324 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py2; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+sse2 | FileCheck %s -check-prefixes=SSE,SSE23; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+ssse3 | FileCheck %s -check-prefixes=SSE,SSSE34; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+sse4.2 | FileCheck %s -check-prefixes=SSE,SSE425; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx | FileCheck %s -check-prefixes=AVX,AVX16; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx2 | FileCheck %s -check-prefixes=AVX,AVX27; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F8; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512,AVX512BW9; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+avx512f,+avx512bw,+avx512vbmi | FileCheck %s --check-prefixes=AVX512,AVX512VBMI10;11; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mcpu=slm | FileCheck %s --check-prefixes=SSE,SSE4212; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mcpu=goldmont | FileCheck %s --check-prefixes=SSE,SSE4213; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mcpu=btver2 | FileCheck %s --check-prefixes=AVX,AVX114 15;16; Verify the cost model for splice shuffles.17;18 19define void @test_vXf64(<2 x double> %a128, <2 x double> %b128, <4 x double> %a256, <4 x double> %b256, <8 x double> %a512, <8 x double> %b512) {20; SSE-LABEL: 'test_vXf64'21; SSE-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %a128, <2 x double> %b128, <2 x i32> <i32 1, i32 2>22; SSE-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x double> %a256, <4 x double> %b256, <4 x i32> <i32 1, i32 2, i32 3, i32 4>23; SSE-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x double> %a512, <8 x double> %b512, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>24; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void25;26; AVX-LABEL: 'test_vXf64'27; AVX-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %a128, <2 x double> %b128, <2 x i32> <i32 1, i32 2>28; AVX-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x double> %a256, <4 x double> %b256, <4 x i32> <i32 1, i32 2, i32 3, i32 4>29; AVX-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x double> %a512, <8 x double> %b512, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>30; AVX-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void31;32; AVX512-LABEL: 'test_vXf64'33; AVX512-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %a128, <2 x double> %b128, <2 x i32> <i32 1, i32 2>34; AVX512-NEXT: Cost Model: Found costs of 1 for: %V256 = shufflevector <4 x double> %a256, <4 x double> %b256, <4 x i32> <i32 1, i32 2, i32 3, i32 4>35; AVX512-NEXT: Cost Model: Found costs of 1 for: %V512 = shufflevector <8 x double> %a512, <8 x double> %b512, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>36; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void37;38 %V128 = shufflevector <2 x double> %a128, <2 x double> %b128, <2 x i32> <i32 1, i32 2>39 %V256 = shufflevector <4 x double> %a256, <4 x double> %b256, <4 x i32> <i32 1, i32 2, i32 3, i32 4>40 %V512 = shufflevector <8 x double> %a512, <8 x double> %b512, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>41 ret void42}43 44define void @test_vXi64(<2 x i64> %a128, <2 x i64> %b128, <4 x i64> %a256, <4 x i64> %b256, <8 x i64> %a512, <8 x i64> %b512) {45; SSE-LABEL: 'test_vXi64'46; SSE-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %a128, <2 x i64> %b128, <2 x i32> <i32 1, i32 2>47; SSE-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x i64> %a256, <4 x i64> %b256, <4 x i32> <i32 1, i32 2, i32 3, i32 4>48; SSE-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x i64> %a512, <8 x i64> %b512, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>49; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void50;51; AVX-LABEL: 'test_vXi64'52; AVX-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %a128, <2 x i64> %b128, <2 x i32> <i32 1, i32 2>53; AVX-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x i64> %a256, <4 x i64> %b256, <4 x i32> <i32 1, i32 2, i32 3, i32 4>54; AVX-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x i64> %a512, <8 x i64> %b512, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>55; AVX-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void56;57; AVX512-LABEL: 'test_vXi64'58; AVX512-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %a128, <2 x i64> %b128, <2 x i32> <i32 1, i32 2>59; AVX512-NEXT: Cost Model: Found costs of 1 for: %V256 = shufflevector <4 x i64> %a256, <4 x i64> %b256, <4 x i32> <i32 1, i32 2, i32 3, i32 4>60; AVX512-NEXT: Cost Model: Found costs of 1 for: %V512 = shufflevector <8 x i64> %a512, <8 x i64> %b512, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>61; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void62;63 %V128 = shufflevector <2 x i64> %a128, <2 x i64> %b128, <2 x i32> <i32 1, i32 2>64 %V256 = shufflevector <4 x i64> %a256, <4 x i64> %b256, <4 x i32> <i32 1, i32 2, i32 3, i32 4>65 %V512 = shufflevector <8 x i64> %a512, <8 x i64> %b512, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>66 ret void67}68 69define