 %s struct ST { int *a; }; @@ -38,7 +40,7 @@ struct SA { #pragma omp target parallel is_device_ptr(aa) // OK {} #pragma omp target parallel is_device_ptr(raa) // OK - {} + {} #pragma omp target parallel is_device_ptr(e) // expected-error{{expected pointer, array, reference to pointer, or reference to array in 'is_device_ptr clause'}} {} #pragma omp target parallel is_device_ptr(g) // OK diff --git a/test/OpenMP/target_parallel_map_messages.cpp b/test/OpenMP/target_parallel_map_messages.cpp index d3cc742decd56..4bb14ea52f03e 100644 --- a/test/OpenMP/target_parallel_map_messages.cpp +++ b/test/OpenMP/target_parallel_map_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/target_parallel_messages.cpp b/test/OpenMP/target_parallel_messages.cpp index b6763d8f09e03..a7dfbf91ca536 100644 --- a/test/OpenMP/target_parallel_messages.cpp +++ b/test/OpenMP/target_parallel_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -o - %s // RUN: not %clang_cc1 -fopenmp -std=c++11 -fopenmp-targets=aaa-bbb-ccc-ddd -o - %s 2>&1 | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -o - %s // CHECK: error: OpenMP target is invalid: 'aaa-bbb-ccc-ddd' void foo() { diff --git a/test/OpenMP/target_parallel_nowait_messages.cpp b/test/OpenMP/target_parallel_nowait_messages.cpp index 91e26f20819f6..8da03e4756ab4 100644 --- a/test/OpenMP/target_parallel_nowait_messages.cpp +++ b/test/OpenMP/target_parallel_nowait_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_parallel_num_threads_codegen.cpp b/test/OpenMP/target_parallel_num_threads_codegen.cpp index 836a92d9a5db9..e6b93113c5571 100644 --- a/test/OpenMP/target_parallel_num_threads_codegen.cpp +++ b/test/OpenMP/target_parallel_num_threads_codegen.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 @@ -16,13 +24,23 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER -// CHECK-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CHECK-DAG: [[S1:%.+]] = type { double } // CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } @@ -35,19 +53,19 @@ // We have 6 target regions -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] // Check if offloading descriptor is created. // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] @@ -58,7 +76,7 @@ // CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) // Check target registration is registered as a Ctor. -// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] template<typename tx> @@ -147,7 +165,7 @@ int bar(int n){ // CHECK: [[ARG:%.+]] = load i[[SZ]], i[[SZ]]* [[CAPEC_ADDR]], align // CHECK: [[THREADS:%.+]] = load i32, i32* [[CAPE_ADDR]], align // -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 3, {{.*}}, i32 1, i32 [[THREADS]]) +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 4, {{.*}}, i32 1, i32 [[THREADS]]) // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 // CHECK: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] // @@ -158,7 +176,7 @@ int bar(int n){ // // // -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 1, {{.+}}, i32 1, i32 1024) +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 2, {{.+}}, i32 1, i32 1024) // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 // CHECK: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] // @@ -270,15 +288,15 @@ int bar(int n){ // CHECK-64: [[CONV:%.+]] = bitcast i[[SZ]]* [[CAPE_ADDR]] to i32* // CHECK-64: [[NT:%.+]] = load i32, i32* [[CONV]], align // CHECK-32: [[NT:%.+]] = load i32, i32* [[CAPE_ADDR]], align -// CHECK: call void @__kmpc_push_num_threads(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]]) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 2, +// CHECK: call void @__kmpc_push_num_threads(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 2, // // // CHECK: define internal void [[HVT2]]([[S1]]* {{%.+}}) -// CHECK: call void @__kmpc_push_num_threads(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 1024) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 1, +// CHECK: call void @__kmpc_push_num_threads(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 1024) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 1, // // @@ -294,8 +312,8 @@ int bar(int n){ // CHECK-64: [[CONV:%.+]] = bitcast i[[SZ]]* [[CAPE_ADDR]] to i32* // CHECK-64: [[NT:%.+]] = load i32, i32* [[CONV]], align // CHECK-32: [[NT:%.+]] = load i32, i32* [[CAPE_ADDR]], align -// CHECK: call void @__kmpc_push_num_threads(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]]) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 0, +// CHECK: call void @__kmpc_push_num_threads(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 0, // // // CHECK: define internal void [[HVT4]](i[[SZ]] [[PARM:%.+]]) @@ -303,8 +321,8 @@ int bar(int n){ // CHECK-64: [[CONV:%.+]] = bitcast i[[SZ]]* [[CAPE_ADDR]] to i32* // CHECK-64: [[NT:%.+]] = load i32, i32* [[CONV]], align // CHECK-32: [[NT:%.+]] = load i32, i32* [[CAPE_ADDR]], align -// CHECK: call void @__kmpc_push_num_threads(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]]) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 0, +// CHECK: call void @__kmpc_push_num_threads(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 0, // // @@ -313,8 +331,8 @@ int bar(int n){ // CHECK: define internal void [[HVT5]]( -// CHECK: call void @__kmpc_push_num_threads(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 20) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 0, +// CHECK: call void @__kmpc_push_num_threads(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 20) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 0, // // @@ -324,8 +342,8 @@ int bar(int n){ // CHECK: [[CONV:%.+]] = bitcast i[[SZ]]* [[CAPE_ADDR]] to i16* // CHECK: [[T:%.+]] = load i16, i16* [[CONV]], align // CHECK: [[NT:%.+]] = sext i16 [[T]] to i32 -// CHECK: call void @__kmpc_push_num_threads(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]]) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* [[DEF_LOC]], i32 2, +// CHECK: call void @__kmpc_push_num_threads(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEF_LOC]], i32 2, // // diff --git a/test/OpenMP/target_parallel_num_threads_messages.cpp b/test/OpenMP/target_parallel_num_threads_messages.cpp index 95797caf29622..efc1d1327f5ab 100644 --- a/test/OpenMP/target_parallel_num_threads_messages.cpp +++ b/test/OpenMP/target_parallel_num_threads_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/target_parallel_private_messages.cpp b/test/OpenMP/target_parallel_private_messages.cpp index fabd37d4b3950..34652d6cb3cef 100644 --- a/test/OpenMP/target_parallel_private_messages.cpp +++ b/test/OpenMP/target_parallel_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_parallel_proc_bind_messages.cpp b/test/OpenMP/target_parallel_proc_bind_messages.cpp index 56292ad437032..a01b63d7fef0c 100644 --- a/test/OpenMP/target_parallel_proc_bind_messages.cpp +++ b/test/OpenMP/target_parallel_proc_bind_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/target_parallel_reduction_messages.cpp b/test/OpenMP/target_parallel_reduction_messages.cpp index 038f099c5a6c4..6ea08848d5132 100644 --- a/test/OpenMP/target_parallel_reduction_messages.cpp +++ b/test/OpenMP/target_parallel_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/target_parallel_shared_messages.cpp b/test/OpenMP/target_parallel_shared_messages.cpp index 302a09239d341..2ade722dd3a2a 100644 --- a/test/OpenMP/target_parallel_shared_messages.cpp +++ b/test/OpenMP/target_parallel_shared_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/target_private_codegen.cpp b/test/OpenMP/target_private_codegen.cpp index 5c738ee5930f9..213a20fec3aa6 100644 --- a/test/OpenMP/target_private_codegen.cpp +++ b/test/OpenMP/target_private_codegen.cpp @@ -9,6 +9,16 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -35,28 +45,28 @@ int foo(int n) { { } - // TCHECK: define void @__omp_offloading_{{.+}}() + // TCHECK: define weak void @__omp_offloading_{{.+}}() // TCHECK: [[A:%.+]] = alloca i{{[0-9]+}}, // TCHECK-NOT: store {{.+}}, {{.+}} [[A]], - // TCHECK: ret void + // TCHECK: ret void #pragma omp target private(a) { a = 1; } - // TCHECK: define void @__omp_offloading_{{.+}}() + // TCHECK: define weak void @__omp_offloading_{{.+}}() // TCHECK: [[A:%.+]] = alloca i{{[0-9]+}}, // TCHECK: store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A]], // TCHECK: ret void - - #pragma omp target private(a, aa) + +#pragma omp target private(a, aa) { a = 1; aa = 1; } - // TCHECK: define void @__omp_offloading_{{.+}}() + // TCHECK: define weak void @__omp_offloading_{{.+}}() // TCHECK: [[A:%.+]] = alloca i{{[0-9]+}}, // TCHECK: [[A2:%.+]] = alloca i{{[0-9]+}}, // TCHECK: store i{{[0-9]+}} 1, i{{[0-9]+}}* [[A]], @@ -75,7 +85,7 @@ int foo(int n) { } // make sure that private variables are generated in all cases and that we use those instances for operations inside the // target region - // TCHECK: define void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[VLA:%.+]], i{{[0-9]+}} [[VLA1:%.+]], i{{[0-9]+}} [[VLA3:%.+]]) + // TCHECK: define weak void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[VLA:%.+]], i{{[0-9]+}} [[VLA1:%.+]], i{{[0-9]+}} [[VLA3:%.+]]) // TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}}, // TCHECK: [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}}, // TCHECK: [[VLA_ADDR4:%.+]] = alloca i{{[0-9]+}}, @@ -102,7 +112,7 @@ int foo(int n) { // b[2] = 1.0 // TCHECK: [[B_GEP:%.+]] = getelementptr inbounds [10 x float], [10 x float]* [[B]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 // TCHECK: store float 1.0{{.*}}, float* [[B_GEP]], - + // bn[3] = 1.0 // TCHECK: [[BN_GEP:%.+]] = getelementptr inbounds float, float* [[VLA5]], i{{[0-9]+}} 3 // TCHECK: store float 1.0{{.*}}, float* [[BN_GEP]], @@ -121,7 +131,7 @@ int foo(int n) { // d.X = 1 // [[X_FIELD:%.+]] = getelementptr inbounds [[TT]] [[TT]]* [[D]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // store i{{[0-9]+}} 1, i{{[0-9]+}}* [[X_FIELD]], - + // d.Y = 1 // [[Y_FIELD:%.+]] = getelementptr inbounds [[TT]] [[TT]]* [[D]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 // store i{{[0-9]+}} 1, i{{[0-9]+}}* [[Y_FIELD]], @@ -169,7 +179,7 @@ int fstatic(int n) { return a; } -// TCHECK: define void @__omp_offloading_{{.+}}() +// TCHECK: define weak void @__omp_offloading_{{.+}}() // TCHECK: [[A:%.+]] = alloca i{{[0-9]+}}, // TCHECK: [[A2:%.+]] = alloca i{{[0-9]+}}, // TCHECK: [[A3:%.+]] = alloca i{{[0-9]+}}, @@ -197,7 +207,7 @@ struct S1 { return c[1][1] + (int)b; } - // TCHECK: define void @__omp_offloading_{{.+}}([[S1]]* [[TH:%.+]], i{{[0-9]+}} [[VLA:%.+]], i{{[0-9]+}} [[VLA1:%.+]]) + // TCHECK: define weak void @__omp_offloading_{{.+}}([[S1]]* [[TH:%.+]], i{{[0-9]+}} [[VLA:%.+]], i{{[0-9]+}} [[VLA1:%.+]]) // TCHECK: [[TH_ADDR:%.+]] = alloca [[S1]]*, // TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}}, // TCHECK: [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}}, @@ -251,7 +261,7 @@ int bar(int n){ } // template -// TCHECK: define void @__omp_offloading_{{.+}}() +// TCHECK: define weak void @__omp_offloading_{{.+}}() // TCHECK: [[A:%.+]] = alloca i{{[0-9]+}}, // TCHECK: [[A2:%.+]] = alloca i{{[0-9]+}}, // TCHECK: [[B:%.+]] = alloca [10 x i{{[0-9]+}}], diff --git a/test/OpenMP/target_private_messages.cpp b/test/OpenMP/target_private_messages.cpp index a093a87056272..49ee3fed3e94b 100644 --- a/test/OpenMP/target_private_messages.cpp +++ b/test/OpenMP/target_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + struct S1; // expected-note 2 {{declared here}} expected-note 2 {{forward declaration of 'S1'}} extern S1 a; class S2 { @@ -36,7 +38,7 @@ public: S5(int v) : a(v) {} S5 &operator=(S5 &s) { #pragma omp target private(a) private(this->a) private(s.a) // expected-error {{expected variable name or data member of current class}} - for (int k = 0; k < s.a; ++k) + for (int k = 0; k < s.a; ++k) // expected-warning {{Non-trivial type 'S5' is mapped, only trivial types are guaranteed to be mapped correctly}} ++s.a; return *this; } diff --git a/test/OpenMP/target_reduction_codegen.cpp b/test/OpenMP/target_reduction_codegen.cpp index f86719b55fa1c..38508006872ec 100644 --- a/test/OpenMP/target_reduction_codegen.cpp +++ b/test/OpenMP/target_reduction_codegen.cpp @@ -9,6 +9,16 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -35,7 +45,7 @@ int foo(int n) { { } - // TCHECK: define void @__omp_offloading_{{.+}}(i32*{{.+}} %{{.+}}) + // TCHECK: define weak void @__omp_offloading_{{.+}}(i32*{{.+}} %{{.+}}) // TCHECK: [[A:%.+]] = alloca i{{[0-9]+}}*, // TCHECK: store {{.+}}, {{.+}} [[A]], // TCHECK: load i32*, i32** [[A]], @@ -46,7 +56,7 @@ int foo(int n) { a = 1; } - // TCHECK: define void @__omp_offloading_{{.+}}(i32*{{.+}} %{{.+}}) + // TCHECK: define weak void @__omp_offloading_{{.+}}(i32*{{.+}} %{{.+}}) // TCHECK: [[A:%.+]] = alloca i{{[0-9]+}}*, // TCHECK: store {{.+}}, {{.+}} [[A]], // TCHECK: [[REF:%.+]] = load i32*, i32** [[A]], @@ -59,7 +69,7 @@ int foo(int n) { aa = 1; } - // TCHECK: define void @__omp_offloading_{{.+}}(i32*{{.+}} [[A:%.+]], i16*{{.+}} [[AA:%.+]]) + // TCHECK: define weak void @__omp_offloading_{{.+}}(i32*{{.+}} [[A:%.+]], i16*{{.+}} [[AA:%.+]]) // TCHECK: [[A:%.+]] = alloca i{{[0-9]+}}*, // TCHECK: [[AA:%.+]] = alloca i{{[0-9]+}}*, // TCHECK: store {{.+}}, {{.+}} [[A]], @@ -108,7 +118,7 @@ int fstatic(int n) { return a; } -// TCHECK: define void @__omp_offloading_{{.+}}(i32*{{.+}}, i16*{{.+}}, i8*{{.+}}, [10 x i32]*{{.+}}) +// TCHECK: define weak void @__omp_offloading_{{.+}}(i32*{{.+}}, i16*{{.+}}, i8*{{.+}}, [10 x i32]*{{.+}}) // TCHECK: [[A:%.+]] = alloca i{{[0-9]+}}*, // TCHECK: [[A2:%.+]] = alloca i{{[0-9]+}}*, // TCHECK: [[A3:%.+]] = alloca i{{[0-9]+}}*, @@ -144,7 +154,7 @@ struct S1 { return c[1][1] + (int)b; } - // TCHECK: define void @__omp_offloading_{{.+}}([[S1]]* [[TH:%.+]], i32*{{.+}}, i{{[0-9]+}} [[VLA:%.+]], i{{[0-9]+}} [[VLA1:%.+]], i16*{{.+}}) + // TCHECK: define weak void @__omp_offloading_{{.+}}([[S1]]* [[TH:%.+]], i32*{{.+}}, i{{[0-9]+}} [[VLA:%.+]], i{{[0-9]+}} [[VLA1:%.+]], i16*{{.+}}) // TCHECK: [[TH_ADDR:%.+]] = alloca [[S1]]*, // TCHECK: [[B_ADDR:%.+]] = alloca i{{[0-9]+}}*, // TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}}, @@ -196,7 +206,7 @@ int bar(int n){ } // template -// TCHECK: define void @__omp_offloading_{{.+}}(i{{[0-9]+}}*{{.+}}, i{{[0-9]+}}*{{.+}}, [10 x i32]*{{.+}}) +// TCHECK: define weak void @__omp_offloading_{{.+}}(i{{[0-9]+}}*{{.+}}, i{{[0-9]+}}*{{.+}}, [10 x i32]*{{.+}}) // TCHECK: [[A:%.+]] = alloca i{{[0-9]+}}*, // TCHECK: [[A2:%.+]] = alloca i{{[0-9]+}}*, // TCHECK: [[B:%.+]] = alloca [10 x i{{[0-9]+}}]*, diff --git a/test/OpenMP/target_reduction_messages.cpp b/test/OpenMP/target_reduction_messages.cpp index 8cf79f9a5d993..ff936b8c1950d 100644 --- a/test/OpenMP/target_reduction_messages.cpp +++ b/test/OpenMP/target_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/target_simd_aligned_messages.cpp b/test/OpenMP/target_simd_aligned_messages.cpp index 76a37f64984d0..d2628b435796a 100644 --- a/test/OpenMP/target_simd_aligned_messages.cpp +++ b/test/OpenMP/target_simd_aligned_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp %s +// RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp-simd %s + struct B { static int ib[20]; // expected-note 0 {{'B::ib' declared here}} static constexpr int bfoo() { return 8; } diff --git a/test/OpenMP/target_simd_ast_print.cpp b/test/OpenMP/target_simd_ast_print.cpp index 01ce1d21cfbcc..7d3daa36f30ce 100644 --- a/test/OpenMP/target_simd_ast_print.cpp +++ b/test/OpenMP/target_simd_ast_print.cpp @@ -1,6 +1,10 @@ -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s -Wno-openmp-target | FileCheck %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print -Wno-openmp-target | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s -Wno-openmp-target | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print -Wno-openmp-target | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -35,7 +39,7 @@ public: } }; -// CHECK: #pragma omp target simd private(this->a) private(this->a) private(T::a) +// CHECK: #pragma omp target simd private(this->a) private(this->a) private(T::a){{$}} // CHECK: #pragma omp target simd private(this->a) private(this->a) // CHECK: #pragma omp target simd private(this->a) private(this->a) private(this->S::a) diff --git a/test/OpenMP/target_simd_codegen.cpp b/test/OpenMP/target_simd_codegen.cpp index d57e3818559a3..a2ec51c4e22fd 100644 --- a/test/OpenMP/target_simd_codegen.cpp +++ b/test/OpenMP/target_simd_codegen.cpp @@ -1,20 +1,38 @@ // Test host codegen. -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER @@ -35,31 +53,31 @@ // sizes. // CHECK-DAG: [[SIZET2:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 2, i[[SZ]] 4, i[[SZ]] 4] -// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 288] +// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 800] // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2] -// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 288, i64 288] -// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [9 x i64] [i64 288, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547] +// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 800] +// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [9 x i64] [i64 800, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547] // CHECK-DAG: [[SIZET5:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 40] -// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 547] +// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 547] // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [4 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 1, i[[SZ]] 40] -// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [4 x i64] [i64 288, i64 288, i64 288, i64 547] -// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [5 x i64] [i64 547, i64 288, i64 288, i64 288, i64 547] -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 - -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK-NOT: @{{.+}} = constant [[ENTTY]] +// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [4 x i64] [i64 800, i64 800, i64 800, i64 547] +// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [6 x i64] [i64 32, i64 281474976711171, i64 800, i64 288, i64 288, i64 547] +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] // Check if offloading descriptor is created. // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] @@ -70,7 +88,7 @@ // CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) // Check target registration is registered as a Ctor. -// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] template<typename tx, typename ty> @@ -192,13 +210,13 @@ int foo(int n) { // CHECK-32: store i32 [[A_VAL]], i32* [[A_CADDR:%.+]], // CHECK-32: [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]], + // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 + // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] + // CHECK: [[TRY]] // CHECK: [[BNSIZE:%.+]] = mul nuw i[[SZ]] [[VLA0:%.+]], 4 // CHECK: [[CNELEMSIZE2:%.+]] = mul nuw i[[SZ]] 5, [[VLA1:%.+]] // CHECK: [[CNSIZE:%.+]] = mul nuw i[[SZ]] [[CNELEMSIZE2]], 8 - // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 - // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] - // CHECK: [[TRY]] // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target(i64 -1, i8* @{{[^,]+}}, i32 9, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([9 x i64], [9 x i64]* [[MAPT4]], i32 0, i32 0)) // CHECK-DAG: [[BPR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 // CHECK-DAG: [[PR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P:%[^,]+]], i32 0, i32 0 @@ -460,32 +478,39 @@ int bar(int n){ // CHECK-64: store i32 %{{.+}}, i32* [[B_ADDR]], // CHECK-64: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_CADDR]], +// CHECK-32: store i32 %{{.+}}, i32* %__vla_expr // CHECK-32: store i32 %{{.+}}, i32* [[B_ADDR:%.+]], // CHECK-32: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_ADDR]], +// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 +// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] +// CHECK: [[TRY]] // We capture 2 VLA sizes in this target region // CHECK: [[CELEMSIZE2:%.+]] = mul nuw i[[SZ]] 2, [[VLA0:%.+]] // CHECK: [[CSIZE:%.+]] = mul nuw i[[SZ]] [[CELEMSIZE2]], 2 -// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 -// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] -// CHECK: [[TRY]] -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target(i64 -1, i8* @{{[^,]+}}, i32 5, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([5 x i64], [5 x i64]* [[MAPT7]], i32 0, i32 0)) -// CHECK-DAG: [[BPR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP:%.+]], i32 0, i32 0 -// CHECK-DAG: [[PR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SR]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] -// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX0]] -// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX0]] -// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] -// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX1]] -// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX1]] -// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] -// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX2]] -// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX2]] -// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] -// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX3]] -// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target(i64 -1, i8* @{{[^,]+}}, i32 6, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([6 x i64], [6 x i64]* [[MAPT7]], i32 0, i32 0)) +// CHECK-DAG: [[BPR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP:%.+]], i32 0, i32 0 +// CHECK-DAG: [[PR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SR]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] +// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX0]] +// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX0]] +// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] +// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX1]] +// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX1]] +// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] +// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX2]] +// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX2]] +// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] +// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] +// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX4:[0-9]+]] +// CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX4]] +// CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX4]] +// CHECK-DAG: [[SADDR5:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX5:[0-9]+]] +// CHECK-DAG: [[BPADDR5:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX5]] +// CHECK-DAG: [[PADDR5:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX5]] // The names below are not necessarily consistent with the names used for the // addresses above as some are repeated. @@ -508,15 +533,21 @@ int bar(int n){ // CHECK-DAG: store i[[SZ]] 4, i[[SZ]]* {{%[^,]+}} // CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR3:%.+]], -// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CPADDR3:%.+]], +// CHECK-DAG: store double* %{{.+}}, double** [[CPADDR3:%.+]], // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** -// CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** +// CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to double** +// CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* {{%[^,]+}} + +// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR4:%.+]], +// CHECK-DAG: store double* %{{.+}}, double** [[CPADDR4:%.+]], +// CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to [[S1]]** +// CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to double** // CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* {{%[^,]+}} -// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR4:%.+]], -// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR4:%.+]], -// CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to i16** -// CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to i16** +// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR5:%.+]], +// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR5:%.+]], +// CHECK-DAG: [[CBPADDR5]] = bitcast i8** {{%[^,]+}} to i16** +// CHECK-DAG: [[CPADDR5]] = bitcast i8** {{%[^,]+}} to i16** // CHECK-DAG: store i[[SZ]] [[CSIZE]], i[[SZ]]* {{%[^,]+}} // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 diff --git a/test/OpenMP/target_simd_codegen_registration.cpp b/test/OpenMP/target_simd_codegen_registration.cpp index a6e6e9b385650..39e930a5d5547 100644 --- a/test/OpenMP/target_simd_codegen_registration.cpp +++ b/test/OpenMP/target_simd_codegen_registration.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target simd codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,9 +24,22 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK -// Check that no target code is emmitted if no omptests flag was provided. +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// Check that no target code is emitted if no omptests flag was provided. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK-NTARGET +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -60,96 +81,96 @@ // We have 7 target regions -// CHECK-DAG: {{@.+}} = private constant i8 0 -// TCHECK-NOT: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// TCHECK-NOT: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] -// CHECK-NTARGET-NOT: private constant i8 0 +// CHECK-NTARGET-NOT: weak constant i8 0 // CHECK-NTARGET-NOT: private unnamed_addr constant [1 x i // CHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// CHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// CHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// CHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// CHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// CHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// CHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// CHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// CHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// CHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// CHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// CHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// CHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// TCHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// TCHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// TCHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// TCHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// TCHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// TCHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// TCHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// TCHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// TCHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// TCHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// TCHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// TCHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] // CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] @@ -163,7 +184,7 @@ // CHECK-SAME: { i32, void ()*, i8* } { i32 500, void ()* [[P500:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 501, void ()* [[P501:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 65535, void ()* [[PMAX:@[^,]+]], i8* null }, -// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] // CHECK-NTARGET: @llvm.global_ctors = appending global [3 x { i32, void ()*, i8* }] [ @@ -328,18 +349,18 @@ struct ST { //CHECK-DAG: define internal void @[[NAME12]]( //CHECK-DAG: call void @[[NAME12]]( -//TCHECK-DAG: define void @[[NAME1]]( -//TCHECK-DAG: define void @[[NAME2]]( -//TCHECK-DAG: define void @[[NAME3]]( -//TCHECK-DAG: define void @[[NAME4]]( -//TCHECK-DAG: define void @[[NAME5]]( -//TCHECK-DAG: define void @[[NAME6]]( -//TCHECK-DAG: define void @[[NAME7]]( -//TCHECK-DAG: define void @[[NAME8]]( -//TCHECK-DAG: define void @[[NAME9]]( -//TCHECK-DAG: define void @[[NAME10]]( -//TCHECK-DAG: define void @[[NAME11]]( -//TCHECK-DAG: define void @[[NAME12]]( +//TCHECK-DAG: define weak void @[[NAME1]]( +//TCHECK-DAG: define weak void @[[NAME2]]( +//TCHECK-DAG: define weak void @[[NAME3]]( +//TCHECK-DAG: define weak void @[[NAME4]]( +//TCHECK-DAG: define weak void @[[NAME5]]( +//TCHECK-DAG: define weak void @[[NAME6]]( +//TCHECK-DAG: define weak void @[[NAME7]]( +//TCHECK-DAG: define weak void @[[NAME8]]( +//TCHECK-DAG: define weak void @[[NAME9]]( +//TCHECK-DAG: define weak void @[[NAME10]]( +//TCHECK-DAG: define weak void @[[NAME11]]( +//TCHECK-DAG: define weak void @[[NAME12]]( // CHECK-NTARGET-NOT: __tgt_target // CHECK-NTARGET-NOT: __tgt_register_lib @@ -381,7 +402,7 @@ struct ST { //CHECK: ret void //CHECK: declare i32 @__tgt_unregister_lib([[DSCTY]]*) -//CHECK: define linkonce hidden void @[[REGFN]](i8*) +//CHECK: define linkonce hidden void @[[REGFN]]() //CHECK-SAME: comdat { //CHECK: call i32 @__tgt_register_lib([[DSCTY]]* [[DESC]]) //CHECK: call i32 @__cxa_atexit(void (i8*)* @[[UNREGFN]], i8* bitcast ([[DSCTY]]* [[DESC]] to i8*), @@ -421,31 +442,31 @@ int bar(int a){ // Check metadata is properly generated: // CHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 247, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 265, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 272, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 291, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 415, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 298, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 291, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 298, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 221, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 268, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 286, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 293, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 312, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 436, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 319, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 312, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 319, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 242, i32 {{[0-9]+}}} // TCHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 247, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 265, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 272, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 291, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 415, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 298, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 291, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 298, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 221, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 268, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 286, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 293, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 312, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 436, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 319, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 312, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 319, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 242, i32 {{[0-9]+}}} #endif diff --git a/test/OpenMP/target_simd_codegen_registration_naming.cpp b/test/OpenMP/target_simd_codegen_registration_naming.cpp index 2ed74da4f05bf..2de806ddeaf8b 100644 --- a/test/OpenMP/target_simd_codegen_registration_naming.cpp +++ b/test/OpenMP/target_simd_codegen_registration_naming.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target simd codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,6 +24,16 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/target_simd_collapse_messages.cpp b/test/OpenMP/target_simd_collapse_messages.cpp index ce42273cd259b..400844152b440 100644 --- a/test/OpenMP/target_simd_collapse_messages.cpp +++ b/test/OpenMP/target_simd_collapse_messages.cpp @@ -1,6 +1,9 @@ // RUN: %clang_cc1 -verify -fopenmp %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_simd_defaultmap_messages.cpp b/test/OpenMP/target_simd_defaultmap_messages.cpp index 4f622136b13e3..33e12c1e0e84b 100644 --- a/test/OpenMP/target_simd_defaultmap_messages.cpp +++ b/test/OpenMP/target_simd_defaultmap_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_simd_depend_codegen.cpp b/test/OpenMP/target_simd_depend_codegen.cpp new file mode 100644 index 0000000000000..ed34dd12dfb66 --- /dev/null +++ b/test/OpenMP/target_simd_depend_codegen.cpp @@ -0,0 +1,261 @@ +// Test host codegen. +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// CHECK-DAG: [[TT:%.+]] = type { i64, i8 } +// CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } +// CHECK-DAG: [[DEVTY:%.+]] = type { i8*, i8*, [[ENTTY]]*, [[ENTTY]]* } +// CHECK-DAG: [[DSCTY:%.+]] = type { i32, [[DEVTY]]*, [[ENTTY]]*, [[ENTTY]]* } + +// TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 } + +// CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat + +// CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 0, i[[SZ]] 4] +// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 800] +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] + +// Check if offloading descriptor is created. +// CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] +// CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] +// CHECK: [[DEVBEGIN:@.+]] = external constant i8 +// CHECK: [[DEVEND:@.+]] = external constant i8 +// CHECK: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x [[DEVTY]]] [{{.+}} { i8* [[DEVBEGIN]], i8* [[DEVEND]], [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }], comdat($[[REGFN]]) +// CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) + +// Check target registration is registered as a Ctor. +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] + + +template<typename tx, typename ty> +struct TT{ + tx X; + ty Y; +}; + +int global; +extern int global; + +// CHECK: define {{.*}}[[FOO:@.+]]( +int foo(int n) { + int a = 0; + short aa = 0; + float b[10]; + float bn[n]; + double c[5][10]; + double cn[5][n]; + TT<long long, char> d; + static long *plocal; + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID:%.+]], i32 1, i[[SZ]] {{20|40}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY0:%.+]]* + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 3 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* @0, i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + #pragma omp target simd device(global + a) depend(in: global) depend(out: a, b, cn[4]) + for (int i = 0; i < 10; ++i) { + } + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + + // CHECK: [[BOOL:%.+]] = icmp ne i32 %{{.+}}, 0 + // CHECK: br i1 [[BOOL]], label %[[THEN:.+]], label %[[ELSE:.+]] + // CHECK: [[THEN]]: + // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%.+]], i32 0, i32 0 + // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%.+]], i32 0, i32 0 + // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]** + // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]** + // CHECK-DAG: store i[[SZ]]* [[BP0:%[^,]+]], i[[SZ]]** [[CBPADDR0]] + // CHECK-DAG: store i[[SZ]]* [[BP0]], i[[SZ]]** [[CPADDR0]] + + // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1 + // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1 + // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* + // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* + // CHECK-DAG: store i[[SZ]] [[BP1:%[^,]+]], i[[SZ]]* [[CBPADDR1]] + // CHECK-DAG: store i[[SZ]] [[BP1]], i[[SZ]]* [[CPADDR1]] + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID]], i32 1, i[[SZ]] {{104|52}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1_:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1_:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + + // CHECK: [[ELSE]]: + // CHECK-NOT: getelementptr inbounds [2 x i8*], [2 x i8*]* + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1__:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1__:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + // CHECK: [[EXIT]]: + + #pragma omp target simd device(global + a) nowait depend(inout: global, a, bn) if(a) + for (int i = 0; i < *plocal; ++i) { + static int local1; + *plocal = global; + local1 = global; + } + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]* + // CHECK: getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* @0, i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + #pragma omp target simd if(0) firstprivate(global) depend(out:global) + for (int i = 0; i < global; ++i) { + global += 1; + } + + return a; +} + +// Check that the offloading functions are emitted and that the arguments are +// correct and loaded correctly for the target regions in foo(). + +// CHECK: define internal void [[HVT0:@.+]]() + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, [[TASK_TY0]]* noalias) +// CHECK: store void (i8*, ...)* null, void (i8*, ...)** % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 0, i8** null, i8** null, i[[SZ]]* null, i64* null) +// CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: call void [[HVT0]]() +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT1:@.+]](i[[SZ]]* %{{.+}}, i[[SZ]] %{{.+}}) + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1_]](i32{{.*}}, [[TASK_TY1_]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[SZT:%.+]] = getelementptr inbounds [2 x i[[SZ]]], [2 x i[[SZ]]]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_nowait(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 2, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SZT]], i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT]], i32 0, i32 0) + +// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1__]](i32{{.*}}, [[TASK_TY1__]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT2:@.+]](i[[SZ]] %{{.+}}) +// Create stack storage and store argument in there. +// CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align +// CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align +// CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* +// CHECK-64: load i32, i32* [[AA_CADDR]], align +// CHECK-32: load i32, i32* [[AA_ADDR]], align + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, [[TASK_TY2]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT2]](i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + + +#endif diff --git a/test/OpenMP/target_simd_depend_messages.cpp b/test/OpenMP/target_simd_depend_messages.cpp index 89184f25d7a17..80f988d7a0e70 100644 --- a/test/OpenMP/target_simd_depend_messages.cpp +++ b/test/OpenMP/target_simd_depend_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_simd_device_messages.cpp b/test/OpenMP/target_simd_device_messages.cpp index 1e9afc3bb1791..d705f6140285d 100644 --- a/test/OpenMP/target_simd_device_messages.cpp +++ b/test/OpenMP/target_simd_device_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_simd_firstprivate_messages.cpp b/test/OpenMP/target_simd_firstprivate_messages.cpp index 5bfd951ff81ea..85b216b4cbf63 100644 --- a/test/OpenMP/target_simd_firstprivate_messages.cpp +++ b/test/OpenMP/target_simd_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_simd_if_messages.cpp b/test/OpenMP/target_simd_if_messages.cpp index ff68f6f21602e..60dee37ac543a 100644 --- a/test/OpenMP/target_simd_if_messages.cpp +++ b/test/OpenMP/target_simd_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 %s + void foo() { } diff --git a/test/OpenMP/target_simd_is_device_ptr_messages.cpp b/test/OpenMP/target_simd_is_device_ptr_messages.cpp index 02e2b326c7426..6d6bc247c8068 100644 --- a/test/OpenMP/target_simd_is_device_ptr_messages.cpp +++ b/test/OpenMP/target_simd_is_device_ptr_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -std=c++11 -verify -fopenmp %s +// RUN: %clang_cc1 -std=c++11 -verify -fopenmp-simd %s + struct ST { int *a; }; diff --git a/test/OpenMP/target_simd_lastprivate_messages.cpp b/test/OpenMP/target_simd_lastprivate_messages.cpp index 9a6eb241728bd..ae8bd0af7b5f8 100644 --- a/test/OpenMP/target_simd_lastprivate_messages.cpp +++ b/test/OpenMP/target_simd_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_simd_linear_messages.cpp b/test/OpenMP/target_simd_linear_messages.cpp index 6319e31b3fb2f..6a094b8651858 100644 --- a/test/OpenMP/target_simd_linear_messages.cpp +++ b/test/OpenMP/target_simd_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/target_simd_loop_messages.cpp b/test/OpenMP/target_simd_loop_messages.cpp index d33f1f4cdc653..760bfe0496c2f 100644 --- a/test/OpenMP/target_simd_loop_messages.cpp +++ b/test/OpenMP/target_simd_loop_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s -Wno-openmp-target + +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s -Wno-openmp-target class S { int a; diff --git a/test/OpenMP/target_simd_map_messages.cpp b/test/OpenMP/target_simd_map_messages.cpp index 2473bbb49bead..3722ecb95ad1f 100644 --- a/test/OpenMP/target_simd_map_messages.cpp +++ b/test/OpenMP/target_simd_map_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/target_simd_messages.cpp b/test/OpenMP/target_simd_messages.cpp index d1b06e3b9030b..aa259c0696631 100644 --- a/test/OpenMP/target_simd_messages.cpp +++ b/test/OpenMP/target_simd_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_simd_misc_messages.c b/test/OpenMP/target_simd_misc_messages.c index debe38713826d..3965d28af577f 100644 --- a/test/OpenMP/target_simd_misc_messages.c +++ b/test/OpenMP/target_simd_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp target simd'}} #pragma omp target simd diff --git a/test/OpenMP/target_simd_nowait_messages.cpp b/test/OpenMP/target_simd_nowait_messages.cpp index 3635ce25956b7..c7ab302cb0f90 100644 --- a/test/OpenMP/target_simd_nowait_messages.cpp +++ b/test/OpenMP/target_simd_nowait_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_simd_private_messages.cpp b/test/OpenMP/target_simd_private_messages.cpp index b20ff0a426a72..1b7d864c7c1f6 100644 --- a/test/OpenMP/target_simd_private_messages.cpp +++ b/test/OpenMP/target_simd_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } @@ -43,7 +45,7 @@ public: S5(int v) : a(v) {} S5 &operator=(S5 &s) { #pragma omp target simd private(a) private(this->a) private(s.a) // expected-error {{expected variable name or data member of current class}} - for (int k = 0; k < s.a; ++k) + for (int k = 0; k < s.a; ++k) // expected-warning {{Non-trivial type 'S5' is mapped, only trivial types are guaranteed to be mapped correctly}} ++s.a; return *this; } diff --git a/test/OpenMP/target_simd_reduction_messages.cpp b/test/OpenMP/target_simd_reduction_messages.cpp index 9e1bb0521d716..299c81315abef 100644 --- a/test/OpenMP/target_simd_reduction_messages.cpp +++ b/test/OpenMP/target_simd_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_simd_safelen_messages.cpp b/test/OpenMP/target_simd_safelen_messages.cpp index 79bee763498e8..69ceab1a68361 100644 --- a/test/OpenMP/target_simd_safelen_messages.cpp +++ b/test/OpenMP/target_simd_safelen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_simd_simdlen_messages.cpp b/test/OpenMP/target_simd_simdlen_messages.cpp index 8a9a3dc9f9b3f..c65fe20d3a243 100644 --- a/test/OpenMP/target_simd_simdlen_messages.cpp +++ b/test/OpenMP/target_simd_simdlen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_ast_print.cpp b/test/OpenMP/target_teams_ast_print.cpp index a80b8d7dac831..a0cb63535d683 100644 --- a/test/OpenMP/target_teams_ast_print.cpp +++ b/test/OpenMP/target_teams_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -46,7 +50,7 @@ T tmain(T argc, T *argv) { // CHECK-NEXT: T b = argc, c, d, e, f, g; // CHECK-NEXT: static T a; // CHECK-NEXT: S<T> s; -// CHECK-NEXT: #pragma omp target teams +// CHECK-NEXT: #pragma omp target teams{{$}} // CHECK-NEXT: a = 2; // CHECK-NEXT: #pragma omp target teams default(none) private(argc,b) firstprivate(argv) shared(d) reduction(+: c) reduction(max: e) num_teams(C) thread_limit(d * C) // CHECK-NEXT: foo() diff --git a/test/OpenMP/target_teams_codegen.cpp b/test/OpenMP/target_teams_codegen.cpp index e6c250178d064..5c18a0cba3bbb 100644 --- a/test/OpenMP/target_teams_codegen.cpp +++ b/test/OpenMP/target_teams_codegen.cpp @@ -1,28 +1,46 @@ // Test host codegen. -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER -// CHECK-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CHECK-DAG: [[TT:%.+]] = type { i64, i8 } // CHECK-DAG: [[S1:%.+]] = type { double } @@ -34,38 +52,37 @@ // CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat -// We have 8 target regions, but only 7 that actually will generate offloading -// code, only 6 will have mapped arguments, and only 4 have all-constant map -// sizes. +// We have 8 target regions, but only 6 that actually will generate offloading +// code and have mapped arguments, and only 4 have all-constant map sizes. // CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 2, i[[SZ]] 4, i[[SZ]] 4] -// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 288] -// CHECK-DAG: [[SIZET2:@.+]] = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 2] -// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [1 x i64] [i64 288] +// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 800] +// CHECK-DAG: [[SIZET2:@.+]] = private unnamed_addr constant [1 x i{{32|64}}] [i[[SZ]] 2] +// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [1 x i64] [i64 800] // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2] -// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 288, i64 288] -// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [9 x i64] [i64 288, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547] +// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 800] +// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [9 x i64] [i64 800, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547] // CHECK-DAG: [[SIZET5:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 40] -// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 547] +// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 547] // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [4 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 1, i[[SZ]] 40] -// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [4 x i64] [i64 288, i64 288, i64 288, i64 547] -// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [5 x i64] [i64 547, i64 288, i64 288, i64 288, i64 547] -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 - -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK-NOT: @{{.+}} = constant [[ENTTY]] +// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [4 x i64] [i64 800, i64 800, i64 800, i64 547] +// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [6 x i64] [i64 32, i64 281474976711171, i64 800, i64 288, i64 288, i64 547] +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] // Check if offloading descriptor is created. // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] @@ -76,7 +93,7 @@ // CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) // Check target registration is registered as a Ctor. -// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] template<typename tx, typename ty> @@ -202,13 +219,13 @@ int foo(int n) { // CHECK-32: store i32 [[A_VAL]], i32* [[A_CADDR:%.+]], // CHECK-32: [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]], + // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 + // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] + // CHECK: [[TRY]] // CHECK: [[BNSIZE:%.+]] = mul nuw i[[SZ]] [[VLA0:%.+]], 4 // CHECK: [[CNELEMSIZE2:%.+]] = mul nuw i[[SZ]] 5, [[VLA1:%.+]] // CHECK: [[CNSIZE:%.+]] = mul nuw i[[SZ]] [[CNELEMSIZE2]], 8 - // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 - // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] - // CHECK: [[TRY]] // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 9, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([9 x i64], [9 x i64]* [[MAPT4]], i32 0, i32 0), i32 0, i32 0) // CHECK-DAG: [[BPR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 // CHECK-DAG: [[PR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P:%[^,]+]], i32 0, i32 0 @@ -323,11 +340,12 @@ int foo(int n) { // correct and loaded correctly for the target regions in foo(). // CHECK: define internal void [[HVT0]](i[[SZ]] {{[^,]+}}, i[[SZ]] {{[^,]+}}, i[[SZ]] {{[^)]+}}) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] {{[^)]+}}) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] {{[^)]+}}) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] {{[^)]+}}) -// CHECK: alloca i16, +// CHECK: alloca i[[SZ]], +// CHECK: bitcast i[[SZ]]* {{.+}} to i16* // CHECK: ret void // CHECK-NEXT: } @@ -344,7 +362,7 @@ int foo(int n) { // CHECK-64: store i32 [[AA]], i32* [[AA_C]], align // CHECK-32: store i32 [[AA]], i32* [[AA_CASTED]], align // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED1]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) @@ -366,7 +384,7 @@ int foo(int n) { // CHECK: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i16* // CHECK: store i16 [[AA]], i16* [[AA_C]], align // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED2]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) @@ -397,7 +415,7 @@ int foo(int n) { // CHECK-DAG: store i16 [[AA]], i16* [[AA_C]], align // CHECK-DAG: [[PARAM1:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CASTED]], align // CHECK-DAG: [[PARAM2:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK-DAG: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]]) +// CHECK-DAG: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED3]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}) @@ -449,7 +467,7 @@ int foo(int n) { // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 9, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], [10 x float]*, i[[SZ]], float*, [5 x [10 x double]]*, i[[SZ]], i[[SZ]], double*, [[TT]]*)* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], [10 x float]* [[REF_B]], i[[SZ]] [[VAL_VLA1]], float* [[REF_BN]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] [[VAL_VLA2]], i[[SZ]] [[VAL_VLA3]], double* [[REF_CN]], [[TT]]* [[REF_D]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 9, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], [10 x float]*, i[[SZ]], float*, [5 x [10 x double]]*, i[[SZ]], i[[SZ]], double*, [[TT]]*)* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], [10 x float]* [[REF_B]], i[[SZ]] [[VAL_VLA1]], float* [[REF_BN]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] [[VAL_VLA2]], i[[SZ]] [[VAL_VLA3]], double* [[REF_CN]], [[TT]]* [[REF_D]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED4]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, [10 x float]* {{.+}}, i[[SZ]] %{{.+}}, float* {{.+}}, [5 x [10 x double]]* {{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, double* {{.+}}, [[TT]]* {{.+}}) @@ -534,32 +552,36 @@ int bar(int n){ // CHECK-64: store i32 %{{.+}}, i32* [[B_ADDR]], // CHECK-64: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_CADDR]], +// CHECK-32: store i32 %{{.+}}, i32* %__vla_expr // CHECK-32: store i32 %{{.+}}, i32* [[B_ADDR:%.+]], // CHECK-32: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_ADDR]], +// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 +// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] +// CHECK: [[TRY]] // We capture 2 VLA sizes in this target region // CHECK: [[CELEMSIZE2:%.+]] = mul nuw i[[SZ]] 2, [[VLA0:%.+]] // CHECK: [[CSIZE:%.+]] = mul nuw i[[SZ]] [[CELEMSIZE2]], 2 -// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 -// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] -// CHECK: [[TRY]] -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 5, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([5 x i64], [5 x i64]* [[MAPT7]], i32 0, i32 0), i32 0, i32 0) -// CHECK-DAG: [[BPR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP:%.+]], i32 0, i32 0 -// CHECK-DAG: [[PR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SR]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] -// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX0]] -// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX0]] -// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] -// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX1]] -// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX1]] -// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] -// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX2]] -// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX2]] -// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] -// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX3]] -// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 6, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([6 x i64], [6 x i64]* [[MAPT7]], i32 0, i32 0), i32 0, i32 0) +// CHECK-DAG: [[BPR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP:%.+]], i32 0, i32 0 +// CHECK-DAG: [[PR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SR]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] +// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX0]] +// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX0]] +// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] +// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX1]] +// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX1]] +// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] +// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX2]] +// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX2]] +// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] +// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] +// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX4:[0-9]+]] +// CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX4]] +// CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX4]] // The names below are not necessarily consistent with the names used for the // addresses above as some are repeated. @@ -587,10 +609,16 @@ int bar(int n){ // CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** // CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* {{%[^,]+}} -// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR4:%.+]], -// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR4:%.+]], -// CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to i16** -// CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to i16** +// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR4:%.+]], +// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CPADDR4:%.+]], +// CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to [[S1]]** +// CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to [[S1]]** +// CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* {{%[^,]+}} + +// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR5:%.+]], +// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR5:%.+]], +// CHECK-DAG: [[CBPADDR5]] = bitcast i8** {{%[^,]+}} to i16** +// CHECK-DAG: [[CPADDR5]] = bitcast i8** {{%[^,]+}} to i16** // CHECK-DAG: store i[[SZ]] [[CSIZE]], i[[SZ]]* {{%[^,]+}} // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 @@ -726,7 +754,7 @@ int bar(int n){ // CHECK-32-DAG:store i32 [[LOCAL_BV]], i32* [[LOCAL_B_CASTED]], align // CHECK-DAG: [[REF_B:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_B_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]], i[[SZ]], i[[SZ]], i16*)* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), [[S1]]* [[REF_THIS]], i[[SZ]] [[REF_B]], i[[SZ]] [[VAL_VLA1]], i[[SZ]] [[VAL_VLA2]], i16* [[REF_C]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]], i[[SZ]], i[[SZ]], i16*)* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), [[S1]]* [[REF_THIS]], i[[SZ]] [[REF_B]], i[[SZ]] [[VAL_VLA1]], i[[SZ]] [[VAL_VLA2]], i16* [[REF_C]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED5]](i32* noalias %.global_tid., i32* noalias %.bound_tid., [[S1]]* %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i16* {{.+}}) @@ -769,7 +797,7 @@ int bar(int n){ // CHECK-DAG: store i8 [[CONV_AAA]], i8* [[CONV]], align // CHECK-DAG: [[REF_AAA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AAA_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED6:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], i[[SZ]] [[REF_AAA]], [10 x i32]* [[REF_B]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED6:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], i[[SZ]] [[REF_AAA]], [10 x i32]* [[REF_B]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED6]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) @@ -802,7 +830,7 @@ int bar(int n){ // CHECK-DAG: store i16 [[CONV_AA]], i16* [[CONV]], align // CHECK-DAG: [[REF_AA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AA_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED7:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], [10 x i32]* [[REF_B]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED7:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], [10 x i32]* [[REF_B]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED7]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) diff --git a/test/OpenMP/target_teams_codegen_registration.cpp b/test/OpenMP/target_teams_codegen_registration.cpp index 38668a46b161f..e62be6813460e 100644 --- a/test/OpenMP/target_teams_codegen_registration.cpp +++ b/test/OpenMP/target_teams_codegen_registration.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target teams codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,9 +24,22 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK -// Check that no target code is emmitted if no omptests flag was provided. +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// Check that no target code is emitted if no omptests flag was provided. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK-NTARGET +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -60,96 +81,96 @@ // We have 7 target regions -// CHECK-DAG: {{@.+}} = private constant i8 0 -// TCHECK-NOT: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// TCHECK-NOT: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] -// CHECK-NTARGET-NOT: private constant i8 0 +// CHECK-NTARGET-NOT: weak constant i8 0 // CHECK-NTARGET-NOT: private unnamed_addr constant [1 x i // CHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// CHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// CHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// CHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// CHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// CHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// CHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// CHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// CHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// CHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// CHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// CHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// CHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// TCHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// TCHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// TCHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// TCHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// TCHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// TCHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// TCHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// TCHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// TCHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// TCHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// TCHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// TCHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] // CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] @@ -163,7 +184,7 @@ // CHECK-SAME: { i32, void ()*, i8* } { i32 500, void ()* [[P500:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 501, void ()* [[P501:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 65535, void ()* [[PMAX:@[^,]+]], i8* null }, -// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] // CHECK-NTARGET: @llvm.global_ctors = appending global [3 x { i32, void ()*, i8* }] [ @@ -319,18 +340,18 @@ struct ST { //CHECK-DAG: define internal void @[[NAME12]]( //CHECK-DAG: call void @[[NAME12]]( -//TCHECK-DAG: define void @[[NAME1]]( -//TCHECK-DAG: define void @[[NAME2]]( -//TCHECK-DAG: define void @[[NAME3]]( -//TCHECK-DAG: define void @[[NAME4]]( -//TCHECK-DAG: define void @[[NAME5]]( -//TCHECK-DAG: define void @[[NAME6]]( -//TCHECK-DAG: define void @[[NAME7]]( -//TCHECK-DAG: define void @[[NAME8]]( -//TCHECK-DAG: define void @[[NAME9]]( -//TCHECK-DAG: define void @[[NAME10]]( -//TCHECK-DAG: define void @[[NAME11]]( -//TCHECK-DAG: define void @[[NAME12]]( +//TCHECK-DAG: define weak void @[[NAME1]]( +//TCHECK-DAG: define weak void @[[NAME2]]( +//TCHECK-DAG: define weak void @[[NAME3]]( +//TCHECK-DAG: define weak void @[[NAME4]]( +//TCHECK-DAG: define weak void @[[NAME5]]( +//TCHECK-DAG: define weak void @[[NAME6]]( +//TCHECK-DAG: define weak void @[[NAME7]]( +//TCHECK-DAG: define weak void @[[NAME8]]( +//TCHECK-DAG: define weak void @[[NAME9]]( +//TCHECK-DAG: define weak void @[[NAME10]]( +//TCHECK-DAG: define weak void @[[NAME11]]( +//TCHECK-DAG: define weak void @[[NAME12]]( // CHECK-NTARGET-NOT: __tgt_target // CHECK-NTARGET-NOT: __tgt_register_lib @@ -372,7 +393,7 @@ struct ST { //CHECK: ret void //CHECK: declare i32 @__tgt_unregister_lib([[DSCTY]]*) -//CHECK: define linkonce hidden void @[[REGFN]](i8*) +//CHECK: define linkonce hidden void @[[REGFN]]() //CHECK-SAME: comdat { //CHECK: call i32 @__tgt_register_lib([[DSCTY]]* [[DESC]]) //CHECK: call i32 @__cxa_atexit(void (i8*)* @[[UNREGFN]], i8* bitcast ([[DSCTY]]* [[DESC]] to i8*), @@ -411,31 +432,31 @@ int bar(int a){ // Check metadata is properly generated: // CHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 245, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 261, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 267, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 278, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 406, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 290, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 290, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 278, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 220, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 266, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 282, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 288, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 299, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 427, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 311, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 311, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 299, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 241, i32 {{[0-9]+}}} // TCHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 245, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 261, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 267, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 278, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 406, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 290, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 290, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 278, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 220, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 266, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 282, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 288, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 299, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 427, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 311, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 311, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 299, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 241, i32 {{[0-9]+}}} #endif diff --git a/test/OpenMP/target_teams_codegen_registration_naming.cpp b/test/OpenMP/target_teams_codegen_registration_naming.cpp index fb4232897282c..c09442d61d979 100644 --- a/test/OpenMP/target_teams_codegen_registration_naming.cpp +++ b/test/OpenMP/target_teams_codegen_registration_naming.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target teams codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,6 +24,16 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/target_teams_default_messages.cpp b/test/OpenMP/target_teams_default_messages.cpp index 053463749b5a2..fa54416e6774e 100644 --- a/test/OpenMP/target_teams_default_messages.cpp +++ b/test/OpenMP/target_teams_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/target_teams_defaultmap_messages.cpp b/test/OpenMP/target_teams_defaultmap_messages.cpp index f2d30fdc87e25..b1b2bdd389620 100644 --- a/test/OpenMP/target_teams_defaultmap_messages.cpp +++ b/test/OpenMP/target_teams_defaultmap_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_depend_codegen.cpp b/test/OpenMP/target_teams_depend_codegen.cpp new file mode 100644 index 0000000000000..a5fbc5afc163f --- /dev/null +++ b/test/OpenMP/target_teams_depend_codegen.cpp @@ -0,0 +1,261 @@ +// Test host codegen. +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// CHECK-DAG: [[TT:%.+]] = type { i64, i8 } +// CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } +// CHECK-DAG: [[DEVTY:%.+]] = type { i8*, i8*, [[ENTTY]]*, [[ENTTY]]* } +// CHECK-DAG: [[DSCTY:%.+]] = type { i32, [[DEVTY]]*, [[ENTTY]]*, [[ENTTY]]* } + +// TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 } + +// CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat + +// CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 0, i[[SZ]] 4] +// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 800] +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] + +// Check if offloading descriptor is created. +// CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] +// CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] +// CHECK: [[DEVBEGIN:@.+]] = external constant i8 +// CHECK: [[DEVEND:@.+]] = external constant i8 +// CHECK: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x [[DEVTY]]] [{{.+}} { i8* [[DEVBEGIN]], i8* [[DEVEND]], [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }], comdat($[[REGFN]]) +// CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) + +// Check target registration is registered as a Ctor. +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] + + +template<typename tx, typename ty> +struct TT{ + tx X; + ty Y; +}; + +int global; +extern int global; + +// CHECK: define {{.*}}[[FOO:@.+]]( +int foo(int n) { + int a = 0; + short aa = 0; + float b[10]; + float bn[n]; + double c[5][10]; + double cn[5][n]; + TT<long long, char> d; + static long *plocal; + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID:%.+]], i32 1, i[[SZ]] {{20|40}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY0:%.+]]* + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 3 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* @0, i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + #pragma omp target teams device(global + a) depend(in: global) depend(out: a, b, cn[4]) + { + } + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + + // CHECK: [[BOOL:%.+]] = icmp ne i32 %{{.+}}, 0 + // CHECK: br i1 [[BOOL]], label %[[THEN:.+]], label %[[ELSE:.+]] + // CHECK: [[THEN]]: + // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%.+]], i32 0, i32 0 + // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%.+]], i32 0, i32 0 + // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]** + // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]** + // CHECK-DAG: store i[[SZ]]* [[BP0:%[^,]+]], i[[SZ]]** [[CBPADDR0]] + // CHECK-DAG: store i[[SZ]]* [[BP0]], i[[SZ]]** [[CPADDR0]] + + // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1 + // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1 + // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* + // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* + // CHECK-DAG: store i[[SZ]] [[BP1:%[^,]+]], i[[SZ]]* [[CBPADDR1]] + // CHECK-DAG: store i[[SZ]] [[BP1]], i[[SZ]]* [[CPADDR1]] + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID]], i32 1, i[[SZ]] {{104|52}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1_:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1_:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + + // CHECK: [[ELSE]]: + // CHECK-NOT: getelementptr inbounds [2 x i8*], [2 x i8*]* + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1__:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1__:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + // CHECK: [[EXIT]]: + + #pragma omp target teams device(global + a) nowait depend(inout: global, a, bn) if(a) + { + static int local1; + *plocal = global; + local1 = global; + } + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]* + // CHECK: getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* @0, i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK]]) + #pragma omp target teams if(0) firstprivate(global) depend(out:global) + { + global += 1; + } + + return a; +} + +// Check that the offloading functions are emitted and that the arguments are +// correct and loaded correctly for the target regions in foo(). + +// CHECK: define internal void [[HVT0:@.+]]() + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, [[TASK_TY0]]* noalias) +// CHECK: store void (i8*, ...)* null, void (i8*, ...)** % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 0, i8** null, i8** null, i[[SZ]]* null, i64* null, i32 0, i32 0) +// CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: call void [[HVT0]]() +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT1:@.+]](i[[SZ]]* %{{.+}}, i[[SZ]] %{{.+}}) + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1_]](i32{{.*}}, [[TASK_TY1_]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[SZT:%.+]] = getelementptr inbounds [2 x i[[SZ]]], [2 x i[[SZ]]]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams_nowait(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 2, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SZT]], i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT]], i32 0, i32 0), i32 0, i32 0) + +// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1__]](i32{{.*}}, [[TASK_TY1__]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT2:@.+]](i[[SZ]] %{{.+}}) +// Create stack storage and store argument in there. +// CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align +// CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align +// CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* +// CHECK-64: load i32, i32* [[AA_CADDR]], align +// CHECK-32: load i32, i32* [[AA_ADDR]], align + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, [[TASK_TY2]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT2]](i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + + +#endif diff --git a/test/OpenMP/target_teams_depend_messages.cpp b/test/OpenMP/target_teams_depend_messages.cpp index 57983eec8882c..235ce98ee9450 100644 --- a/test/OpenMP/target_teams_depend_messages.cpp +++ b/test/OpenMP/target_teams_depend_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_device_messages.cpp b/test/OpenMP/target_teams_device_messages.cpp index bd75da8351585..da56c470fb6ef 100644 --- a/test/OpenMP/target_teams_device_messages.cpp +++ b/test/OpenMP/target_teams_device_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_ast_print.cpp b/test/OpenMP/target_teams_distribute_ast_print.cpp index ae50366aabb3c..4484451c49076 100644 --- a/test/OpenMP/target_teams_distribute_ast_print.cpp +++ b/test/OpenMP/target_teams_distribute_ast_print.cpp @@ -1,6 +1,10 @@ -// RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp -ast-print %s -Wno-openmp-target | FileCheck %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print -Wno-openmp-target | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s -Wno-openmp-target | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print -Wno-openmp-target | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -81,7 +85,7 @@ T tmain(T argc) { #pragma omp target teams distribute for (int i=0; i < 2; ++i) a = 2; -// CHECK: #pragma omp target teams distribute +// CHECK: #pragma omp target teams distribute{{$}} // CHECK-NEXT: for (int i = 0; i < 2; ++i) // CHECK-NEXT: a = 2; #pragma omp target teams distribute private(argc, b), firstprivate(c, d), collapse(2) diff --git a/test/OpenMP/target_teams_distribute_codegen.cpp b/test/OpenMP/target_teams_distribute_codegen.cpp index 17baa603d12be..f9500a73ef795 100644 --- a/test/OpenMP/target_teams_distribute_codegen.cpp +++ b/test/OpenMP/target_teams_distribute_codegen.cpp @@ -1,28 +1,46 @@ // Test host codegen. -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER -// CHECK-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CHECK-DAG: [[TT:%.+]] = type { i64, i8 } // CHECK-DAG: [[S1:%.+]] = type { double } @@ -39,33 +57,33 @@ // sizes. // CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 2, i[[SZ]] 4, i[[SZ]] 4] -// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 288] +// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 800] // CHECK-DAG: [[SIZET2:@.+]] = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 2] -// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [1 x i64] [i64 288] +// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [1 x i64] [i64 800] // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2] -// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 288, i64 288] -// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [9 x i64] [i64 288, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547] -// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [5 x i64] [i64 547, i64 288, i64 288, i64 288, i64 547] +// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 800] +// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [10 x i64] [i64 800, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547, i64 800] +// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [6 x i64] [i64 32, i64 281474976711171, i64 800, i64 288, i64 288, i64 547] // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [5 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 1, i[[SZ]] 40] -// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [5 x i64] [i64 288, i64 288, i64 288, i64 288, i64 547] +// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [5 x i64] [i64 800, i64 800, i64 800, i64 800, i64 547] // CHECK-DAG: [[SIZET7:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 40] -// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 547] -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 - -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK-NOT: @{{.+}} = constant [[ENTTY]] +// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 547] +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] // Check if offloading descriptor is created. // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] @@ -76,7 +94,7 @@ // CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) // Check target registration is registered as a Ctor. -// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] template<typename tx, typename ty> @@ -193,6 +211,7 @@ int foo(int n) { } // We capture 3 VLA sizes in this target region + // CHECK: load i32, i32* % // CHECK-64: [[A_VAL:%.+]] = load i32, i32* %{{.+}}, // CHECK-64: [[A_ADDR:%.+]] = bitcast i[[SZ]]* [[A_CADDR:%.+]] to i32* // CHECK-64: store i32 [[A_VAL]], i32* [[A_ADDR]], @@ -202,45 +221,48 @@ int foo(int n) { // CHECK-32: store i32 [[A_VAL]], i32* [[A_CADDR:%.+]], // CHECK-32: [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]], + // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 + // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] + // CHECK: [[TRY]] // CHECK: [[BNSIZE:%.+]] = mul nuw i[[SZ]] [[VLA0:%.+]], 4 // CHECK: [[CNELEMSIZE2:%.+]] = mul nuw i[[SZ]] 5, [[VLA1:%.+]] // CHECK: [[CNSIZE:%.+]] = mul nuw i[[SZ]] [[CNELEMSIZE2]], 8 - // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 - // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] - // CHECK: [[TRY]] - // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 9, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([9 x i64], [9 x i64]* [[MAPT4]], i32 0, i32 0), i32 0, i32 0) - // CHECK-DAG: [[BPR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 - // CHECK-DAG: [[PR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P:%[^,]+]], i32 0, i32 0 - // CHECK-DAG: [[SR]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S:%[^,]+]], i32 0, i32 0 - - // CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX0:[0-9]+]] - // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX0]] - // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX0]] - // CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX1:[0-9]+]] - // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX1]] - // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX1]] - // CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX2:[0-9]+]] - // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX2]] - // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX2]] - // CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX3:[0-9]+]] - // CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX3]] - // CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX3]] - // CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX4:[0-9]+]] - // CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX4]] - // CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX4]] - // CHECK-DAG: [[SADDR5:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX5:[0-9]+]] - // CHECK-DAG: [[BPADDR5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX5]] - // CHECK-DAG: [[PADDR5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX5]] - // CHECK-DAG: [[SADDR6:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX6:[0-9]+]] - // CHECK-DAG: [[BPADDR6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX6]] - // CHECK-DAG: [[PADDR6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX6]] - // CHECK-DAG: [[SADDR7:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX7:[0-9]+]] - // CHECK-DAG: [[BPADDR7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX7]] - // CHECK-DAG: [[PADDR7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX7]] - // CHECK-DAG: [[SADDR8:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX8:[0-9]+]] - // CHECK-DAG: [[BPADDR8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP]], i32 0, i32 [[IDX8]] - // CHECK-DAG: [[PADDR8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P]], i32 0, i32 [[IDX8]] + // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 10, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([10 x i64], [10 x i64]* [[MAPT4]], i32 0, i32 0), i32 0, i32 0) + // CHECK-DAG: [[BPR]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 + // CHECK-DAG: [[PR]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P:%[^,]+]], i32 0, i32 0 + // CHECK-DAG: [[SR]] = getelementptr inbounds [10 x i[[SZ]]], [10 x i[[SZ]]]* [[S:%[^,]+]], i32 0, i32 0 + + // CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [10 x i[[SZ]]], [10 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX0:[0-9]+]] + // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX0]] + // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX0]] + // CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [10 x i[[SZ]]], [10 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX1:[0-9]+]] + // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX1]] + // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX1]] + // CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [10 x i[[SZ]]], [10 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX2:[0-9]+]] + // CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX2]] + // CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX2]] + // CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [10 x i[[SZ]]], [10 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX3:[0-9]+]] + // CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX3]] + // CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX3]] + // CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [10 x i[[SZ]]], [10 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX4:[0-9]+]] + // CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX4]] + // CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX4]] + // CHECK-DAG: [[SADDR5:%.