198 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -disable-peephole -mtriple=i686-apple-darwin -mattr=+avx512f,avx512bw | FileCheck %s --check-prefix=X863; RUN: llc < %s -disable-peephole -mtriple=x86_64-apple-darwin -mattr=+avx512f,avx512bw | FileCheck %s --check-prefix=X644 5define <32 x i16> @test_llvm_x86_avx512_pmovsxbw(ptr %a) {6; X86-LABEL: test_llvm_x86_avx512_pmovsxbw:7; X86: ## %bb.0:8; X86-NEXT: movl {{[0-9]+}}(%esp), %eax9; X86-NEXT: vpmovsxbw (%eax), %zmm010; X86-NEXT: retl11;12; X64-LABEL: test_llvm_x86_avx512_pmovsxbw:13; X64: ## %bb.0:14; X64-NEXT: vpmovsxbw (%rdi), %zmm015; X64-NEXT: retq16 %1 = load <32 x i8>, ptr %a, align 117 %2 = sext <32 x i8> %1 to <32 x i16>18 ret <32 x i16> %219}20 21define <16 x i32> @test_llvm_x86_avx512_pmovsxbd(ptr %a) {22; X86-LABEL: test_llvm_x86_avx512_pmovsxbd:23; X86: ## %bb.0:24; X86-NEXT: movl {{[0-9]+}}(%esp), %eax25; X86-NEXT: vpmovsxbd (%eax), %zmm026; X86-NEXT: retl27;28; X64-LABEL: test_llvm_x86_avx512_pmovsxbd:29; X64: ## %bb.0:30; X64-NEXT: vpmovsxbd (%rdi), %zmm031; X64-NEXT: retq32 %1 = load <16 x i8>, ptr %a, align 133 %2 = sext <16 x i8> %1 to <16 x i32>34 ret <16 x i32> %235}36 37define <8 x i64> @test_llvm_x86_avx512_pmovsxbq(ptr %a) {38; X86-LABEL: test_llvm_x86_avx512_pmovsxbq:39; X86: ## %bb.0:40; X86-NEXT: movl {{[0-9]+}}(%esp), %eax41; X86-NEXT: vpmovsxbq (%eax), %zmm042; X86-NEXT: retl43;44; X64-LABEL: test_llvm_x86_avx512_pmovsxbq:45; X64: ## %bb.0:46; X64-NEXT: vpmovsxbq (%rdi), %zmm047; X64-NEXT: retq48 %1 = load <16 x i8>, ptr %a, align 149 %2 = shufflevector <16 x i8> %1, <16 x i8> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>50 %3 = sext <8 x i8> %2 to <8 x i64>51 ret <8 x i64> %352}53 54define <16 x i32> @test_llvm_x86_avx512_pmovsxwd(ptr %a) {55; X86-LABEL: test_llvm_x86_avx512_pmovsxwd:56; X86: ## %bb.0:57; X86-NEXT: movl {{[0-9]+}}(%esp), %eax58; X86-NEXT: vpmovsxwd (%eax), %zmm059; X86-NEXT: retl60;61; X64-LABEL: test_llvm_x86_avx512_pmovsxwd:62; X64: ## %bb.0:63; X64-NEXT: vpmovsxwd (%rdi), %zmm064; X64-NEXT: retq65 %1 = load <16 x i16>, ptr %a, align 166 %2 = sext <16 x i16> %1 to <16 x i32>67 ret <16 x i32> %268}69 70define <8 x i64> @test_llvm_x86_avx512_pmovsxwq(ptr %a) {71; X86-LABEL: test_llvm_x86_avx512_pmovsxwq:72; X86: ## %bb.0:73; X86-NEXT: movl {{[0-9]+}}(%esp), %eax74; X86-NEXT: vpmovsxwq (%eax), %zmm075; X86-NEXT: retl76;77; X64-LABEL: test_llvm_x86_avx512_pmovsxwq:78; X64: ## %bb.0:79; X64-NEXT: vpmovsxwq (%rdi), %zmm080; X64-NEXT: retq81 %1 = load <8 x i16>, ptr %a, align 182 %2 = sext <8 x i16> %1 to <8 x i64>83 ret <8 x i64> %284}85 86define <8 x i64> @test_llvm_x86_avx512_pmovsxdq(ptr %a) {87; X86-LABEL: test_llvm_x86_avx512_pmovsxdq:88; X86: ## %bb.0:89; X86-NEXT: movl {{[0-9]+}}(%esp), %eax90; X86-NEXT: vpmovsxdq (%eax), %zmm091; X86-NEXT: retl92;93; X64-LABEL: test_llvm_x86_avx512_pmovsxdq:94; X64: ## %bb.0:95; X64-NEXT: vpmovsxdq (%rdi), %zmm096; X64-NEXT: retq97 %1 = load <8 x i32>, ptr %a, align 198 %2 = sext <8 x i32> %1 to <8 x i64>99 ret <8 x i64> %2100}101 102define <32 x i16> @test_llvm_x86_avx512_pmovzxbw(ptr %a) {103; X86-LABEL: test_llvm_x86_avx512_pmovzxbw:104; X86: ## %bb.0:105; X86-NEXT: movl {{[0-9]+}}(%esp), %eax106; X86-NEXT: vpmovzxbw {{.*#+}} zmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero,mem[8],zero,mem[9],zero,mem[10],zero,mem[11],zero,mem[12],zero,mem[13],zero,mem[14],zero,mem[15],zero,mem[16],zero,mem[17],zero,mem[18],zero,mem[19],zero,mem[20],zero,mem[21],zero,mem[22],zero,mem[23],zero,mem[24],zero,mem[25],zero,mem[26],zero,mem[27],zero,mem[28],zero,mem[29],zero,mem[30],zero,mem[31],zero107; X86-NEXT: retl108;109; X64-LABEL: test_llvm_x86_avx512_pmovzxbw:110; X64: ## %bb.0:111; X64-NEXT: vpmovzxbw {{.*#+}} zmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero,mem[8],zero,mem[9],zero,mem[10],zero,mem[11],zero,mem[12],zero,mem[13],zero,mem[14],zero,mem[15],zero,mem[16],zero,mem[17],zero,mem[18],zero,mem[19],zero,mem[20],zero,mem[21],zero,mem[22],zero,mem[23],zero,mem[24],zero,mem[25],zero,mem[26],zero,mem[27],zero,mem[28],zero,mem[29],zero,mem[30],zero,mem[31],zero112; X64-NEXT: retq113 %1 = load <32 x i8>, ptr %a, align 1114 %2 = zext <32 x i8> %1 to <32 x i16>115 ret <32 x i16> %2116}117 118define <16 x i32> @test_llvm_x86_avx512_pmovzxbd(ptr %a) {119; X86-LABEL: test_llvm_x86_avx512_pmovzxbd:120; X86: ## %bb.0:121; X86-NEXT: movl {{[0-9]+}}(%esp), %eax122; X86-NEXT: vpmovzxbd {{.*#+}} zmm0 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero,mem[8],zero,zero,zero,mem[9],zero,zero,zero,mem[10],zero,zero,zero,mem[11],zero,zero,zero,mem[12],zero,zero,zero,mem[13],zero,zero,zero,mem[14],zero,zero,zero,mem[15],zero,zero,zero123; X86-NEXT: retl124;125; X64-LABEL: test_llvm_x86_avx512_pmovzxbd:126; X64: ## %bb.0:127; X64-NEXT: vpmovzxbd {{.*#+}} zmm0 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero,mem[8],zero,zero,zero,mem[9],zero,zero,zero,mem[10],zero,zero,zero,mem[11],zero,zero,zero,mem[12],zero,zero,zero,mem[13],zero,zero,zero,mem[14],zero,zero,zero,mem[15],zero,zero,zero128; X64-NEXT: retq129 %1 = load <16 x i8>, ptr %a, align 1130 %2 = zext <16 x i8> %1 to <16 x i32>131 ret <16 x i32> %2132}133 134define <8 x i64> @test_llvm_x86_avx512_pmovzxbq(ptr %a) {135; X86-LABEL: test_llvm_x86_avx512_pmovzxbq:136; X86: ## %bb.0:137; X86-NEXT: movl {{[0-9]+}}(%esp), %eax138; X86-NEXT: vpmovzxbq {{.