549 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -O3 -S -mtriple=x86_64-- -mcpu=x86-64-v2 | FileCheck %s --check-prefixes=CHECK,SSE3; RUN: opt < %s -O3 -S -mtriple=x86_64-- -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=CHECK,AVX24; RUN: opt < %s -O3 -S -mtriple=x86_64-- -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=CHECK,AVX5125 6;7; PR58895 - replace shuffled _mm_blendv_epi8+icmp with select+icmp8;9 10;11; 256-bits types split to use 128-bit PBLENDVB selects12;13 14define <4 x double> @x86_pblendvb_v4f64_v2f64(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x double> %d) {15; CHECK-LABEL: @x86_pblendvb_v4f64_v2f64(16; CHECK-NEXT: [[CMP:%.*]] = fcmp olt <4 x double> [[C:%.*]], [[D:%.*]]17; CHECK-NEXT: [[DOTV:%.*]] = select <4 x i1> [[CMP]], <4 x double> [[B:%.*]], <4 x double> [[A:%.*]]18; CHECK-NEXT: ret <4 x double> [[DOTV]]19;20 %a.bc = bitcast <4 x double> %a to <32 x i8>21 %b.bc = bitcast <4 x double> %b to <32 x i8>22 %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>23 %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>24 %a.hi = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <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>25 %b.hi = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <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>26 %cmp = fcmp olt <4 x double> %c, %d27 %sext = sext <4 x i1> %cmp to <4 x i64>28 %sext.bc = bitcast <4 x i64> %sext to <32 x i8>29 %sext.lo = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>30 %sext.hi = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <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>31 %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)32 %sel.hi = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.hi, <16 x i8> %b.hi, <16 x i8> %sext.hi)33 %concat = shufflevector <16 x i8> %sel.lo, <16 x i8> %sel.hi, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>34 %res = bitcast <32 x i8> %concat to <4 x double>35 ret <4 x double> %res36}37 38define <8 x float> @x86_pblendvb_v8f32_v4f32(<8 x float> %a, <8 x float> %b, <8 x float> %c, <8 x float> %d) {39; CHECK-LABEL: @x86_pblendvb_v8f32_v4f32(40; CHECK-NEXT: [[CMP:%.*]] = fcmp olt <8 x float> [[C:%.*]], [[D:%.*]]41; CHECK-NEXT: [[DOTV:%.*]] = select <8 x i1> [[CMP]], <8 x float> [[B:%.*]], <8 x float> [[A:%.*]]42; CHECK-NEXT: ret <8 x float> [[DOTV]]43;44 %a.bc = bitcast <8 x float> %a to <32 x i8>45 %b.bc = bitcast <8 x float> %b to <32 x i8>46 %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>47 %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>48 %a.hi = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <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>49 %b.hi = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <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>50 %cmp = fcmp olt <8 x float> %c, %d51 %sext = sext <8 x i1> %cmp to <8 x i32>52 %sext.bc = bitcast <8 x i32> %sext to <32 x i8>53 %sext.lo = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>54 %sext.hi = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <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>55 %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)56 %sel.hi = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.hi, <16 x i8> %b.hi, <16 x i8> %sext.hi)57 %concat = shufflevector <16 x i8> %sel.lo, <16 x i8> %sel.hi, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>58 %res = bitcast <32 x i8> %concat to <8 x float>59 ret <8 x float> %res60}61 62define <4 x i64> @x86_pblendvb_v4i64_v2i64(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c, <4 x i64> %d) {63; CHECK-LABEL: @x86_pblendvb_v4i64_v2i64(64; CHECK-NEXT: [[CMP:%.*]] = icmp slt <4 x i64> [[C:%.*]], [[D:%.*]]65; CHECK-NEXT: [[TMP1:%.*]] = select <4 x i1> [[CMP]], <4 x i64> [[B:%.*]], <4 x i64> [[A:%.