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-prefix=X32 3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X64 4 5define <4 x i64> @test_vpaddq(<4 x i64> %i, <4 x i64> %j) nounwind readnone { 6; X32-LABEL: test_vpaddq: 7; X32: # %bb.0: 8; X32-NEXT: vpaddq %ymm1, %ymm0, %ymm0 9; X32-NEXT: retl 10; 11; X64-LABEL: test_vpaddq: 12; X64: # %bb.0: 13; X64-NEXT: vpaddq %ymm1, %ymm0, %ymm0 14; X64-NEXT: retq 15 %x = add <4 x i64> %i, %j 16 ret <4 x i64> %x 17} 18 19define <8 x i32> @test_vpaddd(<8 x i32> %i, <8 x i32> %j) nounwind readnone { 20; X32-LABEL: test_vpaddd: 21; X32: # %bb.0: 22; X32-NEXT: vpaddd %ymm1, %ymm0, %ymm0 23; X32-NEXT: retl 24; 25; X64-LABEL: test_vpaddd: 26; X64: # %bb.0: 27; X64-NEXT: vpaddd %ymm1, %ymm0, %ymm0 28; X64-NEXT: retq 29 %x = add <8 x i32> %i, %j 30 ret <8 x i32> %x 31} 32 33define <16 x i16> @test_vpaddw(<16 x i16> %i, <16 x i16> %j) nounwind readnone { 34; X32-LABEL: test_vpaddw: 35; X32: # %bb.0: 36; X32-NEXT: vpaddw %ymm1, %ymm0, %ymm0 37; X32-NEXT: retl 38; 39; X64-LABEL: test_vpaddw: 40; X64: # %bb.0: 41; X64-NEXT: vpaddw %ymm1, %ymm0, %ymm0 42; X64-NEXT: retq 43 %x = add <16 x i16> %i, %j 44 ret <16 x i16> %x 45} 46 47define <32 x i8> @test_vpaddb(<32 x i8> %i, <32 x i8> %j) nounwind readnone { 48; X32-LABEL: test_vpaddb: 49; X32: # %bb.0: 50; X32-NEXT: vpaddb %ymm1, %ymm0, %ymm0 51; X32-NEXT: retl 52; 53; X64-LABEL: test_vpaddb: 54; X64: # %bb.0: 55; X64-NEXT: vpaddb %ymm1, %ymm0, %ymm0 56; X64-NEXT: retq 57 %x = add <32 x i8> %i, %j 58 ret <32 x i8> %x 59} 60 61define <4 x i64> @test_vpsubq(<4 x i64> %i, <4 x i64> %j) nounwind readnone { 62; X32-LABEL: test_vpsubq: 63; X32: # %bb.0: 64; X32-NEXT: vpsubq %ymm1, %ymm0, %ymm0 65; X32-NEXT: retl 66; 67; X64-LABEL: test_vpsubq: 68; X64: # %bb.0: 69; X64-NEXT: vpsubq %ymm1, %ymm0, %ymm0 70; X64-NEXT: retq 71 %x = sub <4 x i64> %i, %j 72 ret <4 x i64> %x 73} 74 75define <8 x i32> @test_vpsubd(<8 x i32> %i, <8 x i32> %j) nounwind readnone { 76; X32-LABEL: test_vpsubd: 77; X32: # %bb.0: 78; X32-NEXT: vpsubd %ymm1, %ymm0, %ymm0 79; X32-NEXT: retl 80; 81; X64-LABEL: test_vpsubd: 82; X64: # %bb.0: 83; X64-NEXT: vpsubd %ymm1, %ymm0, %ymm0 84; X64-NEXT: retq 85 %x = sub <8 x i32> %i, %j 86 ret <8 x i32> %x 87} 88 89define <16 x i16> @test_vpsubw(<16 x i16> %i, <16 x i16> %j) nounwind readnone { 90; X32-LABEL: test_vpsubw: 91; X32: # %bb.0: 92; X32-NEXT: vpsubw %ymm1, %ymm0, %ymm0 93; X32-NEXT: retl 94; 95; X64-LABEL: test_vpsubw: 96; X64: # %bb.0: 97; X64-NEXT: vpsubw %ymm1, %ymm0, %ymm0 98; X64-NEXT: retq 