void @test_vXf32(<2 x float> %a64, <2 x float> %b64, <4 x float> %a128, <4 x float> %b128, <8 x float> %a256, <8 x float> %b256, <16 x float> %a512, <16 x float> %b512) {70; SSE2-LABEL: 'test_vXf32'71; SSE2-NEXT: Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x float> %a64, <2 x float> %b64, <2 x i32> <i32 1, i32 2>72; SSE2-NEXT: Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x float> %a128, <4 x float> %b128, <4 x i32> <i32 3, i32 4, i32 5, i32 6>73; SSE2-NEXT: Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x float> %a256, <8 x float> %b256, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>74; SSE2-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <16 x float> %a512, <16 x float> %b512, <16 x i32> <i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21>75; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void76;77; SSSE3-LABEL: 'test_vXf32'78; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %a64, <2 x float> %b64, <2 x i32> <i32 1, i32 2>79; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %a128, <4 x float> %b128, <4 x i32> <i32 3, i32 4, i32 5, i32 6>80; SSSE3-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x float> %a256, <8 x float> %b256, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>81; SSSE3-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <16 x float> %a512, <16 x float> %b512, <16 x i32> <i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21>82; SSSE3-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void83;84; SSE42-LABEL: 'test_vXf32'85; SSE42-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %a64, <2 x float> %b64, <2 x i32> <i32 1, i32 2>86; SSE42-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %a128, <4 x float> %b128, <4 x i32> <i32 3, i32 4, i32 5, i32 6>87; SSE42-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x float> %a256, <8 x float> %b256, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>88; SSE42-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <16 x float> %a512, <16 x float> %b512, <16 x i32> <i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21>89; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void90;91; AVX1-LABEL: 'test_vXf32'92; AVX1-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %a64, <2 x float> %b64, <2 x i32> <i32 1, i32 2>93; AVX1-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %a128, <4 x float> %b128, <4 x i32> <i32 3, i32 4, i32 5, i32 6>94; AVX1-NEXT: Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x float> %a256, <8 x float> %b256, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>95; AVX1-NEXT: Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x float> %a512, <16 x float> %b512, <16 x i32> <i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21>96; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void97;98; AVX2-LABEL: 'test_vXf32'99; AVX2-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %a64, <2 x float> %b64, <2 x i32> <i32 1, i32 2>100; AVX2-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %a128, <4 x float> %b128, <4 x i32> <i32 3, i32 4, i32 5, i32 6>101; AVX2-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x float> %a256, <8 x float> %b256, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>102; AVX2-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <16 x float> %a512, <16 x float> %b512, <16 x i32> <i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21>103; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void104;105; AVX512-LABEL: 'test_vXf32'106; AVX512-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %a64, <2 x float> %b64, <2 x i32> <i32 1, i32 2>107; AVX512-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %a128, <4 x float> %b128, <4 x i32> <i32 3, i32 4, i32 5, i32 6>108; AVX512-NEXT: Cost Model: Found costs of 1 for: %V256 = shufflevector <8 x float> %a256, <8 x float> %b256, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>109; AVX512-NEXT: Cost Model: Found costs of 1 for: %V512 = shufflevector <16 x float> %a512, <16 x float> %b512, <16 x i32> <i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21>110; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void111;112 %V64 = shufflevector <2 x float> %a64, <2 x float> %b64, <2 x i32> <i32 1, i32 2>113 %V128 = shufflevector <4 x float> %a128, <4 x float> %b128, <4 x i32> <i32 3, i32 4, i32 5, i32 6>114 %V256 = shufflevector <8 x float> %a256, <8 x float> %b256, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>115 %V512 = shufflevector <16 x float> %a512, <16 x float> %b512, <16 x i32> <i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21>116 ret void117}118 119define void @test_vXi32(<2 x i32> %a64, <2 x i32> %b64, <4 x i32> %a128, <4 x i32> %b128, <8 x i32> %a256, <8 x i32> %b256, <16 x i32> %a512, <16 x i32> %b512) {120; SSE2-LABEL: 'test_vXi32'121; SSE2-NEXT: Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x i32> %a64, <2 x i32> %b64, <2 x i32> <i32 1, i32 2>122; SSE2-NEXT: Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x i32> %a128, <4 x i32> %b128, <4 x i32> <i32 2, i32 3, i32 4, i32 5>123; SSE2-NEXT: Cost Model: Found