845 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -disable-peephole -mtriple=i686-unknown -mattr=+avx | FileCheck %s --check-prefixes=CHECK,X86,AVX,AVX1,X86-AVX13; RUN: llc < %s -disable-peephole -mtriple=i686-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,X86,AVX,AVX2,X86-AVX24; RUN: llc < %s -disable-peephole -mtriple=i686-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,X86,AVX512,X86-AVX5125; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown -mattr=+avx | FileCheck %s --check-prefixes=CHECK,X64,AVX,AVX1,X64-AVX16; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,X64,AVX,AVX2,X64-AVX27; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,X64,AVX512,X64-AVX5128;9; Combine tests involving AVX target shuffles10 11declare <4 x float> @llvm.x86.avx.vpermil.ps(<4 x float>, i8)12declare <8 x float> @llvm.x86.avx.vpermil.ps.256(<8 x float>, i8)13declare <2 x double> @llvm.x86.avx.vpermil.pd(<2 x double>, i8)14declare <4 x double> @llvm.x86.avx.vpermil.pd.256(<4 x double>, i8)15 16declare <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float>, <4 x i32>)17declare <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float>, <8 x i32>)18declare <2 x double> @llvm.x86.avx.vpermilvar.pd(<2 x double>, <2 x i64>)19declare <4 x double> @llvm.x86.avx.vpermilvar.pd.256(<4 x double>, <4 x i64>)20 21declare <8 x i32> @llvm.x86.avx.vperm2f128.si.256(<8 x i32>, <8 x i32>, i8)22declare <8 x float> @llvm.x86.avx.vperm2f128.ps.256(<8 x float>, <8 x float>, i8)23declare <4 x double> @llvm.x86.avx.vperm2f128.pd.256(<4 x double>, <4 x double>, i8)24 25define <4 x float> @combine_vpermilvar_4f32_identity(<4 x float> %a0) {26; CHECK-LABEL: combine_vpermilvar_4f32_identity:27; CHECK: # %bb.0:28; CHECK-NEXT: ret{{[l|q]}}29 %1 = tail call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> %a0, <4 x i32> <i32 3, i32 2, i32 1, i32 0>)30 %2 = tail call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> %1, <4 x i32> <i32 3, i32 2, i32 1, i32 0>)31 ret <4 x float> %232}33 34define <4 x float> @combine_vpermilvar_4f32_movddup(<4 x float> %a0) {35; CHECK-LABEL: combine_vpermilvar_4f32_movddup:36; CHECK: # %bb.0:37; CHECK-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0]38; CHECK-NEXT: ret{{[l|q]}}39 %1 = tail call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> %a0, <4 x i32> <i32 0, i32 1, i32 0, i32 1>)40 ret <4 x float> %141}42define <4 x float> @combine_vpermilvar_4f32_movddup_load(ptr%a0) {43; X86-LABEL: combine_vpermilvar_4f32_movddup_load:44; X86: # %bb.0:45; X86-NEXT: movl {{[0-9]+}}(%esp), %eax46; X86-NEXT: vmovddup {{.*#+}} xmm0 = mem[0,0]47; X86-NEXT: retl48;49; X64-LABEL: combine_vpermilvar_4f32_movddup_load:50; X64: # %bb.0:51; X64-NEXT: vmovddup {{.*#+}} xmm0 = mem[0,0]52; X64-NEXT: retq53 %1 = load <4 x float>, ptr%a054 %2 = tail call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> %1, <4 x i32> <i32 0, i32 1, i32 0, i32 1>)55 ret <4 x float> %256}57 58define <4 x float> @combine_vpermilvar_4f32_movshdup(<4 x float> %a0) {59; CHECK-LABEL: combine_vpermilvar_4f32_movshdup:60; CHECK: # %bb.0:61; CHECK-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]62; CHECK-NEXT: ret{{[l|q]}}63 %1 = tail call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> %a0, <4 x i32> <i32 undef, i32 1, i32 3, i32 3>)64 ret <4 x float> %165}66 67define <4 x float> @combine_vpermilvar_4f32_movsldup(<4 x float> %a0) {68; CHECK-LABEL: combine_vpermilvar_4f32_movsldup:69; CHECK: # %bb.0:70; CHECK-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2]71; CHECK-NEXT: ret{{[l|q]}}72 %1 = tail call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> %a0, <4 x i32> <i32 0, i32 0, i32 2, i32 undef>)73 ret <4 x float> %174}75 76define <4 x float> @combine_vpermilvar_4f32_unpckh(<4 x float> %a0) {77; CHECK-LABEL: combine_vpermilvar_4f32_unpckh:78; CHECK: # %bb.0:79; CHECK-NEXT: vshufps {{.*#+}} xmm0 = xmm0[2,2,3,3]80; CHECK-NEXT: ret{{[l|q]}}81 %1 = tail call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> %a0, <4 x i32> <i32 2, i32 2, i32 3, i32 3>)82 ret <4 x float> %183}84 85define <4 x float> @combine_vpermilvar_4f32_unpckl(<4 x float> %a0) {86; CHECK-LABEL: combine_vpermilvar_4f32_unpckl:87; CHECK: # %bb.0:88; CHECK-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,0,1,1]89; CHECK-NEXT: ret{{[l|q]}}90 %1 = tail call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> %a0, <4 x i32> <i32 0, i32 0, i32 1, i32 1>)91 ret <4 x float> %192}93 94define <8 x float> @combine_vpermilvar_8f32_identity(<8 x float> %a0) {95; CHECK-LABEL: combine_vpermilvar_8f32_identity:96; CHECK: # %bb.0:97; CHECK-NEXT: ret{{[l|q]}}98 %1 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %a0, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 2, i32 3, i32 0, i32 undef>)99 %2 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %1, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 2, i32 3, i32 0, i32 1>)100 ret <8 x float> %2101}102 103define <8 x float> @combine_vpermilvar_8f32_10326u4u(<8 x float> %a0) {104; CHECK-LABEL: combine_vpermilvar_8f32_10326u4u:105; CHECK: # %bb.0:106; CHECK-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,0,3,2,6,u,4,u]107; CHECK-NEXT: ret{{[l|q]}}108 %1 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %a0, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 