124 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,+avx512cd,+prefer-256-bit | FileCheck %s --check-prefix=AVX2563; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512cd,-prefer-256-bit | FileCheck %s --check-prefixes=AVX512,AVX512VL4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512cd,+prefer-256-bit | FileCheck %s --check-prefixes=AVX512,AVX512F5; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512cd,-prefer-256-bit | FileCheck %s --check-prefixes=AVX512,AVX512F6 7define <8 x i16> @testv8i16(<8 x i16> %in) {8; AVX256-LABEL: testv8i16:9; AVX256: # %bb.0:10; AVX256-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero11; AVX256-NEXT: vplzcntd %ymm0, %ymm012; AVX256-NEXT: vpmovdw %ymm0, %xmm013; AVX256-NEXT: vpsubw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm014; AVX256-NEXT: vzeroupper15; AVX256-NEXT: retq16;17; AVX512VL-LABEL: testv8i16:18; AVX512VL: # %bb.0:19; AVX512VL-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero20; AVX512VL-NEXT: vplzcntd %ymm0, %ymm021; AVX512VL-NEXT: vpmovdw %ymm0, %xmm022; AVX512VL-NEXT: vpsubw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm023; AVX512VL-NEXT: vzeroupper24; AVX512VL-NEXT: retq25;26; AVX512F-LABEL: testv8i16:27; AVX512F: # %bb.0:28; AVX512F-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero29; AVX512F-NEXT: vplzcntd %zmm0, %zmm030; AVX512F-NEXT: vpmovdw %zmm0, %ymm031; AVX512F-NEXT: vpsubw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm032; AVX512F-NEXT: vzeroupper33; AVX512F-NEXT: retq34 %out = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %in, i1 false)35 ret <8 x i16> %out36}37 38define <16 x i8> @testv16i8(<16 x i8> %in) {39; AVX256-LABEL: testv16i8:40; AVX256: # %bb.0:41; AVX256-NEXT: vmovq {{.*#+}} xmm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]42; AVX256-NEXT: vpshufb %xmm0, %xmm1, %xmm243; AVX256-NEXT: vpsrlw $4, %xmm0, %xmm044; AVX256-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %xmm0, %xmm045; AVX256-NEXT: vpxor %xmm3, %xmm3, %xmm346; AVX256-NEXT: vpcmpeqb %xmm3, %xmm0, %xmm347; AVX256-NEXT: vpand %xmm3, %xmm2, %xmm248; AVX256-NEXT: vpshufb %xmm0, %xmm1, %xmm049; AVX256-NEXT: vpaddb %xmm0, %xmm2, %xmm050; AVX256-NEXT: retq51;52; AVX512-LABEL: testv16i8:53; AVX512: # %bb.0:54; AVX512-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero55; AVX512-NEXT: vplzcntd %zmm0, %zmm056; AVX512-NEXT: vpmovdb %zmm0, %xmm057; AVX512-NEXT: vpsubb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm058; AVX512-NEXT: vzeroupper59; AVX512-NEXT: retq60 %out = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %in, i1 false)61 ret <16 x i8> %out62}63 64define <16 x i16> @testv16i16(<16 x i16> %in) {65; AVX256-LABEL: testv16i16:66; AVX256: # %bb.0:67; AVX256-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero68; AVX256-NEXT: vplzcntd %ymm1, %ymm169; AVX256-NEXT: vpmovdw %ymm1, %xmm170; AVX256-NEXT: vextracti128 $1, %ymm0, %xmm071; AVX256-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero72; AVX256-NEXT: vplzcntd %ymm0, %ymm073; AVX256-NEXT: vpmovdw %ymm0, %xmm074; AVX256-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm075; AVX256-NEXT: vpsubw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm076; AVX256-NEXT: retq77;78; AVX512-LABEL: testv16i16:79; AVX512: # %bb.0:80; 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],zero81; AVX512-NEXT: vplzcntd %zmm0, %zmm082; AVX512-NEXT: vpmovdw %zmm0, %ymm083; AVX512-NEXT: vpsubw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm084; AVX512-NEXT: retq85 %out = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %in, i1 false)86 ret <16 x i16> %out87}88 89define <32 x i8> @testv32i8(<32 x i8> %in) {90; AVX256-LABEL: testv32i8:91; AVX256: # %bb.0:92; AVX256-NEXT: vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]93; AVX256-NEXT: # ymm1 = mem[0,1,0,1]94; AVX256-NEXT: vpshufb %ymm0, %ymm1, %ymm295; AVX256-NEXT: vpsrlw $4, %ymm0, %ymm096; AVX256-NEXT: vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %ymm097; AVX256-NEXT: vpxor %xmm3, %xmm3, %xmm398; AVX256-NEXT: vpcmpeqb %ymm3, %ymm0, %ymm399; AVX256-NEXT: vpand %ymm3, %ymm2, %ymm2100; AVX256-NEXT: vpshufb %ymm0, %ymm1, %ymm0101; AVX256-NEXT: vpaddb %ymm0, %ymm2, %ymm0102; AVX256-NEXT: retq103;104; AVX512-LABEL: testv32i8:105; AVX512: # %bb.0:106; AVX512-NEXT: vpmovzxbd {{.*#+}} zmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero107; AVX512-NEXT: vplzcntd %zmm1, %zmm1108; AVX512-NEXT: vpmovdb %zmm1, %xmm1109; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm0110; AVX512-NEXT: vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero111; AVX512-NEXT: vplzcntd %zmm0, %zmm0112; AVX512-NEXT: vpmovdb %zmm0, %xmm0113; AVX512-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0114; AVX512-NEXT: vpsubb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0115; AVX512-NEXT: retq116 %out = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %in, i1 false)117 ret <32 x i8> %out118}119 120declare <8 x i16> @llvm.ctlz.v8i16(<8 x i16>, i1)121declare <16 x i8> @llvm.ctlz.v16i8(<16 x i8>, i1)122declare <16 x i16> @llvm.ctlz.v16i16(<16 x i16>, i1)123declare <32 x i8> @llvm.ctlz.v32i8(<32 x i8>, i1)124