| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py |
| ; RUN: llc < %s -mtriple=x86_64-apple-darwin -show-mc-encoding -mattr=+avx512f | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=KNL |
| ; 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 |
| ; RUN: llc < %s -mtriple=x86_64-apple-darwin -show-mc-encoding -mattr=+avx512f,+avx512bw,+avx512vl,+avx512dq | FileCheck %s --check-prefix=CHECK --check-prefix=SKX |
| |
| define <16 x float> @test1(<16 x float> %x, <16 x float> %y) nounwind { |
| ; CHECK-LABEL: test1: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vcmpleps %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xc9,0x02] |
| ; CHECK-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %mask = fcmp ole <16 x float> %x, %y |
| %max = select <16 x i1> %mask, <16 x float> %x, <16 x float> %y |
| ret <16 x float> %max |
| } |
| |
| define <8 x double> @test2(<8 x double> %x, <8 x double> %y) nounwind { |
| ; CHECK-LABEL: test2: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vcmplepd %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0xc9,0x02] |
| ; CHECK-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %mask = fcmp ole <8 x double> %x, %y |
| %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %y |
| ret <8 x double> %max |
| } |
| |
| define <16 x i32> @test3(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %yp) nounwind { |
| ; CHECK-LABEL: test3: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpeqd (%rdi), %zmm0, %k1 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0x0f] |
| ; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %y = load <16 x i32>, <16 x i32>* %yp, align 4 |
| %mask = icmp eq <16 x i32> %x, %y |
| %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 |
| ret <16 x i32> %max |
| } |
| |
| define <16 x i32> @test4_unsigned(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1) nounwind { |
| ; CHECK-LABEL: test4_unsigned: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpnltud %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf3,0x7d,0x48,0x1e,0xc9,0x05] |
| ; CHECK-NEXT: vpblendmd %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc2] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %mask = icmp uge <16 x i32> %x, %y |
| %max = select <16 x i1> %mask, <16 x i32> %x1, <16 x i32> %y |
| ret <16 x i32> %max |
| } |
| |
| define <8 x i64> @test5(<8 x i64> %x, <8 x i64> %y) nounwind { |
| ; CHECK-LABEL: test5: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpeqq %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf2,0xfd,0x48,0x29,0xc9] |
| ; CHECK-NEXT: vpblendmq %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x64,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %mask = icmp eq <8 x i64> %x, %y |
| %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %y |
| ret <8 x i64> %max |
| } |
| |
| define <8 x i64> @test6_unsigned(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1) nounwind { |
| ; CHECK-LABEL: test6_unsigned: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpnleuq %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf3,0xfd,0x48,0x1e,0xc9,0x06] |
| ; CHECK-NEXT: vpblendmq %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x64,0xc2] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %mask = icmp ugt <8 x i64> %x, %y |
| %max = select <8 x i1> %mask, <8 x i64> %x1, <8 x i64> %y |
| ret <8 x i64> %max |
| } |
| |
| define <4 x float> @test7(<4 x float> %a, <4 x float> %b) { |
| ; AVX512-LABEL: test7: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 |
| ; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 |
| ; AVX512-NEXT: vxorps %xmm2, %xmm2, %xmm2 ## encoding: [0xc5,0xe8,0x57,0xd2] |
| ; AVX512-NEXT: vcmpltps %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xca,0x01] |
| ; AVX512-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] |
| ; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 |
| ; AVX512-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test7: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vxorps %xmm2, %xmm2, %xmm2 ## EVEX TO VEX Compression encoding: [0xc5,0xe8,0x57,0xd2] |
| ; SKX-NEXT: vcmpltps %xmm2, %xmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x08,0xc2,0xca,0x01] |
| ; SKX-NEXT: vblendmps %xmm0, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0x65,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| |
| %mask = fcmp olt <4 x float> %a, zeroinitializer |
| %c = select <4 x i1>%mask, <4 x float>%a, <4 x float>%b |
| ret <4 x float>%c |
| } |
| |
| define <2 x double> @test8(<2 x double> %a, <2 x double> %b) { |
| ; AVX512-LABEL: test8: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 |
| ; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 |
| ; AVX512-NEXT: vxorpd %xmm2, %xmm2, %xmm2 ## encoding: [0xc5,0xe9,0x57,0xd2] |
| ; AVX512-NEXT: vcmpltpd %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0xca,0x01] |
| ; AVX512-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] |
| ; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 |
| ; AVX512-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test8: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vxorpd %xmm2, %xmm2, %xmm2 ## EVEX TO VEX Compression encoding: [0xc5,0xe9,0x57,0xd2] |
| ; SKX-NEXT: vcmpltpd %xmm2, %xmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x08,0xc2,0xca,0x01] |
| ; SKX-NEXT: vblendmpd %xmm0, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0x65,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %mask = fcmp olt <2 x double> %a, zeroinitializer |
| %c = select <2 x i1>%mask, <2 x double>%a, <2 x double>%b |
| ret <2 x double>%c |
| } |
| |
| define <8 x i32> @test9(<8 x i32> %x, <8 x i32> %y) nounwind { |
| ; AVX512-LABEL: test9: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 |
| ; AVX512-NEXT: vpcmpeqd %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0xc9] |
| ; AVX512-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test9: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vpcmpeqd %ymm1, %ymm0, %k1 ## encoding: [0x62,0xf1,0x7d,0x28,0x76,0xc9] |
| ; SKX-NEXT: vpblendmd %ymm0, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0x64,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %mask = icmp eq <8 x i32> %x, %y |
| %max = select <8 x i1> %mask, <8 x i32> %x, <8 x i32> %y |
| ret <8 x i32> %max |
| } |
| |
| define <8 x float> @test10(<8 x float> %x, <8 x float> %y) nounwind { |
| ; AVX512-LABEL: test10: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 |
| ; AVX512-NEXT: vcmpeqps %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xc9,0x00] |
| ; AVX512-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test10: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vcmpeqps %ymm1, %ymm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x28,0xc2,0xc9,0x00] |