0, i32 1, i32 2, i32 undef>)109 %2 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %1, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 undef>)110 ret <8 x float> %2111}112 113define <8 x float> @combine_vpermilvar_vperm2f128_8f32(<8 x float> %a0) {114; AVX1-LABEL: combine_vpermilvar_vperm2f128_8f32:115; AVX1: # %bb.0:116; AVX1-NEXT: vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,0,1]117; AVX1-NEXT: ret{{[l|q]}}118;119; AVX2-LABEL: combine_vpermilvar_vperm2f128_8f32:120; AVX2: # %bb.0:121; AVX2-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,1]122; AVX2-NEXT: ret{{[l|q]}}123;124; AVX512-LABEL: combine_vpermilvar_vperm2f128_8f32:125; AVX512: # %bb.0:126; AVX512-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,1]127; AVX512-NEXT: ret{{[l|q]}}128 %1 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %a0, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 3, i32 2, i32 1, i32 0>)129 %2 = shufflevector <8 x float> %1, <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>130 %3 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %2, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 3, i32 2, i32 1, i32 0>)131 ret <8 x float> %3132}133 134define <8 x float> @combine_vpermilvar_vperm2f128_zero_8f32(<8 x float> %a0) {135; AVX-LABEL: combine_vpermilvar_vperm2f128_zero_8f32:136; AVX: # %bb.0:137; AVX-NEXT: vperm2f128 {{.*#+}} ymm0 = zero,zero,ymm0[0,1]138; AVX-NEXT: ret{{[l|q]}}139;140; AVX512-LABEL: combine_vpermilvar_vperm2f128_zero_8f32:141; AVX512: # %bb.0:142; AVX512-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0143; AVX512-NEXT: vpmovsxbd {{.*#+}} ymm1 = [16,17,18,19,3,2,1,0]144; AVX512-NEXT: vxorps %xmm2, %xmm2, %xmm2145; AVX512-NEXT: vpermt2ps %zmm2, %zmm1, %zmm0146; AVX512-NEXT: vshufps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]147; AVX512-NEXT: ret{{[l|q]}}148 %1 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %a0, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 3, i32 2, i32 1, i32 0>)149 %2 = shufflevector <8 x float> %1, <8 x float> zeroinitializer, <8 x i32> <i32 8, i32 8, i32 8, i32 8, i32 0, i32 1, i32 2, i32 3>150 %3 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %2, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 3, i32 2, i32 1, i32 0>)151 ret <8 x float> %3152}153 154define <4 x double> @combine_vperm2f128_vpermilvar_as_vperm2f128(<4 x double> %a0) {155; CHECK-LABEL: combine_vperm2f128_vpermilvar_as_vperm2f128:156; CHECK: # %bb.0:157; CHECK-NEXT: vperm2f128 {{.*#+}} ymm0 = zero,zero,ymm0[0,1]158; CHECK-NEXT: ret{{[l|q]}}159 %1 = tail call <4 x double> @llvm.x86.avx.vpermilvar.pd.256(<4 x double> %a0, <4 x i64> <i64 2, i64 0, i64 2, i64 0>)160 %2 = shufflevector <4 x double> %1, <4 x double> zeroinitializer, <4 x i32> <i32 4, i32 5, i32 0, i32 1>161 %3 = tail call <4 x double> @llvm.x86.avx.vpermilvar.pd.256(<4 x double> %2, <4 x i64> <i64 2, i64 0, i64 2, i64 0>)162 ret <4 x double> %3163}164 165define <4 x double> @combine_vperm2f128_vpermilvar_as_vmovaps(<4 x double> %a0) {166; CHECK-LABEL: combine_vperm2f128_vpermilvar_as_vmovaps:167; CHECK: # %bb.0:168; CHECK-NEXT: vmovaps %xmm0, %xmm0169; CHECK-NEXT: ret{{[l|q]}}170 %1 = tail call <4 x double> @llvm.x86.avx.vpermilvar.pd.256(<4 x double> %a0, <4 x i64> <i64 2, i64 0, i64 2, i64 0>)171 %2 = shufflevector <4 x double> %1, <4 x double> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 4, i32 5>172 %3 = tail call <4 x double> @llvm.x86.avx.vpermilvar.pd.256(<4 x double> %2, <4 x i64> <i64 2, i64 0, i64 2, i64 0>)173 ret <4 x double> %3174}175 176define <8 x float> @combine_vpermilvar_8f32_movddup(<8 x float> %a0) {177; CHECK-LABEL: combine_vpermilvar_8f32_movddup:178; CHECK: # %bb.0:179; CHECK-NEXT: vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2]180; CHECK-NEXT: ret{{[l|q]}}181 %1 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %a0, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5>)182 ret <8 x float> %1183}184define <8 x float> @combine_vpermilvar_8f32_movddup_load(ptr%a0) {185; X86-LABEL: combine_vpermilvar_8f32_movddup_load:186; X86: # %bb.0:187; X86-NEXT: movl {{[0-9]+}}(%esp), %eax188; X86-NEXT: vmovddup {{.*#+}} ymm0 = mem[0,0,2,2]189; X86-NEXT: retl190;191; X64-LABEL: combine_vpermilvar_8f32_movddup_load:192; X64: # %bb.0:193; X64-NEXT: vmovddup {{.*#+}} ymm0 = mem[0,0,2,2]194; X64-NEXT: retq195 %1 = load <8 x float>, ptr%a0196 %2 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %1, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5>)197 ret <8 x float> %2198}199 200define <8 x float> @combine_vpermilvar_8f32_movshdup(<8 x float> %a0) {201; CHECK-LABEL: combine_vpermilvar_8f32_movshdup:202; CHECK: # %bb.0:203; CHECK-NEXT: vmovshdup {{.*#+}} ymm0 = ymm0[1,1,3,3,5,5,7,7]204; CHECK-NEXT: ret{{[l|q]}}205 %1 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %a0, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 undef, i32 5, i32 7, i32 7>)206 ret <8 x float> %1207}208define <8 x float> @demandedelts_vpermilvar_8f32_movshdup(<8 x float> %a0, i32 %a1) {209; CHECK-LABEL: demandedelts_vpermilvar_8f32_movshdup:210; CHECK: # %bb.0:211; CHECK-NEXT: vmovshdup {{.