+]] = getelementptr inbounds [10 x i[[SZ]]], [10 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX5:[0-9]+]] + // CHECK-DAG: [[BPADDR5:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX5]] + // CHECK-DAG: [[PADDR5:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX5]] + // CHECK-DAG: [[SADDR6:%.+]] = getelementptr inbounds [10 x i[[SZ]]], [10 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX6:[0-9]+]] + // CHECK-DAG: [[BPADDR6:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX6]] + // CHECK-DAG: [[PADDR6:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX6]] + // CHECK-DAG: [[SADDR7:%.+]] = getelementptr inbounds [10 x i[[SZ]]], [10 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX7:[0-9]+]] + // CHECK-DAG: [[BPADDR7:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX7]] + // CHECK-DAG: [[PADDR7:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX7]] + // CHECK-DAG: [[SADDR8:%.+]] = getelementptr inbounds [10 x i[[SZ]]], [10 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX8:[0-9]+]] + // CHECK-DAG: [[BPADDR8:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX8]] + // CHECK-DAG: [[PADDR8:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX8]] + // CHECK-DAG: [[SADDR9:%.+]] = getelementptr inbounds [10 x i[[SZ]]], [10 x i[[SZ]]]* [[S]], i32 0, i32 [[IDX9:[0-9]+]] + // CHECK-DAG: [[BPADDR9:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[BP]], i32 0, i32 [[IDX9]] + // CHECK-DAG: [[PADDR9:%.+]] = getelementptr inbounds [10 x i8*], [10 x i8*]* [[P]], i32 0, i32 [[IDX9]] // The names below are not necessarily consistent with the names used for the // addresses above as some are repeated. @@ -298,14 +320,20 @@ int foo(int n) { // CHECK-DAG: [[CPADDR8]] = bitcast i8** {{%[^,]+}} to [[TT]]** // CHECK-DAG: store i[[SZ]] {{12|16}}, i[[SZ]]* {{%[^,]+}} + // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CBPADDR9:%.+]], + // CHECK-DAG: store i[[SZ]] %{{.+}}, i[[SZ]]* [[CPADDR9:%.+]], + // CHECK-DAG: [[CBPADDR9]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* + // CHECK-DAG: [[CPADDR9]] = bitcast i8** {{%[^,]+}} to i[[SZ]]* + // CHECK-DAG: store i[[SZ]] {{4|8}}, i[[SZ]]* {{%[^,]+}} + // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 // CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] // CHECK: [[FAIL]] - // CHECK: call void [[HVT4:@.+]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}) + // CHECK: call void [[HVT4:@.+]]({{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}, {{[^,]+}}) // CHECK-NEXT: br label %[[END]] // CHECK: [[END]] - #pragma omp target teams distribute if(target: n>20) + #pragma omp target teams distribute if(target: n>20) dist_schedule(static, n) for (int i = 0; i < 10; ++i) { a += 1; b[2] += 1.0; @@ -323,11 +351,12 @@ int foo(int n) { // correct and loaded correctly for the target regions in foo(). // CHECK: define internal void [[HVT0]](i[[SZ]] {{[^,]+}}, i[[SZ]] {{[^,]+}}, i[[SZ]] {{[^)]+}}) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] {{[^)]+}}) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] {{[^)]+}}) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] {{[^)]+}}) -// CHECK: alloca i16, +// CHECK: alloca i[[SZ]], +// CHECK: bitcast i[[SZ]]* {{.+}} to i16* // CHECK: ret void // CHECK-NEXT: } @@ -344,7 +373,7 @@ int foo(int n) { // CHECK-64: store i32 [[AA]], i32* [[AA_C]], align // CHECK-32: store i32 [[AA]], i32* [[AA_CASTED]], align // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED1]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) @@ -366,7 +395,7 @@ int foo(int n) { // CHECK: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i16* // CHECK: store i16 [[AA]], i16* [[AA_C]], align // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED2]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) @@ -397,7 +426,7 @@ int foo(int n) { // CHECK-DAG: store i16 [[AA]], i16* [[AA_C]], align // CHECK-DAG: [[PARAM1:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CASTED]], align // CHECK-DAG: [[PARAM2:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK-DAG: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]]) +// CHECK-DAG: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED3]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}) @@ -421,6 +450,7 @@ int foo(int n) { // CHECK: [[LOCAL_VLA3:%.+]] = alloca i[[SZ]] // CHECK: [[LOCAL_CN:%.+]] = alloca double* // CHECK: [[LOCAL_D:%.+]] = alloca [[TT]]* +// CHECK: alloca i[[SZ]], // CHECK: [[LOCAL_A_CASTED:%.+]] = alloca i[[SZ]] // CHECK-DAG: store i[[SZ]] [[ARG_A:%.+]], i[[SZ]]* [[LOCAL_A]] // CHECK-DAG: store [10 x float]* [[ARG_B:%.+]], [10 x float]** [[LOCAL_B]] @@ -449,7 +479,7 @@ int foo(int n) { // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 9, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], [10 x float]*, i[[SZ]], float*, [5 x [10 x double]]*, i[[SZ]], i[[SZ]], double*, [[TT]]*)* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], [10 x float]* [[REF_B]], i[[SZ]] [[VAL_VLA1]], float* [[REF_BN]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] [[VAL_VLA2]], i[[SZ]] [[VAL_VLA3]], double* [[REF_CN]], [[TT]]* [[REF_D]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 10, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], [10 x float]*, i[[SZ]], float*, [5 x [10 x double]]*, i[[SZ]], i[[SZ]], double*, [[TT]]*, i[[SZ]])* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], [10 x float]* [[REF_B]], i[[SZ]] [[VAL_VLA1]], float* [[REF_BN]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] [[VAL_VLA2]], i[[SZ]] [[VAL_VLA3]], double* [[REF_CN]], [[TT]]* [[REF_D]], i[[SZ]] %{{.+}}) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED4]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, [10 x float]* {{.+}}, i[[SZ]] %{{.+}}, float* {{.+}}, [5 x [10 x double]]* {{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, double* {{.+}}, [[TT]]* {{.+}}) @@ -534,32 +564,39 @@ int bar(int n){ // CHECK-64: store i32 %{{.+}}, i32* [[B_ADDR]], // CHECK-64: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_CADDR]], +// CHECK-32: store i32 %{{.+}}, i32* %__vla_expr // CHECK-32: store i32 %{{.+}}, i32* [[B_ADDR:%.+]], // CHECK-32: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_ADDR]], +// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 +// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] +// CHECK: [[TRY]] // We capture 2 VLA sizes in this target region // CHECK: [[CELEMSIZE2:%.+]] = mul nuw i[[SZ]] 2, [[VLA0:%.+]] // CHECK: [[CSIZE:%.+]] = mul nuw i[[SZ]] [[CELEMSIZE2]], 2 -// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 -// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] -// CHECK: [[TRY]] -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 5, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([5 x i64], [5 x i64]* [[MAPT5]], i32 0, i32 0), i32 0, i32 0) -// CHECK-DAG: [[BPR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP:%.+]], i32 0, i32 0 -// CHECK-DAG: [[PR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SR]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] -// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX0]] -// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX0]] -// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] -// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX1]] -// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX1]] -// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] -// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX2]] -// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX2]] -// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] -// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX3]] -// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 6, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([6 x i64], [6 x i64]* [[MAPT5]], i32 0, i32 0), i32 0, i32 0) +// CHECK-DAG: [[BPR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP:%.+]], i32 0, i32 0 +// CHECK-DAG: [[PR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SR]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] +// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX0]] +// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX0]] +// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] +// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX1]] +// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX1]] +// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] +// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX2]] +// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX2]] +// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] +// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] +// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] +// CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] +// CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[SADDR5:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] +// CHECK-DAG: [[BPADDR5:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] +// CHECK-DAG: [[PADDR5:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] // The names below are not necessarily consistent with the names used for the // addresses above as some are repeated. @@ -582,15 +619,21 @@ int bar(int n){ // CHECK-DAG: store i[[SZ]] 4, i[[SZ]]* {{%[^,]+}} // CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR3:%.+]], -// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CPADDR3:%.+]], +// CHECK-DAG: store double* %{{.+}}, double** [[CPADDR3:%.+]], // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** -// CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** +// CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to double** +// CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* {{%[^,]+}} + +// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR4:%.+]], +// CHECK-DAG: store double* %{{.+}}, double** [[CPADDR4:%.+]], +// CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to [[S1]]** +// CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to double** // CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* {{%[^,]+}} -// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR4:%.+]], -// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR4:%.+]], -// CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to i16** -// CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to i16** +// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR5:%.+]], +// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR5:%.+]], +// CHECK-DAG: [[CBPADDR5]] = bitcast i8** {{%[^,]+}} to i16** +// CHECK-DAG: [[CPADDR5]] = bitcast i8** {{%[^,]+}} to i16** // CHECK-DAG: store i[[SZ]] [[CSIZE]], i[[SZ]]* {{%[^,]+}} // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 @@ -733,7 +776,7 @@ int bar(int n){ // CHECK-32-DAG:store i32 [[LOCAL_BV]], i32* [[LOCAL_B_CASTED]], align // CHECK-DAG: [[REF_B:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_B_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]], i[[SZ]], i[[SZ]], i16*)* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), [[S1]]* [[REF_THIS]], i[[SZ]] [[REF_B]], i[[SZ]] [[VAL_VLA1]], i[[SZ]] [[VAL_VLA2]], i16* [[REF_C]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]], i[[SZ]], i[[SZ]], i16*)* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), [[S1]]* [[REF_THIS]], i[[SZ]] [[REF_B]], i[[SZ]] [[VAL_VLA1]], i[[SZ]] [[VAL_VLA2]], i16* [[REF_C]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED5]](i32* noalias %.global_tid., i32* noalias %.bound_tid., [[S1]]* %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i16* {{.+}}) @@ -778,7 +821,7 @@ int bar(int n){ // CHECK-DAG: store i8 [[CONV_AAA]], i8* [[CONV]], align // CHECK-DAG: [[REF_AAA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AAA_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED6:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] {{.+}}, i[[SZ]] [[REF_AA]], i[[SZ]] [[REF_AAA]], [10 x i32]* [[REF_B]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED6:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] {{.+}}, i[[SZ]] [[REF_AA]], i[[SZ]] [[REF_AAA]], [10 x i32]* [[REF_B]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED6]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) @@ -811,7 +854,7 @@ int bar(int n){ // CHECK-DAG: store i16 [[CONV_AA]], i16* [[CONV]], align // CHECK-DAG: [[REF_AA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AA_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED7:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], [10 x i32]* [[REF_B]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED7:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], [10 x i32]* [[REF_B]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED7]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) diff --git a/test/OpenMP/target_teams_distribute_codegen_registration.cpp b/test/OpenMP/target_teams_distribute_codegen_registration.cpp index c031731c80654..e45e040b1fcda 100644 --- a/test/OpenMP/target_teams_distribute_codegen_registration.cpp +++ b/test/OpenMP/target_teams_distribute_codegen_registration.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target teams distribute codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,9 +24,22 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK -// Check that no target code is emmitted if no omptests flag was provided. +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// Check that no target code is emitted if no omptests flag was provided. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK-NTARGET +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -60,96 +81,96 @@ // We have 7 target regions -// CHECK-DAG: {{@.+}} = private constant i8 0 -// TCHECK-NOT: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// TCHECK-NOT: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] -// CHECK-NTARGET-NOT: private constant i8 0 +// CHECK-NTARGET-NOT: weak constant i8 0 // CHECK-NTARGET-NOT: private unnamed_addr constant [1 x i // CHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// CHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// CHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// CHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// CHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// CHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// CHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// CHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// CHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// CHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// CHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// CHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// CHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// TCHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// TCHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// TCHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// TCHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// TCHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// TCHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// TCHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// TCHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// TCHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// TCHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// TCHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// TCHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] // CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] @@ -163,7 +184,7 @@ // CHECK-SAME: { i32, void ()*, i8* } { i32 500, void ()* [[P500:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 501, void ()* [[P501:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 65535, void ()* [[PMAX:@[^,]+]], i8* null }, -// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] // CHECK-NTARGET: @llvm.global_ctors = appending global [3 x { i32, void ()*, i8* }] [ @@ -328,18 +349,18 @@ struct ST { //CHECK-DAG: define internal void @[[NAME12]]( //CHECK-DAG: call void @[[NAME12]]( -//TCHECK-DAG: define void @[[NAME1]]( -//TCHECK-DAG: define void @[[NAME2]]( -//TCHECK-DAG: define void @[[NAME3]]( -//TCHECK-DAG: define void @[[NAME4]]( -//TCHECK-DAG: define void @[[NAME5]]( -//TCHECK-DAG: define void @[[NAME6]]( -//TCHECK-DAG: define void @[[NAME7]]( -//TCHECK-DAG: define void @[[NAME8]]( -//TCHECK-DAG: define void @[[NAME9]]( -//TCHECK-DAG: define void @[[NAME10]]( -//TCHECK-DAG: define void @[[NAME11]]( -//TCHECK-DAG: define void @[[NAME12]]( +//TCHECK-DAG: define weak void @[[NAME1]]( +//TCHECK-DAG: define weak void @[[NAME2]]( +//TCHECK-DAG: define weak void @[[NAME3]]( +//TCHECK-DAG: define weak void @[[NAME4]]( +//TCHECK-DAG: define weak void @[[NAME5]]( +//TCHECK-DAG: define weak void @[[NAME6]]( +//TCHECK-DAG: define weak void @[[NAME7]]( +//TCHECK-DAG: define weak void @[[NAME8]]( +//TCHECK-DAG: define weak void @[[NAME9]]( +//TCHECK-DAG: define weak void @[[NAME10]]( +//TCHECK-DAG: define weak void @[[NAME11]]( +//TCHECK-DAG: define weak void @[[NAME12]]( // CHECK-NTARGET-NOT: __tgt_target // CHECK-NTARGET-NOT: __tgt_register_lib @@ -381,7 +402,7 @@ struct ST { //CHECK: ret void //CHECK: declare i32 @__tgt_unregister_lib([[DSCTY]]*) -//CHECK: define linkonce hidden void @[[REGFN]](i8*) +//CHECK: define linkonce hidden void @[[REGFN]]() //CHECK-SAME: comdat { //CHECK: call i32 @__tgt_register_lib([[DSCTY]]* [[DESC]]) //CHECK: call i32 @__cxa_atexit(void (i8*)* @[[UNREGFN]], i8* bitcast ([[DSCTY]]* [[DESC]] to i8*), @@ -421,31 +442,31 @@ int bar(int a){ // Check metadata is properly generated: // CHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 247, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 265, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 272, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 291, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 415, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 298, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 291, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 298, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 221, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 268, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 286, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 293, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 312, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 436, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 319, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 312, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 319, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 242, i32 {{[0-9]+}}} // TCHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 247, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 265, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 272, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 291, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 415, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 298, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 291, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 298, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 221, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 268, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 286, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 293, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 312, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 436, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 319, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 312, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 319, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 242, i32 {{[0-9]+}}} #endif diff --git a/test/OpenMP/target_teams_distribute_codegen_registration_naming.cpp b/test/OpenMP/target_teams_distribute_codegen_registration_naming.cpp index c03466a52646b..c2195acf3e7e1 100644 --- a/test/OpenMP/target_teams_distribute_codegen_registration_naming.cpp +++ b/test/OpenMP/target_teams_distribute_codegen_registration_naming.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target teams distribute codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,6 +24,16 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/target_teams_distribute_collapse_codegen.cpp b/test/OpenMP/target_teams_distribute_collapse_codegen.cpp index be99186e46816..62fc58980a7c6 100644 --- a/test/OpenMP/target_teams_distribute_collapse_codegen.cpp +++ b/test/OpenMP/target_teams_distribute_collapse_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 template <typename T, int X, long long Y> @@ -60,6 +68,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 template <typename T, int n, int m> diff --git a/test/OpenMP/target_teams_distribute_collapse_messages.cpp b/test/OpenMP/target_teams_distribute_collapse_messages.cpp index 4bc4a7b79ccf1..4010c5a9a9c9f 100644 --- a/test/OpenMP/target_teams_distribute_collapse_messages.cpp +++ b/test/OpenMP/target_teams_distribute_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_default_messages.cpp b/test/OpenMP/target_teams_distribute_default_messages.cpp index 1bd106769784c..2a3229dde5ec6 100644 --- a/test/OpenMP/target_teams_distribute_default_messages.cpp +++ b/test/OpenMP/target_teams_distribute_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/target_teams_distribute_defaultmap_messages.cpp b/test/OpenMP/target_teams_distribute_defaultmap_messages.cpp index 6608cee3de589..a31015255ccf4 100644 --- a/test/OpenMP/target_teams_distribute_defaultmap_messages.cpp +++ b/test/OpenMP/target_teams_distribute_defaultmap_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_depend_codegen.cpp b/test/OpenMP/target_teams_distribute_depend_codegen.cpp new file mode 100644 index 0000000000000..f066574e8768d --- /dev/null +++ b/test/OpenMP/target_teams_distribute_depend_codegen.cpp @@ -0,0 +1,261 @@ +// Test host codegen. +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// CHECK-DAG: [[TT:%.+]] = type { i64, i8 } +// CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } +// CHECK-DAG: [[DEVTY:%.+]] = type { i8*, i8*, [[ENTTY]]*, [[ENTTY]]* } +// CHECK-DAG: [[DSCTY:%.+]] = type { i32, [[DEVTY]]*, [[ENTTY]]*, [[ENTTY]]* } + +// TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 } + +// CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat + +// CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 0, i[[SZ]] 4] +// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 800] +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] + +// Check if offloading descriptor is created. +// CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] +// CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] +// CHECK: [[DEVBEGIN:@.+]] = external constant i8 +// CHECK: [[DEVEND:@.+]] = external constant i8 +// CHECK: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x [[DEVTY]]] [{{.+}} { i8* [[DEVBEGIN]], i8* [[DEVEND]], [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }], comdat($[[REGFN]]) +// CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) + +// Check target registration is registered as a Ctor. +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] + + +template<typename tx, typename ty> +struct TT{ + tx X; + ty Y; +}; + +int global; +extern int global; + +// CHECK: define {{.*}}[[FOO:@.+]]( +int foo(int n) { + int a = 0; + short aa = 0; + float b[10]; + float bn[n]; + double c[5][10]; + double cn[5][n]; + TT<long long, char> d; + static long *plocal; + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID:@.+]], i32 [[GTID:%.+]], i32 1, i[[SZ]] {{20|40}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY0:%.+]]* + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 3 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + #pragma omp target teams distribute device(global + a) depend(in: global) depend(out: a, b, cn[4]) + for (int i = 0; i < 10; ++i) { + } + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + + // CHECK: [[BOOL:%.+]] = icmp ne i32 %{{.+}}, 0 + // CHECK: br i1 [[BOOL]], label %[[THEN:.+]], label %[[ELSE:.+]] + // CHECK: [[THEN]]: + // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%.+]], i32 0, i32 0 + // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%.+]], i32 0, i32 0 + // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]** + // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]** + // CHECK-DAG: store i[[SZ]]* [[BP0:%[^,]+]], i[[SZ]]** [[CBPADDR0]] + // CHECK-DAG: store i[[SZ]]* [[BP0]], i[[SZ]]** [[CPADDR0]] + + // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1 + // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1 + // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* + // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* + // CHECK-DAG: store i[[SZ]] [[BP1:%[^,]+]], i[[SZ]]* [[CBPADDR1]] + // CHECK-DAG: store i[[SZ]] [[BP1]], i[[SZ]]* [[CPADDR1]] + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{104|52}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1_:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1_:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + + // CHECK: [[ELSE]]: + // CHECK-NOT: getelementptr inbounds [2 x i8*], [2 x i8*]* + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1__:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1__:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + // CHECK: [[EXIT]]: + + #pragma omp target teams distribute device(global + a) nowait depend(inout: global, a, bn) if(a) + for (int i = 0; i < *plocal; ++i) { + static int local1; + *plocal = global; + local1 = global; + } + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]* + // CHECK: getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + #pragma omp target teams distribute if(0) firstprivate(global) depend(out:global) + for (int i = 0; i < global; ++i) { + global += 1; + } + + return a; +} + +// Check that the offloading functions are emitted and that the arguments are +// correct and loaded correctly for the target regions in foo(). + +// CHECK: define internal void [[HVT0:@.+]]() + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, [[TASK_TY0]]* noalias) +// CHECK: store void (i8*, ...)* null, void (i8*, ...)** % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 0, i8** null, i8** null, i[[SZ]]* null, i64* null, i32 0, i32 0) +// CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: call void [[HVT0]]() +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT1:@.+]](i[[SZ]]* %{{.+}}, i[[SZ]] %{{.+}}) + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1_]](i32{{.*}}, [[TASK_TY1_]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[SZT:%.+]] = getelementptr inbounds [2 x i[[SZ]]], [2 x i[[SZ]]]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams_nowait(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 2, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SZT]], i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT]], i32 0, i32 0), i32 0, i32 0) + +// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1__]](i32{{.*}}, [[TASK_TY1__]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT2:@.+]](i[[SZ]] %{{.+}}) +// Create stack storage and store argument in there. +// CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align +// CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align +// CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* +// CHECK-64: load i32, i32* [[AA_CADDR]], align +// CHECK-32: load i32, i32* [[AA_ADDR]], align + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, [[TASK_TY2]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT2]](i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + + +#endif diff --git a/test/OpenMP/target_teams_distribute_depend_messages.cpp b/test/OpenMP/target_teams_distribute_depend_messages.cpp index fbdd49590f87f..9daba670203ee 100644 --- a/test/OpenMP/target_teams_distribute_depend_messages.cpp +++ b/test/OpenMP/target_teams_distribute_depend_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_device_messages.cpp b/test/OpenMP/target_teams_distribute_device_messages.cpp index b35d0befce210..aa62e0997168e 100644 --- a/test/OpenMP/target_teams_distribute_device_messages.cpp +++ b/test/OpenMP/target_teams_distribute_device_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_dist_schedule_codegen.cpp b/test/OpenMP/target_teams_distribute_dist_schedule_codegen.cpp index 0c9b2387debf8..3eb153173006a 100644 --- a/test/OpenMP/target_teams_distribute_dist_schedule_codegen.cpp +++ b/test/OpenMP/target_teams_distribute_dist_schedule_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 template <typename T, int X, long long Y> @@ -81,6 +89,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 template <typename T, int n> diff --git a/test/OpenMP/target_teams_distribute_dist_schedule_messages.cpp b/test/OpenMP/target_teams_distribute_dist_schedule_messages.cpp index 87708706a69ee..2544b1b9abbca 100644 --- a/test/OpenMP/target_teams_distribute_dist_schedule_messages.cpp +++ b/test/OpenMP/target_teams_distribute_dist_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp b/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp index 35abffa926004..5487fa86e8f01 100644 --- a/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp +++ b/test/OpenMP/target_teams_distribute_firstprivate_codegen.cpp @@ -1,13 +1,26 @@ -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER @@ -97,16 +110,7 @@ int main() { // LAMBDA: [[G_ADDR:%.+]] = alloca i{{[0-9]+}}, // LAMBDA: [[G1_ADDR:%.+]] = alloca i{{[0-9]+}}, // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, - // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, - // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, - // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, - // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, - // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, - // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, - // LAMBDA: [[G_PRIV_ADDR:%.+]] = alloca i{{[0-9]+}}, - // LAMBDA: [[G1_PRIV_ADDR:%.+]] = alloca i{{[0-9]+}}, // LAMBDA: [[G1_TMP:%.+]] = alloca i32*, - // LAMBDA: [[SIVAR_PRIV_ADDR:%.+]] = alloca i{{[0-9]+}}, // skip loop vars // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_ADDR]], // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_ADDR]], @@ -114,15 +118,15 @@ int main() { // LAMBDA-DAG: [[G_CONV:%.+]] = bitcast {{.+}} [[G_ADDR]] to // LAMBDA-DAG: [[G1_CONV:%.+]] = bitcast {{.+}} [[G1_ADDR]] to // LAMBDA-DAG: [[SIVAR_CONV:%.+]] = bitcast {{.+}} [[SIVAR_ADDR]] to - // LAMBDA-DAG: store{{.+}} [[G1_PRIV_ADDR]], {{.+}} [[G1_TMP]], + // LAMBDA-DAG: store{{.+}} [[G1_CONV]], {{.+}} [[G1_TMP]], g = 1; g1 = 1; sivar = 2; // LAMBDA: call void @__kmpc_for_static_init_4( - // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_PRIV_ADDR]], + // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_CONV]], // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP]] // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1]], - // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_PRIV_ADDR]], + // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_CONV]], // LAMBDA-DAG: [[G1_REF:%.+]] = load{{.+}}, {{.+}} [[G1_TMP]], // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1_REF]], // LAMBDA: call void [[INNER_LAMBDA:@.+]]( @@ -212,8 +216,8 @@ int main() { // CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[SIVAR_ADDR]], // T_VAR and SIVAR -// CHECK-DAG-64: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* -// CHECK-DAG-64: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* // preparation vars // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], @@ -223,7 +227,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], @@ -243,13 +247,13 @@ int main() { // CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) // CHECK: call void @__kmpc_for_static_init_4( -// CHECK-DAG-32: {{.+}} = {{.+}} [[T_VAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] // CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[VAR_PRIV]] -// CHECK-DAG-32: {{.+}} = {{.+}} [[SIVAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_SIVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[SIVAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_SIVAR]] // CHECK: call void @__kmpc_for_static_fini( // CHECK: ret void @@ -309,7 +313,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], @@ -331,8 +335,8 @@ int main() { // CHECK-DAG: store [[S_INT_TY]]* [[VAR_PRIV]], [[S_INT_TY]]** [[TMP]], // CHECK: call void @__kmpc_for_static_init_4( -// CHECK-DAG-32: {{.+}} = {{.+}} [[T_VAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] // CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[TMP]] // CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] diff --git a/test/OpenMP/target_teams_distribute_firstprivate_messages.cpp b/test/OpenMP/target_teams_distribute_firstprivate_messages.cpp index 934149e7c1398..01c1a2818f82d 100644 --- a/test/OpenMP/target_teams_distribute_firstprivate_messages.cpp +++ b/test/OpenMP/target_teams_distribute_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_if_messages.cpp b/test/OpenMP/target_teams_distribute_if_messages.cpp index c9ef03a61ba04..94d82ab5ae655 100644 --- a/test/OpenMP/target_teams_distribute_if_messages.cpp +++ b/test/OpenMP/target_teams_distribute_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_lastprivate_codegen.cpp b/test/OpenMP/target_teams_distribute_lastprivate_codegen.cpp index b41a6016079b4..26a7c0d61da83 100644 --- a/test/OpenMP/target_teams_distribute_lastprivate_codegen.cpp +++ b/test/OpenMP/target_teams_distribute_lastprivate_codegen.cpp @@ -5,12 +5,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -243,7 +259,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK-DAG: [[S_ARR_PTR_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_PTR]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST]], i8* [[TMP_VAL_BCAST]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST]], // CHECK: call void @__kmpc_for_static_fini( // lastprivates @@ -257,7 +273,7 @@ int main() { // CHECK-32: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -268,7 +284,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_FLOAT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -277,7 +293,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_ADDR_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: [[SVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[S_VAR_PRIV]], // CHECK-64: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR_REF]], // CHECK-32: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR]], @@ -345,7 +361,7 @@ int main() { // CHECK-DAG: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK-DAG: [[S_ARR_PTR_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_PTR1]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST1]], i8* [[TMP_VAL_BCAST1]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST1]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST1]], // CHECK: call void @__kmpc_for_static_fini( // lastprivates @@ -359,7 +375,7 @@ int main() { // CHECK-32: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR1]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF1]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_PRIV1]] to [[S_INT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -370,7 +386,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_INT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -379,6 +395,6 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_ADDR1_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: ret void #endif diff --git a/test/OpenMP/target_teams_distribute_lastprivate_messages.cpp b/test/OpenMP/target_teams_distribute_lastprivate_messages.cpp index 898291833d20f..0739846cee0a6 100644 --- a/test/OpenMP/target_teams_distribute_lastprivate_messages.cpp +++ b/test/OpenMP/target_teams_distribute_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_loop_messages.cpp b/test/OpenMP/target_teams_distribute_loop_messages.cpp index 0a825ab1e6095..b090c288a2018 100644 --- a/test/OpenMP/target_teams_distribute_loop_messages.cpp +++ b/test/OpenMP/target_teams_distribute_loop_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s -Wno-openmp-target + +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s -Wno-openmp-target class S { int a; diff --git a/test/OpenMP/target_teams_distribute_map_messages.cpp b/test/OpenMP/target_teams_distribute_map_messages.cpp index eafdccb66bc91..826a09dd76d5d 100644 --- a/test/OpenMP/target_teams_distribute_map_messages.cpp +++ b/test/OpenMP/target_teams_distribute_map_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/target_teams_distribute_messages.cpp b/test/OpenMP/target_teams_distribute_messages.cpp index 91f3c76a4a094..d3be0f632be3e 100644 --- a/test/OpenMP/target_teams_distribute_messages.cpp +++ b/test/OpenMP/target_teams_distribute_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -std=c++11 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -std=c++11 -o - %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_misc_messages.c b/test/OpenMP/target_teams_distribute_misc_messages.c index 33496a70d1405..6d17b512619a3 100644 --- a/test/OpenMP/target_teams_distribute_misc_messages.c +++ b/test/OpenMP/target_teams_distribute_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp target teams distribute'}} #pragma omp target teams distribute diff --git a/test/OpenMP/target_teams_distribute_nowait_messages.cpp b/test/OpenMP/target_teams_distribute_nowait_messages.cpp index 17aa1dcac36e1..b7a9a2548be5c 100644 --- a/test/OpenMP/target_teams_distribute_nowait_messages.cpp +++ b/test/OpenMP/target_teams_distribute_nowait_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -o - %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_num_teams_messages.cpp b/test/OpenMP/target_teams_distribute_num_teams_messages.cpp index 9b36e9c581f0e..8b351dc8fb4af 100644 --- a/test/OpenMP/target_teams_distribute_num_teams_messages.cpp +++ b/test/OpenMP/target_teams_distribute_num_teams_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_ast_print.cpp b/test/OpenMP/target_teams_distribute_parallel_for_ast_print.cpp index f51c4fe60d7f2..a22b561320fa2 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_ast_print.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_ast_print.cpp @@ -1,6 +1,10 @@ -// RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp -ast-print %s -Wno-openmp-target | FileCheck %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print -Wno-openmp-target | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s -Wno-openmp-target | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print -Wno-openmp-target | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -89,7 +93,7 @@ T tmain(T argc) { #pragma omp target teams distribute parallel for for (int i=0; i < 2; ++i) a = 2; -// CHECK: #pragma omp target teams distribute parallel for +// CHECK: #pragma omp target teams distribute parallel for{{$}} // CHECK-NEXT: for (int i = 0; i < 2; ++i) // CHECK-NEXT: a = 2; #pragma omp target teams distribute parallel for private(argc, b), firstprivate(c, d), collapse(2) diff --git a/test/OpenMP/target_teams_distribute_parallel_for_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_codegen.cpp new file mode 100644 index 0000000000000..4d263cf69f02a --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_codegen.cpp @@ -0,0 +1,118 @@ +// expected-no-diagnostics +#ifndef HEADER +#define HEADER +// Test host codegen. +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 --check-prefix HCK1 --check-prefix HCK1-64 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 --check-prefix HCK1 --check-prefix HCK1-64 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 --check-prefix HCK1 --check-prefix HCK1-64 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 --check-prefix HCK1 --check-prefix HCK1-64 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY1 +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY1 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY1 +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY1 + +// Test target codegen - host bc file has to be created first. (no significant differences with host version of target region) +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 --check-prefix TCK1 --check-prefix TCK1-64 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 --check-prefix TCK1 --check-prefix TCK1-64 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 --check-prefix TCK1 --check-prefix TCK1-32 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 --check-prefix TCK1 --check-prefix TCK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix SIMD-ONLY1 +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY1 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix SIMD-ONLY1 +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY1 +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +#ifdef CK1 + +// HCK1: define{{.*}} i32 @{{.+}}target_teams_fun{{.*}}( +int target_teams_fun(int *g){ + int n = 1000; + int a[1000]; + int te = n / 128; + int th = 128; + // discard n_addr + // HCK1: alloca i32, + // HCK1: [[TE:%.+]] alloca i32, + // HCK1: [[TH:%.+]] = alloca i32, + // discard capture expressions for te and th + // HCK1: = alloca i32, + // HCK1: = alloca i32, + // HCK1: [[N_CAST:%.+]] = alloca i{{32|64}}, + // HCK1: [[TE_CAST:%.+]] = alloca i{{32|64}}, + // HCK1: [[TH_CAST:%.+]] = alloca i{{32|64}}, + // HCK1: [[N_PAR:%.+]] = load{{.+}}, {{.+}} [[N_CAST]], + // HCK1: [[TE_PAR:%.+]] = load{{.+}}, {{.+}} [[TE_CAST]], + // HCK1: [[TH_PAR:%.+]] = load{{.+}}, {{.+}} [[TH_CAST]], + // HCK1: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 4, i8** %{{[^,]+}}, i8** %{{[^,]+}}, + + // HCK1: call void @[[OFFL1:.+]](i{{32|64}} [[N_PAR]], {{.+}}, i{{32|64}} [[TE_PAR]], i{{32|64}} [[TH_PAR]]) + #pragma omp target teams distribute parallel for num_teams(te), thread_limit(th) + for(int i = 0; i < n; i++) { + a[i] = 0; + #pragma omp cancel for + } + + // HCK1: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 3, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), + // HCK1: call void @[[OFFL2:.+]](i{{64|32}} %{{.+}}) + {{{ + #pragma omp target teams distribute parallel for is_device_ptr(g) + for(int i = 0; i < n; i++) { + a[i] = g[0]; + } + }}} + + // outlined target regions + // HCK1: define internal void @[[OFFL1]](i{{32|64}} [[N_ARG:%.+]], i{{32|64}} [[TE_ARG:%.+]], i{{32|64}} [[TH_ARG:%.+]]) + // TCK1: define weak void @{{.+}}target_teams_fun{{.*}}(i{{32|64}} [[N_ARG:%.+]], {{.+}}, i{{32|64}} [[TE_ARG:%.+]], i{{32|64}} [[TH_ARG:%.+]]) + // CK1: [[N_ADDR:%.+]] = alloca i{{32|64}}, + // CK1: [[TE_ADDR:%.+]] = alloca i{{32|64}}, + // CK1: [[TH_ADDR:%.+]] = alloca i{{32|64}}, + // TCK1: store {{.+}} [[N_ARG]], {{.+}} [[N_ADDR]], + // CK1: store{{.+}} [[TE_ARG]], {{.+}} [[TE_ADDR]], + // CK1: store{{.+}} [[TH_ARG]], {{.+}} [[TH_ADDR]], + // CK1-64: [[TE_CONV:%.+]] = bitcast{{.+}} [[TE_ADDR]] to + // CK1-64: [[TH_CONV:%.+]] = bitcast{{.+}} [[TH_ADDR]] to + // CK1-64: [[TE_VAL:%.+]] = load i32, i32* [[TE_CONV]], + // CK1-64: [[TH_VAL:%.+]] = load i32, i32* [[TH_CONV]], + // CK1-32: [[TE_VAL:%.+]] = load i32, i32* [[TE_ADDR]], + // CK1-32: [[TH_VAL:%.+]] = load i32, i32* [[TH_ADDR]], + // CK1: {{%.+}} = call i32 @__kmpc_push_num_teams({{.+}}, {{.+}}, i32 [[TE_VAL]], i32 [[TH_VAL]]) + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 2, {{.+}} @[[OUTL1:.+]] to {{.+}}, {{.+}}, {{.+}}) + // CK1: ret void + + // CK1: define internal void @[[OUTL1]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void {{.+}} @__kmpc_fork_call( + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // HCK1: define internal void @[[OFFL2]]( + // TCK1: define weak void @{{.+}}target_teams_fun{{.+}}( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[OUTL2:.+]] to {{.+}}, {{.+}}, {{.+}}) + // CK1: ret void + + // CK1: define internal void @[[OUTL2]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void {{.+}} @__kmpc_fork_call( + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + return a[0]; +} + +#endif // CK1 +#endif // HEADER diff --git a/test/OpenMP/target_teams_distribute_parallel_for_collapse_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_collapse_codegen.cpp new file mode 100644 index 0000000000000..80e38925b39ec --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_collapse_codegen.cpp @@ -0,0 +1,157 @@ +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// Test host codegen. +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} +#ifdef CK1 + +template <typename T, int X, long long Y> +struct SS{ + T a[X][Y]; + + // CK1: define {{.*}}i32 @{{.+}}foo{{.+}}( + int foo(void) { + + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL1:.+]]( + #pragma omp target teams distribute parallel for collapse(2) + for(int i = 0; i < X; i++) { + for(int j = 0; j < Y; j++) { + a[i][j] = (T)0; + } + } + // CK1: define internal void @[[OFFL1]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL1:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL1]]({{.+}}) + // discard loop variables not needed here + // CK1: [[OMP_UB:%.omp.comb.ub]] = alloca i32, + // CK1: store i32 56087, i32* [[OMP_UB]], + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92, {{.+}}, {{.+}}, i32* [[OMP_UB]], + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL1]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, {{.+}}, {{.+}}, + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + return a[0][0]; + } +}; + +int teams_template_struct(void) { + SS<int, 123, 456> V; + return V.foo(); + +} +#endif // CK1 + +// Test host codegen. +// RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64 +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64 +// RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} +#ifdef CK2 + +template <typename T, int n, int m> +int tmain(T argc) { + T a[n][m]; + #pragma omp target teams distribute parallel for collapse(2) + for(int i = 0; i < n; i++) { + for(int j = 0; j < m; j++) { + a[i][j] = (T)0; + } + } + return 0; +} + +int main (int argc, char **argv) { + int n = 100; + int m = 2; + int a[n][m]; + #pragma omp target teams distribute parallel for collapse(2) + for(int i = 0; i < n; i++) { + for(int j = 0; j < m; j++) { + a[i][j] = 0; + } + } + return tmain<int, 10, 2>(argc); +} + +// CK2: define {{.*}}i32 @{{[^,]+}}(i{{.+}}{{.+}} %[[ARGC:.+]], {{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL1:.+]]({{.+}}) +// CK2: {{%.+}} = call{{.*}} i32 @[[TMAIN:.+]]({{.+}}) +// CK2: ret + +// CK2: define {{.*}}void @[[OFFL1]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 5, {{.+}} @[[OUTL1:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL1]]({{.+}}) +// CK2: [[OMP_UB:%.omp.comb.ub]] = alloca i64, +// CK2: store i64 {{.+}}, i64* [[OMP_UB]], +// CK2: call void @__kmpc_for_static_init_8({{.+}}, {{.+}}, i32 92, {{.+}}, {{.+}}, i64* [[OMP_UB]], +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_{{[4|8]}}({{.+}}, {{.+}}, i32 34, {{.+}}, {{.+}}, +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + + +// CK2: define {{.*}}i32 @[[TMAIN]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT1:.+]]({{.+}}) +// CK2: ret +// CK2-NEXT: } + +// CK2: define {{.*}}void @[[OFFLT1]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTLT1:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT1]]({{.+}}) +// discard loop variables not needed here +// CK2: [[OMP_UB:%.omp.comb.ub]] = alloca i32, +// CK2: store i32 {{.+}}, i32* [[OMP_UB]], +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92, {{.+}}, {{.+}}, i32* [[OMP_UB]], +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[TPAR_OUTL1:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[TPAR_OUTL1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, {{.+}}, {{.+}}, +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +#endif // CK2 +#endif // #ifndef HEADER diff --git a/test/OpenMP/target_teams_distribute_parallel_for_collapse_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_collapse_messages.cpp index bd94ba5205ae2..61e76cde020e2 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_collapse_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp %s -std=c++98 // RUN: %clang_cc1 -verify -fopenmp %s -std=c++11 +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s -std=c++98 +// RUN: %clang_cc1 -verify -fopenmp-simd %s -std=c++11 + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_default_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_default_messages.cpp index db47e68a8a321..d7c3fbaa9c75e 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_default_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/target_teams_distribute_parallel_for_defaultmap_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_defaultmap_messages.cpp index ad0295d461c4d..8de99aeb214be 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_defaultmap_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_defaultmap_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_depend_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_depend_codegen.cpp new file mode 100644 index 0000000000000..5cfa2cca51b52 --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_depend_codegen.cpp @@ -0,0 +1,261 @@ +// Test host codegen. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// CHECK-DAG: [[TT:%.+]] = type { i64, i8 } +// CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } +// CHECK-DAG: [[DEVTY:%.+]] = type { i8*, i8*, [[ENTTY]]*, [[ENTTY]]* } +// CHECK-DAG: [[DSCTY:%.+]] = type { i32, [[DEVTY]]*, [[ENTTY]]*, [[ENTTY]]* } + +// TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 } + +// CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat + +// CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 0, i[[SZ]] 4] +// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 800] +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] + +// Check if offloading descriptor is created. +// CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] +// CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] +// CHECK: [[DEVBEGIN:@.+]] = external constant i8 +// CHECK: [[DEVEND:@.+]] = external constant i8 +// CHECK: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x [[DEVTY]]] [{{.+}} { i8* [[DEVBEGIN]], i8* [[DEVEND]], [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }], comdat($[[REGFN]]) +// CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) + +// Check target registration is registered as a Ctor. +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] + + +template<typename tx, typename ty> +struct TT{ + tx X; + ty Y; +}; + +int global; +extern int global; + +// CHECK: define {{.*}}[[FOO:@.+]]( +int foo(int n) { + int a = 0; + short aa = 0; + float b[10]; + float bn[n]; + double c[5][10]; + double cn[5][n]; + TT<long long, char> d; + static long *plocal; + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID:@.+]], i32 [[GTID:%.+]], i32 1, i[[SZ]] {{20|40}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY0:%.+]]* + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 3 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + #pragma omp target teams distribute parallel for device(global + a) depend(in: global) depend(out: a, b, cn[4]) + for (int i = 0; i < 10; ++i) { + } + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + + // CHECK: [[BOOL:%.+]] = icmp ne i32 %{{.+}}, 0 + // CHECK: br i1 [[BOOL]], label %[[THEN:.+]], label %[[ELSE:.+]] + // CHECK: [[THEN]]: + // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%.+]], i32 0, i32 0 + // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%.+]], i32 0, i32 0 + // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]** + // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]** + // CHECK-DAG: store i[[SZ]]* [[BP0:%[^,]+]], i[[SZ]]** [[CBPADDR0]] + // CHECK-DAG: store i[[SZ]]* [[BP0]], i[[SZ]]** [[CPADDR0]] + + // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1 + // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1 + // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* + // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* + // CHECK-DAG: store i[[SZ]] [[BP1:%[^,]+]], i[[SZ]]* [[CBPADDR1]] + // CHECK-DAG: store i[[SZ]] [[BP1]], i[[SZ]]* [[CPADDR1]] + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{104|52}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1_:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1_:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + + // CHECK: [[ELSE]]: + // CHECK-NOT: getelementptr inbounds [2 x i8*], [2 x i8*]* + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1__:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1__:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + // CHECK: [[EXIT]]: + + #pragma omp target teams distribute parallel for device(global + a) nowait depend(inout: global, a, bn) if(target:a) + for (int i = 0; i < *plocal; ++i) { + static int local1; + *plocal = global; + local1 = global; + } + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]* + // CHECK: getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + #pragma omp target teams distribute parallel for if(0) firstprivate(global) depend(out:global) + for (int i = 0; i < global; ++i) { + global += 1; + } + + return a; +} + +// Check that the offloading functions are emitted and that the arguments are +// correct and loaded correctly for the target regions in foo(). + +// CHECK: define internal void [[HVT0:@.+]]() + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, [[TASK_TY0]]* noalias) +// CHECK: store void (i8*, ...)* null, void (i8*, ...)** % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 0, i8** null, i8** null, i[[SZ]]* null, i64* null, i32 0, i32 0) +// CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: call void [[HVT0]]() +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT1:@.+]](i[[SZ]]* %{{.+}}, i[[SZ]] %{{.+}}) + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1_]](i32{{.*}}, [[TASK_TY1_]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[SZT:%.+]] = getelementptr inbounds [2 x i[[SZ]]], [2 x i[[SZ]]]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams_nowait(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 2, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SZT]], i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT]], i32 0, i32 0), i32 0, i32 0) + +// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1__]](i32{{.*}}, [[TASK_TY1__]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT2:@.+]](i[[SZ]] %{{.+}}) +// Create stack storage and store argument in there. +// CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align +// CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align +// CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* +// CHECK-64: load i32, i32* [[AA_CADDR]], align +// CHECK-32: load i32, i32* [[AA_ADDR]], align + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, [[TASK_TY2]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT2]](i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + + +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_depend_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_depend_messages.cpp index c4688958e5545..253e139b82a6f 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_depend_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_depend_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_device_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_device_messages.cpp index b1b86cd4080be..64b682a601b3e 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_device_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_device_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_dist_schedule_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_dist_schedule_codegen.cpp new file mode 100644 index 0000000000000..28619f883f9dd --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_dist_schedule_codegen.cpp @@ -0,0 +1,270 @@ +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// Test host codegen. +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} +#ifdef CK1 + +template <typename T, int X, long long Y> +struct SS{ + T a[X]; + float b; + // CK1: define {{.*}}i32 @{{.+}}foo{{.+}}( + int foo(void) { + + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL1:.+]]( + #pragma omp target teams distribute parallel for + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL2:.+]]( + #pragma omp target teams distribute parallel for dist_schedule(static) + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL3:.+]]( + #pragma omp target teams distribute parallel for dist_schedule(static, X/2) + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + // CK1: define internal void @[[OFFL1]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL1:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL1]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92 + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL1]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[OFFL2]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL2:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL2]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92 + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL2:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL2]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + + // CK1: define internal void @[[OFFL3]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL3:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL3]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 91 + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL3:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL3]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + return a[0]; + } +}; + +int teams_template_struct(void) { + SS<int, 123, 456> V; + return V.foo(); + +} +#endif // CK1 + +// Test host codegen. +// RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64 +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64 +// RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} +#ifdef CK2 + +template <typename T, int n> +int tmain(T argc) { + T a[n]; + int m = 10; +#pragma omp target teams distribute parallel for + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } +#pragma omp target teams distribute parallel for dist_schedule(static) + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } +#pragma omp target teams distribute parallel for dist_schedule(static, m) + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } + return 0; +} + +int main (int argc, char **argv) { + int n = 100; + int a[n]; + int m = 10; +#pragma omp target teams distribute parallel for + for(int i = 0; i < n; i++) { + a[i] = 0; + } +#pragma omp target teams distribute parallel for dist_schedule(static) + for(int i = 0; i < n; i++) { + a[i] = 0; + } +#pragma omp target teams distribute parallel for dist_schedule(static, m) + for(int i = 0; i < n; i++) { + a[i] = 0; + } + return tmain<int, 10>(argc); +} + +// CK2: define {{.*}}i32 @{{[^,]+}}(i{{.+}}{{.+}} %[[ARGC:.+]], {{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL1:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL2:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL3:.+]]({{.+}}) +// CK2: {{%.+}} = call{{.*}} i32 @[[TMAIN:.+]]({{.+}}) +// CK2: ret + +// CK2: define {{.*}}void @[[OFFL1]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[OUTL1:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92 +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFL2]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[OUTL2:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92 +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL2:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFL3]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 4, {{.+}} @[[OUTL3:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 91 +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL3:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}i32 @[[TMAIN]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT1:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT2:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT3:.+]]({{.+}}) +// CK2: ret +// CK2-NEXT: } + +// CK2: define {{.*}}void @[[OFFLT1]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTLT1:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92 +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT1:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFLT2]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTLT2:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92 +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT2:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFLT3]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[OUTLT3:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 91 +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT3:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +#endif // CK2 +#endif // #ifndef HEADER diff --git a/test/OpenMP/target_teams_distribute_parallel_for_dist_schedule_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_dist_schedule_messages.cpp index d9913bc962122..a8abaffb464a6 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_dist_schedule_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_dist_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp new file mode 100644 index 0000000000000..843b667316a56 --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_codegen.cpp @@ -0,0 +1,523 @@ +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix HCHECK --check-prefix HCHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix HCHECK --check-prefix HCHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix HCHECK --check-prefix HCHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix HCHECK --check-prefix HCHECK-32 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 --check-prefix HLAMBDA --check-prefix HLAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 --check-prefix HLAMBDA --check-prefix HLAMBDA-64 + +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY + +// Test target codegen - host bc file has to be created first. (no significant differences with host version of target region) +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 --check-prefix TLAMBDA --check-prefix TLAMBDA-64 + +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +struct St { + int a, b; + St() : a(0), b(0) {} + St(const St &st) : a(st.a + st.b), b(0) {} + ~St() {} +}; + +volatile int g = 1212; +volatile int &g1 = g; + +template <class T> +struct S { + T f; + S(T a) : f(a + g) {} + S() : f(g) {} + S(const S &s, St t = St()) : f(s.f + t.a) {} + operator T() { return T(); } + ~S() {} +}; + +// CHECK-DAG: [[S_FLOAT_TY:%.+]] = type { float } +// CHECK-DAG: [[S_INT_TY:%.+]] = type { i{{[0-9]+}} } +// CHECK-DAG: [[ST_TY:%.+]] = type { i{{[0-9]+}}, i{{[0-9]+}} } + +template <typename T> +T tmain() { + S<T> test; + T t_var = T(); + T vec[] = {1, 2}; + S<T> s_arr[] = {1, 2}; + S<T> &var = test; +#pragma omp target teams distribute parallel for firstprivate(t_var, vec, s_arr, var) + for (int i = 0; i < 2; ++i) { + vec[i] = t_var; + s_arr[i] = var; + } + return T(); +} + +// HCHECK-DAG: [[TEST:@.+]] = global [[S_FLOAT_TY]] zeroinitializer, +S<float> test; +// HCHECK-DAG: [[T_VAR:@.+]] = global i{{[0-9]+}} 333, +int t_var = 333; +// HCHECK-DAG: [[VEC:@.+]] = global [2 x i{{[0-9]+}}] [i{{[0-9]+}} 1, i{{[0-9]+}} 2], +int vec[] = {1, 2}; +// HCHECK-DAG: [[S_ARR:@.+]] = global [2 x [[S_FLOAT_TY]]] zeroinitializer, +S<float> s_arr[] = {1, 2}; +// HCHECK-DAG: [[VAR:@.+]] = global [[S_FLOAT_TY]] zeroinitializer, +S<float> var(3); +// HCHECK-DAG: [[SIVAR:@.+]] = internal global i{{[0-9]+}} 0, + +int main() { + static int sivar; +#ifdef LAMBDA + // HLAMBDA: [[G:@.+]] = global i{{[0-9]+}} 1212, + // HLAMBDA-LABEL: @main + // HLAMBDA: call void [[OUTER_LAMBDA:@.+]]( + [&]() { + // HLAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]]( + // HLAMBDA: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 3, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i32 0, i32 0) + // HLAMBDA: call void @[[LOFFL1:.+]](i{{64|32}} %{{.+}}) + // HLAMBDA: ret +#pragma omp target teams distribute parallel for firstprivate(g, g1, sivar) + for (int i = 0; i < 2; ++i) { + // HLAMBDA: define{{.*}} internal{{.*}} void @[[LOFFL1]](i{{64|32}} {{%.+}}, i{{64|32}} {{%.+}}) + // TLAMBDA: define weak void @[[LOFFL1:.+]](i{{64|32}} {{%.+}}, i{{64|32}} {{%.+}}) + // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, + // LAMBDA: [[G_CAST:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[G1_CAST:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[SIVAR_CAST:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA-DAG: [[G_CAST_VAL:%.+]] = load{{.+}} [[G_CAST]], + // LAMBDA-DAG: [[G1_CAST_VAL:%.+]] = load{{.+}} [[G1_CAST]], + // LAMBDA-DAG: [[SIVAR_CAST_VAL:%.+]] = load{{.+}} [[SIVAR_CAST]], + // LAMBDA: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[LOUTL1:.+]] to {{.+}}, {{.+}} [[G_CAST_VAL]], {{.+}} [[G1_CAST_VAL]], {{.+}} [[SIVAR_CAST_VAL]]) + // LAMBDA: ret void + + // LAMBDA: define internal void @[[LOUTL1]]({{.+}}) + // Skip global and bound tid vars + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: [[G_ADDR:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[G1_ADDR:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[G1_TMP:%.+]] = alloca i32*, + // skip loop vars + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_ADDR]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_ADDR]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[SIVAR_ADDR]], + // LAMBDA-DAG: [[G_CONV:%.+]] = bitcast {{.+}} [[G_ADDR]] to + // LAMBDA-DAG: [[G1_CONV:%.+]] = bitcast {{.+}} [[G1_ADDR]] to + // LAMBDA-DAG: [[SIVAR_CONV:%.+]] = bitcast {{.+}} [[SIVAR_ADDR]] to + // LAMBDA-DAG: store{{.+}} [[G1_CONV]], {{.+}} [[G1_TMP]], + g = 1; + g1 = 1; + sivar = 2; + // LAMBDA: call void @__kmpc_for_static_init_4( + // LAMBDA: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[LPAR_OUTL:.+]] to + // LAMBDA: call void @__kmpc_for_static_fini( + // LAMBDA: ret void + + // LAMBDA: define internal void @[[LPAR_OUTL]]({{.+}}) + // Skip global and bound tid vars, and prev lb and ub vars + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: [[G_ADDR:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[G1_ADDR:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[G1_TMP:%.+]] = alloca i32*, + // skip loop vars + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_ADDR]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_ADDR]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[SIVAR_ADDR]], + // LAMBDA-DAG: [[G_CONV:%.+]] = bitcast {{.+}} [[G_ADDR]] to + // LAMBDA-DAG: [[G1_CONV:%.+]] = bitcast {{.+}} [[G1_ADDR]] to + // LAMBDA-DAG: [[SIVAR_CONV:%.+]] = bitcast {{.+}} [[SIVAR_ADDR]] to + // LAMBDA-DAG: store{{.+}} [[G1_CONV]], {{.+}} [[G1_TMP]], + + // use of private vars + // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_CONV]], + // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP]] + // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1]], + // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_CONV]], + // LAMBDA-DAG: [[G1_REF:%.+]] = load{{.+}}, {{.+}} [[G1_TMP]], + // LAMBDA: call void [[INNER_LAMBDA:@.+]]( + // LAMBDA: call void @__kmpc_for_static_fini( + // LAMBDA: ret void + [&]() { + // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) + // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], + g = 2; + g1 = 2; + sivar = 4; + // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]] + + // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 + // LAMBDA: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_PTR_REF]] + // LAMBDA: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[G_REF]] + // LAMBDA: [[G1_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 + // LAMBDA: [[G1_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G1_PTR_REF]] + // LAMBDA: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[G1_REF]] + // LAMBDA: [[SIVAR_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 + // LAMBDA: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_PTR_REF]] + // LAMBDA: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIVAR_REF]] + }(); + } + }(); + return 0; +#else +#pragma omp target teams distribute parallel for firstprivate(t_var, vec, s_arr, var, sivar) + for (int i = 0; i < 2; ++i) { + vec[i] = t_var; + s_arr[i] = var; + sivar += i; + } + return tmain<int>(); +#endif +} + +// HCHECK: define {{.*}}i{{[0-9]+}} @main() +// HCHECK: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 5, +// HCHECK: call void @[[OFFL1:.+]]( +// HCHECK: {{%.+}} = call{{.*}} i32 @[[TMAIN_INT:.+]]() +// HCHECK: ret + +// HCHECK: define{{.*}} void @[[OFFL1]]( +// TCHECK: define{{.*}} void @[[OFFL1:.+]]( +// CHECK-DAG: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK-DAG: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK-DAG: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*, +// CHECK-DAG: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]]*, +// CHECK: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[T_VAR_CAST:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[SIVAR_CAST:%.+]] = alloca i{{[0-9]+}}, + +// CHECK-DAG: [[VEC_TE_PAR:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_PRIV]], +// CHECK-DAG: [[T_VAR_TE_PAR:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_CAST]], +// CHECK-DAG: [[S_ARR_TE_PAR:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[S_ARR_PRIV]], +// CHECK-DAG: [[VAR_TE_PAR:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[VAR_PRIV]], +// CHECK-DAG: [[SIVAR_TE_PAR:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_CAST]], + +// CHECK: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 5, {{.+}} @[[OUTL1:.+]] to {{.+}}, [2 x i{{[0-9]+}}]* [[VEC_TE_PAR]], i{{[0-9]+}} [[T_VAR_TE_PAR]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_TE_PAR]], [[S_FLOAT_TY]]* [[VAR_TE_PAR]], i{{[0-9]+}} [[SIVAR_TE_PAR]]) +// CHECK: ret void + +// CHECK: define internal void @[[OUTL1]]({{.+}}) +// Skip global and bound tid vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: [[VEC_ADDR:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*, +// CHECK: [[VAR_ADDR:%.+]] = alloca [[S_FLOAT_TY]]*, +// CHECK: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// Skip temp vars for loop +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]], +// CHECK: [[AGG_TMP1:%.+]] = alloca [[ST_TY]], +// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]], +// CHECK: [[AGG_TMP2:%.+]] = alloca [[ST_TY]], + +// param copy +// CHECK: store [2 x i{{[0-9]+}}]* {{.+}}, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[T_VAR_ADDR]], +// CHECK: store [2 x [[S_FLOAT_TY]]]* {{.+}}, [2 x [[S_FLOAT_TY]]]** [[S_ARR_ADDR]], +// CHECK: store [[S_FLOAT_TY]]* {{.+}}, [[S_FLOAT_TY]]** [[VAR_ADDR]], +// CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[SIVAR_ADDR]], + +// T_VAR and SIVAR +// CHECK-64-DAG: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* + +// preparation vars +// CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK-DAG: [[S_ARR_ADDR_REF:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[S_ARR_ADDR]], +// CHECK-DAG: [[VAR_ADDR_REF:%.+]] = load{{.+}} [[VAR_ADDR]], + +// firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 +// CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* +// CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) + +// firstprivate(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_ADDR_BGN:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]] to +// CHECK-DAG: [[S_ARR_FIN:%.+]] = icmp{{.+}} [[S_ARR_PRIV_BGN]], +// CHECK-DAG: [[S_ARR_SRC_COPY:%.+]] = phi{{.+}} [ [[S_ARR_ADDR_BGN]], {{.+}} ], [ [[S_ARR_SRC:%.+]], {{.+}} ] +// CHECK-DAG: [[S_ARR_DST_COPY:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}}], [ [[S_ARR_DST:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_DST_COPY]], {{.+}} [[S_ARR_SRC_COPY]], {{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: [[S_ARR_DST]] = getelementptr {{.+}} [[S_ARR_DST_COPY]], +// CHECK-DAG: [[S_ARR_SRC]] = getelementptr {{.+}} [[S_ARR_SRC_COPY]], + +// firstprivate(var) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]], {{.+}} [[VAR_ADDR_REF]], {{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL:.+]] to +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: define internal void @[[PAR_OUTL]]({{.+}}) +// Skip global and bound tid vars, and prev lb ub vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: [[VEC_ADDR:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*, +// CHECK: [[VAR_ADDR:%.+]] = alloca [[S_FLOAT_TY]]*, +// CHECK: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// Skip temp vars for loop +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]], +// CHECK: [[AGG_TMP1:%.+]] = alloca [[ST_TY]], +// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]], +// CHECK: [[AGG_TMP2:%.+]] = alloca [[ST_TY]], + +// param copy +// CHECK: store [2 x i{{[0-9]+}}]* {{.+}}, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[T_VAR_ADDR]], +// CHECK: store [2 x [[S_FLOAT_TY]]]* {{.+}}, [2 x [[S_FLOAT_TY]]]** [[S_ARR_ADDR]], +// CHECK: store [[S_FLOAT_TY]]* {{.+}}, [[S_FLOAT_TY]]** [[VAR_ADDR]], +// CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[SIVAR_ADDR]], + +// T_VAR and SIVAR +// CHECK-64-DAG: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* + +// preparation vars +// CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK-DAG: [[S_ARR_ADDR_REF:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[S_ARR_ADDR]], +// CHECK-DAG: [[VAR_ADDR_REF:%.+]] = load{{.+}} [[VAR_ADDR]], + +// firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 +// CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* +// CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) + +// firstprivate(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_ADDR_BGN:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]] to +// CHECK-DAG: [[S_ARR_FIN:%.+]] = icmp{{.+}} [[S_ARR_PRIV_BGN]], +// CHECK-DAG: [[S_ARR_SRC_COPY:%.+]] = phi{{.+}} [ [[S_ARR_ADDR_BGN]], {{.+}} ], [ [[S_ARR_SRC:%.+]], {{.+}} ] +// CHECK-DAG: [[S_ARR_DST_COPY:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}}], [ [[S_ARR_DST:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_DST_COPY]], {{.+}} [[S_ARR_SRC_COPY]], {{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: [[S_ARR_DST]] = getelementptr {{.+}} [[S_ARR_DST_COPY]], +// CHECK-DAG: [[S_ARR_SRC]] = getelementptr {{.+}} [[S_ARR_SRC_COPY]], + +// firstprivate(var) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]], {{.+}} [[VAR_ADDR_REF]], {{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[VAR_PRIV]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[SIVAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_SIVAR]] +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// HCHECK: define{{.*}} i{{[0-9]+}} @[[TMAIN_INT]]() +// HCHECK: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 4, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i32 0, i32 0) +// HCHECK: call void @[[TOFFL1:.+]]( +// HCHECK: ret + +// HCHECK: define {{.*}}void @[[TOFFL1]]({{.+}}) +// TCHECK: define {{.*}}void @[[TOFFL1:.+]]({{.+}}) +// CHECK-DAG: [[TT_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK-DAG: [[TVEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK-DAG: [[TS_ARR_PRIV:%.+]] = alloca [2 x [[S_INT_TY]]]*, +// CHECK-DAG: [[TVAR_PRIV:%.+]] = alloca [[S_INT_TY]]*, +// CHECK: [[TT_VAR_CAST:%.+]] = alloca i{{[0-9]+}}, + +// CHECK-DAG: [[TVEC_TE_PAR:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[TVEC_PRIV]], +// CHECK-DAG: [[TT_VAR_TE_PAR:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[TT_VAR_CAST]], +// CHECK-DAG: [[TS_ARR_TE_PAR:%.+]] = load [2 x [[S_INT_TY]]]*, [2 x [[S_INT_TY]]]** [[TS_ARR_PRIV]], +// CHECK-DAG: [[TVAR_TE_PAR:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TVAR_PRIV]], + +// CHECK: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 4, {{.+}} @[[TOUTL1:.+]] to {{.+}}, [2 x i{{[0-9]+}}]* [[TVEC_TE_PAR]], i{{[0-9]+}} [[TT_VAR_TE_PAR]], [2 x [[S_INT_TY]]]* [[TS_ARR_TE_PAR]], [[S_INT_TY]]* [[TVAR_TE_PAR]]) +// CHECK: ret void + +// CHECK: define internal void @[[TOUTL1]]({{.+}}) +// Skip global and bound tid vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: [[VEC_ADDR:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR:%.+]] = alloca [2 x [[S_INT_TY]]]*, +// CHECK: [[VAR_ADDR:%.+]] = alloca [[S_INT_TY]]*, +// Skip temp vars for loop +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_INT_TY]]], +// CHECK: [[AGG_TMP1:%.+]] = alloca [[ST_TY]], +// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]], +// CHECK: [[AGG_TMP2:%.+]] = alloca [[ST_TY]], +// CHECK: [[TMP:%.+]] = alloca [[S_INT_TY]]*, + +// param copy +// CHECK: store [2 x i{{[0-9]+}}]* {{.+}}, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[T_VAR_ADDR]], +// CHECK: store [2 x [[S_INT_TY]]]* {{.+}}, [2 x [[S_INT_TY]]]** [[S_ARR_ADDR]], +// CHECK: store [[S_INT_TY]]* {{.+}}, [[S_INT_TY]]** [[VAR_ADDR]], + +// T_VAR and preparation variables +// CHECK: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK-64: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK: [[S_ARR_ADDR_REF:%.+]] = load [2 x [[S_INT_TY]]]*, [2 x [[S_INT_TY]]]** [[S_ARR_ADDR]], + +// firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 +// CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* +// CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) + +// firstprivate(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_ADDR_BGN:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]] to +// CHECK-DAG: [[S_ARR_FIN:%.+]] = icmp{{.+}} [[S_ARR_PRIV_BGN]], +// CHECK-DAG: [[S_ARR_SRC_COPY:%.+]] = phi{{.+}} [ [[S_ARR_ADDR_BGN]], {{.+}} ], [ [[S_ARR_SRC:%.+]], {{.+}} ] +// CHECK-DAG: [[S_ARR_DST_COPY:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_DST:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_DST_COPY]], {{.+}} [[S_ARR_SRC_COPY]], {{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: [[S_ARR_DST]] = getelementptr {{.+}} [[S_ARR_DST_COPY]], +// CHECK-DAG: [[S_ARR_SRC]] = getelementptr {{.+}} [[S_ARR_SRC_COPY]], + +// firstprivate(var) +// CHECK-DAG: [[VAR_ADDR_REF:%.+]] = load{{.+}} [[VAR_ADDR]], +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]], {{.+}} [[VAR_ADDR_REF]], {{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) +// CHECK-DAG: store [[S_INT_TY]]* [[VAR_PRIV]], [[S_INT_TY]]** [[TMP]], + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[TPAR_OUTL:.+]] to +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: define internal void @[[TPAR_OUTL]]({{.+}}) +// Skip global and bound tid vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: [[VEC_ADDR:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR:%.+]] = alloca [2 x [[S_INT_TY]]]*, +// CHECK: [[VAR_ADDR:%.+]] = alloca [[S_INT_TY]]*, +// Skip temp vars for loop +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_INT_TY]]], +// CHECK: [[AGG_TMP1:%.+]] = alloca [[ST_TY]], +// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]], +// CHECK: [[AGG_TMP2:%.+]] = alloca [[ST_TY]], +// CHECK: [[TMP:%.+]] = alloca [[S_INT_TY]]*, + +// param copy +// CHECK: store [2 x i{{[0-9]+}}]* {{.+}}, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[T_VAR_ADDR]], +// CHECK: store [2 x [[S_INT_TY]]]* {{.+}}, [2 x [[S_INT_TY]]]** [[S_ARR_ADDR]], +// CHECK: store [[S_INT_TY]]* {{.+}}, [[S_INT_TY]]** [[VAR_ADDR]], + +// T_VAR and preparation variables +// CHECK: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK-64: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK: [[S_ARR_ADDR_REF:%.