| ; SKX-NEXT: vblendmps %ymm0, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0x65,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| |
| %mask = fcmp oeq <8 x float> %x, %y |
| %max = select <8 x i1> %mask, <8 x float> %x, <8 x float> %y |
| ret <8 x float> %max |
| } |
| |
| define <8 x i32> @test11_unsigned(<8 x i32> %x, <8 x i32> %y) nounwind { |
| ; AVX512-LABEL: test11_unsigned: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: vpmaxud %ymm1, %ymm0, %ymm0 ## encoding: [0xc4,0xe2,0x7d,0x3f,0xc1] |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test11_unsigned: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vpmaxud %ymm1, %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x7d,0x3f,0xc1] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %mask = icmp ugt <8 x i32> %x, %y |
| %max = select <8 x i1> %mask, <8 x i32> %x, <8 x i32> %y |
| ret <8 x i32> %max |
| } |
| |
| define i16 @test12(<16 x i64> %a, <16 x i64> %b) nounwind { |
| ; KNL-LABEL: test12: |
| ; KNL: ## %bb.0: |
| ; KNL-NEXT: vpcmpeqq %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x48,0x29,0xc2] |
| ; KNL-NEXT: vpcmpeqq %zmm3, %zmm1, %k1 ## encoding: [0x62,0xf2,0xf5,0x48,0x29,0xcb] |
| ; KNL-NEXT: kunpckbw %k0, %k1, %k0 ## encoding: [0xc5,0xf5,0x4b,0xc0] |
| ; KNL-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0] |
| ; KNL-NEXT: ## kill: def $ax killed $ax killed $eax |
| ; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; KNL-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; AVX512BW-LABEL: test12: |
| ; AVX512BW: ## %bb.0: |
| ; AVX512BW-NEXT: vpcmpeqq %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x48,0x29,0xc2] |
| ; AVX512BW-NEXT: vpcmpeqq %zmm3, %zmm1, %k1 ## encoding: [0x62,0xf2,0xf5,0x48,0x29,0xcb] |
| ; AVX512BW-NEXT: kunpckbw %k0, %k1, %k0 ## encoding: [0xc5,0xf5,0x4b,0xc0] |
| ; AVX512BW-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] |
| ; AVX512BW-NEXT: ## kill: def $ax killed $ax killed $eax |
| ; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; AVX512BW-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test12: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vpcmpeqq %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x48,0x29,0xc2] |
| ; SKX-NEXT: vpcmpeqq %zmm3, %zmm1, %k1 ## encoding: [0x62,0xf2,0xf5,0x48,0x29,0xcb] |
| ; SKX-NEXT: kunpckbw %k0, %k1, %k0 ## encoding: [0xc5,0xf5,0x4b,0xc0] |
| ; SKX-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] |
| ; SKX-NEXT: ## kill: def $ax killed $ax killed $eax |
| ; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %res = icmp eq <16 x i64> %a, %b |
| %res1 = bitcast <16 x i1> %res to i16 |
| ret i16 %res1 |
| } |
| |
| define i32 @test12_v32i32(<32 x i32> %a, <32 x i32> %b) nounwind { |
| ; KNL-LABEL: test12_v32i32: |
| ; KNL: ## %bb.0: |
| ; KNL-NEXT: vpcmpeqd %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0xc2] |
| ; KNL-NEXT: kmovw %k0, %ecx ## encoding: [0xc5,0xf8,0x93,0xc8] |
| ; KNL-NEXT: vpcmpeqd %zmm3, %zmm1, %k0 ## encoding: [0x62,0xf1,0x75,0x48,0x76,0xc3] |
| ; KNL-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0] |
| ; KNL-NEXT: shll $16, %eax ## encoding: [0xc1,0xe0,0x10] |
| ; KNL-NEXT: orl %ecx, %eax ## encoding: [0x09,0xc8] |
| ; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; KNL-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; AVX512BW-LABEL: test12_v32i32: |
| ; AVX512BW: ## %bb.0: |
| ; AVX512BW-NEXT: vpcmpeqd %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0xc2] |
| ; AVX512BW-NEXT: vpcmpeqd %zmm3, %zmm1, %k1 ## encoding: [0x62,0xf1,0x75,0x48,0x76,0xcb] |
| ; AVX512BW-NEXT: kunpckwd %k0, %k1, %k0 ## encoding: [0xc5,0xf4,0x4b,0xc0] |
| ; AVX512BW-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] |
| ; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; AVX512BW-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test12_v32i32: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vpcmpeqd %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0xc2] |
| ; SKX-NEXT: vpcmpeqd %zmm3, %zmm1, %k1 ## encoding: [0x62,0xf1,0x75,0x48,0x76,0xcb] |
| ; SKX-NEXT: kunpckwd %k0, %k1, %k0 ## encoding: [0xc5,0xf4,0x4b,0xc0] |
| ; SKX-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] |
| ; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %res = icmp eq <32 x i32> %a, %b |
| %res1 = bitcast <32 x i1> %res to i32 |
| ret i32 %res1 |
| } |
| |
| define i64 @test12_v64i16(<64 x i16> %a, <64 x i16> %b) nounwind { |
| ; KNL-LABEL: test12_v64i16: |
| ; KNL: ## %bb.0: |
| ; KNL-NEXT: vpcmpeqw %ymm4, %ymm0, %ymm0 ## encoding: [0xc5,0xfd,0x75,0xc4] |
| ; KNL-NEXT: vpmovsxwd %ymm0, %zmm0 ## encoding: [0x62,0xf2,0x7d,0x48,0x23,0xc0] |
| ; KNL-NEXT: vptestmd %zmm0, %zmm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x48,0x27,0xc0] |
| ; KNL-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0] |
| ; KNL-NEXT: vpcmpeqw %ymm5, %ymm1, %ymm0 ## encoding: [0xc5,0xf5,0x75,0xc5] |
| ; KNL-NEXT: vpmovsxwd %ymm0, %zmm0 ## encoding: [0x62,0xf2,0x7d,0x48,0x23,0xc0] |
| ; KNL-NEXT: vptestmd %zmm0, %zmm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x48,0x27,0xc0] |
| ; KNL-NEXT: kmovw %k0, %ecx ## encoding: [0xc5,0xf8,0x93,0xc8] |
| ; KNL-NEXT: shll $16, %ecx ## encoding: [0xc1,0xe1,0x10] |
| ; KNL-NEXT: orl %eax, %ecx ## encoding: [0x09,0xc1] |
| ; KNL-NEXT: vpcmpeqw %ymm6, %ymm2, %ymm0 ## encoding: [0xc5,0xed,0x75,0xc6] |
| ; KNL-NEXT: vpmovsxwd %ymm0, %zmm0 ## encoding: [0x62,0xf2,0x7d,0x48,0x23,0xc0] |
| ; KNL-NEXT: vptestmd %zmm0, %zmm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x48,0x27,0xc0] |
| ; KNL-NEXT: kmovw %k0, %edx ## encoding: [0xc5,0xf8,0x93,0xd0] |
| ; KNL-NEXT: vpcmpeqw %ymm7, %ymm3, %ymm0 ## encoding: [0xc5,0xe5,0x75,0xc7] |
| ; KNL-NEXT: vpmovsxwd %ymm0, %zmm0 ## encoding: [0x62,0xf2,0x7d,0x48,0x23,0xc0] |
| ; KNL-NEXT: vptestmd %zmm0, %zmm0, %k0 ## encoding: [0x62,0xf2,0x7d,0x48,0x27,0xc0] |
| ; KNL-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0] |
| ; KNL-NEXT: shll $16, %eax ## encoding: [0xc1,0xe0,0x10] |
| ; KNL-NEXT: orl %edx, %eax ## encoding: [0x09,0xd0] |
| ; KNL-NEXT: shlq $32, %rax ## encoding: [0x48,0xc1,0xe0,0x20] |
| ; KNL-NEXT: orq %rcx, %rax ## encoding: [0x48,0x09,0xc8] |
| ; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; KNL-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; AVX512BW-LABEL: test12_v64i16: |
| ; AVX512BW: ## %bb.0: |
| ; AVX512BW-NEXT: vpcmpeqw %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x75,0xc2] |
| ; AVX512BW-NEXT: vpcmpeqw %zmm3, %zmm1, %k1 ## encoding: [0x62,0xf1,0x75,0x48,0x75,0xcb] |
| ; AVX512BW-NEXT: kunpckdq %k0, %k1, %k0 ## encoding: [0xc4,0xe1,0xf4,0x4b,0xc0] |
