1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=x86_64-apple-darwin -show-mc-encoding -mattr=+avx512f | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=KNL 3; RUN: llc < %s -mtriple=x86_64-apple-darwin -show-mc-encoding -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512BW 4; RUN: llc < %s -mtriple=x86_64-apple-darwin -show-mc-encoding -mattr=+avx512f,+avx512bw,+avx512vl,+avx512dq | FileCheck %s --check-prefix=CHECK --check-prefix=SKX 5 6define <16 x float> @test1(<16 x float> %x, <16 x float> %y) nounwind { 7; CHECK-LABEL: test1: 8; CHECK: ## %bb.0: 9; CHECK-NEXT: vcmpleps %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xc9,0x02] 10; CHECK-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] 11; CHECK-NEXT: retq ## encoding: [0xc3] 12 %mask = fcmp ole <16 x float> %x, %y 13 %max = select <16 x i1> %mask, <16 x float> %x, <16 x float> %y 14 ret <16 x float> %max 15} 16 17define <8 x double> @test2(<8 x double> %x, <8 x double> %y) nounwind { 18; CHECK-LABEL: test2: 19; CHECK: ## %bb.0: 20; CHECK-NEXT: vcmplepd %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0xc9,0x02] 21; CHECK-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] 22; CHECK-NEXT: retq ## encoding: [0xc3] 23 %mask = fcmp ole <8 x double> %x, %y 24 %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %y 25 ret <8 x double> %max 26} 27 28define <16 x i32> @test3(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %yp) nounwind { 29; CHECK-LABEL: test3: 30; CHECK: ## %bb.0: 31; CHECK-NEXT: vpcmpeqd (%rdi), %zmm0, %k1 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0x0f] 32; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] 33; CHECK-NEXT: retq ## encoding: [0xc3] 34 %y = load <16 x i32>, <16 x i32>* %yp, align 4 35 %mask = icmp eq <16 x i32> %x, %y 36 %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 37 ret <16 x i32> %max 38} 39 40define <16 x i32> @test4_unsigned(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1) nounwind { 41; CHECK-LABEL: test4_unsigned: 42; CHECK: ## %bb.0: 43; CHECK-NEXT: vpcmpnltud %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf3,0x7d,0x48,0x1e,0xc9,0x05] 44; CHECK-NEXT: vpblendmd %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc2] 45; CHECK-NEXT: retq ## encoding: [0xc3] 46 %mask = icmp uge <16 x i32> %x, %y 47 %max = select <16 x i1> %mask, <16 x i32> %x1, <16 x i32> %y 48 ret <16 x i32> %max 49} 50 51define <8 x i64> @test5(<8 x i64> %x, <8 x i64> %y) nounwind { 52; CHECK-LABEL: test5: 53; CHECK: ## %bb.0: 54; CHECK-NEXT: vpcmpeqq %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf2,0xfd,0x48,0x29,0xc9] 55; CHECK-NEXT: vpblendmq %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x64,0xc0] 56; CHECK-NEXT: retq ## encoding: [0xc3] 57 %mask = icmp eq <8 x i64> %x, %y 58 %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %y 59 ret <8 x i64> %max 60} 61 62define <8 x i64> @test6_unsigned(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1) nounwind { 63; CHECK-LABEL: test6_unsigned: 64; CHECK: ## %bb.0: 65; CHECK-NEXT: vpcmpnleuq %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf3,0xfd,0x48,0x1e,0xc9,0x06] 66; CHECK-NEXT: vpblendmq %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x64,0xc2] 67; CHECK-NEXT: retq ## encoding: [0xc3] 68 %mask = icmp ugt <8 x i64> %x, %y 69 %max = select <8 x i1> %mask, <8 x i64> %x1, <8 x i64> %y 70 ret <8 x i64> %max 71} 72 73define <4 x float> @test7(<4 x float> %a, <4 x float> %b) { 74; AVX512-LABEL: test7: 75; AVX512: ## %bb.0: 76; AVX512-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 77; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 78; AVX512-NEXT: vxorps %xmm2, %xmm2, %xmm2 ## encoding: [0xc5,0xe8,0x57,0xd2] 79; AVX512-NEXT: vcmpltps %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xca,0x01] 80; AVX512-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] 81; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 82; AVX512-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 83; AVX512-NEXT: retq ## encoding: [0xc3] 84; 85; SKX-LABEL: test7: 86; SKX: ## %bb.0: 87; SKX-NEXT: vxorps %xmm2, %xmm2, %xmm2 ## EVEX TO VEX Compression encoding: [0xc5,0xe8,0x57,0xd2] 88; SKX-NEXT: vcmpltps %xmm2, %xmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x08,0xc2,0xca,0x01] 89; SKX-NEXT: vblendmps %xmm0, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0x65,0xc0] 90; SKX-NEXT: retq ## encoding: [0xc3] 91 92 %mask = fcmp olt <4 x float> %a, zeroinitializer 93 %c = select <4 x i1>%mask, <4 x float>%a, <4 x float>%b 94 ret <4 x float>%c 95} 96 97define <2 x double> @test8(<2 x double> %a, <2 x double> %b) { 98; AVX512-LABEL: test8: 99; AVX512: ## %bb.0: 100; AVX512-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 101; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 102; AVX512-NEXT: vxorpd %xmm2, %xmm2, %xmm2 ## encoding: [0xc5,0xe9,0x57,0xd2] 103; AVX512-NEXT: vcmpltpd %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0xca,0x01] 104; AVX512-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] 105; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 106; AVX512-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 107; AVX512-NEXT: retq ## encoding: [0xc3] 108; 109; SKX-LABEL: test8: 110; SKX: ## %bb.0: 111; SKX-NEXT: vxorpd %xmm2, %xmm2, %xmm2 ## EVEX TO VEX Compression encoding: [0xc5,0xe9,0x57,0xd2] 112; SKX-NEXT: vcmpltpd %xmm2, %xmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x08,0xc2,0xca,0x01] 113; SKX-NEXT: vblendmpd %xmm0, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0x65,0xc0] 114; SKX-NEXT: retq ## encoding: [0xc3] 115 %mask = fcmp olt <2 x double> %a, zeroinitializer 116 %c = select <2 x i1>%mask, <2 x double>%a, <2 x double>%b 117 ret <2 x double>%c 118} 119 120define <8 x i32> @test9(<8 x i32> %x, <8 x i32> %y) nounwind { 121; AVX512-LABEL: test9: 122; AVX512: ## %bb.0: 123; AVX512-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 124; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 125; AVX512-NEXT: vpcmpeqd %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0xc9] 126; AVX512-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] 127; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 128; AVX512-NEXT: retq ## encoding: [0xc3] 129; 130; SKX-LABEL: test9: 131; SKX: ## %bb.0: 132; SKX-NEXT: vpcmpeqd %ymm1, %ymm0, %k1 ## encoding: [0x62,0xf1,0x7d,0x28,0x76,0xc9] 133; SKX-NEXT: vpblendmd %ymm0, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0x64,0xc0] 134; SKX-NEXT: retq ## encoding: [0xc3] 135 %mask = icmp eq <8 x i32> %x, %y 136 %max = select <8 x i1> %mask, <8 x i32> %x, <8 x i32> %y 137 ret <8 x i32> %max 138} 139 140define <8 x float> @test10(<8 x float> %x, <8 x float> %y) nounwind { 141; AVX512-LABEL: test10: 142; AVX512: ## %bb.0: 143; AVX512-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 144; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 145; AVX512-NEXT: vcmpeqps %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xc9,0x00] 146; AVX512-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] 147; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 148; AVX512-NEXT: retq ## encoding: [0xc3] 149; 150; SKX-LABEL: test10: 151; SKX: ## %bb.0: 152; SKX-NEXT: vcmpeqps %ymm1, %ymm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x28,0xc2,0xc9,0x00] 153; SKX-NEXT: vblendmps %ymm0, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0x65,0xc0] 154; SKX-NEXT: retq ## encoding: [0xc3] 155 156 %mask = fcmp oeq <8 x float> %x, %y 157 %max = select <8 x i1> %mask, <8 x float> %x, <8 x float> %y 158 ret <8 x float> %max 159} 160 161define <8 x i32> @test11_unsigned(<8 x i32> %x, <8 x i32> %y) nounwind { 162; AVX512-LABEL: test11_unsigned: 163; AVX512: ## %bb.0: 164; AVX512-NEXT: vpmaxud %ymm1, %ymm0, %ymm0 ## encoding: [0xc4,0xe2,0x7d,0x3f,0xc1] 165; AVX512-NEXT: retq ## encoding: [0xc3] 166; 167; SKX-LABEL: test11_unsigned: 168; SKX: ## %bb.0: 169; SKX-NEXT: vpmaxud %ymm1, %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x7d,0x3f,0xc1] 170; SKX-NEXT: retq ## encoding: [0xc3] 171 %mask = icmp ugt <8 x i32> %x, %y 172 %max = select <8 x i1> %mask, <8 x i32> %x, <8 x i32> %y 173 ret <8 x i32> %max 174} 175 176define i16 @test12(<16 x i64> %a, <16 x i64> %b) nounwind { 177; KNL-LABEL: test12: 178; KNL: ## %bb.0: 179; KNL-NEXT: vpcmpeqq %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x48,0x29,0xc2] 180; KNL-NEXT: vpcmpeqq %zmm3, %zmm1, %k1 ## encoding: [0x62,0xf2,0xf5,0x48,0x29,0xcb] 181; KNL-NEXT: kunpckbw %k0, %k1, %k0 ## encoding: [0xc5,0xf5,0x4b,0xc0] 182; KNL-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0] 183; KNL-NEXT: ## kill: def $ax killed $ax killed $eax 184; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 185; KNL-NEXT: retq ## encoding: [0xc3] 186; 187; AVX512BW-LABEL: test12: 188; AVX512BW: ## %bb.0: 189; AVX512BW-NEXT: vpcmpeqq %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x48,0x29,0xc2] 190; AVX512BW-NEXT: vpcmpeqq %zmm3, %zmm1, %k1 ## encoding: [0x62,0xf2,0xf5,0x48,0x29,0xcb] 191; AVX512BW-NEXT: kunpckbw %k0, %k1, %k0 ## encoding: [0xc5,0xf5,0x4b,0xc0] 192; AVX512BW-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] 193; AVX512BW-NEXT: ## kill: def $ax killed $ax killed $eax 194; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 195; AVX512BW-NEXT: retq ## encoding: [0xc3] 196; 197; SKX-LABEL: test12: 198; SKX: ## %bb.0: 199; SKX-NEXT: vpcmpeqq %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x48,0x29,0xc2] 200; SKX-NEXT: vpcmpeqq %zmm3, %zmm1, %k1 ## encoding: [0x62,0xf2,0xf5,0x48,0x29,0xcb] 201; SKX-NEXT: kunpckbw %k0, %k1, %k0 ## encoding: [0xc5,0xf5,0x4b,0xc0] 202; SKX-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] 203; SKX-NEXT: ## kill: def $ax killed $ax killed $eax 204; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 205; SKX-NEXT: retq ## encoding: [0xc3] 206 %res = icmp eq <16 x i64> %a, %b 207 %res1 = bitcast <16 x i1> %res to i16 208 ret i16 %res1 209} 210 211define i32 @test12_v32i32(<32 x i32> %a, <32 x i32> %b) nounwind { 212; KNL-LABEL: test12_v32i32: 213; KNL: ## %bb.0: 214; KNL-NEXT: vpcmpeqd %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0xc2] 215; KNL-NEXT: kmovw %k0, %ecx ## encoding: [0xc5,0xf8,0x93,0xc8] 216; KNL-NEXT: vpcmpeqd %zmm3, %zmm1, %k0 ## encoding: [0x62,0xf1,0x75,0x48,0x76,0xc3] 217; KNL-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0] 218; KNL-NEXT: shll $16, %eax ## encoding: [0xc1,0xe0,0x10] 219; KNL-NEXT: orl %ecx, %eax ## encoding: [0x09,0xc8] 220; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 221; KNL-NEXT: retq ## encoding: [0xc3] 222; 223; AVX512BW-LABEL: test12_v32i32: 224; AVX512BW: ## %bb.0: 225; AVX512BW-NEXT: vpcmpeqd %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0xc2] 226; AVX512BW-NEXT: vpcmpeqd %zmm3, %zmm1, %k1 ## encoding: [0x62,0xf1,0x75,0x48,0x76,0xcb] 227; AVX512BW-NEXT: kunpckwd %k0, %k1, %k0 ## encoding: [0xc5,0xf4,0x4b,0xc0] 228; AVX512BW-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] 229; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 230; AVX512BW-NEXT: retq ## encoding: [0xc3] 231; 232; SKX-LABEL: test12_v32i32: 233; SKX: ## %bb.0: 234; SKX-NEXT: vpcmpeqd %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0xc2] 235; SKX-NEXT: vpcmpeqd %zmm3, %zmm1, %k1 ## encoding: [0x62,0xf1,0x75,0x48,0x76,0xcb] 236; SKX-NEXT: kunpckwd %k0, %k1, %k0 ## encoding: [0xc5,0xf4,0x4b,0xc0] 237; SKX-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] 238; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 239; SKX-NEXT: retq ## encoding: [0xc3] 240 %res = icmp eq <32 x i32> %a, %b 241 %res1 = bitcast <32 x i1> %res to i32 242 ret i32 %res1 243} 244 245define i64 @test12_v64i16(<64 x i16> %a, <64 x i16> %b) nounwind { 246; KNL-LABEL: test12_v64i16: 247; KNL: ## %bb.0: 248; KNL-NEXT: vpcmpeqw %ymm4, %ymm0, %ymm0 ## encoding: [0xc5,0xfd,0x75,0xc4] 249; KNL-NEXT: vpmovsxwd %ymm0, %zmm0 ## encoding: [0x62,0xf2,0x7d,0x48,0x23,0xc0] 250; KNL-NEXT: vptestmd %zmm0, %zmm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x48,0x27,0xc0] 251; KNL-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0] 252; KNL-NEXT: vpcmpeqw %ymm5, %ymm1, %ymm0 ## encoding: [0xc5,0xf5,0x75,0xc5] 253; KNL-NEXT: vpmovsxwd %ymm0, %zmm0 ## encoding: [0x62,0xf2,0x7d,0x48,0x23,0xc0] 254; KNL-NEXT: vptestmd %zmm0, %zmm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x48,0x27,0xc0] 255; KNL-NEXT: kmovw %k0, %ecx ## encoding: [0xc5,0xf8,0x93,0xc8] 256; KNL-NEXT: shll $16, %ecx ## encoding: [0xc1,0xe1,0x10] 257; KNL-NEXT: orl %eax, %ecx ## encoding: [0x09,0xc1] 258; KNL-NEXT: vpcmpeqw %ymm6, %ymm2, %ymm0 ## encoding: [0xc5,0xed,0x75,0xc6] 259; KNL-NEXT: vpmovsxwd %ymm0, %zmm0 ## encoding: [0x62,0xf2,0x7d,0x48,0x23,0xc0] 260; KNL-NEXT: vptestmd %zmm0, %zmm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x48,0x27,0xc0] 261; KNL-NEXT: kmovw %k0, %edx ## encoding: [0xc5,0xf8,0x93,0xd0] 262; KNL-NEXT: vpcmpeqw %ymm7, %ymm3, %ymm0 ## encoding: [0xc5,0xe5,0x75,0xc7] 263; KNL-NEXT: vpmovsxwd %ymm0, %zmm0 ## encoding: [0x62,0xf2,0x7d,0x48,0x23,0xc0] 264; KNL-NEXT: vptestmd %zmm0, %zmm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x48,0x27,0xc0] 265; KNL-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0] 266; KNL-NEXT: shll $16, %eax ## encoding: [0xc1,0xe0,0x10] 267; KNL-NEXT: orl %edx, %eax ## encoding: [0x09,0xd0] 268; KNL-NEXT: shlq $32, %rax ## encoding: [0x48,0xc1,0xe0,0x20] 269; KNL-NEXT: orq %rcx, %rax ## encoding: [0x48,0x09,0xc8] 270; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 271; KNL-NEXT: retq ## encoding: [0xc3] 272; 273; AVX512BW-LABEL: test12_v64i16: 274; AVX512BW: ## %bb.0: 275; AVX512BW-NEXT: vpcmpeqw %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x75,0xc2] 276; AVX512BW-NEXT: vpcmpeqw %zmm3, %zmm1, %k1 ## encoding: [0x62,0xf1,0x75,0x48,0x75,0xcb] 277; AVX512BW-NEXT: kunpckdq %k0, %k1, %k0 ## encoding: [0xc4,0xe1,0xf4,0x4b,0xc0] 278; AVX512BW-NEXT: kmovq %k0, %rax ## encoding: [0xc4,0xe1,0xfb,0x93,0xc0] 279; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 280; AVX512BW-NEXT: retq ## encoding: [0xc3] 281; 282; SKX-LABEL: test12_v64i16: 283; SKX: ## %bb.0: 284; SKX-NEXT: vpcmpeqw %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x75,0xc2] 285; SKX-NEXT: vpcmpeqw %zmm3, %zmm1, %k1 ## encoding: [0x62,0xf1,0x75,0x48,0x75,0xcb] 286; SKX-NEXT: kunpckdq %k0, %k1, %k0 ## encoding: [0xc4,0xe1,0xf4,0x4b,0xc0] 287; SKX-NEXT: kmovq %k0, %rax ## encoding: [0xc4,0xe1,0xfb,0x93,0xc0] 288; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 289; SKX-NEXT: retq ## encoding: [0xc3] 290 %res = icmp eq <64 x i16> %a, %b 291 %res1 = bitcast <64 x i1> %res to i64 292 ret i64 %res1 293} 294 295define <16 x i32> @test13(<16 x float>%a, <16 x float>%b) 296; AVX512-LABEL: test13: 297; AVX512: ## %bb.0: 298; AVX512-NEXT: vcmpeqps %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xc9,0x00] 299; AVX512-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf3,0x7d,0xc9,0x25,0xc0,0xff] 300; AVX512-NEXT: vpsrld $31, %zmm0, %zmm0 ## encoding: [0x62,0xf1,0x7d,0x48,0x72,0xd0,0x1f] 301; AVX512-NEXT: retq ## encoding: [0xc3] 302; 303; SKX-LABEL: test13: 304; SKX: ## %bb.0: 305; SKX-NEXT: vcmpeqps %zmm1, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xc1,0x00] 306; SKX-NEXT: vpmovm2d %k0, %zmm0 ## encoding: [0x62,0xf2,0x7e,0x48,0x38,0xc0] 307; SKX-NEXT: vpsrld $31, %zmm0, %zmm0 ## encoding: [0x62,0xf1,0x7d,0x48,0x72,0xd0,0x1f] 308; SKX-NEXT: retq ## encoding: [0xc3] 309{ 310 %cmpvector_i = fcmp oeq <16 x float> %a, %b 311 %conv = zext <16 x i1> %cmpvector_i to <16 x i32> 312 ret <16 x i32> %conv 313} 314 315define <16 x i32> @test14(<16 x i32>%a, <16 x i32>%b) { 316; CHECK-LABEL: test14: 317; CHECK: ## %bb.0: 318; CHECK-NEXT: vpsubd %zmm1, %zmm0, %zmm2 ## encoding: [0x62,0xf1,0x7d,0x48,0xfa,0xd1] 319; CHECK-NEXT: vpcmpgtd %zmm0, %zmm2, %k1 ## encoding: [0x62,0xf1,0x6d,0x48,0x66,0xc8] 320; CHECK-NEXT: vpsubd %zmm1, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xc9,0xfa,0xc1] 321; CHECK-NEXT: retq ## encoding: [0xc3] 322 %sub_r = sub <16 x i32> %a, %b 323 %cmp.i2.i = icmp sgt <16 x i32> %sub_r, %a 324 %sext.i3.i = sext <16 x i1> %cmp.i2.i to <16 x i32> 325 %mask = icmp eq <16 x i32> %sext.i3.i, zeroinitializer 326 %res = select <16 x i1> %mask, <16 x i32> zeroinitializer, <16 x i32> %sub_r 327 ret <16 x i32>%res 328} 329 330define <8 x i64> @test15(<8 x i64>%a, <8 x i64>%b) { 331; CHECK-LABEL: test15: 332; CHECK: ## %bb.0: 333; CHECK-NEXT: vpsubq %zmm1, %zmm0, %zmm2 ## encoding: [0x62,0xf1,0xfd,0x48,0xfb,0xd1] 334; CHECK-NEXT: vpcmpgtq %zmm0, %zmm2, %k1 ## encoding: [0x62,0xf2,0xed,0x48,0x37,0xc8] 335; CHECK-NEXT: vpsubq %zmm1, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0xfd,0xc9,0xfb,0xc1] 336; CHECK-NEXT: retq ## encoding: [0xc3] 337 %sub_r = sub <8 x i64> %a, %b 338 %cmp.i2.i = icmp sgt <8 x i64> %sub_r, %a 339 %sext.i3.i = sext <8 x i1> %cmp.i2.i to <8 x i64> 340 %mask = icmp eq <8 x i64> %sext.i3.i, zeroinitializer 341 %res = select <8 x i1> %mask, <8 x i64> zeroinitializer, <8 x i64> %sub_r 342 ret <8 x i64>%res 343} 344 345define <16 x i32> @test16(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1) nounwind { 346; CHECK-LABEL: test16: 347; CHECK: ## %bb.0: 348; CHECK-NEXT: vpcmpnltd %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf3,0x7d,0x48,0x1f,0xc9,0x05] 349; CHECK-NEXT: vpblendmd %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc2] 350; CHECK-NEXT: retq ## encoding: [0xc3] 351 %mask = icmp sge <16 x i32> %x, %y 352 %max = select <16 x i1> %mask, <16 x i32> %x1, <16 x i32> %y 353 ret <16 x i32> %max 354} 355 356define <16 x i32> @test17(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nounwind { 357; CHECK-LABEL: test17: 358; CHECK: ## %bb.0: 359; CHECK-NEXT: vpcmpgtd (%rdi), %zmm0, %k1 ## encoding: [0x62,0xf1,0x7d,0x48,0x66,0x0f] 360; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] 361; CHECK-NEXT: retq ## encoding: [0xc3] 362 %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4 363 %mask = icmp sgt <16 x i32> %x, %y 364 %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 365 ret <16 x i32> %max 366} 367 368define <16 x i32> @test18(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nounwind { 369; CHECK-LABEL: test18: 370; CHECK: ## %bb.0: 371; CHECK-NEXT: vpcmpled (%rdi), %zmm0, %k1 ## encoding: [0x62,0xf3,0x7d,0x48,0x1f,0x0f,0x02] 372; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] 373; CHECK-NEXT: retq ## encoding: [0xc3] 374 %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4 375 %mask = icmp sle <16 x i32> %x, %y 376 %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 377 ret <16 x i32> %max 378} 379 380define <16 x i32> @test19(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nounwind { 381; CHECK-LABEL: test19: 382; CHECK: ## %bb.0: 383; CHECK-NEXT: vpcmpleud (%rdi), %zmm0, %k1 ## encoding: [0x62,0xf3,0x7d,0x48,0x1e,0x0f,0x02] 384; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] 385; CHECK-NEXT: retq ## encoding: [0xc3] 386 %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4 387 %mask = icmp ule <16 x i32> %x, %y 388 %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 389 ret <16 x i32> %max 390} 391 392define <16 x i32> @test20(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1, <16 x i32> %y1) nounwind { 393; CHECK-LABEL: test20: 394; CHECK: ## %bb.0: 395; CHECK-NEXT: vpcmpeqd %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0xc9] 396; CHECK-NEXT: vpcmpeqd %zmm3, %zmm2, %k1 {%k1} ## encoding: [0x62,0xf1,0x6d,0x49,0x76,0xcb] 397; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] 398; CHECK-NEXT: retq ## encoding: [0xc3] 399 %mask1 = icmp eq <16 x i32> %x1, %y1 400 %mask0 = icmp eq <16 x i32> %x, %y 401 %mask = select <16 x i1> %mask0, <16 x i1> %mask1, <16 x i1> zeroinitializer 402 %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %y 403 ret <16 x i32> %max 404} 405 406define <8 x i64> @test21(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1, <8 x i64> %y1) nounwind { 407; CHECK-LABEL: test21: 408; CHECK: ## %bb.0: 409; CHECK-NEXT: vpcmpleq %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf3,0xfd,0x48,0x1f,0xc9,0x02] 410; CHECK-NEXT: vpcmpnltq %zmm3, %zmm2, %k1 {%k1} ## encoding: [0x62,0xf3,0xed,0x49,0x1f,0xcb,0x05] 411; CHECK-NEXT: vpblendmq %zmm0, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x49,0x64,0xc0] 412; CHECK-NEXT: retq ## encoding: [0xc3] 413 %mask1 = icmp sge <8 x i64> %x1, %y1 414 %mask0 = icmp sle <8 x i64> %x, %y 415 %mask = select <8 x i1> %mask0, <8 x i1> %mask1, <8 x i1> zeroinitializer 416 %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %x1 417 ret <8 x i64> %max 418} 419 420define <8 x i64> @test22(<8 x i64> %x, <8 x i64>* %y.ptr, <8 x i64> %x1, <8 x i64> %y1) nounwind { 421; CHECK-LABEL: test22: 422; CHECK: ## %bb.0: 