*#+}} ymm0 = ymm0[1,1,3,3,5,5,7,7]212; CHECK-NEXT: ret{{[l|q]}}213 %1 = insertelement <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 undef, i32 5, i32 7, i32 7>, i32 %a1, i32 7214 %2 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %a0, <8 x i32> %1)215 %3 = shufflevector <8 x float> %2, <8 x float> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 6>216 ret <8 x float> %3217}218 219define <8 x float> @combine_vpermilvar_8f32_movsldup(<8 x float> %a0) {220; CHECK-LABEL: combine_vpermilvar_8f32_movsldup:221; CHECK: # %bb.0:222; CHECK-NEXT: vmovsldup {{.*#+}} ymm0 = ymm0[0,0,2,2,4,4,6,6]223; CHECK-NEXT: ret{{[l|q]}}224 %1 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %a0, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6>)225 ret <8 x float> %1226}227define <8 x float> @demandedelts_vpermilvar_8f32_movsldup(<8 x float> %a0, i32 %a1) {228; CHECK-LABEL: demandedelts_vpermilvar_8f32_movsldup:229; CHECK: # %bb.0:230; CHECK-NEXT: vmovsldup {{.*#+}} ymm0 = ymm0[0,0,2,2,4,4,6,6]231; CHECK-NEXT: ret{{[l|q]}}232 %1 = insertelement <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6>, i32 %a1, i32 0233 %2 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %a0, <8 x i32> %1)234 %3 = shufflevector <8 x float> %2, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>235 ret <8 x float> %3236}237 238define <2 x double> @combine_vpermilvar_2f64_identity(<2 x double> %a0) {239; CHECK-LABEL: combine_vpermilvar_2f64_identity:240; CHECK: # %bb.0:241; CHECK-NEXT: ret{{[l|q]}}242 %1 = tail call <2 x double> @llvm.x86.avx.vpermilvar.pd(<2 x double> %a0, <2 x i64> <i64 2, i64 0>)243 %2 = tail call <2 x double> @llvm.x86.avx.vpermilvar.pd(<2 x double> %1, <2 x i64> <i64 2, i64 0>)244 ret <2 x double> %2245}246 247define <2 x double> @combine_vpermilvar_2f64_movddup(<2 x double> %a0) {248; CHECK-LABEL: combine_vpermilvar_2f64_movddup:249; CHECK: # %bb.0:250; CHECK-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0]251; CHECK-NEXT: ret{{[l|q]}}252 %1 = tail call <2 x double> @llvm.x86.avx.vpermilvar.pd(<2 x double> %a0, <2 x i64> <i64 0, i64 0>)253 ret <2 x double> %1254}255 256define <4 x double> @combine_vpermilvar_4f64_identity(<4 x double> %a0) {257; CHECK-LABEL: combine_vpermilvar_4f64_identity:258; CHECK: # %bb.0:259; CHECK-NEXT: ret{{[l|q]}}260 %1 = tail call <4 x double> @llvm.x86.avx.vpermilvar.pd.256(<4 x double> %a0, <4 x i64> <i64 2, i64 0, i64 2, i64 0>)261 %2 = tail call <4 x double> @llvm.x86.avx.vpermilvar.pd.256(<4 x double> %1, <4 x i64> <i64 2, i64 0, i64 2, i64 0>)262 ret <4 x double> %2263}264 265define <4 x double> @combine_vpermilvar_4f64_movddup(<4 x double> %a0) {266; CHECK-LABEL: combine_vpermilvar_4f64_movddup:267; CHECK: # %bb.0:268; CHECK-NEXT: vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2]269; CHECK-NEXT: ret{{[l|q]}}270 %1 = tail call <4 x double> @llvm.x86.avx.vpermilvar.pd.256(<4 x double> %a0, <4 x i64> <i64 0, i64 0, i64 4, i64 4>)271 ret <4 x double> %1272}273 274define <4 x float> @combine_vpermilvar_4f32_4stage(<4 x float> %a0) {275; CHECK-LABEL: combine_vpermilvar_4f32_4stage:276; CHECK: # %bb.0:277; CHECK-NEXT: vshufps {{.*#+}} xmm0 = xmm0[2,0,3,1]278; CHECK-NEXT: ret{{[l|q]}}279 %1 = tail call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> %a0, <4 x i32> <i32 3, i32 2, i32 1, i32 0>)280 %2 = tail call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> %1, <4 x i32> <i32 2, i32 3, i32 0, i32 1>)281 %3 = tail call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> %2, <4 x i32> <i32 0, i32 2, i32 1, i32 3>)282 %4 = tail call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> %3, <4 x i32> <i32 3, i32 2, i32 1, i32 0>)283 ret <4 x float> %4284}285 286define <8 x float> @combine_vpermilvar_8f32_4stage(<8 x float> %a0) {287; CHECK-LABEL: combine_vpermilvar_8f32_4stage:288; CHECK: # %bb.0:289; CHECK-NEXT: vshufps {{.*#+}} ymm0 = ymm0[2,0,3,1,6,4,7,5]290; CHECK-NEXT: ret{{[l|q]}}291 %1 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %a0, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 3, i32 2, i32 1, i32 0>)292 %2 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %1, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>)293 %3 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %2, <8 x i32> <i32 0, i32 2, i32 1, i32 3, i32 0, i32 2, i32 1, i32 3>)294 %4 = tail call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %3, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 3, i32 2, i32 1, i32 0>)295 ret <8 x float> %4296}297 298define <4 x float> @combine_vpermilvar_4f32_as_insertps(<4 x float> %a0) {299; CHECK-LABEL: combine_vpermilvar_4f32_as_insertps:300; CHECK: # %bb.0:301; CHECK-NEXT: vinsertps {{.*#+}} xmm0 = xmm0[1],zero,xmm0[2],zero302; CHECK-NEXT: ret{{[l|q]}}303 %1 = call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> %a0, <4 x i32> <i32 3, i32 2, i32 1, i32 0>)304 %2 = shufflevector <4 x float> %1, <4 x float> zeroinitializer, <4 x i32> <i32 2, i32 4, i32 1, i32 4>305 ret <4 x float> %2306}307 308define <8 x i32> @combine_blend_of_permutes_v8i32(<4 x i64> %a0, <4 x i64> %a1) {309; AVX1-LABEL: combine_blend_of_permutes_v8i32:310; AVX1: # %bb.0:311; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]312; AVX1-NEXT: vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,0,1]313; AVX1-NEXT: ret{{[l|q]}}314;315; AVX2-LABEL: combine_blend_of_permutes_v8i32:316; AVX2: # %bb.0:317; AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]318; AVX2-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,1]319; AVX2-NEXT: ret{{[l|q]}}320;321; AVX512-LABEL: combine_blend_of_permutes_v8i32:322; AVX512: # %bb.0:323; AVX512-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1324; AVX512-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0325; AVX512-NEXT: vpmovsxbd {{.