*]]66; CHECK-NEXT: ret <4 x i64> [[TMP1]]67;68 %a.bc = bitcast <4 x i64> %a to <32 x i8>69 %b.bc = bitcast <4 x i64> %b to <32 x i8>70 %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>71 %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>72 %a.hi = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <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>73 %b.hi = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <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>74 %cmp = icmp slt <4 x i64> %c, %d75 %sext = sext <4 x i1> %cmp to <4 x i64>76 %sext.bc = bitcast <4 x i64> %sext to <32 x i8>77 %sext.lo = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>78 %sext.hi = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <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>79 %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)80 %sel.hi = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.hi, <16 x i8> %b.hi, <16 x i8> %sext.hi)81 %concat = shufflevector <16 x i8> %sel.lo, <16 x i8> %sel.hi, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>82 %res = bitcast <32 x i8> %concat to <4 x i64>83 ret <4 x i64> %res84}85 86define <4 x i64> @x86_pblendvb_v8i32_v4i32(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c, <4 x i64> %d) {87; SSE-LABEL: @x86_pblendvb_v8i32_v4i32(88; SSE-NEXT: [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <8 x i32>89; SSE-NEXT: [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <8 x i32>90; SSE-NEXT: [[CMP:%.*]] = icmp slt <8 x i32> [[C_BC]], [[D_BC]]91; SSE-NEXT: [[TMP1:%.*]] = bitcast <4 x i64> [[A:%.*]] to <8 x i32>92; SSE-NEXT: [[TMP2:%.*]] = bitcast <4 x i64> [[B:%.*]] to <8 x i32>93; SSE-NEXT: [[TMP3:%.*]] = select <8 x i1> [[CMP]], <8 x i32> [[TMP2]], <8 x i32> [[TMP1]]94; SSE-NEXT: [[RES:%.*]] = bitcast <8 x i32> [[TMP3]] to <4 x i64>95; SSE-NEXT: ret <4 x i64> [[RES]]96;97; AVX2-LABEL: @x86_pblendvb_v8i32_v4i32(98; AVX2-NEXT: [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <8 x i32>99; AVX2-NEXT: [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <8 x i32>100; AVX2-NEXT: [[TMP1:%.*]] = icmp slt <8 x i32> [[C_BC]], [[D_BC]]101; AVX2-NEXT: [[TMP2:%.*]] = bitcast <4 x i64> [[B:%.*]] to <8 x i32>102; AVX2-NEXT: [[TMP3:%.*]] = bitcast <4 x i64> [[A:%.*]] to <8 x i32>103; AVX2-NEXT: [[TMP4:%.*]] = select <8 x i1> [[TMP1]], <8 x i32> [[TMP2]], <8 x i32> [[TMP3]]104; AVX2-NEXT: [[RES:%.*]] = bitcast <8 x i32> [[TMP4]] to <4 x i64>105; AVX2-NEXT: ret <4 x i64> [[RES]]106;107; AVX512-LABEL: @x86_pblendvb_v8i32_v4i32(108; AVX512-NEXT: [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <8 x i32>109; AVX512-NEXT: [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <8 x i32>110; AVX512-NEXT: [[CMP:%.*]] = icmp slt <8 x i32> [[C_BC]], [[D_BC]]111; AVX512-NEXT: [[TMP1:%.*]] = bitcast <4 x i64> [[A:%.*]] to <8 x i32>112; AVX512-NEXT: [[TMP2:%.*]] = bitcast <4 x i64> [[B:%.*]] to <8 x i32>113; AVX512-NEXT: [[TMP3:%.*]] = select <8 x i1> [[CMP]], <8 x i32> [[TMP2]], <8 x i32> [[TMP1]]114; AVX512-NEXT: [[RES:%.*]] = bitcast <8 x i32> [[TMP3]] to <4 x i64>115; AVX512-NEXT: ret <4 x i64> [[RES]]116;117 %a.bc = bitcast <4 x i64> %a to <32 x i8>118 %b.bc = bitcast <4 x i64> %b to <32 x i8>119 %c.bc = bitcast <4 x i64> %c to <8 x i32>120 %d.bc = bitcast <4 x i64> %d to <8 x i32>121 %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>122 %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>123 %a.hi = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <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>124 %b.hi = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <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>125 %cmp = icmp slt <8 x i32> %c.bc, %d.bc126 %sext = sext <8 x i1> %cmp to <8 x i32>127 %sext.bc = bitcast <8 x i32> %sext to <32 x i8>128 %sext.lo = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>129 %sext.hi = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <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>130 %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)131 %sel.hi = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.hi, <16 x i8> %b.hi, <16 x i8> %sext.hi)132 %concat = shufflevector <16 x i8> %sel.lo, <16 x i8> %sel.hi, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>133 %res = bitcast <32 x i8> %concat to <4 x i64>134 ret <4 x i64> %res135}136 137define <4 x i64> @x86_pblendvb_v16i16_v8i16(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c, <4 x i64> %d) {138; SSE-LABEL: @x86_pblendvb_v16i16_v8i16(139; SSE-NEXT: [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <16 x i16>140; SSE-NEXT: [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <16 x i16>141; SSE-NEXT: [[CMP:%.*]] = icmp slt <16 x i16> [[C_BC]], [[D_BC]]142; SSE-NEXT: [[TMP1:%.*]] = bitcast <4 x i64> [[A:%.*]] to <16 x i16>143; SSE-NEXT: [[TMP2:%.