404 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=CHECK,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=CHECK,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=CHECK,SSE,SSE425; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=all -mattr=+xop | FileCheck %s -check-prefixes=CHECK,XOP6; 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=CHECK,AVX17; 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=CHECK,AVX28; 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=CHECK,AVX512,AVX512F9; 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=CHECK,AVX512,AVX512BW10; 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=CHECK,AVX512,AVX512VBMI11;12; 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=CHECK,SSE,SSE4213; 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=CHECK,SSE,SSE4214; 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=CHECK,AVX115 16;17; Verify the cost model for 1 src shuffles18;19 20define void @test_vXf64(<2 x double> %src128, <4 x double> %src256, <8 x double> %src512, <16 x double> %src1024) {21; SSE-LABEL: 'test_vXf64'22; SSE-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> <i32 1, i32 1>23; SSE-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>24; SSE-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>25; SSE-NEXT: Cost Model: Found costs of 8 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 0, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 13, i32 3, i32 11, i32 1, i32 0>26; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void27;28; XOP-LABEL: 'test_vXf64'29; XOP-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> <i32 1, i32 1>30; XOP-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>31; XOP-NEXT: Cost Model: Found costs of 5 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>32; XOP-NEXT: Cost Model: Found costs of 11 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 0, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 13, i32 3, i32 11, i32 1, i32 0>33; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void34;35; AVX1-LABEL: 'test_vXf64'36; AVX1-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> <i32 1, i32 1>37; AVX1-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>38; AVX1-NEXT: Cost Model: Found costs of 5 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>39; AVX1-NEXT: Cost Model: Found costs of 11 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 0, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 13, i32 3, i32 11, i32 1, i32 0>40; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void41;42; AVX2-LABEL: 'test_vXf64'43; AVX2-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> <i32 1, i32 1>44; AVX2-NEXT: Cost Model: Found costs of 1 for: %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>45; AVX2-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>46; AVX2-NEXT: Cost Model: Found costs of 10 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 0, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 13, i32 3, i32 11, i32 1, i32 0>47; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void48;49; AVX512-LABEL: 'test_vXf64'50; AVX512-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> <i32 1, i32 1>51; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>52; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>53; AVX512-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:6 SizeLat:2 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 0, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 13, i32 3, i32 11, i32 1, i32 0>54; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void55;56 %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> <i32 1, i32 1>57 %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>58 %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>59 %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 0, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 13, i32 3, i32 11, i32 1, i32 0>60 ret void61}62 63define void @test_vXi64(<2 x i64> %src128, <4 x i64> %src256, <8 x i64> %src512) {64; SSE-LABEL: 'test_vXi64'65; SSE-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> <i32 1, i32 1>66; SSE-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>67; SSE-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>68; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void69;70; XOP-LABEL: 'test_vXi64'71; XOP-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> <i32 1, i32 1>72; XOP-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>73; XOP-NEXT: Cost Model: Found costs of 5 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>74; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void75;76; AVX1-LABEL: 'test_vXi64'77; AVX1-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> <i32 1, i32 1>78; AVX1-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>79; AVX1-NEXT: Cost Model: Found costs of 5 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>80; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void81;82; AVX2-LABEL: 'test_vXi64'83; AVX2-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> <i32 1, i32 1>84; AVX2-NEXT: Cost Model: Found costs of 1 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>85; AVX2-NEXT: Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>86; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void87;88; AVX512-LABEL: 'test_vXi64'89; AVX512-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> <i32 1, i32 1>90; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>91; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>92; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void93;94 %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> <i32 1, i32 1>95 %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>96 %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>97 ret void98}99 100define void @test_vXf32(<2 x float> %src64, <4 x float> %src128, <8 x float> %src256, <16 x float> %src512) {101; SSE2-LABEL: 'test_vXf32'102; SSE2-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>103; SSE2-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>104; SSE2-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>105; SSE2-NEXT: Cost Model: Found costs of 11 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>106; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void107;108; SSSE3-LABEL: 