*#+}} ymm2 = [4,21,6,23,16,1,2,19]326; AVX512-NEXT: vpermt2d %zmm1, %zmm2, %zmm0327; AVX512-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0328; AVX512-NEXT: ret{{[l|q]}}329 %s0 = shufflevector <4 x i64> %a0, <4 x i64> undef, <4 x i32> <i32 2, i32 3, i32 0, i32 1>330 %s1 = shufflevector <4 x i64> %a1, <4 x i64> undef, <4 x i32> <i32 2, i32 3, i32 0, i32 1>331 %x0 = bitcast <4 x i64> %s0 to <8 x i32>332 %x1 = bitcast <4 x i64> %s1 to <8 x i32>333 %r = shufflevector <8 x i32> %x0, <8 x i32> %x1, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 12, i32 5, i32 6, i32 15>334 ret <8 x i32> %r335}336 337define <2 x double> @constant_fold_vpermilvar_pd() {338; CHECK-LABEL: constant_fold_vpermilvar_pd:339; CHECK: # %bb.0:340; CHECK-NEXT: vmovaps {{.*#+}} xmm0 = [2.0E+0,1.0E+0]341; CHECK-NEXT: ret{{[l|q]}}342 %1 = call <2 x double> @llvm.x86.avx.vpermilvar.pd(<2 x double> <double 1.0, double 2.0>, <2 x i64> <i64 2, i64 0>)343 ret <2 x double> %1344}345 346define <4 x double> @constant_fold_vpermilvar_pd_256() {347; CHECK-LABEL: constant_fold_vpermilvar_pd_256:348; CHECK: # %bb.0:349; CHECK-NEXT: vmovaps {{.*#+}} ymm0 = [2.0E+0,1.0E+0,3.0E+0,4.0E+0]350; CHECK-NEXT: ret{{[l|q]}}351 %1 = call <4 x double> @llvm.x86.avx.vpermilvar.pd.256(<4 x double> <double 1.0, double 2.0, double 3.0, double 4.0>, <4 x i64> <i64 2, i64 0, i64 0, i64 2>)352 ret <4 x double> %1353}354 355define <4 x float> @constant_fold_vpermilvar_ps() {356; CHECK-LABEL: constant_fold_vpermilvar_ps:357; CHECK: # %bb.0:358; CHECK-NEXT: vmovaps {{.*#+}} xmm0 = [4.0E+0,1.0E+0,3.0E+0,2.0E+0]359; CHECK-NEXT: ret{{[l|q]}}360 %1 = call <4 x float> @llvm.x86.avx.vpermilvar.ps(<4 x float> <float 1.0, float 2.0, float 3.0, float 4.0>, <4 x i32> <i32 3, i32 0, i32 2, i32 1>)361 ret <4 x float> %1362}363 364define <8 x float> @constant_fold_vpermilvar_ps_256() {365; CHECK-LABEL: constant_fold_vpermilvar_ps_256:366; CHECK: # %bb.0:367; CHECK-NEXT: vmovaps {{.*#+}} ymm0 = [1.0E+0,1.0E+0,3.0E+0,2.0E+0,5.0E+0,6.0E+0,6.0E+0,6.0E+0]368; CHECK-NEXT: ret{{[l|q]}}369 %1 = call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> <float 1.0, float 2.0, float 3.0, float 4.0, float 5.0, float 6.0, float 7.0, float 8.0>, <8 x i32> <i32 4, i32 0, i32 2, i32 1, i32 0, i32 1, i32 1, i32 1>)370 ret <8 x float> %1371}372 373define <8 x float> @freeze_vpermilvar_ps_256(<8 x float> %a0) {374; CHECK-LABEL: freeze_vpermilvar_ps_256:375; CHECK: # %bb.0:376; CHECK-NEXT: ret{{[l|q]}}377 %s0 = call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %a0, <8 x i32> <i32 0, i32 3, i32 1, i32 2, i32 7, i32 6, i32 5, i32 4>)378 %f0 = freeze <8 x float> %s0379 %s1 = call <8 x float> @llvm.x86.avx.vpermilvar.ps.256(<8 x float> %f0, <8 x i32> <i32 0, i32 2, i32 3, i32 1, i32 7, i32 6, i32 5, i32 4>)380 ret <8 x float> %s1381}382 383define void @PR39483() {384; X86-AVX1-LABEL: PR39483:385; X86-AVX1: # %bb.0: # %entry386; X86-AVX1-NEXT: vmovups 32, %ymm0387; X86-AVX1-NEXT: vmovups 64, %ymm1388; X86-AVX1-NEXT: vperm2f128 {{.*#+}} ymm2 = ymm1[2,3,0,1]389; X86-AVX1-NEXT: vshufps {{.*#+}} ymm1 = ymm2[0,1],ymm1[0,3],ymm2[4,5],ymm1[4,7]390; X86-AVX1-NEXT: vmovups 16, %xmm2391; X86-AVX1-NEXT: vblendps {{.*#+}} ymm3 = ymm0[0,1],mem[2],ymm0[3,4],mem[5],ymm0[6,7]392; X86-AVX1-NEXT: vshufps {{.*#+}} ymm2 = ymm2[1,0],ymm3[2,0],ymm2[5,4],ymm3[6,4]393; X86-AVX1-NEXT: vshufps {{.*#+}} ymm0 = ymm2[2,0],ymm0[0,3],ymm2[6,4],ymm0[4,7]394; X86-AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]395; X86-AVX1-NEXT: vxorps %xmm1, %xmm1, %xmm1396; X86-AVX1-NEXT: vmulps %ymm1, %ymm0, %ymm0397; X86-AVX1-NEXT: vaddps %ymm1, %ymm0, %ymm0398; X86-AVX1-NEXT: vmovups %ymm0, (%eax)399;400; X86-AVX2-LABEL: PR39483:401; X86-AVX2: # %bb.0: # %entry402; X86-AVX2-NEXT: vmovups 32, %ymm0403; X86-AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1],mem[2],ymm0[3,4],mem[5],ymm0[6,7]404; X86-AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]405; X86-AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0406; X86-AVX2-NEXT: vpermilps {{.*#+}} ymm1 = mem[0,1,0,3,4,5,4,7]407; X86-AVX2-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]408; X86-AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]409; X86-AVX2-NEXT: vxorps %xmm1, %xmm1, %xmm1410; X86-AVX2-NEXT: vmulps %ymm1, %ymm0, %ymm0411; X86-AVX2-NEXT: vaddps %ymm1, %ymm0, %ymm0412; X86-AVX2-NEXT: vmovups %ymm0, (%eax)413;414; X86-AVX512-LABEL: PR39483:415; X86-AVX512: # %bb.0: # %entry416; X86-AVX512-NEXT: vmovups 64, %ymm0417; X86-AVX512-NEXT: vpmovsxbd {{.*#+}} ymm1 = [18,21,24,27,30,1,4,7]418; X86-AVX512-NEXT: vpermt2ps 0, %zmm1, %zmm0419; X86-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1420; X86-AVX512-NEXT: vmulps %ymm1, %ymm0, %ymm0421; X86-AVX512-NEXT: vaddps %ymm1, %ymm0, %ymm0422; X86-AVX512-NEXT: vmovups %ymm0, (%eax)423;424; X64-AVX1-LABEL: PR39483:425; X64-AVX1: # %bb.0: # %entry426; X64-AVX1-NEXT: vmovups 32, %ymm0427; X64-AVX1-NEXT: vmovups 64, %ymm1428; X64-AVX1-NEXT: vperm2f128 {{.