*]] = bitcast <4 x i64> [[B:%.*]] to <16 x i16>144; SSE-NEXT: [[TMP3:%.*]] = select <16 x i1> [[CMP]], <16 x i16> [[TMP2]], <16 x i16> [[TMP1]]145; SSE-NEXT: [[RES:%.*]] = bitcast <16 x i16> [[TMP3]] to <4 x i64>146; SSE-NEXT: ret <4 x i64> [[RES]]147;148; AVX2-LABEL: @x86_pblendvb_v16i16_v8i16(149; AVX2-NEXT: [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <16 x i16>150; AVX2-NEXT: [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <16 x i16>151; AVX2-NEXT: [[TMP1:%.*]] = icmp slt <16 x i16> [[C_BC]], [[D_BC]]152; AVX2-NEXT: [[TMP2:%.*]] = bitcast <4 x i64> [[B:%.*]] to <16 x i16>153; AVX2-NEXT: [[TMP3:%.*]] = bitcast <4 x i64> [[A:%.*]] to <16 x i16>154; AVX2-NEXT: [[TMP4:%.*]] = select <16 x i1> [[TMP1]], <16 x i16> [[TMP2]], <16 x i16> [[TMP3]]155; AVX2-NEXT: [[RES:%.*]] = bitcast <16 x i16> [[TMP4]] to <4 x i64>156; AVX2-NEXT: ret <4 x i64> [[RES]]157;158; AVX512-LABEL: @x86_pblendvb_v16i16_v8i16(159; AVX512-NEXT: [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <16 x i16>160; AVX512-NEXT: [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <16 x i16>161; AVX512-NEXT: [[CMP:%.*]] = icmp slt <16 x i16> [[C_BC]], [[D_BC]]162; AVX512-NEXT: [[TMP1:%.*]] = bitcast <4 x i64> [[A:%.*]] to <16 x i16>163; AVX512-NEXT: [[TMP2:%.*]] = bitcast <4 x i64> [[B:%.*]] to <16 x i16>164; AVX512-NEXT: [[TMP3:%.*]] = select <16 x i1> [[CMP]], <16 x i16> [[TMP2]], <16 x i16> [[TMP1]]165; AVX512-NEXT: [[RES:%.*]] = bitcast <16 x i16> [[TMP3]] to <4 x i64>166; AVX512-NEXT: ret <4 x i64> [[RES]]167;168 %a.bc = bitcast <4 x i64> %a to <32 x i8>169 %b.bc = bitcast <4 x i64> %b to <32 x i8>170 %c.bc = bitcast <4 x i64> %c to <16 x i16>171 %d.bc = bitcast <4 x i64> %d to <16 x i16>172 %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>173 %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>174 %a.hi = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <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>175 %b.hi = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <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>176 %cmp = icmp slt <16 x i16> %c.bc, %d.bc177 %sext = sext <16 x i1> %cmp to <16 x i16>178 %sext.bc = bitcast <16 x i16> %sext to <32 x i8>179 %sext.lo = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>180 %sext.hi = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <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>181 %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)182 %sel.hi = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.hi, <16 x i8> %b.hi, <16 x i8> %sext.hi)183 %concat = shufflevector <16 x i8> %sel.lo, <16 x i8> %sel.hi, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>184 %res = bitcast <32 x i8> %concat to <4 x i64>185 ret <4 x i64> %res186}187 188define <4 x i64> @x86_pblendvb_v32i8_v16i8(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c, <4 x i64> %d) {189; CHECK-LABEL: @x86_pblendvb_v32i8_v16i8(190; CHECK-NEXT: [[A_BC:%.*]] = bitcast <4 x i64> [[A:%.*]] to <32 x i8>191; CHECK-NEXT: [[B_BC:%.*]] = bitcast <4 x i64> [[B:%.*]] to <32 x i8>192; CHECK-NEXT: [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <32 x i8>193; CHECK-NEXT: [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <32 x i8>194; CHECK-NEXT: [[CMP:%.*]] = icmp slt <32 x i8> [[C_BC]], [[D_BC]]195; CHECK-NEXT: [[CONCAT:%.*]] = select <32 x i1> [[CMP]], <32 x i8> [[B_BC]], <32 x i8> [[A_BC]]196; CHECK-NEXT: [[RES:%.*]] = bitcast <32 x i8> [[CONCAT]] to <4 x i64>197; CHECK-NEXT: ret <4 x i64> [[RES]]198;199 %a.bc = bitcast <4 x i64> %a to <32 x i8>200 %b.bc = bitcast <4 x i64> %b to <32 x i8>201 %c.bc = bitcast <4 x i64> %c to <32 x i8>202 %d.bc = bitcast <4 x i64> %d to <32 x i8>203 %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>204 %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>205 %a.hi = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <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>206 %b.hi = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <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>207 %cmp = icmp slt <32 x i8> %c.bc, %d.bc208 %sext = sext <32 x i1> %cmp to <32 x i8>209 %sext.lo = shufflevector <32 x i8> %sext, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>210 %sext.hi = shufflevector <32 x i8> %sext, <32 x i8> poison, <16 x