+]] = load [2 x [[S_INT_TY]]]*, [2 x [[S_INT_TY]]]** [[S_ARR_ADDR]], + +// firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 +// CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* +// CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) + +// firstprivate(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_ADDR_BGN:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]] to +// CHECK-DAG: [[S_ARR_FIN:%.+]] = icmp{{.+}} [[S_ARR_PRIV_BGN]], +// CHECK-DAG: [[S_ARR_SRC_COPY:%.+]] = phi{{.+}} [ [[S_ARR_ADDR_BGN]], {{.+}} ], [ [[S_ARR_SRC:%.+]], {{.+}} ] +// CHECK-DAG: [[S_ARR_DST_COPY:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_DST:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_DST_COPY]], {{.+}} [[S_ARR_SRC_COPY]], {{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: [[S_ARR_DST]] = getelementptr {{.+}} [[S_ARR_DST_COPY]], +// CHECK-DAG: [[S_ARR_SRC]] = getelementptr {{.+}} [[S_ARR_SRC_COPY]], + +// firstprivate(var) +// CHECK-DAG: [[VAR_ADDR_REF:%.+]] = load{{.+}} [[VAR_ADDR]], +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]], {{.+}} [[VAR_ADDR_REF]], {{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) +// CHECK-DAG: store [[S_INT_TY]]* [[VAR_PRIV]], [[S_INT_TY]]** [[TMP]], + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[TMP]] +// CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_messages.cpp index 0d83550478304..a278e9f89acd7 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_if_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_if_codegen.cpp new file mode 100644 index 0000000000000..8cd90a610b5cb --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_if_codegen.cpp @@ -0,0 +1,181 @@ +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +void fn1(); +void fn2(); +void fn3(); +void fn4(); +void fn5(); +void fn6(); + +int Arg; + +// CHECK-LABEL: define {{.*}}void @{{.+}}gtid_test +void gtid_test() { +// CHECK: call i{{[0-9]+}} @__tgt_target_teams( +// CHECK: call void [[OFFLOADING_FUN_0:@.+]]( +// CHECK: call i{{[0-9]+}} @__tgt_target_teams( +// CHECK: call void [[OFFLOADING_FUN_1:@.+]]( +#pragma omp target teams distribute parallel for + for(int i = 0 ; i < 100; i++) {} + // CHECK: define internal void [[OFFLOADING_FUN_0]]( + // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}}* [[OMP_TEAMS_OUTLINED_0:@.+]] to {{.+}}) + // CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_0]]( + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, {{.+}}* [[OMP_OUTLINED_0:@.+]] to void + // CHECK: call void @__kmpc_for_static_fini( + + // CHECK: define{{.+}} void [[OMP_OUTLINED_0]]( + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call void @__kmpc_for_static_fini( + // CHECK: ret +#pragma omp target teams distribute parallel for if (parallel: false) + for(int i = 0 ; i < 100; i++) { + // CHECK: define internal void [[OFFLOADING_FUN_1]]( + // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}}* [[OMP_TEAMS_OUTLINED_1:@.+]] to {{.+}}) + // CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_1]]( + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call void @__kmpc_serialized_parallel( + // CHECK: call void [[OMP_OUTLINED_1:@.+]]( + // CHECK: call void @__kmpc_end_serialized_parallel( + // CHECK: call void @__kmpc_for_static_fini( + // CHECK: define{{.+}} void [[OMP_OUTLINED_1]]( + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call void @{{.+}}gtid_test + // CHECK: call void @__kmpc_for_static_fini( + // CHECK: ret + gtid_test(); + } +} + + +template <typename T> +int tmain(T Arg) { +#pragma omp target teams distribute parallel for if (true) + for(int i = 0 ; i < 100; i++) { + fn1(); + } +#pragma omp target teams distribute parallel for if (false) + for(int i = 0 ; i < 100; i++) { + fn2(); + } +#pragma omp target teams distribute parallel for if (parallel: Arg) + for(int i = 0 ; i < 100; i++) { + fn3(); + } + return 0; +} + +// CHECK-LABEL: define {{.*}}i{{[0-9]+}} @main() +int main() { +// CHECK: call i{{[0-9]+}} @__tgt_target_teams( +// CHECK: call void [[OFFLOADING_FUN_0:@.+]]( +// CHECK: call void [[OFFLOADING_FUN_1:@.+]]( +// CHECK: call i{{[0-9]+}} @__tgt_target_teams( +// CHECK: call void [[OFFLOADING_FUN_2:@.+]]( +// CHECK: = call {{.*}}i{{.+}} @{{.+}}tmain +#pragma omp target teams distribute parallel for if (true) + for(int i = 0 ; i < 100; i++) { + // CHECK: define internal void [[OFFLOADING_FUN_0]]( + // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}}* [[OMP_TEAMS_OUTLINED_0:@.+]] to {{.+}}) + // CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_0]]( + + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, {{.+}}* [[OMP_OUTLINED_2:@.+]] to void + // CHECK: call void @__kmpc_for_static_fini( + // CHECK: define{{.+}} void [[OMP_OUTLINED_2]]( + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call {{.*}}void @{{.+}}fn4 + // CHECK: call void @__kmpc_for_static_fini( + + fn4(); + } + +#pragma omp target teams distribute parallel for if (false) + for(int i = 0 ; i < 100; i++) { + // CHECK: define internal void [[OFFLOADING_FUN_1]]( + // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}}* [[OMP_TEAMS_OUTLINED_1:@.+]] to {{.+}}) + // CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_1]]( + + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call void @__kmpc_serialized_parallel( + // CHECK: call void [[OMP_OUTLINED_3:@.+]]( + // CHECK: call void @__kmpc_end_serialized_parallel( + // CHECK: call void @__kmpc_for_static_fini( + + // CHECK: define{{.+}} void [[OMP_OUTLINED_3]]( + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call {{.*}}void @{{.+}}fn5 + // CHECK: call void @__kmpc_for_static_fini( + fn5(); + } + +#pragma omp target teams distribute parallel for if (Arg) + for(int i = 0 ; i < 100; i++) { + // CHECK: define internal void [[OFFLOADING_FUN_2]]( + // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}}* [[OMP_TEAMS_OUTLINED_2:@.+]] to {{.+}}) + // CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_2]]( + + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, {{.+}}* [[OMP_OUTLINED_4:@.+]] to void + // CHECK: call void @__kmpc_serialized_parallel( + // CHECK: call void [[OMP_OUTLINED_4:@.+]]( + // CHECK: call void @__kmpc_end_serialized_parallel( + // CHECK: call void @__kmpc_for_static_fini( + + // CHECK: define{{.+}} void [[OMP_OUTLINED_4]]( + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call {{.*}}void @{{.+}}fn6 + // CHECK: call void @__kmpc_for_static_fini( + fn6(); + } + + return tmain(Arg); +} + +// CHECK-LABEL: define {{.+}} @{{.+}}tmain + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, void {{.+}}* [[T_OUTLINE_FUN_1:@.+]] to void +// CHECK: call void @__kmpc_for_static_fini( + +// CHECK: define internal {{.*}}void [[T_OUTLINE_FUN_1]] +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call {{.*}}void @{{.+}}fn1 +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call {{.*}}void @__kmpc_serialized_parallel( +// CHECK: call void [[T_OUTLINE_FUN_2:@.+]]( +// CHECK: call {{.*}}void @__kmpc_end_serialized_parallel( +// CHECK: call void @__kmpc_for_static_fini( + +// CHECK: define internal {{.*}}void [[T_OUTLINE_FUN_2]] +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call {{.*}}void @{{.+}}fn2 +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, void {{.+}}* [[T_OUTLINE_FUN_3:@.+]] to void +// CHECK: call {{.*}}void @__kmpc_serialized_parallel( +// call void [[T_OUTLINE_FUN_3:@.+]]( +// CHECK: call {{.*}}void @__kmpc_end_serialized_parallel( + +// CHECK: define internal {{.*}}void [[T_OUTLINE_FUN_3]] +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call {{.*}}void @{{.+}}fn3 +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_if_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_if_messages.cpp index 82d39ac4ca8cf..6f990f7bd28e1 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_if_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_is_device_ptr_ast_print.cpp b/test/OpenMP/target_teams_distribute_parallel_for_is_device_ptr_ast_print.cpp index c566319df42d7..b3944bc19530b 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_is_device_ptr_ast_print.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_is_device_ptr_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -56,7 +60,7 @@ struct SA { // CHECK-NEXT: int aa[10]; // CHECK-NEXT: arr &raa = this->aa; // CHECK-NEXT: func( -// CHECK-NEXT: #pragma omp target teams distribute parallel for is_device_ptr(this->k) +// CHECK-NEXT: #pragma omp target teams distribute parallel for is_device_ptr(this->k){{$}} // CHECK-NEXT: for (int i = 0; i < 100; i++) // CHECK-NEXT: ; // CHECK-NEXT: #pragma omp target teams distribute parallel for is_device_ptr(this->z) diff --git a/test/OpenMP/target_teams_distribute_parallel_for_is_device_ptr_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_is_device_ptr_messages.cpp index eaa3190a47f61..e19e46c57b71f 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_is_device_ptr_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_is_device_ptr_messages.cpp @@ -1,4 +1,6 @@ // RUN: %clang_cc1 -std=c++11 -verify -fopenmp %s + +// RUN: %clang_cc1 -std=c++11 -verify -fopenmp-simd %s struct ST { int *a; }; diff --git a/test/OpenMP/target_teams_distribute_parallel_for_lastprivate_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_lastprivate_codegen.cpp new file mode 100644 index 0000000000000..acb1c0b636ece --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_lastprivate_codegen.cpp @@ -0,0 +1,454 @@ +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 + +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +template <class T> +struct S { + T f; + S(T a) : f(a) {} + S() : f() {} + operator T() { return T(); } + ~S() {} +}; + +// CHECK: [[S_FLOAT_TY:%.+]] = type { float } +// CHECK: [[S_INT_TY:%.+]] = type { i{{[0-9]+}} } +template <typename T> +T tmain() { + S<T> test; + T t_var = T(); + T vec[] = {1, 2}; + S<T> s_arr[] = {1, 2}; + S<T> &var = test; + #pragma omp target teams distribute parallel for lastprivate(t_var, vec, s_arr, s_arr, var, var) + for (int i = 0; i < 2; ++i) { + vec[i] = t_var; + s_arr[i] = var; + } + return T(); +} + +int main() { + static int svar; + volatile double g; + volatile double &g1 = g; + + #ifdef LAMBDA + // LAMBDA-LABEL: @main + // LAMBDA: call{{.*}} void [[OUTER_LAMBDA:@.+]]( + [&]() { + static float sfvar; + // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]]( + // LAMBDA: call i{{[0-9]+}} @__tgt_target_teams( + // LAMBDA: call void [[OFFLOADING_FUN:@.+]]( + + // LAMBDA: define{{.+}} void [[OFFLOADING_FUN]]( + // LAMBDA: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 4, {{.+}}* [[OMP_OUTLINED:@.+]] to {{.+}}) + #pragma omp target teams distribute parallel for lastprivate(g, g1, svar, sfvar) + for (int i = 0; i < 2; ++i) { + // LAMBDA: define{{.*}} internal{{.*}} void [[OMP_OUTLINED]](i32* {{.+}}, i32* {{.+}}, {{.+}} [[G1_IN:%.+]], {{.+}} [[SVAR_IN:%.+]], {{.+}} [[SFVAR_IN:%.+]], {{.+}} [[G_IN:%.+]]) + // skip gbl and bound tid + // LAMBDA: alloca + // LAMBDA: alloca + // LAMBDA: [[G1_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA: [[SVAR_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA: [[SFVAR_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA: [[G_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA-64: [[G1_REF:%.+]] = alloca double*, + // loop variables + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: [[OMP_IS_LAST:%.+]] = alloca i{{[0-9]+}}, + + // LAMBDA-DAG: store {{.+}} [[G_IN]], {{.+}} [[G_ADDR]], + // LAMBDA-DAG: store {{.+}} [[G1_IN]], {{.+}} [[G1_ADDR]], + // LAMBDA-DAG: store {{.+}} [[SVAR_IN]], {{.+}} [[SVAR_ADDR]], + // LAMBDA-DAG: store {{.+}} [[SFVAR_IN]], {{.+}} [[SFVAR_ADDR]], + + // LAMBDA-64-DAG: [[G_TGT:%.+]] = bitcast {{.+}} [[G_ADDR]] to + // LAMBDA-32-DAG: [[G_TGT:%.+]] = load {{.+}}, {{.+}} [[G_ADDR]], + // LAMBDA-64-DAG: [[G1_TGT:%.+]] = load {{.+}}, {{.+}} [[G1_REF]], + // LAMBDA-32-DAG: [[G1_TGT:%.+]] = load {{.+}}, {{.+}} [[G1_ADDR]], + // LAMBDA-64-DAG: [[SVAR_TGT:%.+]] = bitcast {{.+}} [[SVAR_ADDR]] to + // LAMBDA-DAG: [[SFVAR_TGT:%.+]] = bitcast {{.+}} [[SFVAR_ADDR]] to + + g1 = 1; + svar = 3; + sfvar = 4.0; + // LAMBDA: call {{.*}}void @__kmpc_for_static_init_4( + // LAMBDA: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[LPAR_OUTL:.+]] to + // LAMBDA: call {{.*}}void @__kmpc_for_static_fini( + + // LAMBDA: [[OMP_IS_LAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[OMP_IS_LAST]], + // LAMBDA: [[IS_LAST_IT:%.+]] = icmp ne i{{[0-9]+}} [[OMP_IS_LAST_VAL]], 0 + // LAMBDA: br i1 [[IS_LAST_IT]], label %[[OMP_LASTPRIV_BLOCK:.+]], label %[[OMP_LASTPRIV_DONE:.+]] + + // LAMBDA: [[OMP_LASTPRIV_BLOCK]]: + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_TGT]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_TGT]], + // LAMBDA-64-DAG: store {{.+}}, {{.+}} [[SVAR_TGT]], + // LAMBDA-32-DAG: store {{.+}}, {{.+}} [[SVAR_ADDR]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[SFVAR_TGT]], + // LAMBDA: br label %[[OMP_LASTPRIV_DONE]] + // LAMBDA: [[OMP_LASTPRIV_DONE]]: + // LAMBDA: ret + + // LAMBDA: define{{.*}} internal{{.*}} void @[[LPAR_OUTL]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, {{.+}}, {{.+}}, {{.+}} [[G1_IN:%.+]], {{.+}} [[SVAR_IN:%.+]], {{.+}} [[SFVAR_IN:%.+]], {{.+}} [[G_IN:%.+]]) + // skip tid and prev variables + // LAMBDA: alloca + // LAMBDA: alloca + // LAMBDA: alloca + // LAMBDA: alloca + // LAMBDA: [[G1_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA: [[SVAR_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA: [[SFVAR_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA: [[G_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA-64: [[G1_REF:%.+]] = alloca double*, + // loop variables + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: [[OMP_IS_LAST:%.+]] = alloca i{{[0-9]+}}, + + // LAMBDA-DAG: store {{.+}} [[G_IN]], {{.+}} [[G_ADDR]], + // LAMBDA-DAG: store {{.+}} [[G1_IN]], {{.+}} [[G1_ADDR]], + // LAMBDA-DAG: store {{.+}} [[SVAR_IN]], {{.+}} [[SVAR_ADDR]], + // LAMBDA-DAG: store {{.+}} [[SFVAR_IN]], {{.+}} [[SFVAR_ADDR]], + + // LAMBDA-64-DAG: [[G_TGT:%.+]] = bitcast {{.+}} [[G_ADDR]] to + // LAMBDA-32-DAG: [[G_TGT:%.+]] = load {{.+}}, {{.+}} [[G_ADDR]], + // LAMBDA-64-DAG: [[G1_TGT:%.+]] = load {{.+}}, {{.+}} [[G1_REF]], + // LAMBDA-32-DAG: [[G1_TGT:%.+]] = load {{.+}}, {{.+}} [[G1_ADDR]], + // LAMBDA-64-DAG: [[SVAR_TGT:%.+]] = bitcast {{.+}} [[SVAR_ADDR]] to + // LAMBDA-DAG: [[SFVAR_TGT:%.+]] = bitcast {{.+}} [[SFVAR_ADDR]] to + + + // LAMBDA: call {{.*}}void @__kmpc_for_static_init_4( + // LAMBDA: call{{.*}} void [[INNER_LAMBDA:@.+]]( + // LAMBDA: call {{.*}}void @__kmpc_for_static_fini( + + // LAMBDA: [[OMP_IS_LAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[OMP_IS_LAST]], + // LAMBDA: [[IS_LAST_IT:%.+]] = icmp ne i{{[0-9]+}} [[OMP_IS_LAST_VAL]], 0 + // LAMBDA: br i1 [[IS_LAST_IT]], label %[[OMP_LASTPRIV_BLOCK:.+]], label %[[OMP_LASTPRIV_DONE:.+]] + + // LAMBDA: [[OMP_LASTPRIV_BLOCK]]: + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_TGT]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_TGT]], + // LAMBDA-64-DAG: store {{.+}}, {{.+}} [[SVAR_TGT]], + // LAMBDA-32-DAG: store {{.+}}, {{.+}} [[SVAR_ADDR]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[SFVAR_TGT]], + // LAMBDA: br label %[[OMP_LASTPRIV_DONE]] + // LAMBDA: [[OMP_LASTPRIV_DONE]]: + // LAMBDA: ret + + [&]() { + // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) + // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], + g = 2; + g1 = 2; + svar = 4; + sfvar = 8.0; + // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]] + // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 + // LAMBDA: [[G_REF:%.+]] = load double*, double** [[G_PTR_REF]] + // LAMBDA: store double 2.0{{.+}}, double* [[G_REF]] + + // LAMBDA: [[TMP_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 + // LAMBDA: [[G1_REF:%.+]] = load double*, double** [[TMP_PTR_REF]] + // LAMBDA: store double 2.0{{.+}}, double* [[G1_REF]], + // LAMBDA: [[SVAR_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 + // LAMBDA: [[SVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SVAR_PTR_REF]] + // LAMBDA: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SVAR_REF]] + // LAMBDA: [[SFVAR_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 + // LAMBDA: [[SFVAR_REF:%.+]] = load float*, float** [[SFVAR_PTR_REF]] + // LAMBDA: store float 8.0{{.+}}, float* [[SFVAR_REF]] + }(); + } + }(); + return 0; + #else + S<float> test; + int t_var = 0; + int vec[] = {1, 2}; + S<float> s_arr[] = {1, 2}; + S<float> &var = test; + + #pragma omp target teams distribute parallel for lastprivate(t_var, vec, s_arr, s_arr, var, var, svar) + for (int i = 0; i < 2; ++i) { + vec[i] = t_var; + s_arr[i] = var; + } + int i; + + return tmain<int>(); + #endif +} + +// CHECK: define{{.*}} i{{[0-9]+}} @main() +// CHECK: [[TEST:%.+]] = alloca [[S_FLOAT_TY]], +// CHECK: call {{.*}} [[S_FLOAT_TY_DEF_CONSTR:@.+]]([[S_FLOAT_TY]]* [[TEST]]) +// CHECK: call i{{[0-9]+}} @__tgt_target_teams( +// CHECK: call void [[OFFLOAD_FUN:@.+]]( +// CHECK: ret + +// CHECK: define{{.+}} [[OFFLOAD_FUN]]( +// CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_teams( +// CHECK: ret +// +// CHECK: define internal void [[OMP_OUTLINED:@.+]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [2 x i{{[0-9]+}}]*{{.+}} [[VEC_IN:%.+]], i{{[0-9]+}}{{.+}} [[T_VAR_IN:%.+]], [2 x [[S_FLOAT_TY]]]*{{.+}} [[S_ARR_IN:%.+]], [[S_FLOAT_TY]]*{{.+}} [[VAR_IN:%.+]], i{{[0-9]+}}{{.*}} [[S_VAR_IN:%.+]]) +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: [[VEC_ADDR:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*, +// CHECK: [[VAR_ADDR:%.+]] = alloca [[S_FLOAT_TY]]*, +// CHECK: [[SVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// skip loop variables +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: [[OMP_IS_LAST:%.+]] = alloca i{{[0-9]+}}, + +// copy from parameters to local address variables +// CHECK: store {{.+}} [[VEC_IN]], {{.+}} [[VEC_ADDR]], +// CHECK: store {{.+}} [[T_VAR_IN]], {{.+}} [[T_VAR_ADDR]], +// CHECK: store {{.+}} [[S_ARR_IN]], {{.+}} [[S_ARR_ADDR]], +// CHECK: store {{.+}} [[VAR_IN]], {{.+}} [[VAR_ADDR]], +// CHECK: store {{.+}} [[S_VAR_IN]], {{.+}} [[SVAR_ADDR]], + +// prepare lastprivate targets +// CHECK-64-DAG: [[TVAR_TGT:%.+]] = bitcast {{.+}} [[T_VAR_ADDR]] to +// CHECK-DAG: [[VEC_TGT:%.+]] = load {{.+}}, {{.+}} [[VEC_ADDR]], +// CHECK-DAG: [[S_ARR_TGT:%.+]] = load {{.+}}, {{.+}} [[S_ARR_ADDR]], +// CHECK-DAG: [[VAR_TGT:%.+]] = load {{.+}}, {{.+}} [[VAR_ADDR]], +// CHECK-64-DAG: [[SVAR_TGT:%.+]] = bitcast {{.+}} [[SVAR_ADDR]] to + +// the distribute loop +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL:.+]] to +// CHECK: call void @__kmpc_for_static_fini( + +// lastprivates +// CHECK: [[OMP_IS_LAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[OMP_IS_LAST]], +// CHECK: [[IS_LAST_IT:%.+]] = icmp ne i{{[0-9]+}} [[OMP_IS_LAST_VAL]], 0 +// CHECK: br i1 [[IS_LAST_IT]], label %[[OMP_LASTPRIV_BLOCK:.+]], label %[[OMP_LASTPRIV_DONE:.+]] + +// CHECK: [[OMP_LASTPRIV_BLOCK]]: +// CHECK-64-DAG: store {{.+}}, {{.+}} [[TVAR_TGT]], +// CHECK-32-DAG: store {{.+}}, {{.+}} [[T_VAR_ADDR]], +// CHECK-DAG: [[VEC_TGT_REF:%.+]] = bitcast {{.+}} [[VEC_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_TGT_REF]], +// CHECK-DAG: [[S_ARR_BEGIN:%.+]] = getelementptr {{.+}} [[S_ARR_TGT]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}( +// CHECK-DAG: [[VAR_TGT_BCAST:%.+]] = bitcast {{.+}} [[VAR_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_TGT_BCAST]], +// CHECK-64-DAG: store {{.+}}, {{.+}} [[SVAR_TGT]], +// CHECK-32-DAG: store {{.+}}, {{.+}} [[SVAR_ADDR]], +// CHECK: ret void + +// CHECK: define internal void [[OMP_OUTLINED:@.+]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, {{.+}}, {{.+}}, [2 x i{{[0-9]+}}]*{{.+}} [[VEC_IN:%.+]], i{{[0-9]+}}{{.+}} [[T_VAR_IN:%.+]], [2 x [[S_FLOAT_TY]]]*{{.+}} [[S_ARR_IN:%.+]], [[S_FLOAT_TY]]*{{.+}} [[VAR_IN:%.+]], i{{[0-9]+}}{{.*}} [[S_VAR_IN:%.+]]) + +// gbl and bound tid vars, prev lb and ub vars +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, + +// CHECK: [[VEC_ADDR:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*, +// CHECK: [[VAR_ADDR:%.+]] = alloca [[S_FLOAT_TY]]*, +// CHECK: [[SVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// skip loop variables +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: [[OMP_IS_LAST:%.+]] = alloca i{{[0-9]+}}, + +// copy from parameters to local address variables +// CHECK: store {{.+}} [[VEC_IN]], {{.+}} [[VEC_ADDR]], +// CHECK: store {{.+}} [[T_VAR_IN]], {{.+}} [[T_VAR_ADDR]], +// CHECK: store {{.+}} [[S_ARR_IN]], {{.+}} [[S_ARR_ADDR]], +// CHECK: store {{.+}} [[VAR_IN]], {{.+}} [[VAR_ADDR]], +// CHECK: store {{.+}} [[S_VAR_IN]], {{.+}} [[SVAR_ADDR]], + +// prepare lastprivate targets +// CHECK-64-DAG: [[TVAR_TGT:%.+]] = bitcast {{.+}} [[T_VAR_ADDR]] to +// CHECK-DAG: [[VEC_TGT:%.+]] = load {{.+}}, {{.+}} [[VEC_ADDR]], +// CHECK-DAG: [[S_ARR_TGT:%.+]] = load {{.+}}, {{.+}} [[S_ARR_ADDR]], +// CHECK-DAG: [[VAR_TGT:%.+]] = load {{.+}}, {{.+}} [[VAR_ADDR]], +// CHECK-64-DAG: [[SVAR_TGT:%.+]] = bitcast {{.+}} [[SVAR_ADDR]] to + +// the distribute loop +// CHECK: call void @__kmpc_for_static_init_4( +// skip body: code generation routine is same as distribute parallel for lastprivate +// CHECK: call void @__kmpc_for_static_fini( + +// lastprivates +// CHECK: [[OMP_IS_LAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[OMP_IS_LAST]], +// CHECK: [[IS_LAST_IT:%.+]] = icmp ne i{{[0-9]+}} [[OMP_IS_LAST_VAL]], 0 +// CHECK: br i1 [[IS_LAST_IT]], label %[[OMP_LASTPRIV_BLOCK:.+]], label %[[OMP_LASTPRIV_DONE:.+]] + +// CHECK: [[OMP_LASTPRIV_BLOCK]]: +// CHECK-64-DAG: store {{.+}}, {{.+}} [[TVAR_TGT]], +// CHECK-32-DAG: store {{.+}}, {{.+}} [[T_VAR_ADDR]], +// CHECK-DAG: [[VEC_TGT_REF:%.+]] = bitcast {{.+}} [[VEC_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_TGT_REF]], +// CHECK-DAG: [[S_ARR_BEGIN:%.+]] = getelementptr {{.+}} [[S_ARR_TGT]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}( +// CHECK-DAG: [[VAR_TGT_BCAST:%.+]] = bitcast {{.+}} [[VAR_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_TGT_BCAST]], +// CHECK-64-DAG: store {{.+}}, {{.+}} [[SVAR_TGT]], +// CHECK-32-DAG: store {{.+}}, {{.+}} [[SVAR_ADDR]], +// CHECK: ret void + +// template tmain +// CHECK: define{{.*}} i{{[0-9]+}} [[TMAIN_INT:@.+]]() +// CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]], +// CHECK: call {{.*}} [[S_INT_TY_DEF_CONSTR:@.+]]([[S_INT_TY]]* [[TEST]]) +// CHECK: call i{{[0-9]+}} @__tgt_target_teams( +// CHECK: call void [[OFFLOAD_FUN_1:@.+]]( +// CHECK: ret + +// CHECK: define internal void [[OFFLOAD_FUN_1]]( +// CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_teams(%{{.+}}* @{{.+}}, i{{[0-9]+}} 4, +// CHECK: ret + +// CHECK: define internal void [[OMP_OUTLINED_1:@.+]](i{{[0-9]+}}* noalias [[GTID_ADDR1:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [2 x i{{[0-9]+}}]*{{.+}} [[VEC_IN1:%.+]], i{{[0-9]+}}{{.+}} [[T_VAR_IN1:%.+]], [2 x [[S_INT_TY]]]*{{.+}} [[S_ARR_IN1:%.+]], [[S_INT_TY]]*{{.+}} [[VAR_IN1:%.+]]) +// skip alloca of global_tid and bound_tid +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: [[VEC_ADDR1:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR1:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR1:%.+]] = alloca [2 x [[S_INT_TY]]]*, +// CHECK: [[VAR_ADDR1:%.+]] = alloca [[S_INT_TY]]*, +// skip loop variables +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: [[OMP_IS_LAST1:%.+]] = alloca i{{[0-9]+}}, + +// copy from parameters to local address variables +// CHECK: store {{.+}} [[VEC_IN1]], {{.+}} [[VEC_ADDR1]], +// CHECK: store {{.+}} [[T_VAR_IN1]], {{.+}} [[T_VAR_ADDR1]], +// CHECK: store {{.+}} [[S_ARR_IN1]], {{.+}} [[S_ARR_ADDR1]], +// CHECK: store {{.+}} [[VAR_IN1]], {{.+}} [[VAR_ADDR1]], + +// prepare lastprivate targets +// CHECK-64-DAG: [[T_VAR_TGT:%.+]] = bitcast {{.+}} [[T_VAR_ADDR1]] to +// CHECK-DAG: [[VEC_TGT:%.+]] = load {{.+}}, {{.+}} [[VEC_ADDR1]], +// CHECK-DAG: [[S_ARR_TGT:%.+]] = load {{.+}}, {{.+}} [[S_ARR_ADDR1]], +// CHECK-DAG: [[VAR_TGT:%.+]] = load {{.+}}, {{.+}} [[VAR_ADDR1]], + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[TPAR_OUTL:.+]] to +// CHECK: call void @__kmpc_for_static_fini( + +// lastprivates +// CHECK: [[OMP_IS_LAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[OMP_IS_LAST1]], +// CHECK: [[IS_LAST_IT:%.+]] = icmp ne i{{[0-9]+}} [[OMP_IS_LAST_VAL]], 0 +// CHECK: br i1 [[IS_LAST_IT]], label %[[OMP_LASTPRIV_BLOCK:.+]], label %[[OMP_LASTPRIV_DONE:.+]] + +// CHECK: [[OMP_LASTPRIV_BLOCK]]: +// CHECK-64-DAG: store {{.+}}, {{.+}} [[T_VAR_TGT]], +// CHECK-32-DAG: store {{.+}}, {{.+}} [[T_VAR_ADDR1]], +// CHECK-DAG: [[VEC_TGT_BCAST:%.+]] = bitcast {{.+}} [[VEC_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_TGT_BCAST]], +// CHECK-DAG: {{.+}} = getelementptr {{.+}} [[S_ARR_TGT]], +// CHECK: call void @llvm.memcpy.{{.+}}( +// CHECK-DAG: [[VAR_ADDR_BCAST:%.+]] = bitcast {{.+}} [[VAR_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_BCAST]], +// CHECK: ret void + +// CHECK: define internal void [[TPAR_OUTL:@.+]](i{{[0-9]+}}* noalias [[GTID_ADDR1:%.+]], i{{[0-9]+}}* noalias %{{.+}}, {{.+}}, {{.+}}, [2 x i{{[0-9]+}}]*{{.+}} [[VEC_IN1:%.+]], i{{[0-9]+}}{{.+}} [[T_VAR_IN1:%.+]], [2 x [[S_INT_TY]]]*{{.+}} [[S_ARR_IN1:%.+]], [[S_INT_TY]]*{{.+}} [[VAR_IN1:%.+]]) +// skip alloca of global_tid and bound_tid, and prev lb and ub vars +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, + +// CHECK: [[VEC_ADDR1:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR1:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR1:%.+]] = alloca [2 x [[S_INT_TY]]]*, +// CHECK: [[VAR_ADDR1:%.+]] = alloca [[S_INT_TY]]*, +// skip loop variables +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: [[OMP_IS_LAST1:%.+]] = alloca i{{[0-9]+}}, + +// copy from parameters to local address variables +// CHECK: store {{.+}} [[VEC_IN1]], {{.+}} [[VEC_ADDR1]], +// CHECK: store {{.+}} [[T_VAR_IN1]], {{.+}} [[T_VAR_ADDR1]], +// CHECK: store {{.+}} [[S_ARR_IN1]], {{.+}} [[S_ARR_ADDR1]], +// CHECK: store {{.+}} [[VAR_IN1]], {{.+}} [[VAR_ADDR1]], + +// prepare lastprivate targets +// CHECK-64-DAG: [[T_VAR_TGT:%.+]] = bitcast {{.+}} [[T_VAR_ADDR1]] to +// CHECK-DAG: [[VEC_TGT:%.+]] = load {{.+}}, {{.+}} [[VEC_ADDR1]], +// CHECK-DAG: [[S_ARR_TGT:%.+]] = load {{.+}}, {{.+}} [[S_ARR_ADDR1]], +// CHECK-DAG: [[VAR_TGT:%.+]] = load {{.+}}, {{.+}} [[VAR_ADDR1]], + +// CHECK: call void @__kmpc_for_static_init_4( +// skip body: code generation routine is same as distribute parallel for lastprivate +// CHECK: call void @__kmpc_for_static_fini( + +// lastprivates +// CHECK: [[OMP_IS_LAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[OMP_IS_LAST1]], +// CHECK: [[IS_LAST_IT:%.+]] = icmp ne i{{[0-9]+}} [[OMP_IS_LAST_VAL]], 0 +// CHECK: br i1 [[IS_LAST_IT]], label %[[OMP_LASTPRIV_BLOCK:.+]], label %[[OMP_LASTPRIV_DONE:.+]] + +// CHECK: [[OMP_LASTPRIV_BLOCK]]: +// CHECK-64-DAG: store {{.+}}, {{.+}} [[T_VAR_TGT]], +// CHECK-32-DAG: store {{.+}}, {{.+}} [[T_VAR_ADDR1]], +// CHECK-DAG: [[VEC_TGT_BCAST:%.+]] = bitcast {{.+}} [[VEC_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_TGT_BCAST]], +// CHECK-DAG: {{.+}} = getelementptr {{.+}} [[S_ARR_TGT]], +// CHECK: call void @llvm.memcpy.{{.+}}( +// CHECK-DAG: [[VAR_ADDR_BCAST:%.+]] = bitcast {{.+}} [[VAR_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_BCAST]], +// CHECK: ret void + +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_lastprivate_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_lastprivate_messages.cpp index 84a6df28174d6..b21b9118f913a 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_lastprivate_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_loop_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_loop_messages.cpp index 57ab273c4601a..01bf38deb97c3 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_loop_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_loop_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s -Wno-openmp-target + +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s -Wno-openmp-target class S { int a; diff --git a/test/OpenMP/target_teams_distribute_parallel_for_map_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_map_messages.cpp index 1fb466ee5518f..5cba1e732d3f8 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_map_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_map_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_messages.cpp index 15135838e981d..551acd2e8f0ff 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_misc_messages.c b/test/OpenMP/target_teams_distribute_parallel_for_misc_messages.c index c48a2dfa9e680..820eb11189a3c 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_misc_messages.c +++ b/test/OpenMP/target_teams_distribute_parallel_for_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp target teams distribute parallel for'}} #pragma omp target teams distribute parallel for diff --git a/test/OpenMP/target_teams_distribute_parallel_for_nowait_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_nowait_messages.cpp index 30a622439dbfc..d0e679936e6c4 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_nowait_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_nowait_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -verify -fopenmp %s +// RUN: %clang_cc1 -fsyntax-only -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_num_teams_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_num_teams_messages.cpp index fdc2d9c259186..ceb4fa567c69d 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_num_teams_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_num_teams_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_num_threads_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_num_threads_messages.cpp index afd8c367a0dc4..f558936afae6b 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_num_threads_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_num_threads_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_private_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_private_codegen.cpp new file mode 100644 index 0000000000000..87a1499689304 --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_private_codegen.cpp @@ -0,0 +1,369 @@ +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix HHECK --check-prefix HCHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix HHECK --check-prefix HCHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix HHECK --check-prefix HCHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix HHECK --check-prefix HCHECK-64 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 --check-prefix HLAMBDA --check-prefix HLAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 --check-prefix HLAMBDA --check-prefix HLAMBDA-64 + +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 + +// Test target codegen - host bc file has to be created first. (no significant differences with host version of target region) +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 --check-prefix TLAMBDA --check-prefix TLAMBDA-64 + +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +struct St { + int a, b; + St() : a(0), b(0) {} + St(const St &st) : a(st.a + st.b), b(0) {} + ~St() {} +}; + +volatile int g = 1212; +volatile int &g1 = g; + +template <class T> +struct S { + T f; + S(T a) : f(a + g) {} + S() : f(g) {} + S(const S &s, St t = St()) : f(s.f + t.a) {} + operator T() { return T(); } + ~S() {} +}; + +// CHECK-DAG: [[S_FLOAT_TY:%.+]] = type { float } +// CHECK-DAG: [[S_INT_TY:%.+]] = type { i{{[0-9]+}} } + +template <typename T> +T tmain() { + S<T> test; + T t_var = T(); + T vec[] = {1, 2}; + S<T> s_arr[] = {1, 2}; + S<T> &var = test; +#pragma omp target teams distribute parallel for private(t_var, vec, s_arr, var) + for (int i = 0; i < 2; ++i) { + vec[i] = t_var; + s_arr[i] = var; + } + return T(); +} + +// HCHECK-DAG: [[TEST:@.+]] ={{.*}} global [[S_FLOAT_TY]] zeroinitializer, +S<float> test; +// HCHECK-DAG: [[T_VAR:@.+]] ={{.+}} global i{{[0-9]+}} 333, +int t_var = 333; +// HCHECK-DAG: [[VEC:@.+]] ={{.+}} global [2 x i{{[0-9]+}}] [i{{[0-9]+}} 1, i{{[0-9]+}} 2], +int vec[] = {1, 2}; +// HCHECK-DAG: [[S_ARR:@.+]] ={{.+}} global [2 x [[S_FLOAT_TY]]] zeroinitializer, +S<float> s_arr[] = {1, 2}; +// HCHECK-DAG: [[VAR:@.+]] ={{.+}} global [[S_FLOAT_TY]] zeroinitializer, +S<float> var(3); +// HCHECK-DAG: [[SIVAR:@.+]] = internal global i{{[0-9]+}} 0, + +int main() { + static int sivar; +#ifdef LAMBDA + // HLAMBDA: [[G:@.+]] = global i{{[0-9]+}} 1212, + // HLAMBDA-LABEL: @main + // HLAMBDA: call void [[OUTER_LAMBDA:@.+]]( + [&]() { + // HLAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]]( + // HLAMBDA: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 0, + // HLAMBDA: call void @[[LOFFL1:.+]]( + // HLAMBDA: ret +#pragma omp target teams distribute parallel for private(g, g1, sivar) + for (int i = 0; i < 2; ++i) { + // HLAMBDA: define{{.*}} internal{{.*}} void @[[LOFFL1]]() + // TLAMBDA: define{{.*}} void @[[LOFFL1:.+]]() + // LAMBDA: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}} @[[LOUTL1:.+]] to {{.+}}) + // LAMBDA: ret void + + // LAMBDA: define internal void @[[LOUTL1]]({{.+}}) + // Skip global, bound tid and loop vars + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: [[G_PRIV:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[G1_PRIV:%.+]] = alloca i{{[0-9]+}} + // LAMBDA: [[TMP:%.+]] = alloca i{{[0-9]+}}*, + // LAMBDA: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: store{{.+}} [[G1_PRIV]], {{.+}} [[TMP]], + + g = 1; + g1 = 1; + sivar = 2; + // LAMBDA: call void @__kmpc_for_static_init_4( + // LAMBDA: call void {{.+}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[LPAR_OUTL:.+]] to + // LAMBDA: call void @__kmpc_for_static_fini( + // LAMBDA: ret void + + // LAMBDA: define internal void @[[LPAR_OUTL]]({{.+}}) + // Skip global, bound tid and loop vars + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: alloca i{{[0-9]+}}, + // LAMBDA: alloca i{{[0-9]+}}, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: [[G_PRIV:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[G1_PRIV:%.+]] = alloca i{{[0-9]+}} + // LAMBDA: [[TMP:%.+]] = alloca i{{[0-9]+}}*, + // LAMBDA: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: store{{.+}} [[G1_PRIV]], {{.+}} [[TMP]], + // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_PRIV]], + // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_PRIV]], + // LAMBDA-DAG: [[G1_REF:%.+]] = load{{.+}}, {{.+}} [[TMP]], + // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1_REF]], + // LAMBDA: call void [[INNER_LAMBDA:@.+]]( + // LAMBDA: call void @__kmpc_for_static_fini( + // LAMBDA: ret void + [&]() { + // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) + // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], + g = 2; + g1 = 2; + sivar = 4; + // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]] + + // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 + // LAMBDA: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_PTR_REF]] + // LAMBDA: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[G_REF]] + // LAMBDA: [[G1_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 + // LAMBDA: [[G1_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G1_PTR_REF]] + // LAMBDA: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[G1_REF]] + // LAMBDA: [[SIVAR_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 + // LAMBDA: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_PTR_REF]] + // LAMBDA: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIVAR_REF]] + }(); + } + }(); + return 0; +#else +#pragma omp target teams distribute parallel for private(t_var, vec, s_arr, var, sivar) + for (int i = 0; i < 2; ++i) { + vec[i] = t_var; + s_arr[i] = var; + sivar += i; + } + return tmain<int>(); +#endif +} + +// HCHECK: define {{.*}}i{{[0-9]+}} @main() +// HCHECK: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 0, i8** null, i8** null, {{.+}} null, {{.+}} null, i32 0, i32 0) +// HCHECK: call void @[[OFFL1:.+]]() +// HCHECK: {{%.+}} = call{{.*}} i32 @[[TMAIN_INT:.+]]() +// HCHECK: ret + +// HCHECK: define{{.*}} void @[[OFFL1]]() +// TCHECK: define{{.*}} void @[[OFFL1:.+]]() +// CHECK: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}} @[[OUTL1:.+]] to {{.+}}) +// CHECK: ret void + +// CHECK: define internal void @[[OUTL1]]({{.+}}) +// Skip global, bound tid and loop vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK-DAG: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK-DAG: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK-DAG: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]], +// CHECK-DAG: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]], +// CHECK-DAG: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK: alloca i32, + +// private(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_PTR_ALLOC:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_NEXT:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_PTR_ALLOC]]) +// CHECK-DAG: [[S_ARR_NEXT]] = getelementptr {{.+}} [[S_ARR_PTR_ALLOC]], + +// private(var) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]]) + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.+}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: define internal void @[[PAR_OUTL1]]({{.+}}) +// Skip global, bound tid and loop vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK-DAG: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK-DAG: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK-DAG: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]], +// CHECK-DAG: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]], +// CHECK-DAG: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK: alloca i32, + +// private(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_PTR_ALLOC:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_NEXT:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_PTR_ALLOC]]) +// CHECK-DAG: [[S_ARR_NEXT]] = getelementptr {{.+}} [[S_ARR_PTR_ALLOC]], + +// private(var) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]]) + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK-DAG: {{.+}} = {{.+}} [[T_VAR_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[VAR_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[SIVAR_PRIV]] +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// HCHECK: define{{.*}} i{{[0-9]+}} @[[TMAIN_INT]]() +// HCHECK: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 0, +// HCHECK: call void @[[TOFFL1:.+]]() +// HCHECK: ret + +// HCHECK: define {{.*}}void @[[TOFFL1]]() +// TCHECK: define weak void @[[TOFFL1:.+]]() +// CHECK: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}} @[[TOUTL1:.+]] to {{.+}}) +// CHECK: ret void + +// CHECK: define internal void @[[TOUTL1]]({{.+}}) +// Skip global, bound tid and loop vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i32, +// CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_INT_TY]]], +// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]], +// CHECK: [[TMP:%.+]] = alloca [[S_INT_TY]]*, + +// private(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_PTR_ALLOC:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_NEXT:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_PTR_ALLOC]]) +// CHECK-DAG: [[S_ARR_NEXT]] = getelementptr {{.+}} [[S_ARR_PTR_ALLOC]], + +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_PTR_ALLOC:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_NEXT:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_PTR_ALLOC]]) +// CHECK-DAG: [[S_ARR_NEXT]] = getelementptr {{.+}} [[S_ARR_PTR_ALLOC]], + +// private(var) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]]) +// CHECK-DAG: store{{.+}} [[VAR_PRIV]], {{.+}} [[TMP]] + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.+}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[TPAR_OUTL1:.+]] to +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: define internal void @[[TPAR_OUTL1]]({{.+}}) +// Skip global, bound tid and loop vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// prev lb and ub +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// iter variables +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i32, +// CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_INT_TY]]], +// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]], +// CHECK: [[TMP:%.+]] = alloca [[S_INT_TY]]*, + +// private(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_PTR_ALLOC:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_NEXT:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_PTR_ALLOC]]) +// CHECK-DAG: [[S_ARR_NEXT]] = getelementptr {{.+}} [[S_ARR_PTR_ALLOC]], + +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_PTR_ALLOC:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_NEXT:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_PTR_ALLOC]]) +// CHECK-DAG: [[S_ARR_NEXT]] = getelementptr {{.+}} [[S_ARR_PTR_ALLOC]], + +// private(var) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]]) +// CHECK-DAG: store{{.+}} [[VAR_PRIV]], {{.+}} [[TMP]] + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK-DAG: {{.+}} = {{.+}} [[T_VAR_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[TMP]] +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + + +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_private_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_private_messages.cpp index d89cb1aa846da..7750238e70b57 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_private_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_proc_bind_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_proc_bind_codegen.cpp new file mode 100644 index 0000000000000..9107c218a436a --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_proc_bind_codegen.cpp @@ -0,0 +1,93 @@ +// add -fopenmp-targets + +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +typedef __INTPTR_TYPE__ intptr_t; + +// CHECK-DAG: [[IDENT_T_TY:%.+]] = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" +// CHECK-DAG: [[DEF_LOC_2:@.+]] = private unnamed_addr constant [[IDENT_T_TY]] { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } + +void foo(); + +struct S { + intptr_t a, b, c; + S(intptr_t a) : a(a) {} + operator char() { return a; } + ~S() {} +}; + +template <typename T> +T tmain() { +#pragma omp target teams distribute parallel for proc_bind(master) + for(int i = 0; i < 1000; i++) {} + return T(); +} + +int main() { + // CHECK-LABEL: @main +#pragma omp target teams distribute parallel for proc_bind(spread) + for(int i = 0; i < 1000; i++) {} +#pragma omp target teams distribute parallel for proc_bind(close) + for(int i = 0; i < 1000; i++) {} + return tmain<int>(); +} + +// CHECK: call {{.*}}@__tgt_target_teams({{.+}}) +// CHECK: call void [[OFFL1:@.+]]() +// CHECK: call {{.*}}@__tgt_target_teams({{.+}}) +// CHECK: call void [[OFFL2:@.+]]() +// CHECK: [[CALL_RET:%.+]] = call{{.+}} i32 [[TMAIN:@.+]]() +// CHECK: ret i32 [[CALL_RET]] + +// CHECK: define{{.+}} void [[OFFL1]]( +// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, {{.+}}, {{.+}}* [[OMP_OUTLINED_1:@.+]] to {{.+}}) + +// CHECK: define{{.+}} [[OMP_OUTLINED_1]](i32* {{.+}} [[GTID_IN:%.+]], +// CHECK: [[GTID_ADDR:%.+]] = alloca i32*, +// CHECK: store i32* [[GTID_IN]], i32** [[GTID_ADDR]], +// CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], +// CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], +// CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 4) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: ret void + +// CHECK: define{{.+}} [[OFFL2]]() +// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, {{.+}}, {{.+}}* [[OMP_OUTLINED_1:@.+]] to {{.+}}) + +// CHECK: define{{.+}} [[OMP_OUTLINED_1]](i32* {{.+}} [[GTID_IN:%.+]], +// CHECK: [[GTID_ADDR:%.+]] = alloca i32*, +// CHECK: store i32* [[GTID_IN]], i32** [[GTID_ADDR]], +// CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], +// CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], +// CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 3) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: ret void + +// CHECK: define{{.+}} [[TMAIN]]() +// CHECK: call {{.*}}@__tgt_target_teams({{.+}}) +// CHECK: call void [[OFFL3:@.+]]() + +// CHECK: define{{.+}} [[OFFL3]]() +// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, {{.+}}, {{.+}}* [[OMP_OUTLINED_3:@.+]] to {{.+}}) + +// CHECK: define{{.+}} [[OMP_OUTLINED_3]](i32* {{.+}} [[GTID_IN:%.+]], +// CHECK: [[GTID_ADDR:%.+]] = alloca i32*, +// CHECK: store i32* [[GTID_IN]], i32** [[GTID_ADDR]], +// CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], +// CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], +// CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 2) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: ret void +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_proc_bind_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_proc_bind_messages.cpp index b2bbfa8bb3371..659684e6f1e3c 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_proc_bind_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_proc_bind_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); template <class T, typename S, int N> diff --git a/test/OpenMP/target_teams_distribute_parallel_for_reduction_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_reduction_codegen.cpp new file mode 100644 index 0000000000000..6575659637a52 --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_reduction_codegen.cpp @@ -0,0 +1,353 @@ +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 + +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +template <typename T> +T tmain() { + T t_var = T(); + T vec[] = {1, 2}; +#pragma omp target teams distribute parallel for reduction(+: t_var) + for (int i = 0; i < 2; ++i) { + t_var += (T) i; + } + return T(); +} + +int main() { + static int sivar; +#ifdef LAMBDA + // LAMBDA: [[RED_VAR:@.+]] = common global [8 x {{.+}}] zeroinitializer + + // LAMBDA-LABEL: @main + // LAMBDA: call void [[OUTER_LAMBDA:@.+]]( + [&]() { + // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]]( + // LAMBDA: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 1, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i32 0, i32 0) + // LAMBDA: call void @[[LOFFL1:.+]]( + // LAMBDA: ret +#pragma omp target teams distribute parallel for reduction(+: sivar) + for (int i = 0; i < 2; ++i) { + // LAMBDA: define{{.*}} internal{{.*}} void @[[LOFFL1]](i{{64|32}}*{{.*}} [[SIVAR_ARG:%.+]]) + // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{.+}}, + // LAMBDA: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]], + // LAMBDA: [[SIVAR_PAR:%.+]] = load{{.+}}, {{.+}} [[SIVAR_ADDR]], + // LAMBDA: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[LOUTL1:.+]] to {{.+}}, {{.+}} [[SIVAR_PAR]]) + // LAMBDA: ret void + + // LAMBDA: define internal void @[[LOUTL1]]({{.+}}, {{.+}}, {{.+}} [[SIVAR_ARG:%.+]]) + // Skip global and bound tid vars + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{.+}}, + // LAMBDA: [[SIVAR_PRIV:%.+]] = alloca i{{.+}}, + // LAMBDA: [[RED_LIST:%.+]] = alloca [1 x {{.+}}], + // LAMBDA: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]], + // LAMBDA: [[SIVAR_REF:%.+]] = load{{.+}}, {{.+}} [[SIVAR_ADDR]] + // LAMBDA: store{{.+}} 0, {{.+}} [[SIVAR_PRIV]], + + // LAMBDA: call void @__kmpc_for_static_init_4( + // LAMBDA: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[LPAR_OUTL:.+]] to + // LAMBDA: call void @__kmpc_for_static_fini( + // LAMBDA: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]], + // LAMBDA: [[SIVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[SIVAR_PRIV]] to + // LAMBDA: store{{.+}} [[SIVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]], + // LAMBDA: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to + // LAMBDA: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce_nowait({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]]) + // LAMBDA: switch{{.+}} [[K_RED_RET]], label{{.+}} [ + // LAMBDA: {{.+}}, label %[[CASE1:.+]] + // LAMBDA: {{.+}}, label %[[CASE2:.+]] + // LAMBDA: ] + // LAMBDA: [[CASE1]]: + // LAMBDA-DAG: [[SIVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_REF]], + // LAMBDA-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], + // LAMBDA-DAG: [[SIVAR_INC:%.+]] = add{{.+}} [[SIVAR_VAL]], [[SIVAR_PRIV_VAL]] + // LAMBDA: store{{.+}} [[SIVAR_INC]], {{.+}} [[SIVAR_REF]], + // LAMBDA: call void @__kmpc_end_reduce_nowait({{.+}}, {{.+}}, {{.+}} [[RED_VAR]]) + // LAMBDA: br + // LAMBDA: [[CASE2]]: + // LAMBDA-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], + // LAMBDA-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[SIVAR_REF]], {{.+}} [[SIVAR_PRIV_VAL]] + // LAMBDA: br + + // LAMBDA: define internal void @[[LPAR_OUTL]]({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[SIVAR_ARG:%.+]]) + + // Skip global and bound tid vars, and prev lb and ub vars + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: alloca i{{[0-9]+}}, + // LAMBDA: alloca i{{[0-9]+}}, + // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{.+}}, + // skip loop vars + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: [[SIVAR_PRIV:%.+]] = alloca i{{.+}}, + // LAMBDA: [[RED_LIST:%.+]] = alloca [1 x {{.+}}], + // LAMBDA: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]], + // LAMBDA: [[SIVAR_REF:%.+]] = load {{.+}} [[SIVAR_ADDR]] + // LAMBDA: store{{.+}} 0, {{.+}} [[SIVAR_PRIV]], + + // LAMBDA: call void @__kmpc_for_static_init_4( + // LAMBDA: store{{.+}}, {{.+}} [[SIVAR_PRIV]], + // LAMBDA: call void [[INNER_LAMBDA:@.+]]( + // LAMBDA: call void @__kmpc_for_static_fini( + // LAMBDA: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]], + // LAMBDA: [[SIVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[SIVAR_PRIV]] to + // LAMBDA: store{{.+}} [[SIVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]], + // LAMBDA: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to + // LAMBDA: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce_nowait({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]]) + // LAMBDA: switch{{.+}} [[K_RED_RET]], label{{.+}} [ + // LAMBDA: {{.+}}, label %[[CASE1:.+]] + // LAMBDA: {{.+}}, label %[[CASE2:.+]] + // LAMBDA: ] + // LAMBDA: [[CASE1]]: + // LAMBDA-64-DAG: [[SIVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_REF]], + // LAMBDA-32-DAG: [[SIVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_ADDR]], + // LAMBDA-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], + // LAMBDA-DAG: [[SIVAR_INC:%.+]] = add{{.+}} [[SIVAR_VAL]], [[SIVAR_PRIV_VAL]] + // LAMBDA: store{{.+}} [[SIVAR_INC]], {{.+}} [[SIVAR_REF]], + // LAMBDA: call void @__kmpc_end_reduce_nowait({{.+}}, {{.+}}, {{.+}} [[RED_VAR]]) + // LAMBDA: br + // LAMBDA: [[CASE2]]: + // LAMBDA-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], + // LAMBDA-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[SIVAR_REF]], {{.+}} [[SIVAR_PRIV_VAL]] + // LAMBDA: br + + sivar += i; + + [&]() { + // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) + // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], + + sivar += 4; + // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]] + + // LAMBDA: [[SIVAR_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 + // LAMBDA: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_PTR_REF]] + // LAMBDA: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_REF]] + // LAMBDA: [[SIVAR_INC:%.+]] = add{{.+}} [[SIVAR_VAL]], 4 + // LAMBDA: store i{{[0-9]+}} [[SIVAR_INC]], i{{[0-9]+}}* [[SIVAR_REF]] + }(); + } + }(); + return 0; +#else +#pragma omp target teams distribute parallel for reduction(+: sivar) + for (int i = 0; i < 2; ++i) { + sivar += i; + } + return tmain<int>(); +#endif +} + +// CHECK: [[RED_VAR:@.+]] = common global [8 x {{.+}}] zeroinitializer + +// CHECK: define {{.*}}i{{[0-9]+}} @main() +// CHECK: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 1, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i32 0, i32 0) +// CHECK: call void @[[OFFL1:.+]](i{{64|32}}* @{{.+}}) +// CHECK: {{%.+}} = call{{.*}} i32 @[[TMAIN_INT:.+]]() +// CHECK: ret + +// CHECK: define{{.*}} void @[[OFFL1]](i{{64|32}}*{{.*}} [[SIVAR_ARG:%.+]]) +// CHECK: [[SIVAR_ADDR:%.+]] = alloca i{{.+}}, +// CHECK: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]], +// CHECK: [[SIVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_ADDR]] +// CHECK: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL1:.+]] to {{.+}}, {{.+}} [[SIVAR_VAL]]) +// CHECK: ret void + +// CHECK: define internal void @[[OUTL1]]({{.+}}, {{.+}}, {{.+}} [[SIVAR_ARG:%.+]]) +// Skip global and bound tid vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: [[SIVAR_ADDR:%.+]] = alloca i{{.+}}, +// CHECK: [[SIVAR_PRIV:%.+]] = alloca i{{.+}}, +// CHECK: [[RED_LIST:%.+]] = alloca [1 x {{.+}}], +// CHECK: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]], +// CHECK: [[SIVAR_REF:%.+]] = load{{.+}} [[SIVAR_ADDR]], +// CHECK: store{{.+}} 0, {{.+}} [[SIVAR_PRIV]], + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL:.+]] to +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]], +// CHECK: [[SIVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[SIVAR_PRIV]] to +// CHECK: store{{.+}} [[SIVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]], +// CHECK: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to +// CHECK: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce_nowait({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]]) +// CHECK: switch{{.+}} [[K_RED_RET]], label{{.+}} [ +// CHECK: {{.+}}, label %[[CASE1:.+]] +// CHECK: {{.+}}, label %[[CASE2:.+]] +// CHECK: ] +// CHECK: [[CASE1]]: +// CHECK-DAG: [[SIVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_REF]], +// CHECK-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], +// CHECK-DAG: [[SIVAR_INC:%.+]] = add{{.+}} [[SIVAR_VAL]], [[SIVAR_PRIV_VAL]] +// CHECK: store{{.+}} [[SIVAR_INC]], {{.+}} [[SIVAR_REF]], +// CHECK: call void @__kmpc_end_reduce_nowait({{.+}}, {{.+}}, {{.+}} [[RED_VAR]]) +// CHECK: br +// CHECK: [[CASE2]]: +// CHECK-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], +// CHECK-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[SIVAR_REF]], {{.+}} [[SIVAR_PRIV_VAL]] +// CHECK: br + +// CHECK: define internal void @[[PAR_OUTL]]({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[SIVAR_ARG:%.+]]) +// Skip global and bound tid vars, and prev lb and ub +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: [[SIVAR_ADDR:%.+]] = alloca i{{.+}}, +// skip loop vars +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: [[SIVAR_PRIV:%.+]] = alloca i{{.+}}, +// CHECK: [[RED_LIST:%.+]] = alloca [1 x {{.+}}], +// CHECK: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]], +// CHECK-64: [[SIVAR_REF:%.+]] = load {{.+}} [[SIVAR_ADDR]], +// CHECK: store{{.+}} 0, {{.+}} [[SIVAR_PRIV]], + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: store{{.+}}, {{.+}} [[SIVAR_PRIV]], +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]], +// CHECK: [[SIVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[SIVAR_PRIV]] to +// CHECK: store{{.+}} [[SIVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]], +// CHECK: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to +// CHECK: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce_nowait({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]]) +// CHECK: switch{{.+}} [[K_RED_RET]], label{{.+}} [ +// CHECK: {{.+}}, label %[[CASE1:.+]] +// CHECK: {{.+}}, label %[[CASE2:.+]] +// CHECK: ] +// CHECK: [[CASE1]]: +// CHECK-DAG: [[SIVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_REF]], +// CHECK-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], +// CHECK-DAG: [[SIVAR_INC:%.+]] = add{{.+}} [[SIVAR_VAL]], [[SIVAR_PRIV_VAL]] +// CHECK: store{{.+}} [[SIVAR_INC]], {{.+}} [[SIVAR_REF]], +// CHECK: call void @__kmpc_end_reduce_nowait({{.+}}, {{.+}}, {{.+}} [[RED_VAR]]) +// CHECK: br +// CHECK: [[CASE2]]: +// CHECK-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], +// CHECK-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[SIVAR_REF]], {{.+}} [[SIVAR_PRIV_VAL]] +// CHECK: br + +// CHECK: define{{.*}} i{{[0-9]+}} @[[TMAIN_INT]]() +// CHECK: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 1, +// CHECK: call void @[[TOFFL1:.+]]({{.+}}) +// CHECK: ret + +// CHECK: define{{.*}} void @[[TOFFL1]](i{{64|32}}*{{.*}} [[TVAR_ARG:%.+]]) +// CHECK: [[TVAR_ADDR:%.+]] = alloca i{{.+}}, +// CHECK: store{{.+}} [[TVAR_ARG]], {{.+}} [[TVAR_ADDR]], +// CHECK: [[TVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_ADDR]] +// CHECK: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[TOUTL1:.+]] to {{.+}}, {{.+}} [[TVAR_VAL]]) +// CHECK: ret void + +// CHECK: define internal void @[[TOUTL1]]({{.+}}, {{.+}}, {{.+}} [[TVAR_ARG:%.+]]) +// Skip global and bound tid vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: [[TVAR_ADDR:%.+]] = alloca i{{.+}}, +// CHECK: [[TVAR_PRIV:%.+]] = alloca i{{.+}}, +// CHECK: [[RED_LIST:%.+]] = alloca [1 x {{.+}}], +// CHECK: store{{.+}} [[TVAR_ARG]], {{.+}} [[TVAR_ADDR]], +// CHECK: [[TVAR_REF:%.+]] = load {{.+}} [[TVAR_ADDR]], +// CHECK: store{{.+}} 0, {{.+}} [[TVAR_PRIV]], + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[TPAR_OUTL:.+]] to +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]], +// CHECK: [[TVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[TVAR_PRIV]] to +// CHECK: store{{.+}} [[TVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]], +// CHECK: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to +// CHECK: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce_nowait({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]]) +// CHECK: switch{{.+}} [[K_RED_RET]], label{{.+}} [ +// CHECK: {{.+}}, label %[[CASE1:.+]] +// CHECK: {{.+}}, label %[[CASE2:.+]] +// CHECK: ] +// CHECK: [[CASE1]]: +// CHECK-DAG: [[TVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_REF]], +// CHECK-DAG: [[TVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_PRIV]], +// CHECK-DAG: [[TVAR_INC:%.+]] = add{{.+}} [[TVAR_VAL]], [[TVAR_PRIV_VAL]] +// CHECK: store{{.+}} [[TVAR_INC]], {{.+}} [[TVAR_REF]], +// CHECK: call void @__kmpc_end_reduce_nowait({{.+}}, {{.+}}, {{.+}} [[RED_VAR]]) +// CHECK: br +// CHECK: [[CASE2]]: +// CHECK-DAG: [[TVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_PRIV]], +// CHECK-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[TVAR_REF]], {{.+}} [[TVAR_PRIV_VAL]] +// CHECK: br + +// CHECK: define internal void @[[TPAR_OUTL]]({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[TVAR_ARG:%.+]]) +// Skip global and bound tid vars, and prev lb and ub vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: [[TVAR_ADDR:%.+]] = alloca i{{.+}}, +// skip loop vars +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: [[TVAR_PRIV:%.+]] = alloca i{{.+}}, +// CHECK: [[RED_LIST:%.+]] = alloca [1 x {{.+}}], +// CHECK: store{{.+}} [[TVAR_ARG]], {{.+}} [[TVAR_ADDR]], +// CHECK: [[TVAR_REF:%.+]] = load {{.+}} [[TVAR_ADDR]], +// CHECK: store{{.+}} 0, {{.+}} [[TVAR_PRIV]], + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: store{{.+}}, {{.+}} [[TVAR_PRIV]], +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]], +// CHECK: [[TVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[TVAR_PRIV]] to +// CHECK: store{{.+}} [[TVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]], +// CHECK: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to +// CHECK: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce_nowait({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]]) +// CHECK: switch{{.+}} [[K_RED_RET]], label{{.+}} [ +// CHECK: {{.+}}, label %[[CASE1:.+]] +// CHECK: {{.+}}, label %[[CASE2:.+]] +// CHECK: ] +// CHECK: [[CASE1]]: +// CHECK-DAG: [[TVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_REF]], +// CHECK-DAG: [[TVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_PRIV]], +// CHECK-DAG: [[TVAR_INC:%.+]] = add{{.+}} [[TVAR_VAL]], [[TVAR_PRIV_VAL]] +// CHECK: store{{.+}} [[TVAR_INC]], {{.+}} [[TVAR_REF]], +// CHECK: call void @__kmpc_end_reduce_nowait({{.+}}, {{.+}}, {{.+}} [[RED_VAR]]) +// CHECK: br +// CHECK: [[CASE2]]: +// CHECK-DAG: [[TVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_PRIV]], +// CHECK-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[TVAR_REF]], {{.+}} [[TVAR_PRIV_VAL]] +// CHECK: br +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_reduction_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_reduction_messages.cpp index 59b35417d401e..53c41b6bb0623 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_reduction_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_schedule_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_schedule_codegen.cpp new file mode 100644 index 0000000000000..8eb6c443ca2d8 --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_schedule_codegen.cpp @@ -0,0 +1,400 @@ +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// Test host codegen. +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +#ifdef CK1 + +template <typename T, int X, long long Y> +struct SS{ + T a[X]; + float b; + // CK1: define {{.*}}i32 @{{.+}}foo{{.+}}( + int foo(void) { + + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL1:.+]]( + #pragma omp target teams distribute parallel for + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL2:.+]]( + #pragma omp target teams distribute parallel for schedule(static) + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL3:.+]]( + #pragma omp target teams distribute parallel for schedule(static, X/2) + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL4:.+]]( + #pragma omp target teams distribute parallel for schedule(dynamic) + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL5:.+]]( + #pragma omp target teams distribute parallel for schedule(dynamic, X/2) + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + + // CK1: define internal void @[[OFFL1]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL1:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL1]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL1]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[OFFL2]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL2:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL2]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL2:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL2]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[OFFL3]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL3:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL3]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL3:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL3]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 33, + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[OFFL4]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL4:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL4]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL4:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL4]]({{.+}}) + // CK1: call void @__kmpc_dispatch_init_4({{.+}}, {{.+}}, i32 35, + // CK1: call {{.+}} @__kmpc_dispatch_next_4( + // CK1: ret void + + // CK1: define internal void @[[OFFL5]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL5:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL5]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL5:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL5]]({{.+}}) + // CK1: call void @__kmpc_dispatch_init_4({{.+}}, {{.+}}, i32 35, + // CK1: call {{.+}} @__kmpc_dispatch_next_4( + // CK1: ret void + + return a[0]; + } +}; + +int teams_template_struct(void) { + SS<int, 123, 456> V; + return V.foo(); + +} +#endif // CK1 + +// Test host codegen. +// RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64 +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64 +// RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} +#ifdef CK2 + +template <typename T, int n> +int tmain(T argc) { + T a[n]; + int m = 10; +#pragma omp target teams distribute parallel for + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } +#pragma omp target teams distribute parallel for schedule(static) + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } +#pragma omp target teams distribute parallel for schedule(static, m) + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } +#pragma omp target teams distribute parallel for schedule(dynamic) + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } +#pragma omp target teams distribute parallel for schedule(dynamic, m) + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } + return 0; +} + +int main (int argc, char **argv) { + int n = 100; + int a[n]; + int m = 10; +#pragma omp target teams distribute parallel for + for(int i = 0; i < n; i++) { + a[i] = 0; + } +#pragma omp target teams distribute parallel for dist_schedule(static) + for(int i = 0; i < n; i++) { + a[i] = 0; + } +#pragma omp target teams distribute parallel for dist_schedule(static, m) + for(int i = 0; i < n; i++) { + a[i] = 0; + } +#pragma omp target teams distribute parallel for schedule(dynamic) + for(int i = 0; i < n; i++) { + a[i] = 0; + } +#pragma omp target teams distribute parallel for schedule(dynamic, m) + for(int i = 0; i < n; i++) { + a[i] = 0; + } + return tmain<int, 10>(argc); +} + +// CK2: define {{.*}}i32 @{{[^,]+}}(i{{.+}}{{.+}} %[[ARGC:.+]], {{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL1:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL2:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL3:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL4:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL5:.+]]({{.+}}) +// CK2: {{%.+}} = call{{.*}} i32 @[[TMAIN:.+]]({{.+}}) +// CK2: ret + +// CK2: define {{.*}}void @[[OFFL1]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[OUTL1:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFL2]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[OUTL2:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL2:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFL3]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 4, {{.+}} @[[OUTL3:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL3:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34 +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFL4]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[OUTL4:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL4]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL4:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL4]]({{.+}}) +// CK2: call void @__kmpc_dispatch_init_4({{.+}}, {{.+}}, i32 35, +// CK2: call {{.+}} @__kmpc_dispatch_next_4( +// CK2: ret void + + +// CK2: define {{.*}}void @[[OFFL5]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[OUTL5:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL5]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL5:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL5]]({{.+}}) +// CK2: call void @__kmpc_dispatch_init_4({{.+}}, {{.+}}, i32 35, +// CK2: call {{.+}} @__kmpc_dispatch_next_4( +// CK2: ret void + +// CK2: define {{.*}}i32 @[[TMAIN]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT1:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT2:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT3:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT4:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT5:.+]]({{.+}}) +// CK2: ret +// CK2-NEXT: } + +// CK2: define {{.*}}void @[[OFFLT1]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTLT1:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT1:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFLT2]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTLT2:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT2:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFLT3]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[OUTLT3:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT3:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 33, +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFLT4]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[OUTLT4:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT4]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT4:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT4]]({{.+}}) +// CK2: call void @__kmpc_dispatch_init_4({{.+}}, {{.+}}, i32 35, +// CK2: call {{.+}} @__kmpc_dispatch_next_4( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFLT5]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[OUTLT5:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT5]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT5:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT5]]({{.+}}) +// CK2: call void @__kmpc_dispatch_init_4({{.+}}, {{.+}}, i32 35, +// CK2: call {{.+}} @__kmpc_dispatch_next_4( +// CK2: ret void + +#endif // CK2 +#endif // #ifndef HEADER diff --git a/test/OpenMP/target_teams_distribute_parallel_for_schedule_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_schedule_messages.cpp index 54e302010992d..1f4f5f12e87b3 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_schedule_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_shared_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_shared_messages.cpp index 4f90000796302..f90a533edfee6 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_shared_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_shared_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_aligned_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_aligned_messages.cpp index f121512733aff..18685d2b30ed0 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_aligned_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_aligned_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp %s +// RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp-simd %s + struct B { static int ib[20]; // expected-note 0 {{'B::ib' declared here}} static constexpr int bfoo() { return 8; } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_ast_print.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_ast_print.cpp index af588f1b81d9f..5de2d4b195dd8 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_ast_print.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_ast_print.cpp @@ -1,6 +1,10 @@ -// RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp -ast-print %s -Wno-openmp-target | FileCheck %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print -Wno-openmp-target | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s -Wno-openmp-target | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print -Wno-openmp-target | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -109,7 +113,7 @@ T tmain(T argc) { #pragma omp target teams distribute parallel for simd for (int i=0; i < 2; ++i) a = 2; -// CHECK: #pragma omp target teams distribute parallel for simd +// CHECK: #pragma omp target teams distribute parallel for simd{{$}} // CHECK-NEXT: for (int i = 0; i < 2; ++i) // CHECK-NEXT: a = 2; #pragma omp target teams distribute parallel for simd private(argc, b), firstprivate(c, d), collapse(2) diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen.cpp new file mode 100644 index 0000000000000..0329d7e2789ec --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen.cpp @@ -0,0 +1,131 @@ +// expected-no-diagnostics +#ifndef HEADER +#define HEADER +// Test host codegen. +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 --check-prefix HCK1 --check-prefix HCK1-64 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 --check-prefix HCK1 --check-prefix HCK1-64 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 --check-prefix HCK1 --check-prefix HCK1-64 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 --check-prefix HCK1 --check-prefix HCK1-64 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY1 +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY1 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY1 +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY1 + +// Test target codegen - host bc file has to be created first. (no significant differences with host version of target region) +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 --check-prefix TCK1 --check-prefix TCK1-64 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 --check-prefix TCK1 --check-prefix TCK1-64 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 --check-prefix TCK1 --check-prefix TCK1-32 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 --check-prefix TCK1 --check-prefix TCK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix SIMD-ONLY1 +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY1 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix SIMD-ONLY1 +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY1 +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +#ifdef CK1 + +// HCK1: define{{.*}} i32 @{{.+}}target_teams_fun{{.*}}( +int target_teams_fun(int *g){ + int n = 1000; + int a[1000]; + int te = n / 128; + int th = 128; +// discard n_addr +// HCK1: alloca i32, +// HCK1: [[TE:%.+]] alloca i32, +// HCK1: [[TH:%.+]] = alloca i32, +// HCK1: [[I:%.+]] = alloca i32, +// discard capture expressions for te and th +// HCK1: = alloca i32, +// HCK1: = alloca i32, +// HCK1: [[I_CAST:%.+]] = alloca i{{32|64}}, +// HCK1: [[N_CAST:%.+]] = alloca i{{32|64}}, +// HCK1: [[TE_CAST:%.+]] = alloca i{{32|64}}, +// HCK1: [[TH_CAST:%.+]] = alloca i{{32|64}}, +// HCK1: [[I_PAR:%.+]] = load{{.+}}, {{.+}} [[I_CAST]], +// HCK1: [[N_PAR:%.+]] = load{{.+}}, {{.+}} [[N_CAST]], +// HCK1: [[TE_PAR:%.+]] = load{{.+}}, {{.+}} [[TE_CAST]], +// HCK1: [[TH_PAR:%.+]] = load{{.+}}, {{.+}} [[TH_CAST]], +// HCK1: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 5, i8** %{{[^,]+}}, i8** %{{[^,]+}}, + +// HCK1: call void @[[OFFL1:.+]](i{{32|64}} [[I_PAR]], i{{32|64}} [[N_PAR]], {{.+}}, i{{32|64}} [[TE_PAR]], i{{32|64}} [[TH_PAR]]) + int i; +#pragma omp target teams distribute parallel for simd num_teams(te), thread_limit(th) aligned(a : 8) safelen(16) simdlen(4) linear(i : n) + for(i = 0; i < n; i++) { + a[i] = 0; + } + + // HCK1: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 3, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), + // HCK1: call void @[[OFFL2:.+]](i{{64|32}} %{{.+}}) + {{{ + #pragma omp target teams distribute parallel for simd is_device_ptr(g) simdlen(8) + for(int i = 0; i < n; i++) { + a[i] = g[0]; + } + }}} + + // outlined target regions + // HCK1: define internal void @[[OFFL1]](i{{32|64}} [[I_ARG:%.+]], i{{32|64}} [[N_ARG:%.+]], {{.+}}, i{{32|64}} [[TE_ARG:%.+]], i{{32|64}} [[TH_ARG:%.+]]) + // TCK1: define weak void @{{.+}}target_teams_fun{{.*}}(i{{32|64}} [[I_ARG:%.+]], i{{32|64}} [[N_ARG:%.+]], {{.+}}, i{{32|64}} [[TE_ARG:%.+]], i{{32|64}} [[TH_ARG:%.+]]) + // CK1: [[I_ADDR:%.+]] = alloca i{{32|64}}, + // CK1: [[N_ADDR:%.+]] = alloca i{{32|64}}, + // CK1: [[TE_ADDR:%.+]] = alloca i{{32|64}}, + // CK1: [[TH_ADDR:%.+]] = alloca i{{32|64}}, + // TCK1: store {{.+}} [[N_ARG]], {{.+}} [[N_ADDR]], + // CK1: store{{.+}} [[TE_ARG]], {{.+}} [[TE_ADDR]], + // CK1: store{{.+}} [[TH_ARG]], {{.+}} [[TH_ADDR]], + // CK1-64: [[TE_CONV:%.+]] = bitcast{{.+}} [[TE_ADDR]] to + // CK1-64: [[TH_CONV:%.+]] = bitcast{{.+}} [[TH_ADDR]] to + // CK1-64: [[TE_VAL:%.+]] = load i32, i32* [[TE_CONV]], + // CK1-64: [[TH_VAL:%.+]] = load i32, i32* [[TH_CONV]], + // CK1-32: [[TE_VAL:%.+]] = load i32, i32* [[TE_ADDR]], + // CK1-32: [[TH_VAL:%.+]] = load i32, i32* [[TH_ADDR]], + // CK1: {{%.+}} = call i32 @__kmpc_push_num_teams({{.+}}, {{.+}}, i32 [[TE_VAL]], i32 [[TH_VAL]]) + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[OUTL1:.+]] to {{.+}}, {{.+}}, {{.+}}) + // CK1: ret void + + // CK1: define internal void @[[OUTL1]]({{.+}}) + // CK1: [[ARRDECAY:%.+]] = getelementptr inbounds [1000 x i32], [1000 x i32]* %{{.+}}, i32 0, i32 0 + // CK1: [[ARR_CAST:%.+]] = ptrtoint i32* [[ARRDECAY]] to i{{32|64}} + // CK1: [[MASKED_PTR:%.+]] = and i{{32|64}} [[ARR_CAST]], 7 + // CK1: [[COND:%.+]] = icmp eq i{{32|64}} [[MASKED_PTR]], 0 + // CK1: call void @llvm.assume(i1 [[COND]]) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void {{.+}} @__kmpc_fork_call( + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // HCK1: define internal void @[[OFFL2]]( + // TCK1: define weak void @{{.+}}target_teams_fun{{.+}}( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[OUTL2:.+]] to {{.+}}, {{.+}}, {{.+}}) + // CK1: ret void + + // CK1: define internal void @[[OUTL2]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void {{.+}} @__kmpc_fork_call( + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + return a[0]; +} + +// CK1-DAG: !{!"llvm.loop.vectorize.width", i32 4} +// CK1-DAG: !{!"llvm.loop.vectorize.enable", i1 true} +// CK1-DAG: !{!"llvm.loop.vectorize.width", i32 8} + +#endif // CK1 +#endif // HEADER diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen_registration.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen_registration.cpp new file mode 100644 index 0000000000000..76a7b4d0bd868 --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen_registration.cpp @@ -0,0 +1,472 @@ +// Test host codegen. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// Test target teams distribute parallel for simd codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s -check-prefix=TCHECK +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// Check that no target code is emitted if no omptests flag was provided. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK-NTARGET + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// CHECK-DAG: [[SA:%.+]] = type { [4 x i32] } +// CHECK-DAG: [[SB:%.+]] = type { [8 x i32] } +// CHECK-DAG: [[SC:%.+]] = type { [16 x i32] } +// CHECK-DAG: [[SD:%.+]] = type { [32 x i32] } +// CHECK-DAG: [[SE:%.+]] = type { [64 x i32] } +// CHECK-DAG: [[ST1:%.+]] = type { [228 x i32] } +// CHECK-DAG: [[ST2:%.+]] = type { [1128 x i32] } +// CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } +// CHECK-DAG: [[DEVTY:%.+]] = type { i8*, i8*, [[ENTTY]]*, [[ENTTY]]* } +// CHECK-DAG: [[DSCTY:%.+]] = type { i32, [[DEVTY]]*, [[ENTTY]]*, [[ENTTY]]* } + +// TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } + +// CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat + +// CHECK-DAG: [[A1:@.+]] = internal global [[SA]] +// CHECK-DAG: [[A2:@.+]] = global [[SA]] +// CHECK-DAG: [[B1:@.+]] = global [[SB]] +// CHECK-DAG: [[B2:@.+]] = global [[SB]] +// CHECK-DAG: [[C1:@.+]] = internal global [[SC]] +// CHECK-DAG: [[D1:@.+]] = global [[SD]] +// CHECK-DAG: [[E1:@.+]] = global [[SE]] +// CHECK-DAG: [[T1:@.+]] = global [[ST1]] +// CHECK-DAG: [[T2:@.+]] = global [[ST2]] + +// CHECK-NTARGET-DAG: [[SA:%.+]] = type { [4 x i32] } +// CHECK-NTARGET-DAG: [[SB:%.+]] = type { [8 x i32] } +// CHECK-NTARGET-DAG: [[SC:%.+]] = type { [16 x i32] } +// CHECK-NTARGET-DAG: [[SD:%.+]] = type { [32 x i32] } +// CHECK-NTARGET-DAG: [[SE:%.+]] = type { [64 x i32] } +// CHECK-NTARGET-DAG: [[ST1:%.+]] = type { [228 x i32] } +// CHECK-NTARGET-DAG: [[ST2:%.+]] = type { [1128 x i32] } +// CHECK-NTARGET-NOT: type { i8*, i8*, % +// CHECK-NTARGET-NOT: type { i32, % + +// We have 7 target regions + +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// TCHECK-NOT: {{@.+}} = weak constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] + +// CHECK-NTARGET-NOT: weak constant i8 0 +// CHECK-NTARGET-NOT: private unnamed_addr constant [1 x i + +// CHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" +// CHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" +// CHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" +// CHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" +// CHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" +// CHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" +// CHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" +// CHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" +// CHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" +// CHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" +// CHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" +// CHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" +// CHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 + +// TCHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" +// TCHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" +// TCHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" +// TCHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" +// TCHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" +// TCHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" +// TCHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" +// TCHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" +// TCHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" +// TCHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" +// TCHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" +// TCHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" +// TCHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 + +// CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] +// CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] +// CHECK: [[DEVBEGIN:@.