diff options
Diffstat (limited to 'test/CodeGen/X86/vec_shift4.ll')
| -rw-r--r-- | test/CodeGen/X86/vec_shift4.ll | 64 |
1 files changed, 63 insertions, 1 deletions
diff --git a/test/CodeGen/X86/vec_shift4.ll b/test/CodeGen/X86/vec_shift4.ll index b266a6987557..66229361990f 100644 --- a/test/CodeGen/X86/vec_shift4.ll +++ b/test/CodeGen/X86/vec_shift4.ll @@ -1,6 +1,23 @@ -; RUN: llc < %s -march=x86 -mcpu=corei7 | FileCheck %s +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=X32 +; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=X64 define <2 x i64> @shl1(<4 x i32> %r, <4 x i32> %a) nounwind readnone ssp { +; X32-LABEL: shl1: +; X32: # BB#0: # %entry +; X32-NEXT: pslld $23, %xmm1 +; X32-NEXT: paddd {{\.LCPI.*}}, %xmm1 +; X32-NEXT: cvttps2dq %xmm1, %xmm1 +; X32-NEXT: pmulld %xmm1, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: shl1: +; X64: # BB#0: # %entry +; X64-NEXT: pslld $23, %xmm1 +; X64-NEXT: paddd {{.*}}(%rip), %xmm1 +; X64-NEXT: cvttps2dq %xmm1, %xmm1 +; X64-NEXT: pmulld %xmm1, %xmm0 +; X64-NEXT: retq entry: ; CHECK-NOT: shll ; CHECK: pslld @@ -14,6 +31,51 @@ entry: } define <2 x i64> @shl2(<16 x i8> %r, <16 x i8> %a) nounwind readnone ssp { +; X32-LABEL: shl2: +; X32: # BB#0: # %entry +; X32-NEXT: movdqa %xmm0, %xmm2 +; X32-NEXT: psllw $5, %xmm1 +; X32-NEXT: movdqa %xmm2, %xmm3 +; X32-NEXT: psllw $4, %xmm3 +; X32-NEXT: pand {{\.LCPI.*}}, %xmm3 +; X32-NEXT: movdqa %xmm1, %xmm0 +; X32-NEXT: pblendvb %xmm3, %xmm2 +; X32-NEXT: movdqa %xmm2, %xmm3 +; X32-NEXT: psllw $2, %xmm3 +; X32-NEXT: pand {{\.LCPI.*}}, %xmm3 +; X32-NEXT: paddb %xmm1, %xmm1 +; X32-NEXT: movdqa %xmm1, %xmm0 +; X32-NEXT: pblendvb %xmm3, %xmm2 +; X32-NEXT: movdqa %xmm2, %xmm3 +; X32-NEXT: paddb %xmm3, %xmm3 +; X32-NEXT: paddb %xmm1, %xmm1 +; X32-NEXT: movdqa %xmm1, %xmm0 +; X32-NEXT: pblendvb %xmm3, %xmm2 +; X32-NEXT: movdqa %xmm2, %xmm0 +; X32-NEXT: retl +; +; X64-LABEL: shl2: +; X64: # BB#0: # %entry +; X64-NEXT: movdqa %xmm0, %xmm2 +; X64-NEXT: psllw $5, %xmm1 +; X64-NEXT: movdqa %xmm2, %xmm3 +; X64-NEXT: psllw $4, %xmm3 +; X64-NEXT: pand {{.*}}(%rip), %xmm3 +; X64-NEXT: movdqa %xmm1, %xmm0 +; X64-NEXT: pblendvb %xmm3, %xmm2 +; X64-NEXT: movdqa %xmm2, %xmm3 +; X64-NEXT: psllw $2, %xmm3 +; X64-NEXT: pand {{.*}}(%rip), %xmm3 +; X64-NEXT: paddb %xmm1, %xmm1 +; X64-NEXT: movdqa %xmm1, %xmm0 +; X64-NEXT: pblendvb %xmm3, %xmm2 +; X64-NEXT: movdqa %xmm2, %xmm3 +; X64-NEXT: paddb %xmm3, %xmm3 +; X64-NEXT: paddb %xmm1, %xmm1 +; X64-NEXT: movdqa %xmm1, %xmm0 +; X64-NEXT: pblendvb %xmm3, %xmm2 +; X64-NEXT: movdqa %xmm2, %xmm0 +; X64-NEXT: retq entry: ; CHECK-NOT: shlb ; CHECK: pblendvb |