235 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+prefer-256-bit | FileCheck %s --check-prefix=AVX256VL3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,-prefer-256-bit | FileCheck %s --check-prefixes=AVX512NOBW,AVX512VL4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+prefer-256-bit | FileCheck %s --check-prefix=AVX256VLBW5; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,-prefer-256-bit | FileCheck %s --check-prefix=AVX512VLBW6; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+prefer-256-bit | FileCheck %s --check-prefixes=AVX512NOBW,AVX512F7; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,-prefer-256-bit | FileCheck %s --check-prefixes=AVX512NOBW,AVX512F8; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+prefer-256-bit | FileCheck %s --check-prefix=AVX512BW9; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,-prefer-256-bit | FileCheck %s --check-prefix=AVX512BW10 11define <16 x i1> @shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0(ptr %a, ptr %b) {12; AVX256VL-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:13; AVX256VL: # %bb.0:14; AVX256VL-NEXT: vmovdqa (%rdi), %ymm015; AVX256VL-NEXT: vmovdqa (%rsi), %ymm116; AVX256VL-NEXT: vptestnmd %ymm0, %ymm0, %k117; AVX256VL-NEXT: vptestnmd %ymm1, %ymm1, %k218; AVX256VL-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm019; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm1 {%k2} {z}20; AVX256VL-NEXT: vpmovdw %ymm1, %xmm121; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm2 {%k1} {z}22; AVX256VL-NEXT: vpmovdw %ymm2, %xmm223; AVX256VL-NEXT: vpblendw {{.*#+}} xmm3 = xmm2[0,1],xmm1[2],xmm2[3],xmm1[4],xmm2[5,6,7]24; AVX256VL-NEXT: vpshufb {{.*#+}} xmm3 = xmm3[6,7,12,13,4,5,8,9,6,7,14,15,14,15,0,1]25; AVX256VL-NEXT: vpmovsxwd %xmm3, %ymm326; AVX256VL-NEXT: vpslld $31, %ymm3, %ymm327; AVX256VL-NEXT: vptestmd %ymm3, %ymm3, %k128; AVX256VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[0,2,1,3]29; AVX256VL-NEXT: vpshufb {{.*#+}} xmm2 = xmm2[6,7,12,13,2,3,u,u,6,7,u,u,14,15,0,1]30; AVX256VL-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1,2],xmm1[3],xmm2[4],xmm1[5],xmm2[6,7]31; AVX256VL-NEXT: vpmovsxwd %xmm1, %ymm132; AVX256VL-NEXT: vpslld $31, %ymm1, %ymm133; AVX256VL-NEXT: vptestmd %ymm1, %ymm1, %k034; AVX256VL-NEXT: kunpckbw %k1, %k0, %k035; AVX256VL-NEXT: kshiftrw $8, %k0, %k236; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm1 {%k2} {z}37; AVX256VL-NEXT: vpmovdw %ymm1, %xmm138; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}39; AVX256VL-NEXT: vpmovdw %ymm0, %xmm040; AVX256VL-NEXT: vpacksswb %xmm1, %xmm0, %xmm041; AVX256VL-NEXT: vzeroupper42; AVX256VL-NEXT: retq43;44; AVX512VL-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:45; AVX512VL: # %bb.0:46; AVX512VL-NEXT: vmovdqa (%rdi), %ymm047; AVX512VL-NEXT: vmovdqa (%rsi), %ymm148; AVX512VL-NEXT: vptestnmd %ymm0, %ymm0, %k149; AVX512VL-NEXT: vptestnmd %ymm1, %ymm1, %k250; AVX512VL-NEXT: vpternlogd {{.*#+}} zmm0 {%k2} {z} = -151; AVX512VL-NEXT: vpternlogd {{.