'test_vXf32'109; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>110; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>111; SSSE3-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>112; SSSE3-NEXT: Cost Model: Found costs of 11 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>113; SSSE3-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void114;115; SSE42-LABEL: 'test_vXf32'116; SSE42-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>117; SSE42-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>118; SSE42-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>119; SSE42-NEXT: Cost Model: Found costs of 10 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>120; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void121;122; XOP-LABEL: 'test_vXf32'123; XOP-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>124; XOP-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>125; XOP-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>126; XOP-NEXT: Cost Model: Found costs of 12 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>127; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void128;129; AVX1-LABEL: 'test_vXf32'130; AVX1-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>131; AVX1-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>132; AVX1-NEXT: Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>133; AVX1-NEXT: Cost Model: Found costs of 12 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>134; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void135;136; AVX2-LABEL: 'test_vXf32'137; AVX2-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>138; AVX2-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>139; AVX2-NEXT: Cost Model: Found costs of 1 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>140; AVX2-NEXT: Cost Model: Found costs of 9 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>141; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void142;143; AVX512-LABEL: 'test_vXf32'144; AVX512-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>145; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>146; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>147; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>148; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void149;150 %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>151 %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>152 %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>153 %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>154 ret void155}156 157define void @test_vXi32(<2 x i32> %src64, <4 x i32> %src128, <8 x i32> %src256, <16 x i32> %src512, <32 x i32> %src1024) {158; SSE-LABEL: 'test_vXi32'159; SSE-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> <i32 1, i32 1>160; SSE-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>161; SSE-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 0, i32 1, i32 0>162; SSE-NEXT: Cost Model: Found costs of 11 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>163; SSE-NEXT: Cost Model: Found costs of 18 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>164; SSE-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void165;166; XOP-LABEL: 'test_vXi32'167; XOP-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> <i32 1, i32 1>168; XOP-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>169; XOP-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 0, i32 1, i32 0>170; XOP-NEXT: Cost Model: Found costs of 12 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>171; XOP-NEXT: Cost Model: Found costs of 26 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>172; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void173;174; AVX1-LABEL: 'test_vXi32'175; AVX1-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> <i32 1, i32 1>176; AVX1-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>177; AVX1-NEXT: Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 0, i32 1, i32 0>178; AVX1-NEXT: Cost Model: Found costs of 12 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>179; AVX1-NEXT: Cost Model: Found costs of 26 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>180; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void181;182; AVX2-LABEL: 'test_vXi32'183; AVX2-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> <i32 1, i32 1>184; AVX2-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>185; AVX2-NEXT: Cost Model: Found costs of 1 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 0, i32 1, i32 0>186; AVX2-NEXT: Cost Model: Found costs of 9 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>187; AVX2-NEXT: Cost Model: Found costs of 20 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>188; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void189;190; AVX512-LABEL: 'test_vXi32'191; AVX512-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> <i32 1, i32 1>192; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>193; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 0, i32 1, i32 0>194; AVX512-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>195; AVX512-NEXT: Cost Model: Found costs of RThru:2 CodeSize:2 Lat:6 SizeLat:2 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>196; AVX512-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void197;198 %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> <i32 1, i32 1>199 %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>200 %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 0, i32 1, i32 0>201 %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>202 %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>203 ret void204}205 206define void @test_vXi16(<2 x i16> %src32, <4 x i16> %src64, <8 x i16> %src128, <16 x i16> %src256, <32 x i16> %src512, <64 x i16> %src1024) {207; SSE2-LABEL: 'test_vXi16'208; SSE2-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>209; SSE2-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>210; SSE2-NEXT: Cost Model: Found costs of RThru:3 CodeSize:5 Lat:5 SizeLat:5 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>211; SSE2-NEXT: Cost Model: Found costs of RThru:18 CodeSize:24 Lat:24 SizeLat:24 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>212; SSE2-NEXT: Cost Model: Found costs of RThru:42 CodeSize:56 Lat:56 SizeLat:56 