+]] = external constant i8 +// CHECK: [[DEVEND:@.+]] = external constant i8 +// CHECK: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x [[DEVTY]]] [{{.+}} { i8* [[DEVBEGIN]], i8* [[DEVEND]], [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }], comdat($[[REGFN]]) +// CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) + +// We have 4 initializers, one for the 500 priority, another one for 501, or more for the default priority, and the last one for the offloading registration function. +// CHECK: @llvm.global_ctors = appending global [4 x { i32, void ()*, i8* }] [ +// CHECK-SAME: { i32, void ()*, i8* } { i32 500, void ()* [[P500:@[^,]+]], i8* null }, +// CHECK-SAME: { i32, void ()*, i8* } { i32 501, void ()* [[P501:@[^,]+]], i8* null }, +// CHECK-SAME: { i32, void ()*, i8* } { i32 65535, void ()* [[PMAX:@[^,]+]], i8* null }, +// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] + +// CHECK-NTARGET: @llvm.global_ctors = appending global [3 x { i32, void ()*, i8* }] [ + +extern int *R; + +struct SA { + int arr[4]; + void foo() { + int a = *R; + a += 1; + *R = a; + } + SA() { + int a = *R; + a += 2; + *R = a; + } + ~SA() { + int a = *R; + a += 3; + *R = a; + } +}; + +struct SB { + int arr[8]; + void foo() { + int a = *R; + #pragma omp target teams distribute parallel for simd + for (int i = 0; i < 10; ++i) + a += 4; + *R = a; + } + SB() { + int a = *R; + a += 5; + *R = a; + } + ~SB() { + int a = *R; + a += 6; + *R = a; + } +}; + +struct SC { + int arr[16]; + void foo() { + int a = *R; + a += 7; + *R = a; + } + SC() { + int a = *R; + #pragma omp target teams distribute parallel for simd + for (int i = 0; i < 10; ++i) + a += 8; + *R = a; + } + ~SC() { + int a = *R; + a += 9; + *R = a; + } +}; + +struct SD { + int arr[32]; + void foo() { + int a = *R; + a += 10; + *R = a; + } + SD() { + int a = *R; + a += 11; + *R = a; + } + ~SD() { + int a = *R; + #pragma omp target teams distribute parallel for simd + for (int i = 0; i < 10; ++i) + a += 12; + *R = a; + } +}; + +struct SE { + int arr[64]; + void foo() { + int a = *R; + #pragma omp target teams distribute parallel for simd if(target: 0) + for (int i = 0; i < 10; ++i) + a += 13; + *R = a; + } + SE() { + int a = *R; + #pragma omp target teams distribute parallel for simd + for (int i = 0; i < 10; ++i) + a += 14; + *R = a; + } + ~SE() { + int a = *R; + #pragma omp target teams distribute parallel for simd + for (int i = 0; i < 10; ++i) + a += 15; + *R = a; + } +}; + +template <int x> +struct ST { + int arr[128 + x]; + void foo() { + int a = *R; + #pragma omp target teams distribute parallel for simd + for (int i = 0; i < 10; ++i) + a += 16 + x; + *R = a; + } + ST() { + int a = *R; + #pragma omp target teams distribute parallel for simd + for (int i = 0; i < 10; ++i) + a += 17 + x; + *R = a; + } + ~ST() { + int a = *R; + #pragma omp target teams distribute parallel for simd + for (int i = 0; i < 10; ++i) + a += 18 + x; + *R = a; + } +}; + +// We have to make sure we us all the target regions: +//CHECK-DAG: define internal void @[[NAME1]]( +//CHECK-DAG: call void @[[NAME1]]( +//CHECK-DAG: define internal void @[[NAME2]]( +//CHECK-DAG: call void @[[NAME2]]( +//CHECK-DAG: define internal void @[[NAME3]]( +//CHECK-DAG: call void @[[NAME3]]( +//CHECK-DAG: define internal void @[[NAME4]]( +//CHECK-DAG: call void @[[NAME4]]( +//CHECK-DAG: define internal void @[[NAME5]]( +//CHECK-DAG: call void @[[NAME5]]( +//CHECK-DAG: define internal void @[[NAME6]]( +//CHECK-DAG: call void @[[NAME6]]( +//CHECK-DAG: define internal void @[[NAME7]]( +//CHECK-DAG: call void @[[NAME7]]( +//CHECK-DAG: define internal void @[[NAME8]]( +//CHECK-DAG: call void @[[NAME8]]( +//CHECK-DAG: define internal void @[[NAME9]]( +//CHECK-DAG: call void @[[NAME9]]( +//CHECK-DAG: define internal void @[[NAME10]]( +//CHECK-DAG: call void @[[NAME10]]( +//CHECK-DAG: define internal void @[[NAME11]]( +//CHECK-DAG: call void @[[NAME11]]( +//CHECK-DAG: define internal void @[[NAME12]]( +//CHECK-DAG: call void @[[NAME12]]( + +//TCHECK-DAG: define weak void @[[NAME1]]( +//TCHECK-DAG: define weak void @[[NAME2]]( +//TCHECK-DAG: define weak void @[[NAME3]]( +//TCHECK-DAG: define weak void @[[NAME4]]( +//TCHECK-DAG: define weak void @[[NAME5]]( +//TCHECK-DAG: define weak void @[[NAME6]]( +//TCHECK-DAG: define weak void @[[NAME7]]( +//TCHECK-DAG: define weak void @[[NAME8]]( +//TCHECK-DAG: define weak void @[[NAME9]]( +//TCHECK-DAG: define weak void @[[NAME10]]( +//TCHECK-DAG: define weak void @[[NAME11]]( +//TCHECK-DAG: define weak void @[[NAME12]]( + +// CHECK-NTARGET-NOT: __tgt_target +// CHECK-NTARGET-NOT: __tgt_register_lib +// CHECK-NTARGET-NOT: __tgt_unregister_lib + +// TCHECK-NOT: __tgt_target +// TCHECK-NOT: __tgt_register_lib +// TCHECK-NOT: __tgt_unregister_lib + +// We have 2 initializers with priority 500 +//CHECK: define internal void [[P500]]( +//CHECK: call void @{{.+}}() +//CHECK: call void @{{.+}}() +//CHECK-NOT: call void @{{.+}}() +//CHECK: ret void + +// We have 1 initializers with priority 501 +//CHECK: define internal void [[P501]]( +//CHECK: call void @{{.+}}() +//CHECK-NOT: call void @{{.+}}() +//CHECK: ret void + +// We have 6 initializers with default priority +//CHECK: define internal void [[PMAX]]( +//CHECK: call void @{{.+}}() +//CHECK: call void @{{.+}}() +//CHECK: call void @{{.+}}() +//CHECK: call void @{{.+}}() +//CHECK: call void @{{.+}}() +//CHECK: call void @{{.+}}() +//CHECK-NOT: call void @{{.+}}() +//CHECK: ret void + +// Check registration and unregistration + +//CHECK: define internal void @[[UNREGFN:.+]](i8*) +//CHECK-SAME: comdat($[[REGFN]]) { +//CHECK: call i32 @__tgt_unregister_lib([[DSCTY]]* [[DESC]]) +//CHECK: ret void +//CHECK: declare i32 @__tgt_unregister_lib([[DSCTY]]*) + +//CHECK: define linkonce hidden void @[[REGFN]]() +//CHECK-SAME: comdat { +//CHECK: call i32 @__tgt_register_lib([[DSCTY]]* [[DESC]]) +//CHECK: call i32 @__cxa_atexit(void (i8*)* @[[UNREGFN]], i8* bitcast ([[DSCTY]]* [[DESC]] to i8*), +//CHECK: ret void +//CHECK: declare i32 @__tgt_register_lib([[DSCTY]]*) + +static __attribute__((init_priority(500))) SA a1; +SA a2; +SB __attribute__((init_priority(500))) b1; +SB __attribute__((init_priority(501))) b2; +static SC c1; +SD d1; +SE e1; +ST<100> t1; +ST<1000> t2; + + +int bar(int a){ + int r = a; + + a1.foo(); + a2.foo(); + b1.foo(); + b2.foo(); + c1.foo(); + d1.foo(); + e1.foo(); + t1.foo(); + t2.foo(); + + #pragma omp target teams distribute parallel for simd + for (int i = 0; i < 10; ++i) + ++r; + + return r + *R; +} + +// Check metadata is properly generated: +// CHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 268, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 286, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 293, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 312, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 436, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 319, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 312, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 319, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 242, i32 {{[0-9]+}}} + +// TCHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 268, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 286, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 293, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 312, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 436, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 319, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 312, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 319, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 242, i32 {{[0-9]+}}} + +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen_registration_naming.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen_registration_naming.cpp new file mode 100644 index 0000000000000..faeaeb949018c --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_codegen_registration_naming.cpp @@ -0,0 +1,86 @@ +// Test host codegen. +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// Test target teams distribute parallel for simd codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s -check-prefix=TCHECK +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// CHECK: [[CA:%.+]] = type { i32* } + +// CHECK: define {{.*}}i32 @[[NNAME:.+]](i32 {{.*}}%{{.+}}) +int nested(int a){ + // CHECK: call void @__omp_offloading_[[FILEID:[0-9a-f]+_[0-9a-f]+]]_[[NNAME]]_l[[T1L:[0-9]+]]( + #pragma omp target teams distribute parallel for simd + for (int i = 0; i < 10; ++i) + ++a; + + // CHECK: call void @"[[LNAME:.+]]"([[CA]]* + auto F = [&](){ + #pragma omp parallel + { + #pragma omp target teams distribute parallel for simd + for (int i = 0; i < 10; ++i) + ++a; + } + }; + + F(); + + return a; +} + +// CHECK: define {{.*}}void @__omp_offloading_[[FILEID]]_[[NNAME]]_l[[T1L]]( +// TCHECK: define {{.*}}void @__omp_offloading_[[FILEID:[0-9a-f]+_[0-9a-f]+]]_[[NNAME:.+]]_l[[T1L:[0-9]+]]( + +// CHECK: define {{.*}}void @"[[LNAME]]"( +// CHECK: call void {{.*}}@__kmpc_fork_call{{.+}}[[PNAME:@.+]] to + +// CHECK: define {{.*}}void [[PNAME]]( +// CHECK: call void @__omp_offloading_[[FILEID]]_[[NNAME]]_l[[T2L:[0-9]+]]( + +// CHECK: define {{.*}}void @__omp_offloading_[[FILEID]]_[[NNAME]]_l[[T2L]]( +// TCHECK: define {{.*}}void @__omp_offloading_[[FILEID]]_[[NNAME:.+]]_l[[T2L:[0-9]+]]( + + +// Check metadata is properly generated: +// CHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 {{-?[0-9]+}}, i32 {{-?[0-9]+}}, !"[[NNAME]]", i32 [[T1L]], i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 {{-?[0-9]+}}, i32 {{-?[0-9]+}}, !"[[NNAME]]", i32 [[T2L]], i32 {{[0-9]+}}} + +// TCHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 {{-?[0-9]+}}, i32 {{-?[0-9]+}}, !"[[NNAME]]", i32 [[T1L]], i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 {{-?[0-9]+}}, i32 {{-?[0-9]+}}, !"[[NNAME]]", i32 [[T2L]], i32 {{[0-9]+}}} +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_collapse_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_collapse_codegen.cpp new file mode 100644 index 0000000000000..3aca29165cd89 --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_collapse_codegen.cpp @@ -0,0 +1,157 @@ +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// Test host codegen. +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} +#ifdef CK1 + +template <typename T, int X, long long Y> +struct SS{ + T a[X][Y]; + + // CK1: define {{.*}}i32 @{{.+}}foo{{.+}}( + int foo(void) { + + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL1:.+]]( + #pragma omp target teams distribute parallel for simd collapse(2) + for(int i = 0; i < X; i++) { + for(int j = 0; j < Y; j++) { + a[i][j] = (T)0; + } + } + // CK1: define internal void @[[OFFL1]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL1:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL1]]({{.+}}) + // discard loop variables not needed here + // CK1: [[OMP_UB:%.omp.comb.ub]] = alloca i32, + // CK1: store i32 56087, i32* [[OMP_UB]], + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92, {{.+}}, {{.+}}, i32* [[OMP_UB]], + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL1]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, {{.+}}, {{.+}}, + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + return a[0][0]; + } +}; + +int teams_template_struct(void) { + SS<int, 123, 456> V; + return V.foo(); + +} +#endif // CK1 + +// Test host codegen. +// RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64 +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64 +// RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} +#ifdef CK2 + +template <typename T, int n, int m> +int tmain(T argc) { + T a[n][m]; + #pragma omp target teams distribute parallel for simd collapse(2) + for(int i = 0; i < n; i++) { + for(int j = 0; j < m; j++) { + a[i][j] = (T)0; + } + } + return 0; +} + +int main (int argc, char **argv) { + int n = 100; + int m = 2; + int a[n][m]; + #pragma omp target teams distribute parallel for simd collapse(2) + for(int i = 0; i < n; i++) { + for(int j = 0; j < m; j++) { + a[i][j] = 0; + } + } + return tmain<int, 10, 2>(argc); +} + +// CK2: define {{.*}}i32 @{{[^,]+}}(i{{.+}}{{.+}} %[[ARGC:.+]], {{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL1:.+]]({{.+}}) +// CK2: {{%.+}} = call{{.*}} i32 @[[TMAIN:.+]]({{.+}}) +// CK2: ret + +// CK2: define {{.*}}void @[[OFFL1]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 5, {{.+}} @[[OUTL1:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL1]]({{.+}}) +// CK2: [[OMP_UB:%.omp.comb.ub]] = alloca i64, +// CK2: store i64 {{.+}}, i64* [[OMP_UB]], +// CK2: call void @__kmpc_for_static_init_8({{.+}}, {{.+}}, i32 92, {{.+}}, {{.+}}, i64* [[OMP_UB]], +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_{{[4|8]}}({{.+}}, {{.+}}, i32 34, {{.+}}, {{.+}}, +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + + +// CK2: define {{.*}}i32 @[[TMAIN]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT1:.+]]({{.+}}) +// CK2: ret +// CK2-NEXT: } + +// CK2: define {{.*}}void @[[OFFLT1]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTLT1:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT1]]({{.+}}) +// discard loop variables not needed here +// CK2: [[OMP_UB:%.omp.comb.ub]] = alloca i32, +// CK2: store i32 {{.+}}, i32* [[OMP_UB]], +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92, {{.+}}, {{.+}}, i32* [[OMP_UB]], +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[TPAR_OUTL1:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[TPAR_OUTL1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, {{.+}}, {{.+}}, +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +#endif // CK2 +#endif // #ifndef HEADER diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_collapse_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_collapse_messages.cpp index d0c8a5823cddd..069ea65ad29d6 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_collapse_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp %s -std=c++98 // RUN: %clang_cc1 -verify -fopenmp %s -std=c++11 +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s -std=c++98 +// RUN: %clang_cc1 -verify -fopenmp-simd %s -std=c++11 + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_default_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_default_messages.cpp index 7d07280d089d3..a81aacea4af04 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_default_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_defaultmap_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_defaultmap_messages.cpp index 6f0ea7e62fe87..c2c0e7be21302 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_defaultmap_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_defaultmap_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_codegen.cpp new file mode 100644 index 0000000000000..2a10ae872375c --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_codegen.cpp @@ -0,0 +1,261 @@ +// Test host codegen. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// CHECK-DAG: [[TT:%.+]] = type { i64, i8 } +// CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } +// CHECK-DAG: [[DEVTY:%.+]] = type { i8*, i8*, [[ENTTY]]*, [[ENTTY]]* } +// CHECK-DAG: [[DSCTY:%.+]] = type { i32, [[DEVTY]]*, [[ENTTY]]*, [[ENTTY]]* } + +// TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 } + +// CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat + +// CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 0, i[[SZ]] 4] +// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 800] +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] + +// Check if offloading descriptor is created. +// CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] +// CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] +// CHECK: [[DEVBEGIN:@.+]] = external constant i8 +// CHECK: [[DEVEND:@.+]] = external constant i8 +// CHECK: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x [[DEVTY]]] [{{.+}} { i8* [[DEVBEGIN]], i8* [[DEVEND]], [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }], comdat($[[REGFN]]) +// CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) + +// Check target registration is registered as a Ctor. +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] + + +template<typename tx, typename ty> +struct TT{ + tx X; + ty Y; +}; + +int global; +extern int global; + +// CHECK: define {{.*}}[[FOO:@.+]]( +int foo(int n) { + int a = 0; + short aa = 0; + float b[10]; + float bn[n]; + double c[5][10]; + double cn[5][n]; + TT<long long, char> d; + static long *plocal; + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID:@.+]], i32 [[GTID:%.+]], i32 1, i[[SZ]] {{20|40}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY0:%.+]]* + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 3 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + #pragma omp target teams distribute parallel for simd device(global + a) depend(in: global) depend(out: a, b, cn[4]) + for (int i = 0; i < 10; ++i) { + } + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + + // CHECK: [[BOOL:%.+]] = icmp ne i32 %{{.+}}, 0 + // CHECK: br i1 [[BOOL]], label %[[THEN:.+]], label %[[ELSE:.+]] + // CHECK: [[THEN]]: + // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%.+]], i32 0, i32 0 + // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%.+]], i32 0, i32 0 + // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]** + // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]** + // CHECK-DAG: store i[[SZ]]* [[BP0:%[^,]+]], i[[SZ]]** [[CBPADDR0]] + // CHECK-DAG: store i[[SZ]]* [[BP0]], i[[SZ]]** [[CPADDR0]] + + // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1 + // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1 + // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* + // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* + // CHECK-DAG: store i[[SZ]] [[BP1:%[^,]+]], i[[SZ]]* [[CBPADDR1]] + // CHECK-DAG: store i[[SZ]] [[BP1]], i[[SZ]]* [[CPADDR1]] + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{104|52}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1_:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1_:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + + // CHECK: [[ELSE]]: + // CHECK-NOT: getelementptr inbounds [2 x i8*], [2 x i8*]* + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1__:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1__:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + // CHECK: [[EXIT]]: + + #pragma omp target teams distribute parallel for simd device(global + a) nowait depend(inout: global, a, bn) if(target:a) + for (int i = 0; i < *plocal; ++i) { + static int local1; + *plocal = global; + local1 = global; + } + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]* + // CHECK: getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + #pragma omp target teams distribute parallel for simd if(0) firstprivate(global) depend(out:global) + for (int i = 0; i < global; ++i) { + global += 1; + } + + return a; +} + +// Check that the offloading functions are emitted and that the arguments are +// correct and loaded correctly for the target regions in foo(). + +// CHECK: define internal void [[HVT0:@.+]]() + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, [[TASK_TY0]]* noalias) +// CHECK: store void (i8*, ...)* null, void (i8*, ...)** % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 0, i8** null, i8** null, i[[SZ]]* null, i64* null, i32 0, i32 0) +// CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: call void [[HVT0]]() +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT1:@.+]](i[[SZ]]* %{{.+}}, i[[SZ]] %{{.+}}) + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1_]](i32{{.*}}, [[TASK_TY1_]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[SZT:%.+]] = getelementptr inbounds [2 x i[[SZ]]], [2 x i[[SZ]]]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams_nowait(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 2, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SZT]], i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT]], i32 0, i32 0), i32 0, i32 0) + +// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1__]](i32{{.*}}, [[TASK_TY1__]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT2:@.+]](i[[SZ]] %{{.+}}) +// Create stack storage and store argument in there. +// CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align +// CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align +// CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* +// CHECK-64: load i32, i32* [[AA_CADDR]], align +// CHECK-32: load i32, i32* [[AA_ADDR]], align + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, [[TASK_TY2]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT2]](i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + + +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_messages.cpp index 4a9bf01e40bf7..7cc0d412341a9 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_depend_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_device_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_device_messages.cpp index 6a725797af6eb..c7b3bc9098db6 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_device_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_device_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp new file mode 100644 index 0000000000000..281a8d7425343 --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp @@ -0,0 +1,270 @@ +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// Test host codegen. +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} +#ifdef CK1 + +template <typename T, int X, long long Y> +struct SS{ + T a[X]; + float b; + // CK1: define {{.*}}i32 @{{.+}}foo{{.+}}( + int foo(void) { + + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL1:.+]]( + #pragma omp target teams distribute parallel for simd + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL2:.+]]( + #pragma omp target teams distribute parallel for simd dist_schedule(static) + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL3:.+]]( + #pragma omp target teams distribute parallel for simd dist_schedule(static, X/2) + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + // CK1: define internal void @[[OFFL1]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL1:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL1]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92 + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL1]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[OFFL2]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL2:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL2]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92 + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL2:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL2]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + + // CK1: define internal void @[[OFFL3]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL3:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL3]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 91 + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL3:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL3]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + return a[0]; + } +}; + +int teams_template_struct(void) { + SS<int, 123, 456> V; + return V.foo(); + +} +#endif // CK1 + +// Test host codegen. +// RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64 +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64 +// RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} +#ifdef CK2 + +template <typename T, int n> +int tmain(T argc) { + T a[n]; + int m = 10; +#pragma omp target teams distribute parallel for simd + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } +#pragma omp target teams distribute parallel for simd dist_schedule(static) + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } +#pragma omp target teams distribute parallel for simd dist_schedule(static, m) + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } + return 0; +} + +int main (int argc, char **argv) { + int n = 100; + int a[n]; + int m = 10; +#pragma omp target teams distribute parallel for simd + for(int i = 0; i < n; i++) { + a[i] = 0; + } +#pragma omp target teams distribute parallel for simd dist_schedule(static) + for(int i = 0; i < n; i++) { + a[i] = 0; + } +#pragma omp target teams distribute parallel for simd dist_schedule(static, m) + for(int i = 0; i < n; i++) { + a[i] = 0; + } + return tmain<int, 10>(argc); +} + +// CK2: define {{.*}}i32 @{{[^,]+}}(i{{.+}}{{.+}} %[[ARGC:.+]], {{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL1:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL2:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL3:.+]]({{.+}}) +// CK2: {{%.+}} = call{{.*}} i32 @[[TMAIN:.+]]({{.+}}) +// CK2: ret + +// CK2: define {{.*}}void @[[OFFL1]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[OUTL1:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92 +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFL2]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[OUTL2:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92 +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL2:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFL3]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 4, {{.+}} @[[OUTL3:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 91 +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL3:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}i32 @[[TMAIN]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT1:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT2:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT3:.+]]({{.+}}) +// CK2: ret +// CK2-NEXT: } + +// CK2: define {{.*}}void @[[OFFLT1]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTLT1:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92 +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT1:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFLT2]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTLT2:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 92 +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT2:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFLT3]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[OUTLT3:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 91 +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT3:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +#endif // CK2 +#endif // #ifndef HEADER diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_dist_schedule_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_dist_schedule_messages.cpp index 4fb37532f095e..079aaf69f556b 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_dist_schedule_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_dist_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_codegen.cpp new file mode 100644 index 0000000000000..0b09cc6c4027b --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_codegen.cpp @@ -0,0 +1,523 @@ +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix HCHECK --check-prefix HCHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix HCHECK --check-prefix HCHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix HCHECK --check-prefix HCHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix HCHECK --check-prefix HCHECK-32 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 --check-prefix HLAMBDA --check-prefix HLAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 --check-prefix HLAMBDA --check-prefix HLAMBDA-64 + +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY + +// Test target codegen - host bc file has to be created first. (no significant differences with host version of target region) +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 --check-prefix TLAMBDA --check-prefix TLAMBDA-64 + +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +struct St { + int a, b; + St() : a(0), b(0) {} + St(const St &st) : a(st.a + st.b), b(0) {} + ~St() {} +}; + +volatile int g = 1212; +volatile int &g1 = g; + +template <class T> +struct S { + T f; + S(T a) : f(a + g) {} + S() : f(g) {} + S(const S &s, St t = St()) : f(s.f + t.a) {} + operator T() { return T(); } + ~S() {} +}; + +// CHECK-DAG: [[S_FLOAT_TY:%.+]] = type { float } +// CHECK-DAG: [[S_INT_TY:%.+]] = type { i{{[0-9]+}} } +// CHECK-DAG: [[ST_TY:%.+]] = type { i{{[0-9]+}}, i{{[0-9]+}} } + +template <typename T> +T tmain() { + S<T> test; + T t_var = T(); + T vec[] = {1, 2}; + S<T> s_arr[] = {1, 2}; + S<T> &var = test; +#pragma omp target teams distribute parallel for simd firstprivate(t_var, vec, s_arr, var) + for (int i = 0; i < 2; ++i) { + vec[i] = t_var; + s_arr[i] = var; + } + return T(); +} + +// HCHECK-DAG: [[TEST:@.+]] = global [[S_FLOAT_TY]] zeroinitializer, +S<float> test; +// HCHECK-DAG: [[T_VAR:@.+]] = global i{{[0-9]+}} 333, +int t_var = 333; +// HCHECK-DAG: [[VEC:@.+]] = global [2 x i{{[0-9]+}}] [i{{[0-9]+}} 1, i{{[0-9]+}} 2], +int vec[] = {1, 2}; +// HCHECK-DAG: [[S_ARR:@.+]] = global [2 x [[S_FLOAT_TY]]] zeroinitializer, +S<float> s_arr[] = {1, 2}; +// HCHECK-DAG: [[VAR:@.+]] = global [[S_FLOAT_TY]] zeroinitializer, +S<float> var(3); +// HCHECK-DAG: [[SIVAR:@.+]] = internal global i{{[0-9]+}} 0, + +int main() { + static int sivar; +#ifdef LAMBDA + // HLAMBDA: [[G:@.+]] = global i{{[0-9]+}} 1212, + // HLAMBDA-LABEL: @main + // HLAMBDA: call void [[OUTER_LAMBDA:@.+]]( + [&]() { + // HLAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]]( + // HLAMBDA: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 3, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i32 0, i32 0) + // HLAMBDA: call void @[[LOFFL1:.+]](i{{64|32}} %{{.+}}) + // HLAMBDA: ret +#pragma omp target teams distribute parallel for simd firstprivate(g, g1, sivar) + for (int i = 0; i < 2; ++i) { + // HLAMBDA: define{{.*}} internal{{.*}} void @[[LOFFL1]](i{{64|32}} {{%.+}}, i{{64|32}} {{%.+}}) + // TLAMBDA: define weak void @[[LOFFL1:.+]](i{{64|32}} {{%.+}}, i{{64|32}} {{%.+}}) + // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, + // LAMBDA: [[G_CAST:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[G1_CAST:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[SIVAR_CAST:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA-DAG: [[G_CAST_VAL:%.+]] = load{{.+}} [[G_CAST]], + // LAMBDA-DAG: [[G1_CAST_VAL:%.+]] = load{{.+}} [[G1_CAST]], + // LAMBDA-DAG: [[SIVAR_CAST_VAL:%.+]] = load{{.+}} [[SIVAR_CAST]], + // LAMBDA: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[LOUTL1:.+]] to {{.+}}, {{.+}} [[G_CAST_VAL]], {{.+}} [[G1_CAST_VAL]], {{.+}} [[SIVAR_CAST_VAL]]) + // LAMBDA: ret void + + // LAMBDA: define internal void @[[LOUTL1]]({{.+}}) + // Skip global and bound tid vars + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: [[G_ADDR:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[G1_ADDR:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[G1_TMP:%.+]] = alloca i32*, + // skip loop vars + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_ADDR]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_ADDR]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[SIVAR_ADDR]], + // LAMBDA-DAG: [[G_CONV:%.+]] = bitcast {{.+}} [[G_ADDR]] to + // LAMBDA-DAG: [[G1_CONV:%.+]] = bitcast {{.+}} [[G1_ADDR]] to + // LAMBDA-DAG: [[SIVAR_CONV:%.+]] = bitcast {{.+}} [[SIVAR_ADDR]] to + // LAMBDA-DAG: store{{.+}} [[G1_CONV]], {{.+}} [[G1_TMP]], + g = 1; + g1 = 1; + sivar = 2; + // LAMBDA: call void @__kmpc_for_static_init_4( + // LAMBDA: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[LPAR_OUTL:.+]] to + // LAMBDA: call void @__kmpc_for_static_fini( + // LAMBDA: ret void + + // LAMBDA: define internal void @[[LPAR_OUTL]]({{.+}}) + // Skip global and bound tid vars, and prev lb and ub vars + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: [[G_ADDR:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[G1_ADDR:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[G1_TMP:%.+]] = alloca i32*, + // skip loop vars + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_ADDR]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_ADDR]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[SIVAR_ADDR]], + // LAMBDA-DAG: [[G_CONV:%.+]] = bitcast {{.+}} [[G_ADDR]] to + // LAMBDA-DAG: [[G1_CONV:%.+]] = bitcast {{.+}} [[G1_ADDR]] to + // LAMBDA-DAG: [[SIVAR_CONV:%.+]] = bitcast {{.+}} [[SIVAR_ADDR]] to + // LAMBDA-DAG: store{{.+}} [[G1_CONV]], {{.+}} [[G1_TMP]], + + // use of private vars + // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_CONV]], + // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP]] + // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1]], + // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_CONV]], + // LAMBDA-DAG: [[G1_REF:%.+]] = load{{.+}}, {{.+}} [[G1_TMP]], + // LAMBDA: call void [[INNER_LAMBDA:@.+]]( + // LAMBDA: call void @__kmpc_for_static_fini( + // LAMBDA: ret void + [&]() { + // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) + // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], + g = 2; + g1 = 2; + sivar = 4; + // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]] + + // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 + // LAMBDA: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_PTR_REF]] + // LAMBDA: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[G_REF]] + // LAMBDA: [[G1_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 + // LAMBDA: [[G1_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G1_PTR_REF]] + // LAMBDA: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[G1_REF]] + // LAMBDA: [[SIVAR_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 + // LAMBDA: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_PTR_REF]] + // LAMBDA: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIVAR_REF]] + }(); + } + }(); + return 0; +#else +#pragma omp target teams distribute parallel for simd firstprivate(t_var, vec, s_arr, var, sivar) + for (int i = 0; i < 2; ++i) { + vec[i] = t_var; + s_arr[i] = var; + sivar += i; + } + return tmain<int>(); +#endif +} + +// HCHECK: define {{.*}}i{{[0-9]+}} @main() +// HCHECK: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 5, +// HCHECK: call void @[[OFFL1:.+]]( +// HCHECK: {{%.+}} = call{{.*}} i32 @[[TMAIN_INT:.+]]() +// HCHECK: ret + +// HCHECK: define{{.*}} void @[[OFFL1]]( +// TCHECK: define{{.*}} void @[[OFFL1:.+]]( +// CHECK-DAG: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK-DAG: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK-DAG: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*, +// CHECK-DAG: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]]*, +// CHECK: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[T_VAR_CAST:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[SIVAR_CAST:%.+]] = alloca i{{[0-9]+}}, + +// CHECK-DAG: [[VEC_TE_PAR:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_PRIV]], +// CHECK-DAG: [[T_VAR_TE_PAR:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[T_VAR_CAST]], +// CHECK-DAG: [[S_ARR_TE_PAR:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[S_ARR_PRIV]], +// CHECK-DAG: [[VAR_TE_PAR:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[VAR_PRIV]], +// CHECK-DAG: [[SIVAR_TE_PAR:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_CAST]], + +// CHECK: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 5, {{.+}} @[[OUTL1:.+]] to {{.+}}, [2 x i{{[0-9]+}}]* [[VEC_TE_PAR]], i{{[0-9]+}} [[T_VAR_TE_PAR]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_TE_PAR]], [[S_FLOAT_TY]]* [[VAR_TE_PAR]], i{{[0-9]+}} [[SIVAR_TE_PAR]]) +// CHECK: ret void + +// CHECK: define internal void @[[OUTL1]]({{.+}}) +// Skip global and bound tid vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: [[VEC_ADDR:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*, +// CHECK: [[VAR_ADDR:%.+]] = alloca [[S_FLOAT_TY]]*, +// CHECK: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// Skip temp vars for loop +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]], +// CHECK: [[AGG_TMP1:%.+]] = alloca [[ST_TY]], +// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]], +// CHECK: [[AGG_TMP2:%.+]] = alloca [[ST_TY]], + +// param copy +// CHECK: store [2 x i{{[0-9]+}}]* {{.+}}, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[T_VAR_ADDR]], +// CHECK: store [2 x [[S_FLOAT_TY]]]* {{.+}}, [2 x [[S_FLOAT_TY]]]** [[S_ARR_ADDR]], +// CHECK: store [[S_FLOAT_TY]]* {{.+}}, [[S_FLOAT_TY]]** [[VAR_ADDR]], +// CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[SIVAR_ADDR]], + +// T_VAR and SIVAR +// CHECK-64-DAG: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* + +// preparation vars +// CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK-DAG: [[S_ARR_ADDR_REF:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[S_ARR_ADDR]], +// CHECK-DAG: [[VAR_ADDR_REF:%.+]] = load{{.+}} [[VAR_ADDR]], + +// firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 +// CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* +// CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) + +// firstprivate(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_ADDR_BGN:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]] to +// CHECK-DAG: [[S_ARR_FIN:%.+]] = icmp{{.+}} [[S_ARR_PRIV_BGN]], +// CHECK-DAG: [[S_ARR_SRC_COPY:%.+]] = phi{{.+}} [ [[S_ARR_ADDR_BGN]], {{.+}} ], [ [[S_ARR_SRC:%.+]], {{.+}} ] +// CHECK-DAG: [[S_ARR_DST_COPY:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}}], [ [[S_ARR_DST:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_DST_COPY]], {{.+}} [[S_ARR_SRC_COPY]], {{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: [[S_ARR_DST]] = getelementptr {{.+}} [[S_ARR_DST_COPY]], +// CHECK-DAG: [[S_ARR_SRC]] = getelementptr {{.+}} [[S_ARR_SRC_COPY]], + +// firstprivate(var) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]], {{.+}} [[VAR_ADDR_REF]], {{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL:.+]] to +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: define internal void @[[PAR_OUTL]]({{.+}}) +// Skip global and bound tid vars, and prev lb ub vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: [[VEC_ADDR:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*, +// CHECK: [[VAR_ADDR:%.+]] = alloca [[S_FLOAT_TY]]*, +// CHECK: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// Skip temp vars for loop +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]], +// CHECK: [[AGG_TMP1:%.+]] = alloca [[ST_TY]], +// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]], +// CHECK: [[AGG_TMP2:%.+]] = alloca [[ST_TY]], + +// param copy +// CHECK: store [2 x i{{[0-9]+}}]* {{.+}}, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[T_VAR_ADDR]], +// CHECK: store [2 x [[S_FLOAT_TY]]]* {{.+}}, [2 x [[S_FLOAT_TY]]]** [[S_ARR_ADDR]], +// CHECK: store [[S_FLOAT_TY]]* {{.+}}, [[S_FLOAT_TY]]** [[VAR_ADDR]], +// CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[SIVAR_ADDR]], + +// T_VAR and SIVAR +// CHECK-64-DAG: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* + +// preparation vars +// CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK-DAG: [[S_ARR_ADDR_REF:%.+]] = load [2 x [[S_FLOAT_TY]]]*, [2 x [[S_FLOAT_TY]]]** [[S_ARR_ADDR]], +// CHECK-DAG: [[VAR_ADDR_REF:%.+]] = load{{.+}} [[VAR_ADDR]], + +// firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 +// CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* +// CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) + +// firstprivate(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_ADDR_BGN:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]] to +// CHECK-DAG: [[S_ARR_FIN:%.+]] = icmp{{.+}} [[S_ARR_PRIV_BGN]], +// CHECK-DAG: [[S_ARR_SRC_COPY:%.+]] = phi{{.+}} [ [[S_ARR_ADDR_BGN]], {{.+}} ], [ [[S_ARR_SRC:%.+]], {{.+}} ] +// CHECK-DAG: [[S_ARR_DST_COPY:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}}], [ [[S_ARR_DST:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_DST_COPY]], {{.+}} [[S_ARR_SRC_COPY]], {{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: [[S_ARR_DST]] = getelementptr {{.+}} [[S_ARR_DST_COPY]], +// CHECK-DAG: [[S_ARR_SRC]] = getelementptr {{.+}} [[S_ARR_SRC_COPY]], + +// firstprivate(var) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]], {{.+}} [[VAR_ADDR_REF]], {{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[VAR_PRIV]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[SIVAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_SIVAR]] +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// HCHECK: define{{.*}} i{{[0-9]+}} @[[TMAIN_INT]]() +// HCHECK: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 4, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i32 0, i32 0) +// HCHECK: call void @[[TOFFL1:.+]]( +// HCHECK: ret + +// HCHECK: define {{.*}}void @[[TOFFL1]]({{.+}}) +// TCHECK: define {{.*}}void @[[TOFFL1:.+]]({{.+}}) +// CHECK-DAG: [[TT_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK-DAG: [[TVEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK-DAG: [[TS_ARR_PRIV:%.+]] = alloca [2 x [[S_INT_TY]]]*, +// CHECK-DAG: [[TVAR_PRIV:%.+]] = alloca [[S_INT_TY]]*, +// CHECK: [[TT_VAR_CAST:%.+]] = alloca i{{[0-9]+}}, + +// CHECK-DAG: [[TVEC_TE_PAR:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[TVEC_PRIV]], +// CHECK-DAG: [[TT_VAR_TE_PAR:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[TT_VAR_CAST]], +// CHECK-DAG: [[TS_ARR_TE_PAR:%.+]] = load [2 x [[S_INT_TY]]]*, [2 x [[S_INT_TY]]]** [[TS_ARR_PRIV]], +// CHECK-DAG: [[TVAR_TE_PAR:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TVAR_PRIV]], + +// CHECK: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 4, {{.+}} @[[TOUTL1:.+]] to {{.+}}, [2 x i{{[0-9]+}}]* [[TVEC_TE_PAR]], i{{[0-9]+}} [[TT_VAR_TE_PAR]], [2 x [[S_INT_TY]]]* [[TS_ARR_TE_PAR]], [[S_INT_TY]]* [[TVAR_TE_PAR]]) +// CHECK: ret void + +// CHECK: define internal void @[[TOUTL1]]({{.+}}) +// Skip global and bound tid vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: [[VEC_ADDR:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR:%.+]] = alloca [2 x [[S_INT_TY]]]*, +// CHECK: [[VAR_ADDR:%.+]] = alloca [[S_INT_TY]]*, +// Skip temp vars for loop +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_INT_TY]]], +// CHECK: [[AGG_TMP1:%.+]] = alloca [[ST_TY]], +// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]], +// CHECK: [[AGG_TMP2:%.+]] = alloca [[ST_TY]], +// CHECK: [[TMP:%.+]] = alloca [[S_INT_TY]]*, + +// param copy +// CHECK: store [2 x i{{[0-9]+}}]* {{.+}}, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[T_VAR_ADDR]], +// CHECK: store [2 x [[S_INT_TY]]]* {{.+}}, [2 x [[S_INT_TY]]]** [[S_ARR_ADDR]], +// CHECK: store [[S_INT_TY]]* {{.+}}, [[S_INT_TY]]** [[VAR_ADDR]], + +// T_VAR and preparation variables +// CHECK: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK-64: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK: [[S_ARR_ADDR_REF:%.+]] = load [2 x [[S_INT_TY]]]*, [2 x [[S_INT_TY]]]** [[S_ARR_ADDR]], + +// firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 +// CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* +// CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) + +// firstprivate(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_ADDR_BGN:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]] to +// CHECK-DAG: [[S_ARR_FIN:%.+]] = icmp{{.+}} [[S_ARR_PRIV_BGN]], +// CHECK-DAG: [[S_ARR_SRC_COPY:%.+]] = phi{{.+}} [ [[S_ARR_ADDR_BGN]], {{.+}} ], [ [[S_ARR_SRC:%.+]], {{.+}} ] +// CHECK-DAG: [[S_ARR_DST_COPY:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_DST:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_DST_COPY]], {{.+}} [[S_ARR_SRC_COPY]], {{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: [[S_ARR_DST]] = getelementptr {{.+}} [[S_ARR_DST_COPY]], +// CHECK-DAG: [[S_ARR_SRC]] = getelementptr {{.+}} [[S_ARR_SRC_COPY]], + +// firstprivate(var) +// CHECK-DAG: [[VAR_ADDR_REF:%.+]] = load{{.+}} [[VAR_ADDR]], +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]], {{.+}} [[VAR_ADDR_REF]], {{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) +// CHECK-DAG: store [[S_INT_TY]]* [[VAR_PRIV]], [[S_INT_TY]]** [[TMP]], + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[TPAR_OUTL:.+]] to +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: define internal void @[[TPAR_OUTL]]({{.+}}) +// Skip global and bound tid vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: [[VEC_ADDR:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR:%.+]] = alloca [2 x [[S_INT_TY]]]*, +// CHECK: [[VAR_ADDR:%.+]] = alloca [[S_INT_TY]]*, +// Skip temp vars for loop +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_INT_TY]]], +// CHECK: [[AGG_TMP1:%.+]] = alloca [[ST_TY]], +// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]], +// CHECK: [[AGG_TMP2:%.+]] = alloca [[ST_TY]], +// CHECK: [[TMP:%.+]] = alloca [[S_INT_TY]]*, + +// param copy +// CHECK: store [2 x i{{[0-9]+}}]* {{.+}}, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[T_VAR_ADDR]], +// CHECK: store [2 x [[S_INT_TY]]]* {{.+}}, [2 x [[S_INT_TY]]]** [[S_ARR_ADDR]], +// CHECK: store [[S_INT_TY]]* {{.+}}, [[S_INT_TY]]** [[VAR_ADDR]], + +// T_VAR and preparation variables +// CHECK: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], +// CHECK-64: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK: [[S_ARR_ADDR_REF:%.+]] = load [2 x [[S_INT_TY]]]*, [2 x [[S_INT_TY]]]** [[S_ARR_ADDR]], + +// firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 +// CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* +// CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) + +// firstprivate(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_ADDR_BGN:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]] to +// CHECK-DAG: [[S_ARR_FIN:%.+]] = icmp{{.+}} [[S_ARR_PRIV_BGN]], +// CHECK-DAG: [[S_ARR_SRC_COPY:%.+]] = phi{{.+}} [ [[S_ARR_ADDR_BGN]], {{.+}} ], [ [[S_ARR_SRC:%.+]], {{.+}} ] +// CHECK-DAG: [[S_ARR_DST_COPY:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_DST:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_DST_COPY]], {{.+}} [[S_ARR_SRC_COPY]], {{.+}} [[AGG_TMP1]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP1]]) +// CHECK-DAG: [[S_ARR_DST]] = getelementptr {{.+}} [[S_ARR_DST_COPY]], +// CHECK-DAG: [[S_ARR_SRC]] = getelementptr {{.+}} [[S_ARR_SRC_COPY]], + +// firstprivate(var) +// CHECK-DAG: [[VAR_ADDR_REF:%.+]] = load{{.+}} [[VAR_ADDR]], +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]], {{.+}} [[VAR_ADDR_REF]], {{.+}} [[AGG_TMP2]]) +// CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) +// CHECK-DAG: store [[S_INT_TY]]* [[VAR_PRIV]], [[S_INT_TY]]** [[TMP]], + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[TMP]] +// CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_messages.cpp index 4790b4bffdaf1..69292a7c6183b 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_if_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_if_codegen.cpp new file mode 100644 index 0000000000000..a3a14fbf1e685 --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_if_codegen.cpp @@ -0,0 +1,181 @@ +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +void fn1(); +void fn2(); +void fn3(); +void fn4(); +void fn5(); +void fn6(); + +int Arg; + +// CHECK-LABEL: define {{.*}}void @{{.+}}gtid_test +void gtid_test() { +// CHECK: call i{{[0-9]+}} @__tgt_target_teams( +// CHECK: call void [[OFFLOADING_FUN_0:@.+]]( +// CHECK: call i{{[0-9]+}} @__tgt_target_teams( +// CHECK: call void [[OFFLOADING_FUN_1:@.+]]( +#pragma omp target teams distribute parallel for simd + for(int i = 0 ; i < 100; i++) {} + // CHECK: define internal void [[OFFLOADING_FUN_0]]( + // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}}* [[OMP_TEAMS_OUTLINED_0:@.+]] to {{.+}}) + // CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_0]]( + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, {{.+}}* [[OMP_OUTLINED_0:@.+]] to void + // CHECK: call void @__kmpc_for_static_fini( + + // CHECK: define{{.+}} void [[OMP_OUTLINED_0]]( + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call void @__kmpc_for_static_fini( + // CHECK: ret +#pragma omp target teams distribute parallel for simd if (parallel: false) + for(int i = 0 ; i < 100; i++) { + // CHECK: define internal void [[OFFLOADING_FUN_1]]( + // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}}* [[OMP_TEAMS_OUTLINED_1:@.+]] to {{.+}}) + // CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_1]]( + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call void @__kmpc_serialized_parallel( + // CHECK: call void [[OMP_OUTLINED_1:@.+]]( + // CHECK: call void @__kmpc_end_serialized_parallel( + // CHECK: call void @__kmpc_for_static_fini( + // CHECK: define{{.+}} void [[OMP_OUTLINED_1]]( + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call void @{{.+}}gtid_test + // CHECK: call void @__kmpc_for_static_fini( + // CHECK: ret + gtid_test(); + } +} + + +template <typename T> +int tmain(T Arg) { +#pragma omp target teams distribute parallel for simd if (true) + for(int i = 0 ; i < 100; i++) { + fn1(); + } +#pragma omp target teams distribute parallel for simd if (false) + for(int i = 0 ; i < 100; i++) { + fn2(); + } +#pragma omp target teams distribute parallel for simd if (parallel: Arg) + for(int i = 0 ; i < 100; i++) { + fn3(); + } + return 0; +} + +// CHECK-LABEL: define {{.*}}i{{[0-9]+}} @main() +int main() { +// CHECK: call i{{[0-9]+}} @__tgt_target_teams( +// CHECK: call void [[OFFLOADING_FUN_0:@.+]]( +// CHECK: call void [[OFFLOADING_FUN_1:@.+]]( +// CHECK: call i{{[0-9]+}} @__tgt_target_teams( +// CHECK: call void [[OFFLOADING_FUN_2:@.+]]( +// CHECK: = call {{.*}}i{{.+}} @{{.+}}tmain +#pragma omp target teams distribute parallel for simd if (true) + for(int i = 0 ; i < 100; i++) { + // CHECK: define internal void [[OFFLOADING_FUN_0]]( + // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}}* [[OMP_TEAMS_OUTLINED_0:@.+]] to {{.+}}) + // CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_0]]( + + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, {{.+}}* [[OMP_OUTLINED_2:@.+]] to void + // CHECK: call void @__kmpc_for_static_fini( + // CHECK: define{{.+}} void [[OMP_OUTLINED_2]]( + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call {{.*}}void @{{.+}}fn4 + // CHECK: call void @__kmpc_for_static_fini( + + fn4(); + } + +#pragma omp target teams distribute parallel for simd if (false) + for(int i = 0 ; i < 100; i++) { + // CHECK: define internal void [[OFFLOADING_FUN_1]]( + // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}}* [[OMP_TEAMS_OUTLINED_1:@.+]] to {{.+}}) + // CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_1]]( + + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call void @__kmpc_serialized_parallel( + // CHECK: call void [[OMP_OUTLINED_3:@.+]]( + // CHECK: call void @__kmpc_end_serialized_parallel( + // CHECK: call void @__kmpc_for_static_fini( + + // CHECK: define{{.+}} void [[OMP_OUTLINED_3]]( + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call {{.*}}void @{{.+}}fn5 + // CHECK: call void @__kmpc_for_static_fini( + fn5(); + } + +#pragma omp target teams distribute parallel for simd if (Arg) + for(int i = 0 ; i < 100; i++) { + // CHECK: define internal void [[OFFLOADING_FUN_2]]( + // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}}* [[OMP_TEAMS_OUTLINED_2:@.+]] to {{.+}}) + // CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_2]]( + + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, {{.+}}* [[OMP_OUTLINED_4:@.+]] to void + // CHECK: call void @__kmpc_serialized_parallel( + // CHECK: call void [[OMP_OUTLINED_4:@.+]]( + // CHECK: call void @__kmpc_end_serialized_parallel( + // CHECK: call void @__kmpc_for_static_fini( + + // CHECK: define{{.+}} void [[OMP_OUTLINED_4]]( + // CHECK: call void @__kmpc_for_static_init_4( + // CHECK: call {{.*}}void @{{.+}}fn6 + // CHECK: call void @__kmpc_for_static_fini( + fn6(); + } + + return tmain(Arg); +} + +// CHECK-LABEL: define {{.+}} @{{.+}}tmain + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, void {{.+}}* [[T_OUTLINE_FUN_1:@.+]] to void +// CHECK: call void @__kmpc_for_static_fini( + +// CHECK: define internal {{.*}}void [[T_OUTLINE_FUN_1]] +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call {{.*}}void @{{.+}}fn1 +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call {{.*}}void @__kmpc_serialized_parallel( +// CHECK: call void [[T_OUTLINE_FUN_2:@.+]]( +// CHECK: call {{.*}}void @__kmpc_end_serialized_parallel( +// CHECK: call void @__kmpc_for_static_fini( + +// CHECK: define internal {{.*}}void [[T_OUTLINE_FUN_2]] +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call {{.*}}void @{{.+}}fn2 +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, void {{.+}}* [[T_OUTLINE_FUN_3:@.+]] to void +// CHECK: call {{.*}}void @__kmpc_serialized_parallel( +// call void [[T_OUTLINE_FUN_3:@.+]]( +// CHECK: call {{.*}}void @__kmpc_end_serialized_parallel( + +// CHECK: define internal {{.*}}void [[T_OUTLINE_FUN_3]] +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call {{.*}}void @{{.+}}fn3 +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_if_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_if_messages.cpp index 96db02eea7a34..7554e5bc0590a 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_if_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_is_device_ptr_ast_print.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_is_device_ptr_ast_print.cpp index e73f40ef1ccf2..510751967230d 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_is_device_ptr_ast_print.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_is_device_ptr_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -56,7 +60,7 @@ struct SA { // CHECK-NEXT: int aa[10]; // CHECK-NEXT: arr &raa = this->aa; // CHECK-NEXT: func( -// CHECK-NEXT: #pragma omp target teams distribute parallel for simd is_device_ptr(this->k) +// CHECK-NEXT: #pragma omp target teams distribute parallel for simd is_device_ptr(this->k){{$}} // CHECK-NEXT: for (int i = 0; i < 100; i++) // CHECK-NEXT: ; // CHECK-NEXT: #pragma omp target teams distribute parallel for simd is_device_ptr(this->z) diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_is_device_ptr_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_is_device_ptr_messages.cpp index 520f746efdd59..57abff3fb48ed 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_is_device_ptr_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_is_device_ptr_messages.cpp @@ -1,4 +1,6 @@ // RUN: %clang_cc1 -std=c++11 -verify -fopenmp %s + +// RUN: %clang_cc1 -std=c++11 -verify -fopenmp-simd %s struct ST { int *a; }; diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_lastprivate_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_lastprivate_codegen.cpp new file mode 100644 index 0000000000000..c3d38612d63d9 --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_lastprivate_codegen.cpp @@ -0,0 +1,456 @@ +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 + +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +template <class T> +struct S { + T f; + S(T a) : f(a) {} + S() : f() {} + operator T() { return T(); } + ~S() {} +}; + +// CHECK: [[S_FLOAT_TY:%.+]] = type { float } +// CHECK: [[S_INT_TY:%.+]] = type { i{{[0-9]+}} } +template <typename T> +T tmain() { + S<T> test; + T t_var = T(); + T vec[] = {1, 2}; + S<T> s_arr[] = {1, 2}; + S<T> &var = test; + #pragma omp target teams distribute parallel for simd lastprivate(t_var, vec, s_arr, s_arr, var, var) + for (int i = 0; i < 2; ++i) { + vec[i] = t_var; + s_arr[i] = var; + } + return T(); +} + +int main() { + static int svar; + volatile double g; + volatile double &g1 = g; + + #ifdef LAMBDA + // LAMBDA-LABEL: @main + // LAMBDA: call{{.*}} void [[OUTER_LAMBDA:@.+]]( + [&]() { + static float sfvar; + // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]]( + // LAMBDA: call i{{[0-9]+}} @__tgt_target_teams( + // LAMBDA: call void [[OFFLOADING_FUN:@.+]]( + + // LAMBDA: define{{.+}} void [[OFFLOADING_FUN]]( + // LAMBDA: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 4, {{.+}}* [[OMP_OUTLINED:@.+]] to {{.+}}) + #pragma omp target teams distribute parallel for simd lastprivate(g, g1, svar, sfvar) + for (int i = 0; i < 2; ++i) { + // LAMBDA: define{{.*}} internal{{.*}} void [[OMP_OUTLINED]](i32* {{.+}}, i32* {{.+}}, {{.+}} [[G1_IN:%.+]], {{.+}} [[SVAR_IN:%.+]], {{.+}} [[SFVAR_IN:%.+]], {{.+}} [[G_IN:%.+]]) + // skip gbl and bound tid + // LAMBDA: alloca + // LAMBDA: alloca + // LAMBDA: [[G1_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA: [[SVAR_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA: [[SFVAR_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA: [[G_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA-64: [[G1_REF:%.+]] = alloca double*, + // loop variables + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: [[OMP_IS_LAST:%.+]] = alloca i{{[0-9]+}}, + + // LAMBDA-DAG: store {{.+}} [[G_IN]], {{.+}} [[G_ADDR]], + // LAMBDA-DAG: store {{.+}} [[G1_IN]], {{.+}} [[G1_ADDR]], + // LAMBDA-DAG: store {{.+}} [[SVAR_IN]], {{.+}} [[SVAR_ADDR]], + // LAMBDA-DAG: store {{.+}} [[SFVAR_IN]], {{.+}} [[SFVAR_ADDR]], + + // LAMBDA-64-DAG: [[G_TGT:%.+]] = bitcast {{.+}} [[G_ADDR]] to + // LAMBDA-32-DAG: [[G_TGT:%.+]] = load {{.+}}, {{.+}} [[G_ADDR]], + // LAMBDA-64-DAG: [[G1_TGT:%.+]] = load {{.+}}, {{.+}} [[G1_REF]], + // LAMBDA-32-DAG: [[G1_TGT:%.+]] = load {{.+}}, {{.+}} [[G1_ADDR]], + // LAMBDA-64-DAG: [[SVAR_TGT:%.+]] = bitcast {{.+}} [[SVAR_ADDR]] to + // LAMBDA-DAG: [[SFVAR_TGT:%.+]] = bitcast {{.+}} [[SFVAR_ADDR]] to + + g1 = 1; + svar = 3; + sfvar = 4.0; + // LAMBDA: call {{.*}}void @__kmpc_for_static_init_4( + // LAMBDA: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[LPAR_OUTL:.+]] to + // LAMBDA: call {{.*}}void @__kmpc_for_static_fini( + + // LAMBDA: store i32 2, i32* % + // LAMBDA: [[OMP_IS_LAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[OMP_IS_LAST]], + // LAMBDA: [[IS_LAST_IT:%.+]] = icmp ne i{{[0-9]+}} [[OMP_IS_LAST_VAL]], 0 + // LAMBDA: br i1 [[IS_LAST_IT]], label %[[OMP_LASTPRIV_BLOCK:.+]], label %[[OMP_LASTPRIV_DONE:.+]] + + // LAMBDA: [[OMP_LASTPRIV_BLOCK]]: + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_TGT]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_TGT]], + // LAMBDA-64-DAG: store {{.+}}, {{.+}} [[SVAR_TGT]], + // LAMBDA-32-DAG: store {{.+}}, {{.+}} [[SVAR_ADDR]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[SFVAR_TGT]], + // LAMBDA: br label %[[OMP_LASTPRIV_DONE]] + // LAMBDA: [[OMP_LASTPRIV_DONE]]: + // LAMBDA: ret + + // LAMBDA: define{{.*}} internal{{.*}} void @[[LPAR_OUTL]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, {{.+}}, {{.+}}, {{.+}} [[G1_IN:%.+]], {{.+}} [[SVAR_IN:%.+]], {{.+}} [[SFVAR_IN:%.+]], {{.+}} [[G_IN:%.+]]) + // skip tid and prev variables + // LAMBDA: alloca + // LAMBDA: alloca + // LAMBDA: alloca + // LAMBDA: alloca + // LAMBDA: [[G1_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA: [[SVAR_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA: [[SFVAR_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA: [[G_ADDR:%.+]] = alloca {{.+}}, + // LAMBDA-64: [[G1_REF:%.+]] = alloca double*, + // loop variables + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: {{.+}} = alloca i{{[0-9]+}}, + // LAMBDA: [[OMP_IS_LAST:%.+]] = alloca i{{[0-9]+}}, + + // LAMBDA-DAG: store {{.+}} [[G_IN]], {{.+}} [[G_ADDR]], + // LAMBDA-DAG: store {{.+}} [[G1_IN]], {{.+}} [[G1_ADDR]], + // LAMBDA-DAG: store {{.+}} [[SVAR_IN]], {{.+}} [[SVAR_ADDR]], + // LAMBDA-DAG: store {{.+}} [[SFVAR_IN]], {{.+}} [[SFVAR_ADDR]], + + // LAMBDA-64-DAG: [[G_TGT:%.+]] = bitcast {{.+}} [[G_ADDR]] to + // LAMBDA-32-DAG: [[G_TGT:%.+]] = load {{.+}}, {{.+}} [[G_ADDR]], + // LAMBDA-64-DAG: [[G1_TGT:%.+]] = load {{.+}}, {{.+}} [[G1_REF]], + // LAMBDA-32-DAG: [[G1_TGT:%.+]] = load {{.+}}, {{.+}} [[G1_ADDR]], + // LAMBDA-64-DAG: [[SVAR_TGT:%.+]] = bitcast {{.+}} [[SVAR_ADDR]] to + // LAMBDA-DAG: [[SFVAR_TGT:%.+]] = bitcast {{.+}} [[SFVAR_ADDR]] to + + + // LAMBDA: call {{.*}}void @__kmpc_for_static_init_4( + // LAMBDA: call{{.*}} void [[INNER_LAMBDA:@.+]]( + // LAMBDA: call {{.*}}void @__kmpc_for_static_fini( + + // LAMBDA: store i32 2, i32* % + // LAMBDA: [[OMP_IS_LAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[OMP_IS_LAST]], + // LAMBDA: [[IS_LAST_IT:%.+]] = icmp ne i{{[0-9]+}} [[OMP_IS_LAST_VAL]], 0 + // LAMBDA: br i1 [[IS_LAST_IT]], label %[[OMP_LASTPRIV_BLOCK:.+]], label %[[OMP_LASTPRIV_DONE:.+]] + + // LAMBDA: [[OMP_LASTPRIV_BLOCK]]: + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_TGT]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_TGT]], + // LAMBDA-64-DAG: store {{.+}}, {{.+}} [[SVAR_TGT]], + // LAMBDA-32-DAG: store {{.+}}, {{.+}} [[SVAR_ADDR]], + // LAMBDA-DAG: store {{.+}}, {{.+}} [[SFVAR_TGT]], + // LAMBDA: br label %[[OMP_LASTPRIV_DONE]] + // LAMBDA: [[OMP_LASTPRIV_DONE]]: + // LAMBDA: ret + + [&]() { + // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) + // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], + g = 2; + g1 = 2; + svar = 4; + sfvar = 8.0; + // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]] + // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 + // LAMBDA: [[G_REF:%.+]] = load double*, double** [[G_PTR_REF]] + // LAMBDA: store double 2.0{{.+}}, double* [[G_REF]] + + // LAMBDA: [[TMP_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 + // LAMBDA: [[G1_REF:%.+]] = load double*, double** [[TMP_PTR_REF]] + // LAMBDA: store double 2.0{{.+}}, double* [[G1_REF]], + // LAMBDA: [[SVAR_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 + // LAMBDA: [[SVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SVAR_PTR_REF]] + // LAMBDA: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SVAR_REF]] + // LAMBDA: [[SFVAR_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 3 + // LAMBDA: [[SFVAR_REF:%.+]] = load float*, float** [[SFVAR_PTR_REF]] + // LAMBDA: store float 8.0{{.+}}, float* [[SFVAR_REF]] + }(); + } + }(); + return 0; + #else + S<float> test; + int t_var = 0; + int vec[] = {1, 2}; + S<float> s_arr[] = {1, 2}; + S<float> &var = test; + + #pragma omp target teams distribute parallel for simd lastprivate(t_var, vec, s_arr, s_arr, var, var, svar) + for (int i = 0; i < 2; ++i) { + vec[i] = t_var; + s_arr[i] = var; + } + int i; + + return tmain<int>(); + #endif +} + +// CHECK: define{{.*}} i{{[0-9]+}} @main() +// CHECK: [[TEST:%.+]] = alloca [[S_FLOAT_TY]], +// CHECK: call {{.*}} [[S_FLOAT_TY_DEF_CONSTR:@.+]]([[S_FLOAT_TY]]* [[TEST]]) +// CHECK: call i{{[0-9]+}} @__tgt_target_teams( +// CHECK: call void [[OFFLOAD_FUN:@.+]]( +// CHECK: ret + +// CHECK: define{{.+}} [[OFFLOAD_FUN]]( +// CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_teams( +// CHECK: ret +// +// CHECK: define internal void [[OMP_OUTLINED:@.+]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [2 x i{{[0-9]+}}]*{{.+}} [[VEC_IN:%.+]], i{{[0-9]+}}{{.+}} [[T_VAR_IN:%.+]], [2 x [[S_FLOAT_TY]]]*{{.+}} [[S_ARR_IN:%.+]], [[S_FLOAT_TY]]*{{.+}} [[VAR_IN:%.+]], i{{[0-9]+}}{{.*}} [[S_VAR_IN:%.+]]) +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: [[VEC_ADDR:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*, +// CHECK: [[VAR_ADDR:%.+]] = alloca [[S_FLOAT_TY]]*, +// CHECK: [[SVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// skip loop variables +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: [[OMP_IS_LAST:%.+]] = alloca i{{[0-9]+}}, + +// copy from parameters to local address variables +// CHECK: store {{.+}} [[VEC_IN]], {{.+}} [[VEC_ADDR]], +// CHECK: store {{.+}} [[T_VAR_IN]], {{.+}} [[T_VAR_ADDR]], +// CHECK: store {{.+}} [[S_ARR_IN]], {{.+}} [[S_ARR_ADDR]], +// CHECK: store {{.+}} [[VAR_IN]], {{.+}} [[VAR_ADDR]], +// CHECK: store {{.+}} [[S_VAR_IN]], {{.+}} [[SVAR_ADDR]], + +// prepare lastprivate targets +// CHECK-64-DAG: [[TVAR_TGT:%.+]] = bitcast {{.+}} [[T_VAR_ADDR]] to +// CHECK-DAG: [[VEC_TGT:%.+]] = load {{.+}}, {{.+}} [[VEC_ADDR]], +// CHECK-DAG: [[S_ARR_TGT:%.+]] = load {{.+}}, {{.+}} [[S_ARR_ADDR]], +// CHECK-DAG: [[VAR_TGT:%.+]] = load {{.+}}, {{.+}} [[VAR_ADDR]], +// CHECK-64-DAG: [[SVAR_TGT:%.+]] = bitcast {{.+}} [[SVAR_ADDR]] to + +// the distribute loop +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL:.+]] to +// CHECK: call void @__kmpc_for_static_fini( + +// lastprivates +// CHECK: [[OMP_IS_LAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[OMP_IS_LAST]], +// CHECK: [[IS_LAST_IT:%.+]] = icmp ne i{{[0-9]+}} [[OMP_IS_LAST_VAL]], 0 +// CHECK: br i1 [[IS_LAST_IT]], label %[[OMP_LASTPRIV_BLOCK:.+]], label %[[OMP_LASTPRIV_DONE:.+]] + +// CHECK: [[OMP_LASTPRIV_BLOCK]]: +// CHECK-64-DAG: store {{.+}}, {{.+}} [[TVAR_TGT]], +// CHECK-32-DAG: store {{.+}}, {{.+}} [[T_VAR_ADDR]], +// CHECK-DAG: [[VEC_TGT_REF:%.+]] = bitcast {{.+}} [[VEC_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_TGT_REF]], +// CHECK-DAG: [[S_ARR_BEGIN:%.+]] = getelementptr {{.+}} [[S_ARR_TGT]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}( +// CHECK-DAG: [[VAR_TGT_BCAST:%.+]] = bitcast {{.+}} [[VAR_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_TGT_BCAST]], +// CHECK-64-DAG: store {{.+}}, {{.+}} [[SVAR_TGT]], +// CHECK-32-DAG: store {{.+}}, {{.+}} [[SVAR_ADDR]], +// CHECK: ret void + +// CHECK: define internal void [[OMP_OUTLINED:@.+]](i{{[0-9]+}}* noalias [[GTID_ADDR:%.+]], i{{[0-9]+}}* noalias %{{.+}}, {{.+}}, {{.+}}, [2 x i{{[0-9]+}}]*{{.+}} [[VEC_IN:%.+]], i{{[0-9]+}}{{.+}} [[T_VAR_IN:%.+]], [2 x [[S_FLOAT_TY]]]*{{.+}} [[S_ARR_IN:%.+]], [[S_FLOAT_TY]]*{{.+}} [[VAR_IN:%.+]], i{{[0-9]+}}{{.*}} [[S_VAR_IN:%.+]]) + +// gbl and bound tid vars, prev lb and ub vars +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, + +// CHECK: [[VEC_ADDR:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR:%.+]] = alloca [2 x [[S_FLOAT_TY]]]*, +// CHECK: [[VAR_ADDR:%.+]] = alloca [[S_FLOAT_TY]]*, +// CHECK: [[SVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, +// skip loop variables +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: [[OMP_IS_LAST:%.+]] = alloca i{{[0-9]+}}, + +// copy from parameters to local address variables +// CHECK: store {{.+}} [[VEC_IN]], {{.+}} [[VEC_ADDR]], +// CHECK: store {{.+}} [[T_VAR_IN]], {{.+}} [[T_VAR_ADDR]], +// CHECK: store {{.+}} [[S_ARR_IN]], {{.+}} [[S_ARR_ADDR]], +// CHECK: store {{.+}} [[VAR_IN]], {{.+}} [[VAR_ADDR]], +// CHECK: store {{.+}} [[S_VAR_IN]], {{.+}} [[SVAR_ADDR]], + +// prepare lastprivate targets +// CHECK-64-DAG: [[TVAR_TGT:%.+]] = bitcast {{.+}} [[T_VAR_ADDR]] to +// CHECK-DAG: [[VEC_TGT:%.+]] = load {{.+}}, {{.+}} [[VEC_ADDR]], +// CHECK-DAG: [[S_ARR_TGT:%.+]] = load {{.+}}, {{.+}} [[S_ARR_ADDR]], +// CHECK-DAG: [[VAR_TGT:%.+]] = load {{.+}}, {{.+}} [[VAR_ADDR]], +// CHECK-64-DAG: [[SVAR_TGT:%.+]] = bitcast {{.+}} [[SVAR_ADDR]] to + +// the distribute loop +// CHECK: call void @__kmpc_for_static_init_4( +// skip body: code generation routine is same as distribute parallel for lastprivate +// CHECK: call void @__kmpc_for_static_fini( + +// lastprivates +// CHECK: [[OMP_IS_LAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[OMP_IS_LAST]], +// CHECK: [[IS_LAST_IT:%.+]] = icmp ne i{{[0-9]+}} [[OMP_IS_LAST_VAL]], 0 +// CHECK: br i1 [[IS_LAST_IT]], label %[[OMP_LASTPRIV_BLOCK:.+]], label %[[OMP_LASTPRIV_DONE:.+]] + +// CHECK: [[OMP_LASTPRIV_BLOCK]]: +// CHECK-64-DAG: store {{.+}}, {{.+}} [[TVAR_TGT]], +// CHECK-32-DAG: store {{.+}}, {{.+}} [[T_VAR_ADDR]], +// CHECK-DAG: [[VEC_TGT_REF:%.+]] = bitcast {{.+}} [[VEC_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_TGT_REF]], +// CHECK-DAG: [[S_ARR_BEGIN:%.+]] = getelementptr {{.+}} [[S_ARR_TGT]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}( +// CHECK-DAG: [[VAR_TGT_BCAST:%.+]] = bitcast {{.+}} [[VAR_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_TGT_BCAST]], +// CHECK-64-DAG: store {{.+}}, {{.+}} [[SVAR_TGT]], +// CHECK-32-DAG: store {{.+}}, {{.+}} [[SVAR_ADDR]], +// CHECK: ret void + +// template tmain +// CHECK: define{{.*}} i{{[0-9]+}} [[TMAIN_INT:@.+]]() +// CHECK: [[TEST:%.+]] = alloca [[S_INT_TY]], +// CHECK: call {{.*}} [[S_INT_TY_DEF_CONSTR:@.+]]([[S_INT_TY]]* [[TEST]]) +// CHECK: call i{{[0-9]+}} @__tgt_target_teams( +// CHECK: call void [[OFFLOAD_FUN_1:@.+]]( +// CHECK: ret + +// CHECK: define internal void [[OFFLOAD_FUN_1]]( +// CHECK: call void (%{{.+}}*, i{{[0-9]+}}, void (i{{[0-9]+}}*, i{{[0-9]+}}*, ...)*, ...) @__kmpc_fork_teams(%{{.+}}* @{{.+}}, i{{[0-9]+}} 4, +// CHECK: ret + +// CHECK: define internal void [[OMP_OUTLINED_1:@.+]](i{{[0-9]+}}* noalias [[GTID_ADDR1:%.+]], i{{[0-9]+}}* noalias %{{.+}}, [2 x i{{[0-9]+}}]*{{.+}} [[VEC_IN1:%.+]], i{{[0-9]+}}{{.+}} [[T_VAR_IN1:%.+]], [2 x [[S_INT_TY]]]*{{.+}} [[S_ARR_IN1:%.+]], [[S_INT_TY]]*{{.+}} [[VAR_IN1:%.+]]) +// skip alloca of global_tid and bound_tid +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: [[VEC_ADDR1:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR1:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR1:%.+]] = alloca [2 x [[S_INT_TY]]]*, +// CHECK: [[VAR_ADDR1:%.+]] = alloca [[S_INT_TY]]*, +// skip loop variables +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: [[OMP_IS_LAST1:%.+]] = alloca i{{[0-9]+}}, + +// copy from parameters to local address variables +// CHECK: store {{.+}} [[VEC_IN1]], {{.+}} [[VEC_ADDR1]], +// CHECK: store {{.+}} [[T_VAR_IN1]], {{.+}} [[T_VAR_ADDR1]], +// CHECK: store {{.+}} [[S_ARR_IN1]], {{.+}} [[S_ARR_ADDR1]], +// CHECK: store {{.+}} [[VAR_IN1]], {{.+}} [[VAR_ADDR1]], + +// prepare lastprivate targets +// CHECK-64-DAG: [[T_VAR_TGT:%.+]] = bitcast {{.+}} [[T_VAR_ADDR1]] to +// CHECK-DAG: [[VEC_TGT:%.+]] = load {{.+}}, {{.+}} [[VEC_ADDR1]], +// CHECK-DAG: [[S_ARR_TGT:%.+]] = load {{.+}}, {{.+}} [[S_ARR_ADDR1]], +// CHECK-DAG: [[VAR_TGT:%.+]] = load {{.+}}, {{.+}} [[VAR_ADDR1]], + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[TPAR_OUTL:.+]] to +// CHECK: call void @__kmpc_for_static_fini( + +// lastprivates +// CHECK: [[OMP_IS_LAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[OMP_IS_LAST1]], +// CHECK: [[IS_LAST_IT:%.+]] = icmp ne i{{[0-9]+}} [[OMP_IS_LAST_VAL]], 0 +// CHECK: br i1 [[IS_LAST_IT]], label %[[OMP_LASTPRIV_BLOCK:.+]], label %[[OMP_LASTPRIV_DONE:.+]] + +// CHECK: [[OMP_LASTPRIV_BLOCK]]: +// CHECK-64-DAG: store {{.+}}, {{.+}} [[T_VAR_TGT]], +// CHECK-32-DAG: store {{.+}}, {{.+}} [[T_VAR_ADDR1]], +// CHECK-DAG: [[VEC_TGT_BCAST:%.+]] = bitcast {{.+}} [[VEC_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_TGT_BCAST]], +// CHECK-DAG: {{.+}} = getelementptr {{.+}} [[S_ARR_TGT]], +// CHECK: call void @llvm.memcpy.{{.+}}( +// CHECK-DAG: [[VAR_ADDR_BCAST:%.+]] = bitcast {{.+}} [[VAR_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_BCAST]], +// CHECK: ret void + +// CHECK: define internal void [[TPAR_OUTL:@.+]](i{{[0-9]+}}* noalias [[GTID_ADDR1:%.+]], i{{[0-9]+}}* noalias %{{.+}}, {{.+}}, {{.+}}, [2 x i{{[0-9]+}}]*{{.+}} [[VEC_IN1:%.+]], i{{[0-9]+}}{{.+}} [[T_VAR_IN1:%.+]], [2 x [[S_INT_TY]]]*{{.+}} [[S_ARR_IN1:%.+]], [[S_INT_TY]]*{{.+}} [[VAR_IN1:%.+]]) +// skip alloca of global_tid and bound_tid, and prev lb and ub vars +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: {{.+}} = alloca i{{[0-9]+}}*, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, + +// CHECK: [[VEC_ADDR1:%.+]] = alloca [2 x i{{[0-9]+}}]*, +// CHECK: [[T_VAR_ADDR1:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[S_ARR_ADDR1:%.+]] = alloca [2 x [[S_INT_TY]]]*, +// CHECK: [[VAR_ADDR1:%.+]] = alloca [[S_INT_TY]]*, +// skip loop variables +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: [[OMP_IS_LAST1:%.+]] = alloca i{{[0-9]+}}, + +// copy from parameters to local address variables +// CHECK: store {{.+}} [[VEC_IN1]], {{.+}} [[VEC_ADDR1]], +// CHECK: store {{.+}} [[T_VAR_IN1]], {{.+}} [[T_VAR_ADDR1]], +// CHECK: store {{.+}} [[S_ARR_IN1]], {{.+}} [[S_ARR_ADDR1]], +// CHECK: store {{.+}} [[VAR_IN1]], {{.+}} [[VAR_ADDR1]], + +// prepare lastprivate targets +// CHECK-64-DAG: [[T_VAR_TGT:%.+]] = bitcast {{.+}} [[T_VAR_ADDR1]] to +// CHECK-DAG: [[VEC_TGT:%.+]] = load {{.+}}, {{.+}} [[VEC_ADDR1]], +// CHECK-DAG: [[S_ARR_TGT:%.+]] = load {{.+}}, {{.+}} [[S_ARR_ADDR1]], +// CHECK-DAG: [[VAR_TGT:%.+]] = load {{.+}}, {{.+}} [[VAR_ADDR1]], + +// CHECK: call void @__kmpc_for_static_init_4( +// skip body: code generation routine is same as distribute parallel for lastprivate +// CHECK: call void @__kmpc_for_static_fini( + +// lastprivates +// CHECK: [[OMP_IS_LAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[OMP_IS_LAST1]], +// CHECK: [[IS_LAST_IT:%.+]] = icmp ne i{{[0-9]+}} [[OMP_IS_LAST_VAL]], 0 +// CHECK: br i1 [[IS_LAST_IT]], label %[[OMP_LASTPRIV_BLOCK:.+]], label %[[OMP_LASTPRIV_DONE:.+]] + +// CHECK: [[OMP_LASTPRIV_BLOCK]]: +// CHECK-64-DAG: store {{.+}}, {{.+}} [[T_VAR_TGT]], +// CHECK-32-DAG: store {{.+}}, {{.+}} [[T_VAR_ADDR1]], +// CHECK-DAG: [[VEC_TGT_BCAST:%.+]] = bitcast {{.+}} [[VEC_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_TGT_BCAST]], +// CHECK-DAG: {{.+}} = getelementptr {{.+}} [[S_ARR_TGT]], +// CHECK: call void @llvm.memcpy.{{.+}}( +// CHECK-DAG: [[VAR_ADDR_BCAST:%.+]] = bitcast {{.+}} [[VAR_TGT]] to +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_BCAST]], +// CHECK: ret void + +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_lastprivate_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_lastprivate_messages.cpp index 156f1a66c1913..b65c22ba8a400 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_lastprivate_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_linear_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_linear_messages.cpp index 257244ede4443..3013480d6ea9f 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_linear_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_loop_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_loop_messages.cpp index a16dec190dd2a..1378ea4eb65c5 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_loop_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_loop_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s -Wno-openmp-target + +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s -Wno-openmp-target class S { int a; diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_map_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_map_messages.cpp index b230d9ab24fea..63d85ebc37b99 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_map_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_map_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_messages.cpp index 7e16c794f01f7..58e88623473f9 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_misc_messages.c b/test/OpenMP/target_teams_distribute_parallel_for_simd_misc_messages.c index 1ea02f6d2b574..7005c7b027a8e 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_misc_messages.c +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp target teams distribute parallel for simd'}} #pragma omp target teams distribute parallel for simd diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_nowait_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_nowait_messages.cpp index 4c65130b9d918..458eaf72c3403 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_nowait_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_nowait_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -verify -fopenmp %s +// RUN: %clang_cc1 -fsyntax-only -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_num_teams_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_num_teams_messages.cpp index 240d7d7b66b8a..6564f104a126c 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_num_teams_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_num_teams_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_num_threads_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_num_threads_messages.cpp index 9032996dc28d5..58fab1e8df63e 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_num_threads_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_num_threads_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_private_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_private_codegen.cpp new file mode 100644 index 0000000000000..88169197d1c61 --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_private_codegen.cpp @@ -0,0 +1,369 @@ +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix HHECK --check-prefix HCHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix HHECK --check-prefix HCHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix HHECK --check-prefix HCHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix HHECK --check-prefix HCHECK-64 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 --check-prefix HLAMBDA --check-prefix HLAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 --check-prefix HLAMBDA --check-prefix HLAMBDA-64 + +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 + +// Test target codegen - host bc file has to be created first. (no significant differences with host version of target region) +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 --check-prefix TLAMBDA --check-prefix TLAMBDA-64 + +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix SIMD-ONLY0 +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +struct St { + int a, b; + St() : a(0), b(0) {} + St(const St &st) : a(st.a + st.b), b(0) {} + ~St() {} +}; + +volatile int g = 1212; +volatile int &g1 = g; + +template <class T> +struct S { + T f; + S(T a) : f(a + g) {} + S() : f(g) {} + S(const S &s, St t = St()) : f(s.f + t.a) {} + operator T() { return T(); } + ~S() {} +}; + +// CHECK-DAG: [[S_FLOAT_TY:%.+]] = type { float } +// CHECK-DAG: [[S_INT_TY:%.+]] = type { i{{[0-9]+}} } + +template <typename T> +T tmain() { + S<T> test; + T t_var = T(); + T vec[] = {1, 2}; + S<T> s_arr[] = {1, 2}; + S<T> &var = test; +#pragma omp target teams distribute parallel for simd private(t_var, vec, s_arr, var) + for (int i = 0; i < 2; ++i) { + vec[i] = t_var; + s_arr[i] = var; + } + return T(); +} + +// HCHECK-DAG: [[TEST:@.+]] ={{.*}} global [[S_FLOAT_TY]] zeroinitializer, +S<float> test; +// HCHECK-DAG: [[T_VAR:@.+]] ={{.+}} global i{{[0-9]+}} 333, +int t_var = 333; +// HCHECK-DAG: [[VEC:@.+]] ={{.+}} global [2 x i{{[0-9]+}}] [i{{[0-9]+}} 1, i{{[0-9]+}} 2], +int vec[] = {1, 2}; +// HCHECK-DAG: [[S_ARR:@.+]] ={{.+}} global [2 x [[S_FLOAT_TY]]] zeroinitializer, +S<float> s_arr[] = {1, 2}; +// HCHECK-DAG: [[VAR:@.+]] ={{.+}} global [[S_FLOAT_TY]] zeroinitializer, +S<float> var(3); +// HCHECK-DAG: [[SIVAR:@.+]] = internal global i{{[0-9]+}} 0, + +int main() { + static int sivar; +#ifdef LAMBDA + // HLAMBDA: [[G:@.+]] = global i{{[0-9]+}} 1212, + // HLAMBDA-LABEL: @main + // HLAMBDA: call void [[OUTER_LAMBDA:@.+]]( + [&]() { + // HLAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]]( + // HLAMBDA: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 0, + // HLAMBDA: call void @[[LOFFL1:.+]]( + // HLAMBDA: ret +#pragma omp target teams distribute parallel for simd private(g, g1, sivar) + for (int i = 0; i < 2; ++i) { + // HLAMBDA: define{{.*}} internal{{.*}} void @[[LOFFL1]]() + // TLAMBDA: define{{.*}} void @[[LOFFL1:.+]]() + // LAMBDA: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}} @[[LOUTL1:.+]] to {{.+}}) + // LAMBDA: ret void + + // LAMBDA: define internal void @[[LOUTL1]]({{.+}}) + // Skip global, bound tid and loop vars + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: [[G_PRIV:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[G1_PRIV:%.+]] = alloca i{{[0-9]+}} + // LAMBDA: [[TMP:%.+]] = alloca i{{[0-9]+}}*, + // LAMBDA: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: store{{.+}} [[G1_PRIV]], {{.