diff options
| author | Craig Topper <craig.topper@intel.com> | 2018-02-09 23:32:27 +0000 |
|---|---|---|
| committer | Craig Topper <craig.topper@intel.com> | 2018-02-09 23:32:27 +0000 |
| commit | 99db883d55313ac98439b8031383abf82dcaafde (patch) | |
| tree | 7444ba59d9a9bb4c8856732c5c40ce9ec2c48d54 /llvm/test | |
| parent | 7512c3e15f4c2da39bd4b62b03083b7c7facd405 (diff) | |
| download | bcm5719-llvm-99db883d55313ac98439b8031383abf82dcaafde.tar.gz bcm5719-llvm-99db883d55313ac98439b8031383abf82dcaafde.zip | |
[X86] Teach lower1BitVectorShuffle to recognize shuffles that are just filling upper elements with zero. Replace with insert_subvector.
There's still some extra kshifts in one of the modified test cases here, but hopefully that's only a DAG combine away.
llvm-svn: 324782
Diffstat (limited to 'llvm/test')
| -rw-r--r-- | llvm/test/CodeGen/X86/avx512-mask-op.ll | 22 |
1 files changed, 8 insertions, 14 deletions
diff --git a/llvm/test/CodeGen/X86/avx512-mask-op.ll b/llvm/test/CodeGen/X86/avx512-mask-op.ll index 40d91356843..44e0a74db2e 100644 --- a/llvm/test/CodeGen/X86/avx512-mask-op.ll +++ b/llvm/test/CodeGen/X86/avx512-mask-op.ll @@ -2757,19 +2757,16 @@ define <8 x i64> @mask_widening(<2 x i64> %a, <2 x i64> %b, <2 x i64> %c, <2 x i ; KNL-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 ; KNL-NEXT: vpcmpeqd %zmm1, %zmm0, %k0 ; KNL-NEXT: kshiftlw $12, %k0, %k0 -; KNL-NEXT: kshiftrw $12, %k0, %k1 -; KNL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} -; KNL-NEXT: vmovdqa %ymm0, %ymm0 -; KNL-NEXT: vptestmd %zmm0, %zmm0, %k1 +; KNL-NEXT: kshiftrw $12, %k0, %k0 +; KNL-NEXT: kshiftlw $8, %k0, %k0 +; KNL-NEXT: kshiftrw $8, %k0, %k1 ; KNL-NEXT: vpblendmd %zmm5, %zmm4, %zmm0 {%k1} ; KNL-NEXT: retq ; ; SKX-LABEL: mask_widening: ; SKX: ## %bb.0: ## %entry ; SKX-NEXT: vpcmpeqd %xmm1, %xmm0, %k0 -; SKX-NEXT: vpmovm2d %k0, %zmm0 -; SKX-NEXT: vmovdqa %ymm0, %ymm0 -; SKX-NEXT: vpmovd2m %zmm0, %k1 +; SKX-NEXT: kmovb %k0, %k1 ; SKX-NEXT: vpblendmd %zmm5, %zmm4, %zmm0 {%k1} ; SKX-NEXT: retq ; @@ -2779,10 +2776,9 @@ define <8 x i64> @mask_widening(<2 x i64> %a, <2 x i64> %b, <2 x i64> %c, <2 x i ; AVX512BW-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 ; AVX512BW-NEXT: vpcmpeqd %zmm1, %zmm0, %k0 ; AVX512BW-NEXT: kshiftlw $12, %k0, %k0 -; AVX512BW-NEXT: kshiftrw $12, %k0, %k1 -; AVX512BW-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} -; AVX512BW-NEXT: vmovdqa %ymm0, %ymm0 -; AVX512BW-NEXT: vptestmd %zmm0, %zmm0, %k1 +; AVX512BW-NEXT: kshiftrw $12, %k0, %k0 +; AVX512BW-NEXT: kshiftlw $8, %k0, %k0 +; AVX512BW-NEXT: kshiftrw $8, %k0, %k1 ; AVX512BW-NEXT: vpblendmd %zmm5, %zmm4, %zmm0 {%k1} ; AVX512BW-NEXT: retq ; @@ -2793,9 +2789,7 @@ define <8 x i64> @mask_widening(<2 x i64> %a, <2 x i64> %b, <2 x i64> %c, <2 x i ; AVX512DQ-NEXT: vpcmpeqd %zmm1, %zmm0, %k0 ; AVX512DQ-NEXT: kshiftlb $4, %k0, %k0 ; AVX512DQ-NEXT: kshiftrb $4, %k0, %k0 -; AVX512DQ-NEXT: vpmovm2d %k0, %zmm0 -; AVX512DQ-NEXT: vmovdqa %ymm0, %ymm0 -; AVX512DQ-NEXT: vpmovd2m %zmm0, %k1 +; AVX512DQ-NEXT: kmovb %k0, %k1 ; AVX512DQ-NEXT: vpblendmd %zmm5, %zmm4, %zmm0 {%k1} ; AVX512DQ-NEXT: retq entry: |
