diff options
| author | Simon Pilgrim <llvm-dev@redking.me.uk> | 2016-03-26 09:44:27 +0000 |
|---|---|---|
| committer | Simon Pilgrim <llvm-dev@redking.me.uk> | 2016-03-26 09:44:27 +0000 |
| commit | 7379a70677180605fce5e34ea007c3edb5345537 (patch) | |
| tree | 7a96798b3c420f4a15881f049187a4f91426f895 /llvm/test/CodeGen/X86/pmul.ll | |
| parent | b549ab02b4659c2e6164cb1a39ef798b80d643df (diff) | |
| download | bcm5719-llvm-7379a70677180605fce5e34ea007c3edb5345537.tar.gz bcm5719-llvm-7379a70677180605fce5e34ea007c3edb5345537.zip | |
[X86][AVX512BW] AVX512BW can sign-extend v32i8 to v32i16 for simpler v32i8 multiplies.
Only pre-AVX512BW targets need to split v32i8 vectors.
llvm-svn: 264509
Diffstat (limited to 'llvm/test/CodeGen/X86/pmul.ll')
| -rw-r--r-- | llvm/test/CodeGen/X86/pmul.ll | 25 |
1 files changed, 7 insertions, 18 deletions
diff --git a/llvm/test/CodeGen/X86/pmul.ll b/llvm/test/CodeGen/X86/pmul.ll index 89c35ba1803..d5fffa8826c 100644 --- a/llvm/test/CodeGen/X86/pmul.ll +++ b/llvm/test/CodeGen/X86/pmul.ll @@ -473,15 +473,11 @@ define <32 x i8> @mul_v32i8c(<32 x i8> %i) nounwind { ; ; AVX512BW-LABEL: mul_v32i8c: ; AVX512BW: # BB#0: # %entry -; AVX512BW-NEXT: vpmovsxbw %xmm0, %ymm1 -; AVX512BW-NEXT: vpmovsxbw {{.*}}(%rip), %ymm2 -; AVX512BW-NEXT: vpmullw %ymm2, %ymm1, %ymm1 -; AVX512BW-NEXT: vpmovwb %zmm1, %ymm1 -; AVX512BW-NEXT: vextracti128 $1, %ymm0, %xmm0 -; AVX512BW-NEXT: vpmovsxbw %xmm0, %ymm0 -; AVX512BW-NEXT: vpmullw %ymm2, %ymm0, %ymm0 +; AVX512BW-NEXT: vmovaps {{.*#+}} ymm1 = [117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117] +; AVX512BW-NEXT: vpmovsxbw %ymm1, %zmm1 +; AVX512BW-NEXT: vpmovsxbw %ymm0, %zmm0 +; AVX512BW-NEXT: vpmullw %zmm1, %zmm0, %zmm0 ; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 ; AVX512BW-NEXT: retq entry: %A = mul <32 x i8> %i, < i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117 > @@ -678,17 +674,10 @@ define <32 x i8> @mul_v32i8(<32 x i8> %i, <32 x i8> %j) nounwind { ; ; AVX512BW-LABEL: mul_v32i8: ; AVX512BW: # BB#0: # %entry -; AVX512BW-NEXT: vpmovsxbw %xmm1, %ymm2 -; AVX512BW-NEXT: vpmovsxbw %xmm0, %ymm3 -; AVX512BW-NEXT: vpmullw %ymm2, %ymm3, %ymm2 -; AVX512BW-NEXT: vpmovwb %zmm2, %ymm2 -; AVX512BW-NEXT: vextracti128 $1, %ymm1, %xmm1 -; AVX512BW-NEXT: vpmovsxbw %xmm1, %ymm1 -; AVX512BW-NEXT: vextracti128 $1, %ymm0, %xmm0 -; AVX512BW-NEXT: vpmovsxbw %xmm0, %ymm0 -; AVX512BW-NEXT: vpmullw %ymm1, %ymm0, %ymm0 +; AVX512BW-NEXT: vpmovsxbw %ymm1, %zmm1 +; AVX512BW-NEXT: vpmovsxbw %ymm0, %zmm0 +; AVX512BW-NEXT: vpmullw %zmm1, %zmm0, %zmm0 ; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: vinsertf128 $1, %xmm0, %ymm2, %ymm0 ; AVX512BW-NEXT: retq entry: %A = mul <32 x i8> %i, %j |
