112 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+prefer-256-bit | FileCheck %s --check-prefix=AVX2563; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,-prefer-256-bit | FileCheck %s --check-prefix=AVX5124 5define <16 x i1> @smulo_v16i8(<16 x i8> %a0, <16 x i8> %a1, ptr %p2) nounwind {6; AVX256-LABEL: smulo_v16i8:7; AVX256: # %bb.0:8; AVX256-NEXT: vpmovsxbw %xmm1, %ymm19; AVX256-NEXT: vpmovsxbw %xmm0, %ymm010; AVX256-NEXT: vpmullw %ymm1, %ymm0, %ymm011; AVX256-NEXT: vpsrlw $8, %ymm0, %ymm112; AVX256-NEXT: vextracti128 $1, %ymm1, %xmm213; AVX256-NEXT: vpackuswb %xmm2, %xmm1, %xmm114; AVX256-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %ymm015; AVX256-NEXT: vextracti128 $1, %ymm0, %xmm216; AVX256-NEXT: vpackuswb %xmm2, %xmm0, %xmm017; AVX256-NEXT: vpxor %xmm2, %xmm2, %xmm218; AVX256-NEXT: vpcmpgtb %xmm0, %xmm2, %xmm219; AVX256-NEXT: vpcmpeqb %xmm1, %xmm2, %xmm120; AVX256-NEXT: vpternlogq {{.*#+}} xmm1 = ~xmm121; AVX256-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]22; AVX256-NEXT: vpmovsxbd %xmm2, %ymm223; AVX256-NEXT: vptestmd %ymm2, %ymm2, %k124; AVX256-NEXT: vpmovsxbd %xmm1, %ymm125; AVX256-NEXT: vptestmd %ymm1, %ymm1, %k226; AVX256-NEXT: vmovdqa %xmm0, (%rdi)27; AVX256-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm028; AVX256-NEXT: vmovdqa32 %ymm0, %ymm1 {%k2} {z}29; AVX256-NEXT: vpmovdw %ymm1, %xmm130; AVX256-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}31; AVX256-NEXT: vpmovdw %ymm0, %xmm032; AVX256-NEXT: vpacksswb %xmm0, %xmm1, %xmm033; AVX256-NEXT: vzeroupper34; AVX256-NEXT: retq35;36; AVX512-LABEL: smulo_v16i8:37; AVX512: # %bb.0:38; AVX512-NEXT: vpmovsxbw %xmm1, %ymm139; AVX512-NEXT: vpmovsxbw %xmm0, %ymm040; AVX512-NEXT: vpmullw %ymm1, %ymm0, %ymm041; AVX512-NEXT: vpsraw $8, %ymm0, %ymm142; AVX512-NEXT: vpmovsxwd %ymm1, %zmm143; AVX512-NEXT: vpsllw $8, %ymm0, %ymm244; AVX512-NEXT: vpsraw $15, %ymm2, %ymm245; AVX512-NEXT: vpmovsxwd %ymm2, %zmm246; AVX512-NEXT: vpcmpneqd %zmm1, %zmm2, %k147; AVX512-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero48; AVX512-NEXT: vpmovdb %zmm0, (%rdi)49; AVX512-NEXT: vpternlogd {{.*#+}} zmm0 {%k1} {z} = -150; AVX512-NEXT: vpmovdb %zmm0, %xmm051; AVX512-NEXT: vzeroupper52; AVX512-NEXT: retq53 %t = call {<16 x i8>, <16 x i1>} @llvm.smul.with.overflow.v16i8(<16 x i8> %a0, <16 x i8> %a1)54 %val = extractvalue {<16 x i8>, <16 x i1>} %t, 055 %obit = extractvalue {<16 x i8>, <16 x i1>} %t, 156 store <16 x i8> %val, ptr %p257 ret <16 x i1> %obit58}59declare {<16 x i8>, <16 x i1>} @llvm.smul.with.overflow.v16i8(<16 x i8>, <16 x i8>)60 61define <16 x i1> @umulo_v16i8(<16 x i8> %a0, <16 x i8> %a1, ptr %p2) nounwind {62; AVX256-LABEL: umulo_v16i8:63; AVX256: # %bb.0:64; AVX256-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero65; AVX256-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero66; AVX256-NEXT: vpmullw %ymm1, %ymm0, %ymm067; AVX256-NEXT: vpsrlw $8, %ymm0, %ymm168; AVX256-NEXT: vextracti128 $1, %ymm1, %xmm269; AVX256-NEXT: vpackuswb %xmm2, %xmm1, %xmm170; AVX256-NEXT: vpxor %xmm2, %xmm2, %xmm271; AVX256-NEXT: vpcmpeqb %xmm2, %xmm1, %xmm172; AVX256-NEXT: vpternlogq {{.*#+}} xmm1 = ~xmm173; AVX256-NEXT: vpshufd {{.*#+}} xmm2 = xmm1[2,3,2,3]74; AVX256-NEXT: vpmovsxbd %xmm2, %ymm275; AVX256-NEXT: vptestmd %ymm2, %ymm2, %k176; AVX256-NEXT: vpmovsxbd %xmm1, %ymm177; AVX256-NEXT: vptestmd %ymm1, %ymm1, %k278; AVX256-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %ymm079; AVX256-NEXT: vextracti128 $1, %ymm0, %xmm180; AVX256-NEXT: vpackuswb %xmm1, %xmm0, %xmm081; AVX256-NEXT: vmovdqa %xmm0, (%rdi)82; AVX256-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm083; AVX256-NEXT: vmovdqa32 %ymm0, %ymm1 {%k2} {z}84; AVX256-NEXT: vpmovdw %ymm1, %xmm185; AVX256-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}86; AVX256-NEXT: vpmovdw %ymm0, %xmm087; AVX256-NEXT: vpacksswb %xmm0, %xmm1, %xmm088; AVX256-NEXT: vzeroupper89; AVX256-NEXT: retq90;91; AVX512-LABEL: umulo_v16i8:92; AVX512: # %bb.0:93; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero94; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero95; AVX512-NEXT: vpmullw %ymm1, %ymm0, %ymm096; AVX512-NEXT: vpsrlw $8, %ymm0, %ymm197; AVX512-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero98; AVX512-NEXT: vptestmd %zmm1, %zmm1, %k199; AVX512-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero100; AVX512-NEXT: vpmovdb %zmm0, (%rdi)101; AVX512-NEXT: vpternlogd {{.*#+}} zmm0 {%k1} {z} = -1102; AVX512-NEXT: vpmovdb %zmm0, %xmm0103; AVX512-NEXT: vzeroupper104; AVX512-NEXT: retq105 %t = call {<16 x i8>, <16 x i1>} @llvm.umul.with.overflow.v16i8(<16 x i8> %a0, <16 x i8> %a1)106 %val = extractvalue {<16 x i8>, <16 x i1>} %t, 0107 %obit = extractvalue {<16 x i8>, <16 x i1>} %t, 1108 store <16 x i8> %val, ptr %p2109 ret <16 x i1> %obit110}111declare {<16 x i8>, <16 x i1>} @llvm.umul.with.overflow.v16i8(<16 x i8>, <16 x i8>)112