+}} [[TMP]], + + g = 1; + g1 = 1; + sivar = 2; + // LAMBDA: call void @__kmpc_for_static_init_4( + // LAMBDA: call void {{.+}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[LPAR_OUTL:.+]] to + // LAMBDA: call void @__kmpc_for_static_fini( + // LAMBDA: ret void + + // LAMBDA: define internal void @[[LPAR_OUTL]]({{.+}}) + // Skip global, bound tid and loop vars + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: alloca i{{[0-9]+}}, + // LAMBDA: alloca i{{[0-9]+}}, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: [[G_PRIV:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: [[G1_PRIV:%.+]] = alloca i{{[0-9]+}} + // LAMBDA: [[TMP:%.+]] = alloca i{{[0-9]+}}*, + // LAMBDA: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}}, + // LAMBDA: store{{.+}} [[G1_PRIV]], {{.+}} [[TMP]], + // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_PRIV]], + // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_PRIV]], + // LAMBDA-DAG: [[G1_REF:%.+]] = load{{.+}}, {{.+}} [[TMP]], + // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1_REF]], + // LAMBDA: call void [[INNER_LAMBDA:@.+]]( + // LAMBDA: call void @__kmpc_for_static_fini( + // LAMBDA: ret void + [&]() { + // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) + // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], + g = 2; + g1 = 2; + sivar = 4; + // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]] + + // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 + // LAMBDA: [[G_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G_PTR_REF]] + // LAMBDA: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[G_REF]] + // LAMBDA: [[G1_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 1 + // LAMBDA: [[G1_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[G1_PTR_REF]] + // LAMBDA: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[G1_REF]] + // LAMBDA: [[SIVAR_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 + // LAMBDA: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_PTR_REF]] + // LAMBDA: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIVAR_REF]] + }(); + } + }(); + return 0; +#else +#pragma omp target teams distribute parallel for simd private(t_var, vec, s_arr, var, sivar) + for (int i = 0; i < 2; ++i) { + vec[i] = t_var; + s_arr[i] = var; + sivar += i; + } + return tmain<int>(); +#endif +} + +// HCHECK: define {{.*}}i{{[0-9]+}} @main() +// HCHECK: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 0, i8** null, i8** null, {{.+}} null, {{.+}} null, i32 0, i32 0) +// HCHECK: call void @[[OFFL1:.+]]() +// HCHECK: {{%.+}} = call{{.*}} i32 @[[TMAIN_INT:.+]]() +// HCHECK: ret + +// HCHECK: define{{.*}} void @[[OFFL1]]() +// TCHECK: define{{.*}} void @[[OFFL1:.+]]() +// CHECK: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}} @[[OUTL1:.+]] to {{.+}}) +// CHECK: ret void + +// CHECK: define internal void @[[OUTL1]]({{.+}}) +// Skip global, bound tid and loop vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK-DAG: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK-DAG: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK-DAG: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]], +// CHECK-DAG: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]], +// CHECK-DAG: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK: alloca i32, + +// private(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_PTR_ALLOC:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_NEXT:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_PTR_ALLOC]]) +// CHECK-DAG: [[S_ARR_NEXT]] = getelementptr {{.+}} [[S_ARR_PTR_ALLOC]], + +// private(var) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]]) + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.+}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: define internal void @[[PAR_OUTL1]]({{.+}}) +// Skip global, bound tid and loop vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i{{[0-9]+}}, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK: {{.+}} = alloca i32, +// CHECK-DAG: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK-DAG: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK-DAG: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_FLOAT_TY]]], +// CHECK-DAG: [[VAR_PRIV:%.+]] = alloca [[S_FLOAT_TY]], +// CHECK-DAG: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK: alloca i32, + +// private(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_PTR_ALLOC:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_NEXT:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_PTR_ALLOC]]) +// CHECK-DAG: [[S_ARR_NEXT]] = getelementptr {{.+}} [[S_ARR_PTR_ALLOC]], + +// private(var) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]]) + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK-DAG: {{.+}} = {{.+}} [[T_VAR_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[VAR_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[SIVAR_PRIV]] +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// HCHECK: define{{.*}} i{{[0-9]+}} @[[TMAIN_INT]]() +// HCHECK: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 0, +// HCHECK: call void @[[TOFFL1:.+]]() +// HCHECK: ret + +// HCHECK: define {{.*}}void @[[TOFFL1]]() +// TCHECK: define weak void @[[TOFFL1:.+]]() +// CHECK: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}} @[[TOUTL1:.+]] to {{.+}}) +// CHECK: ret void + +// CHECK: define internal void @[[TOUTL1]]({{.+}}) +// Skip global, bound tid and loop vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i32, +// CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_INT_TY]]], +// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]], +// CHECK: [[TMP:%.+]] = alloca [[S_INT_TY]]*, + +// private(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_PTR_ALLOC:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_NEXT:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_PTR_ALLOC]]) +// CHECK-DAG: [[S_ARR_NEXT]] = getelementptr {{.+}} [[S_ARR_PTR_ALLOC]], + +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_PTR_ALLOC:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_NEXT:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_PTR_ALLOC]]) +// CHECK-DAG: [[S_ARR_NEXT]] = getelementptr {{.+}} [[S_ARR_PTR_ALLOC]], + +// private(var) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]]) +// CHECK-DAG: store{{.+}} [[VAR_PRIV]], {{.+}} [[TMP]] + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.+}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[TPAR_OUTL1:.+]] to +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + +// CHECK: define internal void @[[TPAR_OUTL1]]({{.+}}) +// Skip global, bound tid and loop vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// prev lb and ub +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// iter variables +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i32, +// CHECK: [[T_VAR_PRIV:%.+]] = alloca i{{[0-9]+}}, +// CHECK: [[VEC_PRIV:%.+]] = alloca [2 x i{{[0-9]+}}], +// CHECK: [[S_ARR_PRIV:%.+]] = alloca [2 x [[S_INT_TY]]], +// CHECK: [[VAR_PRIV:%.+]] = alloca [[S_INT_TY]], +// CHECK: [[TMP:%.+]] = alloca [[S_INT_TY]]*, + +// private(s_arr) +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_PTR_ALLOC:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_NEXT:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_PTR_ALLOC]]) +// CHECK-DAG: [[S_ARR_NEXT]] = getelementptr {{.+}} [[S_ARR_PTR_ALLOC]], + +// CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], +// CHECK-DAG: [[S_ARR_PTR_ALLOC:%.+]] = phi{{.+}} [ [[S_ARR_PRIV_BGN]], {{.+}} ], [ [[S_ARR_NEXT:%.+]], {{.+}} ] +// CHECK-DAG: call void @{{.+}}({{.+}} [[S_ARR_PTR_ALLOC]]) +// CHECK-DAG: [[S_ARR_NEXT]] = getelementptr {{.+}} [[S_ARR_PTR_ALLOC]], + +// private(var) +// CHECK-DAG: call void @{{.+}}({{.+}} [[VAR_PRIV]]) +// CHECK-DAG: store{{.+}} [[VAR_PRIV]], {{.+}} [[TMP]] + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK-DAG: {{.+}} = {{.+}} [[T_VAR_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] +// CHECK-DAG: {{.+}} = {{.+}} [[TMP]] +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: ret void + + +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_private_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_private_messages.cpp index 20e185003ba0e..f1a8e0bfc7c3f 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_private_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_proc_bind_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_proc_bind_codegen.cpp new file mode 100644 index 0000000000000..40bf51f340280 --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_proc_bind_codegen.cpp @@ -0,0 +1,93 @@ +// add -fopenmp-targets + +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +typedef __INTPTR_TYPE__ intptr_t; + +// CHECK-DAG: [[IDENT_T_TY:%.+]] = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" +// CHECK-DAG: [[DEF_LOC_2:@.+]] = private unnamed_addr constant [[IDENT_T_TY]] { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } + +void foo(); + +struct S { + intptr_t a, b, c; + S(intptr_t a) : a(a) {} + operator char() { return a; } + ~S() {} +}; + +template <typename T> +T tmain() { +#pragma omp target teams distribute parallel for simd proc_bind(master) + for(int i = 0; i < 1000; i++) {} + return T(); +} + +int main() { + // CHECK-LABEL: @main +#pragma omp target teams distribute parallel for simd proc_bind(spread) + for(int i = 0; i < 1000; i++) {} +#pragma omp target teams distribute parallel for simd proc_bind(close) + for(int i = 0; i < 1000; i++) {} + return tmain<int>(); +} + +// CHECK: call {{.*}}@__tgt_target_teams({{.+}}) +// CHECK: call void [[OFFL1:@.+]]() +// CHECK: call {{.*}}@__tgt_target_teams({{.+}}) +// CHECK: call void [[OFFL2:@.+]]() +// CHECK: [[CALL_RET:%.+]] = call{{.+}} i32 [[TMAIN:@.+]]() +// CHECK: ret i32 [[CALL_RET]] + +// CHECK: define{{.+}} void [[OFFL1]]( +// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, {{.+}}, {{.+}}* [[OMP_OUTLINED_1:@.+]] to {{.+}}) + +// CHECK: define{{.+}} [[OMP_OUTLINED_1]](i32* {{.+}} [[GTID_IN:%.+]], +// CHECK: [[GTID_ADDR:%.+]] = alloca i32*, +// CHECK: store i32* [[GTID_IN]], i32** [[GTID_ADDR]], +// CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], +// CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], +// CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 4) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: ret void + +// CHECK: define{{.+}} [[OFFL2]]() +// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, {{.+}}, {{.+}}* [[OMP_OUTLINED_1:@.+]] to {{.+}}) + +// CHECK: define{{.+}} [[OMP_OUTLINED_1]](i32* {{.+}} [[GTID_IN:%.+]], +// CHECK: [[GTID_ADDR:%.+]] = alloca i32*, +// CHECK: store i32* [[GTID_IN]], i32** [[GTID_ADDR]], +// CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], +// CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], +// CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 3) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: ret void + +// CHECK: define{{.+}} [[TMAIN]]() +// CHECK: call {{.*}}@__tgt_target_teams({{.+}}) +// CHECK: call void [[OFFL3:@.+]]() + +// CHECK: define{{.+}} [[OFFL3]]() +// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, {{.+}}, {{.+}}* [[OMP_OUTLINED_3:@.+]] to {{.+}}) + +// CHECK: define{{.+}} [[OMP_OUTLINED_3]](i32* {{.+}} [[GTID_IN:%.+]], +// CHECK: [[GTID_ADDR:%.+]] = alloca i32*, +// CHECK: store i32* [[GTID_IN]], i32** [[GTID_ADDR]], +// CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], +// CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], +// CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 2) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: ret void +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_proc_bind_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_proc_bind_messages.cpp index 14714124bc4e5..3ae5268667d34 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_proc_bind_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_proc_bind_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -o - %s + void foo(); template <class T, typename S, int N> diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_reduction_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_reduction_codegen.cpp new file mode 100644 index 0000000000000..e938a7746313d --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_reduction_codegen.cpp @@ -0,0 +1,353 @@ +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 + +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +template <typename T> +T tmain() { + T t_var = T(); + T vec[] = {1, 2}; +#pragma omp target teams distribute parallel for simd reduction(+: t_var) + for (int i = 0; i < 2; ++i) { + t_var += (T) i; + } + return T(); +} + +int main() { + static int sivar; +#ifdef LAMBDA + // LAMBDA: [[RED_VAR:@.+]] = common global [8 x {{.+}}] zeroinitializer + + // LAMBDA-LABEL: @main + // LAMBDA: call void [[OUTER_LAMBDA:@.+]]( + [&]() { + // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]]( + // LAMBDA: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 1, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i32 0, i32 0) + // LAMBDA: call void @[[LOFFL1:.+]]( + // LAMBDA: ret +#pragma omp target teams distribute parallel for simd reduction(+: sivar) + for (int i = 0; i < 2; ++i) { + // LAMBDA: define{{.*}} internal{{.*}} void @[[LOFFL1]](i{{64|32}}*{{.*}} [[SIVAR_ARG:%.+]]) + // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{.+}}, + // LAMBDA: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]], + // LAMBDA: [[SIVAR_PAR:%.+]] = load{{.+}}, {{.+}} [[SIVAR_ADDR]], + // LAMBDA: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[LOUTL1:.+]] to {{.+}}, {{.+}} [[SIVAR_PAR]]) + // LAMBDA: ret void + + // LAMBDA: define internal void @[[LOUTL1]]({{.+}}, {{.+}}, {{.+}} [[SIVAR_ARG:%.+]]) + // Skip global and bound tid vars + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{.+}}, + // LAMBDA: [[SIVAR_PRIV:%.+]] = alloca i{{.+}}, + // LAMBDA: [[RED_LIST:%.+]] = alloca [1 x {{.+}}], + // LAMBDA: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]], + // LAMBDA: [[SIVAR_REF:%.+]] = load{{.+}}, {{.+}} [[SIVAR_ADDR]] + // LAMBDA: store{{.+}} 0, {{.+}} [[SIVAR_PRIV]], + + // LAMBDA: call void @__kmpc_for_static_init_4( + // LAMBDA: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[LPAR_OUTL:.+]] to + // LAMBDA: call void @__kmpc_for_static_fini( + // LAMBDA: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]], + // LAMBDA: [[SIVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[SIVAR_PRIV]] to + // LAMBDA: store{{.+}} [[SIVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]], + // LAMBDA: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to + // LAMBDA: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce_nowait({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]]) + // LAMBDA: switch{{.+}} [[K_RED_RET]], label{{.+}} [ + // LAMBDA: {{.+}}, label %[[CASE1:.+]] + // LAMBDA: {{.+}}, label %[[CASE2:.+]] + // LAMBDA: ] + // LAMBDA: [[CASE1]]: + // LAMBDA-DAG: [[SIVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_REF]], + // LAMBDA-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], + // LAMBDA-DAG: [[SIVAR_INC:%.+]] = add{{.+}} [[SIVAR_VAL]], [[SIVAR_PRIV_VAL]] + // LAMBDA: store{{.+}} [[SIVAR_INC]], {{.+}} [[SIVAR_REF]], + // LAMBDA: call void @__kmpc_end_reduce_nowait({{.+}}, {{.+}}, {{.+}} [[RED_VAR]]) + // LAMBDA: br + // LAMBDA: [[CASE2]]: + // LAMBDA-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], + // LAMBDA-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[SIVAR_REF]], {{.+}} [[SIVAR_PRIV_VAL]] + // LAMBDA: br + + // LAMBDA: define internal void @[[LPAR_OUTL]]({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[SIVAR_ARG:%.+]]) + + // Skip global and bound tid vars, and prev lb and ub vars + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: {{.+}} = alloca i32*, + // LAMBDA: alloca i{{[0-9]+}}, + // LAMBDA: alloca i{{[0-9]+}}, + // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{.+}}, + // skip loop vars + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: alloca i32, + // LAMBDA: [[SIVAR_PRIV:%.+]] = alloca i{{.+}}, + // LAMBDA: [[RED_LIST:%.+]] = alloca [1 x {{.+}}], + // LAMBDA: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]], + // LAMBDA: [[SIVAR_REF:%.+]] = load {{.+}} [[SIVAR_ADDR]] + // LAMBDA: store{{.+}} 0, {{.+}} [[SIVAR_PRIV]], + + // LAMBDA: call void @__kmpc_for_static_init_4( + // LAMBDA: store{{.+}}, {{.+}} [[SIVAR_PRIV]], + // LAMBDA: call void [[INNER_LAMBDA:@.+]]( + // LAMBDA: call void @__kmpc_for_static_fini( + // LAMBDA: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]], + // LAMBDA: [[SIVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[SIVAR_PRIV]] to + // LAMBDA: store{{.+}} [[SIVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]], + // LAMBDA: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to + // LAMBDA: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce_nowait({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]]) + // LAMBDA: switch{{.+}} [[K_RED_RET]], label{{.+}} [ + // LAMBDA: {{.+}}, label %[[CASE1:.+]] + // LAMBDA: {{.+}}, label %[[CASE2:.+]] + // LAMBDA: ] + // LAMBDA: [[CASE1]]: + // LAMBDA-64-DAG: [[SIVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_REF]], + // LAMBDA-32-DAG: [[SIVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_ADDR]], + // LAMBDA-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], + // LAMBDA-DAG: [[SIVAR_INC:%.+]] = add{{.+}} [[SIVAR_VAL]], [[SIVAR_PRIV_VAL]] + // LAMBDA: store{{.+}} [[SIVAR_INC]], {{.+}} [[SIVAR_REF]], + // LAMBDA: call void @__kmpc_end_reduce_nowait({{.+}}, {{.+}}, {{.+}} [[RED_VAR]]) + // LAMBDA: br + // LAMBDA: [[CASE2]]: + // LAMBDA-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], + // LAMBDA-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[SIVAR_REF]], {{.+}} [[SIVAR_PRIV_VAL]] + // LAMBDA: br + + sivar += i; + + [&]() { + // LAMBDA: define {{.+}} void [[INNER_LAMBDA]](%{{.+}}* [[ARG_PTR:%.+]]) + // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]], + + sivar += 4; + // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]] + + // LAMBDA: [[SIVAR_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 + // LAMBDA: [[SIVAR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[SIVAR_PTR_REF]] + // LAMBDA: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR_REF]] + // LAMBDA: [[SIVAR_INC:%.+]] = add{{.+}} [[SIVAR_VAL]], 4 + // LAMBDA: store i{{[0-9]+}} [[SIVAR_INC]], i{{[0-9]+}}* [[SIVAR_REF]] + }(); + } + }(); + return 0; +#else +#pragma omp target teams distribute parallel for simd reduction(+: sivar) + for (int i = 0; i < 2; ++i) { + sivar += i; + } + return tmain<int>(); +#endif +} + +// CHECK: [[RED_VAR:@.+]] = common global [8 x {{.+}}] zeroinitializer + +// CHECK: define {{.*}}i{{[0-9]+}} @main() +// CHECK: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 1, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i32 0, i32 0) +// CHECK: call void @[[OFFL1:.+]](i{{64|32}}* @{{.+}}) +// CHECK: {{%.+}} = call{{.*}} i32 @[[TMAIN_INT:.+]]() +// CHECK: ret + +// CHECK: define{{.*}} void @[[OFFL1]](i{{64|32}}*{{.*}} [[SIVAR_ARG:%.+]]) +// CHECK: [[SIVAR_ADDR:%.+]] = alloca i{{.+}}, +// CHECK: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]], +// CHECK: [[SIVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_ADDR]] +// CHECK: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL1:.+]] to {{.+}}, {{.+}} [[SIVAR_VAL]]) +// CHECK: ret void + +// CHECK: define internal void @[[OUTL1]]({{.+}}, {{.+}}, {{.+}} [[SIVAR_ARG:%.+]]) +// Skip global and bound tid vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: [[SIVAR_ADDR:%.+]] = alloca i{{.+}}, +// CHECK: [[SIVAR_PRIV:%.+]] = alloca i{{.+}}, +// CHECK: [[RED_LIST:%.+]] = alloca [1 x {{.+}}], +// CHECK: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]], +// CHECK: [[SIVAR_REF:%.+]] = load{{.+}} [[SIVAR_ADDR]], +// CHECK: store{{.+}} 0, {{.+}} [[SIVAR_PRIV]], + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL:.+]] to +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]], +// CHECK: [[SIVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[SIVAR_PRIV]] to +// CHECK: store{{.+}} [[SIVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]], +// CHECK: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to +// CHECK: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce_nowait({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]]) +// CHECK: switch{{.+}} [[K_RED_RET]], label{{.+}} [ +// CHECK: {{.+}}, label %[[CASE1:.+]] +// CHECK: {{.+}}, label %[[CASE2:.+]] +// CHECK: ] +// CHECK: [[CASE1]]: +// CHECK-DAG: [[SIVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_REF]], +// CHECK-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], +// CHECK-DAG: [[SIVAR_INC:%.+]] = add{{.+}} [[SIVAR_VAL]], [[SIVAR_PRIV_VAL]] +// CHECK: store{{.+}} [[SIVAR_INC]], {{.+}} [[SIVAR_REF]], +// CHECK: call void @__kmpc_end_reduce_nowait({{.+}}, {{.+}}, {{.+}} [[RED_VAR]]) +// CHECK: br +// CHECK: [[CASE2]]: +// CHECK-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], +// CHECK-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[SIVAR_REF]], {{.+}} [[SIVAR_PRIV_VAL]] +// CHECK: br + +// CHECK: define internal void @[[PAR_OUTL]]({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[SIVAR_ARG:%.+]]) +// Skip global and bound tid vars, and prev lb and ub +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: [[SIVAR_ADDR:%.+]] = alloca i{{.+}}, +// skip loop vars +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: [[SIVAR_PRIV:%.+]] = alloca i{{.+}}, +// CHECK: [[RED_LIST:%.+]] = alloca [1 x {{.+}}], +// CHECK: store{{.+}} [[SIVAR_ARG]], {{.+}} [[SIVAR_ADDR]], +// CHECK-64: [[SIVAR_REF:%.+]] = load {{.+}} [[SIVAR_ADDR]], +// CHECK: store{{.+}} 0, {{.+}} [[SIVAR_PRIV]], + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: store{{.+}}, {{.+}} [[SIVAR_PRIV]], +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]], +// CHECK: [[SIVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[SIVAR_PRIV]] to +// CHECK: store{{.+}} [[SIVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]], +// CHECK: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to +// CHECK: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce_nowait({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]]) +// CHECK: switch{{.+}} [[K_RED_RET]], label{{.+}} [ +// CHECK: {{.+}}, label %[[CASE1:.+]] +// CHECK: {{.+}}, label %[[CASE2:.+]] +// CHECK: ] +// CHECK: [[CASE1]]: +// CHECK-DAG: [[SIVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_REF]], +// CHECK-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], +// CHECK-DAG: [[SIVAR_INC:%.+]] = add{{.+}} [[SIVAR_VAL]], [[SIVAR_PRIV_VAL]] +// CHECK: store{{.+}} [[SIVAR_INC]], {{.+}} [[SIVAR_REF]], +// CHECK: call void @__kmpc_end_reduce_nowait({{.+}}, {{.+}}, {{.+}} [[RED_VAR]]) +// CHECK: br +// CHECK: [[CASE2]]: +// CHECK-DAG: [[SIVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[SIVAR_PRIV]], +// CHECK-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[SIVAR_REF]], {{.+}} [[SIVAR_PRIV_VAL]] +// CHECK: br + +// CHECK: define{{.*}} i{{[0-9]+}} @[[TMAIN_INT]]() +// CHECK: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 1, +// CHECK: call void @[[TOFFL1:.+]]({{.+}}) +// CHECK: ret + +// CHECK: define{{.*}} void @[[TOFFL1]](i{{64|32}}*{{.*}} [[TVAR_ARG:%.+]]) +// CHECK: [[TVAR_ADDR:%.+]] = alloca i{{.+}}, +// CHECK: store{{.+}} [[TVAR_ARG]], {{.+}} [[TVAR_ADDR]], +// CHECK: [[TVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_ADDR]] +// CHECK: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[TOUTL1:.+]] to {{.+}}, {{.+}} [[TVAR_VAL]]) +// CHECK: ret void + +// CHECK: define internal void @[[TOUTL1]]({{.+}}, {{.+}}, {{.+}} [[TVAR_ARG:%.+]]) +// Skip global and bound tid vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: [[TVAR_ADDR:%.+]] = alloca i{{.+}}, +// CHECK: [[TVAR_PRIV:%.+]] = alloca i{{.+}}, +// CHECK: [[RED_LIST:%.+]] = alloca [1 x {{.+}}], +// CHECK: store{{.+}} [[TVAR_ARG]], {{.+}} [[TVAR_ADDR]], +// CHECK: [[TVAR_REF:%.+]] = load {{.+}} [[TVAR_ADDR]], +// CHECK: store{{.+}} 0, {{.+}} [[TVAR_PRIV]], + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[TPAR_OUTL:.+]] to +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]], +// CHECK: [[TVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[TVAR_PRIV]] to +// CHECK: store{{.+}} [[TVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]], +// CHECK: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to +// CHECK: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce_nowait({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]]) +// CHECK: switch{{.+}} [[K_RED_RET]], label{{.+}} [ +// CHECK: {{.+}}, label %[[CASE1:.+]] +// CHECK: {{.+}}, label %[[CASE2:.+]] +// CHECK: ] +// CHECK: [[CASE1]]: +// CHECK-DAG: [[TVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_REF]], +// CHECK-DAG: [[TVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_PRIV]], +// CHECK-DAG: [[TVAR_INC:%.+]] = add{{.+}} [[TVAR_VAL]], [[TVAR_PRIV_VAL]] +// CHECK: store{{.+}} [[TVAR_INC]], {{.+}} [[TVAR_REF]], +// CHECK: call void @__kmpc_end_reduce_nowait({{.+}}, {{.+}}, {{.+}} [[RED_VAR]]) +// CHECK: br +// CHECK: [[CASE2]]: +// CHECK-DAG: [[TVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_PRIV]], +// CHECK-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[TVAR_REF]], {{.+}} [[TVAR_PRIV_VAL]] +// CHECK: br + +// CHECK: define internal void @[[TPAR_OUTL]]({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[TVAR_ARG:%.+]]) +// Skip global and bound tid vars, and prev lb and ub vars +// CHECK: {{.+}} = alloca i32*, +// CHECK: {{.+}} = alloca i32*, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: alloca i{{[0-9]+}}, +// CHECK: [[TVAR_ADDR:%.+]] = alloca i{{.+}}, +// skip loop vars +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: alloca i32, +// CHECK: [[TVAR_PRIV:%.+]] = alloca i{{.+}}, +// CHECK: [[RED_LIST:%.+]] = alloca [1 x {{.+}}], +// CHECK: store{{.+}} [[TVAR_ARG]], {{.+}} [[TVAR_ADDR]], +// CHECK: [[TVAR_REF:%.+]] = load {{.+}} [[TVAR_ADDR]], +// CHECK: store{{.+}} 0, {{.+}} [[TVAR_PRIV]], + +// CHECK: call void @__kmpc_for_static_init_4( +// CHECK: store{{.+}}, {{.+}} [[TVAR_PRIV]], +// CHECK: call void @__kmpc_for_static_fini( +// CHECK: [[RED_LIST_GEP:%.+]] = getelementptr{{.+}} [[RED_LIST]], +// CHECK: [[TVAR_PRIV_CAST:%.+]] = bitcast{{.+}} [[TVAR_PRIV]] to +// CHECK: store{{.+}} [[TVAR_PRIV_CAST]], {{.+}} [[RED_LIST_GEP]], +// CHECK: [[RED_LIST_BCAST:%.+]] = bitcast{{.+}} [[RED_LIST]] to +// CHECK: [[K_RED_RET:%.+]] = call{{.+}} @__kmpc_reduce_nowait({{.+}}, {{.+}}, {{.+}}, {{.+}}, {{.+}} [[RED_LIST_BCAST]], {{.+}} [[RED_FUN:@.+]], {{.+}} [[RED_VAR]]) +// CHECK: switch{{.+}} [[K_RED_RET]], label{{.+}} [ +// CHECK: {{.+}}, label %[[CASE1:.+]] +// CHECK: {{.+}}, label %[[CASE2:.+]] +// CHECK: ] +// CHECK: [[CASE1]]: +// CHECK-DAG: [[TVAR_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_REF]], +// CHECK-DAG: [[TVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_PRIV]], +// CHECK-DAG: [[TVAR_INC:%.+]] = add{{.+}} [[TVAR_VAL]], [[TVAR_PRIV_VAL]] +// CHECK: store{{.+}} [[TVAR_INC]], {{.+}} [[TVAR_REF]], +// CHECK: call void @__kmpc_end_reduce_nowait({{.+}}, {{.+}}, {{.+}} [[RED_VAR]]) +// CHECK: br +// CHECK: [[CASE2]]: +// CHECK-DAG: [[TVAR_PRIV_VAL:%.+]] = load{{.+}}, {{.+}} [[TVAR_PRIV]], +// CHECK-DAG: [[ATOMIC_RES:%.+]] = atomicrmw add{{.+}} [[TVAR_REF]], {{.+}} [[TVAR_PRIV_VAL]] +// CHECK: br +#endif diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_reduction_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_reduction_messages.cpp index 63744c77fb3a7..8cd4acd878d04 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_reduction_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_safelen_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_safelen_messages.cpp index beb3c1e845809..f068e6b2e475d 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_safelen_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_safelen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_schedule_codegen.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_schedule_codegen.cpp new file mode 100644 index 0000000000000..799769739d976 --- /dev/null +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_schedule_codegen.cpp @@ -0,0 +1,400 @@ +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// Test host codegen. +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64 +// RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} + +#ifdef CK1 + +template <typename T, int X, long long Y> +struct SS{ + T a[X]; + float b; + // CK1: define {{.*}}i32 @{{.+}}foo{{.+}}( + int foo(void) { + + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL1:.+]]( + #pragma omp target teams distribute parallel for simd + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL2:.+]]( + #pragma omp target teams distribute parallel for simd schedule(static) + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL3:.+]]( + #pragma omp target teams distribute parallel for simd schedule(static, X/2) + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL4:.+]]( + #pragma omp target teams distribute parallel for simd schedule(dynamic) + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + + // CK1: call i32 @__tgt_target_teams( + // CK1: call void @[[OFFL5:.+]]( + #pragma omp target teams distribute parallel for simd schedule(dynamic, X/2) + for(int i = 0; i < X; i++) { + a[i] = (T)0; + } + + // CK1: define internal void @[[OFFL1]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL1:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL1]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL1]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[OFFL2]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL2:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL2]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL2:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL2]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[OFFL3]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL3:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL3]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL3:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL3]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 33, + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[OFFL4]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL4:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL4]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL4:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL4]]({{.+}}) + // CK1: call void @__kmpc_dispatch_init_4({{.+}}, {{.+}}, i32 35, + // CK1: call {{.+}} @__kmpc_dispatch_next_4( + // CK1: ret void + + // CK1: define internal void @[[OFFL5]]( + // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL5:.+]] to {{.+}}, + // CK1: ret void + + // CK1: define internal void @[[OUTL5]]({{.+}}) + // CK1: call void @__kmpc_for_static_init_4( + // CK1: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL5:.+]] to + // CK1: call void @__kmpc_for_static_fini( + // CK1: ret void + + // CK1: define internal void @[[PAR_OUTL5]]({{.+}}) + // CK1: call void @__kmpc_dispatch_init_4({{.+}}, {{.+}}, i32 35, + // CK1: call {{.+}} @__kmpc_dispatch_next_4( + // CK1: ret void + + return a[0]; + } +}; + +int teams_template_struct(void) { + SS<int, 123, 456> V; + return V.foo(); + +} +#endif // CK1 + +// Test host codegen. +// RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64 +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64 +// RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix SIMD-ONLY +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix SIMD-ONLY +// SIMD-ONLY-NOT: {{__kmpc|__tgt}} +#ifdef CK2 + +template <typename T, int n> +int tmain(T argc) { + T a[n]; + int m = 10; +#pragma omp target teams distribute parallel for simd + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } +#pragma omp target teams distribute parallel for simd schedule(static) + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } +#pragma omp target teams distribute parallel for simd schedule(static, m) + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } +#pragma omp target teams distribute parallel for simd schedule(dynamic) + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } +#pragma omp target teams distribute parallel for simd schedule(dynamic, m) + for(int i = 0; i < n; i++) { + a[i] = (T)0; + } + return 0; +} + +int main (int argc, char **argv) { + int n = 100; + int a[n]; + int m = 10; +#pragma omp target teams distribute parallel for simd + for(int i = 0; i < n; i++) { + a[i] = 0; + } +#pragma omp target teams distribute parallel for simd dist_schedule(static) + for(int i = 0; i < n; i++) { + a[i] = 0; + } +#pragma omp target teams distribute parallel for simd dist_schedule(static, m) + for(int i = 0; i < n; i++) { + a[i] = 0; + } +#pragma omp target teams distribute parallel for simd schedule(dynamic) + for(int i = 0; i < n; i++) { + a[i] = 0; + } +#pragma omp target teams distribute parallel for simd schedule(dynamic, m) + for(int i = 0; i < n; i++) { + a[i] = 0; + } + return tmain<int, 10>(argc); +} + +// CK2: define {{.*}}i32 @{{[^,]+}}(i{{.+}}{{.+}} %[[ARGC:.+]], {{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL1:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL2:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL3:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL4:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFL5:.+]]({{.+}}) +// CK2: {{%.+}} = call{{.*}} i32 @[[TMAIN:.+]]({{.+}}) +// CK2: ret + +// CK2: define {{.*}}void @[[OFFL1]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[OUTL1:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL1:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFL2]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[OUTL2:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL2:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFL3]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 4, {{.+}} @[[OUTL3:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL3:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34 +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFL4]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[OUTL4:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL4]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL4:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL4]]({{.+}}) +// CK2: call void @__kmpc_dispatch_init_4({{.+}}, {{.+}}, i32 35, +// CK2: call {{.+}} @__kmpc_dispatch_next_4( +// CK2: ret void + + +// CK2: define {{.*}}void @[[OFFL5]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[OUTL5:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTL5]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTL5:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTL5]]({{.+}}) +// CK2: call void @__kmpc_dispatch_init_4({{.+}}, {{.+}}, i32 35, +// CK2: call {{.+}} @__kmpc_dispatch_next_4( +// CK2: ret void + +// CK2: define {{.*}}i32 @[[TMAIN]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT1:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT2:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT3:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT4:.+]]({{.+}}) +// CK2: call i32 @__tgt_target_teams( +// CK2: call void @[[OFFLT5:.+]]({{.+}}) +// CK2: ret +// CK2-NEXT: } + +// CK2: define {{.*}}void @[[OFFLT1]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTLT1:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT1:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT1]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFLT2]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTLT2:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT2:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT2]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 34, +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFLT3]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[OUTLT3:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT3:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT3]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, i32 33, +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFLT4]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[OUTLT4:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT4]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT4:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT4]]({{.+}}) +// CK2: call void @__kmpc_dispatch_init_4({{.+}}, {{.+}}, i32 35, +// CK2: call {{.+}} @__kmpc_dispatch_next_4( +// CK2: ret void + +// CK2: define {{.*}}void @[[OFFLT5]]({{.+}}) +// CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 {{.+}}, {{.+}} @[[OUTLT5:.+]] to {{.+}}, +// CK2: ret void + +// CK2: define internal void @[[OUTLT5]]({{.+}}) +// CK2: call void @__kmpc_for_static_init_4( +// CK2: call void {{.*}} @__kmpc_fork_call({{.+}}, {{.+}}, {{.+}} @[[PAR_OUTLT5:.+]] to +// CK2: call void @__kmpc_for_static_fini( +// CK2: ret void + +// CK2: define internal void @[[PAR_OUTLT5]]({{.+}}) +// CK2: call void @__kmpc_dispatch_init_4({{.+}}, {{.+}}, i32 35, +// CK2: call {{.+}} @__kmpc_dispatch_next_4( +// CK2: ret void + +#endif // CK2 +#endif // #ifndef HEADER diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_schedule_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_schedule_messages.cpp index 773a148f9bba9..c24a543184281 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_schedule_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_shared_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_shared_messages.cpp index cbef07a4d0697..f59ecbd58378d 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_shared_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_shared_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_simdlen_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_simdlen_messages.cpp index c2a82d5bfc28f..d5f714c9c3088 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_simdlen_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_simdlen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_simd_thread_limit_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_simd_thread_limit_messages.cpp index 159d1b0f518df..ea8b1a77513e5 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_simd_thread_limit_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_simd_thread_limit_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_parallel_for_thread_limit_messages.cpp b/test/OpenMP/target_teams_distribute_parallel_for_thread_limit_messages.cpp index 1db07262931ee..bd04e934a2451 100644 --- a/test/OpenMP/target_teams_distribute_parallel_for_thread_limit_messages.cpp +++ b/test/OpenMP/target_teams_distribute_parallel_for_thread_limit_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_private_codegen.cpp b/test/OpenMP/target_teams_distribute_private_codegen.cpp index 7c88c01c4ab33..f52d3b040f04b 100644 --- a/test/OpenMP/target_teams_distribute_private_codegen.cpp +++ b/test/OpenMP/target_teams_distribute_private_codegen.cpp @@ -1,13 +1,26 @@ -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER diff --git a/test/OpenMP/target_teams_distribute_private_messages.cpp b/test/OpenMP/target_teams_distribute_private_messages.cpp index fcbd4f87162a8..df7b68919ca55 100644 --- a/test/OpenMP/target_teams_distribute_private_messages.cpp +++ b/test/OpenMP/target_teams_distribute_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_reduction_codegen.cpp b/test/OpenMP/target_teams_distribute_reduction_codegen.cpp index 188b753064537..704bf00120168 100644 --- a/test/OpenMP/target_teams_distribute_reduction_codegen.cpp +++ b/test/OpenMP/target_teams_distribute_reduction_codegen.cpp @@ -5,10 +5,23 @@ // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/target_teams_distribute_reduction_messages.cpp b/test/OpenMP/target_teams_distribute_reduction_messages.cpp index 3cfb08651a612..38878fc74c744 100644 --- a/test/OpenMP/target_teams_distribute_reduction_messages.cpp +++ b/test/OpenMP/target_teams_distribute_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_shared_messages.cpp b/test/OpenMP/target_teams_distribute_shared_messages.cpp index 13f4d471f1d12..37e0aa1b5b6ef 100644 --- a/test/OpenMP/target_teams_distribute_shared_messages.cpp +++ b/test/OpenMP/target_teams_distribute_shared_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_aligned_messages.cpp b/test/OpenMP/target_teams_distribute_simd_aligned_messages.cpp index d91cfa9b54ae3..2ec53ee99e09c 100644 --- a/test/OpenMP/target_teams_distribute_simd_aligned_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_aligned_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp %s +// RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp-simd %s + struct B { static int ib[20]; // expected-note 0 {{'B::ib' declared here}} static constexpr int bfoo() { return 8; } diff --git a/test/OpenMP/target_teams_distribute_simd_ast_print.cpp b/test/OpenMP/target_teams_distribute_simd_ast_print.cpp index 31b9ad0c5fc13..a96863572a0f4 100644 --- a/test/OpenMP/target_teams_distribute_simd_ast_print.cpp +++ b/test/OpenMP/target_teams_distribute_simd_ast_print.cpp @@ -1,6 +1,10 @@ -// RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp -ast-print %s -Wno-openmp-target | FileCheck %s +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print -Wno-openmp-target | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s -Wno-openmp-target | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print -Wno-openmp-target | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -103,7 +107,7 @@ T tmain(T argc) { #pragma omp target teams distribute simd for (int i=0; i < 2; ++i) a = 2; -// CHECK: #pragma omp target teams distribute simd +// CHECK: #pragma omp target teams distribute simd{{$}} // CHECK-NEXT: for (int i = 0; i < 2; ++i) // CHECK-NEXT: a = 2; #pragma omp target teams distribute simd private(argc, b), firstprivate(c, d), collapse(2) diff --git a/test/OpenMP/target_teams_distribute_simd_codegen.cpp b/test/OpenMP/target_teams_distribute_simd_codegen.cpp index 7fb76d9198db6..c400e0867154e 100644 --- a/test/OpenMP/target_teams_distribute_simd_codegen.cpp +++ b/test/OpenMP/target_teams_distribute_simd_codegen.cpp @@ -1,28 +1,46 @@ // Test host codegen. -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-64 // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc -// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s -// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER -// CHECK-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CHECK-DAG: [[TT:%.+]] = type { i64, i8 } // CHECK-DAG: [[S1:%.+]] = type { double } @@ -39,33 +57,33 @@ // sizes. // CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 2, i[[SZ]] 4, i[[SZ]] 4] -// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 288] +// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 800] // CHECK-DAG: [[SIZET2:@.+]] = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 2] -// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [1 x i64] [i64 288] +// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [1 x i64] [i64 800] // CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2] -// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 288, i64 288] -// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [9 x i64] [i64 288, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547] -// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [5 x i64] [i64 547, i64 288, i64 288, i64 288, i64 547] +// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 800] +// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [9 x i64] [i64 800, i64 547, i64 288, i64 547, i64 547, i64 288, i64 288, i64 547, i64 547] +// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [6 x i64] [i64 32, i64 281474976711171, i64 800, i64 288, i64 288, i64 547] // CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [5 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 1, i[[SZ]] 40] -// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [5 x i64] [i64 288, i64 288, i64 288, i64 288, i64 547] +// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [5 x i64] [i64 800, i64 800, i64 800, i64 800, i64 547] // CHECK-DAG: [[SIZET7:@.+]] = private unnamed_addr constant [3 x i[[SZ]]] [i[[SZ]] 4, i[[SZ]] 2, i[[SZ]] 40] -// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [3 x i64] [i64 288, i64 288, i64 547] -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 - -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK-NOT: @{{.+}} = constant [[ENTTY]] +// CHECK-DAG: [[MAPT7:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 547] +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] // Check if offloading descriptor is created. // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] @@ -76,7 +94,7 @@ // CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) // Check target registration is registered as a Ctor. -// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] template<typename tx, typename ty> @@ -202,13 +220,13 @@ int foo(int n) { // CHECK-32: store i32 [[A_VAL]], i32* [[A_CADDR:%.+]], // CHECK-32: [[A_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CADDR]], + // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 + // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] + // CHECK: [[TRY]] // CHECK: [[BNSIZE:%.+]] = mul nuw i[[SZ]] [[VLA0:%.+]], 4 // CHECK: [[CNELEMSIZE2:%.+]] = mul nuw i[[SZ]] 5, [[VLA1:%.+]] // CHECK: [[CNSIZE:%.+]] = mul nuw i[[SZ]] [[CNELEMSIZE2]], 8 - // CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 20 - // CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] - // CHECK: [[TRY]] // CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 9, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([9 x i64], [9 x i64]* [[MAPT4]], i32 0, i32 0), i32 0, i32 0) // CHECK-DAG: [[BPR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BP:%[^,]+]], i32 0, i32 0 // CHECK-DAG: [[PR]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[P:%[^,]+]], i32 0, i32 0 @@ -323,11 +341,12 @@ int foo(int n) { // correct and loaded correctly for the target regions in foo(). // CHECK: define internal void [[HVT0]](i[[SZ]] {{[^,]+}}, i[[SZ]] {{[^,]+}}, i[[SZ]] {{[^)]+}}) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] {{[^)]+}}) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] {{[^)]+}}) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] {{[^)]+}}) -// CHECK: alloca i16, +// CHECK: alloca i[[SZ]], +// CHECK: bitcast i[[SZ]]* {{.+}} to i16* // CHECK: ret void // CHECK-NEXT: } @@ -344,7 +363,7 @@ int foo(int n) { // CHECK-64: store i32 [[AA]], i32* [[AA_C]], align // CHECK-32: store i32 [[AA]], i32* [[AA_CASTED]], align // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED1]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) @@ -366,7 +385,7 @@ int foo(int n) { // CHECK: [[AA_C:%.+]] = bitcast i[[SZ]]* [[AA_CASTED]] to i16* // CHECK: store i16 [[AA]], i16* [[AA_C]], align // CHECK: [[PARAM:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]])* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED2]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}) @@ -397,7 +416,7 @@ int foo(int n) { // CHECK-DAG: store i16 [[AA]], i16* [[AA_C]], align // CHECK-DAG: [[PARAM1:%.+]] = load i[[SZ]], i[[SZ]]* [[A_CASTED]], align // CHECK-DAG: [[PARAM2:%.+]] = load i[[SZ]], i[[SZ]]* [[AA_CASTED]], align -// CHECK-DAG: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]]) +// CHECK-DAG: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]])* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[PARAM1]], i[[SZ]] [[PARAM2]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED3]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}) @@ -449,7 +468,7 @@ int foo(int n) { // CHECK-32-DAG:store i32 [[LOCAL_AV]], i32* [[LOCAL_A_CASTED]], align // CHECK-DAG: [[REF_A:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_A_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 9, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], [10 x float]*, i[[SZ]], float*, [5 x [10 x double]]*, i[[SZ]], i[[SZ]], double*, [[TT]]*)* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], [10 x float]* [[REF_B]], i[[SZ]] [[VAL_VLA1]], float* [[REF_BN]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] [[VAL_VLA2]], i[[SZ]] [[VAL_VLA3]], double* [[REF_CN]], [[TT]]* [[REF_D]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 9, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], [10 x float]*, i[[SZ]], float*, [5 x [10 x double]]*, i[[SZ]], i[[SZ]], double*, [[TT]]*)* [[OMP_OUTLINED4:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], [10 x float]* [[REF_B]], i[[SZ]] [[VAL_VLA1]], float* [[REF_BN]], [5 x [10 x double]]* [[REF_C]], i[[SZ]] [[VAL_VLA2]], i[[SZ]] [[VAL_VLA3]], double* [[REF_CN]], [[TT]]* [[REF_D]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED4]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, [10 x float]* {{.+}}, i[[SZ]] %{{.+}}, float* {{.+}}, [5 x [10 x double]]* {{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, double* {{.+}}, [[TT]]* {{.+}}) @@ -534,32 +553,39 @@ int bar(int n){ // CHECK-64: store i32 %{{.+}}, i32* [[B_ADDR]], // CHECK-64: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_CADDR]], +// CHECK-32: store i32 %{{.+}}, i32* %__vla_expr // CHECK-32: store i32 %{{.+}}, i32* [[B_ADDR:%.+]], // CHECK-32: [[B_CVAL:%.+]] = load i[[SZ]], i[[SZ]]* [[B_ADDR]], +// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 +// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] +// CHECK: [[TRY]] // We capture 2 VLA sizes in this target region // CHECK: [[CELEMSIZE2:%.+]] = mul nuw i[[SZ]] 2, [[VLA0:%.+]] // CHECK: [[CSIZE:%.+]] = mul nuw i[[SZ]] [[CELEMSIZE2]], 2 -// CHECK: [[IF:%.+]] = icmp sgt i32 {{[^,]+}}, 60 -// CHECK: br i1 [[IF]], label %[[TRY:[^,]+]], label %[[FAIL:[^,]+]] -// CHECK: [[TRY]] -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 5, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([5 x i64], [5 x i64]* [[MAPT5]], i32 0, i32 0), i32 0, i32 0) -// CHECK-DAG: [[BPR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP:%.+]], i32 0, i32 0 -// CHECK-DAG: [[PR]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SR]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 -// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] -// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX0]] -// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX0]] -// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] -// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX1]] -// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX1]] -// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] -// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX2]] -// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX2]] -// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [5 x i[[SZ]]], [5 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] -// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[BP]], i32 [[IDX3]] -// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 6, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SR:%[^,]+]], i64* getelementptr inbounds ([6 x i64], [6 x i64]* [[MAPT5]], i32 0, i32 0), i32 0, i32 0) +// CHECK-DAG: [[BPR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP:%.+]], i32 0, i32 0 +// CHECK-DAG: [[PR]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SR]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S:%.+]], i32 0, i32 0 +// CHECK-DAG: [[SADDR0:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX0:[0-9]+]] +// CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX0]] +// CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX0]] +// CHECK-DAG: [[SADDR1:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX1:[0-9]+]] +// CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX1]] +// CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX1]] +// CHECK-DAG: [[SADDR2:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX2:[0-9]+]] +// CHECK-DAG: [[BPADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX2]] +// CHECK-DAG: [[PADDR2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX2]] +// CHECK-DAG: [[SADDR3:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] +// CHECK-DAG: [[BPADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] +// CHECK-DAG: [[PADDR3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[SADDR4:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] +// CHECK-DAG: [[BPADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] +// CHECK-DAG: [[PADDR4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] +// CHECK-DAG: [[SADDR5:%.+]] = getelementptr inbounds [6 x i[[SZ]]], [6 x i[[SZ]]]* [[S]], i32 [[IDX3:[0-9]+]] +// CHECK-DAG: [[BPADDR5:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BP]], i32 [[IDX3]] +// CHECK-DAG: [[PADDR5:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[P]], i32 [[IDX3]] // The names below are not necessarily consistent with the names used for the // addresses above as some are repeated. @@ -582,15 +608,21 @@ int bar(int n){ // CHECK-DAG: store i[[SZ]] 4, i[[SZ]]* {{%[^,]+}} // CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR3:%.+]], -// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CPADDR3:%.+]], +// CHECK-DAG: store double* %{{.+}}, double** [[CPADDR3:%.+]], // CHECK-DAG: [[CBPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** -// CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to [[S1]]** +// CHECK-DAG: [[CPADDR3]] = bitcast i8** {{%[^,]+}} to double** +// CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* {{%[^,]+}} + +// CHECK-DAG: store [[S1]]* %{{.+}}, [[S1]]** [[CBPADDR4:%.+]], +// CHECK-DAG: store double* %{{.+}}, double** [[CPADDR4:%.+]], +// CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to [[S1]]** +// CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to double** // CHECK-DAG: store i[[SZ]] 8, i[[SZ]]* {{%[^,]+}} -// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR4:%.+]], -// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR4:%.+]], -// CHECK-DAG: [[CBPADDR4]] = bitcast i8** {{%[^,]+}} to i16** -// CHECK-DAG: [[CPADDR4]] = bitcast i8** {{%[^,]+}} to i16** +// CHECK-DAG: store i16* %{{.+}}, i16** [[CBPADDR5:%.+]], +// CHECK-DAG: store i16* %{{.+}}, i16** [[CPADDR5:%.+]], +// CHECK-DAG: [[CBPADDR5]] = bitcast i8** {{%[^,]+}} to i16** +// CHECK-DAG: [[CPADDR5]] = bitcast i8** {{%[^,]+}} to i16** // CHECK-DAG: store i[[SZ]] [[CSIZE]], i[[SZ]]* {{%[^,]+}} // CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 @@ -733,7 +765,7 @@ int bar(int n){ // CHECK-32-DAG:store i32 [[LOCAL_BV]], i32* [[LOCAL_B_CASTED]], align // CHECK-DAG: [[REF_B:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_B_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]], i[[SZ]], i[[SZ]], i16*)* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), [[S1]]* [[REF_THIS]], i[[SZ]] [[REF_B]], i[[SZ]] [[VAL_VLA1]], i[[SZ]] [[VAL_VLA2]], i16* [[REF_C]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, [[S1]]*, i[[SZ]], i[[SZ]], i[[SZ]], i16*)* [[OMP_OUTLINED5:@.+]] to void (i32*, i32*, ...)*), [[S1]]* [[REF_THIS]], i[[SZ]] [[REF_B]], i[[SZ]] [[VAL_VLA1]], i[[SZ]] [[VAL_VLA2]], i16* [[REF_C]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED5]](i32* noalias %.global_tid., i32* noalias %.bound_tid., [[S1]]* %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i16* {{.+}}) @@ -778,7 +810,7 @@ int bar(int n){ // CHECK-DAG: store i8 [[CONV_AAA]], i8* [[CONV]], align // CHECK-DAG: [[REF_AAA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AAA_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED6:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] {{.+}}, i[[SZ]] [[REF_AA]], i[[SZ]] [[REF_AAA]], [10 x i32]* [[REF_B]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED6:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] {{.+}}, i[[SZ]] [[REF_AA]], i[[SZ]] [[REF_AAA]], [10 x i32]* [[REF_B]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED6]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) @@ -811,7 +843,7 @@ int bar(int n){ // CHECK-DAG: store i16 [[CONV_AA]], i16* [[CONV]], align // CHECK-DAG: [[REF_AA:%.+]] = load i[[SZ]], i[[SZ]]* [[LOCAL_AA_CASTED]], -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED7:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], [10 x i32]* [[REF_B]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i[[SZ]], i[[SZ]], [10 x i32]*)* [[OMP_OUTLINED7:@.+]] to void (i32*, i32*, ...)*), i[[SZ]] [[REF_A]], i[[SZ]] [[REF_AA]], [10 x i32]* [[REF_B]]) // // // CHECK: define internal {{.*}}void [[OMP_OUTLINED7]](i32* noalias %.global_tid., i32* noalias %.bound_tid., i[[SZ]] %{{.+}}, i[[SZ]] %{{.+}}, [10 x i32]* {{.+}}) diff --git a/test/OpenMP/target_teams_distribute_simd_codegen_registration.cpp b/test/OpenMP/target_teams_distribute_simd_codegen_registration.cpp index d138114633ea9..5020bcc87aaff 100644 --- a/test/OpenMP/target_teams_distribute_simd_codegen_registration.cpp +++ b/test/OpenMP/target_teams_distribute_simd_codegen_registration.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target teams distribute simd codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,9 +24,22 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK -// Check that no target code is emmitted if no omptests flag was provided. +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// Check that no target code is emitted if no omptests flag was provided. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck %s -check-prefix=CHECK-NTARGET +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -60,96 +81,96 @@ // We have 7 target regions -// CHECK-DAG: {{@.+}} = private constant i8 0 -// TCHECK-NOT: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = weak constant i8 0 +// TCHECK-NOT: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] -// CHECK-DAG: {{@.+}} = private constant i8 0 +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] +// CHECK-DAG: {{@.+}} = weak constant i8 0 // CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i[[SZ]]] [i[[SZ]] 4] -// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 288] +// CHECK-DAG: {{@.+}} = private unnamed_addr constant [1 x i64] [i64 800] -// CHECK-NTARGET-NOT: private constant i8 0 +// CHECK-NTARGET-NOT: weak constant i8 0 // CHECK-NTARGET-NOT: private unnamed_addr constant [1 x i // CHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// CHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// CHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// CHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// CHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// CHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// CHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// CHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// CHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// CHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// CHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// CHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// CHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// CHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* @{{.*}}, i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR1:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME1:__omp_offloading_[0-9a-f]+_[0-9a-f]+__Z.+_l[0-9]+]]\00" -// TCHECK-DAG: [[ENTRY1:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY1:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR1]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR2:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME2:.+]]\00" -// TCHECK-DAG: [[ENTRY2:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY2:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR2]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR3:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME3:.+]]\00" -// TCHECK-DAG: [[ENTRY3:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY3:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR3]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR4:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME4:.+]]\00" -// TCHECK-DAG: [[ENTRY4:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY4:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR4]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR5:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME5:.+]]\00" -// TCHECK-DAG: [[ENTRY5:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY5:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR5]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR6:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME6:.+]]\00" -// TCHECK-DAG: [[ENTRY6:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY6:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR6]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR7:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME7:.+]]\00" -// TCHECK-DAG: [[ENTRY7:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY7:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR7]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR8:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME8:.+]]\00" -// TCHECK-DAG: [[ENTRY8:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY8:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR8]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR9:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME9:.+]]\00" -// TCHECK-DAG: [[ENTRY9:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY9:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR9]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR10:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME10:.+]]\00" -// TCHECK-DAG: [[ENTRY10:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY10:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR10]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR11:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME11:.+]]\00" -// TCHECK-DAG: [[ENTRY11:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY11:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR11]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // TCHECK-DAG: [[NAMEPTR12:@.+]] = internal unnamed_addr constant [{{.*}} x i8] c"[[NAME12:.+]]\00" -// TCHECK-DAG: [[ENTRY12:@.+]] = constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 +// TCHECK-DAG: [[ENTRY12:@.+]] = weak constant [[ENTTY]] { i8* bitcast (void (i[[SZ]])* @{{.*}} to i8*), i8* getelementptr inbounds ([{{.*}} x i8], [{{.*}} x i8]* [[NAMEPTR12]], i32 0, i32 0), i[[SZ]] 0, i32 0, i32 0 }, section ".omp_offloading.entries", align 1 // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] // CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] @@ -163,7 +184,7 @@ // CHECK-SAME: { i32, void ()*, i8* } { i32 500, void ()* [[P500:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 501, void ()* [[P501:@[^,]+]], i8* null }, // CHECK-SAME: { i32, void ()*, i8* } { i32 65535, void ()* [[PMAX:@[^,]+]], i8* null }, -// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK-SAME: { i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] // CHECK-NTARGET: @llvm.global_ctors = appending global [3 x { i32, void ()*, i8* }] [ @@ -328,18 +349,18 @@ struct ST { //CHECK-DAG: define internal void @[[NAME12]]( //CHECK-DAG: call void @[[NAME12]]( -//TCHECK-DAG: define void @[[NAME1]]( -//TCHECK-DAG: define void @[[NAME2]]( -//TCHECK-DAG: define void @[[NAME3]]( -//TCHECK-DAG: define void @[[NAME4]]( -//TCHECK-DAG: define void @[[NAME5]]( -//TCHECK-DAG: define void @[[NAME6]]( -//TCHECK-DAG: define void @[[NAME7]]( -//TCHECK-DAG: define void @[[NAME8]]( -//TCHECK-DAG: define void @[[NAME9]]( -//TCHECK-DAG: define void @[[NAME10]]( -//TCHECK-DAG: define void @[[NAME11]]( -//TCHECK-DAG: define void @[[NAME12]]( +//TCHECK-DAG: define weak void @[[NAME1]]( +//TCHECK-DAG: define weak void @[[NAME2]]( +//TCHECK-DAG: define weak void @[[NAME3]]( +//TCHECK-DAG: define weak void @[[NAME4]]( +//TCHECK-DAG: define weak void @[[NAME5]]( +//TCHECK-DAG: define weak void @[[NAME6]]( +//TCHECK-DAG: define weak void @[[NAME7]]( +//TCHECK-DAG: define weak void @[[NAME8]]( +//TCHECK-DAG: define weak void @[[NAME9]]( +//TCHECK-DAG: define weak void @[[NAME10]]( +//TCHECK-DAG: define weak void @[[NAME11]]( +//TCHECK-DAG: define weak void @[[NAME12]]( // CHECK-NTARGET-NOT: __tgt_target // CHECK-NTARGET-NOT: __tgt_register_lib @@ -381,7 +402,7 @@ struct ST { //CHECK: ret void //CHECK: declare i32 @__tgt_unregister_lib([[DSCTY]]*) -//CHECK: define linkonce hidden void @[[REGFN]](i8*) +//CHECK: define linkonce hidden void @[[REGFN]]() //CHECK-SAME: comdat { //CHECK: call i32 @__tgt_register_lib([[DSCTY]]* [[DESC]]) //CHECK: call i32 @__cxa_atexit(void (i8*)* @[[UNREGFN]], i8* bitcast ([[DSCTY]]* [[DESC]] to i8*), @@ -421,32 +442,32 @@ int bar(int a){ // Check metadata is properly generated: // CHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 247, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 265, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 272, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 291, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 415, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 298, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 291, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 298, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 284, i32 {{[0-9]+}}} -// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 221, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 268, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 286, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 293, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 312, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 436, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 319, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 312, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 319, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 305, i32 {{[0-9]+}}} +// CHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 242, i32 {{[0-9]+}}} // TCHECK: !omp_offload.info = !{!{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}, !{{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 195, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 247, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 265, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 272, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 291, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 415, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 298, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 291, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 298, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 284, i32 {{[0-9]+}}} -// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 221, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID:-?[0-9]+]], i32 [[FILEID:-?[0-9]+]], !"_ZN2SB3fooEv", i32 216, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SDD1Ev", i32 268, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SEC1Ev", i32 286, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SED1Ev", i32 293, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EE3fooEv", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EEC1Ev", i32 312, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_Z3bari", i32 436, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EED1Ev", i32 319, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EEC1Ev", i32 312, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi1000EED1Ev", i32 319, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2STILi100EE3fooEv", i32 305, i32 {{[0-9]+}}} +// TCHECK-DAG: = !{i32 0, i32 [[DEVID]], i32 [[FILEID]], !"_ZN2SCC1Ev", i32 242, i32 {{[0-9]+}}} // TCHECK-DAG: !{!"llvm.loop.vectorize.enable", i1 true} // CHECK-DAG: !{!"llvm.loop.vectorize.enable", i1 true} diff --git a/test/OpenMP/target_teams_distribute_simd_codegen_registration_naming.cpp b/test/OpenMP/target_teams_distribute_simd_codegen_registration_naming.cpp index 76618accbd685..856664b4c0e25 100644 --- a/test/OpenMP/target_teams_distribute_simd_codegen_registration_naming.cpp +++ b/test/OpenMP/target_teams_distribute_simd_codegen_registration_naming.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target teams distribute simd codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s -check-prefix=TCHECK @@ -16,6 +24,16 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s -check-prefix=TCHECK +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/target_teams_distribute_simd_collapse_codegen.cpp b/test/OpenMP/target_teams_distribute_simd_collapse_codegen.cpp index 703136f0652e5..8a80774ef187d 100644 --- a/test/OpenMP/target_teams_distribute_simd_collapse_codegen.cpp +++ b/test/OpenMP/target_teams_distribute_simd_collapse_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 template <typename T, int X, long long Y> @@ -60,6 +68,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 template <typename T, int n, int m> diff --git a/test/OpenMP/target_teams_distribute_simd_collapse_messages.cpp b/test/OpenMP/target_teams_distribute_simd_collapse_messages.cpp index 39937f90fb34a..a2d3bde44afbc 100644 --- a/test/OpenMP/target_teams_distribute_simd_collapse_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_defaultmap_messages.cpp b/test/OpenMP/target_teams_distribute_simd_defaultmap_messages.cpp index 3bea411540a2f..b8ac8aee672af 100644 --- a/test/OpenMP/target_teams_distribute_simd_defaultmap_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_defaultmap_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_depend_codegen.cpp b/test/OpenMP/target_teams_distribute_simd_depend_codegen.cpp new file mode 100644 index 0000000000000..0c1c13e1e908c --- /dev/null +++ b/test/OpenMP/target_teams_distribute_simd_depend_codegen.cpp @@ -0,0 +1,261 @@ +// Test host codegen. +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// Test target codegen - host bc file has to be created first. +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// expected-no-diagnostics +#ifndef HEADER +#define HEADER + +// CHECK-DAG: [[TT:%.+]] = type { i64, i8 } +// CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } +// CHECK-DAG: [[DEVTY:%.+]] = type { i8*, i8*, [[ENTTY]]*, [[ENTTY]]* } +// CHECK-DAG: [[DSCTY:%.+]] = type { i32, [[DEVTY]]*, [[ENTTY]]*, [[ENTTY]]* } + +// TCHECK: [[ENTTY:%.+]] = type { i8*, i8*, i{{32|64}}, i32, i32 } + +// CHECK-DAG: $[[REGFN:\.omp_offloading\..+]] = comdat + +// CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [2 x i[[SZ]]] [i[[SZ]] 0, i[[SZ]] 4] +// CHECK-DAG: [[MAPT:@.+]] = private unnamed_addr constant [2 x i64] [i64 544, i64 800] +// CHECK-DAG: @{{.*}} = weak constant i8 0 + +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = {{.*}}constant [[ENTTY]] +// TCHECK-NOT: @{{.+}} = weak constant [[ENTTY]] + +// Check if offloading descriptor is created. +// CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] +// CHECK: [[ENTEND:@.+]] = external constant [[ENTTY]] +// CHECK: [[DEVBEGIN:@.+]] = external constant i8 +// CHECK: [[DEVEND:@.+]] = external constant i8 +// CHECK: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x [[DEVTY]]] [{{.+}} { i8* [[DEVBEGIN]], i8* [[DEVEND]], [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }], comdat($[[REGFN]]) +// CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) + +// Check target registration is registered as a Ctor. +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] + + +template<typename tx, typename ty> +struct TT{ + tx X; + ty Y; +}; + +int global; +extern int global; + +// CHECK: define {{.*}}[[FOO:@.+]]( +int foo(int n) { + int a = 0; + short aa = 0; + float b[10]; + float bn[n]; + double c[5][10]; + double cn[5][n]; + TT<long long, char> d; + static long *plocal; + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID:@.+]], i32 [[GTID:%.+]], i32 1, i[[SZ]] {{20|40}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY0:%.+]]* + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 3 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 4, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY0]](i32 [[GTID]], [[TASK_TY0]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + #pragma omp target teams distribute simd device(global + a) depend(in: global) depend(out: a, b, cn[4]) + for (int i = 0; i < 10; ++i) { + } + + // CHECK: [[ADD:%.+]] = add nsw i32 + // CHECK: store i32 [[ADD]], i32* [[DEVICE_CAP:%.+]], + + // CHECK: [[BOOL:%.+]] = icmp ne i32 %{{.+}}, 0 + // CHECK: br i1 [[BOOL]], label %[[THEN:.+]], label %[[ELSE:.+]] + // CHECK: [[THEN]]: + // CHECK-DAG: [[BPADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP:%.+]], i32 0, i32 0 + // CHECK-DAG: [[PADDR0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P:%.+]], i32 0, i32 0 + // CHECK-DAG: [[CBPADDR0:%.+]] = bitcast i8** [[BPADDR0]] to i[[SZ]]** + // CHECK-DAG: [[CPADDR0:%.+]] = bitcast i8** [[PADDR0]] to i[[SZ]]** + // CHECK-DAG: store i[[SZ]]* [[BP0:%[^,]+]], i[[SZ]]** [[CBPADDR0]] + // CHECK-DAG: store i[[SZ]]* [[BP0]], i[[SZ]]** [[CPADDR0]] + + // CHECK-DAG: [[BPADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 1 + // CHECK-DAG: [[PADDR1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 1 + // CHECK-DAG: [[CBPADDR1:%.+]] = bitcast i8** [[BPADDR1]] to i[[SZ]]* + // CHECK-DAG: [[CPADDR1:%.+]] = bitcast i8** [[PADDR1]] to i[[SZ]]* + // CHECK-DAG: store i[[SZ]] [[BP1:%[^,]+]], i[[SZ]]* [[CBPADDR1]] + // CHECK-DAG: store i[[SZ]] [[BP1]], i[[SZ]]* [[CPADDR1]] + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 + // CHECK-DAG: getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{104|52}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1_:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1_:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + + // CHECK: [[ELSE]]: + // CHECK-NOT: getelementptr inbounds [2 x i8*], [2 x i8*]* + // CHECK: [[GEP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 + // CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], + // CHECK: store i32 [[DEV]], i32* [[GEP]], + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{56|28}}, i[[SZ]] {{16|12}}, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY1__:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY1__:%.+]]* + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 1 + // CHECK: getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 2 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]], i32 3, i8* [[DEP]], i32 0, i8* null) + // CHECK: br label %[[EXIT:.+]] + // CHECK: [[EXIT]]: + + #pragma omp target teams distribute simd device(global + a) nowait depend(inout: global, a, bn) if(a) + for (int i = 0; i < *plocal; ++i) { + static int local1; + *plocal = global; + local1 = global; + } + + // CHECK: [[TASK:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i[[SZ]] {{48|24}}, i[[SZ]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) + // CHECK: [[BC_TASK:%.+]] = bitcast i8* [[TASK]] to [[TASK_TY2:%.+]]* + // CHECK: getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 + // CHECK: [[DEP_START:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* %{{.+}}, i32 0, i32 0 + // CHECK: [[DEP:%.+]] = bitcast %struct.kmp_depend_info* [[DEP_START]] to i8* + // CHECK: call void @__kmpc_omp_wait_deps(%struct.ident_t* [[ID]], i32 [[GTID]], i32 1, i8* [[DEP]], i32 0, i8* null) + // CHECK: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + // CHECK: call i32 [[TASK_ENTRY2]](i32 [[GTID]], [[TASK_TY2]]* [[BC_TASK]]) + // CHECK: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* [[ID]], i32 [[GTID]], i8* [[TASK]]) + #pragma omp target teams distribute simd if(0) firstprivate(global) depend(out:global) + for (int i = 0; i < global; ++i) { + global += 1; + } + + return a; +} + +// Check that the offloading functions are emitted and that the arguments are +// correct and loaded correctly for the target regions in foo(). + +// CHECK: define internal void [[HVT0:@.+]]() + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY0]](i32{{.*}}, [[TASK_TY0]]* noalias) +// CHECK: store void (i8*, ...)* null, void (i8*, ...)** % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 0 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 0, i8** null, i8** null, i[[SZ]]* null, i64* null, i32 0, i32 0) +// CHECK-NEXT: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: call void [[HVT0]]() +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT1:@.+]](i[[SZ]]* %{{.+}}, i[[SZ]] %{{.+}}) + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1_]](i32{{.*}}, [[TASK_TY1_]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[SZT:%.+]] = getelementptr inbounds [2 x i[[SZ]]], [2 x i[[SZ]]]* %{{.+}}, i[[SZ]] 0, i[[SZ]] 0 +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[DEV:%.+]] = load i32, i32* [[DEVICE_CAP]], +// CHECK: [[DEVICE:%.+]] = sext i32 [[DEV]] to i64 +// CHECK: [[RET:%.+]] = call i32 @__tgt_target_teams_nowait(i64 [[DEVICE]], i8* @{{[^,]+}}, i32 2, i8** [[BPR:%[^,]+]], i8** [[PR:%[^,]+]], i[[SZ]]* [[SZT]], i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT]], i32 0, i32 0), i32 0, i32 0) + +// CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 +// CHECK-NEXT: br i1 [[ERROR]], label %[[FAIL:[^,]+]], label %[[END:[^,]+]] +// CHECK: [[FAIL]] +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK-NEXT: br label %[[END]] +// CHECK: [[END]] +// CHECK: ret i32 0 + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY1__]](i32{{.*}}, [[TASK_TY1__]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[DEVICE_CAP:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i32 0, i32 2 +// CHECK: [[BP0:%.+]] = load i[[SZ]]*, i[[SZ]]** % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT1]](i[[SZ]]* [[BP0]], i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + +// CHECK: define internal void [[HVT2:@.+]](i[[SZ]] %{{.+}}) +// Create stack storage and store argument in there. +// CHECK: [[AA_ADDR:%.+]] = alloca i[[SZ]], align +// CHECK: store i[[SZ]] %{{.+}}, i[[SZ]]* [[AA_ADDR]], align +// CHECK-64: [[AA_CADDR:%.+]] = bitcast i[[SZ]]* [[AA_ADDR]] to i32* +// CHECK-64: load i32, i32* [[AA_CADDR]], align +// CHECK-32: load i32, i32* [[AA_ADDR]], align + +// CHECK: define internal{{.*}} i32 [[TASK_ENTRY2]](i32{{.*}}, [[TASK_TY2]]* noalias) +// CHECK: call void (i8*, ...) % +// CHECK: [[BP1_I32:%.+]] = load i32, i32* % +// CHECK-64: [[BP1_CAST:%.+]] = bitcast i[[SZ]]* [[BP1_PTR:%.+]] to i32* +// CHECK-64: store i32 [[BP1_I32]], i32* [[BP1_CAST]], +// CHECK-32: store i32 [[BP1_I32]], i32* [[BP1_PTR:%.+]], +// CHECK: [[BP1:%.+]] = load i[[SZ]], i[[SZ]]* [[BP1_PTR]], +// CHECK: call void [[HVT2]](i[[SZ]] [[BP1]]) +// CHECK: ret i32 0 + + +#endif diff --git a/test/OpenMP/target_teams_distribute_simd_depend_messages.cpp b/test/OpenMP/target_teams_distribute_simd_depend_messages.cpp index c24c7438b0b5f..9c70d051d7b3e 100644 --- a/test/OpenMP/target_teams_distribute_simd_depend_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_depend_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_device_messages.cpp b/test/OpenMP/target_teams_distribute_simd_device_messages.cpp index e9246cb29def3..1f47d1c726362 100644 --- a/test/OpenMP/target_teams_distribute_simd_device_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_device_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_dist_schedule_codegen.cpp b/test/OpenMP/target_teams_distribute_simd_dist_schedule_codegen.cpp index d096da329abcb..874d4fc92c667 100644 --- a/test/OpenMP/target_teams_distribute_simd_dist_schedule_codegen.cpp +++ b/test/OpenMP/target_teams_distribute_simd_dist_schedule_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 template <typename T, int X, long long Y> @@ -81,6 +89,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 template <typename T, int n> diff --git a/test/OpenMP/target_teams_distribute_simd_dist_schedule_messages.cpp b/test/OpenMP/target_teams_distribute_simd_dist_schedule_messages.cpp index 9b9239d64e9d5..b9fdc18a0ec0d 100644 --- a/test/OpenMP/target_teams_distribute_simd_dist_schedule_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_dist_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp b/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp index f2871f8118334..b2b911e888f9a 100644 --- a/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp +++ b/test/OpenMP/target_teams_distribute_simd_firstprivate_codegen.cpp @@ -1,13 +1,26 @@ -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER @@ -97,16 +110,7 @@ int main() { // LAMBDA: [[G_ADDR:%.