| ; AVX512BW-NEXT: kmovq %k0, %rax ## encoding: [0xc4,0xe1,0xfb,0x93,0xc0] |
| ; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; AVX512BW-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test12_v64i16: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vpcmpeqw %zmm2, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x75,0xc2] |
| ; SKX-NEXT: vpcmpeqw %zmm3, %zmm1, %k1 ## encoding: [0x62,0xf1,0x75,0x48,0x75,0xcb] |
| ; SKX-NEXT: kunpckdq %k0, %k1, %k0 ## encoding: [0xc4,0xe1,0xf4,0x4b,0xc0] |
| ; SKX-NEXT: kmovq %k0, %rax ## encoding: [0xc4,0xe1,0xfb,0x93,0xc0] |
| ; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %res = icmp eq <64 x i16> %a, %b |
| %res1 = bitcast <64 x i1> %res to i64 |
| ret i64 %res1 |
| } |
| |
| define <16 x i32> @test13(<16 x float>%a, <16 x float>%b) |
| ; AVX512-LABEL: test13: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: vcmpeqps %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xc9,0x00] |
| ; AVX512-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf3,0x7d,0xc9,0x25,0xc0,0xff] |
| ; AVX512-NEXT: vpsrld $31, %zmm0, %zmm0 ## encoding: [0x62,0xf1,0x7d,0x48,0x72,0xd0,0x1f] |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test13: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vcmpeqps %zmm1, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xc1,0x00] |
| ; SKX-NEXT: vpmovm2d %k0, %zmm0 ## encoding: [0x62,0xf2,0x7e,0x48,0x38,0xc0] |
| ; SKX-NEXT: vpsrld $31, %zmm0, %zmm0 ## encoding: [0x62,0xf1,0x7d,0x48,0x72,0xd0,0x1f] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| { |
| %cmpvector_i = fcmp oeq <16 x float> %a, %b |
| %conv = zext <16 x i1> %cmpvector_i to <16 x i32> |
| ret <16 x i32> %conv |
| } |
| |
| define <16 x i32> @test14(<16 x i32>%a, <16 x i32>%b) { |
| ; CHECK-LABEL: test14: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpsubd %zmm1, %zmm0, %zmm2 ## encoding: [0x62,0xf1,0x7d,0x48,0xfa,0xd1] |
| ; CHECK-NEXT: vpcmpgtd %zmm0, %zmm2, %k1 ## encoding: [0x62,0xf1,0x6d,0x48,0x66,0xc8] |
| ; CHECK-NEXT: vpsubd %zmm1, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xc9,0xfa,0xc1] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %sub_r = sub <16 x i32> %a, %b |
| %cmp.i2.i = icmp sgt <16 x i32> %sub_r, %a |
| %sext.i3.i = sext <16 x i1> %cmp.i2.i to <16 x i32> |
| %mask = icmp eq <16 x i32> %sext.i3.i, zeroinitializer |
| %res = select <16 x i1> %mask, <16 x i32> zeroinitializer, <16 x i32> %sub_r |
| ret <16 x i32>%res |
| } |
| |
| define <8 x i64> @test15(<8 x i64>%a, <8 x i64>%b) { |
| ; CHECK-LABEL: test15: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpsubq %zmm1, %zmm0, %zmm2 ## encoding: [0x62,0xf1,0xfd,0x48,0xfb,0xd1] |
| ; CHECK-NEXT: vpcmpgtq %zmm0, %zmm2, %k1 ## encoding: [0x62,0xf2,0xed,0x48,0x37,0xc8] |
| ; CHECK-NEXT: vpsubq %zmm1, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0xfd,0xc9,0xfb,0xc1] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %sub_r = sub <8 x i64> %a, %b |
| %cmp.i2.i = icmp sgt <8 x i64> %sub_r, %a |
| %sext.i3.i = sext <8 x i1> %cmp.i2.i to <8 x i64> |
| %mask = icmp eq <8 x i64> %sext.i3.i, zeroinitializer |
| %res = select <8 x i1> %mask, <8 x i64> zeroinitializer, <8 x i64> %sub_r |
| ret <8 x i64>%res |
| } |
| |
| define <16 x i32> @test16(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1) nounwind { |
| ; CHECK-LABEL: test16: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpnltd %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf3,0x7d,0x48,0x1f,0xc9,0x05] |
| ; CHECK-NEXT: vpblendmd %zmm2, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc2] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %mask = icmp sge <16 x i32> %x, %y |
| %max = select <16 x i1> %mask, <16 x i32> %x1, <16 x i32> %y |
| ret <16 x i32> %max |
| } |
| |
| define <16 x i32> @test17(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nounwind { |
| ; CHECK-LABEL: test17: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpgtd (%rdi), %zmm0, %k1 ## encoding: [0x62,0xf1,0x7d,0x48,0x66,0x0f] |
| ; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4 |
| %mask = icmp sgt <16 x i32> %x, %y |
| %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 |
| ret <16 x i32> %max |
| } |
| |
| define <16 x i32> @test18(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nounwind { |
| ; CHECK-LABEL: test18: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpled (%rdi), %zmm0, %k1 ## encoding: [0x62,0xf3,0x7d,0x48,0x1f,0x0f,0x02] |
| ; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4 |
| %mask = icmp sle <16 x i32> %x, %y |
| %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 |
| ret <16 x i32> %max |
| } |
| |
| define <16 x i32> @test19(<16 x i32> %x, <16 x i32> %x1, <16 x i32>* %y.ptr) nounwind { |
| ; CHECK-LABEL: test19: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpleud (%rdi), %zmm0, %k1 ## encoding: [0x62,0xf3,0x7d,0x48,0x1e,0x0f,0x02] |
| ; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4 |
| %mask = icmp ule <16 x i32> %x, %y |
| %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 |
| ret <16 x i32> %max |
| } |
| |
| define <16 x i32> @test20(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1, <16 x i32> %y1) nounwind { |
| ; CHECK-LABEL: test20: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpeqd %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0xc9] |
| ; CHECK-NEXT: vpcmpeqd %zmm3, %zmm2, %k1 {%k1} ## encoding: [0x62,0xf1,0x6d,0x49,0x76,0xcb] |
| ; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %mask1 = icmp eq <16 x i32> %x1, %y1 |
| %mask0 = icmp eq <16 x i32> %x, %y |
| %mask = select <16 x i1> %mask0, <16 x i1> %mask1, <16 x i1> zeroinitializer |
| %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %y |
| ret <16 x i32> %max |
| } |
| |
| define <8 x i64> @test21(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1, <8 x i64> %y1) nounwind { |
| ; CHECK-LABEL: test21: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpleq %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf3,0xfd,0x48,0x1f,0xc9,0x02] |
| ; CHECK-NEXT: vpcmpnltq %zmm3, %zmm2, %k1 {%k1} ## encoding: [0x62,0xf3,0xed,0x49,0x1f,0xcb,0x05] |
| ; CHECK-NEXT: vpblendmq %zmm0, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x49,0x64,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %mask1 = icmp sge <8 x i64> %x1, %y1 |
| %mask0 = icmp sle <8 x i64> %x, %y |
| %mask = select <8 x i1> %mask0, <8 x i1> %mask1, <8 x i1> zeroinitializer |
| %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %x1 |
| ret <8 x i64> %max |
| } |
| |
| define <8 x i64> @test22(<8 x i64> %x, <8 x i64>* %y.ptr, <8 x i64> %x1, <8 x i64> %y1) nounwind { |