423; CHECK-NEXT: vpcmpgtq %zmm2, %zmm1, %k1 ## encoding: [0x62,0xf2,0xf5,0x48,0x37,0xca] 424; CHECK-NEXT: vpcmpgtq (%rdi), %zmm0, %k1 {%k1} ## encoding: [0x62,0xf2,0xfd,0x49,0x37,0x0f] 425; CHECK-NEXT: vpblendmq %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x64,0xc0] 426; CHECK-NEXT: retq ## encoding: [0xc3] 427 %mask1 = icmp sgt <8 x i64> %x1, %y1 428 %y = load <8 x i64>, <8 x i64>* %y.ptr, align 4 429 %mask0 = icmp sgt <8 x i64> %x, %y 430 %mask = select <8 x i1> %mask0, <8 x i1> %mask1, <8 x i1> zeroinitializer 431 %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %x1 432 ret <8 x i64> %max 433} 434 435define <16 x i32> @test23(<16 x i32> %x, <16 x i32>* %y.ptr, <16 x i32> %x1, <16 x i32> %y1) nounwind { 436; CHECK-LABEL: test23: 437; CHECK: ## %bb.0: 438; CHECK-NEXT: vpcmpnltd %zmm2, %zmm1, %k1 ## encoding: [0x62,0xf3,0x75,0x48,0x1f,0xca,0x05] 439; CHECK-NEXT: vpcmpleud (%rdi), %zmm0, %k1 {%k1} ## encoding: [0x62,0xf3,0x7d,0x49,0x1e,0x0f,0x02] 440; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] 441; CHECK-NEXT: retq ## encoding: [0xc3] 442 %mask1 = icmp sge <16 x i32> %x1, %y1 443 %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4 444 %mask0 = icmp ule <16 x i32> %x, %y 445 %mask = select <16 x i1> %mask0, <16 x i1> %mask1, <16 x i1> zeroinitializer 446 %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 447 ret <16 x i32> %max 448} 449 450define <8 x i64> @test24(<8 x i64> %x, <8 x i64> %x1, i64* %yb.ptr) nounwind { 451; CHECK-LABEL: test24: 452; CHECK: ## %bb.0: 453; CHECK-NEXT: vpcmpeqq (%rdi){1to8}, %zmm0, %k1 ## encoding: [0x62,0xf2,0xfd,0x58,0x29,0x0f] 454; CHECK-NEXT: vpblendmq %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x64,0xc0] 455; CHECK-NEXT: retq ## encoding: [0xc3] 456 %yb = load i64, i64* %yb.ptr, align 4 457 %y.0 = insertelement <8 x i64> undef, i64 %yb, i32 0 458 %y = shufflevector <8 x i64> %y.0, <8 x i64> undef, <8 x i32> zeroinitializer 459 %mask = icmp eq <8 x i64> %x, %y 460 %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %x1 461 ret <8 x i64> %max 462} 463 464define <16 x i32> @test25(<16 x i32> %x, i32* %yb.ptr, <16 x i32> %x1) nounwind { 465; CHECK-LABEL: test25: 466; CHECK: ## %bb.0: 467; CHECK-NEXT: vpcmpled (%rdi){1to16}, %zmm0, %k1 ## encoding: [0x62,0xf3,0x7d,0x58,0x1f,0x0f,0x02] 468; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] 469; CHECK-NEXT: retq ## encoding: [0xc3] 470 %yb = load i32, i32* %yb.ptr, align 4 471 %y.0 = insertelement <16 x i32> undef, i32 %yb, i32 0 472 %y = shufflevector <16 x i32> %y.0, <16 x i32> undef, <16 x i32> zeroinitializer 473 %mask = icmp sle <16 x i32> %x, %y 474 %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 475 ret <16 x i32> %max 476} 477 478define <16 x i32> @test26(<16 x i32> %x, i32* %yb.ptr, <16 x i32> %x1, <16 x i32> %y1) nounwind { 479; CHECK-LABEL: test26: 480; CHECK: ## %bb.0: 481; CHECK-NEXT: vpcmpnltd %zmm2, %zmm1, %k1 ## encoding: [0x62,0xf3,0x75,0x48,0x1f,0xca,0x05] 482; CHECK-NEXT: vpcmpgtd (%rdi){1to16}, %zmm0, %k1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x59,0x66,0x0f] 483; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] 484; CHECK-NEXT: retq ## encoding: [0xc3] 485 %mask1 = icmp sge <16 x i32> %x1, %y1 486 %yb = load i32, i32* %yb.ptr, align 4 487 %y.0 = insertelement <16 x i32> undef, i32 %yb, i32 0 488 %y = shufflevector <16 x i32> %y.0, <16 x i32> undef, <16 x i32> zeroinitializer 489 %mask0 = icmp sgt <16 x i32> %x, %y 490 %mask = select <16 x i1> %mask0, <16 x i1> %mask1, <16 x i1> zeroinitializer 491 %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 492 ret <16 x i32> %max 493} 494 495define <8 x i64> @test27(<8 x i64> %x, i64* %yb.ptr, <8 x i64> %x1, <8 x i64> %y1) nounwind { 496; CHECK-LABEL: test27: 497; CHECK: ## %bb.0: 498; CHECK-NEXT: vpcmpnltq %zmm2, %zmm1, %k1 ## encoding: [0x62,0xf3,0xf5,0x48,0x1f,0xca,0x05] 499; CHECK-NEXT: vpcmpleq (%rdi){1to8}, %zmm0, %k1 {%k1} ## encoding: [0x62,0xf3,0xfd,0x59,0x1f,0x0f,0x02] 500; CHECK-NEXT: vpblendmq %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x64,0xc0] 501; CHECK-NEXT: retq ## encoding: [0xc3] 502 %mask1 = icmp sge <8 x i64> %x1, %y1 503 %yb = load i64, i64* %yb.ptr, align 4 504 %y.0 = insertelement <8 x i64> undef, i64 %yb, i32 0 505 %y = shufflevector <8 x i64> %y.0, <8 x i64> undef, <8 x i32> zeroinitializer 506 %mask0 = icmp sle <8 x i64> %x, %y 507 %mask = select <8 x i1> %mask0, <8 x i1> %mask1, <8 x i1> zeroinitializer 508 %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %x1 509 ret <8 x i64> %max 510} 511 512define <8 x i32>@test28(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1, <8 x i64> %y1) { 513; AVX512-LABEL: test28: 514; AVX512: ## %bb.0: 515; AVX512-NEXT: vpcmpgtq %zmm1, %zmm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x48,0x37,0xc1] 516; AVX512-NEXT: vpcmpgtq %zmm3, %zmm2, %k1 ## encoding: [0x62,0xf2,0xed,0x48,0x37,0xcb] 517; AVX512-NEXT: kxnorw %k1, %k0, %k1 ## encoding: [0xc5,0xfc,0x46,0xc9] 518; AVX512-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf3,0x7d,0xc9,0x25,0xc0,0xff] 519; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 520; AVX512-NEXT: retq ## encoding: [0xc3] 521; 522; SKX-LABEL: test28: 523; SKX: ## %bb.0: 524; SKX-NEXT: vpcmpgtq %zmm1, %zmm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x48,0x37,0xc1] 525; SKX-NEXT: vpcmpgtq %zmm3, %zmm2, %k1 ## encoding: [0x62,0xf2,0xed,0x48,0x37,0xcb] 526; SKX-NEXT: kxnorb %k1, %k0, %k0 ## encoding: [0xc5,0xfd,0x46,0xc1] 527; SKX-NEXT: vpmovm2d %k0, %ymm0 ## encoding: [0x62,0xf2,0x7e,0x28,0x38,0xc0] 528; SKX-NEXT: retq ## encoding: [0xc3] 529 %x_gt_y = icmp sgt <8 x i64> %x, %y 530 %x1_gt_y1 = icmp sgt <8 x i64> %x1, %y1 531 %res = icmp eq <8 x i1>%x_gt_y, %x1_gt_y1 532 %resse = sext <8 x i1>%res to <8 x i32> 533 ret <8 x i32> %resse 534} 535 536define <16 x i8>@test29(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1, <16 x i32> %y1) { 537; KNL-LABEL: test29: 538; KNL: ## %bb.0: 539; KNL-NEXT: vpcmpgtd %zmm1, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x66,0xc1] 540; KNL-NEXT: vpcmpgtd %zmm3, %zmm2, %k1 ## encoding: [0x62,0xf1,0x6d,0x48,0x66,0xcb] 541; KNL-NEXT: kxorw %k1, %k0, %k1 ## encoding: [0xc5,0xfc,0x47,0xc9] 542; KNL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf3,0x7d,0xc9,0x25,0xc0,0xff] 543; KNL-NEXT: vpmovdb %zmm0, %xmm0 ## encoding: [0x62,0xf2,0x7e,0x48,0x31,0xc0] 544; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 545; KNL-NEXT: retq ## encoding: [0xc3] 546; 547; AVX512BW-LABEL: test29: 548; AVX512BW: ## %bb.0: 549; AVX512BW-NEXT: vpcmpgtd %zmm1, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x66,0xc1] 550; AVX512BW-NEXT: vpcmpgtd %zmm3, %zmm2, %k1 ## encoding: [0x62,0xf1,0x6d,0x48,0x66,0xcb] 551; AVX512BW-NEXT: kxorw %k1, %k0, %k0 ## encoding: [0xc5,0xfc,0x47,0xc1] 552; AVX512BW-NEXT: vpmovm2b %k0, %zmm0 ## encoding: [0x62,0xf2,0x7e,0x48,0x28,0xc0] 553; AVX512BW-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 554; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 555; AVX512BW-NEXT: retq ## encoding: [0xc3] 556; 557; SKX-LABEL: test29: 558; SKX: ## %bb.0: 559; SKX-NEXT: vpcmpgtd %zmm1, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x66,0xc1] 560; SKX-NEXT: vpcmpgtd %zmm3, %zmm2, %k1 ## encoding: [0x62,0xf1,0x6d,0x48,0x66,0xcb] 561; SKX-NEXT: kxorw %k1, %k0, %k0 ## encoding: [0xc5,0xfc,0x47,0xc1] 562; SKX-NEXT: vpmovm2b %k0, %xmm0 ## encoding: [0x62,0xf2,0x7e,0x08,0x28,0xc0] 563; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 564; SKX-NEXT: retq ## encoding: [0xc3] 565 %x_gt_y = icmp sgt <16 x i32> %x, %y 566 %x1_gt_y1 = icmp sgt <16 x i32> %x1, %y1 567 %res = icmp ne <16 x i1>%x_gt_y, %x1_gt_y1 568 %resse = sext <16 x i1>%res to <16 x i8> 569 ret <16 x i8> %resse 570} 571 572define <4 x double> @test30(<4 x double> %x, <4 x double> %y) nounwind { 573; AVX512-LABEL: test30: 574; AVX512: ## %bb.0: 575; AVX512-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 576; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 577; AVX512-NEXT: vcmpeqpd %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0xc9,0x00] 578; AVX512-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] 579; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 580; AVX512-NEXT: retq ## encoding: [0xc3] 581; 582; SKX-LABEL: test30: 583; SKX: ## %bb.0: 584; SKX-NEXT: vcmpeqpd %ymm1, %ymm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x28,0xc2,0xc9,0x00] 585; SKX-NEXT: vblendmpd %ymm0, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0x65,0xc0] 586; SKX-NEXT: retq ## encoding: [0xc3] 587 588 %mask = fcmp oeq <4 x double> %x, %y 589 %max = select <4 x i1> %mask, <4 x double> %x, <4 x double> %y 590 ret <4 x double> %max 591} 592 593define <2 x double> @test31(<2 x double> %x, <2 x double> %x1, <2 x double>* %yp) nounwind { 594; AVX512-LABEL: test31: 595; AVX512: ## %bb.0: 596; AVX512-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 597; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 598; AVX512-NEXT: vmovupd (%rdi), %xmm2 ## encoding: [0xc5,0xf9,0x10,0x17] 599; AVX512-NEXT: vcmpltpd %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0xca,0x01] 600; AVX512-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] 601; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 602; AVX512-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 603; AVX512-NEXT: retq ## encoding: [0xc3] 604; 605; SKX-LABEL: test31: 606; SKX: ## %bb.0: 607; SKX-NEXT: vcmpltpd (%rdi), %xmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x08,0xc2,0x0f,0x01] 608; SKX-NEXT: vblendmpd %xmm0, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0x65,0xc0] 609; SKX-NEXT: retq ## encoding: [0xc3] 610 611 %y = load <2 x double>, <2 x double>* %yp, align 4 612 %mask = fcmp olt <2 x double> %x, %y 613 %max = select <2 x i1> %mask, <2 x double> %x, <2 x double> %x1 614 ret <2 x double> %max 615} 616 617define <4 x double> @test32(<4 x double> %x, <4 x double> %x1, <4 x double>* %yp) nounwind { 618; AVX512-LABEL: test32: 619; AVX512: ## %bb.0: 620; AVX512-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 621; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 622; AVX512-NEXT: vmovupd (%rdi), %ymm2 ## encoding: [0xc5,0xfd,0x10,0x17] 623; AVX512-NEXT: vcmpltpd %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0xca,0x01] 624; AVX512-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] 625; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 626; AVX512-NEXT: retq ## encoding: [0xc3] 627; 628; SKX-LABEL: test32: 629; SKX: ## %bb.0: 630; SKX-NEXT: vcmpltpd (%rdi), %ymm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x28,0xc2,0x0f,0x01] 631; SKX-NEXT: vblendmpd %ymm0, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0x65,0xc0] 632; SKX-NEXT: retq ## encoding: [0xc3] 633 634 %y = load <4 x double>, <4 x double>* %yp, align 4 635 %mask = fcmp ogt <4 x double> %y, %x 636 %max = select <4 x i1> %mask, <4 x double> %x, <4 x double> %x1 637 ret <4 x double> %max 638} 639 640define <8 x double> @test33(<8 x double> %x, <8 x double> %x1, <8 x double>* %yp) nounwind { 641; CHECK-LABEL: test33: 642; CHECK: ## %bb.0: 643; CHECK-NEXT: vcmpltpd (%rdi), %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0x0f,0x01] 644; CHECK-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] 645; CHECK-NEXT: retq ## encoding: [0xc3] 646 %y = load <8 x double>, <8 x double>* %yp, align 4 647 %mask = fcmp olt <8 x double> %x, %y 648 %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %x1 649 ret <8 x double> %max 650} 651 652define <4 x float> @test34(<4 x float> %x, <4 x float> %x1, <4 x float>* %yp) nounwind { 653; AVX512-LABEL: test34: 654; AVX512: ## %bb.0: 655; AVX512-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 656; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 657; AVX512-NEXT: vmovups (%rdi), %xmm2 ## encoding: [0xc5,0xf8,0x10,0x17] 658; AVX512-NEXT: vcmpltps %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xca,0x01] 659; AVX512-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] 660; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 661; AVX512-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 662; AVX512-NEXT: retq ## encoding: [0xc3] 663; 664; SKX-LABEL: test34: 665; SKX: ## %bb.0: 666; SKX-NEXT: vcmpltps (%rdi), %xmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x08,0xc2,0x0f,0x01] 667; SKX-NEXT: vblendmps %xmm0, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0x65,0xc0] 668; SKX-NEXT: retq ## encoding: [0xc3] 669 %y = load <4 x float>, <4 x float>* %yp, align 4 670 %mask = fcmp olt <4 x float> %x, %y 671 %max = select <4 x i1> %mask, <4 x float> %x, <4 x float> %x1 672 ret <4 x float> %max 673} 674 675define <8 x float> @test35(<8 x float> %x, <8 x float> %x1, <8 x float>* %yp) nounwind { 676; AVX512-LABEL: test35: 677; AVX512: ## %bb.0: 678; AVX512-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 679; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 680; AVX512-NEXT: vmovups (%rdi), %ymm2 ## encoding: [0xc5,0xfc,0x10,0x17] 681; AVX512-NEXT: vcmpltps %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xca,0x01] 682; AVX512-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] 683; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 684; AVX512-NEXT: retq ## encoding: [0xc3] 685; 686; SKX-LABEL: test35: 687; SKX: ## %bb.0: 688; SKX-NEXT: vcmpltps (%rdi), %ymm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x28,0xc2,0x0f,0x01] 689; SKX-NEXT: vblendmps %ymm0, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0x65,0xc0] 690; SKX-NEXT: retq ## encoding: [0xc3] 691 692 %y = load <8 x float>, <8 x float>* %yp, align 4 693 %mask = fcmp ogt <8 x float> %y, %x 694 %max = select <8 x i1> %mask, <8 x float> %x, <8 x float> %x1 695 ret <8 x float> %max 696} 697 698define <16 x float> @test36(<16 x float> %x, <16 x float> %x1, <16 x float>* %yp) nounwind { 