+]] = alloca i{{[0-9]+}}, // LAMBDA: [[G1_ADDR:%.+]] = alloca i{{[0-9]+}}, // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, - // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, - // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, - // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, - // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, - // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, - // LAMBDA: {{%.+}} = alloca i{{[0-9]+}}, - // LAMBDA: [[G_PRIV_ADDR:%.+]] = alloca i{{[0-9]+}}, - // LAMBDA: [[G1_PRIV_ADDR:%.+]] = alloca i{{[0-9]+}}, // LAMBDA: [[G1_TMP:%.+]] = alloca i32*, - // LAMBDA: [[SIVAR_PRIV_ADDR:%.+]] = alloca i{{[0-9]+}}, // skip loop vars // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_ADDR]], // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_ADDR]], @@ -114,15 +118,15 @@ int main() { // LAMBDA-DAG: [[G_CONV:%.+]] = bitcast {{.+}} [[G_ADDR]] to // LAMBDA-DAG: [[G1_CONV:%.+]] = bitcast {{.+}} [[G1_ADDR]] to // LAMBDA-DAG: [[SIVAR_CONV:%.+]] = bitcast {{.+}} [[SIVAR_ADDR]] to - // LAMBDA-DAG: store{{.+}} [[G1_PRIV_ADDR]], {{.+}} [[G1_TMP]], + // LAMBDA-DAG: store{{.+}} [[G1_CONV]], {{.+}} [[G1_TMP]], g = 1; g1 = 1; sivar = 2; // LAMBDA: call void @__kmpc_for_static_init_4( - // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_PRIV_ADDR]], + // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_CONV]], // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP]] // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1]], - // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_PRIV_ADDR]], + // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_CONV]], // LAMBDA-DAG: [[G1_REF:%.+]] = load{{.+}}, {{.+}} [[G1_TMP]], // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1_REF]], // LAMBDA: call void [[INNER_LAMBDA:@.+]]( @@ -212,8 +216,8 @@ int main() { // CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[SIVAR_ADDR]], // T_VAR and SIVAR -// CHECK-DAG-64: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* -// CHECK-DAG-64: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* // preparation vars // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], @@ -223,7 +227,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], @@ -243,13 +247,13 @@ int main() { // CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) // CHECK: call void @__kmpc_for_static_init_4( -// CHECK-DAG-32: {{.+}} = {{.+}} [[T_VAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] // CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[VAR_PRIV]] -// CHECK-DAG-32: {{.+}} = {{.+}} [[SIVAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_SIVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[SIVAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_SIVAR]] // CHECK: call void @__kmpc_for_static_fini( // CHECK: ret void @@ -309,7 +313,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], @@ -331,8 +335,8 @@ int main() { // CHECK-DAG: store [[S_INT_TY]]* [[VAR_PRIV]], [[S_INT_TY]]** [[TMP]], // CHECK: call void @__kmpc_for_static_init_4( -// CHECK-DAG-32: {{.+}} = {{.+}} [[T_VAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] // CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[TMP]] // CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] diff --git a/test/OpenMP/target_teams_distribute_simd_firstprivate_messages.cpp b/test/OpenMP/target_teams_distribute_simd_firstprivate_messages.cpp index 235533b49d16f..a4aaba7f8373e 100644 --- a/test/OpenMP/target_teams_distribute_simd_firstprivate_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_if_messages.cpp b/test/OpenMP/target_teams_distribute_simd_if_messages.cpp index 6cb2ad44aa110..43d8957f00657 100644 --- a/test/OpenMP/target_teams_distribute_simd_if_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_is_device_ptr_ast_print.cpp b/test/OpenMP/target_teams_distribute_simd_is_device_ptr_ast_print.cpp index 61aede4cb9e67..eb5347d61d65c 100644 --- a/test/OpenMP/target_teams_distribute_simd_is_device_ptr_ast_print.cpp +++ b/test/OpenMP/target_teams_distribute_simd_is_device_ptr_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -56,7 +60,7 @@ struct SA { // CHECK-NEXT: int aa[10]; // CHECK-NEXT: arr &raa = this->aa; // CHECK-NEXT: func( -// CHECK-NEXT: #pragma omp target teams distribute simd is_device_ptr(this->k) +// CHECK-NEXT: #pragma omp target teams distribute simd is_device_ptr(this->k){{$}} // CHECK-NEXT: for (int i = 0; i < 100; i++) // CHECK-NEXT: ; // CHECK-NEXT: #pragma omp target teams distribute simd is_device_ptr(this->z) diff --git a/test/OpenMP/target_teams_distribute_simd_is_device_ptr_messages.cpp b/test/OpenMP/target_teams_distribute_simd_is_device_ptr_messages.cpp index 013b703a91e9d..3408305692635 100644 --- a/test/OpenMP/target_teams_distribute_simd_is_device_ptr_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_is_device_ptr_messages.cpp @@ -1,4 +1,6 @@ // RUN: %clang_cc1 -std=c++11 -verify -fopenmp %s + +// RUN: %clang_cc1 -std=c++11 -verify -fopenmp-simd %s struct ST { int *a; }; diff --git a/test/OpenMP/target_teams_distribute_simd_lastprivate_codegen.cpp b/test/OpenMP/target_teams_distribute_simd_lastprivate_codegen.cpp index 8a978a2219548..fdf6d2f3a7dc0 100644 --- a/test/OpenMP/target_teams_distribute_simd_lastprivate_codegen.cpp +++ b/test/OpenMP/target_teams_distribute_simd_lastprivate_codegen.cpp @@ -5,12 +5,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -244,7 +260,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK-DAG: [[S_ARR_PTR_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_PTR]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST]], i8* [[TMP_VAL_BCAST]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST]], // CHECK: call void @__kmpc_for_static_fini( // lastprivates @@ -258,7 +274,7 @@ int main() { // CHECK-32: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -269,7 +285,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_FLOAT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -278,7 +294,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_ADDR_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: [[SVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[S_VAR_PRIV]], // CHECK-64: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR_REF]], // CHECK-32: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR]], @@ -346,7 +362,7 @@ int main() { // CHECK-DAG: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK-DAG: [[S_ARR_PTR_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_PTR1]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST1]], i8* [[TMP_VAL_BCAST1]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST1]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST1]], // CHECK: call void @__kmpc_for_static_fini( // lastprivates @@ -360,7 +376,7 @@ int main() { // CHECK-32: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR1]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF1]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_PRIV1]] to [[S_INT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -371,7 +387,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_INT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -380,6 +396,6 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_ADDR1_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: ret void #endif diff --git a/test/OpenMP/target_teams_distribute_simd_lastprivate_messages.cpp b/test/OpenMP/target_teams_distribute_simd_lastprivate_messages.cpp index a143dc2fd8830..3a9abfb24a1a3 100644 --- a/test/OpenMP/target_teams_distribute_simd_lastprivate_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_linear_messages.cpp b/test/OpenMP/target_teams_distribute_simd_linear_messages.cpp index 2be792bbda745..5bcd77da9e586 100644 --- a/test/OpenMP/target_teams_distribute_simd_linear_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/target_teams_distribute_simd_loop_messages.cpp b/test/OpenMP/target_teams_distribute_simd_loop_messages.cpp index 09aa548bd5c63..463c88df2c3e8 100644 --- a/test/OpenMP/target_teams_distribute_simd_loop_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_loop_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s -Wno-openmp-target + +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s -Wno-openmp-target class S { int a; diff --git a/test/OpenMP/target_teams_distribute_simd_map_messages.cpp b/test/OpenMP/target_teams_distribute_simd_map_messages.cpp index 29d48d958a3d5..73671d7284431 100644 --- a/test/OpenMP/target_teams_distribute_simd_map_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_map_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_messages.cpp b/test/OpenMP/target_teams_distribute_simd_messages.cpp index aa82c947010f5..6ee8073357f4c 100644 --- a/test/OpenMP/target_teams_distribute_simd_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } @@ -62,7 +64,7 @@ L1: // expected-error@+1 {{unexpected OpenMP clause 'default' in directive '#pragma omp target teams distribute simd'}} #pragma omp target teams distribute simd default(none) for (int i = 0; i < 10; ++i) - ++argc; // expected-error {{ariable 'argc' must have explicitly specified data sharing attributes}} + ++argc; goto L2; // expected-error {{use of undeclared label 'L2'}} #pragma omp target teams distribute simd diff --git a/test/OpenMP/target_teams_distribute_simd_misc_messages.c b/test/OpenMP/target_teams_distribute_simd_misc_messages.c index 70ec508f67987..397795ff6e0a9 100644 --- a/test/OpenMP/target_teams_distribute_simd_misc_messages.c +++ b/test/OpenMP/target_teams_distribute_simd_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp target teams distribute simd'}} #pragma omp target teams distribute simd diff --git a/test/OpenMP/target_teams_distribute_simd_nowait_messages.cpp b/test/OpenMP/target_teams_distribute_simd_nowait_messages.cpp index 6c2c2a50ab551..25daa81eb7447 100644 --- a/test/OpenMP/target_teams_distribute_simd_nowait_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_nowait_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -verify -fopenmp %s +// RUN: %clang_cc1 -fsyntax-only -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_num_teams_messages.cpp b/test/OpenMP/target_teams_distribute_simd_num_teams_messages.cpp index 63926bcfcbe95..cdc5244fe1ef5 100644 --- a/test/OpenMP/target_teams_distribute_simd_num_teams_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_num_teams_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_private_codegen.cpp b/test/OpenMP/target_teams_distribute_simd_private_codegen.cpp index c8883e18d7d9f..4fa4d04c706b3 100644 --- a/test/OpenMP/target_teams_distribute_simd_private_codegen.cpp +++ b/test/OpenMP/target_teams_distribute_simd_private_codegen.cpp @@ -1,13 +1,26 @@ -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER diff --git a/test/OpenMP/target_teams_distribute_simd_private_messages.cpp b/test/OpenMP/target_teams_distribute_simd_private_messages.cpp index 3d351bdc7d827..743ac160ce5de 100644 --- a/test/OpenMP/target_teams_distribute_simd_private_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_reduction_codegen.cpp b/test/OpenMP/target_teams_distribute_simd_reduction_codegen.cpp index 76a1299981cbc..c533bb30f813c 100644 --- a/test/OpenMP/target_teams_distribute_simd_reduction_codegen.cpp +++ b/test/OpenMP/target_teams_distribute_simd_reduction_codegen.cpp @@ -5,10 +5,23 @@ // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/target_teams_distribute_simd_reduction_messages.cpp b/test/OpenMP/target_teams_distribute_simd_reduction_messages.cpp index 35d4e12dc3392..b77082b44a9e0 100644 --- a/test/OpenMP/target_teams_distribute_simd_reduction_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_safelen_messages.cpp b/test/OpenMP/target_teams_distribute_simd_safelen_messages.cpp index 1f41c537ced6f..67b22216b97d2 100644 --- a/test/OpenMP/target_teams_distribute_simd_safelen_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_safelen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_shared_messages.cpp b/test/OpenMP/target_teams_distribute_simd_shared_messages.cpp index 9f98ab9c80b68..3147baedae8f9 100644 --- a/test/OpenMP/target_teams_distribute_simd_shared_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_shared_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_simdlen_messages.cpp b/test/OpenMP/target_teams_distribute_simd_simdlen_messages.cpp index 1b7ad66af69e5..6fea25a57ab64 100644 --- a/test/OpenMP/target_teams_distribute_simd_simdlen_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_simdlen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_distribute_simd_thread_limit_messages.cpp b/test/OpenMP/target_teams_distribute_simd_thread_limit_messages.cpp index 357f5dde313b6..593650b38cc7e 100644 --- a/test/OpenMP/target_teams_distribute_simd_thread_limit_messages.cpp +++ b/test/OpenMP/target_teams_distribute_simd_thread_limit_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_teams_firstprivate_messages.cpp b/test/OpenMP/target_teams_firstprivate_messages.cpp index 84fc4ab1063ac..492796981adf2 100644 --- a/test/OpenMP/target_teams_firstprivate_messages.cpp +++ b/test/OpenMP/target_teams_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_if_messages.cpp b/test/OpenMP/target_teams_if_messages.cpp index 0ab8fc3c94e8d..84c900a41ab3d 100644 --- a/test/OpenMP/target_teams_if_messages.cpp +++ b/test/OpenMP/target_teams_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 %s + void foo() { } diff --git a/test/OpenMP/target_teams_is_device_ptr_ast_print.cpp b/test/OpenMP/target_teams_is_device_ptr_ast_print.cpp index 3a9cee5eee193..19fe955ee76b8 100644 --- a/test/OpenMP/target_teams_is_device_ptr_ast_print.cpp +++ b/test/OpenMP/target_teams_is_device_ptr_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -49,7 +53,7 @@ struct SA { // CHECK-NEXT: int aa[10]; // CHECK-NEXT: arr &raa = this->aa; // CHECK-NEXT: func( -// CHECK-NEXT: #pragma omp target teams is_device_ptr(this->k) +// CHECK-NEXT: #pragma omp target teams is_device_ptr(this->k){{$}} // CHECK-NEXT: { // CHECK-NEXT: } // CHECK-NEXT: #pragma omp target teams is_device_ptr(this->z) diff --git a/test/OpenMP/target_teams_is_device_ptr_messages.cpp b/test/OpenMP/target_teams_is_device_ptr_messages.cpp index 0f5be43aaee79..8ae6d7fed0e16 100644 --- a/test/OpenMP/target_teams_is_device_ptr_messages.cpp +++ b/test/OpenMP/target_teams_is_device_ptr_messages.cpp @@ -1,4 +1,6 @@ // RUN: %clang_cc1 -std=c++11 -verify -fopenmp -ferror-limit 200 %s + +// RUN: %clang_cc1 -std=c++11 -verify -fopenmp-simd -ferror-limit 200 %s struct ST { int *a; }; diff --git a/test/OpenMP/target_teams_map_messages.cpp b/test/OpenMP/target_teams_map_messages.cpp index 7874c2e105cf9..57d9396394c7b 100644 --- a/test/OpenMP/target_teams_map_messages.cpp +++ b/test/OpenMP/target_teams_map_messages.cpp @@ -1,5 +1,7 @@ -// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 200 %s -// rUN: %clang_cc1 -DCCODE -verify -fopenmp -ferror-limit 200 -x c %s +// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 200 %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 200 %s -Wno-openmp-target +// RUN: %clang_cc1 -DCCODE -verify -fopenmp -ferror-limit 200 -x c %s -Wno-openmp-target #ifdef CCODE void foo(int arg) { const int n = 0; @@ -223,9 +225,17 @@ void SAclient(int arg) { {} #pragma omp target teams map(r.ArrS[0].A, t.ArrS[1].A) {} - #pragma omp target teams map(r.PtrS[0], r.PtrS->B) // expected-error {{same pointer derreferenced in multiple different ways in map clause expressions}} expected-note {{used here}} + #pragma omp target teams map(r.PtrS[0], r.PtrS->B) // expected-error {{same pointer dereferenced in multiple different ways in map clause expressions}} expected-note {{used here}} + {} + #pragma omp target teams map(r.PtrS, r.PtrS->B) // expected-error {{pointer cannot be mapped along with a section derived from itself}} expected-note {{used here}} + {} + #pragma omp target teams map(r.PtrS->A, r.PtrS->B) + {} + #pragma omp target teams map(r.RPtrS[0], r.RPtrS->B) // expected-error {{same pointer dereferenced in multiple different ways in map clause expressions}} expected-note {{used here}} + {} + #pragma omp target teams map(r.RPtrS, r.RPtrS->B) // expected-error {{pointer cannot be mapped along with a section derived from itself}} expected-note {{used here}} {} - #pragma omp target teams map(r.RPtrS[0], r.RPtrS->B) // expected-error {{same pointer derreferenced in multiple different ways in map clause expressions}} expected-note {{used here}} + #pragma omp target teams map(r.RPtrS->A, r.RPtrS->B) {} #pragma omp target teams map(r.S.Arr[:12]) {} diff --git a/test/OpenMP/target_teams_messages.cpp b/test/OpenMP/target_teams_messages.cpp index 7fc7d1f1aab07..3a367bfc7e4ed 100644 --- a/test/OpenMP/target_teams_messages.cpp +++ b/test/OpenMP/target_teams_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -o - %s + void foo() { } diff --git a/test/OpenMP/target_teams_nowait_messages.cpp b/test/OpenMP/target_teams_nowait_messages.cpp index 9cf8cc34ca068..aeb1829403157 100644 --- a/test/OpenMP/target_teams_nowait_messages.cpp +++ b/test/OpenMP/target_teams_nowait_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -o - %s + void foo() { } diff --git a/test/OpenMP/target_teams_num_teams_codegen.cpp b/test/OpenMP/target_teams_num_teams_codegen.cpp index 91288458c5aa1..0961c03e30f08 100644 --- a/test/OpenMP/target_teams_num_teams_codegen.cpp +++ b/test/OpenMP/target_teams_num_teams_codegen.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 @@ -16,13 +24,23 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER -// CHECK-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CHECK-DAG: [[S1:%.+]] = type { double } // CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } @@ -35,19 +53,19 @@ // We have 6 target regions -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] // Check if offloading descriptor is created. // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] @@ -58,7 +76,7 @@ // CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) // Check target registration is registered as a Ctor. -// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] template<typename tx> @@ -147,7 +165,7 @@ int bar(int n){ // CHECK: [[ARG:%.+]] = load i[[SZ]], i[[SZ]]* [[CAPEC_ADDR]], align // CHECK: [[TEAMS:%.+]] = load i32, i32* [[CAPE_ADDR]], align // -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 3, {{.*}}, i32 [[TEAMS]], i32 0) +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 4, {{.*}}, i32 [[TEAMS]], i32 0) // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 // CHECK: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] // @@ -158,7 +176,7 @@ int bar(int n){ // // // -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 1, {{.+}}, i32 1024, i32 0) +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 2, {{.+}}, i32 1024, i32 0) // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 // CHECK: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] // @@ -270,15 +288,15 @@ int bar(int n){ // CHECK-64: [[CONV:%.+]] = bitcast i[[SZ]]* [[CAPE_ADDR]] to i32* // CHECK-64: [[NT:%.+]] = load i32, i32* [[CONV]], align // CHECK-32: [[NT:%.+]] = load i32, i32* [[CAPE_ADDR]], align -// CHECK: call i32 @__kmpc_push_num_teams(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]], i32 0) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 2, +// CHECK: call i32 @__kmpc_push_num_teams(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]], i32 0) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 2, // // // CHECK: define internal void [[HVT2]]([[S1]]* {{%.+}}) -// CHECK: call i32 @__kmpc_push_num_teams(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 1024, i32 0) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 1, +// CHECK: call i32 @__kmpc_push_num_teams(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 1024, i32 0) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, // // @@ -294,8 +312,8 @@ int bar(int n){ // CHECK-64: [[CONV:%.+]] = bitcast i[[SZ]]* [[CAPE_ADDR]] to i32* // CHECK-64: [[NT:%.+]] = load i32, i32* [[CONV]], align // CHECK-32: [[NT:%.+]] = load i32, i32* [[CAPE_ADDR]], align -// CHECK: call i32 @__kmpc_push_num_teams(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]], i32 0) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 0, +// CHECK: call i32 @__kmpc_push_num_teams(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]], i32 0) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 0, // // // CHECK: define internal void [[HVT4]](i[[SZ]] [[PARM:%.+]]) @@ -303,8 +321,8 @@ int bar(int n){ // CHECK-64: [[CONV:%.+]] = bitcast i[[SZ]]* [[CAPE_ADDR]] to i32* // CHECK-64: [[NT:%.+]] = load i32, i32* [[CONV]], align // CHECK-32: [[NT:%.+]] = load i32, i32* [[CAPE_ADDR]], align -// CHECK: call i32 @__kmpc_push_num_teams(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]], i32 0) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 0, +// CHECK: call i32 @__kmpc_push_num_teams(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]], i32 0) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 0, // // @@ -313,8 +331,8 @@ int bar(int n){ // CHECK: define internal void [[HVT5]]( -// CHECK: call i32 @__kmpc_push_num_teams(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 20, i32 0) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 0, +// CHECK: call i32 @__kmpc_push_num_teams(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 20, i32 0) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 0, // // @@ -324,8 +342,8 @@ int bar(int n){ // CHECK: [[CONV:%.+]] = bitcast i[[SZ]]* [[CAPE_ADDR]] to i16* // CHECK: [[T:%.+]] = load i16, i16* [[CONV]], align // CHECK: [[NT:%.+]] = sext i16 [[T]] to i32 -// CHECK: call i32 @__kmpc_push_num_teams(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]], i32 0) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 2, +// CHECK: call i32 @__kmpc_push_num_teams(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]], i32 0) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 2, // // diff --git a/test/OpenMP/target_teams_num_teams_messages.cpp b/test/OpenMP/target_teams_num_teams_messages.cpp index e220fb67477dc..9134af640c902 100644 --- a/test/OpenMP/target_teams_num_teams_messages.cpp +++ b/test/OpenMP/target_teams_num_teams_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_teams_private_messages.cpp b/test/OpenMP/target_teams_private_messages.cpp index 69551ff96cc0e..7ee509c8acaa1 100644 --- a/test/OpenMP/target_teams_private_messages.cpp +++ b/test/OpenMP/target_teams_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_reduction_messages.cpp b/test/OpenMP/target_teams_reduction_messages.cpp index b5876b108508f..c62f2c58440b8 100644 --- a/test/OpenMP/target_teams_reduction_messages.cpp +++ b/test/OpenMP/target_teams_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -o - %s + void foo() { } diff --git a/test/OpenMP/target_teams_shared_messages.cpp b/test/OpenMP/target_teams_shared_messages.cpp index 84e412eb4c4bf..3fb84967b9d92 100644 --- a/test/OpenMP/target_teams_shared_messages.cpp +++ b/test/OpenMP/target_teams_shared_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/target_teams_thread_limit_codegen.cpp b/test/OpenMP/target_teams_thread_limit_codegen.cpp index 5d93c1de77fe6..21c9c5a68e70b 100644 --- a/test/OpenMP/target_teams_thread_limit_codegen.cpp +++ b/test/OpenMP/target_teams_thread_limit_codegen.cpp @@ -6,6 +6,14 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // Test target codegen - host bc file has to be created first. // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64 @@ -16,13 +24,23 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER -// CHECK-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CHECK-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CHECK-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CHECK-DAG: [[DEF_LOC:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CHECK-DAG: [[S1:%.+]] = type { double } // CHECK-DAG: [[ENTTY:%.+]] = type { i8*, i8*, i[[SZ:32|64]], i32, i32 } @@ -35,19 +53,19 @@ // We have 6 target regions -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 -// CHECK-DAG: @{{.*}} = private constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 +// CHECK-DAG: @{{.*}} = weak constant i8 0 -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] -// TCHECK: @{{.+}} = constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] +// TCHECK: @{{.+}} = weak constant [[ENTTY]] // Check if offloading descriptor is created. // CHECK: [[ENTBEGIN:@.+]] = external constant [[ENTTY]] @@ -58,7 +76,7 @@ // CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 1, [[DEVTY]]* getelementptr inbounds ([1 x [[DEVTY]]], [1 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]]) // Check target registration is registered as a Ctor. -// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* bitcast (void (i8*)* @[[REGFN]] to void ()*), i8* bitcast (void (i8*)* @[[REGFN]] to i8*) }] +// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }] template<typename tx> @@ -147,7 +165,7 @@ int bar(int n){ // CHECK: [[ARG:%.+]] = load i[[SZ]], i[[SZ]]* [[CAPEC_ADDR]], align // CHECK: [[TL:%.+]] = load i32, i32* [[CAPE_ADDR]], align // -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 3, {{.*}}, i32 0, i32 [[TL]]) +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 4, {{.*}}, i32 0, i32 [[TL]]) // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 // CHECK: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] // @@ -158,7 +176,7 @@ int bar(int n){ // // // -// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 1, {{.+}}, i32 0, i32 1024) +// CHECK-DAG: [[RET:%.+]] = call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 2, {{.+}}, i32 0, i32 1024) // CHECK: [[ERROR:%.+]] = icmp ne i32 [[RET]], 0 // CHECK: br i1 [[ERROR]], label %[[FAIL:.+]], label %[[END:[^,]+]] // @@ -279,15 +297,15 @@ int bar(int n){ // CHECK-64: [[CONV:%.+]] = bitcast i[[SZ]]* [[CAPE_ADDR]] to i32* // CHECK-64: [[TL:%.+]] = load i32, i32* [[CONV]], align // CHECK-32: [[TL:%.+]] = load i32, i32* [[CAPE_ADDR]], align -// CHECK: call i32 @__kmpc_push_num_teams(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 0, i32 [[TL]]) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 2, +// CHECK: call i32 @__kmpc_push_num_teams(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 0, i32 [[TL]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 2, // // // CHECK: define internal void [[HVT2]]([[S1]]* {{%.+}}) -// CHECK: call i32 @__kmpc_push_num_teams(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 0, i32 1024) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 1, +// CHECK: call i32 @__kmpc_push_num_teams(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 0, i32 1024) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 1, // // @@ -307,8 +325,8 @@ int bar(int n){ // CHECK-64: [[TL:%.+]] = load i32, i32* [[CONV2]], align // CHECK-32: [[NT:%.+]] = load i32, i32* [[CAPE_ADDR1]], align // CHECK-32: [[TL:%.+]] = load i32, i32* [[CAPE_ADDR2]], align -// CHECK: call i32 @__kmpc_push_num_teams(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]], i32 [[TL]]) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 0, +// CHECK: call i32 @__kmpc_push_num_teams(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]], i32 [[TL]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 0, // // // CHECK: define internal void [[HVT4]](i[[SZ]] [[PARM:%.+]]) @@ -316,8 +334,8 @@ int bar(int n){ // CHECK-64: [[CONV:%.+]] = bitcast i[[SZ]]* [[CAPE_ADDR]] to i32* // CHECK-64: [[TL:%.+]] = load i32, i32* [[CONV]], align // CHECK-32: [[TL:%.+]] = load i32, i32* [[CAPE_ADDR]], align -// CHECK: call i32 @__kmpc_push_num_teams(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 0, i32 [[TL]]) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 0, +// CHECK: call i32 @__kmpc_push_num_teams(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 0, i32 [[TL]]) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 0, // // @@ -326,8 +344,8 @@ int bar(int n){ // CHECK: define internal void [[HVT5]]( -// CHECK: call i32 @__kmpc_push_num_teams(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 0, i32 20) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 0, +// CHECK: call i32 @__kmpc_push_num_teams(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 0, i32 20) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 0, // // @@ -337,8 +355,8 @@ int bar(int n){ // CHECK: [[CONV:%.+]] = bitcast i[[SZ]]* [[CAPE_ADDR]] to i16* // CHECK: [[T:%.+]] = load i16, i16* [[CONV]], align // CHECK: [[NT:%.+]] = sext i16 [[T]] to i32 -// CHECK: call i32 @__kmpc_push_num_teams(%ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]], i32 1024) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC]], i32 2, +// CHECK: call i32 @__kmpc_push_num_teams(%struct.ident_t* {{[^,]+}}, i32 {{[^,]+}}, i32 [[NT]], i32 1024) +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC]], i32 2, // // diff --git a/test/OpenMP/target_teams_thread_limit_messages.cpp b/test/OpenMP/target_teams_thread_limit_messages.cpp index 2685f242196c1..4b0a4b05c8289 100644 --- a/test/OpenMP/target_teams_thread_limit_messages.cpp +++ b/test/OpenMP/target_teams_thread_limit_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/target_update_ast_print.cpp b/test/OpenMP/target_update_ast_print.cpp index cbff907acaeeb..e60e081b32104 100644 --- a/test/OpenMP/target_update_ast_print.cpp +++ b/test/OpenMP/target_update_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -21,7 +25,7 @@ T foo(T targ, U uarg) { // CHECK: static T a; // CHECK-NEXT: U b; // CHECK-NEXT: int l; -// CHECK-NEXT: #pragma omp target update to(a) if(l > 5) device(l) nowait depend(inout : l) +// CHECK-NEXT: #pragma omp target update to(a) if(l > 5) device(l) nowait depend(inout : l){{$}} // CHECK-NEXT: #pragma omp target update from(b) if(l < 5) device(l - 1) nowait depend(inout : l) // CHECK: static int a; // CHECK-NEXT: float b; diff --git a/test/OpenMP/target_update_codegen.cpp b/test/OpenMP/target_update_codegen.cpp index 7f45c313a8a1c..5beb853a2f6dd 100644 --- a/test/OpenMP/target_update_codegen.cpp +++ b/test/OpenMP/target_update_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 // CK1: [[ST:%.+]] = type { i32, double* } @@ -29,8 +37,9 @@ double gc[100]; // CK1: [[MTYPE03:@.+]] = {{.+}}constant [1 x i64] [i64 34] -// CK1: [[SIZE04:@.+]] = {{.+}}constant [2 x i[[sz]]] [i[[sz]] {{8|4}}, i[[sz]] 24] -// CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i64] [i64 33, i64 17] +// CK1-64: [[SIZE04:@.+]] = {{.+}}constant [2 x i64] [i64 sdiv exact (i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), i64 ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)), i64 24] +// CK1-32: [[SIZE04:@.+]] = {{.+}}constant [2 x i32] [i32 trunc (i64 sdiv exact (i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), i64 ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)) to i32), i32 24] +// CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i64] [i64 32, i64 281474976710673] // CK1-LABEL: _Z3fooi void foo(int arg) { @@ -143,6 +152,14 @@ void foo(int arg) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 // CK2: [[ST:%.+]] = type { i32, double* } @@ -159,8 +176,7 @@ struct ST { } }; -// CK2: [[SIZE00:@.+]] = {{.+}}constant [2 x i[[sz:64|32]]] [i{{64|32}} {{8|4}}, i{{64|32}} 24] -// CK2: [[MTYPE00:@.+]] = {{.+}}constant [2 x i64] [i64 34, i64 18] +// CK2: [[MTYPE00:@.+]] = {{.+}}constant [2 x i64] [i64 32, i64 281474976710674] // CK2-LABEL: _Z3bari int bar(int arg){ @@ -171,18 +187,21 @@ int bar(int arg){ // Region 00 // CK2: br i1 %{{[^,]+}}, label %[[IFTHEN:[^,]+]], label %[[IFELSE:[^,]+]] // CK2: [[IFTHEN]] -// CK2-DAG: call void @__tgt_target_data_update(i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[SIZE00]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}) +// CK2-DAG: call void @__tgt_target_data_update(i64 [[DEV:%[^,]+]], i32 2, i8** [[GEPBP:%.+]], i8** [[GEPP:%.+]], i[[sz:64|32]]* [[GEPS:%.+]], {{.+}}getelementptr {{.+}}[2 x i{{.+}}]* [[MTYPE00]]{{.+}}) // CK2-DAG: [[DEV]] = sext i32 [[DEVi32:%[^,]+]] to i64 // CK2-DAG: [[DEVi32]] = load i32, i32* %{{[^,]+}}, // CK2-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK2-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] +// CK2-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] // CK2-DAG: [[BP0:%.+]] = getelementptr inbounds {{.+}}[[BP]], i{{.+}} 0, i{{.+}} 0 // CK2-DAG: [[P0:%.+]] = getelementptr inbounds {{.+}}[[P]], i{{.+}} 0, i{{.+}} 0 +// CK2-DAG: [[S0:%.+]] = getelementptr inbounds {{.+}}[[S]], i{{.+}} 0, i{{.+}} 0 // CK2-DAG: [[CBP0:%.+]] = bitcast i8** [[BP0]] to [[ST]]** // CK2-DAG: [[CP0:%.+]] = bitcast i8** [[P0]] to double*** // CK2-DAG: store [[ST]]* [[VAR0:%[^,]+]], [[ST]]** [[CBP0]] // CK2-DAG: store double** [[SEC0:%[^,]+]], double*** [[CP0]] +// CK2-DAG: store i[[sz]] {{%.+}}, i[[sz]]* [[S0]] // CK2-DAG: [[SEC0]] = getelementptr inbounds {{.*}}[[ST]]* [[VAR0]], i32 0, i32 1 @@ -210,6 +229,14 @@ int bar(int arg){ // RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 + +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -verify -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} #ifdef CK3 // CK3-LABEL: no_target_devices @@ -229,6 +256,14 @@ void no_target_devices(int arg) { // RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-32 +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// SIMD-ONLY3-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc // RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-64 // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s @@ -237,6 +272,16 @@ void no_target_devices(int arg) { // RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-32 // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCK4 --check-prefix TCK4-32 + +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY4 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY4 %s +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY4 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY4 %s +// SIMD-ONLY4-NOT: {{__kmpc|__tgt}} #ifdef CK4 // CK4-LABEL: device_side_scan diff --git a/test/OpenMP/target_update_depend_codegen.cpp b/test/OpenMP/target_update_depend_codegen.cpp index 6542aa1958d3e..ee1b0bfceeeea 100644 --- a/test/OpenMP/target_update_depend_codegen.cpp +++ b/test/OpenMP/target_update_depend_codegen.cpp @@ -5,6 +5,14 @@ // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -std=c++11 -triple i386-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-targets=i386-pc-linux-gnu -x c++ -triple i386-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics // CK1: [[ST:%.+]] = type { i32, double* } // CK1: %struct.kmp_depend_info = type { i[[sz:64|32]], @@ -29,8 +37,9 @@ double gc[100]; // CK1: [[MTYPE03:@.+]] = {{.+}}constant [1 x i64] [i64 34] -// CK1: [[SIZE04:@.+]] = {{.+}}constant [2 x i[[sz]]] [i[[sz]] {{8|4}}, i[[sz]] 24] -// CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i64] [i64 33, i64 17] +// CK1-64: [[SIZE04:@.+]] = {{.+}}constant [2 x i64] [i64 sdiv exact (i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), i64 ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)), i64 24] +// CK1-32: [[SIZE04:@.+]] = {{.+}}constant [2 x i32] [i32 trunc (i64 sdiv exact (i64 sub (i64 ptrtoint (double** getelementptr (double*, double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1), i32 1) to i64), i64 ptrtoint (double** getelementptr inbounds (%struct.ST, %struct.ST* @gb, i32 0, i32 1) to i64)), i64 ptrtoint (i8* getelementptr (i8, i8* null, i32 1) to i64)) to i32), i32 24] +// CK1: [[MTYPE04:@.+]] = {{.+}}constant [2 x i64] [i64 32, i64 281474976710673] // CK1-LABEL: _Z3fooi void foo(int arg) { @@ -54,25 +63,25 @@ void foo(int arg) { // CK1: [[CAP_DEVICE:%.+]] = getelementptr inbounds %struct.anon, %struct.anon* [[CAPTURES:%.+]], i32 0, i32 0 // CK1: [[DEVICE:%.+]] = load i32, i32* %{{.+}} // CK1: store i32 [[DEVICE]], i32* [[CAP_DEVICE]], - // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) + // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 4, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* [[TASK_ENTRY0:@.+]] to i32 (i32, i8*)*)) // CK1: [[BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates* // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, %struct.kmp_task_t_with_privates* [[BC]], i32 0, i32 0 // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, %struct.kmp_task_t* [[TASK_T]], i32 0, i32 0 // CK1: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS]], // CK1: [[BC1:%.+]] = bitcast %struct.anon* [[CAPTURES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[SHAREDS_REF]], i8* [[BC1]], i[[sz]] 4, i32 4, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 4 [[SHAREDS_REF]], i8* align 4 [[BC1]], i[[sz]] 4, i1 false) // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates, %struct.kmp_task_t_with_privates* [[BC]], i32 0, i32 1 // CK1: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 0 // CK1: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8* // CK1: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_BASEPTRS]], i8* [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 1 // CK1: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8* // CK1: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_PTRS]], i8* [[BC_PTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t, %struct..kmp_privates.t* [[PRIVS]], i32 0, i32 2 // CK1: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_SIZES]], i8* bitcast ([1 x i[[sz]]]* [[SIZE00]] to i8*), i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i[[sz]]]* [[SIZE00]] to i8*), i[[sz]] {{8|4}}, i1 false) // CK1: [[DEP:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* [[MAIN_DEP:%.+]], i[[sz]] 0, i[[sz]] 0 // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0 // CK1: [[BC_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]] @@ -83,7 +92,7 @@ void foo(int arg) { // CK1: store i8 1, i8* [[DEP_ATTRS]] // CK1: [[DEP:%.+]] = getelementptr inbounds [1 x %struct.kmp_depend_info], [1 x %struct.kmp_depend_info]* [[MAIN_DEP]], i32 0, i32 0 // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[DEP]] to i8* - // CK1: = call i32 @__kmpc_omp_task_with_deps(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i8* [[BC]], i32 0, i8* null) + // CK1: = call i32 @__kmpc_omp_task_with_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]], i32 1, i8* [[BC]], i32 0, i8* null) // CK1: %{{.+}} = add nsw i32 %{{[^,]+}}, 1 #pragma omp target update if(1+3-5) device(arg) from(gc) nowait depend(in: arg) @@ -110,25 +119,25 @@ void foo(int arg) { // CK1: [[IF_BOOL:%.+]] = trunc i8 [[IF]] to i1 // CK1: [[IF:%.+]] = zext i1 [[IF_BOOL]] to i8 // CK1: store i8 [[IF]], i8* [[IF_DEVICE]], - // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) + // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*)) // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}* // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0 // CK1: [[SHAREDS:%.+]] = getelementptr inbounds %struct.kmp_task_t, %struct.kmp_task_t* [[TASK_T]], i32 0, i32 0 // CK1: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS]], // CK1: [[BC1:%.+]] = bitcast %struct.anon{{.+}}* [[CAPTURES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[SHAREDS_REF]], i8* [[BC1]], i[[sz]] 1, i32 1, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align 1 [[SHAREDS_REF]], i8* align 1 [[BC1]], i[[sz]] 1, i1 false) // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1 // CK1: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0 // CK1: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8* // CK1: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_BASEPTRS]], i8* [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1 // CK1: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8* // CK1: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_PTRS]], i8* [[BC_PTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2 // CK1: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_SIZES]], i8* bitcast ([1 x i[[sz]]]* [[SIZE02]] to i8*), i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([1 x i[[sz]]]* [[SIZE02]] to i8*), i[[sz]] {{8|4}}, i1 false) // CK1: [[DEP:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* [[MAIN_DEP:%.+]], i[[sz]] 0, i[[sz]] 0 // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0 // CK1: [[BC_ADR:%.+]] = ptrtoint i32* %{{.+}} to i[[sz]] @@ -154,10 +163,10 @@ void foo(int arg) { // CK1: store i8 3, i8* [[DEP_ATTRS]] // CK1: [[DEP:%.+]] = getelementptr inbounds [3 x %struct.kmp_depend_info], [3 x %struct.kmp_depend_info]* [[MAIN_DEP]], i32 0, i32 0 // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[DEP]] to i8* - // CK1: call void @__kmpc_omp_wait_deps(%ident_t* @{{.+}}, i32 %{{.+}}, i32 3, i8* [[BC]], i32 0, i8* null) - // CK1: call void @__kmpc_omp_task_begin_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 3, i8* [[BC]], i32 0, i8* null) + // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) // CK1: = call i32 [[TASK_ENTRY2]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]]) - // CK1: call void @__kmpc_omp_task_complete_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) // CK1: br label %[[IFEND:[^,]+]] @@ -183,22 +192,22 @@ void foo(int arg) { // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BP]], i32 0, i32 0 // CK1: [[GEPP0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[P]], i32 0, i32 0 // CK1: [[GEPS0:%.+]] = getelementptr inbounds [1 x i[[sz]]], [1 x i[[sz]]]* [[S]], i32 0, i32 0 - // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY3:@.+]] to i32 (i32, i8*)*)) + // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] [[sz]], i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY3:@.+]] to i32 (i32, i8*)*)) // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}* // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0 // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1 // CK1: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0 // CK1: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_BASEPTRS]] to i8* // CK1: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_BASEPTRS]], i8* [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1 // CK1: [[BC_PRIVS_PTRS:%.+]] = bitcast [1 x i8*]* [[PRIVS_PTRS]] to i8* // CK1: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_PTRS]], i8* [[BC_PTRS]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{8|4}}, i1 false) // CK1: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2 // CK1: [[BC_PRIVS_SIZES:%.+]] = bitcast [1 x i[[sz]]]* [[PRIVS_SIZES]] to i8* // CK1: [[BC_SIZES:%.+]] = bitcast i[[sz]]* [[GEPS0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_SIZES]], i8* [[BC_SIZES]], i[[sz]] {{8|4}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} [[BC_SIZES]], i[[sz]] {{8|4}}, i1 false) // CK1: [[DEP:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* [[MAIN_DEP:%.+]], i[[sz]] 0, i[[sz]] 0 // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0 // CK1: [[BC_ADR:%.+]] = ptrtoint float* %{{.+}} to i[[sz]] @@ -232,10 +241,10 @@ void foo(int arg) { // CK1: store i8 3, i8* [[DEP_ATTRS]] // CK1: [[DEP:%.+]] = getelementptr inbounds [4 x %struct.kmp_depend_info], [4 x %struct.kmp_depend_info]* [[MAIN_DEP]], i32 0, i32 0 // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[DEP]] to i8* - // CK1: call void @__kmpc_omp_wait_deps(%ident_t* @{{.+}}, i32 %{{.+}}, i32 4, i8* [[BC]], i32 0, i8* null) - // CK1: call void @__kmpc_omp_task_begin_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 4, i8* [[BC]], i32 0, i8* null) + // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) // CK1: = call i32 [[TASK_ENTRY3]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]]) - // CK1: call void @__kmpc_omp_task_complete_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) #pragma omp target update from(lb) depend(out: lb, arg, la, gc) {++arg;} @@ -257,21 +266,21 @@ void foo(int arg) { // CK1: store double* %{{.+}}, double** [[P1_BC]], // CK1: [[GEPBP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BP]], i32 0, i32 0 // CK1: [[GEPP0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[P]], i32 0, i32 0 - // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{88|44}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY4:@.+]] to i32 (i32, i8*)*)) + // CK1: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* {{.+}}, i32 {{.+}}, i32 1, i[[sz]] {{88|44}}, i[[sz]] 1, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates{{.+}}*)* [[TASK_ENTRY4:@.+]] to i32 (i32, i8*)*)) // CK1: [[RES_BC:%.+]] = bitcast i8* [[RES]] to %struct.kmp_task_t_with_privates{{.+}}* // CK1: [[TASK_T:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 0 // CK1: [[PRIVS:%.+]] = getelementptr inbounds %struct.kmp_task_t_with_privates{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]], i32 0, i32 1 // CK1: [[PRIVS_BASEPTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 0 // CK1: [[BC_PRIVS_BASEPTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_BASEPTRS]] to i8* // CK1: [[BC_BASEPTRS:%.+]] = bitcast i8** [[GEPBP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_BASEPTRS]], i8* [[BC_BASEPTRS]], i[[sz]] {{16|8}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_BASEPTRS]], i8* align {{8|4}} [[BC_BASEPTRS]], i[[sz]] {{16|8}}, i1 false) // CK1: [[PRIVS_PTRS:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 1 // CK1: [[BC_PRIVS_PTRS:%.+]] = bitcast [2 x i8*]* [[PRIVS_PTRS]] to i8* // CK1: [[BC_PTRS:%.+]] = bitcast i8** [[GEPP0]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_PTRS]], i8* [[BC_PTRS]], i[[sz]] {{16|8}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_PTRS]], i8* align {{8|4}} [[BC_PTRS]], i[[sz]] {{16|8}}, i1 false) // CK1: [[PRIVS_SIZES:%.+]] = getelementptr inbounds %struct..kmp_privates.t{{.+}}, %struct..kmp_privates.t{{.+}}* [[PRIVS]], i32 0, i32 2 // CK1: [[BC_PRIVS_SIZES:%.+]] = bitcast [2 x i[[sz]]]* [[PRIVS_SIZES]] to i8* - // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* [[BC_PRIVS_SIZES]], i8* bitcast ([2 x i[[sz]]]* [[SIZE04]] to i8*), i[[sz]] {{16|8}}, i32 {{8|4}}, i1 false) + // CK1: call void @llvm.memcpy.p0i8.p0i8.i[[sz]](i8* align {{8|4}} [[BC_PRIVS_SIZES]], i8* align {{8|4}} bitcast ([2 x i[[sz]]]* [[SIZE04]] to i8*), i[[sz]] {{16|8}}, i1 false) // CK1: [[DEP:%.+]] = getelementptr inbounds [5 x %struct.kmp_depend_info], [5 x %struct.kmp_depend_info]* [[MAIN_DEP:%.+]], i[[sz]] 0, i[[sz]] 0 // CK1: [[DEP_ADR:%.+]] = getelementptr inbounds %struct.kmp_depend_info, %struct.kmp_depend_info* [[DEP]], i32 0, i32 0 // CK1: [[BC_ADR:%.+]] = ptrtoint double* %{{.+}} to i[[sz]] @@ -313,10 +322,10 @@ void foo(int arg) { // CK1: store i8 1, i8* [[DEP_ATTRS]] // CK1: [[DEP:%.+]] = getelementptr inbounds [5 x %struct.kmp_depend_info], [5 x %struct.kmp_depend_info]* [[MAIN_DEP]], i32 0, i32 0 // CK1: [[BC:%.+]] = bitcast %struct.kmp_depend_info* [[DEP]] to i8* - // CK1: call void @__kmpc_omp_wait_deps(%ident_t* @{{.+}}, i32 %{{.+}}, i32 5, i8* [[BC]], i32 0, i8* null) - // CK1: call void @__kmpc_omp_task_begin_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_wait_deps(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i32 5, i8* [[BC]], i32 0, i8* null) + // CK1: call void @__kmpc_omp_task_begin_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) // CK1: = call i32 [[TASK_ENTRY4]](i32 %{{.+}}, %struct.kmp_task_t_with_privates{{.+}}* [[RES_BC]]) - // CK1: call void @__kmpc_omp_task_complete_if0(%ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) + // CK1: call void @__kmpc_omp_task_complete_if0(%struct.ident_t* @{{.+}}, i32 %{{.+}}, i8* [[RES]]) #pragma omp target update to(gb.b[:3]) depend(in: gb.b[:3], la, lb, gc, arg) {++arg;} } @@ -328,9 +337,9 @@ void foo(int arg) { // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] -// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%.+]], -// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%.+]], -// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%.+]], +// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]], +// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]], +// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%[^,]+]], // CK1-DAG: call void (i8*, ...) %{{.+}}(i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i[[sz]]]** [[S_PRIV]]) // CK1: ret i32 0 // CK1: } @@ -340,9 +349,9 @@ void foo(int arg) { // CK1-DAG: [[GEPBP]] = getelementptr inbounds {{.+}}[[BP:%[^,]+]] // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] -// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%.+]], -// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%.+]], -// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%.+]], +// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]], +// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]], +// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%[^,]+]], // CK1-DAG: call void (i8*, ...) %{{.+}}(i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i[[sz]]]** [[S_PRIV]]) // CK1: ret i32 0 // CK1: } @@ -353,9 +362,9 @@ void foo(int arg) { // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] -// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%.+]], -// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%.+]], -// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%.+]], +// CK1-DAG: [[BP]] = load [1 x i8*]*, [1 x i8*]** [[BP_PRIV:%[^,]+]], +// CK1-DAG: [[P]] = load [1 x i8*]*, [1 x i8*]** [[P_PRIV:%[^,]+]], +// CK1-DAG: [[S]] = load [1 x i[[sz]]]*, [1 x i[[sz]]]** [[S_PRIV:%[^,]+]], // CK1-DAG: call void (i8*, ...) %{{.+}}(i8* %{{[^,]+}}, [1 x i8*]** [[BP_PRIV]], [1 x i8*]** [[P_PRIV]], [1 x i[[sz]]]** [[S_PRIV]]) // CK1-NOT: __tgt_target_data_end // CK1: ret i32 0 @@ -367,9 +376,9 @@ void foo(int arg) { // CK1-DAG: [[GEPP]] = getelementptr inbounds {{.+}}[[P:%[^,]+]] // CK1-DAG: [[GEPS]] = getelementptr inbounds {{.+}}[[S:%[^,]+]] -// CK1-DAG: [[BP]] = load [2 x i8*]*, [2 x i8*]** [[BP_PRIV:%.+]], -// CK1-DAG: [[P]] = load [2 x i8*]*, [2 x i8*]** [[P_PRIV:%.+]], -// CK1-DAG: [[S]] = load [2 x i[[sz]]]*, [2 x i[[sz]]]** [[S_PRIV:%.+]], +// CK1-DAG: [[BP]] = load [2 x i8*]*, [2 x i8*]** [[BP_PRIV:%[^,]+]], +// CK1-DAG: [[P]] = load [2 x i8*]*, [2 x i8*]** [[P_PRIV:%[^,]+]], +// CK1-DAG: [[S]] = load [2 x i[[sz]]]*, [2 x i[[sz]]]** [[S_PRIV:%[^,]+]], // CK1-DAG: call void (i8*, ...) %{{.+}}(i8* %{{[^,]+}}, [2 x i8*]** [[BP_PRIV]], [2 x i8*]** [[P_PRIV]], [2 x i[[sz]]]** [[S_PRIV]]) // CK1-NOT: __tgt_target_data_end // CK1: ret i32 0 diff --git a/test/OpenMP/target_update_depend_messages.cpp b/test/OpenMP/target_update_depend_messages.cpp index 59e159e99fdba..d52f56684431a 100644 --- a/test/OpenMP/target_update_depend_messages.cpp +++ b/test/OpenMP/target_update_depend_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/target_update_device_messages.cpp b/test/OpenMP/target_update_device_messages.cpp index 37112755805cc..228b4a800a040 100644 --- a/test/OpenMP/target_update_device_messages.cpp +++ b/test/OpenMP/target_update_device_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/target_update_from_messages.cpp b/test/OpenMP/target_update_from_messages.cpp index c42bd12f75b87..e5a1f045e2307 100644 --- a/test/OpenMP/target_update_from_messages.cpp +++ b/test/OpenMP/target_update_from_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/target_update_if_messages.cpp b/test/OpenMP/target_update_if_messages.cpp index 3d076a6389d1e..4f229a768c7e5 100644 --- a/test/OpenMP/target_update_if_messages.cpp +++ b/test/OpenMP/target_update_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/target_update_messages.cpp b/test/OpenMP/target_update_messages.cpp index 73f1eeca8efec..da71ff6a5a996 100644 --- a/test/OpenMP/target_update_messages.cpp +++ b/test/OpenMP/target_update_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/target_update_nowait_messages.cpp b/test/OpenMP/target_update_nowait_messages.cpp index 19bc58ea467e6..fb3d035af9d33 100644 --- a/test/OpenMP/target_update_nowait_messages.cpp +++ b/test/OpenMP/target_update_nowait_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + int main(int argc, char **argv) { int i; diff --git a/test/OpenMP/target_update_to_messages.cpp b/test/OpenMP/target_update_to_messages.cpp index 9bde51e9839b2..58f6aee234b3a 100644 --- a/test/OpenMP/target_update_to_messages.cpp +++ b/test/OpenMP/target_update_to_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/target_vla_messages.cpp b/test/OpenMP/target_vla_messages.cpp index c970d0def661f..b744081e9837a 100644 --- a/test/OpenMP/target_vla_messages.cpp +++ b/test/OpenMP/target_vla_messages.cpp @@ -2,10 +2,15 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-unknown-unknown -emit-llvm-bc %s -o %t-ppc-host-ppc.bc // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-unknown-unknown -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-ppc.bc -o %t-ppc-device.ll +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-unknown-unknown -emit-llvm-bc %s -o %t-ppc-host-ppc.bc +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-unknown-unknown -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-ppc.bc -o %t-ppc-device.ll + // Nvidia GPUs don't support VLAs. // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host-nvptx.bc // RUN: %clang_cc1 -verify -DNO_VLA -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host-nvptx.bc -o %t-nvptx-device.ll +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host-nvptx.bc + #ifndef NO_VLA // expected-no-diagnostics #endif @@ -38,7 +43,7 @@ void declare_parallel_reduction(int arg) { #pragma omp end declare target template <typename T> -void target_template(int arg) { +void target_template(int arg) { #pragma omp target { #ifdef NO_VLA @@ -86,7 +91,7 @@ void teams_reduction(int arg) { #pragma omp target map(vla) #pragma omp teams reduction(+: vla) { } - + #pragma omp target map(a[0:2]) #pragma omp teams reduction(+: a[0:2]) { } diff --git a/test/OpenMP/task_ast_print.cpp b/test/OpenMP/task_ast_print.cpp index 54c05d8ed4106..01ffda9f87607 100644 --- a/test/OpenMP/task_ast_print.cpp +++ b/test/OpenMP/task_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -38,7 +42,7 @@ public: }; // CHECK: #pragma omp taskgroup task_reduction(+: this->b) -// CHECK: #pragma omp task private(this->a) private(this->a) private(T::a) in_reduction(+: this->b) +// CHECK: #pragma omp task private(this->a) private(this->a) private(T::a) in_reduction(+: this->b){{$}} // CHECK: #pragma omp task private(this->a) private(this->a) // CHECK: #pragma omp task private(this->a) private(this->a) private(this->S1::a) diff --git a/test/OpenMP/task_codegen.c b/test/OpenMP/task_codegen.c index 579a6f1b92297..6669c7e736163 100644 --- a/test/OpenMP/task_codegen.c +++ b/test/OpenMP/task_codegen.c @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -x c -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c -triple x86_64-apple-darwin10 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -x c -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/task_codegen.cpp b/test/OpenMP/task_codegen.cpp index 4743ca3dc13c9..fefe8b4329c21 100644 --- a/test/OpenMP/task_codegen.cpp +++ b/test/OpenMP/task_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -x c++ -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -33,7 +38,7 @@ int main() { // CHECK: [[SHAREDS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]], [[KMP_TASK_T]]* [[TASK_PTR:%.+]], i32 0, i32 0 // CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_PTR]] // CHECK: [[BITCAST:%.+]] = bitcast [[STRUCT_SHAREDS]]* [[CAPTURES]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[SHAREDS_REF]], i8* [[BITCAST]], i64 16, i32 8, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[SHAREDS_REF]], i8* align 8 [[BITCAST]], i64 16, i1 false) // CHECK: [[PRIORITY_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]], [[KMP_TASK_T]]* [[TASK_PTR]], i32 0, i32 4 // CHECK: [[PRIORITY:%.+]] = bitcast %union{{.+}}* [[PRIORITY_REF_PTR]] to i32* // CHECK: store i32 {{.+}}, i32* [[PRIORITY]] @@ -50,7 +55,7 @@ int main() { // CHECK: [[SHAREDS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]], [[KMP_TASK_T]]* [[TASK_PTR:%.+]], i32 0, i32 0 // CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_PTR]] // CHECK: [[BITCAST:%.+]] = bitcast [[STRUCT_SHAREDS1]]* [[CAPTURES]] to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[SHAREDS_REF]], i8* [[BITCAST]], i64 8, i32 8, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[SHAREDS_REF]], i8* align 8 [[BITCAST]], i64 8, i1 false) // CHECK: [[DEP:%.*]] = getelementptr inbounds [4 x [[KMP_DEPEND_INFO]]], [4 x [[KMP_DEPEND_INFO]]]* [[DEPENDENCIES:%.*]], i64 0, i64 0 // CHECK: [[T0:%.*]] = getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP]], i32 0, i32 0 // CHECK: store i64 ptrtoint (i32* @{{.+}} to i64), i64* [[T0]] diff --git a/test/OpenMP/task_default_messages.cpp b/test/OpenMP/task_default_messages.cpp index b3df295a8fa29..45bd8df82151b 100644 --- a/test/OpenMP/task_default_messages.cpp +++ b/test/OpenMP/task_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/task_depend_messages.cpp b/test/OpenMP/task_depend_messages.cpp index 1e51b479dd96a..469785e97cae4 100644 --- a/test/OpenMP/task_depend_messages.cpp +++ b/test/OpenMP/task_depend_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/task_final_messages.cpp b/test/OpenMP/task_final_messages.cpp index a217ea2ef30e1..49b9bfe204fd1 100644 --- a/test/OpenMP/task_final_messages.cpp +++ b/test/OpenMP/task_final_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/task_firstprivate_codegen.cpp b/test/OpenMP/task_firstprivate_codegen.cpp index 7c15a18e7d402..09b1deabde409 100644 --- a/test/OpenMP/task_firstprivate_codegen.cpp +++ b/test/OpenMP/task_firstprivate_codegen.cpp @@ -4,6 +4,14 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics // It doesn't pass on win32. // REQUIRES: shell @@ -199,7 +207,7 @@ int main() { // CHECK: [[SHAREDS_REF_ADDR:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_ADDR]], // CHECK: [[CAPTURES_ADDR:%.+]] = bitcast [[CAP_MAIN_TY]]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[SHAREDS_REF]], i8* [[CAPTURES_ADDR]], i64 40, i32 8, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[SHAREDS_REF]], i8* align 8 [[CAPTURES_ADDR]], i64 40, i1 false) // Initialize kmp_task_t->privates with default values (no init for simple types, default constructors for classes). // Also copy address of private copy to the corresponding shareds reference. @@ -353,7 +361,7 @@ int main() { // CHECK: [[SHAREDS_REF_ADDR:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_ADDR]], // CHECK: [[CAPTURES_ADDR:%.+]] = bitcast [[CAP_TMAIN_TY]]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[SHAREDS_REF]], i8* [[CAPTURES_ADDR]], i64 32, i32 8, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[SHAREDS_REF]], i8* align 8 [[CAPTURES_ADDR]], i64 32, i1 false) // Initialize kmp_task_t->privates with default values (no init for simple types, default constructors for classes). // CHECK: [[PRIVATES:%.+]] = getelementptr inbounds [[KMP_TASK_TMAIN_TY]], [[KMP_TASK_TMAIN_TY]]* [[RES_KMP_TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 diff --git a/test/OpenMP/task_firstprivate_messages.cpp b/test/OpenMP/task_firstprivate_messages.cpp index 11d8c5789879a..1a64f2abf3c67 100644 --- a/test/OpenMP/task_firstprivate_messages.cpp +++ b/test/OpenMP/task_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/task_if_codegen.cpp b/test/OpenMP/task_if_codegen.cpp index 5952085b6ec15..6f02942905b48 100644 --- a/test/OpenMP/task_if_codegen.cpp +++ b/test/OpenMP/task_if_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/task_if_messages.cpp b/test/OpenMP/task_if_messages.cpp index 89b10f71d3d90..fca57bb05d0d7 100644 --- a/test/OpenMP/task_if_messages.cpp +++ b/test/OpenMP/task_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/task_in_reduction_codegen.cpp b/test/OpenMP/task_in_reduction_codegen.cpp index 2ad013a14a57d..e96d12689fb16 100644 --- a/test/OpenMP/task_in_reduction_codegen.cpp +++ b/test/OpenMP/task_in_reduction_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -x c++ -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -33,18 +38,18 @@ int main(int argc, char **argv) { } // CHECK-LABEL: @main -// CHECK: void @__kmpc_taskgroup(%ident_t* @0, i32 [[GTID:%.+]]) +// CHECK: void @__kmpc_taskgroup(%struct.ident_t* @0, i32 [[GTID:%.+]]) // CHECK: [[TD1:%.+]] = call i8* @__kmpc_task_reduction_init(i32 [[GTID]], i32 3, i8* % // CHECK-NEXT: store i8* [[TD1]], i8** [[TD1_ADDR:%[^,]+]], -// CHECK-NEXT: call void @__kmpc_taskgroup(%ident_t* @0, i32 [[GTID]]) +// CHECK-NEXT: call void @__kmpc_taskgroup(%struct.ident_t* @0, i32 [[GTID]]) // CHECK: [[TD2:%.+]] = call i8* @__kmpc_task_reduction_init(i32 [[GTID]], i32 2, i8* % // CHECK-NEXT: store i8* [[TD2]], i8** [[TD2_ADDR:%[^,]+]], -// CHECK-NEXT: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* @0, i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i64, i16*, i8**, i8**)* [[OMP_PARALLEL:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i64 %{{.+}}, i16* %{{.+}}, i8** [[TD1_ADDR]], i8** [[TD2_ADDR]]) -// CHECK-NEXT: call void @__kmpc_end_taskgroup(%ident_t* @0, i32 [[GTID]]) -// CHECK-NEXT: call void @__kmpc_end_taskgroup(%ident_t* @0, i32 [[GTID]]) +// CHECK-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @0, i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i64, i16*, i8**, i8**)* [[OMP_PARALLEL:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i64 %{{.+}}, i16* %{{.+}}, i8** [[TD1_ADDR]], i8** [[TD2_ADDR]]) +// CHECK-NEXT: call void @__kmpc_end_taskgroup(%struct.ident_t* @0, i32 [[GTID]]) +// CHECK-NEXT: call void @__kmpc_end_taskgroup(%struct.ident_t* @0, i32 [[GTID]]) // CHECK: define internal void [[OMP_PARALLEL]]( -// CHECK: [[TASK_T:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* @0, i32 [[GTID:%.+]], i32 1, i64 56, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[T:%.+]]*)* [[OMP_TASK:@.+]] to i32 (i32, i8*)*)) +// CHECK: [[TASK_T:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID:%.+]], i32 1, i64 56, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[T:%.+]]*)* [[OMP_TASK:@.+]] to i32 (i32, i8*)*)) // CHECK-NEXT: [[TASK_T_WITH_PRIVS:%.+]] = bitcast i8* [[TASK_T]] to [[T]]* // CHECK: [[PRIVS:%.+]] = getelementptr inbounds [[T]], [[T]]* [[TASK_T_WITH_PRIVS]], i32 0, i32 1 // CHECK: [[TD1_REF:%.+]] = getelementptr inbounds [[PRIVATES]], [[PRIVATES]]* [[PRIVS]], i32 0, i32 0 @@ -57,7 +62,7 @@ int main(int argc, char **argv) { // CHECK-NEXT: [[TD2_ADDR:%.+]] = load i8**, i8*** [[TD2_SHAR]], // CHECK-NEXT: [[TD2:%.+]] = load i8*, i8** [[TD2_ADDR]], // CHECK-NEXT: store i8* [[TD2]], i8** [[TD2_REF]], -// CHECK-NEXT: call i32 @__kmpc_omp_task(%ident_t* @0, i32 [[GTID]], i8* [[TASK_T]]) +// CHECK-NEXT: call i32 @__kmpc_omp_task(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK_T]]) // CHECK-NEXT: ret void // CHECK-NEXT: } @@ -73,9 +78,11 @@ int main(int argc, char **argv) { // CHECK-NEXT: call i8* @__kmpc_task_reduction_get_th_data(i32 [[GTID]], i8* [[TD1]], i8* [[A_PTR]]) // CHECK: [[D_REF:%.+]] = getelementptr inbounds % // CHECK-NEXT: [[D_ADDR:%.+]] = load i16*, i16** [[D_REF]], +// CHECK: call i8* @__kmpc_threadprivate_cached( // CHECK: [[TD2:%.+]] = load i8*, i8** [[TD2_ADDR]], // CHECK-NEXT: [[D_PTR:%.+]] = bitcast i16* [[D_ADDR]] to i8* // CHECK-NEXT: call i8* @__kmpc_task_reduction_get_th_data(i32 [[GTID]], i8* [[TD2]], i8* [[D_PTR]]) // CHECK: add nsw i32 // CHECK: store i32 % +// CHECK-NOT: call i8* @__kmpc_threadprivate_cached( #endif diff --git a/test/OpenMP/task_in_reduction_message.cpp b/test/OpenMP/task_in_reduction_message.cpp index e176dc9305b39..d99d2e42d762b 100644 --- a/test/OpenMP/task_in_reduction_message.cpp +++ b/test/OpenMP/task_in_reduction_message.cpp @@ -2,6 +2,11 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + void foo() { } diff --git a/test/OpenMP/task_messages.cpp b/test/OpenMP/task_messages.cpp index f42a37ae1daad..832d8fa57d677 100644 --- a/test/OpenMP/task_messages.cpp +++ b/test/OpenMP/task_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -std=c++11 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -std=c++11 -o - %s + void foo() { } diff --git a/test/OpenMP/task_priority_messages.cpp b/test/OpenMP/task_priority_messages.cpp index 8a5553eb08c55..2e184ce1a911f 100644 --- a/test/OpenMP/task_priority_messages.cpp +++ b/test/OpenMP/task_priority_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/task_private_codegen.cpp b/test/OpenMP/task_private_codegen.cpp index 021749912a056..1484c6f7e376c 100644 --- a/test/OpenMP/task_private_codegen.cpp +++ b/test/OpenMP/task_private_codegen.cpp @@ -4,6 +4,14 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics // It doesn't pass on win32. Investigating. // REQUIRES: shell diff --git a/test/OpenMP/task_private_messages.cpp b/test/OpenMP/task_private_messages.cpp index 6bce1352b13eb..5663a436e43f9 100644 --- a/test/OpenMP/task_private_messages.cpp +++ b/test/OpenMP/task_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/task_shared_messages.cpp b/test/OpenMP/task_shared_messages.cpp index bf3b8ba283732..bdd46ab428257 100644 --- a/test/OpenMP/task_shared_messages.cpp +++ b/test/OpenMP/task_shared_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/taskgroup_ast_print.cpp b/test/OpenMP/taskgroup_ast_print.cpp index 4292eedfb56e9..a15600cb779bc 100644 --- a/test/OpenMP/taskgroup_ast_print.cpp +++ b/test/OpenMP/taskgroup_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -70,7 +74,7 @@ int main (int argc, char **argv) { // CHECK: static int a; #pragma omp taskgroup a=2; -// CHECK-NEXT: #pragma omp taskgroup +// CHECK-NEXT: #pragma omp taskgroup{{$}} // CHECK-NEXT: a = 2; // CHECK-NEXT: ++a; ++a; diff --git a/test/OpenMP/taskgroup_codegen.cpp b/test/OpenMP/taskgroup_codegen.cpp index 4b7d89e70391f..f672ab17fd59a 100644 --- a/test/OpenMP/taskgroup_codegen.cpp +++ b/test/OpenMP/taskgroup_codegen.cpp @@ -2,6 +2,12 @@ // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=TERM_DEBUG + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/taskgroup_messages.cpp b/test/OpenMP/taskgroup_messages.cpp index 23820734a551d..dfb4725280bb4 100644 --- a/test/OpenMP/taskgroup_messages.cpp +++ b/test/OpenMP/taskgroup_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + int foo(); int main() { diff --git a/test/OpenMP/taskgroup_task_reduction_codegen.cpp b/test/OpenMP/taskgroup_task_reduction_codegen.cpp index 63e015182420c..90b03c653ef02 100644 --- a/test/OpenMP/taskgroup_task_reduction_codegen.cpp +++ b/test/OpenMP/taskgroup_task_reduction_codegen.cpp @@ -2,10 +2,19 @@ // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ %s -verify -debug-info-kind=limited -emit-llvm -o - -triple x86_64-apple-darwin10 | FileCheck %s --check-prefix=CHECK --check-prefix=DEBUG + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ %s -verify -debug-info-kind=limited -emit-llvm -o - -triple x86_64-apple-darwin10 | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER +// CHECK-DAG: @reduction_size.[[ID:.+]]_[[CID:[0-9]+]].artificial. +// CHECK-DAG: @reduction_size.[[ID]]_[[CID]].artificial..cache. + struct S { int a; S() : a(0) {} @@ -15,7 +24,6 @@ struct S { friend S operator+(const S&a, const S&b) {return a;} }; - int main(int argc, char **argv) { int a; float b; @@ -35,7 +43,7 @@ int main(int argc, char **argv) { // CHECK: [[A:%.+]] = alloca i32, // CHECK: [[B:%.+]] = alloca float, // CHECK: [[C:%.+]] = alloca [5 x %struct.S], -// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%ident_t* +// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* // CHECK: [[RD_IN1:%.+]] = alloca [3 x [[T1:%[^,]+]]], // CHECK: [[TD1:%.+]] = alloca i8*, // CHECK: [[RD_IN2:%.+]] = alloca [2 x [[T2:%[^,]+]]], @@ -43,7 +51,7 @@ int main(int argc, char **argv) { // CHECK: [[VLA:%.+]] = alloca i16, i64 [[VLA_SIZE:%[^,]+]], -// CHECK: call void @__kmpc_taskgroup(%ident_t* {{[^,]+}}, i32 [[GTID]]) +// CHECK: call void @__kmpc_taskgroup(%struct.ident_t* {{[^,]+}}, i32 [[GTID]]) // CHECK-DAG: [[BC_A:%.+]] = bitcast i32* [[A]] to i8* // CHECK-DAG: store i8* [[BC_A]], i8** [[A_REF:[^,]+]], // CHECK-DAG: [[A_REF]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPA:%[^,]+]], i32 0, i32 0 @@ -51,14 +59,14 @@ int main(int argc, char **argv) { // CHECK-DAG: [[TMP6:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPA]], i32 0, i32 1 // CHECK-DAG: store i64 4, i64* [[TMP6]], // CHECK-DAG: [[TMP7:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPA]], i32 0, i32 2 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[AINIT:@.+]] to i8*), i8** [[TMP7]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[AINIT:.+]] to i8*), i8** [[TMP7]], // CHECK-DAG: [[TMP8:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPA]], i32 0, i32 3 // CHECK-DAG: store i8* null, i8** [[TMP8]], // CHECK-DAG: [[TMP9:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPA]], i32 0, i32 4 -// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* [[ACOMB:@.+]] to i8*), i8** [[TMP9]], +// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[ACOMB:.+]] to i8*), i8** [[TMP9]], // CHECK-DAG: [[TMP10:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPA]], i32 0, i32 5 // CHECK-DAG: [[TMP11:%.+]] = bitcast i32* [[TMP10]] to i8* -// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* [[TMP11]], i8 0, i64 4, i32 8, i1 false) +// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP11]], i8 0, i64 4, i1 false) // CHECK-DAG: [[TMP13:%.+]] = bitcast float* [[B]] to i8* // CHECK-DAG: store i8* [[TMP13]], i8** [[TMP12:%[^,]+]], // CHECK-DAG: [[TMP12]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPB:%[^,]+]], i32 0, i32 0 @@ -66,14 +74,14 @@ int main(int argc, char **argv) { // CHECK-DAG: [[TMP14:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPB]], i32 0, i32 1 // CHECK-DAG: store i64 4, i64* [[TMP14]], // CHECK-DAG: [[TMP15:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPB]], i32 0, i32 2 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[BINIT:@.+]] to i8*), i8** [[TMP15]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[BINIT:.+]] to i8*), i8** [[TMP15]], // CHECK-DAG: [[TMP16:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPB]], i32 0, i32 3 // CHECK-DAG: store i8* null, i8** [[TMP16]], // CHECK-DAG: [[TMP17:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPB]], i32 0, i32 4 -// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* [[BCOMB:@.+]] to i8*), i8** [[TMP17]], +// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[BCOMB:.+]] to i8*), i8** [[TMP17]], // CHECK-DAG: [[TMP18:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPB]], i32 0, i32 5 // CHECK-DAG: [[TMP19:%.+]] = bitcast i32* [[TMP18]] to i8* -// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* [[TMP19]], i8 0, i64 4, i32 8, i1 false) +// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP19]], i8 0, i64 4, i1 false) // CHECK-DAG: [[TMP21:%.+]] = bitcast i32* [[ARGC_ADDR]] to i8* // CHECK-DAG: store i8* [[TMP21]], i8** [[TMP20:%[^,]+]], // CHECK-DAG: [[TMP20]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPARGC:%[^,]+]], i32 0, i32 0 @@ -81,19 +89,19 @@ int main(int argc, char **argv) { // CHECK-DAG: [[TMP22:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPARGC]], i32 0, i32 1 // CHECK-DAG: store i64 4, i64* [[TMP22]], // CHECK-DAG: [[TMP23:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPARGC]], i32 0, i32 2 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[ARGCINIT:@.+]] to i8*), i8** [[TMP23]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[ARGCINIT:.+]] to i8*), i8** [[TMP23]], // CHECK-DAG: [[TMP24:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPARGC]], i32 0, i32 3 // CHECK-DAG: store i8* null, i8** [[TMP24]], // CHECK-DAG: [[TMP25:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPARGC]], i32 0, i32 4 -// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* [[ARGCCOMB:@.+]] to i8*), i8** [[TMP25]], +// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[ARGCCOMB:.+]] to i8*), i8** [[TMP25]], // CHECK-DAG: [[TMP26:%.+]] = getelementptr inbounds [[T1]], [[T1]]* [[GEPARGC]], i32 0, i32 5 // CHECK-DAG: [[TMP27:%.+]] = bitcast i32* [[TMP26]] to i8* -// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* [[TMP27]], i8 0, i64 4, i32 8, i1 false) +// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP27]], i8 0, i64 4, i1 false) // CHECK-DAG: [[TMP28:%.+]] = bitcast [3 x [[T1]]]* [[RD_IN1]] to i8* // CHECK-DAG: [[TMP29:%.+]] = call i8* @__kmpc_task_reduction_init(i32 [[GTID]], i32 3, i8* [[TMP28]]) // DEBUG-DAG: call void @llvm.dbg.declare(metadata i8** [[TD1]], // CHECK-DAG: store i8* [[TMP29]], i8** [[TD1]], -// CHECK-DAG: call void @__kmpc_taskgroup(%ident_t* {{[^,]+}}, i32 [[GTID]]) +// CHECK-DAG: call void @__kmpc_taskgroup(%struct.ident_t* {{[^,]+}}, i32 [[GTID]]) // CHECK-DAG: [[TMP31:%.+]] = bitcast [5 x %struct.S]* [[C]] to i8* // CHECK-DAG: store i8* [[TMP31]], i8** [[TMP30:%[^,]+]], // CHECK-DAG: [[TMP30]] = getelementptr inbounds [[T2]], [[T2]]* [[GEPC:%[^,]+]], i32 0, i32 0 @@ -101,14 +109,14 @@ int main(int argc, char **argv) { // CHECK-DAG: [[TMP32:%.+]] = getelementptr inbounds [[T2]], [[T2]]* [[GEPC]], i32 0, i32 1 // CHECK-DAG: store i64 20, i64* [[TMP32]], // CHECK-DAG: [[TMP33:%.+]] = getelementptr inbounds [[T2]], [[T2]]* [[GEPC]], i32 0, i32 2 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[CINIT:@.+]] to i8*), i8** [[TMP33]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[CINIT:.+]] to i8*), i8** [[TMP33]], // CHECK-DAG: [[TMP34:%.+]] = getelementptr inbounds [[T2]], [[T2]]* [[GEPC]], i32 0, i32 3 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[CFINI:@.+]] to i8*), i8** [[TMP34]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[CFINI:.+]] to i8*), i8** [[TMP34]], // CHECK-DAG: [[TMP35:%.+]] = getelementptr inbounds [[T2]], [[T2]]* [[GEPC]], i32 0, i32 4 -// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* [[CCOMB:@.+]] to i8*), i8** [[TMP35]], +// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[CCOMB:.+]] to i8*), i8** [[TMP35]], // CHECK-DAG: [[TMP36:%.+]] = getelementptr inbounds [[T2]], [[T2]]* [[GEPC]], i32 0, i32 5 // CHECK-DAG: [[TMP37:%.+]] = bitcast i32* [[TMP36]] to i8* -// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* [[TMP37]], i8 0, i64 4, i32 8, i1 false) +// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP37]], i8 0, i64 4, i1 false) // CHECK-DAG: [[TMP39:%.+]] = bitcast i16* [[VLA]] to i8* // CHECK-DAG: store i8* [[TMP39]], i8** [[TMP38:%[^,]+]], // CHECK-DAG: [[TMP38]] = getelementptr inbounds [[T2]], [[T2]]* [[GEPVLA:%[^,]+]], i32 0, i32 0 @@ -118,67 +126,67 @@ int main(int argc, char **argv) { // CHECK-DAG: [[TMP42:%.+]] = getelementptr inbounds [[T2]], [[T2]]* [[GEPVLA]], i32 0, i32 1 // CHECK-DAG: store i64 [[TMP40]], i64* [[TMP42]], // CHECK-DAG: [[TMP43:%.+]] = getelementptr inbounds [[T2]], [[T2]]* [[GEPVLA]], i32 0, i32 2 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[VLAINIT:@.+]] to i8*), i8** [[TMP43]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[VLAINIT:.+]] to i8*), i8** [[TMP43]], // CHECK-DAG: [[TMP44:%.+]] = getelementptr inbounds [[T2]], [[T2]]* [[GEPVLA]], i32 0, i32 3 // CHECK-DAG: store i8* null, i8** [[TMP44]], // CHECK-DAG: [[TMP45:%.+]] = getelementptr inbounds [[T2]], [[T2]]* [[GEPVLA]], i32 0, i32 4 -// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* [[VLACOMB:@.+]] to i8*), i8** [[TMP45]], +// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[VLACOMB:.+]] to i8*), i8** [[TMP45]], // CHECK-DAG: [[TMP46:%.+]] = getelementptr inbounds [[T2]], [[T2]]* [[GEPVLA]], i32 0, i32 5 // CHECK-DAG: store i32 1, i32* [[TMP46]], // CHECK: [[TMP47:%.+]] = bitcast [2 x [[T2]]]* [[RD_IN2]] to i8* // CHECK: [[TMP48:%.+]] = call i8* @__kmpc_task_reduction_init(i32 [[GTID]], i32 2, i8* [[TMP47]]) // CHECK: store i8* [[TMP48]], i8** [[TD2]], -// CHECK: call void @__kmpc_end_taskgroup(%ident_t* {{[^,]+}}, i32 [[GTID]]) -// CHECK: call void @__kmpc_end_taskgroup(%ident_t* {{[^,]+}}, i32 [[GTID]]) +// CHECK: call void @__kmpc_end_taskgroup(%struct.ident_t* {{[^,]+}}, i32 [[GTID]]) +// CHECK: call void @__kmpc_end_taskgroup(%struct.ident_t* {{[^,]+}}, i32 [[GTID]]) -// CHECK-DAG: define internal void [[AINIT]](i8*) +// CHECK-DAG: define internal void @[[AINIT]](i8*) // CHECK-DAG: store i32 0, i32* % // CHECK-DAG: ret void // CHECK-DAG: } -// CHECK-DAG: define internal void [[ACOMB]](i8*, i8*) +// CHECK-DAG: define internal void @[[ACOMB]](i8*, i8*) // CHECK-DAG: add nsw i32 % // CHECK-DAG: store i32 % // CHECK-DAG: ret void // CHECK-DAG: } -// CHECK-DAG: define internal void [[BINIT]](i8*) +// CHECK-DAG: define internal void @[[BINIT]](i8*) // CHECK-DAG: store float 0.000000e+00, float* % // CHECK-DAG: ret void // CHECK-DAG: } -// CHECK-DAG: define internal void [[BCOMB]](i8*, i8*) +// CHECK-DAG: define internal void @[[BCOMB]](i8*, i8*) // CHECK-DAG: fadd float % // CHECK-DAG: store float % // CHECK-DAG: ret void // CHECK-DAG: } -// CHECK-DAG: define internal void [[ARGCINIT]](i8*) +// CHECK-DAG: define internal void @[[ARGCINIT]](i8*) // CHECK-DAG: store i32 0, i32* % // CHECK-DAG: ret void // CHECK-DAG: } -// CHECK-DAG: define internal void [[ARGCCOMB]](i8*, i8*) +// CHECK-DAG: define internal void @[[ARGCCOMB]](i8*, i8*) // CHECK-DAG: add nsw i32 % // CHECK-DAG: store i32 % // CHECK-DAG: ret void // CHECK-DAG: } -// CHECK-DAG: define internal void [[CINIT]](i8*) +// CHECK-DAG: define internal void @[[CINIT]](i8*) // CHECK-DAG: phi %struct.S* [ // CHECK-DAG: call {{.+}}(%struct.S* {{.+}}) // CHECK-DAG: br i1 % // CHECK-DAG: ret void // CHECK-DAG: } -// CHECK-DAG: define internal void [[CFINI]](i8*) +// CHECK-DAG: define internal void @[[CFINI]](i8*) // CHECK-DAG: phi %struct.S* [ // CHECK-DAG: call {{.+}}(%struct.S* {{.+}}) // CHECK-DAG: br i1 % // CHECK-DAG: ret void // CHECK-DAG: } -// CHECK-DAG: define internal void [[CCOMB]](i8*, i8*) +// CHECK-DAG: define internal void @[[CCOMB]](i8*, i8*) // CHECK-DAG: phi %struct.S* [ // CHECK-DAG: phi %struct.S* [ // CHECK-DAG: call {{.+}}(%struct.S* {{.+}}, %struct.S* {{.+}}, %struct.S* {{.+}}) @@ -188,18 +196,18 @@ int main(int argc, char **argv) { // CHECK-DAG: ret void // CHECK_DAG: } -// CHECK-DAG: define internal void [[VLAINIT]](i8*) -// CHECK-DAG: call i32 @__kmpc_global_thread_num(%ident_t* {{[^,]+}}) -// CHECK-DAG: call i8* @__kmpc_threadprivate_cached(%ident_t* +// CHECK-DAG: define internal void @[[VLAINIT]](i8*) +// CHECK-DAG: call i32 @__kmpc_global_thread_num(%struct.ident_t* {{[^,]+}}) +// CHECK-DAG: call i8* @__kmpc_threadprivate_cached(%struct.ident_t* // CHECK-DAG: phi i16* [ // CHECK-DAG: store i16 0, i16* % // CHECK-DAG: br i1 % // CHECK-DAG: ret void // CHECK-DAG: } -// CHECK-DAG: define internal void [[VLACOMB]](i8*, i8*) -// CHECK-DAG: call i32 @__kmpc_global_thread_num(%ident_t* {{[^,]+}}) -// CHECK-DAG: call i8* @__kmpc_threadprivate_cached(%ident_t* +// CHECK-DAG: define internal void @[[VLACOMB]](i8*, i8*) +// CHECK-DAG: call i32 @__kmpc_global_thread_num(%struct.ident_t* {{[^,]+}}) +// CHECK-DAG: call i8* @__kmpc_threadprivate_cached(%struct.ident_t* // CHECK-DAG: phi i16* [ // CHECK-DAG: phi i16* [ // CHECK-DAG: sext i16 %{{.