137 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx,+xop | FileCheck %s --check-prefixes=CHECK,X863; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+xop | FileCheck %s --check-prefixes=CHECK,X644 5;6; VPPERM7;8 9define <16 x i8> @vpperm_shuffle_unary(<16 x i8> %a0) {10; CHECK-LABEL: vpperm_shuffle_unary:11; CHECK: # %bb.0:12; CHECK-NEXT: vpperm {{.*#+}} xmm0 = xmm0[15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0]13; CHECK-NEXT: ret{{[l|q]}}14 %1 = tail call <16 x i8> @llvm.x86.xop.vpperm(<16 x i8> %a0, <16 x i8> %a0, <16 x i8> <i8 31, i8 14, i8 29, i8 12, i8 27, i8 10, i8 25, i8 8, i8 23, i8 6, i8 21, i8 4, i8 19, i8 2, i8 17, i8 0>)15 ret <16 x i8> %116}17 18define <16 x i8> @vpperm_shuffle_unary_undef(<16 x i8> %a0) {19; CHECK-LABEL: vpperm_shuffle_unary_undef:20; CHECK: # %bb.0:21; CHECK-NEXT: vpperm {{.*#+}} xmm0 = xmm0[15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0]22; CHECK-NEXT: ret{{[l|q]}}23 %1 = tail call <16 x i8> @llvm.x86.xop.vpperm(<16 x i8> %a0, <16 x i8> undef, <16 x i8> <i8 31, i8 14, i8 29, i8 12, i8 27, i8 10, i8 25, i8 8, i8 23, i8 6, i8 21, i8 4, i8 19, i8 2, i8 17, i8 0>)24 ret <16 x i8> %125}26 27define <16 x i8> @vpperm_shuffle_unary_zero(<16 x i8> %a0) {28; CHECK-LABEL: vpperm_shuffle_unary_zero:29; CHECK: # %bb.0:30; CHECK-NEXT: vpperm {{.*#+}} xmm0 = xmm0[15,14,13,12,11,10,9,8,7,6,5,4,3],zero,xmm0[1],zero31; CHECK-NEXT: ret{{[l|q]}}32 %1 = tail call <16 x i8> @llvm.x86.xop.vpperm(<16 x i8> %a0, <16 x i8> %a0, <16 x i8> <i8 31, i8 14, i8 29, i8 12, i8 27, i8 10, i8 25, i8 8, i8 23, i8 6, i8 21, i8 4, i8 19, i8 130, i8 17, i8 128>)33 ret <16 x i8> %134}35 36define <16 x i8> @vpperm_shuffle_binary(<16 x i8> %a0, <16 x i8> %a1) {37; CHECK-LABEL: vpperm_shuffle_binary:38; CHECK: # %bb.0:39; CHECK-NEXT: vpperm {{.*#+}} xmm0 = xmm1[15],xmm0[14],xmm1[13],xmm0[12],xmm1[11],xmm0[10],xmm1[9],xmm0[8],xmm1[7],xmm0[6],xmm1[5],xmm0[4],xmm1[3],xmm0[2],xmm1[1],xmm0[0]40; CHECK-NEXT: ret{{[l|q]}}41 %1 = tail call <16 x i8> @llvm.x86.xop.vpperm(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> <i8 31, i8 14, i8 29, i8 12, i8 27, i8 10, i8 25, i8 8, i8 23, i8 6, i8 21, i8 4, i8 19, i8 2, i8 17, i8 0>)42 ret <16 x i8> %143}44 45define <16 x i8> @vpperm_shuffle_binary_zero(<16 x i8> %a0, <16 x i8> %a1) {46; CHECK-LABEL: vpperm_shuffle_binary_zero:47; CHECK: # %bb.0:48; CHECK-NEXT: vpperm {{.*#+}} xmm0 = xmm1[15],xmm0[14],xmm1[13],xmm0[12],xmm1[11],xmm0[10],xmm1[9],xmm0[8],xmm1[7],xmm0[6],xmm1[5],xmm0[4],zero,zero,zero,zero49; CHECK-NEXT: ret{{[l|q]}}50 %1 = tail call <16 x i8> @llvm.x86.xop.vpperm(<16 x i8> %a0, <16 x i8> %a1, <16 x i8> <i8 31, i8 14, i8 29, i8 12, i8 27, i8 10, i8 25, i8 8, i8 23, i8 6, i8 21, i8 4, i8 147, i8 130, i8 145, i8 128>)51 ret <16 x i8> %152}53 54; we can't decode vpperm's other permute ops55define <16 x i8> @vpperm_shuffle_general(<16 x i8> %a0, <16 x i8> %a1) {56; X86-LABEL: vpperm_shuffle_general:57; X86: # %bb.0:58; X86-NEXT: vpperm {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0, %xmm0, %xmm059; X86-NEXT: retl60;61; X64-LABEL: vpperm_shuffle_general:62; X64: # %bb.0:63; X64-NEXT: vpperm {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0, %xmm064; X64-NEXT: retq65 %1 = tail call <16 x i8> @llvm.x86.xop.vpperm(<16 x i8> %a0, <16 x i8> %a0, <16 x i8> <i8 31, i8 14, i8 29, i8 12, i8 27, i8 10, i8 25, i8 8, i8 23, i8 6, i8 21, i8 4, i8 179, i8 162, i8 177, i8 160>)66 ret <16 x i8> %167}68 69;70; VPERMIL271;72 73; Note: _mm_permute2_pd shouldn't be used for constant shuffles as there will always74; be a quicker (and smaller) alternative.75define <2 x double> @vpermil2pd_21(<2 x double> %a0, <2 x double> %a1) {76; CHECK-LABEL: vpermil2pd_21:77; CHECK: # %bb.0:78; CHECK-NEXT: vxorps %xmm1, %xmm1, %xmm179; CHECK-NEXT: vmovlhps {{.