i32> <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>211 %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)212 %sel.hi = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.hi, <16 x i8> %b.hi, <16 x i8> %sext.hi)213 %concat = shufflevector <16 x i8> %sel.lo, <16 x i8> %sel.hi, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>214 %res = bitcast <32 x i8> %concat to <4 x i64>215 ret <4 x i64> %res216}217 218;219; 512-bits types split to use 256-bit PBLENDVB selects220;221 222define <8 x double> @x86_pblendvb_v8f64_v4f64(<8 x double> %a, <8 x double> %b, <8 x double> %c, <8 x double> %d) {223; CHECK-LABEL: @x86_pblendvb_v8f64_v4f64(224; CHECK-NEXT: [[CMP:%.*]] = fcmp olt <8 x double> [[C:%.*]], [[D:%.*]]225; CHECK-NEXT: [[DOTV:%.*]] = select <8 x i1> [[CMP]], <8 x double> [[B:%.*]], <8 x double> [[A:%.*]]226; CHECK-NEXT: ret <8 x double> [[DOTV]]227;228 %a.bc = bitcast <8 x double> %a to <64 x i8>229 %b.bc = bitcast <8 x double> %b to <64 x i8>230 %a.lo = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>231 %b.lo = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>232 %a.hi = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <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>233 %b.hi = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <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>234 %cmp = fcmp olt <8 x double> %c, %d235 %sext = sext <8 x i1> %cmp to <8 x i64>236 %sext.bc = bitcast <8 x i64> %sext to <64 x i8>237 %sext.lo = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>238 %sext.hi = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <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>239 %sel.lo = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.lo, <32 x i8> %b.lo, <32 x i8> %sext.lo)240 %sel.hi = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.hi, <32 x i8> %b.hi, <32 x i8> %sext.hi)241 %concat = shufflevector <32 x i8> %sel.lo, <32 x i8> %sel.hi, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, 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>242 %res = bitcast <64 x i8> %concat to <8 x double>243 ret <8 x double> %res244}245 246define <16 x float> @x86_pblendvb_v16f32_v8f32(<16 x float> %a, <16 x float> %b, <16 x float> %c, <16 x float> %d) {247; CHECK-LABEL: @x86_pblendvb_v16f32_v8f32(248; CHECK-NEXT: [[CMP:%.*]] = fcmp olt <16 x float> [[C:%.*]], [[D:%.*]]249; CHECK-NEXT: [[DOTV:%.*]] = select <16 x i1> [[CMP]], <16 x float> [[B:%.*]], <16 x float> [[A:%.*]]250; CHECK-NEXT: ret <16 x float> [[DOTV]]251;252 %a.bc = bitcast <16 x float> %a to <64 x i8>253 %b.bc = bitcast <16 x float> %b to <64 x i8>254 %a.lo = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>255 %b.lo = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>256 %a.hi = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <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>257 %b.hi = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <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>258 %cmp = fcmp olt <16 x float> %c, %d259 %sext = sext <16 x i1> %cmp to <16 x i32>260 %sext.bc = bitcast <16 x i32> %sext to <64 x i8>261 %sext.lo = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>262 %sext.hi = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <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>263 %sel.lo = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.lo, <32 x i8> %b.lo, <32 x i8> %sext.lo)264 %sel.hi = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.hi, <32 x i8> %b.hi, <32 x i8> %sext.hi)265 %concat = shufflevector <32 x i8> %sel.lo, <32 x i8> %sel.hi, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, 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>266 %res = bitcast <64 x i8> %concat to <16 x float>267 ret <16 x float> %res268}269 270define <8 x i64> @x86_pblendvb_v8i64_v4i64(<8 x i64> %a, <8 x i64> %b, <8 x i64> %c, <8 x i64> %d) {271; CHECK-LABEL: @x86_pblendvb_v8i64_v4i64(272; CHECK-NEXT: [[CMP:%.*]] = icmp slt <8 x i64> [[C:%.*]], [[D:%.*]]273; CHECK-NEXT: [[TMP1:%.*]] = select <8 x i1> [[CMP]], <8 x i64> [[B:%.*]], <8 x i64> [[A:%.