| ; CHECK-LABEL: test22: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpgtq %zmm2, %zmm1, %k1 ## encoding: [0x62,0xf2,0xf5,0x48,0x37,0xca] |
| ; CHECK-NEXT: vpcmpgtq (%rdi), %zmm0, %k1 {%k1} ## encoding: [0x62,0xf2,0xfd,0x49,0x37,0x0f] |
| ; CHECK-NEXT: vpblendmq %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x64,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %mask1 = icmp sgt <8 x i64> %x1, %y1 |
| %y = load <8 x i64>, <8 x i64>* %y.ptr, align 4 |
| %mask0 = icmp sgt <8 x i64> %x, %y |
| %mask = select <8 x i1> %mask0, <8 x i1> %mask1, <8 x i1> zeroinitializer |
| %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %x1 |
| ret <8 x i64> %max |
| } |
| |
| define <16 x i32> @test23(<16 x i32> %x, <16 x i32>* %y.ptr, <16 x i32> %x1, <16 x i32> %y1) nounwind { |
| ; CHECK-LABEL: test23: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpnltd %zmm2, %zmm1, %k1 ## encoding: [0x62,0xf3,0x75,0x48,0x1f,0xca,0x05] |
| ; CHECK-NEXT: vpcmpleud (%rdi), %zmm0, %k1 {%k1} ## encoding: [0x62,0xf3,0x7d,0x49,0x1e,0x0f,0x02] |
| ; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %mask1 = icmp sge <16 x i32> %x1, %y1 |
| %y = load <16 x i32>, <16 x i32>* %y.ptr, align 4 |
| %mask0 = icmp ule <16 x i32> %x, %y |
| %mask = select <16 x i1> %mask0, <16 x i1> %mask1, <16 x i1> zeroinitializer |
| %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 |
| ret <16 x i32> %max |
| } |
| |
| define <8 x i64> @test24(<8 x i64> %x, <8 x i64> %x1, i64* %yb.ptr) nounwind { |
| ; CHECK-LABEL: test24: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpeqq (%rdi){1to8}, %zmm0, %k1 ## encoding: [0x62,0xf2,0xfd,0x58,0x29,0x0f] |
| ; CHECK-NEXT: vpblendmq %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x64,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %yb = load i64, i64* %yb.ptr, align 4 |
| %y.0 = insertelement <8 x i64> undef, i64 %yb, i32 0 |
| %y = shufflevector <8 x i64> %y.0, <8 x i64> undef, <8 x i32> zeroinitializer |
| %mask = icmp eq <8 x i64> %x, %y |
| %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %x1 |
| ret <8 x i64> %max |
| } |
| |
| define <16 x i32> @test25(<16 x i32> %x, i32* %yb.ptr, <16 x i32> %x1) nounwind { |
| ; CHECK-LABEL: test25: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpled (%rdi){1to16}, %zmm0, %k1 ## encoding: [0x62,0xf3,0x7d,0x58,0x1f,0x0f,0x02] |
| ; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %yb = load i32, i32* %yb.ptr, align 4 |
| %y.0 = insertelement <16 x i32> undef, i32 %yb, i32 0 |
| %y = shufflevector <16 x i32> %y.0, <16 x i32> undef, <16 x i32> zeroinitializer |
| %mask = icmp sle <16 x i32> %x, %y |
| %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 |
| ret <16 x i32> %max |
| } |
| |
| define <16 x i32> @test26(<16 x i32> %x, i32* %yb.ptr, <16 x i32> %x1, <16 x i32> %y1) nounwind { |
| ; CHECK-LABEL: test26: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpnltd %zmm2, %zmm1, %k1 ## encoding: [0x62,0xf3,0x75,0x48,0x1f,0xca,0x05] |
| ; CHECK-NEXT: vpcmpgtd (%rdi){1to16}, %zmm0, %k1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x59,0x66,0x0f] |
| ; CHECK-NEXT: vpblendmd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x64,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %mask1 = icmp sge <16 x i32> %x1, %y1 |
| %yb = load i32, i32* %yb.ptr, align 4 |
| %y.0 = insertelement <16 x i32> undef, i32 %yb, i32 0 |
| %y = shufflevector <16 x i32> %y.0, <16 x i32> undef, <16 x i32> zeroinitializer |
| %mask0 = icmp sgt <16 x i32> %x, %y |
| %mask = select <16 x i1> %mask0, <16 x i1> %mask1, <16 x i1> zeroinitializer |
| %max = select <16 x i1> %mask, <16 x i32> %x, <16 x i32> %x1 |
| ret <16 x i32> %max |
| } |
| |
| define <8 x i64> @test27(<8 x i64> %x, i64* %yb.ptr, <8 x i64> %x1, <8 x i64> %y1) nounwind { |
| ; CHECK-LABEL: test27: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vpcmpnltq %zmm2, %zmm1, %k1 ## encoding: [0x62,0xf3,0xf5,0x48,0x1f,0xca,0x05] |
| ; CHECK-NEXT: vpcmpleq (%rdi){1to8}, %zmm0, %k1 {%k1} ## encoding: [0x62,0xf3,0xfd,0x59,0x1f,0x0f,0x02] |
| ; CHECK-NEXT: vpblendmq %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x64,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %mask1 = icmp sge <8 x i64> %x1, %y1 |
| %yb = load i64, i64* %yb.ptr, align 4 |
| %y.0 = insertelement <8 x i64> undef, i64 %yb, i32 0 |
| %y = shufflevector <8 x i64> %y.0, <8 x i64> undef, <8 x i32> zeroinitializer |
| %mask0 = icmp sle <8 x i64> %x, %y |
| %mask = select <8 x i1> %mask0, <8 x i1> %mask1, <8 x i1> zeroinitializer |
| %max = select <8 x i1> %mask, <8 x i64> %x, <8 x i64> %x1 |
| ret <8 x i64> %max |
| } |
| |
| define <8 x i32>@test28(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1, <8 x i64> %y1) { |
| ; AVX512-LABEL: test28: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: vpcmpgtq %zmm1, %zmm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x48,0x37,0xc1] |
| ; AVX512-NEXT: vpcmpgtq %zmm3, %zmm2, %k1 ## encoding: [0x62,0xf2,0xed,0x48,0x37,0xcb] |
| ; AVX512-NEXT: kxnorw %k1, %k0, %k1 ## encoding: [0xc5,0xfc,0x46,0xc9] |
| ; AVX512-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf3,0x7d,0xc9,0x25,0xc0,0xff] |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test28: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vpcmpgtq %zmm1, %zmm0, %k0 ## encoding: [0x62,0xf2,0xfd,0x48,0x37,0xc1] |
| ; SKX-NEXT: vpcmpgtq %zmm3, %zmm2, %k1 ## encoding: [0x62,0xf2,0xed,0x48,0x37,0xcb] |
| ; SKX-NEXT: kxnorb %k1, %k0, %k0 ## encoding: [0xc5,0xfd,0x46,0xc1] |
| ; SKX-NEXT: vpmovm2d %k0, %ymm0 ## encoding: [0x62,0xf2,0x7e,0x28,0x38,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %x_gt_y = icmp sgt <8 x i64> %x, %y |
| %x1_gt_y1 = icmp sgt <8 x i64> %x1, %y1 |
| %res = icmp eq <8 x i1>%x_gt_y, %x1_gt_y1 |
| %resse = sext <8 x i1>%res to <8 x i32> |
| ret <8 x i32> %resse |
| } |
| |
| define <16 x i8>@test29(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1, <16 x i32> %y1) { |
| ; KNL-LABEL: test29: |
| ; KNL: ## %bb.0: |
| ; KNL-NEXT: vpcmpgtd %zmm1, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x66,0xc1] |
| ; KNL-NEXT: vpcmpgtd %zmm3, %zmm2, %k1 ## encoding: [0x62,0xf1,0x6d,0x48,0x66,0xcb] |
| ; KNL-NEXT: kxorw %k1, %k0, %k1 ## encoding: [0xc5,0xfc,0x47,0xc9] |
| ; KNL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf3,0x7d,0xc9,0x25,0xc0,0xff] |
| ; KNL-NEXT: vpmovdb %zmm0, %xmm0 ## encoding: [0x62,0xf2,0x7e,0x48,0x31,0xc0] |
| ; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; KNL-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; AVX512BW-LABEL: test29: |
| ; AVX512BW: ## %bb.0: |
| ; AVX512BW-NEXT: vpcmpgtd %zmm1, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x66,0xc1] |
| ; AVX512BW-NEXT: vpcmpgtd %zmm3, %zmm2, %k1 ## encoding: [0x62,0xf1,0x6d,0x48,0x66,0xcb] |