*#+}} zmm1 {%k1} {z} = -152; AVX512VL-NEXT: vpmovsxbd {{.*#+}} zmm2 = [3,6,18,20,3,7,7,0,3,6,1,21,3,19,7,0]53; AVX512VL-NEXT: vpermi2d %zmm0, %zmm1, %zmm254; AVX512VL-NEXT: vptestmd %zmm2, %zmm2, %k155; AVX512VL-NEXT: vpternlogd {{.*#+}} zmm0 {%k1} {z} = -156; AVX512VL-NEXT: vpmovdb %zmm0, %xmm057; AVX512VL-NEXT: vzeroupper58; AVX512VL-NEXT: retq59;60; AVX256VLBW-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:61; AVX256VLBW: # %bb.0:62; AVX256VLBW-NEXT: vmovdqa (%rdi), %ymm063; AVX256VLBW-NEXT: vmovdqa (%rsi), %ymm164; AVX256VLBW-NEXT: vptestnmd %ymm0, %ymm0, %k065; AVX256VLBW-NEXT: vptestnmd %ymm1, %ymm1, %k166; AVX256VLBW-NEXT: vpmovm2w %k1, %ymm067; AVX256VLBW-NEXT: vpmovm2w %k0, %ymm168; AVX256VLBW-NEXT: vpmovsxbw {{.*#+}} ymm2 = [3,6,18,20,3,7,7,0,3,6,1,21,3,19,7,0]69; AVX256VLBW-NEXT: vpermi2w %ymm0, %ymm1, %ymm270; AVX256VLBW-NEXT: vpmovw2m %ymm2, %k071; AVX256VLBW-NEXT: vpmovm2b %k0, %xmm072; AVX256VLBW-NEXT: vzeroupper73; AVX256VLBW-NEXT: retq74;75; AVX512VLBW-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:76; AVX512VLBW: # %bb.0:77; AVX512VLBW-NEXT: vmovdqa (%rdi), %ymm078; AVX512VLBW-NEXT: vmovdqa (%rsi), %ymm179; AVX512VLBW-NEXT: vptestnmd %ymm0, %ymm0, %k180; AVX512VLBW-NEXT: vptestnmd %ymm1, %ymm1, %k281; AVX512VLBW-NEXT: vpternlogd {{.*#+}} zmm0 {%k2} {z} = -182; AVX512VLBW-NEXT: vpternlogd {{.*#+}} zmm1 {%k1} {z} = -183; AVX512VLBW-NEXT: vpmovsxbd {{.*#+}} zmm2 = [3,6,18,20,3,7,7,0,3,6,1,21,3,19,7,0]84; AVX512VLBW-NEXT: vpermi2d %zmm0, %zmm1, %zmm285; AVX512VLBW-NEXT: vptestmd %zmm2, %zmm2, %k086; AVX512VLBW-NEXT: vpmovm2b %k0, %xmm087; AVX512VLBW-NEXT: vzeroupper88; AVX512VLBW-NEXT: retq89;90; AVX512F-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:91; AVX512F: # %bb.0:92; AVX512F-NEXT: vmovdqa (%rdi), %ymm093; AVX512F-NEXT: vmovdqa (%rsi), %ymm194; AVX512F-NEXT: vptestnmd %zmm0, %zmm0, %k195; AVX512F-NEXT: vptestnmd %zmm1, %zmm1, %k296; AVX512F-NEXT: vpternlogd {{.*#+}} zmm0 {%k2} {z} = -197; AVX512F-NEXT: vpternlogd {{.*#+}} zmm1 {%k1} {z} = -198; AVX512F-NEXT: vpmovsxbd {{.*#+}} zmm2 = [3,6,18,20,3,7,7,0,3,6,1,21,3,19,7,0]99; AVX512F-NEXT: vpermi2d %zmm0, %zmm1, %zmm2100; AVX512F-NEXT: vptestmd %zmm2, %zmm2, %k1101; AVX512F-NEXT: vpternlogd {{.*#+}} zmm0 {%k1} {z} = -1102; AVX512F-NEXT: vpmovdb %zmm0, %xmm0103; AVX512F-NEXT: vzeroupper104; AVX512F-NEXT: retq105;106; AVX512BW-LABEL: shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:107; AVX512BW: # %bb.0:108; AVX512BW-NEXT: vmovdqa (%rdi), %ymm0109; AVX512BW-NEXT: vmovdqa (%rsi), %ymm1110; AVX512BW-NEXT: vptestnmd %zmm0, %zmm0, %k1111; AVX512BW-NEXT: vptestnmd %zmm1, %zmm1, %k2112; AVX512BW-NEXT: vpternlogd {{.*#+}} zmm0 {%k2} {z} = -1113; AVX512BW-NEXT: vpternlogd {{.*#+}} zmm1 {%k1} {z} = -1114; AVX512BW-NEXT: vpmovsxbd {{.