225 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=icelake-server | FileCheck %s --check-prefixes=CHECK,CHECK-ICX,CHECK-ICX-NO-BYPASS-DELAY3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=icelake-server -mattr=-no-bypass-delay-shuffle | FileCheck %s --check-prefixes=CHECK,CHECK-ICX,CHECK-ICX-BYPASS-DELAY4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=CHECK,CHECK-V45; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512bw,+avx512dq | FileCheck %s --check-prefixes=CHECK,CHECK-AVX5126; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=znver4 | FileCheck %s --check-prefixes=CHECK,CHECK-ZNVER47; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=znver5 | FileCheck %s --check-prefixes=CHECK,CHECK-ZNVER48 9define <8 x double> @transform_VPERMILPSZrr(<8 x double> %a) nounwind {10; CHECK-LABEL: transform_VPERMILPSZrr:11; CHECK: # %bb.0:12; CHECK-NEXT: vshufpd {{.*#+}} zmm0 = zmm0[0,1,2,3,4,5,7,6]13; CHECK-NEXT: retq14 %shufp = shufflevector <8 x double> %a, <8 x double> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 7, i32 6>15 ret <8 x double> %shufp16}17 18define <4 x double> @transform_VPERMILPSYrr(<4 x double> %a) nounwind {19; CHECK-LABEL: transform_VPERMILPSYrr:20; CHECK: # %bb.0:21; CHECK-NEXT: vshufpd {{.*#+}} ymm0 = ymm0[1,0,2,3]22; CHECK-NEXT: retq23 %shufp = shufflevector <4 x double> %a, <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 2, i32 3>24 ret <4 x double> %shufp25}26 27define <2 x double> @transform_VPERMILPSrr(<2 x double> %a) nounwind {28; CHECK-LABEL: transform_VPERMILPSrr:29; CHECK: # %bb.0:30; CHECK-NEXT: vshufpd {{.*#+}} xmm0 = xmm0[1,0]31; CHECK-NEXT: retq32 %shufp = shufflevector <2 x double> %a, <2 x double> poison, <2 x i32> <i32 1, i32 0>33 ret <2 x double> %shufp34}35 36define <8 x double> @transform_VPERMILPSZrrkz(<8 x double> %a, i8 %mask_int) nounwind {37; CHECK-LABEL: transform_VPERMILPSZrrkz:38; CHECK: # %bb.0:39; CHECK-NEXT: kmovd %edi, %k140; CHECK-NEXT: vshufpd {{.*#+}} zmm0 {%k1} {z} = zmm0[0,1,2,3,5,4,6,7]41; CHECK-NEXT: retq42 %mask = bitcast i8 %mask_int to <8 x i1>43 %shufp = shufflevector <8 x double> %a, <8 x double> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7>44 %res = select <8 x i1> %mask, <8 x double> %shufp, <8 x double> zeroinitializer45 ret <8 x double> %res46}47 48define <4 x double> @transform_VPERMILPSYrrkz(<4 x double> %a, i4 %mask_int) nounwind {49; CHECK-LABEL: transform_VPERMILPSYrrkz:50; CHECK: # %bb.0:51; CHECK-NEXT: kmovd %edi, %k152; CHECK-NEXT: vshufpd {{.*#+}} ymm0 {%k1} {z} = ymm0[0,1,3,2]53; CHECK-NEXT: retq54 %mask = bitcast i4 %mask_int to <4 x i1>55 %shufp = shufflevector <4 x double> %a, <4 x double> poison, <4 x i32> <i32 0, i32 1, i32 3, i32 2>56 %res = select <4 x i1> %mask, <4 x double> %shufp, <4 x double> zeroinitializer57 ret <4 x double> %res58}59 60define <2 x double> @transform_VPERMILPSrrkz(<2 x double> %a, i2 %mask_int) nounwind {61; CHECK-LABEL: transform_VPERMILPSrrkz:62; CHECK: # %bb.0:63; CHECK-NEXT: kmovd %edi, %k164; CHECK-NEXT: vshufpd {{.