; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+avx | FileCheck %s --check-prefix=X86
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx | FileCheck %s --check-prefix=X64
; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+avx2 | FileCheck %s --check-prefix=X86-AVX2
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx2 | FileCheck %s --check-prefix=X64-AVX2
; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+avx512vl | FileCheck %s --check-prefix=X86-AVX512
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx512vl | FileCheck %s --check-prefix=X64-AVX512
define void @and_masks(ptr %a, ptr %b, ptr %c) nounwind uwtable noinline ssp {
; X86-LABEL: and_masks:
; X86: ## %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NEXT: vmovups (%edx), %ymm0
; X86-NEXT: vmovups (%ecx), %ymm1
; X86-NEXT: vcmpltps %ymm0, %ymm1, %ymm1
; X86-NEXT: vmovups (%eax), %ymm2
; X86-NEXT: vcmpltps %ymm0, %ymm2, %ymm0
; X86-NEXT: vandps %ymm1, %ymm0, %ymm0
; X86-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: vmovaps %ymm0, (%eax)
; X86-NEXT: vzeroupper
; X86-NEXT: retl
;
; X64-LABEL: and_masks:
; X64: ## %bb.0:
; X64-NEXT: vmovups (%rdi), %ymm0
; X64-NEXT: vmovups (%rsi), %ymm1
; X64-NEXT: vcmpltps %ymm0, %ymm1, %ymm1
; X64-NEXT: vmovups (%rdx), %ymm2
; X64-NEXT: vcmpltps %ymm0, %ymm2, %ymm0
; X64-NEXT: vandps %ymm1, %ymm0, %ymm0
; X64-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: vmovaps %ymm0, (%rax)
; X64-NEXT: vzeroupper
; X64-NEXT: retq
;
; X86-AVX2-LABEL: and_masks:
; X86-AVX2: ## %bb.0:
; X86-AVX2-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-AVX2-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-AVX2-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-AVX2-NEXT: vmovups (%edx), %ymm0
; X86-AVX2-NEXT: vmovups (%ecx), %ymm1
; X86-AVX2-NEXT: vcmpltps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vmovups (%eax), %ymm2
; X86-AVX2-NEXT: vcmpltps %ymm0, %ymm2, %ymm0
; X86-AVX2-NEXT: vandps %ymm1, %ymm0, %ymm0
; X86-AVX2-NEXT: vpsrld $31, %ymm0, %ymm0
; X86-AVX2-NEXT: vmovdqa %ymm0, (%eax)
; X86-AVX2-NEXT: vzeroupper
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: and_masks:
; X64-AVX2: ## %bb.0:
; X64-AVX2-NEXT: vmovups (%rdi), %ymm0
; X64-AVX2-NEXT: vmovups (%rsi), %ymm1
; X64-AVX2-NEXT: vcmpltps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vmovups (%rdx), %ymm2
; X64-AVX2-NEXT: vcmpltps %ymm0, %ymm2, %ymm0
; X64-AVX2-NEXT: vandps %ymm1, %ymm0, %ymm0
; X64-AVX2-NEXT: vpsrld $31, %ymm0, %ymm0
; X64-AVX2-NEXT: vmovdqa %ymm0, (%rax)
; X64-AVX2-NEXT: vzeroupper
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: and_masks:
; X86-AVX512: ## %bb.0:
; X86-AVX512-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-AVX512-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-AVX512-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-AVX512-NEXT: vmovups (%edx), %ymm0
; X86-AVX512-NEXT: vcmpgtps (%ecx), %ymm0, %k1
; X86-AVX512-NEXT: vcmpgtps (%eax), %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: vpsrld $31, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa %ymm0, (%eax)
; X86-AVX512-NEXT: vzeroupper
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: and_masks:
; X64-AVX512: ## %bb.0:
; X64-AVX512-NEXT: vmovups (%rdi), %ymm0
; X64-AVX512-NEXT: vcmpgtps (%rdx), %ymm0, %k1
; X64-AVX512-NEXT: vcmpgtps (%rsi), %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: vpsrld $31, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa %ymm0, (%rax)
; X64-AVX512-NEXT: vzeroupper
; X64-AVX512-NEXT: retq