| ; AVX512BW-NEXT: kxorw %k1, %k0, %k0 ## encoding: [0xc5,0xfc,0x47,0xc1] |
| ; AVX512BW-NEXT: vpmovm2b %k0, %zmm0 ## encoding: [0x62,0xf2,0x7e,0x48,0x28,0xc0] |
| ; AVX512BW-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 |
| ; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; AVX512BW-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test29: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vpcmpgtd %zmm1, %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x66,0xc1] |
| ; SKX-NEXT: vpcmpgtd %zmm3, %zmm2, %k1 ## encoding: [0x62,0xf1,0x6d,0x48,0x66,0xcb] |
| ; SKX-NEXT: kxorw %k1, %k0, %k0 ## encoding: [0xc5,0xfc,0x47,0xc1] |
| ; SKX-NEXT: vpmovm2b %k0, %xmm0 ## encoding: [0x62,0xf2,0x7e,0x08,0x28,0xc0] |
| ; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %x_gt_y = icmp sgt <16 x i32> %x, %y |
| %x1_gt_y1 = icmp sgt <16 x i32> %x1, %y1 |
| %res = icmp ne <16 x i1>%x_gt_y, %x1_gt_y1 |
| %resse = sext <16 x i1>%res to <16 x i8> |
| ret <16 x i8> %resse |
| } |
| |
| define <4 x double> @test30(<4 x double> %x, <4 x double> %y) nounwind { |
| ; AVX512-LABEL: test30: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 |
| ; AVX512-NEXT: vcmpeqpd %zmm1, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0xc9,0x00] |
| ; AVX512-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test30: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vcmpeqpd %ymm1, %ymm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x28,0xc2,0xc9,0x00] |
| ; SKX-NEXT: vblendmpd %ymm0, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0x65,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| |
| %mask = fcmp oeq <4 x double> %x, %y |
| %max = select <4 x i1> %mask, <4 x double> %x, <4 x double> %y |
| ret <4 x double> %max |
| } |
| |
| define <2 x double> @test31(<2 x double> %x, <2 x double> %x1, <2 x double>* %yp) nounwind { |
| ; AVX512-LABEL: test31: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 |
| ; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 |
| ; AVX512-NEXT: vmovupd (%rdi), %xmm2 ## encoding: [0xc5,0xf9,0x10,0x17] |
| ; AVX512-NEXT: vcmpltpd %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0xca,0x01] |
| ; AVX512-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] |
| ; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 |
| ; AVX512-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test31: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vcmpltpd (%rdi), %xmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x08,0xc2,0x0f,0x01] |
| ; SKX-NEXT: vblendmpd %xmm0, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0x65,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| |
| %y = load <2 x double>, <2 x double>* %yp, align 4 |
| %mask = fcmp olt <2 x double> %x, %y |
| %max = select <2 x i1> %mask, <2 x double> %x, <2 x double> %x1 |
| ret <2 x double> %max |
| } |
| |
| define <4 x double> @test32(<4 x double> %x, <4 x double> %x1, <4 x double>* %yp) nounwind { |
| ; AVX512-LABEL: test32: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 |
| ; AVX512-NEXT: vmovupd (%rdi), %ymm2 ## encoding: [0xc5,0xfd,0x10,0x17] |
| ; AVX512-NEXT: vcmpltpd %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0xca,0x01] |
| ; AVX512-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test32: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vcmpltpd (%rdi), %ymm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x28,0xc2,0x0f,0x01] |
| ; SKX-NEXT: vblendmpd %ymm0, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0x65,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| |
| %y = load <4 x double>, <4 x double>* %yp, align 4 |
| %mask = fcmp ogt <4 x double> %y, %x |
| %max = select <4 x i1> %mask, <4 x double> %x, <4 x double> %x1 |
| ret <4 x double> %max |
| } |
| |
| define <8 x double> @test33(<8 x double> %x, <8 x double> %x1, <8 x double>* %yp) nounwind { |
| ; CHECK-LABEL: test33: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vcmpltpd (%rdi), %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0x0f,0x01] |
| ; CHECK-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %y = load <8 x double>, <8 x double>* %yp, align 4 |
| %mask = fcmp olt <8 x double> %x, %y |
| %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %x1 |
| ret <8 x double> %max |
| } |
| |
| define <4 x float> @test34(<4 x float> %x, <4 x float> %x1, <4 x float>* %yp) nounwind { |
| ; AVX512-LABEL: test34: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 |
| ; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 |
| ; AVX512-NEXT: vmovups (%rdi), %xmm2 ## encoding: [0xc5,0xf8,0x10,0x17] |
| ; AVX512-NEXT: vcmpltps %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xca,0x01] |
| ; AVX512-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] |
| ; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 |
| ; AVX512-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test34: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vcmpltps (%rdi), %xmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x08,0xc2,0x0f,0x01] |
| ; SKX-NEXT: vblendmps %xmm0, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0x65,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %y = load <4 x float>, <4 x float>* %yp, align 4 |
| %mask = fcmp olt <4 x float> %x, %y |
| %max = select <4 x i1> %mask, <4 x float> %x, <4 x float> %x1 |
| ret <4 x float> %max |
| } |
| |
| define <8 x float> @test35(<8 x float> %x, <8 x float> %x1, <8 x float>* %yp) nounwind { |
| ; AVX512-LABEL: test35: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 |
| ; AVX512-NEXT: vmovups (%rdi), %ymm2 ## encoding: [0xc5,0xfc,0x10,0x17] |
| ; AVX512-NEXT: vcmpltps %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xca,0x01] |
| ; AVX512-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test35: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vcmpltps (%rdi), %ymm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x28,0xc2,0x0f,0x01] |
| ; SKX-NEXT: vblendmps %ymm0, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0x65,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| |
| %y = load <8 x float>, <8 x float>* %yp, align 4 |
| %mask = fcmp ogt <8 x float> %y, %x |
| %max = select <8 x i1> %mask, <8 x float> %x, <8 x float> %x1 |
| ret <8 x float> %max |
| } |
| |
| define <16 x float> @test36(<16 x float> %x, <16 x float> %x1, <16 x float>* %yp) nounwind { |
| ; CHECK-LABEL: test36: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vcmpltps (%rdi), %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0x0f,0x01] |
| ; CHECK-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| %y = load <16 x float>, <16 x float>* %yp, align 4 |