*]]274; CHECK-NEXT: ret <8 x i64> [[TMP1]]275;276 %a.bc = bitcast <8 x i64> %a to <64 x i8>277 %b.bc = bitcast <8 x i64> %b to <64 x i8>278 %a.lo = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>279 %b.lo = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>280 %a.hi = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <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>281 %b.hi = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <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>282 %cmp = icmp slt <8 x i64> %c, %d283 %sext = sext <8 x i1> %cmp to <8 x i64>284 %sext.bc = bitcast <8 x i64> %sext to <64 x i8>285 %sext.lo = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>286 %sext.hi = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <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>287 %sel.lo = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.lo, <32 x i8> %b.lo, <32 x i8> %sext.lo)288 %sel.hi = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.hi, <32 x i8> %b.hi, <32 x i8> %sext.hi)289 %concat = shufflevector <32 x i8> %sel.lo, <32 x i8> %sel.hi, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, 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>290 %res = bitcast <64 x i8> %concat to <8 x i64>291 ret <8 x i64> %res292}293 294define <8 x i64> @x86_pblendvb_v16i32_v8i32(<8 x i64> %a, <8 x i64> %b, <8 x i64> %c, <8 x i64> %d) {295; CHECK-LABEL: @x86_pblendvb_v16i32_v8i32(296; CHECK-NEXT: [[C_BC:%.*]] = bitcast <8 x i64> [[C:%.*]] to <16 x i32>297; CHECK-NEXT: [[D_BC:%.*]] = bitcast <8 x i64> [[D:%.*]] to <16 x i32>298; CHECK-NEXT: [[CMP:%.*]] = icmp slt <16 x i32> [[C_BC]], [[D_BC]]299; CHECK-NEXT: [[TMP1:%.*]] = bitcast <8 x i64> [[A:%.*]] to <16 x i32>300; CHECK-NEXT: [[TMP2:%.*]] = bitcast <8 x i64> [[B:%.*]] to <16 x i32>301; CHECK-NEXT: [[TMP3:%.*]] = select <16 x i1> [[CMP]], <16 x i32> [[TMP2]], <16 x i32> [[TMP1]]302; CHECK-NEXT: [[RES:%.*]] = bitcast <16 x i32> [[TMP3]] to <8 x i64>303; CHECK-NEXT: ret <8 x i64> [[RES]]304;305 %a.bc = bitcast <8 x i64> %a to <64 x i8>306 %b.bc = bitcast <8 x i64> %b to <64 x i8>307 %c.bc = bitcast <8 x i64> %c to <16 x i32>308 %d.bc = bitcast <8 x i64> %d to <16 x i32>309 %a.lo = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>310 %b.lo = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>311 %a.hi = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <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>312 %b.hi = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <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>313 %cmp = icmp slt <16 x i32> %c.bc, %d.bc314 %sext = sext <16 x i1> %cmp to <16 x i32>315 %sext.bc = bitcast <16 x i32> %sext to <64 x i8>316 %sext.lo = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>317 %sext.hi = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <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>318 %sel.lo = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.lo, <32 x i8> %b.lo, <32 x i8> %sext.lo)319 %sel.hi = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.hi, <32 x i8> %b.hi, <32 x i8> %sext.hi)320 %concat = shufflevector <32 x i8> %sel.lo, <32 x i8> %sel.hi, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, 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>321 %res = bitcast <64 x i8> %concat to <8 x i64>322 ret <8 x i64> %res323}324 325define <8 x i64> @x86_pblendvb_v32i16_v16i16(<8 x i64> %a, <8 x i64> %b, <8 x i64> %c, <8 x i64> %d) {326; SSE-LABEL: @x86_pblendvb_v32i16_v16i16(327; SSE-NEXT: [[C_BC:%.*]] = bitcast <8 x i64> [[C:%.*]] to <32 x i16>328; SSE-NEXT: [[D_BC:%.*]] = bitcast <8 x i64> [[D:%.*]] to <32 x i16>329; SSE-NEXT: [[TMP1:%.*]] = icmp slt <32 x i16> [[C_BC]], [[D_BC]]330; SSE-NEXT: [[TMP2:%.*]] = bitcast <8 x i64> [[B:%.*]] to <32 x i16>331; SSE-NEXT: [[TMP3:%.*]] = bitcast <8 x i64> [[A:%.*]] to <32 x i16>332; SSE-NEXT: [[TMP4:%.*]] = select <32 x i1> [[TMP1]], <32 x i16> [[TMP2]], <32 x i16> [[TMP3]]333; SSE-NEXT: [[RES:%.*]] = bitcast <32 x i16> [[TMP4]] to <8 x i64>334; SSE-NEXT: ret <8 x i64> [[RES]]335;336; AVX2-LABEL: @x86_pblendvb_v32i16_v16i16(337; AVX2-NEXT: [[C_BC:%.*]] = bitcast <8 x i64> [[C:%.*]] to <32 x i16>338; AVX2-NEXT: [[D_BC:%.*]] = bitcast <8 x i64> [[D:%.*]] to <32 x i16>339; AVX2-NEXT: [[CMP:%.*]] = icmp slt <32 x i16> [[C_BC]], [[D_BC]]340; AVX2-NEXT: [[TMP1:%.*]] = bitcast <8 x i64> [[A:%.*]] to <32 x i16>341; AVX2-NEXT: [[TMP2:%.*]] = bitcast <8 x i64> [[B:%.*]] to <32 x i16>342; AVX2-NEXT: [[TMP3:%.