costs of 0 for: %V256 = shufflevector <8 x i32> %a256, <8 x i32> %b256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>124; SSE2-NEXT: Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x i32> %a512, <16 x i32> %b512, <16 x i32> <i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20>125; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void126;127; SSSE3-LABEL: 'test_vXi32'128; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %a64, <2 x i32> %b64, <2 x i32> <i32 1, i32 2>129; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x i32> %a128, <4 x i32> %b128, <4 x i32> <i32 2, i32 3, i32 4, i32 5>130; SSSE3-NEXT: Cost Model: Found costs of 0 for: %V256 = shufflevector <8 x i32> %a256, <8 x i32> %b256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>131; SSSE3-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <16 x i32> %a512, <16 x i32> %b512, <16 x i32> <i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20>132; SSSE3-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void133;134; SSE42-LABEL: 'test_vXi32'135; SSE42-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %a64, <2 x i32> %b64, <2 x i32> <i32 1, i32 2>136; SSE42-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x i32> %a128, <4 x i32> %b128, <4 x i32> <i32 2, i32 3, i32 4, i32 5>137; SSE42-NEXT: Cost Model: Found costs of 0 for: %V256 = shufflevector <8 x i32> %a256, <8 x i32> %b256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>138; SSE42-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <16 x i32> %a512, <16 x i32> %b512, <16 x i32> <i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20>139; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void140;141; AVX1-LABEL: 'test_vXi32'142; AVX1-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %a64, <2 x i32> %b64, <2 x i32> <i32 1, i32 2>143; AVX1-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x i32> %a128, <4 x i32> %b128, <4 x i32> <i32 2, i32 3, i32 4, i32 5>144; AVX1-NEXT: Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x i32> %a256, <8 x i32> %b256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>145; AVX1-NEXT: Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x i32> %a512, <16 x i32> %b512, <16 x i32> <i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20>146; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void147;148; AVX2-LABEL: 'test_vXi32'149; AVX2-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %a64, <2 x i32> %b64, <2 x i32> <i32 1, i32 2>150; AVX2-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x i32> %a128, <4 x i32> %b128, <4 x i32> <i32 2, i32 3, i32 4, i32 5>151; AVX2-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x i32> %a256, <8 x i32> %b256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>152; AVX2-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <16 x i32> %a512, <16 x i32> %b512, <16 x i32> <i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20>153; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void154;155; AVX512-LABEL: 'test_vXi32'156; AVX512-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %a64, <2 x i32> %b64, <2 x i32> <i32 1, i32 2>157; AVX512-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x i32> %a128, <4 x i32> %b128, <4 x i32> <i32 2, i32 3, i32 4, i32 5>158; AVX512-NEXT: Cost Model: Found costs of 1 for: %V256 = shufflevector <8 x i32> %a256, <8 x i32> %b256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>159; AVX512-NEXT: Cost Model: Found costs of 1 for: %V512 = shufflevector <16 x i32> %a512, <16 x i32> %b512, <16 x i32> <i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20>160; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void161;162 %V64 = shufflevector <2 x i32> %a64, <2 x i32> %b64, <2 x i32> <i32 1, i32 2>163 %V128 = shufflevector <4 x i32> %a128, <4 x i32> %b128, <4 x i32> <i32 2, i32 3, i32 4, i32 5>164 %V256 = shufflevector <8 x i32> %a256, <8 x i32> %b256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>165 %V512 = shufflevector <16 x i32> %a512, <16 x i32> %b512, <16 x i32> <i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20>166 ret void167}168 169define void @test_vXi16(<2 x i16> %a32, <2 x i16> %b32, <4 x i16> %a64, <4 x i16> %b64, <8 x i16> %a128, <8 x i16> %b128, <16 x i16> %a256, <16 x i16> %b256, <32 x i16> %a512, <32 x i16> %b512) {170; SSE2-LABEL: 'test_vXi16'171; SSE2-NEXT: Cost Model: Found costs of 2 for: %V32 = shufflevector <2 x i16> %a32, <2 x i16> %b32, <2 x i32> <i32 1, i32 2>172; SSE2-NEXT: Cost Model: Found costs of 2 for: %V64 = shufflevector <4 x i16> %a64, <4 x i16> %b64, <4 x i32> <i32 2, i32 3, i32 4, i32 5>173; SSE2-NEXT: Cost Model: Found costs of 3 for: %V128 = shufflevector <8 x i16> %a128, <8 x i16> %b128, <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>174; SSE2-NEXT: Cost Model: Found costs of 6 for: %V256 = shufflevector <16 x i16> %a256, <16 x i16> %b256, <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>175; SSE2-NEXT: Cost Model: Found costs of 12 for: %V512 = shufflevector <32 x i16> %a512, <32 x i16> %b512, <32 x