99 %x = sub <16 x i16> %i, %j 100 ret <16 x i16> %x 101} 102 103define <32 x i8> @test_vpsubb(<32 x i8> %i, <32 x i8> %j) nounwind readnone { 104; X32-LABEL: test_vpsubb: 105; X32: # %bb.0: 106; X32-NEXT: vpsubb %ymm1, %ymm0, %ymm0 107; X32-NEXT: retl 108; 109; X64-LABEL: test_vpsubb: 110; X64: # %bb.0: 111; X64-NEXT: vpsubb %ymm1, %ymm0, %ymm0 112; X64-NEXT: retq 113 %x = sub <32 x i8> %i, %j 114 ret <32 x i8> %x 115} 116 117define <8 x i32> @test_vpmulld(<8 x i32> %i, <8 x i32> %j) nounwind readnone { 118; X32-LABEL: test_vpmulld: 119; X32: # %bb.0: 120; X32-NEXT: vpmulld %ymm1, %ymm0, %ymm0 121; X32-NEXT: retl 122; 123; X64-LABEL: test_vpmulld: 124; X64: # %bb.0: 125; X64-NEXT: vpmulld %ymm1, %ymm0, %ymm0 126; X64-NEXT: retq 127 %x = mul <8 x i32> %i, %j 128 ret <8 x i32> %x 129} 130 131define <16 x i16> @test_vpmullw(<16 x i16> %i, <16 x i16> %j) nounwind readnone { 132; X32-LABEL: test_vpmullw: 133; X32: # %bb.0: 134; X32-NEXT: vpmullw %ymm1, %ymm0, %ymm0 135; X32-NEXT: retl 136; 137; X64-LABEL: test_vpmullw: 138; X64: # %bb.0: 139; X64-NEXT: vpmullw %ymm1, %ymm0, %ymm0 140; X64-NEXT: retq 141 %x = mul <16 x i16> %i, %j 142 ret <16 x i16> %x 143} 144 145define <16 x i8> @mul_v16i8(<16 x i8> %i, <16 x i8> %j) nounwind readnone { 146; X32-LABEL: mul_v16i8: 147; X32: # %bb.0: 148; X32-NEXT: vpmovsxbw %xmm1, %ymm1 149; X32-NEXT: vpmovsxbw %xmm0, %ymm0 150; X32-NEXT: vpmullw %ymm1, %ymm0, %ymm0 151; X32-NEXT: vextracti128 $1, %ymm0, %xmm1 152; X32-NEXT: vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u> 153; X32-NEXT: vpshufb %xmm2, %xmm1, %xmm1 154; X32-NEXT: vpshufb %xmm2, %xmm0, %xmm0 155; X32-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] 156; X32-NEXT: vzeroupper 157; X32-NEXT: retl 158; 159; X64-LABEL: mul_v16i8: 160; X64: # %bb.0: 161; X64-NEXT: vpmovsxbw %xmm1, %ymm1 162; X64-NEXT: vpmovsxbw %xmm0, %ymm0 163; X64-NEXT: vpmullw %ymm1, %ymm0, %ymm0 164; X64-NEXT: vextracti128 $1, %ymm0, %xmm1 165; X64-NEXT: vmovdqa {{.*#+}} xmm2 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u> 166; X64-NEXT: vpshufb %xmm2, %xmm1, %xmm1 167; X64-NEXT: vpshufb %xmm2, %xmm0, %xmm0 168; X64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] 169; X64-NEXT: vzeroupper 170; X64-NEXT: retq 171 %x = mul <16 x i8> %i, %j 172 ret <16 x i8> %x 173} 174 175define <32 x i8> @mul_v32i8(<32 x i8> %i, <32 x i8> %j) nounwind readnone { 176; X32-LABEL: mul_v32i8: 177; X32: # %bb.0: 178; X32-NEXT: vextracti128 $1, %ymm1, %xmm2 179; X32-NEXT: vpmovsxbw %xmm2, %ymm2 180; X32-NEXT: vextracti128 $1, %ymm0, %xmm3 181; X32-NEXT: vpmovsxbw %xmm3, %ymm3 182; X32-NEXT: vpmullw %ymm2, %ymm3, %ymm2 183; X32-NEXT: vextracti128 $1, %ymm2, %xmm3 184; X32-NEXT: vmovdqa {{.