*#+}} ymm2 = ymm1[2,3,0,1]429; X64-AVX1-NEXT: vshufps {{.*#+}} ymm1 = ymm2[0,1],ymm1[0,3],ymm2[4,5],ymm1[4,7]430; X64-AVX1-NEXT: vmovups 16, %xmm2431; X64-AVX1-NEXT: vblendps {{.*#+}} ymm3 = ymm0[0,1],mem[2],ymm0[3,4],mem[5],ymm0[6,7]432; X64-AVX1-NEXT: vshufps {{.*#+}} ymm2 = ymm2[1,0],ymm3[2,0],ymm2[5,4],ymm3[6,4]433; X64-AVX1-NEXT: vshufps {{.*#+}} ymm0 = ymm2[2,0],ymm0[0,3],ymm2[6,4],ymm0[4,7]434; X64-AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]435; X64-AVX1-NEXT: vxorps %xmm1, %xmm1, %xmm1436; X64-AVX1-NEXT: vmulps %ymm1, %ymm0, %ymm0437; X64-AVX1-NEXT: vaddps %ymm1, %ymm0, %ymm0438; X64-AVX1-NEXT: vmovups %ymm0, (%rax)439;440; X64-AVX2-LABEL: PR39483:441; X64-AVX2: # %bb.0: # %entry442; X64-AVX2-NEXT: vmovups 32, %ymm0443; X64-AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1],mem[2],ymm0[3,4],mem[5],ymm0[6,7]444; X64-AVX2-NEXT: vmovaps {{.*#+}} ymm1 = [2,5,0,3,6,u,u,u]445; X64-AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0446; X64-AVX2-NEXT: vpermilps {{.*#+}} ymm1 = mem[0,1,0,3,4,5,4,7]447; X64-AVX2-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]448; X64-AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]449; X64-AVX2-NEXT: vxorps %xmm1, %xmm1, %xmm1450; X64-AVX2-NEXT: vmulps %ymm1, %ymm0, %ymm0451; X64-AVX2-NEXT: vaddps %ymm1, %ymm0, %ymm0452; X64-AVX2-NEXT: vmovups %ymm0, (%rax)453;454; X64-AVX512-LABEL: PR39483:455; X64-AVX512: # %bb.0: # %entry456; X64-AVX512-NEXT: vmovups 64, %ymm0457; X64-AVX512-NEXT: vpmovsxbd {{.*#+}} ymm1 = [18,21,24,27,30,1,4,7]458; X64-AVX512-NEXT: vpermt2ps 0, %zmm1, %zmm0459; X64-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1460; X64-AVX512-NEXT: vmulps %ymm1, %ymm0, %ymm0461; X64-AVX512-NEXT: vaddps %ymm1, %ymm0, %ymm0462; X64-AVX512-NEXT: vmovups %ymm0, (%rax)463entry:464 %wide.vec = load <24 x float>, ptr null, align 4465 %strided.vec18 = shufflevector <24 x float> %wide.vec, <24 x float> undef, <8 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23>466 %0 = fmul <8 x float> %strided.vec18, zeroinitializer467 %1 = fadd <8 x float> zeroinitializer, %0468 store <8 x float> %1, ptr undef, align 16469 unreachable470}471 472define void @PR48908(<4 x double> %v0, <4 x double> %v1, <4 x double> %v2, ptr noalias %out0, ptr noalias %out1, ptr noalias %out2) {473; X86-AVX1-LABEL: PR48908:474; X86-AVX1: # %bb.0:475; X86-AVX1-NEXT: movl {{[0-9]+}}(%esp), %eax476; X86-AVX1-NEXT: movl {{[0-9]+}}(%esp), %ecx477; X86-AVX1-NEXT: movl {{[0-9]+}}(%esp), %edx478; X86-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm3479; X86-AVX1-NEXT: vshufpd {{.*#+}} ymm3 = ymm3[0,1,2,2]480; X86-AVX1-NEXT: vperm2f128 {{.*#+}} ymm4 = ymm1[2,3],ymm2[0,1]481; X86-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm1, %ymm5482; X86-AVX1-NEXT: vshufpd {{.*#+}} ymm4 = ymm5[1],ymm4[0],ymm5[2],ymm4[3]483; X86-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm5484; X86-AVX1-NEXT: vperm2f128 {{.*#+}} ymm3 = ymm3[2,3,0,1]485; X86-AVX1-NEXT: vblendpd {{.*#+}} ymm3 = ymm5[0],ymm3[1],ymm5[2],ymm3[3]486; X86-AVX1-NEXT: vmovapd %ymm3, (%edx)487; X86-AVX1-NEXT: vperm2f128 {{.*#+}} ymm3 = ymm4[2,3,0,1]488; X86-AVX1-NEXT: vblendpd {{.*#+}} ymm4 = ymm4[0,1],ymm0[2,3]489; X86-AVX1-NEXT: vblendpd {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2],ymm3[3]490; X86-AVX1-NEXT: vmovapd %ymm3, (%ecx)491; X86-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0492; X86-AVX1-NEXT: vperm2f128 {{.*#+}} ymm1 = ymm2[2,3],ymm1[2,3]493; X86-AVX1-NEXT: vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm2[1],ymm0[3],ymm2[3]494; X86-AVX1-NEXT: vshufpd {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[3],ymm0[3]495; X86-AVX1-NEXT: vmovapd %ymm0, (%eax)496; X86-AVX1-NEXT: vzeroupper497; X86-AVX1-NEXT: retl498;499; X86-AVX2-LABEL: PR48908:500; X86-AVX2: # %bb.0:501; X86-AVX2-NEXT: movl {{[0-9]+}}(%esp), %eax502; X86-AVX2-NEXT: movl {{[0-9]+}}(%esp), %ecx503; X86-AVX2-NEXT: movl {{[0-9]+}}(%esp), %edx504; X86-AVX2-NEXT: vbroadcastsd %xmm1, %ymm3505; X86-AVX2-NEXT: vperm2f128 {{.*#+}} ymm4 = ymm1[2,3],ymm2[0,1]506; X86-AVX2-NEXT: vshufpd {{.*#+}} xmm5 = xmm1[1,0]507; X86-AVX2-NEXT: vperm2f128 {{.*#+}} ymm6 = ymm0[0,1],ymm2[0,1]508; X86-AVX2-NEXT: vperm2f128 {{.*#+}} ymm3 = ymm3[2,3],ymm0[0,1]509; X86-AVX2-NEXT: vblendpd {{.*#+}} ymm3 = ymm6[0],ymm3[1],ymm6[2],ymm3[3]510; X86-AVX2-NEXT: vmovapd %ymm3, (%edx)511; X86-AVX2-NEXT: vblendpd {{.*#+}} ymm3 = ymm5[0,1],ymm0[2,3]512; X86-AVX2-NEXT: vpermpd {{.*#+}} ymm4 = ymm4[0,3,2,0]513; X86-AVX2-NEXT: vblendpd {{.*#+}} ymm3 = ymm3[0],ymm4[1],ymm3[2],ymm4[3]514; X86-AVX2-NEXT: vmovapd %ymm3, (%ecx)515; X86-AVX2-NEXT: vextractf128 $1, %ymm0, %xmm0516; X86-AVX2-NEXT: vperm2f128 {{.*#+}} ymm1 = ymm2[2,3],ymm1[2,3]517; X86-AVX2-NEXT: vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm2[1],ymm0[3],ymm2[3]518; X86-AVX2-NEXT: vshufpd {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[3],ymm0[3]519; X86-AVX2-NEXT: vmovapd %ymm0, (%eax)520; X86-AVX2-NEXT: vzeroupper521; X86-AVX2-NEXT: retl522;523; X86-AVX512-LABEL: PR48908:524; X86-AVX512: # %bb.0:525; X86-AVX512-NEXT: # kill: def $ymm2 killed $ymm2 def $zmm2526; X86-AVX512-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1527; X86-AVX512-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0528; X86-AVX512-NEXT: movl {{[0-9]+}}(%esp), %eax529; X86-AVX512-NEXT: movl {{[0-9]+}}(%esp), %ecx530; X86-AVX512-NEXT: movl {{[0-9]+}}(%esp), %edx531; X86-AVX512-NEXT: vpmovsxbq {{.