i32> <i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38>176; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void177;178; SSSE3-LABEL: 'test_vXi16'179; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %a32, <2 x i16> %b32, <2 x i32> <i32 1, i32 2>180; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %a64, <4 x i16> %b64, <4 x i32> <i32 2, i32 3, i32 4, i32 5>181; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %a128, <8 x i16> %b128, <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>182; SSSE3-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %a256, <16 x i16> %b256, <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>183; SSSE3-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <32 x i16> %a512, <32 x i16> %b512, <32 x i32> <i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38>184; SSSE3-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void185;186; SSE42-LABEL: 'test_vXi16'187; SSE42-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %a32, <2 x i16> %b32, <2 x i32> <i32 1, i32 2>188; SSE42-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %a64, <4 x i16> %b64, <4 x i32> <i32 2, i32 3, i32 4, i32 5>189; SSE42-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %a128, <8 x i16> %b128, <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>190; SSE42-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %a256, <16 x i16> %b256, <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>191; SSE42-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <32 x i16> %a512, <32 x i16> %b512, <32 x i32> <i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38>192; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void193;194; AVX1-LABEL: 'test_vXi16'195; AVX1-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %a32, <2 x i16> %b32, <2 x i32> <i32 1, i32 2>196; AVX1-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %a64, <4 x i16> %b64, <4 x i32> <i32 2, i32 3, i32 4, i32 5>197; AVX1-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %a128, <8 x i16> %b128, <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>198; AVX1-NEXT: Cost Model: Found costs of 5 for: %V256 = shufflevector <16 x i16> %a256, <16 x i16> %b256, <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>199; AVX1-NEXT: Cost Model: Found costs of 10 for: %V512 = shufflevector <32 x i16> %a512, <32 x i16> %b512, <32 x i32> <i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38>200; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void201;202; AVX2-LABEL: 'test_vXi16'203; AVX2-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %a32, <2 x i16> %b32, <2 x i32> <i32 1, i32 2>204; AVX2-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %a64, <4 x i16> %b64, <4 x i32> <i32 2, i32 3, i32 4, i32 5>205; AVX2-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %a128, <8 x i16> %b128, <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>206; AVX2-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %a256, <16 x i16> %b256, <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>207; AVX2-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <32 x i16> %a512, <32 x i16> %b512, <32 x i32> <i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38>208; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void209;210; AVX512F-LABEL: 'test_vXi16'211; AVX512F-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %a32, <2 x i16> %b32, <2 x i32> <i32 1, i32 2>212; AVX512F-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %a64, <4 x i16> %b64, <4 x i32> <i32 2, i32 3, i32 4, i32 5>213; AVX512F-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %a128, <8 x i16> %b128, <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>214; AVX512F-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %a256, <16 x i16> %b256, <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>215; AVX512F-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <32 x i16> %a512, <32 x i16> %b512, <32 x i32> <i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38>216; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void217;218; AVX512BW-LABEL: 'test_vXi16'219; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %a32, <2 x i16> %b32, <2 x i32> <i32 1, i32 2>220; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %a64, <4 x i16> %b64, <4 x i32> <i32 2, i32 3, i32 4, i32 5>221; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %a128, <8 x i16> %b128, <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>222; AVX512BW-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %a256, <16 x i16> %b256, <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>223; AVX512BW-NEXT: Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x i16> %a512, <32 x i16> %b512, <32 x i32> <i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38>224; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void225;226; AVX512VBMI-LABEL: 'test_vXi16'227; AVX512VBMI-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %a32, <2 x i16> %b32, <2 x i32> <i32 1, i32 2>228; AVX512VBMI-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %a64, <4 x i16> %b64, <4 x i32> <i32 2, i32 3, i32 4, i32 