217 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE3; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX14; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX25; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX,AVX5126 7; Combine tests involving SSE41 target shuffles (BLEND,INSERTPS,MOVZX)8 9declare <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8>, <16 x i8>)10declare <4 x float> @llvm.x86.sse41.insertps(<4 x float>, <4 x float>, i8)11 12define <16 x i8> @combine_vpshufb_as_movzx(<16 x i8> %a0) {13; SSE-LABEL: combine_vpshufb_as_movzx:14; SSE: # %bb.0:15; SSE-NEXT: pmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero16; SSE-NEXT: retq17;18; AVX-LABEL: combine_vpshufb_as_movzx:19; AVX: # %bb.0:20; AVX-NEXT: vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero21; AVX-NEXT: retq22 %res0 = call <16 x i8> @llvm.x86.ssse3.pshuf.b.128(<16 x i8> %a0, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 -1, i8 -1, i8 -1, i8 -1, i8 undef, i8 undef, i8 undef, i8 undef, i8 -1, i8 -1, i8 -1, i8 -1>)23 ret <16 x i8> %res024}25 26define <4 x i32> @combine_blend_of_permutes_v4i32(<2 x i64> %a0, <2 x i64> %a1) {27; SSE-LABEL: combine_blend_of_permutes_v4i32:28; SSE: # %bb.0:29; SSE-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]30; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]31; SSE-NEXT: retq32;33; AVX1-LABEL: combine_blend_of_permutes_v4i32:34; AVX1: # %bb.0:35; AVX1-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]36; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[2,3,0,1]37; AVX1-NEXT: retq38;39; AVX2-LABEL: combine_blend_of_permutes_v4i32:40; AVX2: # %bb.0:41; AVX2-NEXT: vblendps {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]42; AVX2-NEXT: vshufps {{.*#+}} xmm0 = xmm0[2,3,0,1]43; AVX2-NEXT: retq44;45; AVX512-LABEL: combine_blend_of_permutes_v4i32:46; AVX512: # %bb.0:47; AVX512-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm148; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm049; AVX512-NEXT: vpmovsxbd {{.*#+}} xmm2 = [2,19,0,17]50; AVX512-NEXT: vpermt2d %zmm1, %zmm2, %zmm051; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm052; AVX512-NEXT: vzeroupper53; AVX512-NEXT: retq54 %s0 = shufflevector <2 x i64> %a0, <2 x i64> undef, <2 x i32> <i32 1, i32 0>55 %s1 = shufflevector <2 x i64> %a1, <2 x i64> undef, <2 x i32> <i32 1, i32 0>56 %x0 = bitcast <2 x i64> %s0 to <4 x i32>57 %x1 = bitcast <2 x i64> %s1 to <4 x i32>58 %r = shufflevector <4 x i32> %x0, <4 x i32> %x1, <4 x i32> <i32 0, i32 5, i32 2, i32 7>59 ret <4 x i32> %r60}61 62define <4 x float> @freeze_insertps(<4 x float> %a0, <4 x float> %a1) {63; SSE-LABEL: freeze_insertps:64; SSE: # %bb.0:65; SSE-NEXT: movaps %xmm1, %xmm066; SSE-NEXT: retq67;68; AVX-LABEL: freeze_insertps:69; AVX: # %bb.0:70; AVX-NEXT: vmovaps %xmm1, %xmm071; AVX-NEXT: retq72 %s0 = call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a0, <4 x float> %a1, i8 16)73 %f0 = freeze <4 x float> %s074 %s1 = call <4 x float> @llvm.x86.sse41.insertps(<4 x float> %a1, <4 x float> %f0, i8 64)75 ret <4 x float> %s176}77 78define <16 x i8> @PR50049(ptr %p1, ptr %p2) {79; SSE-LABEL: PR50049:80; SSE: # %bb.0:81; SSE-NEXT: movdqa (%rdi), %xmm282; SSE-NEXT: movdqa 16(%rdi), %xmm383; SSE-NEXT: movdqa 32(%rdi), %xmm084; SSE-NEXT: movdqa (%rsi), %xmm485; SSE-NEXT: movdqa 16(%rsi), %xmm586; SSE-NEXT: movdqa 32(%rsi), %xmm187; SSE-NEXT: movdqa {{.