*#+}} xmm4 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u> 185; X32-NEXT: vpshufb %xmm4, %xmm3, %xmm3 186; X32-NEXT: vpshufb %xmm4, %xmm2, %xmm2 187; X32-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0] 188; X32-NEXT: vpmovsxbw %xmm1, %ymm1 189; X32-NEXT: vpmovsxbw %xmm0, %ymm0 190; X32-NEXT: vpmullw %ymm1, %ymm0, %ymm0 191; X32-NEXT: vextracti128 $1, %ymm0, %xmm1 192; X32-NEXT: vpshufb %xmm4, %xmm1, %xmm1 193; X32-NEXT: vpshufb %xmm4, %xmm0, %xmm0 194; X32-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] 195; X32-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0 196; X32-NEXT: retl 197; 198; X64-LABEL: mul_v32i8: 199; X64: # %bb.0: 200; X64-NEXT: vextracti128 $1, %ymm1, %xmm2 201; X64-NEXT: vpmovsxbw %xmm2, %ymm2 202; X64-NEXT: vextracti128 $1, %ymm0, %xmm3 203; X64-NEXT: vpmovsxbw %xmm3, %ymm3 204; X64-NEXT: vpmullw %ymm2, %ymm3, %ymm2 205; X64-NEXT: vextracti128 $1, %ymm2, %xmm3 206; X64-NEXT: vmovdqa {{.*#+}} xmm4 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u> 207; X64-NEXT: vpshufb %xmm4, %xmm3, %xmm3 208; X64-NEXT: vpshufb %xmm4, %xmm2, %xmm2 209; X64-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0] 210; X64-NEXT: vpmovsxbw %xmm1, %ymm1 211; X64-NEXT: vpmovsxbw %xmm0, %ymm0 212; X64-NEXT: vpmullw %ymm1, %ymm0, %ymm0 213; X64-NEXT: vextracti128 $1, %ymm0, %xmm1 214; X64-NEXT: vpshufb %xmm4, %xmm1, %xmm1 215; X64-NEXT: vpshufb %xmm4, %xmm0, %xmm0 216; X64-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0] 217; X64-NEXT: vinserti128 $1, %xmm2, %ymm0, %ymm0 218; X64-NEXT: retq 219 %x = mul <32 x i8> %i, %j 220 ret <32 x i8> %x 221} 222 223define <4 x i64> @mul_v4i64(<4 x i64> %i, <4 x i64> %j) nounwind readnone { 224; X32-LABEL: mul_v4i64: 225; X32: # %bb.0: 226; X32-NEXT: vpsrlq $32, %ymm0, %ymm2 227; X32-NEXT: vpmuludq %ymm1, %ymm2, %ymm2 228; X32-NEXT: vpsrlq $32, %ymm1, %ymm3 229; X32-NEXT: vpmuludq %ymm3, %ymm0, %ymm3 230; X32-NEXT: vpaddq %ymm2, %ymm3, %ymm2 231; X32-NEXT: vpsllq $32, %ymm2, %ymm2 232; X32-NEXT: vpmuludq %ymm1, %ymm0, %ymm0 233; X32-NEXT: vpaddq %ymm2, %ymm0, %ymm0 234; X32-NEXT: retl 235; 236; X64-LABEL: mul_v4i64: 237; X64: # %bb.0: 238; X64-NEXT: vpsrlq $32, %ymm0, %ymm2 239; X64-NEXT: vpmuludq %ymm1, %ymm2, %ymm2 240; X64-NEXT: vpsrlq $32, %ymm1, %ymm3 241; X64-NEXT: vpmuludq %ymm3, %ymm0, %ymm3 242; X64-NEXT: vpaddq %ymm2, %ymm3, %ymm2 243; X64-NEXT: vpsllq $32, %ymm2, %ymm2 244; X64-NEXT: vpmuludq %ymm1, %ymm0, %ymm0 245; X64-NEXT: vpaddq %ymm2, %ymm0, %ymm0 246; X64-NEXT: retq 247 %x = mul <4 x i64> %i, %j 248 ret <4 x i64> %x 249} 250 251define <8 x i32> @mul_const1(<8 x i32> %x) { 252; X32-LABEL: mul_const1: 253; X32: # %bb.0: 254; X32-NEXT: vpaddd %ymm0, %ymm0, %ymm0 255; X32-NEXT: retl 256; 257; X64-LABEL: mul_const1: 258; X64: # %bb.0: 259; X64-NEXT: vpaddd %ymm0, %ymm0, %ymm0 260; X64-NEXT: retq 261 %y = mul <8 x i32> %x, <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2> 262 ret <8 x i32> %y 263} 264 265define <4 x i64> @mul_const2(<4 x i64> %x) { 266; X32-LABEL: mul_const2: 267; X32: # %bb.0: 268; X32-NEXT: vpsllq $2, %ymm0, %ymm0 269; X32-NEXT: retl 270; 271; X64-LABEL: mul_const2: 272; X64: # %bb.0: 273; X64-NEXT: vpsllq $2, %ymm0, %ymm0 274; X64-NEXT: retq 275 %y = mul <4 x i64> %x, <i64 4, i64 4, i64 4, i64 4> 276 ret <4 x i64> %y 277} 278 279define <16 x i16> @mul_const3(<16 x i16> %x) { 280; X32-LABEL: mul_const3: 281; X32: # %bb.0: 282; X32-NEXT: vpsllw $3, %ymm0, %ymm0 283; X32-NEXT: retl 284; 285; X64-LABEL: mul_const3: 286; X64: # %bb.0: 287; X64-NEXT: vpsllw $3, %ymm0, %ymm0 288; X64-NEXT: retq 289 %y = mul <16 x i16> %x, <i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8> 290 ret <16 x i16> %y 291} 292 293define <4 x i64> @mul_const4(<4 x i64> %x) { 294; X32-LABEL: mul_const4: 295; X32: # %bb.0: 296; X32-NEXT: vpxor %xmm1, %xmm1, %xmm1 297; X32-NEXT: vpsubq %ymm0, %ymm1, %ymm0 298; X32-NEXT: retl 299; 300; X64-LABEL: mul_const4: 301; X64: # %bb.0: 302; X64-NEXT: vpxor %xmm1, %xmm1, %xmm1 303; X64-NEXT: vpsubq %ymm0, %ymm1, %ymm0 304; X64-NEXT: retq 305 %y = mul <4 x i64> %x, <i64 -1, i64 -1, i64 -1, i64 -1> 306 ret <4 x i64> %y 307} 308 309define <8 x i32> @mul_const5(<8 x i32> %x) { 310; X32-LABEL: mul_const5: 311; X32: # %bb.0: 312; X32-NEXT: vxorps %xmm0, %xmm0, %xmm0 313; X32-NEXT: retl 314; 315; X64-LABEL: mul_const5: 316; X64: # %bb.0: 317; X64-NEXT: vxorps %xmm0, %xmm0, %xmm0 318; X64-NEXT: retq 319 %y = mul <8 x i32> %x, <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> 320 ret <8 x i32> %y 321} 322 323define <8 x i32> @mul_const6(<8 x i32> %x) { 324; X32-LABEL: mul_const6: 325; X32: # %bb.0: 326; X32-NEXT: vpmulld {{\.LCPI.*}}, %ymm0, %ymm0 327; X32-NEXT: retl 328; 329; X64-LABEL: mul_const6: 330; X64: # %bb.0: 331; X64-NEXT: vpmulld {{.