diff options
Diffstat (limited to 'llvm/test/CodeGen/X86/vector-shuffle-v1.ll')
-rw-r--r-- | llvm/test/CodeGen/X86/vector-shuffle-v1.ll | 50 |
1 files changed, 32 insertions, 18 deletions
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-v1.ll b/llvm/test/CodeGen/X86/vector-shuffle-v1.ll index d926e5427dd..94b00fbd937 100644 --- a/llvm/test/CodeGen/X86/vector-shuffle-v1.ll +++ b/llvm/test/CodeGen/X86/vector-shuffle-v1.ll @@ -267,42 +267,50 @@ define <32 x i1> @shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0 define <32 x i16> @shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_icmp_v32i16(<32 x i16> %a, <32 x i16> %c, <32 x i16> %d) { ; AVX512F-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_icmp_v32i16: ; AVX512F: # %bb.0: +; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm3 +; AVX512F-NEXT: vextracti64x4 $1, %zmm2, %ymm4 +; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm5 ; AVX512F-NEXT: vpxor %xmm6, %xmm6, %xmm6 ; AVX512F-NEXT: vpcmpeqw %ymm6, %ymm0, %ymm0 ; AVX512F-NEXT: vpmovsxwd %ymm0, %zmm0 ; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k1 -; AVX512F-NEXT: vpcmpeqw %ymm6, %ymm1, %ymm0 +; AVX512F-NEXT: vpcmpeqw %ymm6, %ymm5, %ymm0 ; AVX512F-NEXT: vpmovsxwd %ymm0, %zmm0 ; AVX512F-NEXT: vptestmd %zmm0, %zmm0, %k2 ; AVX512F-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z} -; AVX512F-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z} +; AVX512F-NEXT: vpternlogd $255, %zmm5, %zmm5, %zmm5 {%k1} {z} ; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm6 = [3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0] -; AVX512F-NEXT: vpermi2d %zmm0, %zmm1, %zmm6 +; AVX512F-NEXT: vpermi2d %zmm0, %zmm5, %zmm6 ; AVX512F-NEXT: vptestmd %zmm6, %zmm6, %k1 ; AVX512F-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} -; AVX512F-NEXT: vpmovdw %zmm0, %ymm1 -; AVX512F-NEXT: vpblendvb %ymm1, %ymm2, %ymm4, %ymm0 -; AVX512F-NEXT: vpblendvb %ymm1, %ymm3, %ymm5, %ymm1 +; AVX512F-NEXT: vpmovdw %zmm0, %ymm0 +; AVX512F-NEXT: vpblendvb %ymm0, %ymm3, %ymm4, %ymm3 +; AVX512F-NEXT: vpblendvb %ymm0, %ymm1, %ymm2, %ymm0 +; AVX512F-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0 ; AVX512F-NEXT: retq ; ; AVX512VL-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_icmp_v32i16: ; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm3 +; AVX512VL-NEXT: vextracti64x4 $1, %zmm2, %ymm4 +; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm5 ; AVX512VL-NEXT: vpxor %xmm6, %xmm6, %xmm6 ; AVX512VL-NEXT: vpcmpeqw %ymm6, %ymm0, %ymm0 ; AVX512VL-NEXT: vpmovsxwd %ymm0, %zmm0 ; AVX512VL-NEXT: vptestmd %zmm0, %zmm0, %k1 -; AVX512VL-NEXT: vpcmpeqw %ymm6, %ymm1, %ymm0 +; AVX512VL-NEXT: vpcmpeqw %ymm6, %ymm5, %ymm0 ; AVX512VL-NEXT: vpmovsxwd %ymm0, %zmm0 ; AVX512VL-NEXT: vptestmd %zmm0, %zmm0, %k2 ; AVX512VL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z} -; AVX512VL-NEXT: vpternlogd $255, %zmm1, %zmm1, %zmm1 {%k1} {z} +; AVX512VL-NEXT: vpternlogd $255, %zmm5, %zmm5, %zmm5 {%k1} {z} ; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm6 = [3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0] -; AVX512VL-NEXT: vpermi2d %zmm0, %zmm1, %zmm6 +; AVX512VL-NEXT: vpermi2d %zmm0, %zmm5, %zmm6 ; AVX512VL-NEXT: vptestmd %zmm6, %zmm6, %k1 ; AVX512VL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} -; AVX512VL-NEXT: vpmovdw %zmm0, %ymm1 -; AVX512VL-NEXT: vpblendvb %ymm1, %ymm2, %ymm4, %ymm0 -; AVX512VL-NEXT: vpblendvb %ymm1, %ymm3, %ymm5, %ymm1 +; AVX512VL-NEXT: vpmovdw %zmm0, %ymm0 +; AVX512VL-NEXT: vpblendvb %ymm0, %ymm3, %ymm4, %ymm3 +; AVX512VL-NEXT: vpblendvb %ymm0, %ymm1, %ymm2, %ymm0 +; AVX512VL-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0 ; AVX512VL-NEXT: retq ; ; VL_BW_DQ-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_icmp_v32i16: @@ -381,6 +389,8 @@ define <32 x i8> @shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0 define <32 x i16> @shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_icmp_v32i16_split(<16 x i32> %a, <16 x i32> %b, <32 x i16> %c, <32 x i16> %d) { ; AVX512F-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_icmp_v32i16_split: ; AVX512F: # %bb.0: +; AVX512F-NEXT: vextracti64x4 $1, %zmm2, %ymm4 +; AVX512F-NEXT: vextracti64x4 $1, %zmm3, %ymm5 ; AVX512F-NEXT: vptestnmd %zmm0, %zmm0, %k1 ; AVX512F-NEXT: vptestnmd %zmm1, %zmm1, %k2 ; AVX512F-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z} @@ -389,13 +399,16 @@ define <32 x i16> @shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_ ; AVX512F-NEXT: vpermi2d %zmm0, %zmm1, %zmm6 ; AVX512F-NEXT: vptestmd %zmm6, %zmm6, %k1 ; AVX512F-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} -; AVX512F-NEXT: vpmovdw %zmm0, %ymm1 -; AVX512F-NEXT: vpblendvb %ymm1, %ymm2, %ymm4, %ymm0 -; AVX512F-NEXT: vpblendvb %ymm1, %ymm3, %ymm5, %ymm1 +; AVX512F-NEXT: vpmovdw %zmm0, %ymm0 +; AVX512F-NEXT: vpblendvb %ymm0, %ymm4, %ymm5, %ymm1 +; AVX512F-NEXT: vpblendvb %ymm0, %ymm2, %ymm3, %ymm0 +; AVX512F-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0 ; AVX512F-NEXT: retq ; ; AVX512VL-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_icmp_v32i16_split: ; AVX512VL: # %bb.0: +; AVX512VL-NEXT: vextracti64x4 $1, %zmm2, %ymm4 +; AVX512VL-NEXT: vextracti64x4 $1, %zmm3, %ymm5 ; AVX512VL-NEXT: vptestnmd %zmm0, %zmm0, %k1 ; AVX512VL-NEXT: vptestnmd %zmm1, %zmm1, %k2 ; AVX512VL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k2} {z} @@ -404,9 +417,10 @@ define <32 x i16> @shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_ ; AVX512VL-NEXT: vpermi2d %zmm0, %zmm1, %zmm6 ; AVX512VL-NEXT: vptestmd %zmm6, %zmm6, %k1 ; AVX512VL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} -; AVX512VL-NEXT: vpmovdw %zmm0, %ymm1 -; AVX512VL-NEXT: vpblendvb %ymm1, %ymm2, %ymm4, %ymm0 -; AVX512VL-NEXT: vpblendvb %ymm1, %ymm3, %ymm5, %ymm1 +; AVX512VL-NEXT: vpmovdw %zmm0, %ymm0 +; AVX512VL-NEXT: vpblendvb %ymm0, %ymm4, %ymm5, %ymm1 +; AVX512VL-NEXT: vpblendvb %ymm0, %ymm2, %ymm3, %ymm0 +; AVX512VL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0 ; AVX512VL-NEXT: retq ; ; VL_BW_DQ-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_icmp_v32i16_split: |