*#+}} xmm6 = [128,128,128,128,128,128,2,5,8,11,14,u,u,u,u,u]88; SSE-NEXT: pshufb %xmm6, %xmm389; SSE-NEXT: movdqa {{.*#+}} xmm7 = [0,3,6,9,12,15,128,128,128,128,128,u,u,u,u,u]90; SSE-NEXT: pshufb %xmm7, %xmm291; SSE-NEXT: por %xmm3, %xmm292; SSE-NEXT: movdqa {{.*#+}} xmm3 = [0,1,2,3,4,5,6,7,8,9,10,128,128,128,128,128]93; SSE-NEXT: pshufb %xmm3, %xmm294; SSE-NEXT: movdqa {{.*#+}} xmm8 = [128,128,128,128,128,128,128,128,128,128,128,1,4,7,10,13]95; SSE-NEXT: pshufb %xmm8, %xmm096; SSE-NEXT: por %xmm2, %xmm097; SSE-NEXT: pshufb %xmm6, %xmm598; SSE-NEXT: pshufb %xmm7, %xmm499; SSE-NEXT: por %xmm5, %xmm4100; SSE-NEXT: pshufb %xmm3, %xmm4101; SSE-NEXT: pshufb %xmm8, %xmm1102; SSE-NEXT: por %xmm4, %xmm1103; SSE-NEXT: movdqa %xmm0, %xmm2104; SSE-NEXT: pmullw %xmm1, %xmm2105; SSE-NEXT: pmovzxbw {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]106; SSE-NEXT: pand %xmm3, %xmm2107; SSE-NEXT: pandn %xmm1, %xmm3108; SSE-NEXT: pmaddubsw %xmm3, %xmm0109; SSE-NEXT: psllw $8, %xmm0110; SSE-NEXT: por %xmm2, %xmm0111; SSE-NEXT: retq112;113; AVX1-LABEL: PR50049:114; AVX1: # %bb.0:115; AVX1-NEXT: vmovdqa (%rdi), %xmm0116; AVX1-NEXT: vmovdqa 16(%rdi), %xmm1117; AVX1-NEXT: vmovdqa 32(%rdi), %xmm2118; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [1,4,7,10,13,128,128,128,128,128,128,u,u,u,u,u]119; AVX1-NEXT: vpshufb %xmm3, %xmm2, %xmm2120; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [128,128,128,128,128,0,3,6,9,12,15,u,u,u,u,u]121; AVX1-NEXT: vpshufb %xmm4, %xmm0, %xmm0122; AVX1-NEXT: vpor %xmm2, %xmm0, %xmm0123; AVX1-NEXT: vmovdqa (%rsi), %xmm2124; AVX1-NEXT: vmovdqa 16(%rsi), %xmm5125; AVX1-NEXT: vmovdqa 32(%rsi), %xmm6126; AVX1-NEXT: vpshufb %xmm3, %xmm6, %xmm3127; AVX1-NEXT: vpshufb %xmm4, %xmm2, %xmm2128; AVX1-NEXT: vpor %xmm3, %xmm2, %xmm2129; AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [5,6,7,8,9,10,128,128,128,128,128,0,1,2,3,4]130; AVX1-NEXT: vpshufb %xmm3, %xmm2, %xmm2131; AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [128,128,128,128,128,128,2,5,8,11,14,128,128,128,128,128]132; AVX1-NEXT: vpshufb %xmm4, %xmm5, %xmm5133; AVX1-NEXT: vpor %xmm5, %xmm2, %xmm2134; AVX1-NEXT: vpshufb %xmm3, %xmm0, %xmm0135; AVX1-NEXT: vpshufb %xmm4, %xmm1, %xmm1136; AVX1-NEXT: vpor %xmm1, %xmm0, %xmm0137; AVX1-NEXT: vpmullw %xmm2, %xmm0, %xmm1138; AVX1-NEXT: vbroadcastss {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]139; AVX1-NEXT: vpand %xmm3, %xmm1, %xmm1140; AVX1-NEXT: vpandn %xmm2, %xmm3, %xmm2141; AVX1-NEXT: vpmaddubsw %xmm2, %xmm0, %xmm0142; AVX1-NEXT: vpsllw $8, %xmm0, %xmm0143; AVX1-NEXT: vpor %xmm0, %xmm1, %xmm0144; AVX1-NEXT: retq145;146; AVX2-LABEL: PR50049:147; AVX2: # %bb.0:148; AVX2-NEXT: vmovdqa (%rdi), %xmm0149; AVX2-NEXT: vmovdqa 16(%rdi), %xmm1150; AVX2-NEXT: vmovdqa 32(%rdi), %xmm2151; AVX2-NEXT: vmovdqa {{.*#+}} xmm3 = [1,4,7,10,13,128,128,128,128,128,128,u,u,u,u,u]152; AVX2-NEXT: vpshufb %xmm3, %xmm2, %xmm2153; AVX2-NEXT: vmovdqa {{.