1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx,+xop | FileCheck %s --check-prefix=X32 3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+xop | FileCheck %s --check-prefix=X64 4 5; 6; VPPERM 7; 8 9define <16 x i8> @vpperm_shuffle_unary(<16 x i8> %a0) { 10; X32-LABEL: vpperm_shuffle_unary: 11; X32: # %bb.0: 12; X32-NEXT: vpperm {{.*#+}} xmm0 = xmm0[15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0] 13; X32-NEXT: retl 14; 15; X64-LABEL: vpperm_shuffle_unary: 16; X64: # %bb.0: 17; X64-NEXT: vpperm {{.*#+}} xmm0 = xmm0[15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0] 18; X64-NEXT: retq 19 %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>) 20 ret <16 x i8> %1 21} 22 23define <16 x i8> @vpperm_shuffle_unary_undef(<16 x i8> %a0) { 24; X32-LABEL: vpperm_shuffle_unary_undef: 25; X32: # %bb.0: 26; X32-NEXT: vpperm {{.*#+}} xmm0 = xmm0[15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0] 27; X32-NEXT: retl 28; 29; X64-LABEL: vpperm_shuffle_unary_undef: 30; X64: # %bb.0: 31; X64-NEXT: vpperm {{.*#+}} xmm0 = xmm0[15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0] 32; X64-NEXT: retq 33 %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>) 34 ret <16 x i8> %1 35} 36 37define <16 x i8> @vpperm_shuffle_unary_zero(<16 x i8> %a0) { 38; X32-LABEL: vpperm_shuffle_unary_zero: 39; X32: # %bb.0: 40; X32-NEXT: vpperm {{.*#+}} xmm0 = xmm0[15,14,13,12,11,10,9,8,7,6,5,4,3],zero,xmm0[1],zero 41; X32-NEXT: retl 42; 43; X64-LABEL: vpperm_shuffle_unary_zero: 44; X64: # %bb.0: 45; X64-NEXT: vpperm {{.*#+}} xmm0 = xmm0[15,14,13,12,11,10,9,8,7,6,5,4,3],zero,xmm0[1],zero 46; X64-NEXT: retq 47 %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>) 48 ret <16 x i8> %1 49} 50 51define <16 x i8> @vpperm_shuffle_binary(<16 x i8> %a0, <16 x i8> %a1) { 52; X32-LABEL: vpperm_shuffle_binary: 53; X32: # %bb.0: 54; X32-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] 55; X32-NEXT: retl 56; 57; X64-LABEL: vpperm_shuffle_binary: 58; X64: # %bb.0: 59; X64-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] 60; X64-NEXT: retq 61 %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>) 62 ret <16 x i8> %1 63} 64 65define <16 x i8> @vpperm_shuffle_binary_zero(<16 x i8> %a0, <16 x i8> %a1) { 66; X32-LABEL: vpperm_shuffle_binary_zero: 67; X32: # %bb.0: 68; X32-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,zero 69; X32-NEXT: retl 70; 71; X64-LABEL: vpperm_shuffle_binary_zero: 72; X64: # %bb.0: 73; X64-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,zero 74; X64-NEXT: retq 75 %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>) 76 ret <16 x i8> %1 77} 78 79; we can't decode vpperm's other permute ops 80define <16 x i8> @vpperm_shuffle_general(<16 x i8> %a0, <16 x i8> %a1) { 81; X32-LABEL: vpperm_shuffle_general: 82; X32: # %bb.0: 83; X32-NEXT: vpperm {{\.LCPI.*}}, %xmm0, %xmm0, %xmm0 84; X32-NEXT: retl 85; 86; X64-LABEL: vpperm_shuffle_general: 87; X64: # %bb.0: 88; X64-NEXT: vpperm {{.*}}(%rip), %xmm0, %xmm0, %xmm0 89; X64-NEXT: retq 90 %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>) 91 ret <16 x i8> %1 92} 93 94; 95; VPERMIL2 96; 97 98; Note: _mm_permute2_pd shouldn't be used for constant shuffles as there will always 99; be a quicker (and smaller) alternative. 100define <2 x double> @vpermil2pd_21(<2 x double> %a0, <2 x double> %a1) { 101; X32-LABEL: vpermil2pd_21: 102; X32: # %bb.0: 103; X32-NEXT: vxorps %xmm1, %xmm1, %xmm1 104; X32-NEXT: vmovlhps {{.