| %mask = fcmp olt <16 x float> %x, %y |
| %max = select <16 x i1> %mask, <16 x float> %x, <16 x float> %x1 |
| ret <16 x float> %max |
| } |
| |
| define <8 x double> @test37(<8 x double> %x, <8 x double> %x1, double* %ptr) nounwind { |
| ; CHECK-LABEL: test37: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vcmpltpd (%rdi){1to8}, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x58,0xc2,0x0f,0x01] |
| ; CHECK-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| |
| %a = load double, double* %ptr |
| %v = insertelement <8 x double> undef, double %a, i32 0 |
| %shuffle = shufflevector <8 x double> %v, <8 x double> undef, <8 x i32> zeroinitializer |
| |
| %mask = fcmp ogt <8 x double> %shuffle, %x |
| %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %x1 |
| ret <8 x double> %max |
| } |
| |
| define <4 x double> @test38(<4 x double> %x, <4 x double> %x1, double* %ptr) nounwind { |
| ; AVX512-LABEL: test38: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 |
| ; AVX512-NEXT: vbroadcastsd (%rdi), %ymm2 ## encoding: [0xc4,0xe2,0x7d,0x19,0x17] |
| ; AVX512-NEXT: vcmpltpd %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0xca,0x01] |
| ; AVX512-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test38: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vcmpltpd (%rdi){1to4}, %ymm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x38,0xc2,0x0f,0x01] |
| ; SKX-NEXT: vblendmpd %ymm0, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0x65,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| |
| %a = load double, double* %ptr |
| %v = insertelement <4 x double> undef, double %a, i32 0 |
| %shuffle = shufflevector <4 x double> %v, <4 x double> undef, <4 x i32> zeroinitializer |
| |
| %mask = fcmp ogt <4 x double> %shuffle, %x |
| %max = select <4 x i1> %mask, <4 x double> %x, <4 x double> %x1 |
| ret <4 x double> %max |
| } |
| |
| define <2 x double> @test39(<2 x double> %x, <2 x double> %x1, double* %ptr) nounwind { |
| ; AVX512-LABEL: test39: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 |
| ; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 |
| ; AVX512-NEXT: vmovddup (%rdi), %xmm2 ## encoding: [0xc5,0xfb,0x12,0x17] |
| ; AVX512-NEXT: ## xmm2 = mem[0,0] |
| ; AVX512-NEXT: vcmpltpd %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x48,0xc2,0xca,0x01] |
| ; AVX512-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] |
| ; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 |
| ; AVX512-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test39: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vcmpltpd (%rdi){1to2}, %xmm0, %k1 ## encoding: [0x62,0xf1,0xfd,0x18,0xc2,0x0f,0x01] |
| ; SKX-NEXT: vblendmpd %xmm0, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0x65,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| |
| %a = load double, double* %ptr |
| %v = insertelement <2 x double> undef, double %a, i32 0 |
| %shuffle = shufflevector <2 x double> %v, <2 x double> undef, <2 x i32> <i32 0, i32 0> |
| |
| %mask = fcmp ogt <2 x double> %shuffle, %x |
| %max = select <2 x i1> %mask, <2 x double> %x, <2 x double> %x1 |
| ret <2 x double> %max |
| } |
| |
| |
| define <16 x float> @test40(<16 x float> %x, <16 x float> %x1, float* %ptr) nounwind { |
| ; CHECK-LABEL: test40: |
| ; CHECK: ## %bb.0: |
| ; CHECK-NEXT: vcmpltps (%rdi){1to16}, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x58,0xc2,0x0f,0x01] |
| ; CHECK-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] |
| ; CHECK-NEXT: retq ## encoding: [0xc3] |
| |
| %a = load float, float* %ptr |
| %v = insertelement <16 x float> undef, float %a, i32 0 |
| %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> |
| |
| %mask = fcmp ogt <16 x float> %shuffle, %x |
| %max = select <16 x i1> %mask, <16 x float> %x, <16 x float> %x1 |
| ret <16 x float> %max |
| } |
| |
| define <8 x float> @test41(<8 x float> %x, <8 x float> %x1, float* %ptr) nounwind { |
| ; AVX512-LABEL: test41: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 def $zmm0 |
| ; AVX512-NEXT: vbroadcastss (%rdi), %ymm2 ## encoding: [0xc4,0xe2,0x7d,0x18,0x17] |
| ; AVX512-NEXT: vcmpltps %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xca,0x01] |
| ; AVX512-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] |
| ; AVX512-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test41: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vcmpltps (%rdi){1to8}, %ymm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x38,0xc2,0x0f,0x01] |
| ; SKX-NEXT: vblendmps %ymm0, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0x65,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| |
| %a = load float, float* %ptr |
| %v = insertelement <8 x float> undef, float %a, i32 0 |
| %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> |
| |
| %mask = fcmp ogt <8 x float> %shuffle, %x |
| %max = select <8 x i1> %mask, <8 x float> %x, <8 x float> %x1 |
| ret <8 x float> %max |
| } |
| |
| define <4 x float> @test42(<4 x float> %x, <4 x float> %x1, float* %ptr) nounwind { |
| ; AVX512-LABEL: test42: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 |
| ; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 def $zmm0 |
| ; AVX512-NEXT: vbroadcastss (%rdi), %xmm2 ## encoding: [0xc4,0xe2,0x79,0x18,0x17] |
| ; AVX512-NEXT: vcmpltps %zmm2, %zmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x48,0xc2,0xca,0x01] |
| ; AVX512-NEXT: vblendmps %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x49,0x65,0xc0] |
| ; AVX512-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 |
| ; AVX512-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test42: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vcmpltps (%rdi){1to4}, %xmm0, %k1 ## encoding: [0x62,0xf1,0x7c,0x18,0xc2,0x0f,0x01] |
| ; SKX-NEXT: vblendmps %xmm0, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0x65,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| |
| %a = load float, float* %ptr |
| %v = insertelement <4 x float> undef, float %a, i32 0 |
| %shuffle = shufflevector <4 x float> %v, <4 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> |
| |
| %mask = fcmp ogt <4 x float> %shuffle, %x |
| %max = select <4 x i1> %mask, <4 x float> %x, <4 x float> %x1 |
| ret <4 x float> %max |
| } |
| |
| define <8 x double> @test43(<8 x double> %x, <8 x double> %x1, double* %ptr,<8 x i1> %mask_in) nounwind { |
| ; KNL-LABEL: test43: |
| ; KNL: ## %bb.0: |
| ; KNL-NEXT: vpmovsxwq %xmm2, %zmm2 ## encoding: [0x62,0xf2,0x7d,0x48,0x24,0xd2] |
| ; KNL-NEXT: vpsllq $63, %zmm2, %zmm2 ## encoding: [0x62,0xf1,0xed,0x48,0x73,0xf2,0x3f] |
| ; KNL-NEXT: vptestmq %zmm2, %zmm2, %k1 ## encoding: [0x62,0xf2,0xed,0x48,0x27,0xca] |