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>213; SSE2-NEXT: Cost Model: Found costs of RThru:72 CodeSize:94 Lat:94 SizeLat:94 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>214; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void215;216; SSSE3-LABEL: 'test_vXi16'217; SSSE3-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>218; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>219; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>220; SSSE3-NEXT: Cost Model: Found costs of 9 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>221; SSSE3-NEXT: Cost Model: Found costs of 21 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>222; SSSE3-NEXT: Cost Model: Found costs of 39 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>223; SSSE3-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void224;225; SSE42-LABEL: 'test_vXi16'226; SSE42-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>227; SSE42-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>228; SSE42-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>229; SSE42-NEXT: Cost Model: Found costs of 9 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>230; SSE42-NEXT: Cost Model: Found costs of 21 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>231; SSE42-NEXT: Cost Model: Found costs of 35 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>232; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void233;234; XOP-LABEL: 'test_vXi16'235; XOP-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>236; XOP-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>237; XOP-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>238; XOP-NEXT: Cost Model: Found costs of 9 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>239; XOP-NEXT: Cost Model: Found costs of 18 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>240; XOP-NEXT: Cost Model: Found costs of 63 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>241; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void242;243; AVX1-LABEL: 'test_vXi16'244; AVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>245; AVX1-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>246; AVX1-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>247; AVX1-NEXT: Cost Model: Found costs of 15 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>248; AVX1-NEXT: Cost Model: Found costs of 30 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>249; AVX1-NEXT: Cost Model: Found costs of 99 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>250; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void251;252; AVX2-LABEL: 'test_vXi16'253; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>254; AVX2-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>255; AVX2-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>256; AVX2-NEXT: Cost Model: Found costs of 7 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>257; AVX2-NEXT: Cost Model: Found costs of 14 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>258; AVX2-NEXT: Cost Model: Found costs of 44 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>259; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void260;261; AVX512F-LABEL: 'test_vXi16'262; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>263; AVX512F-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>264; AVX512F-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>265; AVX512F-NEXT: Cost Model: Found costs of 7 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>266; AVX512F-NEXT: Cost Model: Found costs of 14 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>267; AVX512F-NEXT: Cost Model: Found costs of 84 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>268; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void269;270; AVX512BW-LABEL: 'test_vXi16'271; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>272; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>273; AVX512BW-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>274; AVX512BW-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>275; AVX512BW-NEXT: Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>276; AVX512BW-NEXT: Cost Model: Found costs of 4 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>277; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void278;279; AVX512VBMI-LABEL: 'test_vXi16'280; AVX512VBMI-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>281; AVX512VBMI-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>282; AVX512VBMI-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>283; AVX512VBMI-NEXT: Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>284; AVX512VBMI-NEXT: Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>285; AVX512VBMI-NEXT: Cost Model: Found costs of 4 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>286; AVX512VBMI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void287;288 %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>289 %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>290 %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>291 %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>292 %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>293 %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>294 ret void295}296 297define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x i8> %src128, <32 x i8> %src256, <64 x i8> %src512) {298; SSE2-LABEL: 'test_vXi8'299; SSE2-NEXT: Cost Model: Found costs of 1 for: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> <i32 1, i32 1>300; SSE2-NEXT: Cost Model: Found costs of 3 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>301; SSE2-NEXT: Cost Model: Found costs of 5 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>302; SSE2-NEXT: Cost Model: Found costs of RThru:8 CodeSize:10 Lat:10 SizeLat:10 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>303; SSE2-NEXT: Cost Model: Found costs of RThru:22 CodeSize:26 Lat:26 SizeLat:26 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>304; SSE2-NEXT: Cost Model: Found costs of RThru:77 CodeSize:91 Lat:91 SizeLat:91 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>305; SSE2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void306;307; SSSE3-LABEL: 'test_vXi8'308; SSSE3-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> <i32 1, i32 