5>229; AVX512VBMI-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %a128, <8 x i16> %b128, <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>230; AVX512VBMI-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %a256, <16 x i16> %b256, <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>231; AVX512VBMI-NEXT: Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x i16> %a512, <32 x i16> %b512, <32 x i32> <i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38>232; AVX512VBMI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void233;234 %V32 = shufflevector <2 x i16> %a32, <2 x i16> %b32, <2 x i32> <i32 1, i32 2>235 %V64 = shufflevector <4 x i16> %a64, <4 x i16> %b64, <4 x i32> <i32 2, i32 3, i32 4, i32 5>236 %V128 = shufflevector <8 x i16> %a128, <8 x i16> %b128, <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>237 %V256 = shufflevector <16 x i16> %a256, <16 x i16> %b256, <16 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30>238 %V512 = shufflevector <32 x i16> %a512, <32 x i16> %b512, <32 x i32> <i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38>239 ret void240}241 242define void @test_vXi8(<2 x i8> %a16, <2 x i8> %b16, <4 x i8> %a32, <4 x i8> %b32, <8 x i8> %a64, <8 x i8> %b64, <16 x i8> %a128, <16 x i8> %b128, <32 x i8> %a256, <32 x i8> %b256, <64 x i8> %a512, <64 x i8> %b512) {243; SSE2-LABEL: 'test_vXi8'244; SSE2-NEXT: Cost Model: Found costs of 2 for: %V16 = shufflevector <2 x i8> %a16, <2 x i8> %b16, <2 x i32> <i32 1, i32 2>245; SSE2-NEXT: Cost Model: Found costs of 2 for: %V32 = shufflevector <4 x i8> %a32, <4 x i8> %b32, <4 x i32> <i32 1, i32 2, i32 3, i32 4>246; SSE2-NEXT: Cost Model: Found costs of 3 for: %V64 = shufflevector <8 x i8> %a64, <8 x i8> %b64, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>247; SSE2-NEXT: Cost Model: Found costs of 3 for: %V128 = shufflevector <16 x i8> %a128, <16 x i8> %b128, <16 x i32> <i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29>248; SSE2-NEXT: Cost Model: Found costs of 6 for: %V256 = shufflevector <32 x i8> %a256, <32 x i8> %b256, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39>249; SSE2-NEXT: Cost Model: Found costs of 12 for: %V512 = shufflevector <64 x i8> %a512, <64 x i8> %b512, <64 x i32> <i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 72, i32 73, i32 74, i32 75, i32 76>250; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void251;252; SSSE3-LABEL: 'test_vXi8'253; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V16 = shufflevector <2 x i8> %a16, <2 x i8> %b16, <2 x i32> <i32 1, i32 2>254; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %a32, <4 x i8> %b32, <4 x i32> <i32 1, i32 2, i32 3, i32 4>255; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %a64, <8 x i8> %b64, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>256; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %a128, <16 x i8> %b128, <16 x i32> <i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29>257; SSSE3-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <32 x i8> %a256, <32 x i8> %b256, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39>258; SSSE3-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <64 x i8> %a512, <64 x i8> %b512, <64 x i32> <i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 72, i32 73, i32 74, i32 75, i32 76>259; SSSE3-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void260;261; SSE42-LABEL: 'test_vXi8'262; SSE42-NEXT: Cost Model: Found costs of 1 for: %V16 = shufflevector <2 x i8> %a16, <2 x i8> %b16, <2 x i32> <i32 1, i32 2>263; SSE42-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %a32, <4 x i8> %b32, <4 x i32> <i32 1, i32 2, i32 3, i32 4>264; SSE42-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %a64, <8 x i8> %b64, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>265; SSE42-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %a128, <16 x i8> %b128, <16 x i32> <i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29>266; SSE42-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <32 x i8> %a256, <32 x i8> %b256, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39>267; SSE42-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <64 x i8> %a512, <64 x i8> %b512, <64 x i32> <i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 72, i32 73, i32 74, i32 75, i32 76>268; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void269;270; AVX1-LABEL: 'test_vXi8'271; AVX1-NEXT: Cost Model: Found costs of 1 for: %V16 = shufflevector <2 x i8> %a16, <2 x i8> %b16, <2 x i32> <i32 1, i32 2>272; AVX1-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %a32, <4 x i8> %b32, <4 x i32> <i32 1, i32 2, i32 3, i32 4>273; AVX1-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %a64, <8 x i8> %b64, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>274; AVX1-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %a128, <16 x i8> %b128, <16 x i32> <i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29>275; AVX1-NEXT: Cost Model: Found costs of 5 for: %V256 = shufflevector <32 x i8> %a256, <32 x i8> %b256, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39>276; AVX1-NEXT: Cost Model: Found costs of 10 for: %V512 = shufflevector <64 x i8> %a512, <64 x i8> %b512, <64 x i32> <i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 72, i32 73, i32 74, i32 75, i32 76>277; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void278;279; AVX2-LABEL: 'test_vXi8'280; AVX2-NEXT: Cost Model: Found costs of 1 for: %V16 = shufflevector <2 x i8> %a16, <2 x i8> %b16, <2 x i32> <i32 1, i32 2>281; AVX2-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %a32, <4 x i8> %b32, <4 x i32> <i32 1, i32 2, i32 3, i32 4>282; AVX2-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %a64, <8 x i8> %b64, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>283; AVX2-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %a128, <16 x i8> %b128, <16 x i32> <i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29>284; AVX2-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <32 x i8> %a256, <32 x i8> %b256, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39>285; AVX2-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <64 x i8> %a512, <64 x i8> %b512, <64 x i32> <i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 72, i32 73, i32 74, i32 75, i32 76>286; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void287;288; AVX512F-LABEL: 'test_vXi8'289; AVX512F-NEXT: Cost Model: Found costs of 1 for: %V16 = shufflevector <2 x i8> %a16, <2 x i8> %b16, <2 x i32> <i32 1, i32 2>290; AVX512F-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %a32, <4 x i8> %b32, <4 x i32> <i32 1, i32 2, i32 3, i32 4>291; AVX512F-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %a64, <8 x i8> %b64, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>292; AVX512F-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %a128, <16 x i8> %b128, <16 x i32> <i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29>293; AVX512F-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <32 x i8> %a256, <32 x i8> %b256, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39>294; AVX512F-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <64 x i8> %a512, <64 x i8> %b512, <64 x i32> <i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 72, i32 73, i32 74, i32 75, i32 76>295; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void296;297; AVX512BW-LABEL: 'test_vXi8'298; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %V16 = shufflevector <2 x i8> %a16, <2 x i8> %b16, <2 x i32> <i32 1, i32 2>299; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %a32, <4 x i8> %b32, <4 x i32> <i32 1, i32 2, i32 3, i32 4>300; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %a64, <8 x i8> %b64, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>301; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %a128, <16 x i8> %b128, <16 x i32> <i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29>302; AVX512BW-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <32 x i8> %a256, <32 x i8> %b256, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39>303; AVX512BW-NEXT: Cost Model: Found costs of 2 for: %V512 = shufflevector <64 x i8> %a512, <64 x i8> %b512, <64 x i32> <i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 72, i32 73, i32 74, i32 75, i32 76>304; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void305;306; AVX512VBMI-LABEL: 'test_vXi8'307; AVX512VBMI-NEXT: Cost Model: Found costs of 1 for: %V16 = shufflevector <2 x i8> %a16, <2 x i8> %b16, <2 x i32> <i32 1, i32 2>308; AVX512VBMI-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %a32, <4 x i8> %b32, <4 x i32> <i32 1, i32 2, i32 3, i32 4>309; AVX512VBMI-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %a64, <8 x i8> %b64, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>310; AVX512VBMI-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %a128, <16 x i8> %b128, <16 x i32> <i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29>311; AVX512VBMI-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <32 x i8> %a256, <32 x i8> %b256, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39>312; AVX512VBMI-NEXT: Cost Model: Found costs of 2 for: %V512 = shufflevector <64 x i8> %a512, <64 x i8> %b512, <64 x i32> <i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 72, i32 73, i32 74, i32 75, i32 76>313; AVX512VBMI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void314;315 %V16 = shufflevector <2 x i8> %a16, <2 x i8> %b16, <2 x i32> <i32 1, i32 2>316 %V32 = shufflevector <4 x i8> %a32, <4 x i8> %b32, <4 x i32> <i32 1, i32 2, i32 3, i32 4>317 %V64 = shufflevector <8 x i8> %a64, <8 x i8> %b64, <8 x i32> <i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10>318 %V128 = shufflevector <16 x i8> %a128, <16 x i8> %b128, <16 x i32> <i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29>319 %V256 = shufflevector <32 x i8> %a256, <32 x i8> %b256, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39>320 %V512 = shufflevector <64 x i8> %a512, <64 x i8> %b512, <64 x i32> <i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63, i32 64, i32 65, i32 66, i32 67, i32 68, i32 69, i32 70, i32 71, i32 72, i32 73, i32 74, i32 75, i32 76>321 ret void322}323 324