*]] = select <32 x i1> [[CMP]], <32 x i16> [[TMP2]], <32 x i16> [[TMP1]]343; AVX2-NEXT: [[RES:%.*]] = bitcast <32 x i16> [[TMP3]] to <8 x i64>344; AVX2-NEXT: ret <8 x i64> [[RES]]345;346; AVX512-LABEL: @x86_pblendvb_v32i16_v16i16(347; AVX512-NEXT: [[C_BC:%.*]] = bitcast <8 x i64> [[C:%.*]] to <32 x i16>348; AVX512-NEXT: [[D_BC:%.*]] = bitcast <8 x i64> [[D:%.*]] to <32 x i16>349; AVX512-NEXT: [[CMP:%.*]] = icmp slt <32 x i16> [[C_BC]], [[D_BC]]350; AVX512-NEXT: [[TMP1:%.*]] = bitcast <8 x i64> [[A:%.*]] to <32 x i16>351; AVX512-NEXT: [[TMP2:%.*]] = bitcast <8 x i64> [[B:%.*]] to <32 x i16>352; AVX512-NEXT: [[TMP3:%.*]] = select <32 x i1> [[CMP]], <32 x i16> [[TMP2]], <32 x i16> [[TMP1]]353; AVX512-NEXT: [[RES:%.*]] = bitcast <32 x i16> [[TMP3]] to <8 x i64>354; AVX512-NEXT: ret <8 x i64> [[RES]]355;356 %a.bc = bitcast <8 x i64> %a to <64 x i8>357 %b.bc = bitcast <8 x i64> %b to <64 x i8>358 %c.bc = bitcast <8 x i64> %c to <32 x i16>359 %d.bc = bitcast <8 x i64> %d to <32 x i16>360 %a.lo = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>361 %b.lo = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>362 %a.hi = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <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>363 %b.hi = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <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>364 %cmp = icmp slt <32 x i16> %c.bc, %d.bc365 %sext = sext <32 x i1> %cmp to <32 x i16>366 %sext.bc = bitcast <32 x i16> %sext to <64 x i8>367 %sext.lo = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>368 %sext.hi = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <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>369 %sel.lo = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.lo, <32 x i8> %b.lo, <32 x i8> %sext.lo)370 %sel.hi = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.hi, <32 x i8> %b.hi, <32 x i8> %sext.hi)371 %concat = shufflevector <32 x i8> %sel.lo, <32 x i8> %sel.hi, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, 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>372 %res = bitcast <64 x i8> %concat to <8 x i64>373 ret <8 x i64> %res374}375 376define <8 x i64> @x86_pblendvb_v64i8_v32i8(<8 x i64> %a, <8 x i64> %b, <8 x i64> %c, <8 x i64> %d) {377; CHECK-LABEL: @x86_pblendvb_v64i8_v32i8(378; CHECK-NEXT: [[A_BC:%.*]] = bitcast <8 x i64> [[A:%.*]] to <64 x i8>379; CHECK-NEXT: [[B_BC:%.*]] = bitcast <8 x i64> [[B:%.*]] to <64 x i8>380; CHECK-NEXT: [[C_BC:%.*]] = bitcast <8 x i64> [[C:%.*]] to <64 x i8>381; CHECK-NEXT: [[D_BC:%.*]] = bitcast <8 x i64> [[D:%.*]] to <64 x i8>382; CHECK-NEXT: [[CMP:%.*]] = icmp slt <64 x i8> [[C_BC]], [[D_BC]]383; CHECK-NEXT: [[CONCAT:%.*]] = select <64 x i1> [[CMP]], <64 x i8> [[B_BC]], <64 x i8> [[A_BC]]384; CHECK-NEXT: [[RES:%.*]] = bitcast <64 x i8> [[CONCAT]] to <8 x i64>385; CHECK-NEXT: ret <8 x i64> [[RES]]386;387 %a.bc = bitcast <8 x i64> %a to <64 x i8>388 %b.bc = bitcast <8 x i64> %b to <64 x i8>389 %c.bc = bitcast <8 x i64> %c to <64 x i8>390 %d.bc = bitcast <8 x i64> %d to <64 x i8>391 %a.lo = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>392 %b.lo = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>393 %a.hi = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <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>394 %b.hi = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <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>395 %cmp = icmp slt <64 x i8> %c.bc, %d.bc396 %sext = sext <64 x i1> %cmp to <64 x i8>397 %sext.lo = shufflevector <64 x i8> %sext, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>398 %sext.hi = shufflevector <64 x i8> %sext, <64 x i8> poison, <32 x i32> <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>399 %sel.lo = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.lo, <32 x i8> %b.lo, <32 x i8> %sext.lo)400 %sel.hi = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.hi, <32 x i8> %b.hi, <32 x i8> %sext.hi)401 %concat = shufflevector <32 x i8> %sel.lo, <32 x i8> %sel.hi, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, 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, 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>402 %res = bitcast <64 x i8> %concat to <8 x i64>403 ret <8 x i64> %res404}405 406define <2 x i64> @PR66513(<2 x i64> %a, <2 x i64> %b, <2 x i64> %c, <2 x i64> %src) {407; CHECK-LABEL: @PR66513(408; CHECK-NEXT: [[I:%.