diff options
| author | Craig Topper <craig.topper@intel.com> | 2019-08-19 18:15:50 +0000 |
|---|---|---|
| committer | Craig Topper <craig.topper@intel.com> | 2019-08-19 18:15:50 +0000 |
| commit | a0d92c72620c49aa36b1738a272a2715f7909a6a (patch) | |
| tree | 17423b1397d57b7dee76e2b5f5ec93536b512103 /llvm/test/CodeGen/X86/avg.ll | |
| parent | a8abe1f82899847e29c4f1d66c32fad17dacb62f (diff) | |
| download | bcm5719-llvm-a0d92c72620c49aa36b1738a272a2715f7909a6a.tar.gz bcm5719-llvm-a0d92c72620c49aa36b1738a272a2715f7909a6a.zip | |
[X86] Teach lowerV4I32Shuffle to only use broadcasts if the mask has more than one undef element. Prioritize shifts over broadcast in lowerV8I16Shuffle.
The motivating case are the changes in vector-reduce-add.ll where
we were doing extra work in the scalar domain instead of shuffling.
There may be some one use check that needs to be looked into there,
but this patch sidesteps the issue by avoiding broadcasts that
aren't really broadcasting.
Differential Revision: https://reviews.llvm.org/D66071
llvm-svn: 369287
Diffstat (limited to 'llvm/test/CodeGen/X86/avg.ll')
| -rw-r--r-- | llvm/test/CodeGen/X86/avg.ll | 8 |
1 files changed, 4 insertions, 4 deletions
diff --git a/llvm/test/CodeGen/X86/avg.ll b/llvm/test/CodeGen/X86/avg.ll index f5047b1d640..e2819151ce7 100644 --- a/llvm/test/CodeGen/X86/avg.ll +++ b/llvm/test/CodeGen/X86/avg.ll @@ -2481,23 +2481,23 @@ define void @not_avg_v16i8_wide_constants(<16 x i8>* %a, <16 x i8>* %b) nounwind ; AVX2-NEXT: # xmm7 = mem[0],zero ; AVX2-NEXT: vpunpcklbw {{.*#+}} xmm8 = xmm9[0],xmm8[0],xmm9[1],xmm8[1],xmm9[2],xmm8[2],xmm9[3],xmm8[3],xmm9[4],xmm8[4],xmm9[5],xmm8[5],xmm9[6],xmm8[6],xmm9[7],xmm8[7] ; AVX2-NEXT: vpunpcklbw {{.*#+}} xmm9 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] -; AVX2-NEXT: vpbroadcastw %xmm8, %xmm8 +; AVX2-NEXT: vpslldq {{.*#+}} xmm8 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm8[0,1] ; AVX2-NEXT: vpbroadcastw %xmm9, %xmm0 ; AVX2-NEXT: vpblendw {{.*#+}} xmm8 = xmm0[0,1,2,3,4,5,6],xmm8[7] ; AVX2-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3],xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7] ; AVX2-NEXT: vpunpcklbw {{.*#+}} xmm9 = xmm15[0],xmm14[0],xmm15[1],xmm14[1],xmm15[2],xmm14[2],xmm15[3],xmm14[3],xmm15[4],xmm14[4],xmm15[5],xmm14[5],xmm15[6],xmm14[6],xmm15[7],xmm14[7] -; AVX2-NEXT: vpbroadcastw %xmm0, %xmm0 +; AVX2-NEXT: vpslldq {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5] ; AVX2-NEXT: vpbroadcastw %xmm9, %xmm1 ; AVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3,4],xmm0[5],xmm1[6,7] ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm0[0,1,2],xmm8[3] ; AVX2-NEXT: vpunpcklbw {{.*#+}} xmm1 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3],xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7] ; AVX2-NEXT: vpunpcklbw {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7] -; AVX2-NEXT: vpbroadcastw %xmm1, %xmm1 +; AVX2-NEXT: vpsllq $48, %xmm1, %xmm1 ; AVX2-NEXT: vpbroadcastw %xmm2, %xmm2 ; AVX2-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1,2],xmm1[3],xmm2[4,5,6,7] ; AVX2-NEXT: vpunpcklbw {{.*#+}} xmm2 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3],xmm5[4],xmm4[4],xmm5[5],xmm4[5],xmm5[6],xmm4[6],xmm5[7],xmm4[7] ; AVX2-NEXT: vpunpcklbw {{.*#+}} xmm3 = xmm7[0],xmm6[0],xmm7[1],xmm6[1],xmm7[2],xmm6[2],xmm7[3],xmm6[3],xmm7[4],xmm6[4],xmm7[5],xmm6[5],xmm7[6],xmm6[6],xmm7[7],xmm6[7] -; AVX2-NEXT: vpbroadcastw %xmm3, %xmm3 +; AVX2-NEXT: vpslld $16, %xmm3, %xmm3 ; AVX2-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0],xmm3[1],xmm2[2,3,4,5,6,7] ; AVX2-NEXT: vpblendd {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2,3] ; AVX2-NEXT: vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3] |