*#+}} ymm3 = [1,2,8,9]532; X86-AVX512-NEXT: vpermi2pd %zmm2, %zmm1, %zmm3533; X86-AVX512-NEXT: vpmovsxbq {{.*#+}} ymm4 = [0,8,2,1]534; X86-AVX512-NEXT: vpermi2pd %zmm1, %zmm0, %zmm4535; X86-AVX512-NEXT: vperm2f128 {{.*#+}} ymm5 = ymm0[0,1],ymm2[0,1]536; X86-AVX512-NEXT: vblendpd {{.*#+}} ymm4 = ymm5[0],ymm4[1],ymm5[2],ymm4[3]537; X86-AVX512-NEXT: vmovapd %ymm4, (%edx)538; X86-AVX512-NEXT: vpmovsxbq {{.*#+}} ymm4 = [0,3,10,1]539; X86-AVX512-NEXT: vpermi2pd %zmm0, %zmm3, %zmm4540; X86-AVX512-NEXT: vmovapd %ymm4, (%ecx)541; X86-AVX512-NEXT: vpmovsxbq {{.*#+}} ymm3 = [3,11,0,0]542; X86-AVX512-NEXT: vpermi2pd %zmm1, %zmm0, %zmm3543; X86-AVX512-NEXT: vpmovsxbq {{.*#+}} ymm0 = [2,8,9,3]544; X86-AVX512-NEXT: vpermi2pd %zmm3, %zmm2, %zmm0545; X86-AVX512-NEXT: vmovapd %ymm0, (%eax)546; X86-AVX512-NEXT: vzeroupper547; X86-AVX512-NEXT: retl548;549; X64-AVX1-LABEL: PR48908:550; X64-AVX1: # %bb.0:551; X64-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm3552; X64-AVX1-NEXT: vshufpd {{.*#+}} ymm3 = ymm3[0,1,2,2]553; X64-AVX1-NEXT: vperm2f128 {{.*#+}} ymm4 = ymm1[2,3],ymm2[0,1]554; X64-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm1, %ymm5555; X64-AVX1-NEXT: vshufpd {{.*#+}} ymm4 = ymm5[1],ymm4[0],ymm5[2],ymm4[3]556; X64-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm5557; X64-AVX1-NEXT: vperm2f128 {{.*#+}} ymm3 = ymm3[2,3,0,1]558; X64-AVX1-NEXT: vblendpd {{.*#+}} ymm3 = ymm5[0],ymm3[1],ymm5[2],ymm3[3]559; X64-AVX1-NEXT: vmovapd %ymm3, (%rdi)560; X64-AVX1-NEXT: vperm2f128 {{.*#+}} ymm3 = ymm4[2,3,0,1]561; X64-AVX1-NEXT: vblendpd {{.*#+}} ymm4 = ymm4[0,1],ymm0[2,3]562; X64-AVX1-NEXT: vblendpd {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2],ymm3[3]563; X64-AVX1-NEXT: vmovapd %ymm3, (%rsi)564; X64-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0565; X64-AVX1-NEXT: vperm2f128 {{.*#+}} ymm1 = ymm2[2,3],ymm1[2,3]566; X64-AVX1-NEXT: vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm2[1],ymm0[3],ymm2[3]567; X64-AVX1-NEXT: vshufpd {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[3],ymm0[3]568; X64-AVX1-NEXT: vmovapd %ymm0, (%rdx)569; X64-AVX1-NEXT: vzeroupper570; X64-AVX1-NEXT: retq571;572; X64-AVX2-LABEL: PR48908:573; X64-AVX2: # %bb.0:574; X64-AVX2-NEXT: vbroadcastsd %xmm1, %ymm3575; X64-AVX2-NEXT: vperm2f128 {{.*#+}} ymm4 = ymm1[2,3],ymm2[0,1]576; X64-AVX2-NEXT: vshufpd {{.*#+}} xmm5 = xmm1[1,0]577; X64-AVX2-NEXT: vperm2f128 {{.*#+}} ymm6 = ymm0[0,1],ymm2[0,1]578; X64-AVX2-NEXT: vperm2f128 {{.*#+}} ymm3 = ymm3[2,3],ymm0[0,1]579; X64-AVX2-NEXT: vblendpd {{.*#+}} ymm3 = ymm6[0],ymm3[1],ymm6[2],ymm3[3]580; X64-AVX2-NEXT: vmovapd %ymm3, (%rdi)581; X64-AVX2-NEXT: vblendpd {{.*#+}} ymm3 = ymm5[0,1],ymm0[2,3]582; X64-AVX2-NEXT: vpermpd {{.*#+}} ymm4 = ymm4[0,3,2,0]583; X64-AVX2-NEXT: vblendpd {{.*#+}} ymm3 = ymm3[0],ymm4[1],ymm3[2],ymm4[3]584; X64-AVX2-NEXT: vmovapd %ymm3, (%rsi)585; X64-AVX2-NEXT: vextractf128 $1, %ymm0, %xmm0586; X64-AVX2-NEXT: vperm2f128 {{.*#+}} ymm1 = ymm2[2,3],ymm1[2,3]587; X64-AVX2-NEXT: vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm2[1],ymm0[3],ymm2[3]588; X64-AVX2-NEXT: vshufpd {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[3],ymm0[3]589; X64-AVX2-NEXT: vmovapd %ymm0, (%rdx)590; X64-AVX2-NEXT: vzeroupper591; X64-AVX2-NEXT: retq592;593; X64-AVX512-LABEL: PR48908:594; X64-AVX512: # %bb.0:595; X64-AVX512-NEXT: # kill: def $ymm2 killed $ymm2 def $zmm2596; X64-AVX512-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1597; X64-AVX512-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0598; X64-AVX512-NEXT: vpmovsxbq {{.*#+}} ymm3 = [1,2,8,9]599; X64-AVX512-NEXT: vpermi2pd %zmm2, %zmm1, %zmm3600; X64-AVX512-NEXT: vpmovsxbq {{.*#+}} ymm4 = [0,8,2,1]601; X64-AVX512-NEXT: vpermi2pd %zmm1, %zmm0, %zmm4602; X64-AVX512-NEXT: vperm2f128 {{.*#+}} ymm5 = ymm0[0,1],ymm2[0,1]603; X64-AVX512-NEXT: vblendpd {{.*#+}} ymm4 = ymm5[0],ymm4[1],ymm5[2],ymm4[3]604; X64-AVX512-NEXT: vmovapd %ymm4, (%rdi)605; X64-AVX512-NEXT: vpmovsxbq {{.*#+}} ymm4 = [0,3,10,1]606; X64-AVX512-NEXT: vpermi2pd %zmm0, %zmm3, %zmm4607; X64-AVX512-NEXT: vmovapd %ymm4, (%rsi)608; X64-AVX512-NEXT: vpmovsxbq {{.*#+}} xmm3 = [3,11]609; X64-AVX512-NEXT: vpermi2pd %zmm1, %zmm0, %zmm3610; X64-AVX512-NEXT: vpmovsxbq {{.*#+}} ymm0 = [2,8,9,3]611; X64-AVX512-NEXT: vpermi2pd %zmm3, %zmm2, %zmm0612; X64-AVX512-NEXT: vmovapd %ymm0, (%rdx)613; X64-AVX512-NEXT: vzeroupper614; X64-AVX512-NEXT: retq615 %t0 = shufflevector <4 x double> %v0, <4 x double> %v1, <4 x i32> <i32 0, i32 1, i32 2, i32 4>616 %t1 = shufflevector <4 x double> %v1, <4 x double> %v2, <4 x i32> <i32 1, i32 2, i32 4, i32 5>617 %r0 = shufflevector <4 x double> %t0, <4 x double> %t1, <4 x i32> <i32 0, i32 3, i32 6, i32 1>618 store <4 x double> %r0, ptr %out0, align 32619 %r1 = shufflevector <4 x double> %t0, <4 x double> %t1, <4 x i32> <i32 4, i32 7, i32 2, i32 5>620 store <4 x double> %r1, ptr %out1, align 32621 %t2 = shufflevector <4 x double> %v0, <4 x double> %v1, <4 x i32> <i32 3, i32 7, i32 undef, i32 undef>622 %r2 = shufflevector <4 x double> %t2, <4 x double> %v2, <4 x i32> <i32 6, i32 0, i32 1, i32 7>623 store <4 x double> %r2, ptr %out2, align 32624 ret