*]] = bitcast <2 x i64> [[A:%.*]] to <4 x i32>409; CHECK-NEXT: [[CMP_I23:%.*]] = icmp sgt <4 x i32> [[I]], zeroinitializer410; CHECK-NEXT: [[I2:%.*]] = bitcast <2 x i64> [[B:%.*]] to <4 x i32>411; CHECK-NEXT: [[CMP_I21:%.*]] = icmp sgt <4 x i32> [[I2]], zeroinitializer412; CHECK-NEXT: [[I4:%.*]] = bitcast <2 x i64> [[C:%.*]] to <4 x i32>413; CHECK-NEXT: [[CMP_I:%.*]] = icmp sgt <4 x i32> [[I4]], zeroinitializer414; CHECK-NEXT: [[NARROW:%.*]] = select <4 x i1> [[CMP_I21]], <4 x i1> [[CMP_I23]], <4 x i1> zeroinitializer415; CHECK-NEXT: [[XOR_I_INNER1:%.*]] = xor <4 x i1> [[NARROW]], [[CMP_I]]416; CHECK-NEXT: [[NARROW3:%.*]] = select <4 x i1> [[CMP_I23]], <4 x i1> [[XOR_I_INNER1]], <4 x i1> zeroinitializer417; CHECK-NEXT: [[AND_I25_INNER2:%.*]] = and <4 x i1> [[XOR_I_INNER1]], [[CMP_I21]]418; CHECK-NEXT: [[TMP1:%.*]] = bitcast <2 x i64> [[SRC:%.*]] to <4 x i32>419; CHECK-NEXT: [[TMP2:%.*]] = select <4 x i1> [[NARROW]], <4 x i32> [[TMP1]], <4 x i32> zeroinitializer420; CHECK-NEXT: [[TMP3:%.*]] = bitcast <2 x i64> [[A]] to <4 x i32>421; CHECK-NEXT: [[TMP4:%.*]] = select <4 x i1> [[NARROW3]], <4 x i32> [[TMP3]], <4 x i32> [[TMP2]]422; CHECK-NEXT: [[TMP5:%.*]] = bitcast <2 x i64> [[B]] to <4 x i32>423; CHECK-NEXT: [[SEXT_I:%.*]] = select <4 x i1> [[AND_I25_INNER2]], <4 x i32> [[TMP5]], <4 x i32> [[TMP4]]424; CHECK-NEXT: [[I5:%.*]] = bitcast <4 x i32> [[SEXT_I]] to <2 x i64>425; CHECK-NEXT: ret <2 x i64> [[I5]]426;427 %i = bitcast <2 x i64> %a to <4 x i32>428 %cmp.i23 = icmp sgt <4 x i32> %i, zeroinitializer429 %sext.i24 = sext <4 x i1> %cmp.i23 to <4 x i32>430 %i1 = bitcast <4 x i32> %sext.i24 to <2 x i64>431 %i2 = bitcast <2 x i64> %b to <4 x i32>432 %cmp.i21 = icmp sgt <4 x i32> %i2, zeroinitializer433 %sext.i22 = sext <4 x i1> %cmp.i21 to <4 x i32>434 %i3 = bitcast <4 x i32> %sext.i22 to <2 x i64>435 %i4 = bitcast <2 x i64> %c to <4 x i32>436 %cmp.i = icmp sgt <4 x i32> %i4, zeroinitializer437 %sext.i = sext <4 x i1> %cmp.i to <4 x i32>438 %i5 = bitcast <4 x i32> %sext.i to <2 x i64>439 %and.i27 = and <2 x i64> %i1, %i3440 %xor.i = xor <2 x i64> %and.i27, %i5441 %and.i26 = and <2 x i64> %xor.i, %i1442 %and.i25 = and <2 x i64> %xor.i, %i3443 %and.i = and <2 x i64> %src, %and.i27444 %i6 = bitcast <2 x i64> %and.i to <16 x i8>445 %i7 = bitcast <2 x i64> %a to <16 x i8>446 %i8 = bitcast <2 x i64> %and.i26 to <16 x i8>447 %i9 = call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %i6, <16 x i8> %i7, <16 x i8> %i8)448 %i10 = bitcast <16 x i8> %i9 to <2 x i64>449 %i11 = bitcast <2 x i64> %i10 to <16 x i8>450 %i12 = bitcast <2 x i64> %b to <16 x i8>451 %i13 = bitcast <2 x i64> %and.i25 to <16 x i8>452 %i14 = call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %i11, <16 x i8> %i12, <16 x i8> %i13)453 %i15 = bitcast <16 x i8> %i14 to <2 x i64>454 ret <2 x i64> %i15455}456 457;458; Negative Tests459;460 461; non-pow2 source462define <2 x double> @x86_pblendvb_v3f64_v2f64(<3 x double> %a, <3 x double> %b, <3 x double> %c, <3 x double> %d) {463; CHECK-LABEL: @x86_pblendvb_v3f64_v2f64(464; CHECK-NEXT: [[A_BC:%.*]] = bitcast <3 x double> [[A:%.*]] to <24 x i8>465; CHECK-NEXT: [[B_BC:%.*]] = bitcast <3 x double> [[B:%.*]] to <24 x i8>466; CHECK-NEXT: [[A_LO:%.*]] = shufflevector <24 x i8> [[A_BC]], <24 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>467; CHECK-NEXT: [[B_LO:%.*]] = shufflevector <24 x i8> [[B_BC]], <24 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>468; CHECK-NEXT: [[CMP:%.*]] = fcmp olt <3 x double> [[C:%.*]], [[D:%.*]]469; CHECK-NEXT: [[SEXT:%.*]] = sext <3 x i1> [[CMP]] to <3 x i64>470; CHECK-NEXT: [[SEXT_BC:%.*]] = bitcast <3 x i64> [[SEXT]] to <24 x i8>471; CHECK-NEXT: [[SEXT_LO:%.*]] = shufflevector <24 x i8> [[SEXT_BC]], <24 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>472; CHECK-NEXT: [[SEL_LO:%.*]] = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> [[A_LO]], <16 x i8> [[B_LO]], <16 x i8> [[SEXT_LO]])473; CHECK-NEXT: [[RES:%.