1>309; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>310; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>311; SSSE3-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>312; SSSE3-NEXT: Cost Model: Found costs of 6 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>313; SSSE3-NEXT: Cost Model: Found costs of 21 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>314; SSSE3-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void315;316; SSE42-LABEL: 'test_vXi8'317; SSE42-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> <i32 1, i32 1>318; SSE42-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>319; SSE42-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>320; SSE42-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>321; SSE42-NEXT: Cost Model: Found costs of 6 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>322; SSE42-NEXT: Cost Model: Found costs of 21 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>323; SSE42-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void324;325; XOP-LABEL: 'test_vXi8'326; XOP-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> <i32 1, i32 1>327; XOP-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>328; XOP-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>329; XOP-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>330; XOP-NEXT: Cost Model: Found costs of 4 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>331; XOP-NEXT: Cost Model: Found costs of 18 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>332; XOP-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void333;334; AVX1-LABEL: 'test_vXi8'335; AVX1-NEXT: Cost Model: Found costs of RThru:1 CodeSize:2 Lat:3 SizeLat:2 for: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> <i32 1, i32 1>336; AVX1-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>337; AVX1-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>338; AVX1-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>339; AVX1-NEXT: Cost Model: Found costs of 8 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>340; AVX1-NEXT: Cost Model: Found costs of 30 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>341; AVX1-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void342;343; AVX2-LABEL: 'test_vXi8'344; AVX2-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> <i32 1, i32 1>345; AVX2-NEXT: Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>346; AVX2-NEXT: Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>347; AVX2-NEXT: Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>348; AVX2-NEXT: Cost Model: Found costs of 4 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>349; AVX2-NEXT: Cost Model: Found costs of 14 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>350; AVX2-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void351;352; AVX512F-LABEL: 'test_vXi8'353; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> <i32 1, i32 1>354; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>355; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>356; AVX512F-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>357; AVX512F-NEXT: Cost Model: Found costs of 4 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>358; AVX512F-NEXT: Cost Model: Found costs of 14 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>359; AVX512F-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void360;361; AVX512BW-LABEL: 'test_vXi8'362; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> <i32 1, i32 1>363; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>364; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>365; AVX512BW-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>366; AVX512BW-NEXT: Cost Model: Found costs of 4 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>367; AVX512BW-NEXT: Cost Model: Found costs of 8 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>368; AVX512BW-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void369;370; AVX512VBMI-LABEL: 'test_vXi8'371; AVX512VBMI-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> <i32 1, i32 1>372; AVX512VBMI-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>373; AVX512VBMI-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>374; AVX512VBMI-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>375; AVX512VBMI-NEXT: Cost Model: Found costs of 1 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>376; AVX512VBMI-NEXT: Cost Model: Found costs of 1 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>377; AVX512VBMI-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void378;379 %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> <i32 1, i32 1>380 %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>381 %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>382 %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>383 %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>384 %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>385 ret void386}387 388;389; Test some Identity Shuffles390;391 392define void @identity_vXf32(<4 x float> %a128, <4 x float> %b128, <8 x float> %a256, <8 x float> %b256, <16 x float> %a512, <16 x float> %b512) {393; CHECK-LABEL: 'identity_vXf32'394; CHECK-NEXT: Cost Model: Found costs of 0 for: %V128 = shufflevector <4 x float> %a128, <4 x float> %b128, <4 x i32> <i32 0, i32 poison, i32 2, i32 poison>395; CHECK-NEXT: Cost Model: Found costs of 0 for: %V256 = shufflevector <8 x float> %a256, <8 x float> %b256, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 poison, i32 poison, i32 poison, i32 poison>396; CHECK-NEXT: Cost Model: Found costs of 0 for: %V512 = shufflevector <16 x float> %a512, <16 x float> %b512, <16 x i32> <i32 0, i32 poison, i32 2, i32 poison, i32 4, i32 poison, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 poison>397; CHECK-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void398;399 %V128 = shufflevector <4 x float> %a128, <4 x float> %b128, <4 x i32> <i32 0, i32 undef, i32 2, i32 undef>400 %V256 = shufflevector <8 x float> %a256, <8 x float> %b256, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 undef, i32 undef, i32 undef, i32 undef>401 %V512 = shufflevector <16 x float> %a512, <16 x float> %b512, <16 x i32> <i32 0, i32 undef, i32 2, i32 undef, i32 4, i32 undef, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 undef>402 ret void403}404