void625}626 627define i32 @PR164107(<16 x i1> %0) {628; AVX1-LABEL: PR164107:629; AVX1: # %bb.0:630; AVX1-NEXT: vpsllw $15, %xmm0, %xmm0631; AVX1-NEXT: vpsraw $15, %xmm0, %xmm0632; AVX1-NEXT: vpmovsxwq %xmm0, %xmm0633; AVX1-NEXT: vmovd %xmm0, %eax634; AVX1-NEXT: ret{{[l|q]}}635;636; AVX2-LABEL: PR164107:637; AVX2: # %bb.0:638; AVX2-NEXT: vpbroadcastb %xmm0, %xmm0639; AVX2-NEXT: vpsllw $15, %xmm0, %xmm0640; AVX2-NEXT: vpsraw $15, %xmm0, %xmm0641; AVX2-NEXT: vpmovsxwq %xmm0, %ymm0642; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0643; AVX2-NEXT: vmovd %xmm0, %eax644; AVX2-NEXT: vzeroupper645; AVX2-NEXT: ret{{[l|q]}}646;647; AVX512-LABEL: PR164107:648; AVX512: # %bb.0:649; AVX512-NEXT: vpmovsxbd %xmm0, %zmm0650; AVX512-NEXT: vpslld $31, %zmm0, %zmm0651; AVX512-NEXT: vptestmd %zmm0, %zmm0, %k1652; AVX512-NEXT: vpternlogq {{.*#+}} zmm0 {%k1} {z} = -1653; AVX512-NEXT: vpbroadcastq %xmm0, %zmm0654; AVX512-NEXT: vptestmq %zmm0, %zmm0, %k1655; AVX512-NEXT: vpternlogq {{.*#+}} zmm0 {%k1} {z} = -1656; AVX512-NEXT: vextracti32x4 $3, %zmm0, %xmm0657; AVX512-NEXT: vpbroadcastw %xmm0, %xmm0658; AVX512-NEXT: vpmovsxwq %xmm0, %zmm0659; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm0660; AVX512-NEXT: vmovd %xmm0, %eax661; AVX512-NEXT: vzeroupper662; AVX512-NEXT: ret{{[l|q]}}663 %cmp = shufflevector <16 x i1> %0, <16 x i1> zeroinitializer, <16 x i32> zeroinitializer664 %sext = sext <16 x i1> %cmp to <16 x i64>665 %bc.1 = bitcast <16 x i64> %sext to <64 x i16>666 %vecinit15.i = shufflevector <64 x i16> %bc.1, <64 x i16> zeroinitializer, <16 x i32> <i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56>667 %conv16.i = sext <16 x i16> %vecinit15.i to <16 x i64>668 %bc.2 = bitcast <16 x i64> %conv16.i to <32 x i32>669 %conv22.i = extractelement <32 x i32> %bc.2, i64 4670 ret i32 %conv22.i671}672 673define <4 x i64> @concat_self_v4i64(<2 x i64> %x) {674; AVX1-LABEL: concat_self_v4i64:675; AVX1: # %bb.0:676; AVX1-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0677; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0678; AVX1-NEXT: vshufpd {{.*#+}} ymm0 = ymm0[0,0,3,3]679; AVX1-NEXT: ret{{[l|q]}}680;681; AVX2-LABEL: concat_self_v4i64:682; AVX2: # %bb.0:683; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0684; AVX2-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,1]685; AVX2-NEXT: ret{{[l|q]}}686;687; AVX512-LABEL: concat_self_v4i64:688; AVX512: # %bb.0:689; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0690; AVX512-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,1]691; AVX512-NEXT: ret{{[l|q]}}692 %cat = shufflevector <2 x i64> %x, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>693 %s = shufflevector <4 x i64> %cat, <4 x i64> undef, <4 x i32> <i32 0, i32 2, i32 1, i32 3>694 ret <4 x i64> %s695}696 697define <8 x i32> @concat_self_v8i32(<4 x i32> %x) {698; AVX1-LABEL: concat_self_v8i32:699; AVX1: # %bb.0:700; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[3,2,1,0]701; AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm0[0,2,1,3]702; AVX1-NEXT: vpaddd %xmm0, %xmm2, %xmm2703; AVX1-NEXT: vpaddd %xmm0, %xmm1, %xmm0704; AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0705; AVX1-NEXT: ret{{[l|q]}}706;707; AVX2-LABEL: concat_self_v8i32:708; AVX2: # %bb.0:709; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0710; AVX2-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm1711; AVX2-NEXT: vpmovsxbd {{.*#+}} ymm2 = [3,2,1,0,0,2,1,3]712; AVX2-NEXT: vpermd %ymm0, %ymm2, %ymm0713; AVX2-NEXT: vpaddd %ymm1, %ymm0, %ymm0714; AVX2-NEXT: ret{{[l|q]}}715;716; AVX512-LABEL: concat_self_v8i32:717; AVX512: # %bb.0:718; AVX512-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0719; AVX512-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm1720; AVX512-NEXT: vpmovsxbd {{.*#+}} ymm2 = [3,2,1,0,0,2,1,3]721; AVX512-NEXT: vpermd %ymm0, %ymm2, %ymm0722; AVX512-NEXT: vpaddd %ymm1, %ymm0, %ymm0723; AVX512-NEXT: ret{{[l|q]}}724 %cat = shufflevector <4 x i32> %x, <4 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>725 %s = shufflevector <8 x i32> %cat, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 0, i32 2, i32 1, i32 3>726 %a = add <8 x i32> %s, %cat727 ret <8 x i32> %a728}729 730define <4 x double> @concat_vpermilvar_v4f64_v2f64(<2 x double> %a0, <2 x double> %a1, <4 x i64> %m) {731; CHECK-LABEL: concat_vpermilvar_v4f64_v2f64:732; CHECK: # %bb.0:733; CHECK-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0734; CHECK-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0735; CHECK-NEXT: vpermilpd %ymm2, %ymm0, %ymm0736; CHECK-NEXT: ret{{[l|q]}}737 %m0 = shufflevector <4 x i64> %m, <4 x i64> poison, <2 x i32> <i32 0, i32 1>738 %m1 = shufflevector <4 x i64> %m, <4 x i64> poison, <2 x i32> <i32 2, i32 3>739 %v0 = tail call noundef <2 x double> @llvm.x86.avx.vpermilvar.pd(<2 x double> %a0, <2 x i64> %m0)740 %v1 = tail call noundef <2 x double> @llvm.x86.avx.vpermilvar.pd(<2 x double> %a1, <2 x i64> %m1)741 %res = shufflevector <2 x double> %v0, <2 x double> %v1, <4 x i32> <i32 0, i32 1, i32 2, i32 3>742 ret <4 x double> %res743}744 745define <16 x i64> @bit_reversal_permutation(<16 x i64> %a0) nounwind {746; X86-AVX1-LABEL: bit_reversal_permutation:747; X86-AVX1: # %bb.0:748; X86-AVX1-NEXT: pushl %ebp749; X86-AVX1-NEXT: movl %esp, %ebp750; X86-AVX1-NEXT: andl $-32, %esp751; X86-AVX1-NEXT: subl $32, %esp752; X86-AVX1-NEXT: vmovaps 8(%ebp), %ymm5753; X86-AVX1-NEXT: vperm2f128 {{.