+}} to i32 @@ -210,3 +218,16 @@ int main(int argc, char **argv) { // CHECK-DAG: ret void // CHECK-DAG: } #endif + +// DEBUG-LABEL: distinct !DICompileUnit +// DEBUG-DAG: distinct !DISubprogram(linkageName: "[[AINIT]]", +// DEBUG-DAG: distinct !DISubprogram(linkageName: "[[ACOMB]]", +// DEBUG-DAG: distinct !DISubprogram(linkageName: "[[BINIT]]", +// DEBUG-DAG: distinct !DISubprogram(linkageName: "[[BCOMB]]", +// DEBUG-DAG: distinct !DISubprogram(linkageName: "[[ARGCINIT]]", +// DEBUG-DAG: distinct !DISubprogram(linkageName: "[[ARGCCOMB]]", +// DEBUG-DAG: distinct !DISubprogram(linkageName: "[[CINIT]]", +// DEBUG-DAG: distinct !DISubprogram(linkageName: "[[CFINI]]", +// DEBUG-DAG: distinct !DISubprogram(linkageName: "[[CCOMB]]", +// DEBUG-DAG: distinct !DISubprogram(linkageName: "[[VLAINIT]]", +// DEBUG-DAG: distinct !DISubprogram(linkageName: "[[VLACOMB]]", diff --git a/test/OpenMP/taskgroup_task_reduction_messages.cpp b/test/OpenMP/taskgroup_task_reduction_messages.cpp index 819dc5bdaf680..59ff2190bb9ea 100644 --- a/test/OpenMP/taskgroup_task_reduction_messages.cpp +++ b/test/OpenMP/taskgroup_task_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/taskloop_ast_print.cpp b/test/OpenMP/taskloop_ast_print.cpp index 891b31dce4db1..97f797b7c9869 100644 --- a/test/OpenMP/taskloop_ast_print.cpp +++ b/test/OpenMP/taskloop_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -16,7 +20,7 @@ T tmain(T argc) { #pragma omp taskgroup task_reduction(+: d) #pragma omp taskloop if(taskloop: argc > N) default(shared) untied priority(N) grainsize(N) reduction(+:g) in_reduction(+: d) // CHECK-NEXT: #pragma omp taskgroup task_reduction(+: d) - // CHECK-NEXT: #pragma omp taskloop if(taskloop: argc > N) default(shared) untied priority(N) grainsize(N) reduction(+: g) in_reduction(+: d) + // CHECK-NEXT: #pragma omp taskloop if(taskloop: argc > N) default(shared) untied priority(N) grainsize(N) reduction(+: g) in_reduction(+: d){{$}} for (int i = 0; i < 2; ++i) a = 2; // CHECK-NEXT: for (int i = 0; i < 2; ++i) diff --git a/test/OpenMP/taskloop_codegen.cpp b/test/OpenMP/taskloop_codegen.cpp index 94cf5365d872f..82b8e1b292bd0 100644 --- a/test/OpenMP/taskloop_codegen.cpp +++ b/test/OpenMP/taskloop_codegen.cpp @@ -1,19 +1,24 @@ // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -x c++ -emit-llvm %s -o - -femit-all-decls | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls | FileCheck %s + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -x c++ -emit-llvm %s -o - -femit-all-decls | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER // CHECK-LABEL: @main int main(int argc, char **argv) { -// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%ident_t* [[DEFLOC:@.+]]) -// CHECK: call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], -// CHECK: call i32 @__kmpc_omp_task(%ident_t* [[DEFLOC]], i32 [[GTID]], +// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* [[DEFLOC:@.+]]) +// CHECK: call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], +// CHECK: call i32 @__kmpc_omp_task(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], #pragma omp task ; -// CHECK: call void @__kmpc_taskgroup(%ident_t* [[DEFLOC]], i32 [[GTID]]) -// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 33, i64 80, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK1:@.+]] to i32 (i32, i8*)*)) +// CHECK: call void @__kmpc_taskgroup(%struct.ident_t* [[DEFLOC]], i32 [[GTID]]) +// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i32 33, i64 80, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK1:@.+]] to i32 (i32, i8*)*)) // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]* // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0 // CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5 @@ -23,12 +28,12 @@ int main(int argc, char **argv) { // CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 7 // CHECK: store i64 1, i64* [[ST]], // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]], -// CHECK: call void @__kmpc_taskloop(%ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 1, i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 0, i64 0, i8* null) -// CHECK: call void @__kmpc_end_taskgroup(%ident_t* [[DEFLOC]], i32 [[GTID]]) +// CHECK: call void @__kmpc_taskloop(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 1, i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 0, i64 0, i8* null) +// CHECK: call void @__kmpc_end_taskgroup(%struct.ident_t* [[DEFLOC]], i32 [[GTID]]) #pragma omp taskloop priority(argc) for (int i = 0; i < 10; ++i) ; -// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 80, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK2:@.+]] to i32 (i32, i8*)*)) +// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 80, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK2:@.+]] to i32 (i32, i8*)*)) // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]* // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0 // CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5 @@ -39,12 +44,12 @@ int main(int argc, char **argv) { // CHECK: store i64 1, i64* [[ST]], // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]], // CHECK: [[GRAINSIZE:%.+]] = zext i32 %{{.+}} to i64 -// CHECK: call void @__kmpc_taskloop(%ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 1, i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 1, i64 [[GRAINSIZE]], i8* null) +// CHECK: call void @__kmpc_taskloop(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 1, i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 1, i64 [[GRAINSIZE]], i8* null) #pragma omp taskloop nogroup grainsize(argc) for (int i = 0; i < 10; ++i) ; -// CHECK: call void @__kmpc_taskgroup(%ident_t* [[DEFLOC]], i32 [[GTID]]) -// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 80, i64 24, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK3:@.+]] to i32 (i32, i8*)*)) +// CHECK: call void @__kmpc_taskgroup(%struct.ident_t* [[DEFLOC]], i32 [[GTID]]) +// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 80, i64 24, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK3:@.+]] to i32 (i32, i8*)*)) // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]* // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0 // CHECK: [[IF:%.+]] = icmp ne i32 %{{.+}}, 0 @@ -56,8 +61,8 @@ int main(int argc, char **argv) { // CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 7 // CHECK: store i64 1, i64* [[ST]], // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]], -// CHECK: call void @__kmpc_taskloop(%ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 [[IF_INT]], i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 2, i64 4, i8* null) -// CHECK: call void @__kmpc_end_taskgroup(%ident_t* [[DEFLOC]], i32 [[GTID]]) +// CHECK: call void @__kmpc_taskloop(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 [[IF_INT]], i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 2, i64 4, i8* null) +// CHECK: call void @__kmpc_end_taskgroup(%struct.ident_t* [[DEFLOC]], i32 [[GTID]]) int i; #pragma omp taskloop if(argc) shared(argc, argv) collapse(2) num_tasks(4) for (i = 0; i < argc; ++i) @@ -147,8 +152,8 @@ int main(int argc, char **argv) { struct S { int a; S(int c) { -// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%ident_t* [[DEFLOC:@.+]]) -// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 80, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK4:@.+]] to i32 (i32, i8*)*)) +// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* [[DEFLOC:@.+]]) +// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 80, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK4:@.+]] to i32 (i32, i8*)*)) // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]* // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0 // CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5 @@ -159,7 +164,7 @@ struct S { // CHECK: store i64 1, i64* [[ST]], // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]], // CHECK: [[NUM_TASKS:%.+]] = zext i32 %{{.+}} to i64 -// CHECK: call void @__kmpc_taskloop(%ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 1, i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 2, i64 [[NUM_TASKS]], i8* null) +// CHECK: call void @__kmpc_taskloop(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 1, i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 2, i64 [[NUM_TASKS]], i8* null) #pragma omp taskloop shared(c) num_tasks(a) for (a = 0; a < c; ++a) ; diff --git a/test/OpenMP/taskloop_collapse_messages.cpp b/test/OpenMP/taskloop_collapse_messages.cpp index 1a5620e77db0a..4de0f38dae8b4 100644 --- a/test/OpenMP/taskloop_collapse_messages.cpp +++ b/test/OpenMP/taskloop_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/taskloop_final_messages.cpp b/test/OpenMP/taskloop_final_messages.cpp index a2d7cdae9673f..d53b12322e93e 100644 --- a/test/OpenMP/taskloop_final_messages.cpp +++ b/test/OpenMP/taskloop_final_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/taskloop_firstprivate_codegen.cpp b/test/OpenMP/taskloop_firstprivate_codegen.cpp index 6e8453050987c..d617835299ccd 100644 --- a/test/OpenMP/taskloop_firstprivate_codegen.cpp +++ b/test/OpenMP/taskloop_firstprivate_codegen.cpp @@ -4,6 +4,14 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics // It doesn't pass on win32. // REQUIRES: shell @@ -198,7 +206,7 @@ int main() { // CHECK: [[SHAREDS_REF_ADDR:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_ADDR]], // CHECK: [[CAPTURES_ADDR:%.+]] = bitcast [[CAP_MAIN_TY]]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[SHAREDS_REF]], i8* [[CAPTURES_ADDR]], i64 40, i32 8, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[SHAREDS_REF]], i8* align 8 [[CAPTURES_ADDR]], i64 40, i1 false) // Initialize kmp_task_t->privates with default values (no init for simple types, default constructors for classes). // Also copy address of private copy to the corresponding shareds reference. @@ -368,7 +376,7 @@ int main() { // CHECK: [[SHAREDS_REF_ADDR:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_ADDR]], // CHECK: [[CAPTURES_ADDR:%.+]] = bitcast [[CAP_TMAIN_TY]]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[SHAREDS_REF]], i8* [[CAPTURES_ADDR]], i64 32, i32 8, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[SHAREDS_REF]], i8* align 8 [[CAPTURES_ADDR]], i64 32, i1 false) // Initialize kmp_task_t->privates with default values (no init for simple types, default constructors for classes). // CHECK: [[PRIVATES:%.+]] = getelementptr inbounds [[KMP_TASK_TMAIN_TY]], [[KMP_TASK_TMAIN_TY]]* [[RES_KMP_TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 diff --git a/test/OpenMP/taskloop_firstprivate_messages.cpp b/test/OpenMP/taskloop_firstprivate_messages.cpp index e63d6a678b8df..7e1b818800da8 100644 --- a/test/OpenMP/taskloop_firstprivate_messages.cpp +++ b/test/OpenMP/taskloop_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/taskloop_grainsize_messages.cpp b/test/OpenMP/taskloop_grainsize_messages.cpp index 047f925b82506..a1cfea474c977 100644 --- a/test/OpenMP/taskloop_grainsize_messages.cpp +++ b/test/OpenMP/taskloop_grainsize_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/taskloop_in_reduction_codegen.cpp b/test/OpenMP/taskloop_in_reduction_codegen.cpp index e9ee9a32b97b2..93f15932d7da9 100644 --- a/test/OpenMP/taskloop_in_reduction_codegen.cpp +++ b/test/OpenMP/taskloop_in_reduction_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -x c++ -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -34,18 +39,18 @@ int main(int argc, char **argv) { } // CHECK-LABEL: @main -// CHECK: void @__kmpc_taskgroup(%ident_t* @0, i32 [[GTID:%.+]]) +// CHECK: void @__kmpc_taskgroup(%struct.ident_t* @0, i32 [[GTID:%.+]]) // CHECK: [[TD1:%.+]] = call i8* @__kmpc_task_reduction_init(i32 [[GTID]], i32 3, i8* % // CHECK-NEXT: store i8* [[TD1]], i8** [[TD1_ADDR:%[^,]+]], -// CHECK-NEXT: call void @__kmpc_taskgroup(%ident_t* @0, i32 [[GTID]]) +// CHECK-NEXT: call void @__kmpc_taskgroup(%struct.ident_t* @0, i32 [[GTID]]) // CHECK: [[TD2:%.+]] = call i8* @__kmpc_task_reduction_init(i32 [[GTID]], i32 2, i8* % // CHECK-NEXT: store i8* [[TD2]], i8** [[TD2_ADDR:%[^,]+]], -// CHECK-NEXT: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* @0, i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i64, i16*, i8**, i8**)* [[OMP_PARALLEL:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i64 %{{.+}}, i16* %{{.+}}, i8** [[TD1_ADDR]], i8** [[TD2_ADDR]]) -// CHECK-NEXT: call void @__kmpc_end_taskgroup(%ident_t* @0, i32 [[GTID]]) -// CHECK-NEXT: call void @__kmpc_end_taskgroup(%ident_t* @0, i32 [[GTID]]) +// CHECK-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @0, i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i64, i16*, i8**, i8**)* [[OMP_PARALLEL:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i64 %{{.+}}, i16* %{{.+}}, i8** [[TD1_ADDR]], i8** [[TD2_ADDR]]) +// CHECK-NEXT: call void @__kmpc_end_taskgroup(%struct.ident_t* @0, i32 [[GTID]]) +// CHECK-NEXT: call void @__kmpc_end_taskgroup(%struct.ident_t* @0, i32 [[GTID]]) // CHECK: define internal void [[OMP_PARALLEL]]( -// CHECK: [[TASK_T:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* @0, i32 [[GTID:%.+]], i32 1, i64 96, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[T:%.+]]*)* [[OMP_TASK:@.+]] to i32 (i32, i8*)*)) +// CHECK: [[TASK_T:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID:%.+]], i32 1, i64 96, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[T:%.+]]*)* [[OMP_TASK:@.+]] to i32 (i32, i8*)*)) // CHECK-NEXT: [[TASK_T_WITH_PRIVS:%.+]] = bitcast i8* [[TASK_T]] to [[T]]* // CHECK: [[PRIVS:%.+]] = getelementptr inbounds [[T]], [[T]]* [[TASK_T_WITH_PRIVS]], i32 0, i32 1 // CHECK: [[TD1_REF:%.+]] = getelementptr inbounds [[PRIVATES]], [[PRIVATES]]* [[PRIVS]], i32 0, i32 0 @@ -58,7 +63,7 @@ int main(int argc, char **argv) { // CHECK-NEXT: [[TD2_ADDR:%.+]] = load i8**, i8*** [[TD2_SHAR]], // CHECK-NEXT: [[TD2:%.+]] = load i8*, i8** [[TD2_ADDR]], // CHECK-NEXT: store i8* [[TD2]], i8** [[TD2_REF]], -// CHECK: call void @__kmpc_taskloop(%ident_t* @0, i32 [[GTID]], i8* [[TASK_T]], i32 1, +// CHECK: call void @__kmpc_taskloop(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK_T]], i32 1, // CHECK: ret void // CHECK-NEXT: } diff --git a/test/OpenMP/taskloop_in_reduction_messages.cpp b/test/OpenMP/taskloop_in_reduction_messages.cpp index 409975565a91c..dc19bac907d56 100644 --- a/test/OpenMP/taskloop_in_reduction_messages.cpp +++ b/test/OpenMP/taskloop_in_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/taskloop_lastprivate_codegen.cpp b/test/OpenMP/taskloop_lastprivate_codegen.cpp index 837977eb57c05..a6f7abcc338b7 100644 --- a/test/OpenMP/taskloop_lastprivate_codegen.cpp +++ b/test/OpenMP/taskloop_lastprivate_codegen.cpp @@ -4,6 +4,14 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics // It doesn't pass on win32. // REQUIRES: shell @@ -190,7 +198,7 @@ int main() { // CHECK: [[SHAREDS_REF_ADDR:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_ADDR]], // CHECK: [[CAPTURES_ADDR:%.+]] = bitcast [[CAP_MAIN_TY]]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[SHAREDS_REF]], i8* [[CAPTURES_ADDR]], i64 40, i32 8, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[SHAREDS_REF]], i8* align 8 [[CAPTURES_ADDR]], i64 40, i1 false) // Initialize kmp_task_t->privates with default values (no init for simple types, default constructors for classes). // Also copy address of private copy to the corresponding shareds reference. @@ -279,18 +287,18 @@ int main() { // CHECK-NEXT: br i1 // CHECK: bitcast [[S_DOUBLE_TY]]* %{{.+}} to i8* // CHECK: bitcast [[S_DOUBLE_TY]]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* % +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align {{[0-9]+}} % // CHECK: load i32, i32* % // CHECK: store i32 %{{.+}}, i32* % // CHECK: getelementptr inbounds [2 x [[S_DOUBLE_TY]]], [2 x [[S_DOUBLE_TY]]]* % // CHECK: phi [[S_DOUBLE_TY]]* // CHECK: phi [[S_DOUBLE_TY]]* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* % +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align {{[0-9]+}} % // CHECK: icmp eq [[S_DOUBLE_TY]]* % // CHECK-NEXT: br i1 // CHECK: bitcast [2 x i32]* %{{.+}} to i8* // CHECK: bitcast [2 x i32]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* % +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align {{[0-9]+}} % // CHECK: load i32, i32* % // CHECK: store i32 %{{.+}}, i32* % // CHECK: br label @@ -364,7 +372,7 @@ int main() { // CHECK: [[SHAREDS_REF_ADDR:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_ADDR]], // CHECK: [[CAPTURES_ADDR:%.+]] = bitcast [[CAP_TMAIN_TY]]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[SHAREDS_REF]], i8* [[CAPTURES_ADDR]], i64 32, i32 8, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[SHAREDS_REF]], i8* align 8 [[CAPTURES_ADDR]], i64 32, i1 false) // Initialize kmp_task_t->privates with default values (no init for simple types, default constructors for classes). // CHECK: [[PRIVATES:%.+]] = getelementptr inbounds [[KMP_TASK_TMAIN_TY]], [[KMP_TASK_TMAIN_TY]]* [[RES_KMP_TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 @@ -445,16 +453,16 @@ int main() { // CHECK: store i32 %{{.+}}, i32* % // CHECK: bitcast [2 x i32]* %{{.+}} to i8* // CHECK: bitcast [2 x i32]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* % +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align {{[0-9]+}} % // CHECK: getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* % // CHECK: phi [[S_INT_TY]]* // CHECK: phi [[S_INT_TY]]* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* % +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align {{[0-9]+}} % // CHECK: icmp eq [[S_INT_TY]]* % // CHECK-NEXT: br i1 // CHECK: bitcast [[S_INT_TY]]* %{{.+}} to i8* // CHECK: bitcast [[S_INT_TY]]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* % +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align {{[0-9]+}} % // CHECK: br label // CHECK: ret diff --git a/test/OpenMP/taskloop_lastprivate_messages.cpp b/test/OpenMP/taskloop_lastprivate_messages.cpp index 1d238ea083e63..85aaa00c81899 100644 --- a/test/OpenMP/taskloop_lastprivate_messages.cpp +++ b/test/OpenMP/taskloop_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/taskloop_loop_messages.cpp b/test/OpenMP/taskloop_loop_messages.cpp index edfc2a8dcde55..4ce83b4ae1aeb 100644 --- a/test/OpenMP/taskloop_loop_messages.cpp +++ b/test/OpenMP/taskloop_loop_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s + class S { int a; S() : a(0) {} @@ -292,10 +294,8 @@ int test_iteration_spaces() { c[ii] = a[ii]; #pragma omp parallel -// expected-error@+2 {{unexpected OpenMP clause 'linear' in directive '#pragma omp taskloop'}} -// expected-note@+1 {{defined as linear}} +// expected-error@+1 {{unexpected OpenMP clause 'linear' in directive '#pragma omp taskloop'}} #pragma omp taskloop linear(ii) -// expected-error@+1 {{loop iteration variable in the associated loop of 'omp taskloop' directive may not be linear, predetermined as private}} for (ii = 0; ii < 10; ii++) c[ii] = a[ii]; diff --git a/test/OpenMP/taskloop_misc_messages.c b/test/OpenMP/taskloop_misc_messages.c index f9e084e0edd1c..4e2ac0aab7c69 100644 --- a/test/OpenMP/taskloop_misc_messages.c +++ b/test/OpenMP/taskloop_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -triple x86_64-unknown-unknown -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -triple x86_64-unknown-unknown -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp taskloop'}} #pragma omp taskloop diff --git a/test/OpenMP/taskloop_num_tasks_messages.cpp b/test/OpenMP/taskloop_num_tasks_messages.cpp index 94d74eb41eb02..cf506ff5df922 100644 --- a/test/OpenMP/taskloop_num_tasks_messages.cpp +++ b/test/OpenMP/taskloop_num_tasks_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/taskloop_priority_messages.cpp b/test/OpenMP/taskloop_priority_messages.cpp index 6c1584570eb99..cbfd2579f587f 100644 --- a/test/OpenMP/taskloop_priority_messages.cpp +++ b/test/OpenMP/taskloop_priority_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/taskloop_private_codegen.cpp b/test/OpenMP/taskloop_private_codegen.cpp index 6ca7c402a12b4..21ffcba97d9e6 100644 --- a/test/OpenMP/taskloop_private_codegen.cpp +++ b/test/OpenMP/taskloop_private_codegen.cpp @@ -4,6 +4,14 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics // It doesn't pass on win32. Investigating. // REQUIRES: shell diff --git a/test/OpenMP/taskloop_private_messages.cpp b/test/OpenMP/taskloop_private_messages.cpp index 367d59da69597..ffe731d2c7ef0 100644 --- a/test/OpenMP/taskloop_private_messages.cpp +++ b/test/OpenMP/taskloop_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/taskloop_reduction_codegen.cpp b/test/OpenMP/taskloop_reduction_codegen.cpp index 6af067642fea0..65bc50a4d2553 100644 --- a/test/OpenMP/taskloop_reduction_codegen.cpp +++ b/test/OpenMP/taskloop_reduction_codegen.cpp @@ -1,4 +1,7 @@ // RUN: %clang_cc1 -fopenmp -x c++ %s -verify -debug-info-kind=limited -emit-llvm -o - -triple powerpc64le-unknown-linux-gnu -std=c++98 | FileCheck %s + +// RUN: %clang_cc1 -fopenmp-simd -x c++ %s -verify -debug-info-kind=limited -emit-llvm -o - -triple powerpc64le-unknown-linux-gnu -std=c++98 | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics struct S { @@ -49,7 +52,7 @@ sum = 0.0; // CHECK: [[C:%.*]] = alloca [100 x %struct.S], // CHECK: [[D:%.*]] = alloca float*, // CHECK: [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]], -// CHECK: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%ident_t* +// CHECK: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* // CHECK: [[DOTRD_INPUT_:%.*]] = alloca [4 x %struct.kmp_task_red_input_t], // CHECK: alloca i32, // CHECK: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, @@ -62,21 +65,21 @@ sum = 0.0; // CHECK: [[TMP2:%.*]] = zext i32 [[ADD]] to i64 // CHECK: [[VLA:%.+]] = alloca float, i64 % -// CHECK: call void @__kmpc_taskgroup(%ident_t* +// CHECK: call void @__kmpc_taskgroup(%struct.ident_t* // CHECK-DAG: [[TMP21:%.*]] = bitcast float* [[SUM]] to i8* // CHECK-DAG: store i8* [[TMP21]], i8** [[TMP20:%[^,]+]], // CHECK-DAG: [[TMP20]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T:%.+]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_:%.+]], i32 0, i32 0 // CHECK-DAG: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 1 // CHECK-DAG: store i64 4, i64* [[TMP22]], // CHECK-DAG: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 2 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[RED_INIT1:@.+]] to i8*), i8** [[TMP23]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_INIT1:.+]] to i8*), i8** [[TMP23]], // CHECK-DAG: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 3 // CHECK-DAG: store i8* null, i8** [[TMP24]], // CHECK-DAG: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 4 -// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* [[RED_COMB1:@.+]] to i8*), i8** [[TMP25]], +// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[RED_COMB1:.+]] to i8*), i8** [[TMP25]], // CHECK-DAG: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 5 // CHECK-DAG: [[TMP27:%.*]] = bitcast i32* [[TMP26]] to i8* -// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* [[TMP27]], i8 0, i64 4, i32 8, i1 false) +// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP27]], i8 0, i64 4, i1 false) // CHECK-DAG: [[ARRAYIDX5:%.*]] = getelementptr inbounds [100 x %struct.S], [100 x %struct.S]* [[C]], i64 0, i64 0 // CHECK-DAG: [[LB_ADD_LEN:%.*]] = add nsw i64 -1, % // CHECK-DAG: [[ARRAYIDX6:%.*]] = getelementptr inbounds [100 x %struct.S], [100 x %struct.S]* [[C]], i64 0, i64 [[LB_ADD_LEN]] @@ -92,11 +95,11 @@ sum = 0.0; // CHECK-DAG: store i64 [[TMP37]], i64* [[TMP38:%[^,]+]], // CHECK-DAG: [[TMP38]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 1 // CHECK-DAG: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 2 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[RED_INIT2:@.+]] to i8*), i8** [[TMP39]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_INIT2:.+]] to i8*), i8** [[TMP39]], // CHECK-DAG: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 3 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[RED_FINI2:@.+]] to i8*), i8** [[TMP40]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_FINI2:.+]] to i8*), i8** [[TMP40]], // CHECK-DAG: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 4 -// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* [[RED_COMB2:@.+]] to i8*), i8** [[TMP41]], +// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[RED_COMB2:.+]] to i8*), i8** [[TMP41]], // CHECK-DAG: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 5 // CHECK-DAG: store i32 1, i32* [[TMP42]], // CHECK-DAG: [[TMP44:%.*]] = load float*, float** [[D]], @@ -106,14 +109,14 @@ sum = 0.0; // CHECK-DAG: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 1 // CHECK-DAG: store i64 4, i64* [[TMP46]], // CHECK-DAG: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 2 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[RED_INIT3:@.+]] to i8*), i8** [[TMP47]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_INIT3:.+]] to i8*), i8** [[TMP47]], // CHECK-DAG: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 3 // CHECK-DAG: store i8* null, i8** [[TMP48]], // CHECK-DAG: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 4 -// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* [[RED_COMB3:@.+]] to i8*), i8** [[TMP49]], +// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[RED_COMB3:.+]] to i8*), i8** [[TMP49]], // CHECK-DAG: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 5 // CHECK-DAG: [[TMP51:%.*]] = bitcast i32* [[TMP50]] to i8* -// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* [[TMP51]], i8 0, i64 4, i32 8, i1 false) +// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP51]], i8 0, i64 4, i1 false) // CHECK-DAG: [[TMP53:%.*]] = bitcast float* [[VLA]] to i8* // CHECK-DAG: store i8* [[TMP53]], i8** [[TMP52:%[^,]+]], // CHECK-DAG: [[TMP52]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8:%.+]], i32 0, i32 0 @@ -122,11 +125,11 @@ sum = 0.0; // CHECK-DAG: store i64 [[TMP54]], i64* [[TMP56:%[^,]+]], // CHECK-DAG: [[TMP56]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 1 // CHECK-DAG: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 2 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[RED_INIT4:@.+]] to i8*), i8** [[TMP57]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_INIT4:.+]] to i8*), i8** [[TMP57]], // CHECK-DAG: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 3 // CHECK-DAG: store i8* null, i8** [[TMP58]], // CHECK-DAG: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 4 -// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* [[RED_COMB4:@.+]] to i8*), i8** [[TMP59]], +// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[RED_COMB4:.+]] to i8*), i8** [[TMP59]], // CHECK-DAG: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 5 // CHECK-DAG: store i32 1, i32* [[TMP60]], // CHECK-DAG: [[DOTRD_INPUT_GEP_]] = getelementptr inbounds [4 x %struct.kmp_task_red_input_t], [4 x %struct.kmp_task_red_input_t]* [[DOTRD_INPUT_]], i64 0, i64 @@ -144,56 +147,78 @@ sum = 0.0; // CHECK: [[DIV:%.*]] = sdiv i32 [[ADD11]], 1 // CHECK: [[SUB12:%.*]] = sub nsw i32 [[DIV]], 1 // CHECK: store i32 [[SUB12]], i32* [[DOTCAPTURE_EXPR_9]], -// CHECK: [[TMP65:%.*]] = call i8* @__kmpc_omp_task_alloc(%ident_t* %{{.+}}, i32 [[TMP0]], i32 1, i64 888, i64 72, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* @[[TASK:.+]] to i32 (i32, i8*)*)) -// CHECK: call void @__kmpc_taskloop(%ident_t* %{{.+}}, i32 [[TMP0]], i8* [[TMP65]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* null) -// CHECK: call void @__kmpc_end_taskgroup(%ident_t* +// CHECK: [[TMP65:%.*]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* %{{.+}}, i32 [[TMP0]], i32 1, i64 888, i64 72, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* @[[TASK:.+]] to i32 (i32, i8*)*)) +// CHECK: call void @__kmpc_taskloop(%struct.ident_t* %{{.+}}, i32 [[TMP0]], i8* [[TMP65]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* null) +// CHECK: call void @__kmpc_end_taskgroup(%struct.ident_t* // CHECK: ret i32 -// CHECK: define internal void [[RED_INIT1]](i8*) +// CHECK: define internal void @[[RED_INIT1]](i8*) // CHECK: store float 0.000000e+00, float* % // CHECK: ret void -// CHECK: define internal void [[RED_COMB1]](i8*, i8*) +// CHECK: define internal void @[[RED_COMB1]](i8*, i8*) // CHECK: fadd float % // CHECK: store float %{{.+}}, float* % // CHECK: ret void -// CHECK: define internal void [[RED_INIT2]](i8*) -// CHECK: call i8* @__kmpc_threadprivate_cached( +// CHECK: define internal void @[[RED_INIT2]](i8*) // CHECK: call i8* @__kmpc_threadprivate_cached( +// CHECK: [[ORIG_PTR_ADDR:%.+]] = call i8* @__kmpc_threadprivate_cached( +// CHECK: [[ORIG_PTR_REF:%.+]] = bitcast i8* [[ORIG_PTR_ADDR]] to i8** +// CHECK: load i8*, i8** [[ORIG_PTR_REF]], // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64( // CHECK: ret void -// CHECK: define internal void [[RED_FINI2]](i8*) +// CHECK: define internal void @[[RED_FINI2]](i8*) // CHECK: call i8* @__kmpc_threadprivate_cached( // CHECK: call void @ // CHECK: ret void -// CHECK: define internal void [[RED_COMB2]](i8*, i8*) +// CHECK: define internal void @[[RED_COMB2]](i8*, i8*) // CHECK: call i8* @__kmpc_threadprivate_cached( // CHECK: fadd float % // CHECK: store float %{{.+}}, float* % // CHECK: ret void -// CHECK: define internal void [[RED_INIT3]](i8*) +// CHECK: define internal void @[[RED_INIT3]](i8*) // CHECK: store float 0.000000e+00, float* % // CHECK: ret void -// CHECK: define internal void [[RED_COMB3]](i8*, i8*) +// CHECK: define internal void @[[RED_COMB3]](i8*, i8*) // CHECK: fadd float % // CHECK: store float %{{.+}}, float* % // CHECK: ret void -// CHECK: define internal void [[RED_INIT4]](i8*) +// CHECK: define internal void @[[RED_INIT4]](i8*) // CHECK: call i8* @__kmpc_threadprivate_cached( // CHECK: store float 0.000000e+00, float* % // CHECK: ret void -// CHECK: define internal void [[RED_COMB4]](i8*, i8*) +// CHECK: define internal void @[[RED_COMB4]](i8*, i8*) // CHECK: call i8* @__kmpc_threadprivate_cached( // CHECK: fadd float % // CHECK: store float %{{.+}}, float* % // CHECK: ret void -// CHECK: distinct !DISubprogram(linkageName: "[[TASK]]", scope: ! +// CHECK-NOT: call i8* @__kmpc_threadprivate_cached( +// CHECK: call i8* @__kmpc_task_reduction_get_th_data( +// CHECK: call i8* @__kmpc_threadprivate_cached( +// CHECK: call i8* @__kmpc_threadprivate_cached( +// CHECK: call i8* @__kmpc_task_reduction_get_th_data( +// CHECK-NOT: call i8* @__kmpc_threadprivate_cached( +// CHECK: call i8* @__kmpc_task_reduction_get_th_data( +// CHECK: call i8* @__kmpc_threadprivate_cached( +// CHECK: call i8* @__kmpc_task_reduction_get_th_data( +// CHECK-NOT: call i8* @__kmpc_threadprivate_cached( + +// CHECK-DAG: distinct !DISubprogram(linkageName: "[[TASK]]", scope: ! +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_INIT1]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_COMB1]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_INIT2]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_FINI2]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_COMB2]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_INIT3]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_COMB3]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_INIT4]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_COMB4]]" diff --git a/test/OpenMP/taskloop_reduction_messages.cpp b/test/OpenMP/taskloop_reduction_messages.cpp index 23f70b5685252..c83a8b30c52c8 100644 --- a/test/OpenMP/taskloop_reduction_messages.cpp +++ b/test/OpenMP/taskloop_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/taskloop_simd_aligned_messages.cpp b/test/OpenMP/taskloop_simd_aligned_messages.cpp index c4c41ebbafe48..24cdd00e6b347 100644 --- a/test/OpenMP/taskloop_simd_aligned_messages.cpp +++ b/test/OpenMP/taskloop_simd_aligned_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp %s +// RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp-simd %s + struct B { static int ib[20]; // expected-note 0 {{'B::ib' declared here}} static constexpr int bfoo() { return 8; } diff --git a/test/OpenMP/taskloop_simd_ast_print.cpp b/test/OpenMP/taskloop_simd_ast_print.cpp index 6f13c7e0baa5e..0a847f8fad060 100644 --- a/test/OpenMP/taskloop_simd_ast_print.cpp +++ b/test/OpenMP/taskloop_simd_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -17,7 +21,7 @@ T tmain(T argc) { #pragma omp taskgroup task_reduction(+: d) #pragma omp taskloop simd if(taskloop: argc > N) default(shared) untied priority(N) safelen(N) linear(c) aligned(ptr) grainsize(N) reduction(+:g) in_reduction(+: d) // CHECK-NEXT: #pragma omp taskgroup task_reduction(+: d) - // CHECK-NEXT: #pragma omp taskloop simd if(taskloop: argc > N) default(shared) untied priority(N) safelen(N) linear(c) aligned(ptr) grainsize(N) reduction(+: g) in_reduction(+: d) + // CHECK-NEXT: #pragma omp taskloop simd if(taskloop: argc > N) default(shared) untied priority(N) safelen(N) linear(c) aligned(ptr) grainsize(N) reduction(+: g) in_reduction(+: d){{$}} for (int i = 0; i < 2; ++i) a = 2; // CHECK-NEXT: for (int i = 0; i < 2; ++i) diff --git a/test/OpenMP/taskloop_simd_codegen.cpp b/test/OpenMP/taskloop_simd_codegen.cpp index 006fd59af3fa2..d3efec5ee441d 100644 --- a/test/OpenMP/taskloop_simd_codegen.cpp +++ b/test/OpenMP/taskloop_simd_codegen.cpp @@ -1,15 +1,20 @@ // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -x c++ -emit-llvm %s -o - -femit-all-decls | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls | FileCheck %s + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -x c++ -emit-llvm %s -o - -femit-all-decls | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - -femit-all-decls | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER // CHECK-LABEL: @main int main(int argc, char **argv) { -// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%ident_t* [[DEFLOC:@.+]]) -// CHECK: call void @__kmpc_taskgroup(%ident_t* [[DEFLOC]], i32 [[GTID]]) -// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 33, i64 80, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK1:@.+]] to i32 (i32, i8*)*)) +// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* [[DEFLOC:@.+]]) +// CHECK: call void @__kmpc_taskgroup(%struct.ident_t* [[DEFLOC]], i32 [[GTID]]) +// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i32 33, i64 80, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK1:@.+]] to i32 (i32, i8*)*)) // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]* // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0 // CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5 @@ -19,12 +24,12 @@ int main(int argc, char **argv) { // CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 7 // CHECK: store i64 1, i64* [[ST]], // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]], -// CHECK: call void @__kmpc_taskloop(%ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 1, i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 0, i64 0, i8* null) -// CHECK: call void @__kmpc_end_taskgroup(%ident_t* [[DEFLOC]], i32 [[GTID]]) +// CHECK: call void @__kmpc_taskloop(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 1, i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 0, i64 0, i8* null) +// CHECK: call void @__kmpc_end_taskgroup(%struct.ident_t* [[DEFLOC]], i32 [[GTID]]) #pragma omp taskloop simd priority(argc) for (int i = 0; i < 10; ++i) ; -// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 80, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK2:@.+]] to i32 (i32, i8*)*)) +// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 80, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK2:@.+]] to i32 (i32, i8*)*)) // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]* // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0 // CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5 @@ -35,12 +40,12 @@ int main(int argc, char **argv) { // CHECK: store i64 1, i64* [[ST]], // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]], // CHECK: [[GRAINSIZE:%.+]] = zext i32 %{{.+}} to i64 -// CHECK: call void @__kmpc_taskloop(%ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 1, i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 1, i64 [[GRAINSIZE]], i8* null) +// CHECK: call void @__kmpc_taskloop(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 1, i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 1, i64 [[GRAINSIZE]], i8* null) #pragma omp taskloop simd nogroup grainsize(argc) simdlen(4) for (int i = 0; i < 10; ++i) ; -// CHECK: call void @__kmpc_taskgroup(%ident_t* [[DEFLOC]], i32 [[GTID]]) -// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 80, i64 24, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK3:@.+]] to i32 (i32, i8*)*)) +// CHECK: call void @__kmpc_taskgroup(%struct.ident_t* [[DEFLOC]], i32 [[GTID]]) +// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 80, i64 24, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK3:@.+]] to i32 (i32, i8*)*)) // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]* // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0 // CHECK: [[IF:%.+]] = icmp ne i32 %{{.+}}, 0 @@ -52,8 +57,8 @@ int main(int argc, char **argv) { // CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 7 // CHECK: store i64 1, i64* [[ST]], // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]], -// CHECK: call void @__kmpc_taskloop(%ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 [[IF_INT]], i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 2, i64 4, i8* null) -// CHECK: call void @__kmpc_end_taskgroup(%ident_t* [[DEFLOC]], i32 [[GTID]]) +// CHECK: call void @__kmpc_taskloop(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 [[IF_INT]], i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 2, i64 4, i8* null) +// CHECK: call void @__kmpc_end_taskgroup(%struct.ident_t* [[DEFLOC]], i32 [[GTID]]) int i; #pragma omp taskloop simd if(argc) shared(argc, argv) collapse(2) num_tasks(4) safelen(32) for (i = 0; i < argc; ++i) @@ -145,8 +150,8 @@ int main(int argc, char **argv) { struct S { int a; S(int c) { -// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%ident_t* [[DEFLOC:@.+]]) -// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 80, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK4:@.+]] to i32 (i32, i8*)*)) +// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* [[DEFLOC:@.+]]) +// CHECK: [[TASKV:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i32 1, i64 80, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, [[TDP_TY:%.+]]*)* [[TASK4:@.+]] to i32 (i32, i8*)*)) // CHECK: [[TASK:%.+]] = bitcast i8* [[TASKV]] to [[TDP_TY]]* // CHECK: [[TASK_DATA:%.+]] = getelementptr inbounds [[TDP_TY]], [[TDP_TY]]* [[TASK]], i32 0, i32 0 // CHECK: [[DOWN:%.+]] = getelementptr inbounds [[TD_TY:%.+]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 5 @@ -157,7 +162,7 @@ struct S { // CHECK: store i64 1, i64* [[ST]], // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]], // CHECK: [[NUM_TASKS:%.+]] = zext i32 %{{.+}} to i64 -// CHECK: call void @__kmpc_taskloop(%ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 1, i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 2, i64 [[NUM_TASKS]], i8* null) +// CHECK: call void @__kmpc_taskloop(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 1, i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 0, i32 2, i64 [[NUM_TASKS]], i8* null) #pragma omp taskloop simd shared(c) num_tasks(a) simdlen(8) safelen(64) for (a = 0; a < c; ++a) ; diff --git a/test/OpenMP/taskloop_simd_collapse_messages.cpp b/test/OpenMP/taskloop_simd_collapse_messages.cpp index e4ce0c1df41ab..3f7e66a277021 100644 --- a/test/OpenMP/taskloop_simd_collapse_messages.cpp +++ b/test/OpenMP/taskloop_simd_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/taskloop_simd_final_messages.cpp b/test/OpenMP/taskloop_simd_final_messages.cpp index c6580fbaa9b45..b7393661c940d 100644 --- a/test/OpenMP/taskloop_simd_final_messages.cpp +++ b/test/OpenMP/taskloop_simd_final_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/taskloop_simd_firstprivate_codegen.cpp b/test/OpenMP/taskloop_simd_firstprivate_codegen.cpp index 9e268d8a1739a..4f406b2a583ab 100644 --- a/test/OpenMP/taskloop_simd_firstprivate_codegen.cpp +++ b/test/OpenMP/taskloop_simd_firstprivate_codegen.cpp @@ -4,6 +4,14 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics // It doesn't pass on win32. // REQUIRES: shell @@ -198,7 +206,7 @@ int main() { // CHECK: [[SHAREDS_REF_ADDR:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_ADDR]], // CHECK: [[CAPTURES_ADDR:%.+]] = bitcast [[CAP_MAIN_TY]]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[SHAREDS_REF]], i8* [[CAPTURES_ADDR]], i64 40, i32 8, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[SHAREDS_REF]], i8* align 8 [[CAPTURES_ADDR]], i64 40, i1 false) // Initialize kmp_task_t->privates with default values (no init for simple types, default constructors for classes). // Also copy address of private copy to the corresponding shareds reference. @@ -368,7 +376,7 @@ int main() { // CHECK: [[SHAREDS_REF_ADDR:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_ADDR]], // CHECK: [[CAPTURES_ADDR:%.+]] = bitcast [[CAP_TMAIN_TY]]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[SHAREDS_REF]], i8* [[CAPTURES_ADDR]], i64 32, i32 8, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[SHAREDS_REF]], i8* align 8 [[CAPTURES_ADDR]], i64 32, i1 false) // Initialize kmp_task_t->privates with default values (no init for simple types, default constructors for classes). // CHECK: [[PRIVATES:%.+]] = getelementptr inbounds [[KMP_TASK_TMAIN_TY]], [[KMP_TASK_TMAIN_TY]]* [[RES_KMP_TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 diff --git a/test/OpenMP/taskloop_simd_firstprivate_messages.cpp b/test/OpenMP/taskloop_simd_firstprivate_messages.cpp index 176dcd7f8f9cb..66c9e7a2f8fff 100644 --- a/test/OpenMP/taskloop_simd_firstprivate_messages.cpp +++ b/test/OpenMP/taskloop_simd_firstprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/taskloop_simd_grainsize_messages.cpp b/test/OpenMP/taskloop_simd_grainsize_messages.cpp index 45ccb31824fa5..3441e59e56d05 100644 --- a/test/OpenMP/taskloop_simd_grainsize_messages.cpp +++ b/test/OpenMP/taskloop_simd_grainsize_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/taskloop_simd_in_reduction_codegen.cpp b/test/OpenMP/taskloop_simd_in_reduction_codegen.cpp index d894943cc3598..a197b4613e884 100644 --- a/test/OpenMP/taskloop_simd_in_reduction_codegen.cpp +++ b/test/OpenMP/taskloop_simd_in_reduction_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -x c++ -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -34,18 +39,18 @@ int main(int argc, char **argv) { } // CHECK-LABEL: @main -// CHECK: void @__kmpc_taskgroup(%ident_t* @0, i32 [[GTID:%.+]]) +// CHECK: void @__kmpc_taskgroup(%struct.ident_t* @0, i32 [[GTID:%.+]]) // CHECK: [[TD1:%.+]] = call i8* @__kmpc_task_reduction_init(i32 [[GTID]], i32 3, i8* % // CHECK-NEXT: store i8* [[TD1]], i8** [[TD1_ADDR:%[^,]+]], -// CHECK-NEXT: call void @__kmpc_taskgroup(%ident_t* @0, i32 [[GTID]]) +// CHECK-NEXT: call void @__kmpc_taskgroup(%struct.ident_t* @0, i32 [[GTID]]) // CHECK: [[TD2:%.+]] = call i8* @__kmpc_task_reduction_init(i32 [[GTID]], i32 2, i8* % // CHECK-NEXT: store i8* [[TD2]], i8** [[TD2_ADDR:%[^,]+]], -// CHECK-NEXT: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%ident_t* @0, i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i64, i16*, i8**, i8**)* [[OMP_PARALLEL:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i64 %{{.+}}, i16* %{{.+}}, i8** [[TD1_ADDR]], i8** [[TD2_ADDR]]) -// CHECK-NEXT: call void @__kmpc_end_taskgroup(%ident_t* @0, i32 [[GTID]]) -// CHECK-NEXT: call void @__kmpc_end_taskgroup(%ident_t* @0, i32 [[GTID]]) +// CHECK-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @0, i32 5, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i64, i16*, i8**, i8**)* [[OMP_PARALLEL:@.+]] to void (i32*, i32*, ...)*), i32* %{{.+}}, i64 %{{.+}}, i16* %{{.+}}, i8** [[TD1_ADDR]], i8** [[TD2_ADDR]]) +// CHECK-NEXT: call void @__kmpc_end_taskgroup(%struct.ident_t* @0, i32 [[GTID]]) +// CHECK-NEXT: call void @__kmpc_end_taskgroup(%struct.ident_t* @0, i32 [[GTID]]) // CHECK: define internal void [[OMP_PARALLEL]]( -// CHECK: [[TASK_T:%.+]] = call i8* @__kmpc_omp_task_alloc(%ident_t* @0, i32 [[GTID:%.+]], i32 1, i64 96, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[T:%.+]]*)* [[OMP_TASK:@.+]] to i32 (i32, i8*)*)) +// CHECK: [[TASK_T:%.+]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* @0, i32 [[GTID:%.+]], i32 1, i64 96, i64 40, i32 (i32, i8*)* bitcast (i32 (i32, [[T:%.+]]*)* [[OMP_TASK:@.+]] to i32 (i32, i8*)*)) // CHECK-NEXT: [[TASK_T_WITH_PRIVS:%.+]] = bitcast i8* [[TASK_T]] to [[T]]* // CHECK: [[PRIVS:%.+]] = getelementptr inbounds [[T]], [[T]]* [[TASK_T_WITH_PRIVS]], i32 0, i32 1 // CHECK: [[TD1_REF:%.+]] = getelementptr inbounds [[PRIVATES]], [[PRIVATES]]* [[PRIVS]], i32 0, i32 0 @@ -58,7 +63,7 @@ int main(int argc, char **argv) { // CHECK-NEXT: [[TD2_ADDR:%.+]] = load i8**, i8*** [[TD2_SHAR]], // CHECK-NEXT: [[TD2:%.+]] = load i8*, i8** [[TD2_ADDR]], // CHECK-NEXT: store i8* [[TD2]], i8** [[TD2_REF]], -// CHECK: call void @__kmpc_taskloop(%ident_t* @0, i32 [[GTID]], i8* [[TASK_T]], i32 1, +// CHECK: call void @__kmpc_taskloop(%struct.ident_t* @0, i32 [[GTID]], i8* [[TASK_T]], i32 1, // CHECK: ret void // CHECK-NEXT: } diff --git a/test/OpenMP/taskloop_simd_in_reduction_messages.cpp b/test/OpenMP/taskloop_simd_in_reduction_messages.cpp index f3b1a855e3209..ee3d093766108 100644 --- a/test/OpenMP/taskloop_simd_in_reduction_messages.cpp +++ b/test/OpenMP/taskloop_simd_in_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/taskloop_simd_lastprivate_codegen.cpp b/test/OpenMP/taskloop_simd_lastprivate_codegen.cpp index 8d5ebb216808e..be484465e3aea 100644 --- a/test/OpenMP/taskloop_simd_lastprivate_codegen.cpp +++ b/test/OpenMP/taskloop_simd_lastprivate_codegen.cpp @@ -4,6 +4,14 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics // It doesn't pass on win32. // REQUIRES: shell @@ -190,7 +198,7 @@ int main() { // CHECK: [[SHAREDS_REF_ADDR:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_ADDR]], // CHECK: [[CAPTURES_ADDR:%.+]] = bitcast [[CAP_MAIN_TY]]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[SHAREDS_REF]], i8* [[CAPTURES_ADDR]], i64 40, i32 8, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[SHAREDS_REF]], i8* align 8 [[CAPTURES_ADDR]], i64 40, i1 false) // Initialize kmp_task_t->privates with default values (no init for simple types, default constructors for classes). // Also copy address of private copy to the corresponding shareds reference. @@ -279,18 +287,18 @@ int main() { // CHECK-NEXT: br i1 // CHECK: bitcast [[S_DOUBLE_TY]]* %{{.+}} to i8* // CHECK: bitcast [[S_DOUBLE_TY]]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* % +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align {{[0-9]+}} % // CHECK: load i32, i32* % // CHECK: store i32 %{{.+}}, i32* % // CHECK: getelementptr inbounds [2 x [[S_DOUBLE_TY]]], [2 x [[S_DOUBLE_TY]]]* % // CHECK: phi [[S_DOUBLE_TY]]* // CHECK: phi [[S_DOUBLE_TY]]* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* % +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align {{[0-9]+}} % // CHECK: icmp eq [[S_DOUBLE_TY]]* % // CHECK-NEXT: br i1 // CHECK: bitcast [2 x i32]* %{{.+}} to i8* // CHECK: bitcast [2 x i32]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* % +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align {{[0-9]+}} % // CHECK: load i32, i32* % // CHECK: store i32 %{{.+}}, i32* % // CHECK: br label @@ -364,7 +372,7 @@ int main() { // CHECK: [[SHAREDS_REF_ADDR:%.+]] = getelementptr inbounds [[KMP_TASK_T_TY]], [[KMP_TASK_T_TY]]* [[TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_ADDR]], // CHECK: [[CAPTURES_ADDR:%.+]] = bitcast [[CAP_TMAIN_TY]]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[SHAREDS_REF]], i8* [[CAPTURES_ADDR]], i64 32, i32 8, i1 false) +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[SHAREDS_REF]], i8* align 8 [[CAPTURES_ADDR]], i64 32, i1 false) // Initialize kmp_task_t->privates with default values (no init for simple types, default constructors for classes). // CHECK: [[PRIVATES:%.+]] = getelementptr inbounds [[KMP_TASK_TMAIN_TY]], [[KMP_TASK_TMAIN_TY]]* [[RES_KMP_TASK]], i{{[0-9]+}} 0, i{{[0-9]+}} 2 @@ -445,16 +453,16 @@ int main() { // CHECK: store i32 %{{.+}}, i32* % // CHECK: bitcast [2 x i32]* %{{.+}} to i8* // CHECK: bitcast [2 x i32]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* % +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align {{[0-9]+}} % // CHECK: getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* % // CHECK: phi [[S_INT_TY]]* // CHECK: phi [[S_INT_TY]]* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* % +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align {{[0-9]+}} % // CHECK: icmp eq [[S_INT_TY]]* % // CHECK-NEXT: br i1 // CHECK: bitcast [[S_INT_TY]]* %{{.+}} to i8* // CHECK: bitcast [[S_INT_TY]]* %{{.+}} to i8* -// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* % +// CHECK: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align {{[0-9]+}} % // CHECK: br label // CHECK: ret diff --git a/test/OpenMP/taskloop_simd_lastprivate_messages.cpp b/test/OpenMP/taskloop_simd_lastprivate_messages.cpp index e5cba6c64b6bc..8f418ff3b8837 100644 --- a/test/OpenMP/taskloop_simd_lastprivate_messages.cpp +++ b/test/OpenMP/taskloop_simd_lastprivate_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/taskloop_simd_linear_messages.cpp b/test/OpenMP/taskloop_simd_linear_messages.cpp index c9a82b9dc1488..2aea6a041fce4 100644 --- a/test/OpenMP/taskloop_simd_linear_messages.cpp +++ b/test/OpenMP/taskloop_simd_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/taskloop_simd_loop_messages.cpp b/test/OpenMP/taskloop_simd_loop_messages.cpp index 502224629a56b..71aa9dfe1143b 100644 --- a/test/OpenMP/taskloop_simd_loop_messages.cpp +++ b/test/OpenMP/taskloop_simd_loop_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s + class S { int a; S() : a(0) {} diff --git a/test/OpenMP/taskloop_simd_misc_messages.c b/test/OpenMP/taskloop_simd_misc_messages.c index 35cc62cad5857..bb4b2dfbb85e9 100644 --- a/test/OpenMP/taskloop_simd_misc_messages.c +++ b/test/OpenMP/taskloop_simd_misc_messages.c @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -triple x86_64-unknown-unknown -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -triple x86_64-unknown-unknown -verify %s + // expected-error@+1 {{unexpected OpenMP directive '#pragma omp taskloop simd'}} #pragma omp taskloop simd diff --git a/test/OpenMP/taskloop_simd_num_tasks_messages.cpp b/test/OpenMP/taskloop_simd_num_tasks_messages.cpp index f26ee79a87401..3267d294b7569 100644 --- a/test/OpenMP/taskloop_simd_num_tasks_messages.cpp +++ b/test/OpenMP/taskloop_simd_num_tasks_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/taskloop_simd_priority_messages.cpp b/test/OpenMP/taskloop_simd_priority_messages.cpp index a94798f821a6b..44e0b57a00d33 100644 --- a/test/OpenMP/taskloop_simd_priority_messages.cpp +++ b/test/OpenMP/taskloop_simd_priority_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/taskloop_simd_private_codegen.cpp b/test/OpenMP/taskloop_simd_private_codegen.cpp index 3833b5ea9fb36..fd0820ad84a62 100644 --- a/test/OpenMP/taskloop_simd_private_codegen.cpp +++ b/test/OpenMP/taskloop_simd_private_codegen.cpp @@ -4,6 +4,14 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=LAMBDA %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=BLOCKS %s // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck -check-prefix=ARRAY %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics // It doesn't pass on win32. Investigating. // REQUIRES: shell diff --git a/test/OpenMP/taskloop_simd_private_messages.cpp b/test/OpenMP/taskloop_simd_private_messages.cpp index ba9e8da17e205..68e536525122e 100644 --- a/test/OpenMP/taskloop_simd_private_messages.cpp +++ b/test/OpenMP/taskloop_simd_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/taskloop_simd_reduction_codegen.cpp b/test/OpenMP/taskloop_simd_reduction_codegen.cpp index 7e8d54a2a1da6..6c34e34ca5840 100644 --- a/test/OpenMP/taskloop_simd_reduction_codegen.cpp +++ b/test/OpenMP/taskloop_simd_reduction_codegen.cpp @@ -1,4 +1,7 @@ // RUN: %clang_cc1 -fopenmp -x c++ %s -verify -debug-info-kind=limited -emit-llvm -o - -triple powerpc64le-unknown-linux-gnu -std=c++98 | FileCheck %s + +// RUN: %clang_cc1 -fopenmp-simd -x c++ %s -verify -debug-info-kind=limited -emit-llvm -o - -triple powerpc64le-unknown-linux-gnu -std=c++98 | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics struct S { @@ -49,7 +52,7 @@ sum = 0.0; // CHECK: [[C:%.*]] = alloca [100 x %struct.S], // CHECK: [[D:%.*]] = alloca float*, // CHECK: [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]], -// CHECK: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%ident_t* +// CHECK: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* // CHECK: [[DOTRD_INPUT_:%.*]] = alloca [4 x %struct.kmp_task_red_input_t], // CHECK: alloca i32, // CHECK: [[DOTCAPTURE_EXPR_:%.*]] = alloca i32, @@ -62,21 +65,21 @@ sum = 0.0; // CHECK: [[TMP2:%.*]] = zext i32 [[ADD]] to i64 // CHECK: [[VLA:%.+]] = alloca float, i64 % -// CHECK: call void @__kmpc_taskgroup(%ident_t* +// CHECK: call void @__kmpc_taskgroup(%struct.ident_t* // CHECK-DAG: [[TMP21:%.*]] = bitcast float* [[SUM]] to i8* // CHECK-DAG: store i8* [[TMP21]], i8** [[TMP20:%[^,]+]], // CHECK-DAG: [[TMP20]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T:%.+]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_:%.+]], i32 0, i32 0 // CHECK-DAG: [[TMP22:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 1 // CHECK-DAG: store i64 4, i64* [[TMP22]], // CHECK-DAG: [[TMP23:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 2 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[RED_INIT1:@.+]] to i8*), i8** [[TMP23]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_INIT1:.+]] to i8*), i8** [[TMP23]], // CHECK-DAG: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 3 // CHECK-DAG: store i8* null, i8** [[TMP24]], // CHECK-DAG: [[TMP25:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 4 -// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* [[RED_COMB1:@.+]] to i8*), i8** [[TMP25]], +// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[RED_COMB1:.+]] to i8*), i8** [[TMP25]], // CHECK-DAG: [[TMP26:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_]], i32 0, i32 5 // CHECK-DAG: [[TMP27:%.*]] = bitcast i32* [[TMP26]] to i8* -// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* [[TMP27]], i8 0, i64 4, i32 8, i1 false) +// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP27]], i8 0, i64 4, i1 false) // CHECK-DAG: [[ARRAYIDX5:%.*]] = getelementptr inbounds [100 x %struct.S], [100 x %struct.S]* [[C]], i64 0, i64 0 // CHECK-DAG: [[LB_ADD_LEN:%.*]] = add nsw i64 -1, % // CHECK-DAG: [[ARRAYIDX6:%.*]] = getelementptr inbounds [100 x %struct.S], [100 x %struct.S]* [[C]], i64 0, i64 [[LB_ADD_LEN]] @@ -92,11 +95,11 @@ sum = 0.0; // CHECK-DAG: store i64 [[TMP37]], i64* [[TMP38:%[^,]+]], // CHECK-DAG: [[TMP38]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 1 // CHECK-DAG: [[TMP39:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 2 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[RED_INIT2:@.+]] to i8*), i8** [[TMP39]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_INIT2:.+]] to i8*), i8** [[TMP39]], // CHECK-DAG: [[TMP40:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 3 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[RED_FINI2:@.+]] to i8*), i8** [[TMP40]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_FINI2:.+]] to i8*), i8** [[TMP40]], // CHECK-DAG: [[TMP41:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 4 -// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* [[RED_COMB2:@.+]] to i8*), i8** [[TMP41]], +// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[RED_COMB2:.+]] to i8*), i8** [[TMP41]], // CHECK-DAG: [[TMP42:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_4]], i32 0, i32 5 // CHECK-DAG: store i32 1, i32* [[TMP42]], // CHECK-DAG: [[TMP44:%.*]] = load float*, float** [[D]], @@ -106,14 +109,14 @@ sum = 0.0; // CHECK-DAG: [[TMP46:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 1 // CHECK-DAG: store i64 4, i64* [[TMP46]], // CHECK-DAG: [[TMP47:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 2 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[RED_INIT3:@.+]] to i8*), i8** [[TMP47]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_INIT3:.+]] to i8*), i8** [[TMP47]], // CHECK-DAG: [[TMP48:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 3 // CHECK-DAG: store i8* null, i8** [[TMP48]], // CHECK-DAG: [[TMP49:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 4 -// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* [[RED_COMB3:@.+]] to i8*), i8** [[TMP49]], +// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[RED_COMB3:.+]] to i8*), i8** [[TMP49]], // CHECK-DAG: [[TMP50:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_7]], i32 0, i32 5 // CHECK-DAG: [[TMP51:%.*]] = bitcast i32* [[TMP50]] to i8* -// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* [[TMP51]], i8 0, i64 4, i32 8, i1 false) +// CHECK-DAG: call void @llvm.memset.p0i8.i64(i8* align 8 [[TMP51]], i8 0, i64 4, i1 false) // CHECK-DAG: [[TMP53:%.*]] = bitcast float* [[VLA]] to i8* // CHECK-DAG: store i8* [[TMP53]], i8** [[TMP52:%[^,]+]], // CHECK-DAG: [[TMP52]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8:%.+]], i32 0, i32 0 @@ -122,11 +125,11 @@ sum = 0.0; // CHECK-DAG: store i64 [[TMP54]], i64* [[TMP56:%[^,]+]], // CHECK-DAG: [[TMP56]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 1 // CHECK-DAG: [[TMP57:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 2 -// CHECK-DAG: store i8* bitcast (void (i8*)* [[RED_INIT4:@.+]] to i8*), i8** [[TMP57]], +// CHECK-DAG: store i8* bitcast (void (i8*)* @[[RED_INIT4:.+]] to i8*), i8** [[TMP57]], // CHECK-DAG: [[TMP58:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 3 // CHECK-DAG: store i8* null, i8** [[TMP58]], // CHECK-DAG: [[TMP59:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 4 -// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* [[RED_COMB4:@.+]] to i8*), i8** [[TMP59]], +// CHECK-DAG: store i8* bitcast (void (i8*, i8*)* @[[RED_COMB4:.+]] to i8*), i8** [[TMP59]], // CHECK-DAG: [[TMP60:%.*]] = getelementptr inbounds [[STRUCT_KMP_TASK_RED_INPUT_T]], %struct.kmp_task_red_input_t* [[DOTRD_INPUT_GEP_8]], i32 0, i32 5 // CHECK-DAG: store i32 1, i32* [[TMP60]], // CHECK-DAG: [[DOTRD_INPUT_GEP_]] = getelementptr inbounds [4 x %struct.kmp_task_red_input_t], [4 x %struct.kmp_task_red_input_t]* [[DOTRD_INPUT_]], i64 0, i64 @@ -144,55 +147,64 @@ sum = 0.0; // CHECK: [[DIV:%.*]] = sdiv i32 [[ADD11]], 1 // CHECK: [[SUB12:%.*]] = sub nsw i32 [[DIV]], 1 // CHECK: store i32 [[SUB12]], i32* [[DOTCAPTURE_EXPR_9]], -// CHECK: [[TMP65:%.*]] = call i8* @__kmpc_omp_task_alloc(%ident_t* %{{.+}}, i32 [[TMP0]], i32 1, i64 888, i64 72, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* @{{.+}} to i32 (i32, i8*)*)) -// CHECK: call void @__kmpc_taskloop(%ident_t* %{{.+}}, i32 [[TMP0]], i8* [[TMP65]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* null) -// CHECK: call void @__kmpc_end_taskgroup(%ident_t* +// CHECK: [[TMP65:%.*]] = call i8* @__kmpc_omp_task_alloc(%struct.ident_t* %{{.+}}, i32 [[TMP0]], i32 1, i64 888, i64 72, i32 (i32, i8*)* bitcast (i32 (i32, %struct.kmp_task_t_with_privates*)* @{{.+}} to i32 (i32, i8*)*)) +// CHECK: call void @__kmpc_taskloop(%struct.ident_t* %{{.+}}, i32 [[TMP0]], i8* [[TMP65]], i32 1, i64* %{{.+}}, i64* %{{.+}}, i64 %{{.+}}, i32 0, i32 0, i64 0, i8* null) +// CHECK: call void @__kmpc_end_taskgroup(%struct.ident_t* // CHECK: ret i32 -// CHECK: define internal void [[RED_INIT1]](i8*) +// CHECK: define internal void @[[RED_INIT1]](i8*) // CHECK: store float 0.000000e+00, float* % // CHECK: ret void -// CHECK: define internal void [[RED_COMB1]](i8*, i8*) +// CHECK: define internal void @[[RED_COMB1]](i8*, i8*) // CHECK: fadd float % // CHECK: store float %{{.+}}, float* % // CHECK: ret void -// CHECK: define internal void [[RED_INIT2]](i8*) +// CHECK: define internal void @[[RED_INIT2]](i8*) // CHECK: call i8* @__kmpc_threadprivate_cached( // CHECK: call i8* @__kmpc_threadprivate_cached( // CHECK: call void @llvm.memcpy.p0i8.p0i8.i64( // CHECK: ret void -// CHECK: define internal void [[RED_FINI2]](i8*) +// CHECK: define internal void @[[RED_FINI2]](i8*) // CHECK: call i8* @__kmpc_threadprivate_cached( // CHECK: call void @ // CHECK: ret void -// CHECK: define internal void [[RED_COMB2]](i8*, i8*) +// CHECK: define internal void @[[RED_COMB2]](i8*, i8*) // CHECK: call i8* @__kmpc_threadprivate_cached( // CHECK: fadd float % // CHECK: store float %{{.+}}, float* % // CHECK: ret void -// CHECK: define internal void [[RED_INIT3]](i8*) +// CHECK: define internal void @[[RED_INIT3]](i8*) // CHECK: store float 0.000000e+00, float* % // CHECK: ret void -// CHECK: define internal void [[RED_COMB3]](i8*, i8*) +// CHECK: define internal void @[[RED_COMB3]](i8*, i8*) // CHECK: fadd float % // CHECK: store float %{{.+}}, float* % // CHECK: ret void -// CHECK: define internal void [[RED_INIT4]](i8*) +// CHECK: define internal void @[[RED_INIT4]](i8*) // CHECK: call i8* @__kmpc_threadprivate_cached( // CHECK: store float 0.000000e+00, float* % // CHECK: ret void -// CHECK: define internal void [[RED_COMB4]](i8*, i8*) +// CHECK: define internal void @[[RED_COMB4]](i8*, i8*) // CHECK: call i8* @__kmpc_threadprivate_cached( // CHECK: fadd float % // CHECK: store float %{{.+}}, float* % // CHECK: ret void +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_INIT1]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_COMB1]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_INIT2]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_FINI2]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_COMB2]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_INIT3]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_COMB3]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_INIT4]]" +// CHECK-DAG: !DISubprogram(linkageName: "[[RED_COMB4]]" diff --git a/test/OpenMP/taskloop_simd_reduction_messages.cpp b/test/OpenMP/taskloop_simd_reduction_messages.cpp index 886b685f6325a..10ee5ce8c83cf 100644 --- a/test/OpenMP/taskloop_simd_reduction_messages.cpp +++ b/test/OpenMP/taskloop_simd_reduction_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 -ferror-limit 150 -o - %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -ferror-limit 150 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 150 -o - %s + void foo() { } diff --git a/test/OpenMP/taskloop_simd_safelen_messages.cpp b/test/OpenMP/taskloop_simd_safelen_messages.cpp index 729f31407bfaa..87c4fe07ab028 100644 --- a/test/OpenMP/taskloop_simd_safelen_messages.cpp +++ b/test/OpenMP/taskloop_simd_safelen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/taskloop_simd_simdlen_messages.cpp b/test/OpenMP/taskloop_simd_simdlen_messages.cpp index 79655ba9c275b..4170d012dd5de 100644 --- a/test/OpenMP/taskloop_simd_simdlen_messages.cpp +++ b/test/OpenMP/taskloop_simd_simdlen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/taskwait_ast_print.cpp b/test/OpenMP/taskwait_ast_print.cpp index 1a69b58a37ba8..9cbc46ef29f9e 100644 --- a/test/OpenMP/taskwait_ast_print.cpp +++ b/test/OpenMP/taskwait_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -15,7 +19,7 @@ T tmain(T argc) { return a + argc; } // CHECK: static T a; -// CHECK-NEXT: #pragma omp taskwait +// CHECK-NEXT: #pragma omp taskwait{{$}} // CHECK: static int a; // CHECK-NEXT: #pragma omp taskwait // CHECK: static char a; diff --git a/test/OpenMP/taskwait_codegen.cpp b/test/OpenMP/taskwait_codegen.cpp index 492786eb80588..22f381b351996 100644 --- a/test/OpenMP/taskwait_codegen.cpp +++ b/test/OpenMP/taskwait_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -x c++ -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/taskwait_messages.cpp b/test/OpenMP/taskwait_messages.cpp index 06e8e6b8bdf27..46efff5972e8e 100644 --- a/test/OpenMP/taskwait_messages.cpp +++ b/test/OpenMP/taskwait_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + template <class T> T tmain(T argc) { #pragma omp taskwait diff --git a/test/OpenMP/taskyield_ast_print.cpp b/test/OpenMP/taskyield_ast_print.cpp index 1fd9b9c14d4e9..7acf53dd6afe8 100644 --- a/test/OpenMP/taskyield_ast_print.cpp +++ b/test/OpenMP/taskyield_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -15,7 +19,7 @@ T tmain(T argc) { return a + argc; } // CHECK: static T a; -// CHECK-NEXT: #pragma omp taskyield +// CHECK-NEXT: #pragma omp taskyield{{$}} // CHECK: static int a; // CHECK-NEXT: #pragma omp taskyield // CHECK: static char a; diff --git a/test/OpenMP/taskyield_codegen.cpp b/test/OpenMP/taskyield_codegen.cpp index 80b65bff5e49a..d57071f48aa8f 100644 --- a/test/OpenMP/taskyield_codegen.cpp +++ b/test/OpenMP/taskyield_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/taskyield_messages.cpp b/test/OpenMP/taskyield_messages.cpp index cfeaa63815b32..fd98a8786f187 100644 --- a/test/OpenMP/taskyield_messages.cpp +++ b/test/OpenMP/taskyield_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + template <class T> T tmain(T argc) { #pragma omp taskyield diff --git a/test/OpenMP/teams_ast_print.cpp b/test/OpenMP/teams_ast_print.cpp index 861b25da34a3b..80300bfd4f108 100644 --- a/test/OpenMP/teams_ast_print.cpp +++ b/test/OpenMP/teams_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -50,7 +54,7 @@ T tmain(T argc, T *argv) { // CHECK-NEXT: static T a; // CHECK-NEXT: S<T> s; // CHECK-NEXT: #pragma omp target -// CHECK-NEXT: #pragma omp teams +// CHECK-NEXT: #pragma omp teams{{$}} // CHECK-NEXT: a = 2; // CHECK-NEXT: #pragma omp target // CHECK-NEXT: #pragma omp teams default(none) private(argc,b) firstprivate(argv) shared(d) reduction(+: c) reduction(max: e) num_teams(C) thread_limit(d * C) diff --git a/test/OpenMP/teams_codegen.cpp b/test/OpenMP/teams_codegen.cpp index 28f6ca0714d4a..057c557883194 100644 --- a/test/OpenMP/teams_codegen.cpp +++ b/test/OpenMP/teams_codegen.cpp @@ -8,6 +8,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 int Gbla; @@ -105,6 +113,14 @@ int teams_argument_global_local(int a){ // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 // CK2-DAG: [[SSI:%.+]] = type { i32, float } @@ -168,6 +184,14 @@ int teams_template_arg(void) { // RUN: %clang_cc1 -DCK3 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 + +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} #ifdef CK3 // CK3: [[SSI:%.+]] = type { i32, float } @@ -181,7 +205,7 @@ struct SS{ int foo(void) { int comp = 1; - // CK3-DAG: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 2, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i32 [[NT:%[^,]+]], i32 123) + // CK3-DAG: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 3, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* %{{[^,]+}}, i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i32 [[NT:%[^,]+]], i32 123) // CK3-DAG: [[NT]] = load i32, i32* [[NTA:%[^,]+]], // CK3-DAG: [[NTA]] = getelementptr inbounds [[SSI]], [[SSI]]* [[NTB:%[^,]+]], i32 0, i32 0 @@ -194,7 +218,7 @@ struct SS{ ++comp; } - // CK3-DAG: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 2, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i32 456, i32 [[TL:%[^,]+]]) + // CK3-DAG: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 3, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* %{{[^,]+}}, i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i32 456, i32 [[TL:%[^,]+]]) // CK3-DAG: [[TL]] = add nsw i32 [[TLA:%[^,]+]], 123 // CK3-DAG: [[TLA]] = fptosi float [[TLB:%[^,]+]] to i32 @@ -228,11 +252,21 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-32 +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// SIMD-ONLY3-NOT: {{__kmpc|__tgt}} + #ifdef CK4 -// CK4-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CK4-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CK4-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CK4-DAG: [[DEF_LOC_0:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CK4-DAG: [[DEF_LOC_0:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CK4-DEBUG-DAG: [[LOC1:@.+]] = private unnamed_addr constant [{{.+}} x i8] c";{{.*}}teams_codegen.cpp;main;[[@LINE+14]];9;;\00" // CK4-DEBUG-DAG: [[LOC2:@.+]] = private unnamed_addr constant [{{.+}} x i8] c";{{.*}}teams_codegen.cpp;tmain;[[@LINE+7]];9;;\00" @@ -255,15 +289,15 @@ int main (int argc, char **argv) { // CK4: [[ARGCADDR:%.+]] = alloca i{{.+}} // CK4: store i{{.+}} %[[ARGC]], i{{.+}}* [[ARGCADDR]] // CK4-64: [[CONV:%.+]] = bitcast i64* [[ARGCADDR]] to i32* -// CK4-64: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* {{.+}} to void (i32*, i32*, ...)*), i32* [[CONV]]) -// CK4-32: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* {{.+}} to void (i32*, i32*, ...)*), i32* [[ARGCADDR]]) +// CK4-64: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* {{.+}} to void (i32*, i32*, ...)*), i32* [[CONV]]) +// CK4-32: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* {{.+}} to void (i32*, i32*, ...)*), i32* [[ARGCADDR]]) // CK4: ret void // CK4-NEXT: } // CK4: define {{.*}}void @{{[^,]+}}(i8** [[ARGC1:%.+]]) // CK4: [[ARGCADDR1:%.+]] = alloca i8** // CK4: store i8** [[ARGC1]], i8*** [[ARGCADDR1]] -// CK4: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i8***)* {{.+}} to void (i32*, i32*, ...)*), i8*** [[ARGCADDR1]]) +// CK4: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i8***)* {{.+}} to void (i32*, i32*, ...)*), i8*** [[ARGCADDR1]]) #endif // CK4 @@ -278,12 +312,22 @@ int main (int argc, char **argv) { // RUN: %clang_cc1 -DCK5 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s // RUN: %clang_cc1 -DCK5 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK5 --check-prefix CK5-32 +// RUN: %clang_cc1 -DCK5 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc +// RUN: %clang_cc1 -DCK5 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY4 %s +// RUN: %clang_cc1 -DCK5 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s +// RUN: %clang_cc1 -DCK5 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY4 %s +// RUN: %clang_cc1 -DCK5 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc +// RUN: %clang_cc1 -DCK5 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY4 %s +// RUN: %clang_cc1 -DCK5 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s +// RUN: %clang_cc1 -DCK5 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY4 %s +// SIMD-ONLY4-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifdef CK5 -// CK5-DAG: %ident_t = type { i32, i32, i32, i32, i8* } +// CK5-DAG: %struct.ident_t = type { i32, i32, i32, i32, i8* } // CK5-DAG: [[STR:@.+]] = private unnamed_addr constant [23 x i8] c";unknown;unknown;0;0;;\00" -// CK5-DAG: [[DEF_LOC_0:@.+]] = private unnamed_addr constant %ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } +// CK5-DAG: [[DEF_LOC_0:@.+]] = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([23 x i8], [23 x i8]* [[STR]], i32 0, i32 0) } // CK5-DEBUG-DAG: [[LOC1:@.+]] = private unnamed_addr constant [{{.+}} x i8] c";{{.*}}teams_codegen.cpp;main;[[@LINE+14]];9;;\00" // CK5-DEBUG-DAG: [[LOC2:@.+]] = private unnamed_addr constant [{{.+}} x i8] c";{{.*}}teams_codegen.cpp;tmain;[[@LINE+7]];9;;\00" @@ -314,7 +358,7 @@ int main (int argc, char **argv) { // CK5: [[AADDR:%.+]] = alloca i{{.+}} // CK5: [[BADDR:%.+]] = alloca i{{.+}} // CK5: [[ARGCADDR:%.+]] = alloca i{{.+}} -// CK5: [[GBL_TH_NUM:%.+]] = call i32 @__kmpc_global_thread_num(%ident_t* [[DEF_LOC_0]]) +// CK5: [[GBL_TH_NUM:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* [[DEF_LOC_0]]) // CK5: store i{{.+}} [[AP]], i{{.+}}* [[AADDR]] // CK5: store i{{.+}} [[BP]], i{{.+}}* [[BADDR]] // CK5: store i{{.+}} [[ARGC]], i{{.+}}* [[ARGCADDR]] @@ -325,15 +369,15 @@ int main (int argc, char **argv) { // CK5-64: [[BCONVVAL:%.+]] = load i32, i32* [[BCONV]] // CK5-32: [[ACONVVAL:%.+]] = load i32, i32* [[AADDR]] // CK5-32: [[BCONVVAL:%.+]] = load i32, i32* [[BADDR]] -// CK5: {{.+}} = call i32 @__kmpc_push_num_teams(%ident_t* [[DEF_LOC_0]], i32 [[GBL_TH_NUM]], i32 [[ACONVVAL]], i32 [[BCONVVAL]]) -// CK5-64: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined. to void (i32*, i32*, ...)*), i32* [[CONV]]) -// CK5-32: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined. to void (i32*, i32*, ...)*), i32* [[ARGCADDR]]) +// CK5: {{.+}} = call i32 @__kmpc_push_num_teams(%struct.ident_t* [[DEF_LOC_0]], i32 [[GBL_TH_NUM]], i32 [[ACONVVAL]], i32 [[BCONVVAL]]) +// CK5-64: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined. to void (i32*, i32*, ...)*), i32* [[CONV]]) +// CK5-32: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined. to void (i32*, i32*, ...)*), i32* [[ARGCADDR]]) // CK5: define {{.*}}void @{{[^,]+}}(i{{.+}} [[AP:%.+]], i{{.+}} [[BP:%.+]], i{{.+}}** [[ARGC:%.+]]) // CK5: [[AADDR:%.+]] = alloca i{{.+}} // CK5: [[BADDR:%.+]] = alloca i{{.+}} // CK5: [[ARGCADDR:%.+]] = alloca i{{.+}}** -// CK5: [[GBL_TH_NUM:%.+]] = call i32 @__kmpc_global_thread_num(%ident_t* [[DEF_LOC_0]]) +// CK5: [[GBL_TH_NUM:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* [[DEF_LOC_0]]) // CK5: store i{{.+}} [[AP]], i{{.+}}* [[AADDR]] // CK5: store i{{.+}} [[BP]], i{{.+}}* [[BADDR]] // CK5: store i{{.+}}** [[ARGC]], i{{.+}}*** [[ARGCADDR]] @@ -341,11 +385,11 @@ int main (int argc, char **argv) { // CK5-64: [[BCONV:%.+]] = bitcast i64* [[BADDR]] to i32* // CK5-64: [[ACONVVAL:%.+]] = load i32, i32* [[ACONV]] // CK5-64: [[BCONVVAL:%.+]] = load i32, i32* [[BCONV]] -// CK5-64: {{.+}} = call i32 @__kmpc_push_num_teams(%ident_t* [[DEF_LOC_0]], i32 [[GBL_TH_NUM]], i32 [[ACONVVAL]], i32 [[BCONVVAL]]) +// CK5-64: {{.+}} = call i32 @__kmpc_push_num_teams(%struct.ident_t* [[DEF_LOC_0]], i32 [[GBL_TH_NUM]], i32 [[ACONVVAL]], i32 [[BCONVVAL]]) // CK5-32: [[A_VAL:%.+]] = load i32, i32* [[AADDR]] // CK5-32: [[B_VAL:%.+]] = load i32, i32* [[BADDR]] -// CK5-32: {{.+}} = call i32 @__kmpc_push_num_teams(%ident_t* [[DEF_LOC_0]], i32 [[GBL_TH_NUM]], i32 [[A_VAL]], i32 [[B_VAL]]) -// CK5: call void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i{{.+}})* @.omp_outlined.{{.+}} to void (i32*, i32*, ...)*), i{{.+}}*** [[ARGCADDR]]) +// CK5-32: {{.+}} = call i32 @__kmpc_push_num_teams(%struct.ident_t* [[DEF_LOC_0]], i32 [[GBL_TH_NUM]], i32 [[A_VAL]], i32 [[B_VAL]]) +// CK5: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_teams(%struct.ident_t* [[DEF_LOC_0]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i{{.+}})* @.omp_outlined.{{.+}} to void (i32*, i32*, ...)*), i{{.+}}*** [[ARGCADDR]]) // CK5: ret void // CK5-NEXT: } diff --git a/test/OpenMP/teams_default_messages.cpp b/test/OpenMP/teams_default_messages.cpp index 83909ccfe051c..daf48ae562044 100644 --- a/test/OpenMP/teams_default_messages.cpp +++ b/test/OpenMP/teams_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -o - %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/teams_distribute_ast_print.cpp b/test/OpenMP/teams_distribute_ast_print.cpp index f465c3f8e30b2..ef07d51b5a9d9 100644 --- a/test/OpenMP/teams_distribute_ast_print.cpp +++ b/test/OpenMP/teams_distribute_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -96,7 +100,7 @@ T tmain(T argc) { for (int i=0; i < 2; ++i) a = 2; // CHECK: #pragma omp target -// CHECK-NEXT: #pragma omp teams distribute +// CHECK-NEXT: #pragma omp teams distribute{{$}} // CHECK-NEXT: for (int i = 0; i < 2; ++i) // CHECK-NEXT: a = 2; #pragma omp target diff --git a/test/OpenMP/teams_distribute_codegen.cpp b/test/OpenMP/teams_distribute_codegen.cpp index 8a2afbd4341dd..0f6b5f2f25e11 100644 --- a/test/OpenMP/teams_distribute_codegen.cpp +++ b/test/OpenMP/teams_distribute_codegen.cpp @@ -8,6 +8,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 int a[100]; @@ -86,6 +94,14 @@ int teams_argument_global(int n){ // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 // CK2: define {{.*}}i32 @{{.+}}teams_local_argv( @@ -122,6 +138,14 @@ int teams_local_arg(void) { // RUN: %clang_cc1 -DCK3 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 + +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} #ifdef CK3 // CK3: [[SSI:%.+]] = type { [{{.+}} x i32], float } @@ -170,6 +194,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-32 +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// SIMD-ONLY3-NOT: {{__kmpc|__tgt}} + #ifdef CK4 template <typename T, int n> diff --git a/test/OpenMP/teams_distribute_collapse_codegen.cpp b/test/OpenMP/teams_distribute_collapse_codegen.cpp index a65c0a2e63f74..716e766929134 100644 --- a/test/OpenMP/teams_distribute_collapse_codegen.cpp +++ b/test/OpenMP/teams_distribute_collapse_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 template <typename T, int X, long long Y> @@ -61,6 +69,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 template <typename T, int n, int m> diff --git a/test/OpenMP/teams_distribute_collapse_messages.cpp b/test/OpenMP/teams_distribute_collapse_messages.cpp index 37c10e5986bf3..bcbd774e8ea10 100644 --- a/test/OpenMP/teams_distribute_collapse_messages.cpp +++ b/test/OpenMP/teams_distribute_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp %s -std=c++98 // RUN: %clang_cc1 -verify -fopenmp %s -std=c++11 +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s -std=c++98 +// RUN: %clang_cc1 -verify -fopenmp-simd %s -std=c++11 + void foo() { } diff --git a/test/OpenMP/teams_distribute_default_messages.cpp b/test/OpenMP/teams_distribute_default_messages.cpp index bf62930825d9e..5e8702a99d975 100644 --- a/test/OpenMP/teams_distribute_default_messages.cpp +++ b/test/OpenMP/teams_distribute_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/teams_distribute_dist_schedule_codegen.cpp b/test/OpenMP/teams_distribute_dist_schedule_codegen.cpp index 8bd8825e00c47..97ae3871111da 100644 --- a/test/OpenMP/teams_distribute_dist_schedule_codegen.cpp +++ b/test/OpenMP/teams_distribute_dist_schedule_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 template <typename T, int X, long long Y> @@ -84,6 +92,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 template <typename T, int n> diff --git a/test/OpenMP/teams_distribute_dist_schedule_messages.cpp b/test/OpenMP/teams_distribute_dist_schedule_messages.cpp index d86722dd277d8..a86a87a9ef517 100644 --- a/test/OpenMP/teams_distribute_dist_schedule_messages.cpp +++ b/test/OpenMP/teams_distribute_dist_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_firstprivate_codegen.cpp b/test/OpenMP/teams_distribute_firstprivate_codegen.cpp index a126b1d16b775..562e986df034c 100644 --- a/test/OpenMP/teams_distribute_firstprivate_codegen.cpp +++ b/test/OpenMP/teams_distribute_firstprivate_codegen.cpp @@ -1,13 +1,26 @@ -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 - -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER @@ -206,8 +219,8 @@ int main() { // CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[SIVAR_ADDR]], // T_VAR and SIVAR -// CHECK-DAG-64: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* -// CHECK-DAG-64: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* // preparation vars // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], @@ -217,7 +230,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], @@ -237,13 +250,13 @@ int main() { // CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) // CHECK: call void @__kmpc_for_static_init_4( -// CHECK-DAG-32: {{.+}} = {{.+}} [[T_VAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] // CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[VAR_PRIV]] -// CHECK-DAG-32: {{.+}} = {{.+}} [[SIVAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_SIVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[SIVAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_SIVAR]] // CHECK: call void @__kmpc_for_static_fini( // CHECK: ret void @@ -303,7 +316,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], @@ -325,8 +338,8 @@ int main() { // CHECK-DAG: store [[S_INT_TY]]* [[VAR_PRIV]], [[S_INT_TY]]** [[TMP]], // CHECK: call void @__kmpc_for_static_init_4( -// CHECK-DAG-32: {{.+}} = {{.