*#+}} zmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero,mem[1],zero,zero,zero,zero,zero,zero,zero,mem[2],zero,zero,zero,zero,zero,zero,zero,mem[3],zero,zero,zero,zero,zero,zero,zero,mem[4],zero,zero,zero,zero,zero,zero,zero,mem[5],zero,zero,zero,zero,zero,zero,zero,mem[6],zero,zero,zero,zero,zero,zero,zero,mem[7],zero,zero,zero,zero,zero,zero,zero139; X86-NEXT: retl140;141; X64-LABEL: test_llvm_x86_avx512_pmovzxbq:142; X64: ## %bb.0:143; X64-NEXT: vpmovzxbq {{.*#+}} zmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero,mem[1],zero,zero,zero,zero,zero,zero,zero,mem[2],zero,zero,zero,zero,zero,zero,zero,mem[3],zero,zero,zero,zero,zero,zero,zero,mem[4],zero,zero,zero,zero,zero,zero,zero,mem[5],zero,zero,zero,zero,zero,zero,zero,mem[6],zero,zero,zero,zero,zero,zero,zero,mem[7],zero,zero,zero,zero,zero,zero,zero144; X64-NEXT: retq145 %1 = load <16 x i8>, ptr %a, align 1146 %2 = shufflevector <16 x i8> %1, <16 x i8> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>147 %3 = zext <8 x i8> %2 to <8 x i64>148 ret <8 x i64> %3149}150 151define <16 x i32> @test_llvm_x86_avx512_pmovzxwd(ptr %a) {152; X86-LABEL: test_llvm_x86_avx512_pmovzxwd:153; X86: ## %bb.0:154; X86-NEXT: movl {{[0-9]+}}(%esp), %eax155; X86-NEXT: vpmovzxwd {{.*#+}} zmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero,mem[8],zero,mem[9],zero,mem[10],zero,mem[11],zero,mem[12],zero,mem[13],zero,mem[14],zero,mem[15],zero156; X86-NEXT: retl157;158; X64-LABEL: test_llvm_x86_avx512_pmovzxwd:159; X64: ## %bb.0:160; X64-NEXT: vpmovzxwd {{.*#+}} zmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero,mem[8],zero,mem[9],zero,mem[10],zero,mem[11],zero,mem[12],zero,mem[13],zero,mem[14],zero,mem[15],zero161; X64-NEXT: retq162 %1 = load <16 x i16>, ptr %a, align 1163 %2 = zext <16 x i16> %1 to <16 x i32>164 ret <16 x i32> %2165}166 167define <8 x i64> @test_llvm_x86_avx512_pmovzxwq(ptr %a) {168; X86-LABEL: test_llvm_x86_avx512_pmovzxwq:169; X86: ## %bb.0:170; X86-NEXT: movl {{[0-9]+}}(%esp), %eax171; X86-NEXT: vpmovzxwq {{.*#+}} zmm0 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero172; X86-NEXT: retl173;174; X64-LABEL: test_llvm_x86_avx512_pmovzxwq:175; X64: ## %bb.0:176; X64-NEXT: vpmovzxwq {{.*#+}} zmm0 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero177; X64-NEXT: retq178 %1 = load <8 x i16>, ptr %a, align 1179 %2 = zext <8 x i16> %1 to <8 x i64>180 ret <8 x i64> %2181}182 183define <8 x i64> @test_llvm_x86_avx512_pmovzxdq(ptr %a) {184; X86-LABEL: test_llvm_x86_avx512_pmovzxdq:185; X86: ## %bb.0:186; X86-NEXT: movl {{[0-9]+}}(%esp), %eax187; X86-NEXT: vpmovzxdq {{.*#+}} zmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero188; X86-NEXT: retl189;190; X64-LABEL: test_llvm_x86_avx512_pmovzxdq:191; X64: ## %bb.0:192; X64-NEXT: vpmovzxdq {{.*#+}} zmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero193; X64-NEXT: retq194 %1 = load <8 x i32>, ptr %a, align 1195 %2 = zext <8 x i32> %1 to <8 x i64>196 ret <8 x i64> %2197}198