*#+}} ymm3 = ymm2[2,3],ymm5[2,3]754; X86-AVX1-NEXT: vperm2f128 {{.*#+}} ymm6 = ymm0[2,3],ymm1[2,3]755; X86-AVX1-NEXT: vunpcklpd {{.*#+}} ymm4 = ymm6[0],ymm3[0],ymm6[2],ymm3[2]756; X86-AVX1-NEXT: vunpckhpd {{.*#+}} ymm3 = ymm6[1],ymm3[1],ymm6[3],ymm3[3]757; X86-AVX1-NEXT: vinsertf128 $1, %xmm5, %ymm2, %ymm2758; X86-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm1759; X86-AVX1-NEXT: vunpcklpd {{.*#+}} ymm0 = ymm1[0],ymm2[0],ymm1[2],ymm2[2]760; X86-AVX1-NEXT: vunpckhpd {{.*#+}} ymm2 = ymm1[1],ymm2[1],ymm1[3],ymm2[3]761; X86-AVX1-NEXT: vmovaps %ymm4, %ymm1762; X86-AVX1-NEXT: movl %ebp, %esp763; X86-AVX1-NEXT: popl %ebp764; X86-AVX1-NEXT: retl765;766; X86-AVX2-LABEL: bit_reversal_permutation:767; X86-AVX2: # %bb.0:768; X86-AVX2-NEXT: pushl %ebp769; X86-AVX2-NEXT: movl %esp, %ebp770; X86-AVX2-NEXT: andl $-32, %esp771; X86-AVX2-NEXT: subl $32, %esp772; X86-AVX2-NEXT: vmovaps 8(%ebp), %ymm6773; X86-AVX2-NEXT: vperm2f128 {{.*#+}} ymm3 = ymm0[2,3],ymm1[2,3]774; X86-AVX2-NEXT: vunpcklpd {{.*#+}} ymm4 = ymm2[0],ymm6[0],ymm2[2],ymm6[2]775; X86-AVX2-NEXT: vpermpd {{.*#+}} ymm4 = ymm4[0,2,2,3]776; X86-AVX2-NEXT: vblendps {{.*#+}} ymm4 = ymm3[0,1],ymm4[2,3],ymm3[4,5],ymm4[6,7]777; X86-AVX2-NEXT: vperm2f128 {{.*#+}} ymm3 = ymm2[2,3],ymm6[2,3]778; X86-AVX2-NEXT: vunpckhpd {{.*#+}} ymm5 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]779; X86-AVX2-NEXT: vpermpd {{.*#+}} ymm5 = ymm5[2,1,3,3]780; X86-AVX2-NEXT: vblendps {{.*#+}} ymm3 = ymm5[0,1],ymm3[2,3],ymm5[4,5],ymm3[6,7]781; X86-AVX2-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm5782; X86-AVX2-NEXT: vmovlhps {{.*#+}} xmm7 = xmm2[0],xmm6[0]783; X86-AVX2-NEXT: vpermpd {{.*#+}} ymm7 = ymm7[0,0,2,1]784; X86-AVX2-NEXT: vblendps {{.*#+}} ymm5 = ymm5[0,1],ymm7[2,3],ymm5[4,5],ymm7[6,7]785; X86-AVX2-NEXT: vinsertf128 $1, %xmm6, %ymm2, %ymm2786; X86-AVX2-NEXT: vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]787; X86-AVX2-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,3]788; X86-AVX2-NEXT: vblendps {{.*#+}} ymm2 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]789; X86-AVX2-NEXT: vmovaps %ymm5, %ymm0790; X86-AVX2-NEXT: vmovaps %ymm4, %ymm1791; X86-AVX2-NEXT: movl %ebp, %esp792; X86-AVX2-NEXT: popl %ebp793; X86-AVX2-NEXT: retl794;795; AVX512-LABEL: bit_reversal_permutation:796; AVX512: # %bb.0:797; AVX512-NEXT: vpmovsxbq {{.*#+}} zmm2 = [0,8,4,12,2,10,6,14]798; AVX512-NEXT: vpermi2q %zmm1, %zmm0, %zmm2799; AVX512-NEXT: vpmovsxbq {{.*#+}} zmm3 = [1,9,5,13,3,11,7,15]800; AVX512-NEXT: vpermi2q %zmm1, %zmm0, %zmm3801; AVX512-NEXT: vmovdqa64 %zmm2, %zmm0802; AVX512-NEXT: vmovdqa64 %zmm3, %zmm1803; AVX512-NEXT: ret{{[l|q]}}804;805; X64-AVX1-LABEL: bit_reversal_permutation:806; X64-AVX1: # %bb.0:807; X64-AVX1-NEXT: vperm2f128 {{.*#+}} ymm5 = ymm2[2,3],ymm3[2,3]808; X64-AVX1-NEXT: vperm2f128 {{.*#+}} ymm6 = ymm0[2,3],ymm1[2,3]809; X64-AVX1-NEXT: vunpcklpd {{.*#+}} ymm4 = ymm6[0],ymm5[0],ymm6[2],ymm5[2]810; X64-AVX1-NEXT: vunpckhpd {{.*#+}} ymm5 = ymm6[1],ymm5[1],ymm6[3],ymm5[3]811; X64-AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm2, %ymm2812; X64-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm1813; X64-AVX1-NEXT: vunpcklpd {{.*#+}} ymm0 = ymm1[0],ymm2[0],ymm1[2],ymm2[2]814; X64-AVX1-NEXT: vunpckhpd {{.*#+}} ymm2 = ymm1[1],ymm2[1],ymm1[3],ymm2[3]815; X64-AVX1-NEXT: vmovaps %ymm4, %ymm1816; X64-AVX1-NEXT: vmovaps %ymm5, %ymm3817; X64-AVX1-NEXT: retq818;819; X64-AVX2-LABEL: bit_reversal_permutation:820; X64-AVX2: # %bb.0:821; X64-AVX2-NEXT: vperm2f128 {{.*#+}} ymm4 = ymm0[2,3],ymm1[2,3]822; X64-AVX2-NEXT: vunpcklpd {{.*#+}} ymm5 = ymm2[0],ymm3[0],ymm2[2],ymm3[2]823; X64-AVX2-NEXT: vpermpd {{.*#+}} ymm5 = ymm5[0,2,2,3]824; X64-AVX2-NEXT: vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm5[2,3],ymm4[4,5],ymm5[6,7]825; X64-AVX2-NEXT: vperm2f128 {{.*#+}} ymm5 = ymm2[2,3],ymm3[2,3]826; X64-AVX2-NEXT: vunpckhpd {{.*#+}} ymm6 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]827; X64-AVX2-NEXT: vpermpd {{.*#+}} ymm6 = ymm6[2,1,3,3]828; X64-AVX2-NEXT: vblendps {{.*#+}} ymm5 = ymm6[0,1],ymm5[2,3],ymm6[4,5],ymm5[6,7]829; X64-AVX2-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm6830; X64-AVX2-NEXT: vmovlhps {{.*#+}} xmm7 = xmm2[0],xmm3[0]831; X64-AVX2-NEXT: vpermpd {{.*#+}} ymm7 = ymm7[0,0,2,1]832; X64-AVX2-NEXT: vblendps {{.*#+}} ymm6 = ymm6[0,1],ymm7[2,3],ymm6[4,5],ymm7[6,7]833; X64-AVX2-NEXT: vinsertf128 $1, %xmm3, %ymm2, %ymm2834; X64-AVX2-NEXT: vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]835; X64-AVX2-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,3]836; X64-AVX2-NEXT: vblendps {{.*#+}} ymm2 = ymm0[0,1],ymm2[2,3],ymm0[4,5],ymm2[6,7]837; X64-AVX2-NEXT: vmovaps %ymm6, %ymm0838; X64-AVX2-NEXT: vmovaps %ymm4, %ymm1839; X64-AVX2-NEXT: vmovaps %ymm5, %ymm3840; X64-AVX2-NEXT: retq841 %v0 = shufflevector <16 x i64> %a0, <16 x i64> undef, <16 x i32> <i32 0, i32 1, i32 4, i32 5, i32 2, i32 3, i32 6, i32 7, i32 8, i32 9, i32 12, i32 13, i32 10, i32 11, i32 14, i32 15>842 %v1 = shufflevector <16 x i64> %v0, <16 x i64> undef, <16 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14, i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>843 ret <16 x i64> %v1844}845