| ; KNL-NEXT: vcmpltpd (%rdi){1to8}, %zmm0, %k1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x59,0xc2,0x0f,0x01] |
| ; KNL-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] |
| ; KNL-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; AVX512BW-LABEL: test43: |
| ; AVX512BW: ## %bb.0: |
| ; AVX512BW-NEXT: vpsllw $15, %xmm2, %xmm2 ## encoding: [0xc5,0xe9,0x71,0xf2,0x0f] |
| ; AVX512BW-NEXT: vpmovw2m %zmm2, %k1 ## encoding: [0x62,0xf2,0xfe,0x48,0x29,0xca] |
| ; AVX512BW-NEXT: vcmpltpd (%rdi){1to8}, %zmm0, %k1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x59,0xc2,0x0f,0x01] |
| ; AVX512BW-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] |
| ; AVX512BW-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test43: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vpsllw $15, %xmm2, %xmm2 ## EVEX TO VEX Compression encoding: [0xc5,0xe9,0x71,0xf2,0x0f] |
| ; SKX-NEXT: vpmovw2m %xmm2, %k1 ## encoding: [0x62,0xf2,0xfe,0x08,0x29,0xca] |
| ; SKX-NEXT: vcmpltpd (%rdi){1to8}, %zmm0, %k1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x59,0xc2,0x0f,0x01] |
| ; SKX-NEXT: vblendmpd %zmm0, %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0x65,0xc0] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| |
| %a = load double, double* %ptr |
| %v = insertelement <8 x double> undef, double %a, i32 0 |
| %shuffle = shufflevector <8 x double> %v, <8 x double> undef, <8 x i32> zeroinitializer |
| |
| %mask_cmp = fcmp ogt <8 x double> %shuffle, %x |
| %mask = and <8 x i1> %mask_cmp, %mask_in |
| %max = select <8 x i1> %mask, <8 x double> %x, <8 x double> %x1 |
| ret <8 x double> %max |
| } |
| |
| define <4 x i32> @test44(<4 x i16> %x, <4 x i16> %y) #0 { |
| ; AVX512-LABEL: test44: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: vpxor %xmm2, %xmm2, %xmm2 ## encoding: [0xc5,0xe9,0xef,0xd2] |
| ; AVX512-NEXT: vpblendw $170, %xmm2, %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x0e,0xca,0xaa] |
| ; AVX512-NEXT: ## xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] |
| ; AVX512-NEXT: vpblendw $170, %xmm2, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x0e,0xc2,0xaa] |
| ; AVX512-NEXT: ## xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] |
| ; AVX512-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x76,0xc1] |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test44: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vpxor %xmm2, %xmm2, %xmm2 ## EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2] |
| ; SKX-NEXT: vpblendw $170, %xmm2, %xmm1, %xmm1 ## encoding: [0xc4,0xe3,0x71,0x0e,0xca,0xaa] |
| ; SKX-NEXT: ## xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] |
| ; SKX-NEXT: vpblendw $170, %xmm2, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x0e,0xc2,0xaa] |
| ; SKX-NEXT: ## xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] |
| ; SKX-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x76,0xc1] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %mask = icmp eq <4 x i16> %x, %y |
| %1 = sext <4 x i1> %mask to <4 x i32> |
| ret <4 x i32> %1 |
| } |
| |
| define <2 x i64> @test45(<2 x i16> %x, <2 x i16> %y) #0 { |
| ; AVX512-LABEL: test45: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: vpxor %xmm2, %xmm2, %xmm2 ## encoding: [0xc5,0xe9,0xef,0xd2] |
| ; AVX512-NEXT: vpblendw $17, %xmm1, %xmm2, %xmm1 ## encoding: [0xc4,0xe3,0x69,0x0e,0xc9,0x11] |
| ; AVX512-NEXT: ## xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] |
| ; AVX512-NEXT: vpblendw $17, %xmm0, %xmm2, %xmm0 ## encoding: [0xc4,0xe3,0x69,0x0e,0xc0,0x11] |
| ; AVX512-NEXT: ## xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] |
| ; AVX512-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0 ## encoding: [0xc4,0xe2,0x79,0x29,0xc1] |
| ; AVX512-NEXT: vpsrlq $63, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x73,0xd0,0x3f] |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test45: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vpxor %xmm2, %xmm2, %xmm2 ## EVEX TO VEX Compression encoding: [0xc5,0xe9,0xef,0xd2] |
| ; SKX-NEXT: vpblendw $17, %xmm1, %xmm2, %xmm1 ## encoding: [0xc4,0xe3,0x69,0x0e,0xc9,0x11] |
| ; SKX-NEXT: ## xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] |
| ; SKX-NEXT: vpblendw $17, %xmm0, %xmm2, %xmm0 ## encoding: [0xc4,0xe3,0x69,0x0e,0xc0,0x11] |
| ; SKX-NEXT: ## xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] |
| ; SKX-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0 ## encoding: [0xc4,0xe2,0x79,0x29,0xc1] |
| ; SKX-NEXT: vpsrlq $63, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x73,0xd0,0x3f] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %mask = icmp eq <2 x i16> %x, %y |
| %1 = zext <2 x i1> %mask to <2 x i64> |
| ret <2 x i64> %1 |
| } |
| |
| define <2 x i64> @test46(<2 x float> %x, <2 x float> %y) #0 { |
| ; AVX512-LABEL: test46: |
| ; AVX512: ## %bb.0: |
| ; AVX512-NEXT: vcmpeqps %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf8,0xc2,0xc1,0x00] |
| ; AVX512-NEXT: vpmovzxdq %xmm0, %xmm0 ## encoding: [0xc4,0xe2,0x79,0x35,0xc0] |
| ; AVX512-NEXT: ## xmm0 = xmm0[0],zero,xmm0[1],zero |
| ; AVX512-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0xdb,0x05,A,A,A,A] |
| ; AVX512-NEXT: ## fixup A - offset: 4, value: LCPI47_0-4, kind: reloc_riprel_4byte |
| ; AVX512-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test46: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vcmpeqps %xmm1, %xmm0, %k0 ## encoding: [0x62,0xf1,0x7c,0x08,0xc2,0xc1,0x00] |
| ; SKX-NEXT: vpmovm2q %k0, %xmm0 ## encoding: [0x62,0xf2,0xfe,0x08,0x38,0xc0] |
| ; SKX-NEXT: vpsrlq $63, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x73,0xd0,0x3f] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %mask = fcmp oeq <2 x float> %x, %y |
| %1 = zext <2 x i1> %mask to <2 x i64> |
| ret <2 x i64> %1 |
| } |
| |
| define <16 x i8> @test47(<16 x i32> %a, <16 x i8> %b, <16 x i8> %c) { |
| ; KNL-LABEL: test47: |
| ; KNL: ## %bb.0: |
| ; KNL-NEXT: vptestnmd %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0x7e,0x48,0x27,0xc8] |
| ; KNL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf3,0x7d,0xc9,0x25,0xc0,0xff] |
| ; KNL-NEXT: vpmovdb %zmm0, %xmm0 ## encoding: [0x62,0xf2,0x7e,0x48,0x31,0xc0] |
| ; KNL-NEXT: vpblendvb %xmm0, %xmm1, %xmm2, %xmm0 ## encoding: [0xc4,0xe3,0x69,0x4c,0xc1,0x00] |
| ; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; KNL-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; AVX512BW-LABEL: test47: |
| ; AVX512BW: ## %bb.0: |
| ; AVX512BW-NEXT: ## kill: def $xmm2 killed $xmm2 def $zmm2 |
| ; AVX512BW-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 |
| ; AVX512BW-NEXT: vptestnmd %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0x7e,0x48,0x27,0xc8] |