699; CHECK-LABEL: test36: 700; CHECK: ## %bb.0: 701; CHECK-NEXT: vcmpltps (%rdi), %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0x0f,0x01] 702; CHECK-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] 703; CHECK-NEXT: retq ## encoding: [0xc3] 704 %y = load <16 x float>, <16 x float>* %yp, align 4 705 %mask = fcmp olt <16 x float> %x, %y 706 %max = select <16 x i1> %mask, <16 x float> %x, <16 x float> %x1 707 ret <16 x float> %max 708} 709 710define <8 x double> @test37(<8 x double> %x, <8 x double> %x1, double* %ptr) nounwind { 711; CHECK-LABEL: test37: 712; CHECK: ## %bb.0: 713; CHECK-NEXT: vcmpltpd (%rdi){1to8}, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x58,0xc2,0x0f,0x01] 714; CHECK-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] 715; CHECK-NEXT: retq ## encoding: [0xc3] 716 717 %a = load double, double* %ptr 718 %v = insertelement <8 x double> undef, double %a, i32 0 719 %shuffle = shufflevector <8 x double> %v, <8 x double> undef, <8 x i32> zeroinitializer 720 721 %mask = fcmp ogt <8 x double> %shuffle, %x 722 %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %x1 723 ret <8 x double> %max 724} 725 726define <4 x double> @test38(<4 x double> %x, <4 x double> %x1, double* %ptr) nounwind { 727; AVX512-LABEL: test38: 728; AVX512: ## %bb.0: 729; AVX512-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 730; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 731; AVX512-NEXT: vbroadcastsd (%rdi), %ymm2 ## encoding: [0xc4,0xe2,0x7d,0x19,0x17] 732; AVX512-NEXT: vcmpltpd %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0xca,0x01] 733; AVX512-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] 734; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 735; AVX512-NEXT: retq ## encoding: [0xc3] 736; 737; SKX-LABEL: test38: 738; SKX: ## %bb.0: 739; SKX-NEXT: vcmpltpd (%rdi){1to4}, %ymm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x38,0xc2,0x0f,0x01] 740; SKX-NEXT: vblendmpd %ymm0, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0x65,0xc0] 741; SKX-NEXT: retq ## encoding: [0xc3] 742 743 %a = load double, double* %ptr 744 %v = insertelement <4 x double> undef, double %a, i32 0 745 %shuffle = shufflevector <4 x double> %v, <4 x double> undef, <4 x i32> zeroinitializer 746 747 %mask = fcmp ogt <4 x double> %shuffle, %x 748 %max = select <4 x i1> %mask, <4 x double> %x, <4 x double> %x1 749 ret <4 x double> %max 750} 751 752define <2 x double> @test39(<2 x double> %x, <2 x double> %x1, double* %ptr) nounwind { 753; AVX512-LABEL: test39: 754; AVX512: ## %bb.0: 755; AVX512-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 756; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 757; AVX512-NEXT: vmovddup (%rdi), %xmm2 ## encoding: [0xc5,0xfb,0x12,0x17] 758; AVX512-NEXT: ## xmm2 = mem[0,0] 759; AVX512-NEXT: vcmpltpd %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0xca,0x01] 760; AVX512-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] 761; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 762; AVX512-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 763; AVX512-NEXT: retq ## encoding: [0xc3] 764; 765; SKX-LABEL: test39: 766; SKX: ## %bb.0: 767; SKX-NEXT: vcmpltpd (%rdi){1to2}, %xmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x18,0xc2,0x0f,0x01] 768; SKX-NEXT: vblendmpd %xmm0, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0x65,0xc0] 769; SKX-NEXT: retq ## encoding: [0xc3] 770 771 %a = load double, double* %ptr 772 %v = insertelement <2 x double> undef, double %a, i32 0 773 %shuffle = shufflevector <2 x double> %v, <2 x double> undef, <2 x i32> <i32 0, i32 0> 774 775 %mask = fcmp ogt <2 x double> %shuffle, %x 776 %max = select <2 x i1> %mask, <2 x double> %x, <2 x double> %x1 777 ret <2 x double> %max 778} 779 780 781define <16 x float> @test40(<16 x float> %x, <16 x float> %x1, float* %ptr) nounwind { 782; CHECK-LABEL: test40: 783; CHECK: ## %bb.0: 784; CHECK-NEXT: vcmpltps (%rdi){1to16}, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x58,0xc2,0x0f,0x01] 785; CHECK-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] 786; CHECK-NEXT: retq ## encoding: [0xc3] 787 788 %a = load float, float* %ptr 789 %v = insertelement <16 x float> undef, float %a, i32 0 790 %shuffle = shufflevector <16 x float> %v, <16 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> 791 792 %mask = fcmp ogt <16 x float> %shuffle, %x 793 %max = select <16 x i1> %mask, <16 x float> %x, <16 x float> %x1 794 ret <16 x float> %max 795} 796 797define <8 x float> @test41(<8 x float> %x, <8 x float> %x1, float* %ptr) nounwind { 798; AVX512-LABEL: test41: 799; AVX512: ## %bb.0: 800; AVX512-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 801; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 802; AVX512-NEXT: vbroadcastss (%rdi), %ymm2 ## encoding: [0xc4,0xe2,0x7d,0x18,0x17] 803; AVX512-NEXT: vcmpltps %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xca,0x01] 804; AVX512-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] 805; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 806; AVX512-NEXT: retq ## encoding: [0xc3] 807; 808; SKX-LABEL: test41: 809; SKX: ## %bb.0: 810; SKX-NEXT: vcmpltps (%rdi){1to8}, %ymm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x38,0xc2,0x0f,0x01] 811; SKX-NEXT: vblendmps %ymm0, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0x65,0xc0] 812; SKX-NEXT: retq ## encoding: [0xc3] 813 814 %a = load float, float* %ptr 815 %v = insertelement <8 x float> undef, float %a, i32 0 816 %shuffle = shufflevector <8 x float> %v, <8 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> 817 818 %mask = fcmp ogt <8 x float> %shuffle, %x 819 %max = select <8 x i1> %mask, <8 x float> %x, <8 x float> %x1 820 ret <8 x float> %max 821} 822 823define <4 x float> @test42(<4 x float> %x, <4 x float> %x1, float* %ptr) nounwind { 824; AVX512-LABEL: test42: 825; AVX512: ## %bb.0: 826; AVX512-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 827; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 828; AVX512-NEXT: vbroadcastss (%rdi), %xmm2 ## encoding: [0xc4,0xe2,0x79,0x18,0x17] 829; AVX512-NEXT: vcmpltps %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xca,0x01] 830; AVX512-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] 831; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 832; AVX512-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 833; AVX512-NEXT: retq ## encoding: [0xc3] 834; 835; SKX-LABEL: test42: 836; SKX: ## %bb.0: 837; SKX-NEXT: vcmpltps (%rdi){1to4}, %xmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x18,0xc2,0x0f,0x01] 838; SKX-NEXT: vblendmps %xmm0, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0x65,0xc0] 839; SKX-NEXT: retq ## encoding: [0xc3] 840 841 %a = load float, float* %ptr 842 %v = insertelement <4 x float> undef, float %a, i32 0 843 %shuffle = shufflevector <4 x float> %v, <4 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> 844 845 %mask = fcmp