*#+}} zmm2 = [3,6,18,20,3,7,7,0,3,6,1,21,3,19,7,0]115; AVX512BW-NEXT: vpermi2d %zmm0, %zmm1, %zmm2116; AVX512BW-NEXT: vptestmd %zmm2, %zmm2, %k0117; AVX512BW-NEXT: vpmovm2b %k0, %zmm0118; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0119; AVX512BW-NEXT: vzeroupper120; AVX512BW-NEXT: retq121 122 %a1 = load <8 x i32>, ptr %a123 %b1 = load <8 x i32>, ptr %b124 %a2 = icmp eq <8 x i32> %a1, zeroinitializer125 %b2 = icmp eq <8 x i32> %b1, zeroinitializer126 %c = shufflevector <8 x i1> %a2, <8 x i1> %b2, <16 x i32> <i32 3, i32 6, i32 10, i32 12, i32 3, i32 7, i32 7, i32 0, i32 3, i32 6, i32 1, i32 13, i32 3, i32 11, i32 7, i32 0>127 ret <16 x i1> %c128}129 130define <32 x i1> @shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0(<32 x i8> %a) {131; AVX256VL-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:132; AVX256VL: # %bb.0:133; AVX256VL-NEXT: vpxor %xmm1, %xmm1, %xmm1134; AVX256VL-NEXT: vpcmpeqb %ymm1, %ymm0, %ymm0135; AVX256VL-NEXT: vextracti128 $1, %ymm0, %xmm1136; AVX256VL-NEXT: vpmovsxbd %xmm1, %ymm1137; AVX256VL-NEXT: vptestmd %ymm1, %ymm1, %k1138; AVX256VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]139; AVX256VL-NEXT: vpmovsxbd %xmm1, %ymm1140; AVX256VL-NEXT: vptestmd %ymm1, %ymm1, %k2141; AVX256VL-NEXT: vpmovsxbd %xmm0, %ymm0142; AVX256VL-NEXT: vptestmd %ymm0, %ymm0, %k3143; AVX256VL-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0144; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm1 {%k3} {z}145; AVX256VL-NEXT: vpmovdw %ymm1, %xmm1146; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm2 {%k2} {z}147; AVX256VL-NEXT: vpmovdw %ymm2, %xmm2148; AVX256VL-NEXT: vinserti128 $1, %xmm2, %ymm1, %ymm1149; AVX256VL-NEXT: vpermq {{.*#+}} ymm2 = ymm1[2,3,0,1]150; AVX256VL-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7]151; AVX256VL-NEXT: vpshufb {{.*#+}} ymm1 = ymm1[6,7,12,13],zero,zero,ymm1[8,9,6,7,14,15,14,15,0,1,22,23,28,29,18,19,26,27,22,23],zero,zero,ymm1[30,31,16,17]152; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm2 {%k1} {z}153; AVX256VL-NEXT: vpmovdw %ymm2, %xmm2154; AVX256VL-NEXT: vpermq {{.*#+}} ymm2 = ymm2[1,1,1,1]155; AVX256VL-NEXT: vpternlogq {{.*#+}} ymm2 = (ymm2 & ~mem) | ymm1156; AVX256VL-NEXT: vpmovsxwd %xmm2, %ymm1157; AVX256VL-NEXT: vpslld $31, %ymm1, %ymm1158; AVX256VL-NEXT: vptestmd %ymm1, %ymm1, %k1159; AVX256VL-NEXT: vextracti128 $1, %ymm2, %xmm1160; AVX256VL-NEXT: vpmovsxwd %xmm1, %ymm1161; AVX256VL-NEXT: vpslld $31, %ymm1, %ymm1162; AVX256VL-NEXT: vptestmd %ymm1, %ymm1, %k0163; AVX256VL-NEXT: kunpckbw %k1, %k0, %k0164; AVX256VL-NEXT: kshiftrw $8, %k0, %k2165; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm1 {%k2} {z}166; AVX256VL-NEXT: vpmovdw %ymm1, %xmm1167; AVX256VL-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}168; AVX256VL-NEXT: vpmovdw %ymm0, %xmm0169; AVX256VL-NEXT: vpacksswb %xmm1, %xmm0, %xmm0170; AVX256VL-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0171; AVX256VL-NEXT: retq172;173; AVX512NOBW-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:174; AVX512NOBW: # %bb.0:175; AVX512NOBW-NEXT: vpxor %xmm1, %xmm1, %xmm1176; AVX512NOBW-NEXT: vpcmpeqb %ymm1, %ymm0, %ymm0177; AVX512NOBW-NEXT: vpmovsxbd %xmm0, %zmm1178; AVX512NOBW-NEXT: vptestmd %zmm1, %zmm1, %k1179; AVX512NOBW-NEXT: vextracti128 $1, %ymm0, %xmm0180; AVX512NOBW-NEXT: vpmovsxbd %xmm0, %zmm0181; AVX512NOBW-NEXT: vptestmd %zmm0, %zmm0, %k2182; AVX512NOBW-NEXT: vpternlogd {{.