| ; AVX512BW-NEXT: vpblendmb %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x49,0x66,0xc1] |
| ; AVX512BW-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 |
| ; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; AVX512BW-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test47: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vptestnmd %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0x7e,0x48,0x27,0xc8] |
| ; SKX-NEXT: vpblendmb %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x09,0x66,0xc1] |
| ; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %cmp = icmp eq <16 x i32> %a, zeroinitializer |
| %res = select <16 x i1> %cmp, <16 x i8> %b, <16 x i8> %c |
| ret <16 x i8> %res |
| } |
| |
| define <16 x i16> @test48(<16 x i32> %a, <16 x i16> %b, <16 x i16> %c) { |
| ; KNL-LABEL: test48: |
| ; KNL: ## %bb.0: |
| ; KNL-NEXT: vptestnmd %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0x7e,0x48,0x27,0xc8] |
| ; KNL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf3,0x7d,0xc9,0x25,0xc0,0xff] |
| ; KNL-NEXT: vpmovdw %zmm0, %ymm0 ## encoding: [0x62,0xf2,0x7e,0x48,0x33,0xc0] |
| ; KNL-NEXT: vpblendvb %ymm0, %ymm1, %ymm2, %ymm0 ## encoding: [0xc4,0xe3,0x6d,0x4c,0xc1,0x00] |
| ; KNL-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; AVX512BW-LABEL: test48: |
| ; AVX512BW: ## %bb.0: |
| ; AVX512BW-NEXT: ## kill: def $ymm2 killed $ymm2 def $zmm2 |
| ; AVX512BW-NEXT: ## kill: def $ymm1 killed $ymm1 def $zmm1 |
| ; AVX512BW-NEXT: vptestnmd %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0x7e,0x48,0x27,0xc8] |
| ; AVX512BW-NEXT: vpblendmw %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x49,0x66,0xc1] |
| ; AVX512BW-NEXT: ## kill: def $ymm0 killed $ymm0 killed $zmm0 |
| ; AVX512BW-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test48: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vptestnmd %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0x7e,0x48,0x27,0xc8] |
| ; SKX-NEXT: vpblendmw %ymm1, %ymm2, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x29,0x66,0xc1] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %cmp = icmp eq <16 x i32> %a, zeroinitializer |
| %res = select <16 x i1> %cmp, <16 x i16> %b, <16 x i16> %c |
| ret <16 x i16> %res |
| } |
| |
| define <8 x i16> @test49(<8 x i64> %a, <8 x i16> %b, <8 x i16> %c) { |
| ; KNL-LABEL: test49: |
| ; KNL: ## %bb.0: |
| ; KNL-NEXT: vptestnmq %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0xfe,0x48,0x27,0xc8] |
| ; KNL-NEXT: vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf3,0x7d,0xc9,0x25,0xc0,0xff] |
| ; KNL-NEXT: vpmovdw %zmm0, %ymm0 ## encoding: [0x62,0xf2,0x7e,0x48,0x33,0xc0] |
| ; KNL-NEXT: vpblendvb %xmm0, %xmm1, %xmm2, %xmm0 ## encoding: [0xc4,0xe3,0x69,0x4c,0xc1,0x00] |
| ; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; KNL-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; AVX512BW-LABEL: test49: |
| ; AVX512BW: ## %bb.0: |
| ; AVX512BW-NEXT: ## kill: def $xmm2 killed $xmm2 def $zmm2 |
| ; AVX512BW-NEXT: ## kill: def $xmm1 killed $xmm1 def $zmm1 |
| ; AVX512BW-NEXT: vptestnmq %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0xfe,0x48,0x27,0xc8] |
| ; AVX512BW-NEXT: vpblendmw %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x49,0x66,0xc1] |
| ; AVX512BW-NEXT: ## kill: def $xmm0 killed $xmm0 killed $zmm0 |
| ; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; AVX512BW-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: test49: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vptestnmq %zmm0, %zmm0, %k1 ## encoding: [0x62,0xf2,0xfe,0x48,0x27,0xc8] |
| ; SKX-NEXT: vpblendmw %xmm1, %xmm2, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xed,0x09,0x66,0xc1] |
| ; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %cmp = icmp eq <8 x i64> %a, zeroinitializer |
| %res = select <8 x i1> %cmp, <8 x i16> %b, <8 x i16> %c |
| ret <8 x i16> %res |
| } |
| |
| define i16 @pcmpeq_mem_1(<16 x i32> %a, <16 x i32>* %b) { |
| ; KNL-LABEL: pcmpeq_mem_1: |
| ; KNL: ## %bb.0: |
| ; KNL-NEXT: vpcmpeqd (%rdi), %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0x07] |
| ; KNL-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0] |
| ; KNL-NEXT: ## kill: def $ax killed $ax killed $eax |
| ; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; KNL-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; AVX512BW-LABEL: pcmpeq_mem_1: |
| ; AVX512BW: ## %bb.0: |
| ; AVX512BW-NEXT: vpcmpeqd (%rdi), %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0x07] |
| ; AVX512BW-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] |
| ; AVX512BW-NEXT: ## kill: def $ax killed $ax killed $eax |
| ; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; AVX512BW-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: pcmpeq_mem_1: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vpcmpeqd (%rdi), %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0x07] |
| ; SKX-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] |
| ; SKX-NEXT: ## kill: def $ax killed $ax killed $eax |
| ; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %load = load <16 x i32>, <16 x i32>* %b |
| %cmp = icmp eq <16 x i32> %a, %load |
| %cast = bitcast <16 x i1> %cmp to i16 |
| ret i16 %cast |
| } |
| |
| ; Make sure we use the short pcmpeq encoding like the test above when the memoryo |
| ; operand is in the first argument instead of the second. |
| define i16 @pcmpeq_mem_2(<16 x i32> %a, <16 x i32>* %b) { |
| ; KNL-LABEL: pcmpeq_mem_2: |
| ; KNL: ## %bb.0: |
| ; KNL-NEXT: vpcmpeqd (%rdi), %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0x07] |
| ; KNL-NEXT: kmovw %k0, %eax ## encoding: [0xc5,0xf8,0x93,0xc0] |
| ; KNL-NEXT: ## kill: def $ax killed $ax killed $eax |
| ; KNL-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; KNL-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; AVX512BW-LABEL: pcmpeq_mem_2: |
| ; AVX512BW: ## %bb.0: |
| ; AVX512BW-NEXT: vpcmpeqd (%rdi), %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0x07] |
| ; AVX512BW-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] |
| ; AVX512BW-NEXT: ## kill: def $ax killed $ax killed $eax |
| ; AVX512BW-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; AVX512BW-NEXT: retq ## encoding: [0xc3] |
| ; |
| ; SKX-LABEL: pcmpeq_mem_2: |
| ; SKX: ## %bb.0: |
| ; SKX-NEXT: vpcmpeqd (%rdi), %zmm0, %k0 ## encoding: [0x62,0xf1,0x7d,0x48,0x76,0x07] |
| ; SKX-NEXT: kmovd %k0, %eax ## encoding: [0xc5,0xfb,0x93,0xc0] |
| ; SKX-NEXT: ## kill: def $ax killed $ax killed $eax |
| ; SKX-NEXT: vzeroupper ## encoding: [0xc5,0xf8,0x77] |
| ; SKX-NEXT: retq ## encoding: [0xc3] |
| %load = load <16 x i32>, <16 x i32>* %b |
| %cmp = icmp eq <16 x i32> %load, %a |
| %cast = bitcast <16 x i1> %cmp to i16 |
| ret i16 %cast |
| } |