*]] = bitcast <16 x i8> [[SEL_LO]] to <2 x double>474; CHECK-NEXT: ret <2 x double> [[RES]]475;476 %a.bc = bitcast <3 x double> %a to <24 x i8>477 %b.bc = bitcast <3 x double> %b to <24 x i8>478 %a.lo = shufflevector <24 x i8> %a.bc, <24 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>479 %b.lo = shufflevector <24 x i8> %b.bc, <24 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>480 %cmp = fcmp olt <3 x double> %c, %d481 %sext = sext <3 x i1> %cmp to <3 x i64>482 %sext.bc = bitcast <3 x i64> %sext to <24 x i8>483 %sext.lo = shufflevector <24 x i8> %sext.bc, <24 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>484 %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)485 %res = bitcast <16 x i8> %sel.lo to <2 x double>486 ret <2 x double> %res487}488 489; shuffle masks contain undef490define <2 x i64> @x86_pblendvb_v32i8_v16i8_undefs(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c, <4 x i64> %d) {491; CHECK-LABEL: @x86_pblendvb_v32i8_v16i8_undefs(492; CHECK-NEXT: [[A_BC:%.*]] = bitcast <4 x i64> [[A:%.*]] to <32 x i8>493; CHECK-NEXT: [[B_BC:%.*]] = bitcast <4 x i64> [[B:%.*]] to <32 x i8>494; CHECK-NEXT: [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <32 x i8>495; CHECK-NEXT: [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <32 x i8>496; CHECK-NEXT: [[A_LO:%.*]] = shufflevector <32 x i8> [[A_BC]], <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>497; CHECK-NEXT: [[B_LO:%.*]] = shufflevector <32 x i8> [[B_BC]], <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>498; CHECK-NEXT: [[CMP:%.*]] = icmp slt <32 x i8> [[C_BC]], [[D_BC]]499; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <32 x i1> [[CMP]], <32 x i1> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 poison, i32 poison, i32 poison, i32 poison>500; CHECK-NEXT: [[SEL_LO:%.*]] = select <16 x i1> [[TMP1]], <16 x i8> [[B_LO]], <16 x i8> [[A_LO]]501; CHECK-NEXT: [[RES:%.*]] = bitcast <16 x i8> [[SEL_LO]] to <2 x i64>502; CHECK-NEXT: ret <2 x i64> [[RES]]503;504 %a.bc = bitcast <4 x i64> %a to <32 x i8>505 %b.bc = bitcast <4 x i64> %b to <32 x i8>506 %c.bc = bitcast <4 x i64> %c to <32 x i8>507 %d.bc = bitcast <4 x i64> %d to <32 x i8>508 %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>509 %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>510 %cmp = icmp slt <32 x i8> %c.bc, %d.bc511 %sext = sext <32 x i1> %cmp to <32 x i8>512 %sext.lo = shufflevector <32 x i8> %sext, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 undef, i32 undef, i32 undef, i32 undef>513 %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)514 %res = bitcast <16 x i8> %sel.lo to <2 x i64>515 ret <2 x i64> %res516}517 518; masks elements too small519define <2 x i64> @x86_pblendvb_v32i8_v32i4(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c, <4 x i64> %d) {520; CHECK-LABEL: @x86_pblendvb_v32i8_v32i4(521; CHECK-NEXT: [[A_BC:%.*]] = bitcast <4 x i64> [[A:%.*]] to <32 x i8>522; CHECK-NEXT: [[B_BC:%.*]] = bitcast <4 x i64> [[B:%.*]] to <32 x i8>523; CHECK-NEXT: [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <64 x i4>524; CHECK-NEXT: [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <64 x i4>525; CHECK-NEXT: [[A_LO:%.*]] = shufflevector <32 x i8> [[A_BC]], <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>526; CHECK-NEXT: [[B_LO:%.*]] = shufflevector <32 x i8> [[B_BC]], <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>527; CHECK-NEXT: [[CMP:%.*]] = icmp slt <64 x i4> [[C_BC]], [[D_BC]]528; CHECK-NEXT: [[SEXT:%.*]] = sext <64 x i1> [[CMP]] to <64 x i4>529; CHECK-NEXT: [[SEXT_BC:%.*]] = bitcast <64 x i4> [[SEXT]] to <32 x i8>530; CHECK-NEXT: [[SEXT_LO:%.*]] = shufflevector <32 x i8> [[SEXT_BC]], <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>531; CHECK-NEXT: [[SEL_LO:%.*]] = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> [[A_LO]], <16 x i8> [[B_LO]], <16 x i8> [[SEXT_LO]])532; CHECK-NEXT: [[RES:%.*]] = bitcast <16 x i8> [[SEL_LO]] to <2 x i64>533; CHECK-NEXT: ret <2 x i64> [[RES]]534;535 %a.bc = bitcast <4 x i64> %a to <32 x i8>536 %b.bc = bitcast <4 x i64> %b to <32 x i8>537 %c.bc = bitcast <4 x i64> %c to <64 x i4>538 %d.bc = bitcast <4 x i64> %d to <64 x i4>539 %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>540 %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>541 %cmp = icmp slt <64 x i4> %c.bc, %d.bc542 %sext = sext <64 x i1> %cmp to <64 x i4>543 %sext.bc = bitcast <64 x i4> %sext to <32 x i8>544 %sext.lo = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>545 %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)546 %res = bitcast <16 x i8> %sel.lo to <2 x i64>547 ret <2 x i64> %res548}549