*#+}} xmm0 = xmm1[0],xmm0[0]80; CHECK-NEXT: ret{{[l|q]}}81 %1 = call <2 x double> @llvm.x86.xop.vpermil2pd(<2 x double> %a0, <2 x double> %a1, <2 x i64> <i64 10, i64 1>, i8 2)82 ret <2 x double> %183}84 85define <4 x double> @vpermil2pd256_0062(<4 x double> %a0, <4 x double> %a1) {86; CHECK-LABEL: vpermil2pd256_0062:87; CHECK: # %bb.0:88; CHECK-NEXT: vpermil2pd {{.*#+}} ymm0 = ymm0[0,0],ymm1[2],ymm0[2]89; CHECK-NEXT: ret{{[l|q]}}90 %1 = call <4 x double> @llvm.x86.xop.vpermil2pd.256(<4 x double> %a0, <4 x double> %a1, <4 x i64> <i64 0, i64 0, i64 4, i64 0>, i8 0)91 ret <4 x double> %192}93 94define <4 x double> @vpermil2pd256_zz73(<4 x double> %a0, <4 x double> %a1) {95; CHECK-LABEL: vpermil2pd256_zz73:96; CHECK: # %bb.0:97; CHECK-NEXT: vpermil2pd {{.*#+}} ymm0 = zero,zero,ymm1[3],ymm0[3]98; CHECK-NEXT: ret{{[l|q]}}99 %1 = call <4 x double> @llvm.x86.xop.vpermil2pd.256(<4 x double> %a0, <4 x double> %a1, <4 x i64> <i64 0, i64 0, i64 14, i64 10>, i8 3)100 ret <4 x double> %1101}102 103define <4 x float> @vpermil2ps_0561(<4 x float> %a0, <4 x float> %a1) {104; CHECK-LABEL: vpermil2ps_0561:105; CHECK: # %bb.0:106; CHECK-NEXT: vpermil2ps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[1]107; CHECK-NEXT: ret{{[l|q]}}108 %1 = call <4 x float> @llvm.x86.xop.vpermil2ps(<4 x float> %a0, <4 x float> %a1, <4 x i32> <i32 0, i32 5, i32 6, i32 1>, i8 0)109 ret <4 x float> %1110}111 112define <8 x float> @vpermil2ps256_098144FE(<8 x float> %a0, <8 x float> %a1) {113; CHECK-LABEL: vpermil2ps256_098144FE:114; CHECK: # %bb.0:115; CHECK-NEXT: vpermil2ps {{.*#+}} ymm0 = ymm0[0],ymm1[1,0],ymm0[1,4,4],ymm1[7,6]116; CHECK-NEXT: ret{{[l|q]}}117 %1 = call <8 x float> @llvm.x86.xop.vpermil2ps.256(<8 x float> %a0, <8 x float> %a1, <8 x i32> <i32 0, i32 5, i32 4, i32 1, i32 0, i32 0, i32 7, i32 6>, i8 0)118 ret <8 x float> %1119}120 121define <8 x float> @vpermil2ps256_0zz8BzzA(<8 x float> %a0, <8 x float> %a1) {122; CHECK-LABEL: vpermil2ps256_0zz8BzzA:123; CHECK: # %bb.0:124; CHECK-NEXT: vpermil2ps {{.*#+}} ymm0 = ymm0[0],zero,zero,ymm1[0,7],zero,zero,ymm1[6]125; CHECK-NEXT: ret{{[l|q]}}126 %1 = call <8 x float> @llvm.x86.xop.vpermil2ps.256(<8 x float> %a0, <8 x float> %a1, <8 x i32> <i32 0, i32 8, i32 8, i32 4, i32 7, i32 8, i32 8, i32 6>, i8 2)127 ret <8 x float> %1128}129 130declare <2 x double> @llvm.x86.xop.vpermil2pd(<2 x double>, <2 x double>, <2 x i64>, i8) nounwind readnone131declare <4 x double> @llvm.x86.xop.vpermil2pd.256(<4 x double>, <4 x double>, <4 x i64>, i8) nounwind readnone132 133declare <4 x float> @llvm.x86.xop.vpermil2ps(<4 x float>, <4 x float>, <4 x i32>, i8) nounwind readnone134declare <8 x float> @llvm.x86.xop.vpermil2ps.256(<8 x float>, <8 x float>, <8 x i32>, i8) nounwind readnone135 136declare <16 x i8> @llvm.x86.xop.vpperm(<16 x i8>, <16 x i8>, <16 x i8>) nounwind readnone137