*#+}} xmm0 {%k1} {z} = xmm0[1,0]65; CHECK-NEXT: retq66 %mask = bitcast i2 %mask_int to <2 x i1>67 %shufp = shufflevector <2 x double> %a, <2 x double> poison, <2 x i32> <i32 1, i32 0>68 %res = select <2 x i1> %mask, <2 x double> %shufp, <2 x double> zeroinitializer69 ret <2 x double> %res70}71 72define <8 x double> @transform_VPERMILPSZrrk(<8 x double> %a, <8 x double> %b, i8 %mask_int) nounwind {73; CHECK-LABEL: transform_VPERMILPSZrrk:74; CHECK: # %bb.0:75; CHECK-NEXT: kmovd %edi, %k176; CHECK-NEXT: vshufpd {{.*#+}} zmm1 {%k1} = zmm0[0,1,3,2,4,5,6,7]77; CHECK-NEXT: vmovapd %zmm1, %zmm078; CHECK-NEXT: retq79 %mask = bitcast i8 %mask_int to <8 x i1>80 %shufp = shufflevector <8 x double> %a, <8 x double> poison, <8 x i32> <i32 0, i32 1, i32 3, i32 2, i32 4, i32 5, i32 6, i32 7>81 %res = select <8 x i1> %mask, <8 x double> %shufp, <8 x double> %b82 ret <8 x double> %res83}84 85define <4 x double> @transform_VPERMILPSYrrk(<4 x double> %a, <4 x double> %b, i4 %mask_int) nounwind {86; CHECK-LABEL: transform_VPERMILPSYrrk:87; CHECK: # %bb.0:88; CHECK-NEXT: kmovd %edi, %k189; CHECK-NEXT: vshufpd {{.*#+}} ymm1 {%k1} = ymm0[1,0,3,2]90; CHECK-NEXT: vmovapd %ymm1, %ymm091; CHECK-NEXT: retq92 %mask = bitcast i4 %mask_int to <4 x i1>93 %shufp = shufflevector <4 x double> %a, <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 2>94 %res = select <4 x i1> %mask, <4 x double> %shufp, <4 x double> %b95 ret <4 x double> %res96}97 98define <2 x double> @transform_VPERMILPSrrk(<2 x double> %a, <2 x double> %b, i2 %mask_int) nounwind {99; CHECK-LABEL: transform_VPERMILPSrrk:100; CHECK: # %bb.0:101; CHECK-NEXT: kmovd %edi, %k1102; CHECK-NEXT: vshufpd {{.*#+}} xmm1 {%k1} = xmm0[1,0]103; CHECK-NEXT: vmovapd %xmm1, %xmm0104; CHECK-NEXT: retq105 %mask = bitcast i2 %mask_int to <2 x i1>106 %shufp = shufflevector <2 x double> %a, <2 x double> poison, <2 x i32> <i32 1, i32 0>107 %res = select <2 x i1> %mask, <2 x double> %shufp, <2 x double> %b108 ret <2 x double> %res109}110 111define <8 x double> @transform_VPERMILPSZrm(ptr %ap) nounwind {112; CHECK-LABEL: transform_VPERMILPSZrm:113; CHECK: # %bb.0:114; CHECK-NEXT: vpermilpd {{.*#+}} zmm0 = mem[1,0,2,3,4,5,6,7]115; CHECK-NEXT: retq116 %a = load <8 x double>, ptr %ap117 %shufp = shufflevector <8 x double> %a, <8 x double> poison, <8 x i32> <i32 1, i32 0, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>118 ret <8 x double> %shufp119}120 121define <4 x double> @transform_VPERMILPSYrm(ptr %ap) nounwind {122; CHECK-LABEL: transform_VPERMILPSYrm:123; CHECK: # %bb.0:124; CHECK-NEXT: vpermilpd {{.*#+}} ymm0 = mem[0,1,3,2]125; CHECK-NEXT: retq126 %a = load <4 x double>, ptr %ap127 %shufp = shufflevector <4 x double> %a, <4 x double> poison, <4 x i32> <i32 0, i32 1, i32 3, i32 2>128 ret <4 x double> %shufp129}130 131define <2 x double> @transform_VPERMILPSrm(ptr %ap) nounwind {132; CHECK-LABEL: transform_VPERMILPSrm:133; CHECK: # %bb.0:134; CHECK-NEXT: vpermilpd {{.*#+}} xmm0 = mem[1,0]135; CHECK-NEXT: retq136 %a = load <2 x double>, ptr %ap137 %shufp = shufflevector <2 x double> %a, <2 x double> poison, <2 x i32> <i32 1, i32 0>138 ret <2 x double> %shufp139}140 141define <8 x double> @transform_VPERMILPSZrmkz(ptr %ap, i8 %mask_int) nounwind {142; CHECK-LABEL: transform_VPERMILPSZrmkz:143; CHECK: # %bb.0:144; CHECK-NEXT: kmovd %esi, %k1145; CHECK-NEXT: vpermilpd {{.