ogt <4 x float> %shuffle, %x 846 %max = select <4 x i1> %mask, <4 x float> %x, <4 x float> %x1 847 ret <4 x float> %max 848} 849 850define <8 x double> @test43(<8 x double> %x, <8 x double> %x1, double* %ptr,<8 x i1> %mask_in) nounwind { 851; KNL-LABEL: test43: 852; KNL: ## %bb.0: 853; KNL-NEXT: vpmovsxwq %xmm2, %zmm2 ## encoding: [0x62,0xf2,0x7d,0x48,0x24,0xd2] 854; KNL-NEXT: vpsllq $63, %zmm2, %zmm2 ## encoding: [0x62,0xf1,0xed,0x48,0x73,0xf2,0x3f] 855; KNL-NEXT: vptestmq %zmm2, %zmm2, %k1 ## encoding: [0x62,0xf2,0xed,0x48,0x27,0xca] 856; KNL-NEXT: vcmpltpd (%rdi){1to8}, %zmm0, %k1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x59,0xc2,0x0f,0x01] 857; KNL-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] 858; KNL-NEXT: retq ## encoding: [0xc3] 859; 860; AVX512BW-LABEL: test43: 861; AVX512BW: ## %bb.0: 862; AVX512BW-NEXT: vpsllw $15, %xmm2, %xmm2 ## encoding: [0xc5,0xe9,0x71,0xf2,0x0f] 863; AVX512BW-NEXT: vpmovw2m %zmm2, %k1 ## encoding: [0x62,0xf2,0xfe,0x48,0x29,0xca] 864; AVX512BW-NEXT: vcmpltpd (%rdi){1to8}, %zmm0, %k1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x59,0xc2,0x0f,0x01] 865; AVX512BW-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] 866; AVX512BW-NEXT: retq ## encoding: [0xc3] 867; 868; SKX-LABEL: test43: 869; SKX: ## %bb.0: 870; SKX-NEXT: vpsllw $15, %xmm2, %xmm2 ## EVEX TO VEX Compression encoding: [0xc5,0xe9,0x71,0xf2,0x0f] 871; SKX-NEXT: vpmovw2m %xmm2, %k1 ## encoding: [0x62,0xf2,0xfe,0x08,0x29,0xca] 872; SKX-NEXT: vcmpltpd (%rdi){1to8}, %zmm0, %k1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x59,0xc2,0x0f,0x01] 873; SKX-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] 874; SKX-NEXT: retq ## encoding: [0xc3] 875 876 %a = load double, double* %ptr 877 %v = insertelement <8 x double> undef, double %a, i32 0 878 %shuffle = shufflevector <8 x double> %v, <8 x double> undef, <8 x i32> zeroinitializer 879 880 %mask_cmp = fcmp ogt <8 x double> %shuffle, %x 881 %mask = and <8 x i1> %mask_cmp, %mask_in 882 %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %x1 883 ret <8 x double> %max 884} 885 886define <4 x i32> @test44(<4 x i16> %x, <4 x i16> %y) #0 { 887; AVX512-LABEL: test44: 888; AVX512: ## %bb.0: 889; AVX512-NEXT: vpxor %xmm2, %xmm2, %xmm2 ## encoding: [0xc5,0xe9,0xef,0xd2] 890; AVX512-NEXT: vpblendw $170, %xmm2, %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x0e,0xca,0xaa] 891; AVX512-NEXT: ## xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] 892; AVX512-NEXT: vpblendw $170, %xmm2, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x0e,0xc2,0xaa] 893; AVX512-NEXT: ## xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] 894; AVX512-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x76,0xc1] 895; AVX512-NEXT: retq ## encoding: [0xc3] 896; 897; SKX-LABEL: test44: 898; SKX: ## %bb.0: 899; SKX-NEXT: vpxor %xmm2, %xmm2, %xmm2 ## EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2] 900; SKX-NEXT: vpblendw $170, %xmm2, %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x0e,0xca,0xaa] 901; SKX-NEXT: ## xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] 902; SKX-NEXT: vpblendw $170, %xmm2, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x0e,0xc2,0xaa] 903; SKX-NEXT: ## xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] 904; SKX-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x76,0xc1] 905; SKX-NEXT: retq ## encoding: [0xc3] 906 %mask = icmp eq <4 x i16> %x, %y 907 %1 = sext <4 x i1> %mask to <4 x i32> 908 ret <4 x i32> %1 909} 910 911define <2 x i64> @test45(<2 x i16> %x, <2 x i16> %y) #0 { 912; AVX512-LABEL: test45: 913; AVX512: ## %bb.0: 914; AVX512-NEXT: vpxor %xmm2, %xmm2, %xmm2 ## encoding: [0xc5,0xe9,0xef,0xd2] 915; AVX512-NEXT: vpblendw $17, %xmm1, %xmm2, %xmm1 ## encoding: [0xc4,0xe3,0x69,0x0e,0xc9,0x11] 916; AVX512-NEXT: ## xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] 917; AVX512-NEXT: vpblendw $17, %xmm0, %xmm2, %xmm0 ## encoding: [0xc4,0xe3,0x69,0x0e,0xc0,0x11] 918; AVX512-NEXT: ## xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] 919; AVX512-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0 ## encoding: [0xc4,0xe2,0x79,0x29,0xc1] 920; AVX512-NEXT: vpsrlq $63, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x73,0xd0,0x3f] 921; AVX512-NEXT: retq ## encoding: [0xc3] 922; 923; SKX-LABEL: test45: 924; SKX: ## %bb.0: 925; SKX-NEXT: vpxor %xmm2, %xmm2, %xmm2 ## EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2] 926; SKX-NEXT: vpblendw $17, %xmm1, %xmm2, %xmm1 ## encoding: [0xc4,0xe3,0x69,0x0e,0xc9,0x11] 927; SKX-NEXT: ## xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] 928; SKX-NEXT: vpblendw $17, %xmm0, %xmm2, %xmm0 ## encoding: [0xc4,0xe3,0x69,0x0e,0xc0,0x11] 929; SKX-NEXT: ## xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] 930; SKX-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0 ## encoding: [0xc4,0xe2,0x79,0x29,0xc1] 931; SKX-NEXT: vpsrlq $63, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x73,0xd0,0x3f] 932; SKX-NEXT: retq ## encoding: [0xc3] 933 %mask = icmp eq <2 x i16> %x, %y 934 %1 = zext <2 x i1> %mask to <2 x i64> 935 ret <2 x i64> %1 936} 937 938define <2 x i64> @test46(<2 x float> %x, <2 x float> %y) #0 { 939; AVX512-LABEL: test46: 940; AVX512: ## %bb.0: 941; AVX512-NEXT: vcmpeqps %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf8,0xc2,0xc1,0x00] 942; AVX512-NEXT: vpmovzxdq %xmm0, %xmm0 ## encoding: [0xc4,0xe2,0x79,0x35,0xc0] 943; AVX512-NEXT: ## xmm0 = xmm0[0],zero,xmm0[1],zero 944; AVX512-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0xdb,0x05,A,A,A,A] 945; AVX512-NEXT: ## fixup A - offset: 4, value: LCPI47_0-4, kind: reloc_riprel_4byte 946; AVX512-NEXT: retq ## encoding: [0xc3] 947; 948; SKX-LABEL: test46: 949; SKX: ## %bb.0: 950; SKX-NEXT: vcmpeqps %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf1,0x7c,0x08,0xc2,0xc1,0x00] 951; SKX-NEXT: vpmovm2q %k0, %xmm0 ## encoding: [0x62,0xf2,0xfe,0x08,0x38,0xc0] 952; SKX-NEXT: vpsrlq $63, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x73,0xd0,0x3f] 953; SKX-NEXT: retq ## encoding: [0xc3] 954 %mask = fcmp oeq <2 x float> %x, %y 955 %1 = zext <2 x i1> %mask to <2 x i64> 956 ret <2 x i64> %1 957} 958 959define <16 x i8> @test47(<16 x i32> %a, <16 x i8> %b, <16 x i8> %c) { 960; KNL-LABEL: test47: 961; KNL: ## %bb.0: 962; KNL-NEXT: vptestnmd %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0x7e,0x48,0x27,0xc8] 963; KNL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf3,0x7d,0xc9,0x25,0xc0,0xff] 964; KNL-NEXT: vpmovdb %zmm0, %xmm0 ## encoding: [0x62,0xf2,0x7e,0x48,0x31,0xc0] 965; KNL-NEXT: vpblendvb %xmm0, %xmm1, %xmm2, %xmm0 ## encoding: [0xc4,0xe3,0x69,0x4c,0xc1,0x00] 966; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 967; KNL-NEXT: retq ## encoding: [0xc3] 968; 969; AVX512BW-LABEL: test47: 970; AVX512BW: ## %bb.0: 971; AVX512BW-NEXT: ## kill: def $xmm2 killed $xmm2 def $zmm2 972; AVX512BW-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 