+}} [[T_VAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] // CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[TMP]] // CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] diff --git a/test/OpenMP/teams_distribute_firstprivate_messages.cpp b/test/OpenMP/teams_distribute_firstprivate_messages.cpp index 1a2b5f3782f9e..e373a3a870059 100644 --- a/test/OpenMP/teams_distribute_firstprivate_messages.cpp +++ b/test/OpenMP/teams_distribute_firstprivate_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_lastprivate_codegen.cpp b/test/OpenMP/teams_distribute_lastprivate_codegen.cpp index 2eab60b9dd3c7..883234185aae8 100644 --- a/test/OpenMP/teams_distribute_lastprivate_codegen.cpp +++ b/test/OpenMP/teams_distribute_lastprivate_codegen.cpp @@ -1,16 +1,32 @@ -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 -// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -231,7 +247,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK-DAG: [[S_ARR_PTR_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_PTR]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST]], i8* [[TMP_VAL_BCAST]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST]], // CHECK: call void @__kmpc_for_static_fini( // lastprivates @@ -244,7 +260,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -255,7 +271,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_FLOAT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -264,7 +280,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_ADDR_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: [[SVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[S_VAR_PRIV]], // CHECK: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR_REF]], // CHECK: ret void @@ -331,7 +347,7 @@ int main() { // CHECK-DAG: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK-DAG: [[S_ARR_PTR_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_PTR1]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST1]], i8* [[TMP_VAL_BCAST1]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST1]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST1]], // CHECK: call void @__kmpc_for_static_fini( // lastprivates @@ -344,7 +360,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF1]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF1]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_PRIV1]] to [[S_INT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -355,7 +371,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_INT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -364,6 +380,6 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_ADDR1_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: ret void #endif diff --git a/test/OpenMP/teams_distribute_lastprivate_messages.cpp b/test/OpenMP/teams_distribute_lastprivate_messages.cpp index 392607ea14ba6..d5e1a1c94cd0a 100644 --- a/test/OpenMP/teams_distribute_lastprivate_messages.cpp +++ b/test/OpenMP/teams_distribute_lastprivate_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_loop_messages.cpp b/test/OpenMP/teams_distribute_loop_messages.cpp index acab1f76a2d46..13f39d100aa5d 100644 --- a/test/OpenMP/teams_distribute_loop_messages.cpp +++ b/test/OpenMP/teams_distribute_loop_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s + class S { int a; S() : a(0) {} // expected-note {{implicitly declared private here}} @@ -698,7 +700,7 @@ void test_loop_firstprivate_lastprivate() { S s(4); // expected-error@+2 {{lastprivate variable cannot be firstprivate}} expected-note@+2 {{defined as lastprivate}} #pragma omp target -#pragma omp teams distribute lastprivate(s) firstprivate(s) // expected-error {{calling a private constructor of class 'S'}} +#pragma omp teams distribute lastprivate(s) firstprivate(s) // expected-error {{calling a private constructor of class 'S'}} expected-warning {{Non-trivial type 'S' is mapped, only trivial types are guaranteed to be mapped correctly}} for (int i = 0; i < 16; ++i) ; } diff --git a/test/OpenMP/teams_distribute_num_teams_messages.cpp b/test/OpenMP/teams_distribute_num_teams_messages.cpp index 6086abd9200af..103a6d26fd4cc 100644 --- a/test/OpenMP/teams_distribute_num_teams_messages.cpp +++ b/test/OpenMP/teams_distribute_num_teams_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_ast_print.cpp b/test/OpenMP/teams_distribute_parallel_for_ast_print.cpp index 68bd424ade015..246382daff4f4 100644 --- a/test/OpenMP/teams_distribute_parallel_for_ast_print.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -104,7 +108,7 @@ T tmain(T argc) { for (int i=0; i < 2; ++i) a = 2; // CHECK: #pragma omp target -// CHECK-NEXT: #pragma omp teams distribute parallel for +// CHECK-NEXT: #pragma omp teams distribute parallel for{{$}} // CHECK-NEXT: for (int i = 0; i < 2; ++i) // CHECK-NEXT: a = 2; #pragma omp target diff --git a/test/OpenMP/teams_distribute_parallel_for_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_codegen.cpp index 5711ede87c865..865ea129f9790 100644 --- a/test/OpenMP/teams_distribute_parallel_for_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_codegen.cpp @@ -8,6 +8,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 int a[100]; @@ -88,6 +96,14 @@ int teams_argument_global(int n){ // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 // CK2: define {{.*}}i32 @{{.+}}teams_local_argv( @@ -125,6 +141,14 @@ int teams_local_arg(void) { // RUN: %clang_cc1 -DCK3 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 + +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} #ifdef CK3 // CK3: [[SSI:%.+]] = type { [{{.+}} x i32], float } @@ -174,6 +198,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-32 +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// SIMD-ONLY3-NOT: {{__kmpc|__tgt}} + #ifdef CK4 template <typename T, int n> diff --git a/test/OpenMP/teams_distribute_parallel_for_collapse_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_collapse_codegen.cpp index 7522288f75e22..11a97868ef725 100644 --- a/test/OpenMP/teams_distribute_parallel_for_collapse_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_collapse_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 template <typename T, int X, long long Y> @@ -63,6 +71,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 template <typename T, int n, int m> diff --git a/test/OpenMP/teams_distribute_parallel_for_collapse_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_collapse_messages.cpp index ea4455dcc5f25..2e83a37bc95ec 100644 --- a/test/OpenMP/teams_distribute_parallel_for_collapse_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp %s -std=c++98 // RUN: %clang_cc1 -verify -fopenmp %s -std=c++11 +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s -std=c++98 +// RUN: %clang_cc1 -verify -fopenmp-simd %s -std=c++11 + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_copyin_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_copyin_codegen.cpp index 0202a40002b53..399025d8a4f1f 100644 --- a/test/OpenMP/teams_distribute_parallel_for_copyin_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_copyin_codegen.cpp @@ -5,10 +5,23 @@ // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/teams_distribute_parallel_for_copyin_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_copyin_messages.cpp index 30f3e75bf6359..2664b9df95343 100644 --- a/test/OpenMP/teams_distribute_parallel_for_copyin_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_copyin_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_default_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_default_messages.cpp index 4cd7e65510b1e..e5b3331520390 100644 --- a/test/OpenMP/teams_distribute_parallel_for_default_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/teams_distribute_parallel_for_dist_schedule_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_dist_schedule_codegen.cpp index dc48f166b4c03..03276e4b5ecbe 100644 --- a/test/OpenMP/teams_distribute_parallel_for_dist_schedule_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_dist_schedule_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 template <typename T, int X, long long Y> @@ -103,6 +111,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 template <typename T, int n> diff --git a/test/OpenMP/teams_distribute_parallel_for_dist_schedule_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_dist_schedule_messages.cpp index 098abb74870ba..e0abe1114fb1b 100644 --- a/test/OpenMP/teams_distribute_parallel_for_dist_schedule_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_dist_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp index fa99f9cfcf1b4..7e161b1d2a62a 100644 --- a/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_firstprivate_codegen.cpp @@ -1,13 +1,26 @@ -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 - -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER @@ -127,16 +140,6 @@ int main() { // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, // LAMBDA: [[G1_TMP:%.+]] = alloca i32*, // skip loop vars - // LAMBDA: alloca i32, - // LAMBDA: alloca i32, - // LAMBDA: alloca i32, - // LAMBDA: alloca i32, - // LAMBDA: alloca i32, - // LAMBDA: alloca i32, - // LAMBDA: [[G_PRIV:%.+]] = alloca i{{[0-9]+}}, - // LAMBDA: [[G1_PRIV:%.+]] = alloca i{{[0-9]+}}, - // LAMBDA: [[G1_TMP_PRIV:%.+]] = alloca i{{[0-9]+}}*, - // LAMBDA: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}}, // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_ADDR]], // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_ADDR]], // LAMBDA-DAG: store {{.+}}, {{.+}} [[SIVAR_ADDR]], @@ -146,10 +149,10 @@ int main() { // LAMBDA-DAG: store{{.+}} [[G1_CONV]], {{.+}} [[G1_TMP]], // use of private vars - // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_PRIV]], - // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP_PRIV]] + // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_CONV]], + // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP]] // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1]], - // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_PRIV]], + // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_CONV]], // LAMBDA-DAG: [[G1_REF:%.+]] = load{{.+}}, {{.+}} [[G1_TMP]], // LAMBDA: call void [[INNER_LAMBDA:@.+]]( // LAMBDA: call void @__kmpc_for_static_fini( @@ -240,8 +243,8 @@ int main() { // CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[SIVAR_ADDR]], // T_VAR and SIVAR -// CHECK-DAG-64: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* -// CHECK-DAG-64: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* // preparation vars // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], @@ -251,7 +254,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], @@ -306,8 +309,8 @@ int main() { // CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[SIVAR_ADDR]], // T_VAR and SIVAR -// CHECK-DAG-64: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* -// CHECK-DAG-64: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* // preparation vars // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], @@ -317,7 +320,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], @@ -337,13 +340,13 @@ int main() { // CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) // CHECK: call void @__kmpc_for_static_init_4( -// CHECK-DAG-32: {{.+}} = {{.+}} [[T_VAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] // CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[VAR_PRIV]] -// CHECK-DAG-32: {{.+}} = {{.+}} [[SIVAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_SIVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[SIVAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_SIVAR]] // CHECK: call void @__kmpc_for_static_fini( // CHECK: ret void @@ -402,7 +405,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], @@ -463,7 +466,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], @@ -485,8 +488,8 @@ int main() { // CHECK-DAG: store [[S_INT_TY]]* [[VAR_PRIV]], [[S_INT_TY]]** [[TMP]], // CHECK: call void @__kmpc_for_static_init_4( -// CHECK-DAG-32: {{.+}} = {{.+}} [[T_VAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] // CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[TMP]] // CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] diff --git a/test/OpenMP/teams_distribute_parallel_for_firstprivate_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_firstprivate_messages.cpp index 90e78defc5e27..f4955887546ea 100644 --- a/test/OpenMP/teams_distribute_parallel_for_firstprivate_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_firstprivate_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_if_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_if_codegen.cpp index bc40dca83fbe7..f4119fc47d3b9 100644 --- a/test/OpenMP/teams_distribute_parallel_for_if_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_if_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/teams_distribute_parallel_for_if_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_if_messages.cpp index 14253b89f63e5..e92dc65c60a86 100644 --- a/test/OpenMP/teams_distribute_parallel_for_if_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_lastprivate_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_lastprivate_codegen.cpp index 501abb725b31a..451dc0579cd32 100644 --- a/test/OpenMP/teams_distribute_parallel_for_lastprivate_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_lastprivate_codegen.cpp @@ -1,16 +1,32 @@ -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 - -// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -288,7 +304,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -299,7 +315,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_FLOAT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -308,7 +324,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_ADDR_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: [[SVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[S_VAR_PRIV]], // CHECK: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR_REF]], // CHECK: ret void @@ -365,7 +381,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK-DAG: [[S_ARR_PTR_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_PTR]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST]], i8* [[TMP_VAL_BCAST]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST]], // CHECK: call void @__kmpc_for_static_fini( @@ -379,7 +395,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -390,7 +406,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_FLOAT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -399,7 +415,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_ADDR_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: [[SVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[S_VAR_PRIV]], // CHECK: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR_REF]], // CHECK: ret void @@ -468,7 +484,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF1]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF1]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_PRIV1]] to [[S_INT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -479,7 +495,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_INT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -488,7 +504,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_ADDR1_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: ret void // CHECK: define internal void [[TPAR_OUTL:@.+]](i{{[0-9]+}}* noalias [[GTID_ADDR1:%.+]], i{{[0-9]+}}* noalias %{{.+}}, {{.+}}, {{.+}}, [2 x i{{[0-9]+}}]*{{.+}} [[VEC_IN1:%.+]], i{{[0-9]+}}*{{.+}} [[T_VAR_IN1:%.+]], [2 x [[S_INT_TY]]]*{{.+}} [[S_ARR_IN1:%.+]], [[S_INT_TY]]*{{.+}} [[VAR_IN1:%.+]]) @@ -544,7 +560,7 @@ int main() { // CHECK-DAG: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK-DAG: [[S_ARR_PTR_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_PTR1]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST1]], i8* [[TMP_VAL_BCAST1]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST1]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST1]], // CHECK: call void @__kmpc_for_static_fini( @@ -558,7 +574,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF1]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF1]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_PRIV1]] to [[S_INT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -569,7 +585,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_INT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -578,7 +594,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_ADDR1_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: ret void #endif diff --git a/test/OpenMP/teams_distribute_parallel_for_lastprivate_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_lastprivate_messages.cpp index b58bac4fdfc3b..97e9e4f822f0f 100644 --- a/test/OpenMP/teams_distribute_parallel_for_lastprivate_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_lastprivate_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_loop_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_loop_messages.cpp index a2f6f679d3254..0fd3fa263ad26 100644 --- a/test/OpenMP/teams_distribute_parallel_for_loop_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_loop_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s + class S { int a; S() : a(0) {} // expected-note {{implicitly declared private here}} @@ -693,7 +695,7 @@ void test_loop_firstprivate_lastprivate() { S s(4); // expected-error@+2 {{lastprivate variable cannot be firstprivate}} expected-note@+2 {{defined as lastprivate}} #pragma omp target -#pragma omp teams distribute parallel for lastprivate(s) firstprivate(s) // expected-error {{calling a private constructor of class 'S'}} +#pragma omp teams distribute parallel for lastprivate(s) firstprivate(s) // expected-error {{calling a private constructor of class 'S'}} expected-warning {{Non-trivial type 'S' is mapped, only trivial types are guaranteed to be mapped correctly}} for (int i = 0; i < 16; ++i) ; } diff --git a/test/OpenMP/teams_distribute_parallel_for_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_messages.cpp index 9843b442e5b47..57ad66638028f 100644 --- a/test/OpenMP/teams_distribute_parallel_for_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_num_teams_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_num_teams_messages.cpp index 63bc91e8ec004..cc6f1c8e64056 100644 --- a/test/OpenMP/teams_distribute_parallel_for_num_teams_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_num_teams_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_num_threads_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_num_threads_codegen.cpp index 5aed5b172af6a..ea4afa1418cb2 100644 --- a/test/OpenMP/teams_distribute_parallel_for_num_threads_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_num_threads_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-ibm-linux-gnu -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-ibm-linux-gnu -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-ibm-linux-gnu -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-ibm-linux-gnu -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-ibm-linux-gnu -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-ibm-linux-gnu -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/teams_distribute_parallel_for_private_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_private_codegen.cpp index 80a12738cab4b..5fc7ee12eb0ad 100644 --- a/test/OpenMP/teams_distribute_parallel_for_private_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_private_codegen.cpp @@ -1,13 +1,26 @@ -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER diff --git a/test/OpenMP/teams_distribute_parallel_for_private_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_private_messages.cpp index 6e72905d754a3..5fda1a96e41e0 100644 --- a/test/OpenMP/teams_distribute_parallel_for_private_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_proc_bind_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_proc_bind_codegen.cpp index 71e92bd5a5553..0c242f851b7d6 100644 --- a/test/OpenMP/teams_distribute_parallel_for_proc_bind_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_proc_bind_codegen.cpp @@ -3,6 +3,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -57,7 +62,7 @@ int main() { // CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], // CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 4) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: ret void // CHECK: define{{.+}} [[OFFL2]]() @@ -69,7 +74,7 @@ int main() { // CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], // CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 3) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: ret void // CHECK: define{{.+}} [[TMAIN]]() @@ -85,6 +90,6 @@ int main() { // CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], // CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 2) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: ret void #endif diff --git a/test/OpenMP/teams_distribute_parallel_for_proc_bind_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_proc_bind_messages.cpp index 2a7074a0bcd4b..33e9c39be9007 100644 --- a/test/OpenMP/teams_distribute_parallel_for_proc_bind_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_proc_bind_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo(); template <class T, typename S, int N> diff --git a/test/OpenMP/teams_distribute_parallel_for_reduction_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_reduction_codegen.cpp index 55ccfff99a3c5..34cc2874f6f3e 100644 --- a/test/OpenMP/teams_distribute_parallel_for_reduction_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_reduction_codegen.cpp @@ -5,10 +5,23 @@ // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/teams_distribute_parallel_for_reduction_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_reduction_messages.cpp index 889efd76cc3cc..31a91a862f690 100644 --- a/test/OpenMP/teams_distribute_parallel_for_reduction_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_reduction_messages.cpp @@ -1,6 +1,10 @@ -// RUN: %clang_cc1 -verify -fopenmp %s -// RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s -// RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_schedule_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_schedule_codegen.cpp index 201a170ccd4a4..f7b11a7191940 100644 --- a/test/OpenMP/teams_distribute_parallel_for_schedule_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_schedule_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 template <typename T, int X, long long Y> @@ -149,6 +157,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 template <typename T, int n> diff --git a/test/OpenMP/teams_distribute_parallel_for_schedule_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_schedule_messages.cpp index 477d1a0f7d34f..e8cda672c0961 100644 --- a/test/OpenMP/teams_distribute_parallel_for_schedule_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_shared_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_shared_messages.cpp index 1a1af09aa2059..90edfb1b11f0f 100644 --- a/test/OpenMP/teams_distribute_parallel_for_shared_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_shared_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_aligned_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_aligned_messages.cpp index 97b200bb16419..8fc21a086f818 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_aligned_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_aligned_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp %s +// RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp-simd %s + struct B { static int ib[20]; // expected-note 0 {{'B::ib' declared here}} static constexpr int bfoo() { return 8; } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_ast_print.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_ast_print.cpp index 7573fdeafa315..c7a0a358f60dd 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_ast_print.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -128,7 +132,7 @@ T tmain(T argc) { for (int i=0; i < 2; ++i) a = 2; // CHECK: #pragma omp target -// CHECK-NEXT: #pragma omp teams distribute parallel for simd +// CHECK-NEXT: #pragma omp teams distribute parallel for simd{{$}} // CHECK-NEXT: for (int i = 0; i < 2; ++i) // CHECK-NEXT: a = 2; #pragma omp target diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_codegen.cpp index 00f2844735c90..784deeef7cd2f 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_codegen.cpp @@ -8,6 +8,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 int a[100]; @@ -92,6 +100,14 @@ int teams_argument_global(int n){ // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 // CK2: define {{.*}}i32 @{{.+}}teams_local_argv( @@ -133,6 +149,14 @@ int teams_local_arg(void) { // RUN: %clang_cc1 -DCK3 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 + +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} #ifdef CK3 // CK3: [[SSI:%.+]] = type { [{{.+}} x i32], float } @@ -144,7 +168,7 @@ struct SS{ // CK3: define {{.*}}i32 @{{.+}}foo{{.+}}( int foo(void) { int i; - // CK3: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 2, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0) + // CK3: call i32 @__tgt_target_teams(i64 -1, i8* @{{[^,]+}}, i32 3, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* %{{[^,]+}}, i64* {{.+}}@{{[^,]+}}, i32 0, i32 0) // CK3: call void @[[OFFL1:.+]]([[SSI]]* %{{.+}}) #pragma omp target #pragma omp teams distribute parallel for simd safelen(4) aligned(a) linear(i) @@ -185,6 +209,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-32 +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// SIMD-ONLY3-NOT: {{__kmpc|__tgt}} + #ifdef CK4 template <typename T, int n> diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_collapse_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_collapse_codegen.cpp index 91c451d49a083..24bd31e07ba6d 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_collapse_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_collapse_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 template <typename T, int X, long long Y> @@ -66,6 +74,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 template <typename T, int n, int m> diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_collapse_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_collapse_messages.cpp index 197c5f1ad341e..a287da7a36c24 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_collapse_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp %s -std=c++98 // RUN: %clang_cc1 -verify -fopenmp %s -std=c++11 +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s -std=c++98 +// RUN: %clang_cc1 -verify -fopenmp-simd %s -std=c++11 + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_default_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_default_messages.cpp index dc69acbe0b3c0..8db9d668fd34f 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_default_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp index 5c1f73066cd40..4f11dc67437b8 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_dist_schedule_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 template <typename T, int X, long long Y> @@ -106,6 +114,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 template <typename T, int n> diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_dist_schedule_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_dist_schedule_messages.cpp index b879de749dfa2..d53fa578b5c57 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_dist_schedule_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_dist_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp index 1219742c752d9..68b0016a97462 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_codegen.cpp @@ -1,13 +1,26 @@ -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 - -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER @@ -127,16 +140,6 @@ int main() { // LAMBDA: [[SIVAR_ADDR:%.+]] = alloca i{{[0-9]+}}, // LAMBDA: [[G1_TMP:%.+]] = alloca i32*, // skip loop vars - // LAMBDA: alloca i32, - // LAMBDA: alloca i32, - // LAMBDA: alloca i32, - // LAMBDA: alloca i32, - // LAMBDA: alloca i32, - // LAMBDA: alloca i32, - // LAMBDA: [[G_PRIV:%.+]] = alloca i{{[0-9]+}}, - // LAMBDA: [[G1_PRIV:%.+]] = alloca i{{[0-9]+}}, - // LAMBDA: [[G1_TMP_PRIV:%.+]] = alloca i{{[0-9]+}}*, - // LAMBDA: [[SIVAR_PRIV:%.+]] = alloca i{{[0-9]+}}, // LAMBDA-DAG: store {{.+}}, {{.+}} [[G_ADDR]], // LAMBDA-DAG: store {{.+}}, {{.+}} [[G1_ADDR]], // LAMBDA-DAG: store {{.+}}, {{.+}} [[SIVAR_ADDR]], @@ -146,10 +149,10 @@ int main() { // LAMBDA-DAG: store{{.+}} [[G1_CONV]], {{.+}} [[G1_TMP]], // use of private vars - // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_PRIV]], - // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP_PRIV]] + // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G_CONV]], + // LAMBDA-DAG: [[G1:%.+]] = load{{.+}}, {{.+}}* [[G1_TMP]] // LAMBDA-DAG: store{{.+}} 1, {{.+}} [[G1]], - // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_PRIV]], + // LAMBDA-DAG: store{{.+}} 2, {{.+}} [[SIVAR_CONV]], // LAMBDA-DAG: [[G1_REF:%.+]] = load{{.+}}, {{.+}} [[G1_TMP]], // LAMBDA: call void [[INNER_LAMBDA:@.+]]( // LAMBDA: call void @__kmpc_for_static_fini( @@ -243,8 +246,8 @@ int main() { // CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[SIVAR_ADDR]], // T_VAR and SIVAR -// CHECK-DAG-64: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* -// CHECK-DAG-64: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* // preparation vars // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], @@ -254,7 +257,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], @@ -309,8 +312,8 @@ int main() { // CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[SIVAR_ADDR]], // T_VAR and SIVAR -// CHECK-DAG-64: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* -// CHECK-DAG-64: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* // preparation vars // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], @@ -320,7 +323,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], @@ -340,13 +343,13 @@ int main() { // CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) // CHECK: call void @__kmpc_for_static_init_4( -// CHECK-DAG-32: {{.+}} = {{.+}} [[T_VAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] // CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[VAR_PRIV]] -// CHECK-DAG-32: {{.+}} = {{.+}} [[SIVAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_SIVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[SIVAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_SIVAR]] // CHECK: call void @__kmpc_for_static_fini( // CHECK: ret void @@ -405,7 +408,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], @@ -466,7 +469,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], @@ -488,8 +491,8 @@ int main() { // CHECK-DAG: store [[S_INT_TY]]* [[VAR_PRIV]], [[S_INT_TY]]** [[TMP]], // CHECK: call void @__kmpc_for_static_init_4( -// CHECK-DAG-32: {{.+}} = {{.+}} [[T_VAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] // CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[TMP]] // CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_messages.cpp index 668337db454b2..036f201230933 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_firstprivate_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_if_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_if_codegen.cpp index 6f2cef88cbbc3..66794132a9edf 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_if_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_if_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_if_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_if_messages.cpp index 01978d722226d..d8a642300d71f 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_if_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_if_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_lastprivate_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_lastprivate_codegen.cpp index 5df893d525f47..c9141b01494fd 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_lastprivate_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_lastprivate_codegen.cpp @@ -1,16 +1,32 @@ -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 - -// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -295,7 +311,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -306,7 +322,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_FLOAT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -315,7 +331,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_ADDR_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: [[SVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[S_VAR_PRIV]], // CHECK: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR_REF]], // CHECK: ret void @@ -372,7 +388,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK-DAG: [[S_ARR_PTR_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_PTR]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST]], i8* [[TMP_VAL_BCAST]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST]], // CHECK: call void @__kmpc_for_static_fini( @@ -389,7 +405,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -400,7 +416,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_FLOAT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -409,7 +425,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_ADDR_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: [[SVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[S_VAR_PRIV]], // CHECK: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR_REF]], // CHECK: ret void @@ -478,7 +494,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF1]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF1]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_PRIV1]] to [[S_INT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -489,7 +505,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_INT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -498,7 +514,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_ADDR1_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: ret void // CHECK: define internal void [[TPAR_OUTL:@.+]](i{{[0-9]+}}* noalias [[GTID_ADDR1:%.+]], i{{[0-9]+}}* noalias %{{.+}}, {{.+}}, {{.+}}, [2 x i{{[0-9]+}}]*{{.+}} [[VEC_IN1:%.+]], i{{[0-9]+}}*{{.+}} [[T_VAR_IN1:%.+]], [2 x [[S_INT_TY]]]*{{.+}} [[S_ARR_IN1:%.+]], [[S_INT_TY]]*{{.+}} [[VAR_IN1:%.+]]) @@ -554,7 +570,7 @@ int main() { // CHECK-DAG: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK-DAG: [[S_ARR_PTR_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_PTR1]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST1]], i8* [[TMP_VAL_BCAST1]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST1]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST1]], // CHECK: call void @__kmpc_for_static_fini( @@ -571,7 +587,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF1]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF1]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_PRIV1]] to [[S_INT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -582,7 +598,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_INT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -591,7 +607,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_ADDR1_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: ret void // CHECK: !{!"llvm.loop.vectorize.enable", i1 true} diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_lastprivate_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_lastprivate_messages.cpp index 204be4c95e910..416d409acc131 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_lastprivate_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_lastprivate_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_linear_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_linear_messages.cpp index c0a9baf916dd6..e60f2c9dbc681 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_linear_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_loop_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_loop_messages.cpp index 66361b64eb848..9980a1d11a7e1 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_loop_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_loop_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s + class S { int a; S() : a(0) {} // expected-note {{implicitly declared private here}} @@ -695,7 +697,7 @@ void test_loop_firstprivate_lastprivate() { S s(4); // expected-error@+2 {{lastprivate variable cannot be firstprivate}} expected-note@+2 {{defined as lastprivate}} #pragma omp target -#pragma omp teams distribute parallel for simd lastprivate(s) firstprivate(s) // expected-error {{calling a private constructor of class 'S'}} +#pragma omp teams distribute parallel for simd lastprivate(s) firstprivate(s) // expected-error {{calling a private constructor of class 'S'}} expected-warning {{Non-trivial type 'S' is mapped, only trivial types are guaranteed to be mapped correctly}} for (int i = 0; i < 16; ++i) ; } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_messages.cpp index 12ac8f9a9479f..d27e156161f8b 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_num_teams_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_num_teams_messages.cpp index 7affd607b9d03..5e4ce0c864719 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_num_teams_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_num_teams_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_num_threads_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_num_threads_codegen.cpp index aa7b9331da12c..8c0c8208ab80f 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_num_threads_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_num_threads_codegen.cpp @@ -1,6 +1,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-ibm-linux-gnu -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-ibm-linux-gnu -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-ibm-linux-gnu -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-ibm-linux-gnu -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple powerpc64le-ibm-linux-gnu -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple powerpc64le-ibm-linux-gnu -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_num_threads_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_num_threads_messages.cpp index fc71569e65ac1..1f9b61e464376 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_num_threads_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_num_threads_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_private_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_private_codegen.cpp index a6dbd29a5b96a..b2c8c22b84d80 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_private_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_private_codegen.cpp @@ -1,13 +1,26 @@ -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_private_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_private_messages.cpp index dc1ef5794206b..50bf034891d90 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_private_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_proc_bind_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_proc_bind_codegen.cpp index 7f95a877ab857..08b3cd3a47baf 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_proc_bind_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_proc_bind_codegen.cpp @@ -3,6 +3,11 @@ // RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -57,7 +62,7 @@ int main() { // CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], // CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 4) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: ret void // CHECK: define{{.+}} [[OFFL2]]() @@ -69,7 +74,7 @@ int main() { // CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], // CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 3) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: ret void // CHECK: define{{.+}} [[TMAIN]]() @@ -85,7 +90,7 @@ int main() { // CHECK: [[GTID_REF:%.+]] = load i32*, i32** [[GTID_ADDR]], // CHECK: [[GTID_VAL:%.+]] = load i32, i32* [[GTID_REF]], // CHECK: call {{.*}}void @__kmpc_push_proc_bind([[IDENT_T_TY]]* [[DEF_LOC_2]], i32 [[GTID_VAL]], i32 2) -// CHECK: call {{.*}}void (%ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( +// CHECK: call {{.*}}void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call( // CHECK: ret void // CHECK: !{!"llvm.loop.vectorize.enable", i1 true} diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_proc_bind_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_proc_bind_messages.cpp index fd904299cd6db..459f40f494c2d 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_proc_bind_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_proc_bind_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -ferror-limit 100 -o - %s + void foo(); template <class T, typename S, int N> diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_reduction_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_reduction_codegen.cpp index 384ac4b5f32ad..f97f0a050c346 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_reduction_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_reduction_codegen.cpp @@ -5,10 +5,23 @@ // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_reduction_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_reduction_messages.cpp index b0522e8f69d2d..a17ffe392b879 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_reduction_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_reduction_messages.cpp @@ -1,6 +1,10 @@ -// RUN: %clang_cc1 -verify -fopenmp %s -// RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s -// RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_safelen_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_safelen_messages.cpp index 82fd4b0da837e..e673bb382af67 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_safelen_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_safelen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_schedule_codegen.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_schedule_codegen.cpp index 71e8ee8e81105..ea99cc7b569d2 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_schedule_codegen.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_schedule_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 template <typename T, int X, long long Y> @@ -151,6 +159,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 template <typename T, int n> diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_shared_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_shared_messages.cpp index 1089555889f00..90f3cf46b90d4 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_shared_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_shared_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_simdlen_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_simdlen_messages.cpp index 82fd4b0da837e..e673bb382af67 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_simdlen_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_simdlen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_simd_thread_limit_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_simd_thread_limit_messages.cpp index 81ed9475d3566..b5a95760eb0dc 100644 --- a/test/OpenMP/teams_distribute_parallel_for_simd_thread_limit_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_simd_thread_limit_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_parallel_for_thread_limit_messages.cpp b/test/OpenMP/teams_distribute_parallel_for_thread_limit_messages.cpp index 4018f3f07d0ff..ad37b5389c630 100644 --- a/test/OpenMP/teams_distribute_parallel_for_thread_limit_messages.cpp +++ b/test/OpenMP/teams_distribute_parallel_for_thread_limit_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_private_codegen.cpp b/test/OpenMP/teams_distribute_private_codegen.cpp index 8e4a61cd7cb19..8880e2f517698 100644 --- a/test/OpenMP/teams_distribute_private_codegen.cpp +++ b/test/OpenMP/teams_distribute_private_codegen.cpp @@ -1,13 +1,26 @@ -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER diff --git a/test/OpenMP/teams_distribute_private_messages.cpp b/test/OpenMP/teams_distribute_private_messages.cpp index aff1f539894d2..5fa002aadaaa1 100644 --- a/test/OpenMP/teams_distribute_private_messages.cpp +++ b/test/OpenMP/teams_distribute_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_reduction_codegen.cpp b/test/OpenMP/teams_distribute_reduction_codegen.cpp index 961891da98d0e..019306c6106d8 100644 --- a/test/OpenMP/teams_distribute_reduction_codegen.cpp +++ b/test/OpenMP/teams_distribute_reduction_codegen.cpp @@ -5,10 +5,23 @@ // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/teams_distribute_reduction_messages.cpp b/test/OpenMP/teams_distribute_reduction_messages.cpp index f11236f0ce27b..4b9d6e51b2cc7 100644 --- a/test/OpenMP/teams_distribute_reduction_messages.cpp +++ b/test/OpenMP/teams_distribute_reduction_messages.cpp @@ -1,6 +1,10 @@ -// RUN: %clang_cc1 -verify -fopenmp %s -// RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s -// RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_shared_messages.cpp b/test/OpenMP/teams_distribute_shared_messages.cpp index ecf6f2e03673a..b1828eb43c42b 100644 --- a/test/OpenMP/teams_distribute_shared_messages.cpp +++ b/test/OpenMP/teams_distribute_shared_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_simd_aligned_messages.cpp b/test/OpenMP/teams_distribute_simd_aligned_messages.cpp index 885432bb9a1e5..34227c659a3ea 100644 --- a/test/OpenMP/teams_distribute_simd_aligned_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_aligned_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp %s +// RUN: %clang_cc1 -x c++ -std=c++11 -verify -fopenmp-simd %s + struct B { static int ib[20]; // expected-note 0 {{'B::ib' declared here}} static constexpr int bfoo() { return 8; } diff --git a/test/OpenMP/teams_distribute_simd_ast_print.cpp b/test/OpenMP/teams_distribute_simd_ast_print.cpp index e751413c42327..b3d8ce436d4aa 100644 --- a/test/OpenMP/teams_distribute_simd_ast_print.cpp +++ b/test/OpenMP/teams_distribute_simd_ast_print.cpp @@ -1,6 +1,10 @@ // RUN: %clang_cc1 -verify -fopenmp -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s + +// RUN: %clang_cc1 -verify -fopenmp-simd -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print | FileCheck %s // expected-no-diagnostics #ifndef HEADER @@ -122,7 +126,7 @@ T tmain(T argc) { for (int i=0; i < 2; ++i) a = 2; // CHECK: #pragma omp target -// CHECK-NEXT: #pragma omp teams distribute simd +// CHECK-NEXT: #pragma omp teams distribute simd{{$}} // CHECK-NEXT: for (int i = 0; i < 2; ++i) // CHECK-NEXT: a = 2; #pragma omp target diff --git a/test/OpenMP/teams_distribute_simd_codegen.cpp b/test/OpenMP/teams_distribute_simd_codegen.cpp index cad5854a2f1ea..813d3912c3985 100644 --- a/test/OpenMP/teams_distribute_simd_codegen.cpp +++ b/test/OpenMP/teams_distribute_simd_codegen.cpp @@ -8,6 +8,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 int a[100]; @@ -92,6 +100,14 @@ int teams_argument_global(int n) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 // CK2: define {{.*}}i32 @{{.+}}teams_local_argv( @@ -129,6 +145,14 @@ int teams_local_arg(void) { // RUN: %clang_cc1 -DCK3 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 + +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} #ifdef CK3 // CK3: [[SSI:%.+]] = type { [{{.+}} x i32], float } @@ -178,6 +202,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-32 +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s +// SIMD-ONLY3-NOT: {{__kmpc|__tgt}} + #ifdef CK4 template <typename T, int n> diff --git a/test/OpenMP/teams_distribute_simd_collapse_codegen.cpp b/test/OpenMP/teams_distribute_simd_collapse_codegen.cpp index 071d1672fad1b..41c494a562d70 100644 --- a/test/OpenMP/teams_distribute_simd_collapse_codegen.cpp +++ b/test/OpenMP/teams_distribute_simd_collapse_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 template <typename T, int X, long long Y> @@ -63,6 +71,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 template <typename T, int n, int m> diff --git a/test/OpenMP/teams_distribute_simd_collapse_messages.cpp b/test/OpenMP/teams_distribute_simd_collapse_messages.cpp index 0f62be2861d1f..d0b2f814bf8bc 100644 --- a/test/OpenMP/teams_distribute_simd_collapse_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_collapse_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_simd_default_messages.cpp b/test/OpenMP/teams_distribute_simd_default_messages.cpp index a6685b9202ea3..af176f46cbfc5 100644 --- a/test/OpenMP/teams_distribute_simd_default_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_default_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo(); int main(int argc, char **argv) { diff --git a/test/OpenMP/teams_distribute_simd_dist_schedule_codegen.cpp b/test/OpenMP/teams_distribute_simd_dist_schedule_codegen.cpp index 599e18f58a163..8937a44f89e80 100644 --- a/test/OpenMP/teams_distribute_simd_dist_schedule_codegen.cpp +++ b/test/OpenMP/teams_distribute_simd_dist_schedule_codegen.cpp @@ -9,6 +9,14 @@ // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32 + +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} #ifdef CK1 template <typename T, int X, long long Y> @@ -85,6 +93,14 @@ int teams_template_struct(void) { // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32 + +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} #ifdef CK2 template <typename T, int n> diff --git a/test/OpenMP/teams_distribute_simd_dist_schedule_messages.cpp b/test/OpenMP/teams_distribute_simd_dist_schedule_messages.cpp index 5274204b1ac6b..3eb0db923b81e 100644 --- a/test/OpenMP/teams_distribute_simd_dist_schedule_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_dist_schedule_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_simd_firstprivate_codegen.cpp b/test/OpenMP/teams_distribute_simd_firstprivate_codegen.cpp index e503f5267f19b..27320191faefd 100644 --- a/test/OpenMP/teams_distribute_simd_firstprivate_codegen.cpp +++ b/test/OpenMP/teams_distribute_simd_firstprivate_codegen.cpp @@ -1,13 +1,26 @@ -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 - -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER @@ -206,8 +219,8 @@ int main() { // CHECK: store i{{[0-9]+}} {{.+}}, i{{[0-9]+}}* [[SIVAR_ADDR]], // T_VAR and SIVAR -// CHECK-DAG-64: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* -// CHECK-DAG-64: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_TVAR:%.+]] = bitcast i64* [[T_VAR_ADDR]] to i32* +// CHECK-64-DAG: [[CONV_SIVAR:%.+]] = bitcast i64* [[SIVAR_ADDR]] to i32* // preparation vars // CHECK-DAG: [[VEC_ADDR_VAL:%.+]] = load [2 x i{{[0-9]+}}]*, [2 x i{{[0-9]+}}]** [[VEC_ADDR]], @@ -217,7 +230,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], @@ -237,13 +250,13 @@ int main() { // CHECK-DAG: call void @{{.+}}({{.+}} [[AGG_TMP2]]) // CHECK: call void @__kmpc_for_static_init_4( -// CHECK-DAG-32: {{.+}} = {{.+}} [[T_VAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] // CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[VAR_PRIV]] -// CHECK-DAG-32: {{.+}} = {{.+}} [[SIVAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_SIVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[SIVAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_SIVAR]] // CHECK: call void @__kmpc_for_static_fini( // CHECK: ret void @@ -303,7 +316,7 @@ int main() { // firstprivate vec(vec): copy from *_addr into priv1 and then from priv1 into priv2 // CHECK-DAG: [[VEC_DEST_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK-DAG: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST_PRIV]], i8* [[VEC_SRC]], {{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST_PRIV]], i8* align {{[0-9]+}} [[VEC_SRC]], {{.+}}) // firstprivate(s_arr) // CHECK-DAG: [[S_ARR_PRIV_BGN:%.+]] = getelementptr{{.*}} [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], @@ -325,8 +338,8 @@ int main() { // CHECK-DAG: store [[S_INT_TY]]* [[VAR_PRIV]], [[S_INT_TY]]** [[TMP]], // CHECK: call void @__kmpc_for_static_init_4( -// CHECK-DAG-32: {{.+}} = {{.+}} [[T_VAR_ADDR]] -// CHECK-DAG-64: {{.+}} = {{.+}} [[CONV_TVAR]] +// CHECK-32-DAG: {{.+}} = {{.+}} [[T_VAR_ADDR]] +// CHECK-64-DAG: {{.+}} = {{.+}} [[CONV_TVAR]] // CHECK-DAG: {{.+}} = {{.+}} [[VEC_PRIV]] // CHECK-DAG: {{.+}} = {{.+}} [[TMP]] // CHECK-DAG: {{.+}} = {{.+}} [[S_ARR_PRIV]] diff --git a/test/OpenMP/teams_distribute_simd_firstprivate_messages.cpp b/test/OpenMP/teams_distribute_simd_firstprivate_messages.cpp index bcc0e351c66ec..5b074c4cda10e 100644 --- a/test/OpenMP/teams_distribute_simd_firstprivate_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_firstprivate_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_simd_lastprivate_codegen.cpp b/test/OpenMP/teams_distribute_simd_lastprivate_codegen.cpp index 3a18b75cf405f..86cfd00f8c7a3 100644 --- a/test/OpenMP/teams_distribute_simd_lastprivate_codegen.cpp +++ b/test/OpenMP/teams_distribute_simd_lastprivate_codegen.cpp @@ -1,16 +1,32 @@ -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 -// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target| FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -232,7 +248,7 @@ int main() { // CHECK-DAG: [[TMP_VAL:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK-DAG: [[S_ARR_PTR_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_PTR]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST]], i8* [[TMP_VAL_BCAST]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST]], // CHECK: call void @__kmpc_for_static_fini( // lastprivates @@ -245,7 +261,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -256,7 +272,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_FLOAT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_FLOAT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -265,7 +281,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_FLOAT_TY]]*, [[S_FLOAT_TY]]** [[TMP_PRIV]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[VAR_ADDR_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_FLOAT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: [[SVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[S_VAR_PRIV]], // CHECK: store i{{[0-9]+}} [[SVAR_VAL]], i{{[0-9]+}}* [[SVAR_ADDR_REF]], // CHECK: ret void @@ -332,7 +348,7 @@ int main() { // CHECK-DAG: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK-DAG: [[S_ARR_PTR_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_PTR1]] to i8* // CHECK-DAG: [[TMP_VAL_BCAST1:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_PTR_BCAST1]], i8* [[TMP_VAL_BCAST1]], +// CHECK-DAG: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_PTR_BCAST1]], i8* align {{[0-9]+}} [[TMP_VAL_BCAST1]], // CHECK: call void @__kmpc_for_static_fini( // lastprivates @@ -345,7 +361,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_ADDR_REF1]], // CHECK: [[BCAST_VEC_ADDR_REF:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_ADDR_REF1]] to i8* // CHECK: [[BCAST_VEC_PRIV:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[BCAST_VEC_ADDR_REF]], i8* [[BCAST_VEC_PRIV]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BCAST_VEC_ADDR_REF]], i8* align {{[0-9]+}} [[BCAST_VEC_PRIV]], // CHECK: [[S_ARR_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_ADDR_REF]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_PRIV_BCAST:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_PRIV1]] to [[S_INT_TY]]* // CHECK: [[S_ARR_BEGIN_GEP:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_BEGIN]], i{{[0-9]+}} 2 @@ -356,7 +372,7 @@ int main() { // CHECK: [[S_ARR_DST_EL:%.+]] = phi [[S_INT_TY]]*{{.+}} // CHECK: [[S_ARR_DST_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_DST_EL]] to i8* // CHECK: [[S_ARR_SRC_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[S_ARR_SRC_EL]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[S_ARR_DST_BCAST]], i8* [[S_ARR_SRC_BCAST]]{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[S_ARR_DST_BCAST]], i8* align {{[0-9]+}} [[S_ARR_SRC_BCAST]]{{.+}}) // CHECK: [[S_ARR_DST_NEXT:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_DST_EL]], i{{[0-9]+}} 1 // CHECK: [[S_ARR_SRC_NEXT:%.+]] = getelementptr{{.+}} // CHECK: [[CPY_IS_FINISHED:%.+]] = icmp eq [[S_INT_TY]]* [[S_ARR_DST_NEXT]], [[S_ARR_BEGIN_GEP]] @@ -365,7 +381,7 @@ int main() { // CHECK: [[TMP_VAL1:%.+]] = load [[S_INT_TY]]*, [[S_INT_TY]]** [[TMP_PRIV1]], // CHECK: [[VAR_ADDR_REF_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[VAR_ADDR1_REF]] to i8* // CHECK: [[TMP_VAL1_BCAST:%.+]] = bitcast [[S_INT_TY]]* [[TMP_VAL1]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VAR_ADDR_REF_BCAST]], i8* [[TMP_VAL1_BCAST]],{{.+}}) +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VAR_ADDR_REF_BCAST]], i8* align {{[0-9]+}} [[TMP_VAL1_BCAST]],{{.+}}) // CHECK: ret void // CHECK: !{!"llvm.loop.vectorize.enable", i1 true} #endif diff --git a/test/OpenMP/teams_distribute_simd_lastprivate_messages.cpp b/test/OpenMP/teams_distribute_simd_lastprivate_messages.cpp index 51dbfef22933c..08605912d8e4f 100644 --- a/test/OpenMP/teams_distribute_simd_lastprivate_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_lastprivate_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_simd_linear_messages.cpp b/test/OpenMP/teams_distribute_simd_linear_messages.cpp index a53f0e62dda2f..a3984b98da827 100644 --- a/test/OpenMP/teams_distribute_simd_linear_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_linear_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + namespace X { int x; }; diff --git a/test/OpenMP/teams_distribute_simd_loop_messages.cpp b/test/OpenMP/teams_distribute_simd_loop_messages.cpp index 60e07a3db5964..8721dc50795f5 100644 --- a/test/OpenMP/teams_distribute_simd_loop_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_loop_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -fsyntax-only -fopenmp -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s +// RUN: %clang_cc1 -fsyntax-only -fopenmp-simd -x c++ -std=c++11 -fexceptions -fcxx-exceptions -verify %s + class S { int a; S() : a(0) {} // expected-note {{implicitly declared private here}} @@ -695,7 +697,7 @@ void test_loop_firstprivate_lastprivate() { S s(4); // expected-error@+2 {{lastprivate variable cannot be firstprivate}} expected-note@+2 {{defined as lastprivate}} #pragma omp target -#pragma omp teams distribute simd lastprivate(s) firstprivate(s) // expected-error {{calling a private constructor of class 'S'}} +#pragma omp teams distribute simd lastprivate(s) firstprivate(s) // expected-error {{calling a private constructor of class 'S'}} expected-warning {{Non-trivial type 'S' is mapped, only trivial types are guaranteed to be mapped correctly}} for (int i = 0; i < 16; ++i) ; } diff --git a/test/OpenMP/teams_distribute_simd_messages.cpp b/test/OpenMP/teams_distribute_simd_messages.cpp index b00de796b0ed8..79128706bc94a 100644 --- a/test/OpenMP/teams_distribute_simd_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_simd_num_teams_messages.cpp b/test/OpenMP/teams_distribute_simd_num_teams_messages.cpp index e65025a317ee5..86164642b82fe 100644 --- a/test/OpenMP/teams_distribute_simd_num_teams_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_num_teams_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_simd_private_codegen.cpp b/test/OpenMP/teams_distribute_simd_private_codegen.cpp index ca9a673e24417..fc7000a6996ca 100644 --- a/test/OpenMP/teams_distribute_simd_private_codegen.cpp +++ b/test/OpenMP/teams_distribute_simd_private_codegen.cpp @@ -1,13 +1,26 @@ -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap %s --check-prefix LAMBDA --check-prefix LAMBDA-64 + +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck -allow-deprecated-dag-overlap --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER diff --git a/test/OpenMP/teams_distribute_simd_private_messages.cpp b/test/OpenMP/teams_distribute_simd_private_messages.cpp index 90442130d263f..94fb9098a7d15 100644 --- a/test/OpenMP/teams_distribute_simd_private_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_simd_reduction_codegen.cpp b/test/OpenMP/teams_distribute_simd_reduction_codegen.cpp index 3153cce3a9234..10901dea12f96 100644 --- a/test/OpenMP/teams_distribute_simd_reduction_codegen.cpp +++ b/test/OpenMP/teams_distribute_simd_reduction_codegen.cpp @@ -5,10 +5,23 @@ // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DCHECK -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DCHECK -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/teams_distribute_simd_reduction_messages.cpp b/test/OpenMP/teams_distribute_simd_reduction_messages.cpp index 09d22bd4bd818..49b45b7bc6fb6 100644 --- a/test/OpenMP/teams_distribute_simd_reduction_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_reduction_messages.cpp @@ -1,6 +1,10 @@ -// RUN: %clang_cc1 -verify -fopenmp %s -// RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s -// RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_simd_safelen_messages.cpp b/test/OpenMP/teams_distribute_simd_safelen_messages.cpp index a323c0883c5bc..7be581a96b96b 100644 --- a/test/OpenMP/teams_distribute_simd_safelen_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_safelen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_simd_shared_messages.cpp b/test/OpenMP/teams_distribute_simd_shared_messages.cpp index 6b06f5b6f2030..b29e2bb8557e3 100644 --- a/test/OpenMP/teams_distribute_simd_shared_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_shared_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_distribute_simd_simdlen_messages.cpp b/test/OpenMP/teams_distribute_simd_simdlen_messages.cpp index a323c0883c5bc..7be581a96b96b 100644 --- a/test/OpenMP/teams_distribute_simd_simdlen_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_simdlen_messages.cpp @@ -2,6 +2,10 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++98 %s // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_simd_thread_limit_messages.cpp b/test/OpenMP/teams_distribute_simd_thread_limit_messages.cpp index b9c8ac3e36159..732bb34473554 100644 --- a/test/OpenMP/teams_distribute_simd_thread_limit_messages.cpp +++ b/test/OpenMP/teams_distribute_simd_thread_limit_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/teams_distribute_thread_limit_messages.cpp b/test/OpenMP/teams_distribute_thread_limit_messages.cpp index ec4ca7a05f272..d22cc25b5b59d 100644 --- a/test/OpenMP/teams_distribute_thread_limit_messages.cpp +++ b/test/OpenMP/teams_distribute_thread_limit_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 %s + void foo() { } diff --git a/test/OpenMP/teams_firstprivate_codegen.cpp b/test/OpenMP/teams_firstprivate_codegen.cpp index 3248bfeebd91f..a2b41440f38ce 100644 --- a/test/OpenMP/teams_firstprivate_codegen.cpp +++ b/test/OpenMP/teams_firstprivate_codegen.cpp @@ -1,24 +1,48 @@ // Test host codegen. -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 -// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-64 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 -// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 -// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} -// RUN: %clang_cc1 -DARRAY -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix ARRAY --check-prefix ARRAY-64 -// RUN: %clang_cc1 -DARRAY -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DARRAY -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix ARRAY --check-prefix ARRAY-64 -// RUN: %clang_cc1 -DARRAY -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix ARRAY --check-prefix ARRAY-32 -// RUN: %clang_cc1 -DARRAY -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -// RUN: %clang_cc1 -DARRAY -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix ARRAY --check-prefix ARRAY-32 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 +// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + +// RUN: %clang_cc1 -DARRAY -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix ARRAY --check-prefix ARRAY-64 +// RUN: %clang_cc1 -DARRAY -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DARRAY -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix ARRAY --check-prefix ARRAY-64 +// RUN: %clang_cc1 -DARRAY -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck %s --check-prefix ARRAY --check-prefix ARRAY-32 +// RUN: %clang_cc1 -DARRAY -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DARRAY -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck %s --check-prefix ARRAY --check-prefix ARRAY-32 + +// RUN: %clang_cc1 -DARRAY -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DARRAY -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DARRAY -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DARRAY -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY2 %s +// RUN: %clang_cc1 -DARRAY -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s -Wno-openmp-target +// RUN: %clang_cc1 -DARRAY -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -include-pch %t -verify %s -emit-llvm -o - -Wno-openmp-target | FileCheck --check-prefix SIMD-ONLY2 %s +// SIMD-ONLY2-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER @@ -148,7 +172,7 @@ int main() { // CHECK-64: [[SIVAR7_CONV:%.+]] = bitcast i64* [[SIVAR7_PRIV]] to i32* // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_REF]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* [[VEC_SRC]], +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[VEC_DEST]], i8* align {{[0-9]+}} [[VEC_SRC]], // CHECK: [[S_ARR_PRIV_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_FLOAT_TY]]], [2 x [[S_FLOAT_TY]]]* [[S_ARR_PRIV]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_BEGIN:%.+]] = bitcast [2 x [[S_FLOAT_TY]]]* [[S_ARR_REF]] to [[S_FLOAT_TY]]* // CHECK: [[S_ARR_PRIV_END:%.+]] = getelementptr [[S_FLOAT_TY]], [[S_FLOAT_TY]]* [[S_ARR_PRIV_BEGIN]], i{{[0-9]+}} 2 @@ -200,7 +224,7 @@ int main() { // CHECK: store i{{[0-9]+}} [[T_VAR_VAL]], i{{[0-9]+}}* [[T_VAR_PRIV]], align 128 // CHECK: [[VEC_DEST:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_PRIV]] to i8* // CHECK: [[VEC_SRC:%.+]] = bitcast [2 x i{{[0-9]+}}]* [[VEC_REF]] to i8* -// CHECK: call void @llvm.memcpy.{{.+}}(i8* [[VEC_DEST]], i8* [[VEC_SRC]], i{{[0-9]+}} {{[0-9]+}}, i{{[0-9]+}} 128, +// CHECK: call void @llvm.memcpy.{{.+}}(i8* align 128 [[VEC_DEST]], i8* align 128 [[VEC_SRC]], i{{[0-9]+}} {{[0-9]+}}, // CHECK: [[S_ARR_PRIV_BEGIN:%.+]] = getelementptr inbounds [2 x [[S_INT_TY]]], [2 x [[S_INT_TY]]]* [[S_ARR_PRIV]], i{{[0-9]+}} 0, i{{[0-9]+}} 0 // CHECK: [[S_ARR_BEGIN:%.+]] = bitcast [2 x [[S_INT_TY]]]* [[S_ARR_REF]] to [[S_INT_TY]]* // CHECK: [[S_ARR_PRIV_END:%.+]] = getelementptr [[S_INT_TY]], [[S_INT_TY]]* [[S_ARR_PRIV_BEGIN]], i{{[0-9]+}} 2 @@ -260,7 +284,7 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) { // ARRAY-DAG: store double* %{{.+}}, double** [[PRIV_VLA2]], // ARRAY: call i8* @llvm.stacksave() // ARRAY: [[SIZE:%.+]] = mul nuw i{{[0-9]+}} %{{.+}}, 8 -// ARRAY: call void @llvm.memcpy.p0i8.p0i8.i{{[0-9]+}}(i8* %{{.+}}, i8* %{{.+}}, i{{[0-9]+}} [[SIZE]], i32 128, i1 false) +// ARRAY: call void @llvm.memcpy.p0i8.p0i8.i{{[0-9]+}}(i8* align 128 %{{.+}}, i8* align 128 %{{.+}}, i{{[0-9]+}} [[SIZE]], i1 false) #pragma omp target #pragma omp teams firstprivate(a, s, vla1, vla2) s[0].St_func(s, n, vla1); @@ -278,6 +302,6 @@ void array_func(float a[3], St s[2], int n, long double vla1[n]) { // ARRAY-DAG: store double* %{{.+}}, double** [[PRIV_VLA2]], // ARRAY: call i8* @llvm.stacksave() // ARRAY: [[SIZE:%.+]] = mul nuw i{{[0-9]+}} %{{.+}}, 8 -// ARRAY: call void @llvm.memcpy.p0i8.p0i8.i{{[0-9]+}}(i8* %{{.+}}, i8* %{{.+}}, i{{[0-9]+}} [[SIZE]], i32 128, i1 false) +// ARRAY: call void @llvm.memcpy.p0i8.p0i8.i{{[0-9]+}}(i8* align 128 %{{.+}}, i8* align 128 %{{.+}}, i{{[0-9]+}} [[SIZE]], i1 false) #endif #endif diff --git a/test/OpenMP/teams_firstprivate_messages.cpp b/test/OpenMP/teams_firstprivate_messages.cpp index ba12fde86b5c2..d29a2e3960f67 100644 --- a/test/OpenMP/teams_firstprivate_messages.cpp +++ b/test/OpenMP/teams_firstprivate_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_messages.cpp b/test/OpenMP/teams_messages.cpp index 642fe73cac75a..6ed3be9da58d8 100644 --- a/test/OpenMP/teams_messages.cpp +++ b/test/OpenMP/teams_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -o - %s + void foo() { } diff --git a/test/OpenMP/teams_num_teams_messages.cpp b/test/OpenMP/teams_num_teams_messages.cpp index 01871127bda54..47b2b4eafe342 100644 --- a/test/OpenMP/teams_num_teams_messages.cpp +++ b/test/OpenMP/teams_num_teams_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/teams_private_codegen.cpp b/test/OpenMP/teams_private_codegen.cpp index 7ddf39c9b0eaa..0df111c3eb42f 100644 --- a/test/OpenMP/teams_private_codegen.cpp +++ b/test/OpenMP/teams_private_codegen.cpp @@ -5,12 +5,28 @@ // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -DLAMBDA -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix LAMBDA --check-prefix LAMBDA-32 +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -DLAMBDA -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64 // RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32 + +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} // expected-no-diagnostics #ifndef HEADER #define HEADER diff --git a/test/OpenMP/teams_private_messages.cpp b/test/OpenMP/teams_private_messages.cpp index 344ef8daf996f..30c7a0abd68fe 100644 --- a/test/OpenMP/teams_private_messages.cpp +++ b/test/OpenMP/teams_private_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp-simd %s + void foo() { } diff --git a/test/OpenMP/teams_reduction_messages.cpp b/test/OpenMP/teams_reduction_messages.cpp index 7aef97730d9e0..94a47436c123e 100644 --- a/test/OpenMP/teams_reduction_messages.cpp +++ b/test/OpenMP/teams_reduction_messages.cpp @@ -1,6 +1,10 @@ -// RUN: %clang_cc1 -verify -fopenmp -o - %s -// RUN: %clang_cc1 -verify -fopenmp -std=c++98 -o - %s -// RUN: %clang_cc1 -verify -fopenmp -std=c++11 -o - %s +// RUN: %clang_cc1 -verify -fopenmp -o - %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp -std=c++98 -o - %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp -std=c++11 -o - %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd -o - %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++98 -o - %s -Wno-openmp-target +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -o - %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_shared_messages.cpp b/test/OpenMP/teams_shared_messages.cpp index 6b013d099dce8..a83b82f37f33b 100644 --- a/test/OpenMP/teams_shared_messages.cpp +++ b/test/OpenMP/teams_shared_messages.cpp @@ -1,4 +1,6 @@ -// RUN: %clang_cc1 -verify -fopenmp %s +// RUN: %clang_cc1 -verify -fopenmp %s -Wno-openmp-target + +// RUN: %clang_cc1 -verify -fopenmp-simd %s -Wno-openmp-target void foo() { } diff --git a/test/OpenMP/teams_thread_limit_messages.cpp b/test/OpenMP/teams_thread_limit_messages.cpp index 1cb147c60781c..0c5a22cf3031c 100644 --- a/test/OpenMP/teams_thread_limit_messages.cpp +++ b/test/OpenMP/teams_thread_limit_messages.cpp @@ -1,5 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp -std=c++11 -ferror-limit 100 -o - %s +// RUN: %clang_cc1 -verify -fopenmp-simd -std=c++11 -ferror-limit 100 -o - %s + void foo() { } diff --git a/test/OpenMP/threadprivate_ast_print.cpp b/test/OpenMP/threadprivate_ast_print.cpp index 5502c62134b51..a1482031f5bfc 100644 --- a/test/OpenMP/threadprivate_ast_print.cpp +++ b/test/OpenMP/threadprivate_ast_print.cpp @@ -4,6 +4,13 @@ // RUN: %clang_cc1 -verify -fopenmp -triple x86_64-unknown-linux-gnu -ast-print %s | FileCheck %s // RUN: %clang_cc1 -fopenmp -fnoopenmp-use-tls -triple x86_64-unknown-linux-gnu -x c++ -std=c++11 -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fnoopenmp-use-tls -triple x86_64-unknown-linux-gnu -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print + +// RUN: %clang_cc1 -verify -fopenmp-simd -triple x86_64-apple-darwin10.6.0 -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -triple x86_64-apple-darwin10.6.0 -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -triple x86_64-apple-darwin10.6.0 -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print +// RUN: %clang_cc1 -verify -fopenmp-simd -triple x86_64-unknown-linux-gnu -ast-print %s | FileCheck %s +// RUN: %clang_cc1 -fopenmp-simd -fnoopenmp-use-tls -triple x86_64-unknown-linux-gnu -x c++ -std=c++11 -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fnoopenmp-use-tls -triple x86_64-unknown-linux-gnu -std=c++11 -include-pch %t -fsyntax-only -verify %s -ast-print // expected-no-diagnostics #ifndef HEADER @@ -18,7 +25,7 @@ struct St1{ static int b; // CHECK: static int b; #pragma omp threadprivate(b) -// CHECK-NEXT: #pragma omp threadprivate(St1::b) +// CHECK-NEXT: #pragma omp threadprivate(St1::b){{$}} } d; int a, b; diff --git a/test/OpenMP/threadprivate_codegen.cpp b/test/OpenMP/threadprivate_codegen.cpp index 3785b70c19bae..1f403111e7e27 100644 --- a/test/OpenMP/threadprivate_codegen.cpp +++ b/test/OpenMP/threadprivate_codegen.cpp @@ -2,10 +2,20 @@ // RUN: %clang_cc1 -fopenmp -fnoopenmp-use-tls -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -fnoopenmp-use-tls -DBODY -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix=CHECK-DEBUG %s +// RUN: %clang_cc1 -verify -fopenmp-simd -fnoopenmp-use-tls -DBODY -triple x86_64-unknown-unknown -x c++ -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// RUN: %clang_cc1 -fopenmp-simd -fnoopenmp-use-tls -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -fnoopenmp-use-tls -DBODY -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + // RUN: %clang_cc1 -verify -fopenmp -DBODY -triple x86_64-unknown-unknown -x c++ -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s --check-prefix=CHECK-TLS // RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s // RUN: %clang_cc1 -fopenmp -DBODY -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix=CHECK-TLS %s +// RUN: %clang_cc1 -verify -fopenmp-simd -DBODY -triple x86_64-unknown-unknown -x c++ -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s +// RUN: %clang_cc1 -fopenmp-simd -DBODY -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s +// SIMD-ONLY1-NOT: {{__kmpc|__tgt}} + // expected-no-diagnostics // #ifndef HEADER @@ -144,26 +154,26 @@ struct S5 { // CHECK-DEBUG-DAG: [[ST_INT_ST:@.+]] = linkonce_odr global i32 23 // CHECK-DEBUG-DAG: [[ST_FLOAT_ST:@.+]] = linkonce_odr global float 2.300000e+01 // CHECK-DEBUG-DAG: [[ST_S4_ST:@.+]] = linkonce_odr global %struct.S4 zeroinitializer -// CHECK-DEBUG-DAG: [[LOC1:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;;192;9;;\00" -// CHECK-DEBUG-DAG: [[LOC2:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;;247;9;;\00" -// CHECK-DEBUG-DAG: [[LOC3:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;;334;19;;\00" -// CHECK-DEBUG-DAG: [[LOC4:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;371;9;;\00" -// CHECK-DEBUG-DAG: [[LOC5:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;388;9;;\00" -// CHECK-DEBUG-DAG: [[LOC6:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;410;10;;\00" -// CHECK-DEBUG-DAG: [[LOC7:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;433;10;;\00" -// CHECK-DEBUG-DAG: [[LOC8:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;469;10;;\00" -// CHECK-DEBUG-DAG: [[LOC9:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;498;10;;\00" -// CHECK-DEBUG-DAG: [[LOC10:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;518;10;;\00" -// CHECK-DEBUG-DAG: [[LOC11:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;541;27;;\00" -// CHECK-DEBUG-DAG: [[LOC12:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;564;10;;\00" -// CHECK-DEBUG-DAG: [[LOC13:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;foobar;684;9;;\00" -// CHECK-DEBUG-DAG: [[LOC14:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;foobar;707;10;;\00" -// CHECK-DEBUG-DAG: [[LOC15:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;foobar;743;10;;\00" -// CHECK-DEBUG-DAG: [[LOC16:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;foobar;772;10;;\00" -// CHECK-DEBUG-DAG: [[LOC17:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;foobar;792;10;;\00" -// CHECK-DEBUG-DAG: [[LOC18:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;foobar;815;27;;\00" -// CHECK-DEBUG-DAG: [[LOC19:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;foobar;838;10;;\00" -// CHECK-DEBUG-DAG: [[LOC20:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;;306;9;;\00" +// CHECK-DEBUG-DAG: [[LOC1:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;;201;9;;\00" +// CHECK-DEBUG-DAG: [[LOC2:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;;256;9;;\00" +// CHECK-DEBUG-DAG: [[LOC3:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;;343;19;;\00" +// CHECK-DEBUG-DAG: [[LOC4:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;380;9;;\00" +// CHECK-DEBUG-DAG: [[LOC5:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;397;9;;\00" +// CHECK-DEBUG-DAG: [[LOC6:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;419;10;;\00" +// CHECK-DEBUG-DAG: [[LOC7:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;442;10;;\00" +// CHECK-DEBUG-DAG: [[LOC8:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;478;10;;\00" +// CHECK-DEBUG-DAG: [[LOC9:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;507;10;;\00" +// CHECK-DEBUG-DAG: [[LOC10:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;527;10;;\00" +// CHECK-DEBUG-DAG: [[LOC11:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;550;27;;\00" +// CHECK-DEBUG-DAG: [[LOC12:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;main;573;10;;\00" +// CHECK-DEBUG-DAG: [[LOC13:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;foobar;693;9;;\00" +// CHECK-DEBUG-DAG: [[LOC14:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;foobar;716;10;;\00" +// CHECK-DEBUG-DAG: [[LOC15:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;foobar;752;10;;\00" +// CHECK-DEBUG-DAG: [[LOC16:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;foobar;781;10;;\00" +// CHECK-DEBUG-DAG: [[LOC17:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;foobar;801;10;;\00" +// CHECK-DEBUG-DAG: [[LOC18:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;foobar;824;27;;\00" +// CHECK-DEBUG-DAG: [[LOC19:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;foobar;847;10;;\00" +// CHECK-DEBUG-DAG: [[LOC20:@.*]] = private unnamed_addr constant [{{[0-9]+}} x i8] c";{{.*}}threadprivate_codegen.cpp;;315;9;;\00" // CHECK-TLS-DAG: [[GS1:@.+]] = internal thread_local global [[S1]] zeroinitializer // CHECK-TLS-DAG: [[GS2:@.+]] = internal global [[S2]] zeroinitializer // CHECK-TLS-DAG: [[ARR_X:@.+]] = thread_local global [2 x [3 x [[S1]]]] zeroinitializer @@ -180,7 +190,6 @@ struct S5 { // CHECK-TLS-DAG: [[GS1_TLS_INIT:@_ZTHL3gs1]] = internal alias void (), void ()* @__tls_init // CHECK-TLS-DAG: [[ARR_X_TLS_INIT:@_ZTH5arr_x]] = alias void (), void ()* @__tls_init - // CHECK-TLS-DAG: [[ST_S4_ST_TLS_INIT:@_ZTHN2STI2S4E2stE]] = linkonce_odr alias void (), void ()* [[ST_S4_ST_CXX_INIT:@[^, ]*]] struct Static { diff --git a/test/OpenMP/threadprivate_messages.cpp b/test/OpenMP/threadprivate_messages.cpp index 9775bfa458f49..62ddfd1e639c0 100644 --- a/test/OpenMP/threadprivate_messages.cpp +++ b/test/OpenMP/threadprivate_messages.cpp @@ -1,6 +1,9 @@ // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -fnoopenmp-use-tls -ferror-limit 100 -emit-llvm -o - %s // RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp -ferror-limit 100 -emit-llvm -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -fnoopenmp-use-tls -ferror-limit 100 -emit-llvm -o - %s +// RUN: %clang_cc1 -triple x86_64-apple-macos10.7.0 -verify -fopenmp-simd -ferror-limit 100 -emit-llvm -o - %s + #pragma omp threadprivate // expected-error {{expected '(' after 'threadprivate'}} #pragma omp threadprivate( // expected-error {{expected identifier}} expected-error {{expected ')'}} expected-note {{to match this '('}} #pragma omp threadprivate() // expected-error {{expected identifier}} diff --git a/test/OpenMP/varargs.cpp b/test/OpenMP/varargs.cpp index 7738f83fab241..4730c1025d324 100644 --- a/test/OpenMP/varargs.cpp +++ b/test/OpenMP/varargs.cpp @@ -1,4 +1,7 @@ // RUN: %clang_cc1 -verify -fopenmp %s + +// RUN: %clang_cc1 -verify -fopenmp-simd %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} void f(int a, ...) { #pragma omp parallel for for (int i = 0; i < 100; ++i) { diff --git a/test/OpenMP/vla_crash.c b/test/OpenMP/vla_crash.c index 0dcba5503e4cd..00cada4a64533 100644 --- a/test/OpenMP/vla_crash.c +++ b/test/OpenMP/vla_crash.c @@ -1,5 +1,8 @@ // RUN: %clang_cc1 -verify -triple powerpc64le-unknown-linux-gnu -fopenmp -x c -emit-llvm %s -o - | FileCheck %s +// RUN: %clang_cc1 -verify -triple powerpc64le-unknown-linux-gnu -fopenmp-simd -x c -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s +// SIMD-ONLY0-NOT: {{__kmpc|__tgt}} + int a; // CHECK-LABEL: foo |