*#+}} zmm0 {%k1} {z} = mem[1,0,2,3,4,5,7,6]146; CHECK-NEXT: retq147 %mask = bitcast i8 %mask_int to <8 x i1>148 %a = load <8 x double>, ptr %ap149 %shufp = shufflevector <8 x double> %a, <8 x double> poison, <8 x i32> <i32 1, i32 0, i32 2, i32 3, i32 4, i32 5, i32 7, i32 6>150 %res = select <8 x i1> %mask, <8 x double> %shufp, <8 x double> zeroinitializer151 ret <8 x double> %res152}153 154define <4 x double> @transform_VPERMILPSYrmkz(ptr %ap, i4 %mask_int) nounwind {155; CHECK-LABEL: transform_VPERMILPSYrmkz:156; CHECK: # %bb.0:157; CHECK-NEXT: kmovd %esi, %k1158; CHECK-NEXT: vpermilpd {{.*#+}} ymm0 {%k1} {z} = mem[1,0,3,2]159; CHECK-NEXT: retq160 %mask = bitcast i4 %mask_int to <4 x i1>161 %a = load <4 x double>, ptr %ap162 %shufp = shufflevector <4 x double> %a, <4 x double> poison, <4 x i32> <i32 1, i32 0, i32 3, i32 2>163 %res = select <4 x i1> %mask, <4 x double> %shufp, <4 x double> zeroinitializer164 ret <4 x double> %res165}166 167define <2 x double> @transform_VPERMILPSrmkz(ptr %ap, i2 %mask_int) nounwind {168; CHECK-LABEL: transform_VPERMILPSrmkz:169; CHECK: # %bb.0:170; CHECK-NEXT: kmovd %esi, %k1171; CHECK-NEXT: vpermilpd {{.*#+}} xmm0 {%k1} {z} = mem[1,0]172; CHECK-NEXT: retq173 %mask = bitcast i2 %mask_int to <2 x i1>174 %a = load <2 x double>, ptr %ap175 %shufp = shufflevector <2 x double> %a, <2 x double> poison, <2 x i32> <i32 1, i32 0>176 %res = select <2 x i1> %mask, <2 x double> %shufp, <2 x double> zeroinitializer177 ret <2 x double> %res178}179 180define <8 x double> @transform_VPERMILPSZrmk(ptr %ap, <8 x double> %b, i8 %mask_int) nounwind {181; CHECK-LABEL: transform_VPERMILPSZrmk:182; CHECK: # %bb.0:183; CHECK-NEXT: kmovd %esi, %k1184; CHECK-NEXT: vpermilpd {{.*#+}} zmm0 {%k1} = mem[0,1,3,2,4,5,7,6]185; CHECK-NEXT: retq186 %mask = bitcast i8 %mask_int to <8 x i1>187 %a = load <8 x double>, ptr %ap188 %shufp = shufflevector <8 x double> %a, <8 x double> poison, <8 x i32> <i32 0, i32 1, i32 3, i32 2, i32 4, i32 5, i32 7, i32 6>189 %res = select <8 x i1> %mask, <8 x double> %shufp, <8 x double> %b190 ret <8 x double> %res191}192 193define <4 x double> @transform_VPERMILPSYrmk(ptr %ap, <4 x double> %b, i4 %mask_int) nounwind {194; CHECK-LABEL: transform_VPERMILPSYrmk:195; CHECK: # %bb.0:196; CHECK-NEXT: kmovd %esi, %k1197; CHECK-NEXT: vpermilpd {{.*#+}} ymm0 {%k1} = mem[0,1,3,2]198; CHECK-NEXT: retq199 %mask = bitcast i4 %mask_int to <4 x i1>200 %a = load <4 x double>, ptr %ap201 %shufp = shufflevector <4 x double> %a, <4 x double> poison, <4 x i32> <i32 0, i32 1, i32 3, i32 2>202 %res = select <4 x i1> %mask, <4 x double> %shufp, <4 x double> %b203 ret <4 x double> %res204}205 206define <2 x double> @transform_VPERMILPSrmk(ptr %ap, <2 x double> %b, i2 %mask_int) nounwind {207; CHECK-LABEL: transform_VPERMILPSrmk:208; CHECK: # %bb.0:209; CHECK-NEXT: kmovd %esi, %k1210; CHECK-NEXT: vpermilpd {{.*#+}} xmm0 {%k1} = mem[1,0]211; CHECK-NEXT: retq212 %mask = bitcast i2 %mask_int to <2 x i1>213 %a = load <2 x double>, ptr %ap214 %shufp = shufflevector <2 x double> %a, <2 x double> poison, <2 x i32> <i32 1, i32 0>215 %res = select <2 x i1> %mask, <2 x double> %shufp, <2 x double> %b216 ret <2 x double> %res217}218;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:219; CHECK-AVX512: {{.*}}220; CHECK-ICX: {{.*}}221; CHECK-ICX-BYPASS-DELAY: {{.*}}222; CHECK-ICX-NO-BYPASS-DELAY: {{.*}}223; CHECK-V4: {{.*}}224; CHECK-ZNVER4: {{.*}}225