132 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64 | FileCheck %s --check-prefixes=SSE,SSE23; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v2 | FileCheck %s --check-prefixes=SSE,SSE414; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=AVX25; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=AVX5126 7define <8 x i16> @pow2_mask_v16i8(i8 zeroext %0) {8; SSE2-LABEL: pow2_mask_v16i8:9; SSE2: # %bb.0:10; SSE2-NEXT: movd %edi, %xmm011; SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]12; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]13; SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]14; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [128,128,64,64,32,32,16,16,8,8,4,4,2,2,1,1]15; SSE2-NEXT: pand %xmm1, %xmm016; SSE2-NEXT: pcmpeqb %xmm1, %xmm017; SSE2-NEXT: retq18;19; SSE41-LABEL: pow2_mask_v16i8:20; SSE41: # %bb.0:21; SSE41-NEXT: movd %edi, %xmm022; SSE41-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]23; SSE41-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]24; SSE41-NEXT: movq {{.*#+}} xmm1 = [128,64,32,16,8,4,2,1,0,0,0,0,0,0,0,0]25; SSE41-NEXT: pand %xmm1, %xmm026; SSE41-NEXT: pcmpeqb %xmm1, %xmm027; SSE41-NEXT: pmovsxbw %xmm0, %xmm028; SSE41-NEXT: retq29;30; AVX2-LABEL: pow2_mask_v16i8:31; AVX2: # %bb.0:32; AVX2-NEXT: vmovd %edi, %xmm033; AVX2-NEXT: vpbroadcastb %xmm0, %xmm034; AVX2-NEXT: vpbroadcastq {{.*#+}} xmm1 = [128,64,32,16,8,4,2,1,128,64,32,16,8,4,2,1]35; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm036; AVX2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm037; AVX2-NEXT: vpmovsxbw %xmm0, %xmm038; AVX2-NEXT: retq39;40; AVX512-LABEL: pow2_mask_v16i8:41; AVX512: # %bb.0:42; AVX512-NEXT: vpbroadcastb %edi, %xmm043; AVX512-NEXT: vptestmb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %k044; AVX512-NEXT: vpmovm2w %k0, %xmm045; AVX512-NEXT: retq46 %vec = insertelement <1 x i8> poison, i8 %0, i64 047 %splat = shufflevector <1 x i8> %vec, <1 x i8> poison, <8 x i32> zeroinitializer48 %mask = and <8 x i8> %splat, <i8 -128, i8 64, i8 32, i8 16, i8 8, i8 4, i8 2, i8 1>49 %not = icmp ne <8 x i8> %mask, zeroinitializer50 %ext = sext <8 x i1> %not to <8 x i16>51 ret <8 x i16> %ext52}53 54define <16 x i16> @pow2_mask_v16i16(i16 zeroext %0) {55; SSE-LABEL: pow2_mask_v16i16:56; SSE: # %bb.0:57; SSE-NEXT: movd %edi, %xmm058; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]59; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]60; SSE-NEXT: movdqa {{.*#+}} xmm2 = [128,64,32,16,8,4,2,1]61; SSE-NEXT: movdqa %xmm0, %xmm162; SSE-NEXT: pand %xmm2, %xmm163; SSE-NEXT: movdqa {{.*#+}} xmm3 = [32768,16384,8192,4096,2048,1024,512,256]64; SSE-NEXT: pand %xmm3, %xmm065; SSE-NEXT: pcmpeqw %xmm3, %xmm066; SSE-NEXT: pcmpeqw %xmm2, %xmm167; SSE-NEXT: retq68;69; AVX2-LABEL: pow2_mask_v16i16:70; AVX2: # %bb.0:71; AVX2-NEXT: vmovd %edi, %xmm072; AVX2-NEXT: vpbroadcastw %xmm0, %ymm073; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2,1]74; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm075; AVX2-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm076; AVX2-NEXT: retq77;78; AVX512-LABEL: pow2_mask_v16i16:79; AVX512: # %bb.0:80; AVX512-NEXT: vpbroadcastw %edi, %ymm081; AVX512-NEXT: vmovdqa {{.*#+}} ymm1 = [32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2,1]82; AVX512-NEXT: vpand %ymm1, %ymm0, %ymm083; AVX512-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm084; AVX512-NEXT: retq85 %vec = insertelement <1 x i16> poison, i16 %0, i64 086 %splat = shufflevector <1 x i16> %vec, <1 x i16> poison, <16 x i32> zeroinitializer87 %mask = and <16 x i16> %splat, <i16 -32768, i16 16384, i16 8192, i16 4096, i16 2048, i16 1024, i16 512, i16 256, i16 128, i16 64, i16 32, i16 16, i16 8, i16 4, i16 2, i16 1>88 %not = icmp ne <16 x i16> %mask, zeroinitializer89 %ext = sext <16 x i1> %not to <16 x i16>90 ret <16 x i16> %ext91}92 93; PR7888894define i64 @pow2_mask_v8i8(i8 zeroext %0) {95; SSE-LABEL: pow2_mask_v8i8:96; SSE: # %bb.0:97; SSE-NEXT: movd %edi, %xmm098; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]99; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]100; SSE-NEXT: movq {{.*#+}} xmm1 = [128,64,32,16,8,4,2,1,0,0,0,0,0,0,0,0]101; SSE-NEXT: pand %xmm1, %xmm0102; SSE-NEXT: pcmpeqb %xmm1, %xmm0103; SSE-NEXT: movq %xmm0, %rax104; SSE-NEXT: retq105;106; AVX2-LABEL: pow2_mask_v8i8:107; AVX2: # %bb.0:108; AVX2-NEXT: vmovd %edi, %xmm0109; AVX2-NEXT: vpbroadcastb %xmm0, %xmm0110; AVX2-NEXT: vpbroadcastq {{.*#+}} xmm1 = [128,64,32,16,8,4,2,1,128,64,32,16,8,4,2,1]111; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm0112; AVX2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0113; AVX2-NEXT: vmovq %xmm0, %rax114; AVX2-NEXT: retq115;116; AVX512-LABEL: pow2_mask_v8i8:117; AVX512: # %bb.0:118; AVX512-NEXT: vpbroadcastb %edi, %xmm0119; AVX512-NEXT: vpbroadcastq {{.*#+}} xmm1 = [128,64,32,16,8,4,2,1,128,64,32,16,8,4,2,1]120; AVX512-NEXT: vpand %xmm1, %xmm0, %xmm0121; AVX512-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0122; AVX512-NEXT: vmovq %xmm0, %rax123; AVX512-NEXT: retq124 %vec = insertelement <1 x i8> poison, i8 %0, i64 0125 %splat = shufflevector <1 x i8> %vec, <1 x i8> poison, <8 x i32> zeroinitializer126 %mask = and <8 x i8> %splat, <i8 -128, i8 64, i8 32, i8 16, i8 8, i8 4, i8 2, i8 1>127 %not = icmp ne <8 x i8> %mask, zeroinitializer128 %ext = sext <8 x i1> %not to <8 x i8>129 %res = bitcast <8 x i8> %ext to i64130 ret i64 %res131}132