973; AVX512BW-NEXT: vptestnmd %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0x7e,0x48,0x27,0xc8] 974; AVX512BW-NEXT: vpblendmb %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x49,0x66,0xc1] 975; AVX512BW-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 976; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 977; AVX512BW-NEXT: retq ## encoding: [0xc3] 978; 979; SKX-LABEL: test47: 980; SKX: ## %bb.0: 981; SKX-NEXT: vptestnmd %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0x7e,0x48,0x27,0xc8] 982; SKX-NEXT: vpblendmb %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x09,0x66,0xc1] 983; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 984; SKX-NEXT: retq ## encoding: [0xc3] 985 %cmp = icmp eq <16 x i32> %a, zeroinitializer 986 %res = select <16 x i1> %cmp, <16 x i8> %b, <16 x i8> %c 987 ret <16 x i8> %res 988} 989 990define <16 x i16> @test48(<16 x i32> %a, <16 x i16> %b, <16 x i16> %c) { 991; KNL-LABEL: test48: 992; KNL: ## %bb.0: 993; KNL-NEXT: vptestnmd %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0x7e,0x48,0x27,0xc8] 994; KNL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf3,0x7d,0xc9,0x25,0xc0,0xff] 995; KNL-NEXT: vpmovdw %zmm0, %ymm0 ## encoding: [0x62,0xf2,0x7e,0x48,0x33,0xc0] 996; KNL-NEXT: vpblendvb %ymm0, %ymm1, %ymm2, %ymm0 ## encoding: [0xc4,0xe3,0x6d,0x4c,0xc1,0x00] 997; KNL-NEXT: retq ## encoding: [0xc3] 998; 999; AVX512BW-LABEL: test48: 1000; AVX512BW: ## %bb.0: 1001; AVX512BW-NEXT: ## kill: def $ymm2 killed $ymm2 def $zmm2 1002; AVX512BW-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 1003; AVX512BW-NEXT: vptestnmd %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0x7e,0x48,0x27,0xc8] 1004; AVX512BW-NEXT: vpblendmw %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x49,0x66,0xc1] 1005; AVX512BW-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 1006; AVX512BW-NEXT: retq ## encoding: [0xc3] 1007; 1008; SKX-LABEL: test48: 1009; SKX: ## %bb.0: 1010; SKX-NEXT: vptestnmd %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0x7e,0x48,0x27,0xc8] 1011; SKX-NEXT: vpblendmw %ymm1, %ymm2, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x29,0x66,0xc1] 1012; SKX-NEXT: retq ## encoding: [0xc3] 1013 %cmp = icmp eq <16 x i32> %a, zeroinitializer 1014 %res = select <16 x i1> %cmp, <16 x i16> %b, <16 x i16> %c 1015 ret <16 x i16> %res 1016} 1017 1018define <8 x i16> @test49(<8 x i64> %a, <8 x i16> %b, <8 x i16> %c) { 1019; KNL-LABEL: test49: 1020; KNL: ## %bb.0: 1021; KNL-NEXT: vptestnmq %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0xfe,0x48,0x27,0xc8] 1022; KNL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf3,0x7d,0xc9,0x25,0xc0,0xff] 1023; KNL-NEXT: vpmovdw %zmm0, %ymm0 ## encoding: [0x62,0xf2,0x7e,0x48,0x33,0xc0] 1024; KNL-NEXT: vpblendvb %xmm0, %xmm1, %xmm2, %xmm0 ## encoding: [0xc4,0xe3,0x69,0x4c,0xc1,0x00] 1025; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 1026; KNL-NEXT: retq ## encoding: [0xc3] 1027; 1028; AVX512BW-LABEL: test49: 1029; AVX512BW: ## %bb.0: 1030; AVX512BW-NEXT: ## kill: def $xmm2 killed $xmm2 def $zmm2 1031; AVX512BW-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 1032; AVX512BW-NEXT: vptestnmq %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0xfe,0x48,0x27,0xc8] 1033; AVX512BW-NEXT: vpblendmw %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x49,0x66,0xc1] 1034; AVX512BW-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 1035; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 1036; AVX512BW-NEXT: retq ## encoding: [0xc3] 1037; 1038; SKX-LABEL: test49: 1039; SKX: ## %bb.0: 1040; SKX-NEXT: vptestnmq %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0xfe,0x48,0x27,0xc8] 1041; SKX-NEXT: vpblendmw %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x09,0x66,0xc1] 1042; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 1043; SKX-NEXT: retq ## encoding: [0xc3] 1044 %cmp = icmp eq <8 x i64> %a, zeroinitializer 1045 %res = select <8 x i1> %cmp, <8 x i16> %b, <8 x i16> %c 1046 ret <8 x i16> %res 1047} 1048 1049define i16 @pcmpeq_mem_1(<16 x i32> %a, <16 x i32>* %b) { 1050; KNL-LABEL: pcmpeq_mem_1: 1051; KNL: ## %bb.0: 1052; KNL-NEXT: vpcmpeqd (%rdi), %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0x07] 1053; KNL-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0] 1054; KNL-NEXT: ## kill: def $ax killed $ax killed $eax 1055; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 1056; KNL-NEXT: retq ## encoding: [0xc3] 1057; 1058; AVX512BW-LABEL: pcmpeq_mem_1: 1059; AVX512BW: ## %bb.0: 1060; AVX512BW-NEXT: vpcmpeqd (%rdi), %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0x07] 1061; AVX512BW-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] 1062; AVX512BW-NEXT: ## kill: def $ax killed $ax killed $eax 1063; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 1064; AVX512BW-NEXT: retq ## encoding: [0xc3] 1065; 1066; SKX-LABEL: pcmpeq_mem_1: 1067; SKX: ## %bb.0: 1068; SKX-NEXT: vpcmpeqd (%rdi), %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0x07] 1069; SKX-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] 1070; SKX-NEXT: ## kill: def $ax killed $ax killed $eax 1071; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 1072; SKX-NEXT: retq ## encoding: [0xc3] 1073 %load = load <16 x i32>, <16 x i32>* %b 1074 %cmp = icmp eq <16 x i32> %a, %load 1075 %cast = bitcast <16 x i1> %cmp to i16 1076 ret i16 %cast 1077} 1078 1079; Make sure we use the short pcmpeq encoding like the test above when the memoryo 1080; operand is in the first argument instead of the second. 1081define i16 @pcmpeq_mem_2(<16 x i32> %a, <16 x i32>* %b) { 1082; KNL-LABEL: pcmpeq_mem_2: 1083; KNL: ## %bb.0: 1084; KNL-NEXT: vpcmpeqd (%rdi), %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0x07] 1085; KNL-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0] 1086; KNL-NEXT: ## kill: def $ax killed $ax killed $eax 1087; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 1088; KNL-NEXT: retq ## encoding: [0xc3] 1089; 1090; AVX512BW-LABEL: pcmpeq_mem_2: 1091; AVX512BW: ## %bb.0: 1092; AVX512BW-NEXT: vpcmpeqd (%rdi), %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0x07] 1093; AVX512BW-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] 1094; AVX512BW-NEXT: ## kill: def $ax killed $ax killed $eax 1095; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 1096; AVX512BW-NEXT: retq ## encoding: [0xc3] 1097; 1098; SKX-LABEL: pcmpeq_mem_2: 1099; SKX: ## %bb.0: 1100; SKX-NEXT: vpcmpeqd (%rdi), %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0x07] 1101; SKX-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] 1102; SKX-NEXT: ## kill: def $ax killed $ax killed $eax 1103; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] 1104; SKX-NEXT: retq ## encoding: [0xc3] 1105 %load = load <16 x i32>, <16 x i32>* %b 1106 %cmp = icmp eq <16 x i32> %load, %a 1107 %cast = bitcast <16 x i1> %cmp to i16 1108 ret i16 %cast 1109} 1110