*#+}} xmm4 = [128,128,128,128,128,0,3,6,9,12,15,u,u,u,u,u]154; AVX2-NEXT: vpshufb %xmm4, %xmm0, %xmm0155; AVX2-NEXT: vpor %xmm2, %xmm0, %xmm0156; AVX2-NEXT: vmovdqa (%rsi), %xmm2157; AVX2-NEXT: vmovdqa 16(%rsi), %xmm5158; AVX2-NEXT: vmovdqa 32(%rsi), %xmm6159; AVX2-NEXT: vpshufb %xmm3, %xmm6, %xmm3160; AVX2-NEXT: vpshufb %xmm4, %xmm2, %xmm2161; AVX2-NEXT: vpor %xmm3, %xmm2, %xmm2162; AVX2-NEXT: vmovdqa {{.*#+}} xmm3 = [5,6,7,8,9,10,128,128,128,128,128,0,1,2,3,4]163; AVX2-NEXT: vpshufb %xmm3, %xmm2, %xmm2164; AVX2-NEXT: vmovdqa {{.*#+}} xmm4 = [128,128,128,128,128,128,2,5,8,11,14,128,128,128,128,128]165; AVX2-NEXT: vpshufb %xmm4, %xmm5, %xmm5166; AVX2-NEXT: vpor %xmm5, %xmm2, %xmm2167; AVX2-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero168; AVX2-NEXT: vpshufb %xmm3, %xmm0, %xmm0169; AVX2-NEXT: vpshufb %xmm4, %xmm1, %xmm1170; AVX2-NEXT: vpor %xmm1, %xmm0, %xmm0171; AVX2-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],zero172; AVX2-NEXT: vpmullw %ymm2, %ymm0, %ymm0173; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0174; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1175; AVX2-NEXT: vpackuswb %xmm1, %xmm0, %xmm0176; AVX2-NEXT: vzeroupper177; AVX2-NEXT: retq178;179; AVX512-LABEL: PR50049:180; AVX512: # %bb.0:181; AVX512-NEXT: vmovdqa (%rdi), %xmm0182; AVX512-NEXT: vmovdqa 16(%rdi), %xmm1183; AVX512-NEXT: vmovdqa 32(%rdi), %xmm2184; AVX512-NEXT: vmovdqa {{.*#+}} xmm3 = [1,4,7,10,13,128,128,128,128,128,128,u,u,u,u,u]185; AVX512-NEXT: vpshufb %xmm3, %xmm2, %xmm2186; AVX512-NEXT: vmovdqa {{.*#+}} xmm4 = [128,128,128,128,128,0,3,6,9,12,15,u,u,u,u,u]187; AVX512-NEXT: vpshufb %xmm4, %xmm0, %xmm0188; AVX512-NEXT: vpor %xmm2, %xmm0, %xmm0189; AVX512-NEXT: vmovdqa (%rsi), %xmm2190; AVX512-NEXT: vmovdqa 16(%rsi), %xmm5191; AVX512-NEXT: vmovdqa 32(%rsi), %xmm6192; AVX512-NEXT: vpshufb %xmm3, %xmm6, %xmm3193; AVX512-NEXT: vpshufb %xmm4, %xmm2, %xmm2194; AVX512-NEXT: vpor %xmm3, %xmm2, %xmm2195; AVX512-NEXT: vmovdqa {{.*#+}} xmm3 = [5,6,7,8,9,10,128,128,128,128,128,0,1,2,3,4]196; AVX512-NEXT: vpshufb %xmm3, %xmm2, %xmm2197; AVX512-NEXT: vmovdqa {{.*#+}} xmm4 = [128,128,128,128,128,128,2,5,8,11,14,128,128,128,128,128]198; AVX512-NEXT: vpshufb %xmm4, %xmm5, %xmm5199; AVX512-NEXT: vpor %xmm5, %xmm2, %xmm2200; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero201; AVX512-NEXT: vpshufb %xmm3, %xmm0, %xmm0202; AVX512-NEXT: vpshufb %xmm4, %xmm1, %xmm1203; AVX512-NEXT: vpor %xmm1, %xmm0, %xmm0204; AVX512-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],zero205; AVX512-NEXT: vpmullw %ymm2, %ymm0, %ymm0206; AVX512-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero207; AVX512-NEXT: vpmovdb %zmm0, %xmm0208; AVX512-NEXT: vzeroupper209; AVX512-NEXT: retq210 %x1 = load <48 x i8>, ptr %p1, align 16211 %x2 = load <48 x i8>, ptr %p2, align 16212 %s1 = shufflevector <48 x i8> %x1, <48 x i8> poison, <16 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45>213 %s2 = shufflevector <48 x i8> %x2, <48 x i8> poison, <16 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45>214 %r = mul <16 x i8> %s1, %s2215 ret <16 x i8> %r216}217