*}}(%rip), %ymm0, %ymm0 332; X64-NEXT: retq 333 %y = mul <8 x i32> %x, <i32 0, i32 0, i32 0, i32 2, i32 0, i32 2, i32 0, i32 0> 334 ret <8 x i32> %y 335} 336 337define <8 x i64> @mul_const7(<8 x i64> %x) { 338; X32-LABEL: mul_const7: 339; X32: # %bb.0: 340; X32-NEXT: vpaddq %ymm0, %ymm0, %ymm0 341; X32-NEXT: vpaddq %ymm1, %ymm1, %ymm1 342; X32-NEXT: retl 343; 344; X64-LABEL: mul_const7: 345; X64: # %bb.0: 346; X64-NEXT: vpaddq %ymm0, %ymm0, %ymm0 347; X64-NEXT: vpaddq %ymm1, %ymm1, %ymm1 348; X64-NEXT: retq 349 %y = mul <8 x i64> %x, <i64 2, i64 2, i64 2, i64 2, i64 2, i64 2, i64 2, i64 2> 350 ret <8 x i64> %y 351} 352 353define <8 x i16> @mul_const8(<8 x i16> %x) { 354; X32-LABEL: mul_const8: 355; X32: # %bb.0: 356; X32-NEXT: vpsllw $3, %xmm0, %xmm0 357; X32-NEXT: retl 358; 359; X64-LABEL: mul_const8: 360; X64: # %bb.0: 361; X64-NEXT: vpsllw $3, %xmm0, %xmm0 362; X64-NEXT: retq 363 %y = mul <8 x i16> %x, <i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8> 364 ret <8 x i16> %y 365} 366 367define <8 x i32> @mul_const9(<8 x i32> %x) { 368; X32-LABEL: mul_const9: 369; X32: # %bb.0: 370; X32-NEXT: movl $2, %eax 371; X32-NEXT: vmovd %eax, %xmm1 372; X32-NEXT: vpmulld %ymm1, %ymm0, %ymm0 373; X32-NEXT: retl 374; 375; X64-LABEL: mul_const9: 376; X64: # %bb.0: 377; X64-NEXT: movl $2, %eax 378; X64-NEXT: vmovd %eax, %xmm1 379; X64-NEXT: vpmulld %ymm1, %ymm0, %ymm0 380; X64-NEXT: retq 381 %y = mul <8 x i32> %x, <i32 2, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> 382 ret <8 x i32> %y 383} 384 385; %x * 0x01010101 386define <4 x i32> @mul_const10(<4 x i32> %x) { 387; X32-LABEL: mul_const10: 388; X32: # %bb.0: 389; X32-NEXT: vpbroadcastd {{.*#+}} xmm1 = [16843009,16843009,16843009,16843009] 390; X32-NEXT: vpmulld %xmm1, %xmm0, %xmm0 391; X32-NEXT: retl 392; 393; X64-LABEL: mul_const10: 394; X64: # %bb.0: 395; X64-NEXT: vpbroadcastd {{.*#+}} xmm1 = [16843009,16843009,16843009,16843009] 396; X64-NEXT: vpmulld %xmm1, %xmm0, %xmm0 397; X64-NEXT: retq 398 %m = mul <4 x i32> %x, <i32 16843009, i32 16843009, i32 16843009, i32 16843009> 399 ret <4 x i32> %m 400} 401 402; %x * 0x80808080 403define <4 x i32> @mul_const11(<4 x i32> %x) { 404; X32-LABEL: mul_const11: 405; X32: # %bb.0: 406; X32-NEXT: vpbroadcastd {{.*#+}} xmm1 = [2155905152,2155905152,2155905152,2155905152] 407; X32-NEXT: vpmulld %xmm1, %xmm0, %xmm0 408; X32-NEXT: retl 409; 410; X64-LABEL: mul_const11: 411; X64: # %bb.0: 412; X64-NEXT: vpbroadcastd {{.*#+}} xmm1 = [2155905152,2155905152,2155905152,2155905152] 413; X64-NEXT: vpmulld %xmm1, %xmm0, %xmm0 414; X64-NEXT: retq 415 %m = mul <4 x i32> %x, <i32 2155905152, i32 2155905152, i32 2155905152, i32 2155905152> 416 ret <4 x i32> %m 417} 418