1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=X32,X32-SLOW 3; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X32,X32-FAST-ALL 4; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X32,X32-FAST-PERLANE 5; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=X64,X64-SLOW 6; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X64,X64-FAST-ALL 7; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X64,X64-FAST-PERLANE 8 9define <4 x i32> @trunc4(<4 x i64> %A) nounwind { 10; X32-SLOW-LABEL: trunc4: 11; X32-SLOW: # %bb.0: 12; X32-SLOW-NEXT: vextractf128 $1, %ymm0, %xmm1 13; X32-SLOW-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2] 14; X32-SLOW-NEXT: vzeroupper 15; X32-SLOW-NEXT: retl 16; 17; X32-FAST-ALL-LABEL: trunc4: 18; X32-FAST-ALL: # %bb.0: 19; X32-FAST-ALL-NEXT: vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u> 20; X32-FAST-ALL-NEXT: vpermps %ymm0, %ymm1, %ymm0 21; X32-FAST-ALL-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 22; X32-FAST-ALL-NEXT: vzeroupper 23; X32-FAST-ALL-NEXT: retl 24; 25; X32-FAST-PERLANE-LABEL: trunc4: 26; X32-FAST-PERLANE: # %bb.0: 27; X32-FAST-PERLANE-NEXT: vextractf128 $1, %ymm0, %xmm1 28; X32-FAST-PERLANE-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2] 29; X32-FAST-PERLANE-NEXT: vzeroupper 30; X32-FAST-PERLANE-NEXT: retl 31; 32; X64-SLOW-LABEL: trunc4: 33; X64-SLOW: # %bb.0: 34; X64-SLOW-NEXT: vextractf128 $1, %ymm0, %xmm1 35; X64-SLOW-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2] 36; X64-SLOW-NEXT: vzeroupper 37; X64-SLOW-NEXT: retq 38; 39; X64-FAST-ALL-LABEL: trunc4: 40; X64-FAST-ALL: # %bb.0: 41; X64-FAST-ALL-NEXT: vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u> 42; X64-FAST-ALL-NEXT: vpermps %ymm0, %ymm1, %ymm0 43; X64-FAST-ALL-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 44; X64-FAST-ALL-NEXT: vzeroupper 45; X64-FAST-ALL-NEXT: retq 46; 47; X64-FAST-PERLANE-LABEL: trunc4: 48; X64-FAST-PERLANE: # %bb.0: 49; X64-FAST-PERLANE-NEXT: vextractf128 $1, %ymm0, %xmm1 50; X64-FAST-PERLANE-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2] 51; X64-FAST-PERLANE-NEXT: vzeroupper 52; X64-FAST-PERLANE-NEXT: retq 53 %B = trunc <4 x i64> %A to <4 x i32> 54 ret <4 x i32>%B 55} 56 57define <8 x i16> @trunc8(<8 x i32> %A) nounwind { 58; X32-LABEL: trunc8: 59; X32: # %bb.0: 60; X32-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u] 61; X32-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3] 62; X32-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 63; X32-NEXT: vzeroupper 64; X32-NEXT: retl 65; 66; X64-LABEL: trunc8: 67; X64: # %bb.0: 68; X64-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u] 69; X64-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3] 70; X64-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 71; X64-NEXT: vzeroupper 72; X64-NEXT: retq 73 %B = trunc <8 x i32> %A to <8 x i16> 74 ret <8 x i16>%B 75} 76 77define <4 x i64> @sext4(<4 x i32> %A) nounwind { 78; X32-LABEL: sext4: 79; X32: # %bb.0: 80; X32-NEXT: vpmovsxdq %xmm0, %ymm0 81; X32-NEXT: retl 82; 83; X64-LABEL: sext4: 84; X64: # %bb.0: 85; X64-NEXT: vpmovsxdq %xmm0, %ymm0 86; X64-NEXT: retq 87 %B = sext <4 x i32> %A to <4 x i64> 88 ret <4 x i64>%B 89} 90 91define <8 x i32> @sext8(<8 x i16> %A) nounwind { 92; X32-LABEL: sext8: 93; X32: # %bb.0: 94; X32-NEXT: vpmovsxwd %xmm0, %ymm0 95; X32-NEXT: retl 96; 97; X64-LABEL: sext8: 98; X64: # %bb.0: 99; X64-NEXT: vpmovsxwd %xmm0, %ymm0 100; X64-NEXT: retq 101 %B = sext <8 x i16> %A to <8 x i32> 102 ret <8 x i32>%B 103} 104 105define <4 x i64> @zext4(<4 x i32> %A) nounwind { 106; X32-LABEL: zext4: 107; X32: # %bb.0: 108; X32-NEXT: vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero 109; X32-NEXT: retl 110; 111; X64-LABEL: zext4: 112; X64: # %bb.0: 113; X64-NEXT: vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero 114; X64-NEXT: retq 115 %B = zext <4 x i32> %A to <4 x i64> 116 ret <4 x i64>%B 117} 118 119define <8 x i32> @zext8(<8 x i16> %A) nounwind { 120; X32-LABEL: zext8: 121; X32: # %bb.0: 122; X32-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero 123; X32-NEXT: retl 124; 125; X64-LABEL: zext8: 126; X64: # %bb.0: 127; X64-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero 128; X64-NEXT: retq 129 %B = zext <8 x i16> %A to <8 x i32> 130 ret <8 x i32>%B 131} 132 133define <8 x i32> @zext_8i8_8i32(<8 x i8> %A) nounwind { 134; X32-LABEL: zext_8i8_8i32: 135; X32: # %bb.0: 136; X32-NEXT: vpmovzxbd {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero 137; X32-NEXT: retl 138; 139; X64-LABEL: zext_8i8_8i32: 140; X64: # %bb.0: 141; X64-NEXT: vpmovzxbd {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero 142; X64-NEXT: retq 143 %B = zext <8 x i8> %A to <8 x i32> 144 ret <8 x i32>%B 145} 146 147define <16 x i16> @zext_16i8_16i16(<16 x i8> %z) { 148; X32-LABEL: zext_16i8_16i16: 149; X32: # %bb.0: 150; X32-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero 151; X32-NEXT: retl 152; 153; X64-LABEL: zext_16i8_16i16: 154; X64: # %bb.0: 155; X64-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero 156; X64-NEXT: retq 157 %t = zext <16 x i8> %z to <16 x i16> 158 ret <16 x i16> %t 159} 160 161define <16 x i16> @sext_16i8_16i16(<16 x i8> %z) { 162; X32-LABEL: sext_16i8_16i16: 163; X32: # %bb.0: 164; X32-NEXT: vpmovsxbw %xmm0, %ymm0 165; X32-NEXT: retl 166; 167; X64-LABEL: sext_16i8_16i16: 168; X64: # %bb.0: 169; X64-NEXT: vpmovsxbw %xmm0, %ymm0 170; X64-NEXT: retq 171 %t = sext <16 x i8> %z to <16 x i16> 172 ret <16 x i16> %t 173} 174 175define <16 x i8> @trunc_16i16_16i8(<16 x i16> %z) { 176; X32-LABEL: trunc_16i16_16i8: 177; X32: # %bb.0: 178; X32-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0 179; X32-NEXT: vextracti128 $1, %ymm0, %xmm1 180; X32-NEXT: vpackuswb %xmm1, %xmm0, %xmm0 181; X32-NEXT: vzeroupper 182; X32-NEXT: retl 183; 184; X64-LABEL: trunc_16i16_16i8: 185; X64: # %bb.0: 186; X64-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 187; X64-NEXT: vextracti128 $1, %ymm0, %xmm1 188; X64-NEXT: vpackuswb %xmm1, %xmm0, %xmm0 189; X64-NEXT: vzeroupper 190; X64-NEXT: retq 191 %t = trunc <16 x i16> %z to <16 x i8> 192 ret <16 x i8> %t 193} 194 195define <4 x i64> @load_sext_test1(<4 x i32> *%ptr) { 196; X32-LABEL: load_sext_test1: 197; X32: # %bb.0: 198; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 199; X32-NEXT: vpmovsxdq (%eax), %ymm0 200; X32-NEXT: retl 201; 202; X64-LABEL: load_sext_test1: 203; X64: # %bb.0: 204; X64-NEXT: vpmovsxdq (%rdi), %ymm0 205; X64-NEXT: retq 206 %X = load <4 x i32>, <4 x i32>* %ptr 207 %Y = sext <4 x i32> %X to <4 x i64> 208 ret <4 x i64>%Y 209} 210 211define <4 x i64> @load_sext_test2(<4 x i8> *%ptr) { 212; X32-LABEL: load_sext_test2: 213; X32: # %bb.0: 214; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 215; X32-NEXT: vpmovsxbq (%eax), %ymm0 216; X32-NEXT: retl 217; 218; X64-LABEL: load_sext_test2: 219; X64: # %bb.0: 220; X64-NEXT: vpmovsxbq (%rdi), %ymm0 221; X64-NEXT: retq 222 %X = load <4 x i8>, <4 x i8>* %ptr 223 %Y = sext <4 x i8> %X to <4 x i64> 224 ret <4 x i64>%Y 225} 226 227define <4 x i64> @load_sext_test3(<4 x i16> *%ptr) { 228; X32-LABEL: load_sext_test3: 229; X32: # %bb.0: 230; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 231; X32-NEXT: vpmovsxwq (%eax), %ymm0 232; X32-NEXT: retl 233; 234; X64-LABEL: load_sext_test3: 235; X64: # %bb.0: 236; X64-NEXT: vpmovsxwq (%rdi), %ymm0 237; X64-NEXT: retq 238 %X = load <4 x i16>, <4 x i16>* %ptr 239 %Y = sext <4 x i16> %X to <4 x i64> 240 ret <4 x i64>%Y 241} 242 243define <8 x i32> @load_sext_test4(<8 x i16> *%ptr) { 244; X32-LABEL: load_sext_test4: 245; X32: # %bb.0: 246; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 247; X32-NEXT: vpmovsxwd (%eax), %ymm0 248; X32-NEXT: retl 249; 250; X64-LABEL: load_sext_test4: 251; X64: # %bb.0: 252; X64-NEXT: vpmovsxwd (%rdi), %ymm0 253; X64-NEXT: retq 254 %X = load <8 x i16>, <8 x i16>* %ptr 255 %Y = sext <8 x i16> %X to <8 x i32> 256 ret <8 x i32>%Y 257} 258 259define <8 x i32> @load_sext_test5(<8 x i8> *%ptr) { 260; X32-LABEL: load_sext_test5: 261; X32: # %bb.0: 262; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 263; X32-NEXT: vpmovsxbd (%eax), %ymm0 264; X32-NEXT: retl 265; 266; X64-LABEL: load_sext_test5: 267; X64: # %bb.0: 268; X64-NEXT: vpmovsxbd (%rdi), %ymm0 269; X64-NEXT: retq 270 %X = load <8 x i8>, <8 x i8>* %ptr 271 %Y = sext <8 x i8> %X to <8 x i32> 272 ret <8 x i32>%Y 273} 274