*#+}} xmm0 = xmm1[0],xmm0[0] 105; X32-NEXT: retl 106; 107; X64-LABEL: vpermil2pd_21: 108; X64: # %bb.0: 109; X64-NEXT: vxorps %xmm1, %xmm1, %xmm1 110; X64-NEXT: vmovlhps {{.*#+}} xmm0 = xmm1[0],xmm0[0] 111; X64-NEXT: retq 112 %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) 113 ret <2 x double> %1 114} 115 116define <4 x double> @vpermil2pd256_0062(<4 x double> %a0, <4 x double> %a1) { 117; X32-LABEL: vpermil2pd256_0062: 118; X32: # %bb.0: 119; X32-NEXT: vpermil2pd {{.*#+}} ymm0 = ymm0[0,0],ymm1[2],ymm0[2] 120; X32-NEXT: retl 121; 122; X64-LABEL: vpermil2pd256_0062: 123; X64: # %bb.0: 124; X64-NEXT: vpermil2pd {{.*#+}} ymm0 = ymm0[0,0],ymm1[2],ymm0[2] 125; X64-NEXT: retq 126 %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) 127 ret <4 x double> %1 128} 129 130define <4 x double> @vpermil2pd256_zz73(<4 x double> %a0, <4 x double> %a1) { 131; X32-LABEL: vpermil2pd256_zz73: 132; X32: # %bb.0: 133; X32-NEXT: vpermil2pd {{.*#+}} ymm0 = zero,zero,ymm1[3],ymm0[3] 134; X32-NEXT: retl 135; 136; X64-LABEL: vpermil2pd256_zz73: 137; X64: # %bb.0: 138; X64-NEXT: vpermil2pd {{.*#+}} ymm0 = zero,zero,ymm1[3],ymm0[3] 139; X64-NEXT: retq 140 %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) 141 ret <4 x double> %1 142} 143 144define <4 x float> @vpermil2ps_0561(<4 x float> %a0, <4 x float> %a1) { 145; X32-LABEL: vpermil2ps_0561: 146; X32: # %bb.0: 147; X32-NEXT: vpermil2ps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[1] 148; X32-NEXT: retl 149; 150; X64-LABEL: vpermil2ps_0561: 151; X64: # %bb.0: 152; X64-NEXT: vpermil2ps {{.*#+}} xmm0 = xmm0[0],xmm1[1,2],xmm0[1] 153; X64-NEXT: retq 154 %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) 155 ret <4 x float> %1 156} 157 158define <8 x float> @vpermil2ps256_098144FE(<8 x float> %a0, <8 x float> %a1) { 159; X32-LABEL: vpermil2ps256_098144FE: 160; X32: # %bb.0: 161; X32-NEXT: vpermil2ps {{.*#+}} ymm0 = ymm0[0],ymm1[1,0],ymm0[1,4,4],ymm1[7,6] 162; X32-NEXT: retl 163; 164; X64-LABEL: vpermil2ps256_098144FE: 165; X64: # %bb.0: 166; X64-NEXT: vpermil2ps {{.*#+}} ymm0 = ymm0[0],ymm1[1,0],ymm0[1,4,4],ymm1[7,6] 167; X64-NEXT: retq 168 %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) 169 ret <8 x float> %1 170} 171 172define <8 x float> @vpermil2ps256_0zz8BzzA(<8 x float> %a0, <8 x float> %a1) { 173; X32-LABEL: vpermil2ps256_0zz8BzzA: 174; X32: # %bb.0: 175; X32-NEXT: vpermil2ps {{.*#+}} ymm0 = ymm0[0],zero,zero,ymm1[0,7],zero,zero,ymm1[6] 176; X32-NEXT: retl 177; 178; X64-LABEL: vpermil2ps256_0zz8BzzA: 179; X64: # %bb.0: 180; X64-NEXT: vpermil2ps {{.*#+}} ymm0 = ymm0[0],zero,zero,ymm1[0,7],zero,zero,ymm1[6] 181; X64-NEXT: retq 182 %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) 183 ret <8 x float> %1 184} 185 186declare <2 x double> @llvm.x86.xop.vpermil2pd(<2 x double>, <2 x double>, <2 x i64>, i8) nounwind readnone 187declare <4 x double> @llvm.x86.xop.vpermil2pd.256(<4 x double>, <4 x double>, <4 x i64>, i8) nounwind readnone 188 189declare <4 x float> @llvm.x86.xop.vpermil2ps(<4 x float>, <4 x float>, <4 x i32>, i8) nounwind readnone 190declare <8 x float> @llvm.x86.xop.vpermil2ps.256(<8 x float>, <8 x float>, <8 x i32>, i8) nounwind readnone 191 192declare <16 x i8> @llvm.x86.xop.vpperm(<16 x i8>, <16 x i8>, <16 x i8>) nounwind readnone 193