summaryrefslogtreecommitdiff
path: root/test/OpenMP/nvptx_teams_reduction_codegen.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'test/OpenMP/nvptx_teams_reduction_codegen.cpp')
-rw-r--r--test/OpenMP/nvptx_teams_reduction_codegen.cpp177
1 files changed, 79 insertions, 98 deletions
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]]