*#+}} zmm0 {%k2} {z} = -1183; AVX512NOBW-NEXT: vpternlogd {{.*#+}} zmm1 {%k1} {z} = -1184; AVX512NOBW-NEXT: vpmovsxbd {{.*#+}} zmm2 = [3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0]185; AVX512NOBW-NEXT: vpermi2d %zmm0, %zmm1, %zmm2186; AVX512NOBW-NEXT: vptestmd %zmm2, %zmm2, %k1187; AVX512NOBW-NEXT: vpternlogd {{.*#+}} zmm0 {%k1} {z} = -1188; AVX512NOBW-NEXT: vpmovdb %zmm0, %xmm0189; AVX512NOBW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0190; AVX512NOBW-NEXT: retq191;192; AVX256VLBW-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:193; AVX256VLBW: # %bb.0:194; AVX256VLBW-NEXT: vptestnmb %ymm0, %ymm0, %k0195; AVX256VLBW-NEXT: vpmovm2b %k0, %ymm0196; AVX256VLBW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,2,3]197; AVX256VLBW-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[1,1,1,1,5,5,5,5]198; AVX256VLBW-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]199; AVX256VLBW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[3,6,u,12,3,7,7,0,3,6,1,13,3,u,7,0,19,22,u,28,19,23,23,16,19,22,17,29,19,u,23,16]200; AVX256VLBW-NEXT: movl $537141252, %eax # imm = 0x20042004201; AVX256VLBW-NEXT: kmovd %eax, %k1202; AVX256VLBW-NEXT: vmovdqu8 %ymm1, %ymm0 {%k1}203; AVX256VLBW-NEXT: vpmovb2m %ymm0, %k0204; AVX256VLBW-NEXT: vpmovm2b %k0, %ymm0205; AVX256VLBW-NEXT: retq206;207; AVX512VLBW-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:208; AVX512VLBW: # %bb.0:209; AVX512VLBW-NEXT: vptestnmb %ymm0, %ymm0, %k0210; AVX512VLBW-NEXT: vpmovm2w %k0, %zmm0211; AVX512VLBW-NEXT: vbroadcasti64x4 {{.*#+}} zmm1 = [3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0,3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0]212; AVX512VLBW-NEXT: # zmm1 = mem[0,1,2,3,0,1,2,3]213; AVX512VLBW-NEXT: vpermw %zmm0, %zmm1, %zmm0214; AVX512VLBW-NEXT: vpmovw2m %zmm0, %k0215; AVX512VLBW-NEXT: vpmovm2b %k0, %ymm0216; AVX512VLBW-NEXT: retq217;218; AVX512BW-LABEL: shuf32i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0:219; AVX512BW: # %bb.0:220; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0221; AVX512BW-NEXT: vptestnmb %zmm0, %zmm0, %k0222; AVX512BW-NEXT: vpmovm2w %k0, %zmm0223; AVX512BW-NEXT: vbroadcasti64x4 {{.*#+}} zmm1 = [3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0,3,6,22,12,3,7,7,0,3,6,1,13,3,21,7,0]224; AVX512BW-NEXT: # zmm1 = mem[0,1,2,3,0,1,2,3]225; AVX512BW-NEXT: vpermw %zmm0, %zmm1, %zmm0226; AVX512BW-NEXT: vpmovw2m %zmm0, %k0227; AVX512BW-NEXT: vpmovm2b %k0, %zmm0228; AVX512BW-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0229; AVX512BW-NEXT: retq230 %cmp = icmp eq <32 x i8> %a, zeroinitializer231 %b = shufflevector <32 x i1> %cmp, <32 x i1> undef, <32 x i32> <i32 3, i32 6, i32 22, i32 12, i32 3, i32 7, i32 7, i32 0, i32 3, i32 6, i32 1, i32 13, i32 3, i32 21, i32 7, i32 0, i32 3, i32 6, i32 22, i32 12, i32 3, i32 7, i32 7, i32 0, i32 3, i32 6, i32 1, i32 13, i32 3, i32 21, i32 7, i32 0>232 ret <32 x i1> %b233}234 235