%v0 = load <8 x float>, ptr %a, align 16
%v1 = load <8 x float>, ptr %b, align 16
%m0 = fcmp olt <8 x float> %v1, %v0
%v2 = load <8 x float>, ptr %c, align 16
%m1 = fcmp olt <8 x float> %v2, %v0
%mand = and <8 x i1> %m1, %m0
%r = zext <8 x i1> %mand to <8 x i32>
store <8 x i32> %r, ptr undef, align 32
ret void
}
define void @neg_masks(ptr %a, ptr %b, ptr %c) nounwind uwtable noinline ssp {
; X86-LABEL: neg_masks:
; X86: ## %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: vmovups (%ecx), %ymm0
; X86-NEXT: vcmpnltps (%eax), %ymm0, %ymm0
; X86-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: vmovaps %ymm0, (%eax)
; X86-NEXT: vzeroupper
; X86-NEXT: retl
;
; X64-LABEL: neg_masks:
; X64: ## %bb.0:
; X64-NEXT: vmovups (%rsi), %ymm0
; X64-NEXT: vcmpnltps (%rdi), %ymm0, %ymm0
; X64-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: vmovaps %ymm0, (%rax)
; X64-NEXT: vzeroupper
; X64-NEXT: retq
;
; X86-AVX2-LABEL: neg_masks:
; X86-AVX2: ## %bb.0:
; X86-AVX2-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-AVX2-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-AVX2-NEXT: vmovups (%ecx), %ymm0
; X86-AVX2-NEXT: vcmpnltps (%eax), %ymm0, %ymm0
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [1,1,1,1,1,1,1,1]
; X86-AVX2-NEXT: vandps %ymm1, %ymm0, %ymm0
; X86-AVX2-NEXT: vmovaps %ymm0, (%eax)
; X86-AVX2-NEXT: vzeroupper
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: neg_masks:
; X64-AVX2: ## %bb.0:
; X64-AVX2-NEXT: vmovups (%rsi), %ymm0
; X64-AVX2-NEXT: vcmpnltps (%rdi), %ymm0, %ymm0
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [1,1,1,1,1,1,1,1]
; X64-AVX2-NEXT: vandps %ymm1, %ymm0, %ymm0
; X64-AVX2-NEXT: vmovaps %ymm0, (%rax)
; X64-AVX2-NEXT: vzeroupper
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: neg_masks:
; X86-AVX512: ## %bb.0:
; X86-AVX512-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-AVX512-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-AVX512-NEXT: vmovups (%ecx), %ymm0
; X86-AVX512-NEXT: vcmpnltps (%eax), %ymm0, %k1
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: vpsrld $31, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa %ymm0, (%eax)
; X86-AVX512-NEXT: vzeroupper
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: neg_masks:
; X64-AVX512: ## %bb.0:
; X64-AVX512-NEXT: vmovups (%rsi), %ymm0
; X64-AVX512-NEXT: vcmpnltps (%rdi), %ymm0, %k1
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: vpsrld $31, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa %ymm0, (%rax)
; X64-AVX512-NEXT: vzeroupper
; X64-AVX512-NEXT: retq
%v0 = load <8 x float>, ptr %a, align 16
%v1 = load <8 x float>, ptr %b, align 16
%m0 = fcmp olt <8 x float> %v1, %v0
%mand = xor <8 x i1> %m0, <i1 1, i1 1, i1 1, i1 1, i1 1, i1 1, i1 1, i1 1>
%r = zext <8 x i1> %mand to <8 x i32>
store <8 x i32> %r, ptr undef, align 32
ret void
}
define <8 x i32> @and_mask_constant(<8 x i32> %v0, <8 x i32> %v1) {
; X86-LABEL: and_mask_constant:
; X86: ## %bb.0:
; X86-NEXT: vextractf128 $1, %ymm0, %xmm1
; X86-NEXT: vpxor %xmm2, %xmm2, %xmm2
; X86-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm1
; X86-NEXT: vpcmpeqd %xmm2, %xmm0, %xmm0
; X86-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; X86-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: and_mask_constant:
; X64: ## %bb.0:
; X64-NEXT: vextractf128 $1, %ymm0, %xmm1
; X64-NEXT: vpxor %xmm2, %xmm2, %xmm2
; X64-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm1
; X64-NEXT: vpcmpeqd %xmm2, %xmm0, %xmm0
; X64-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; X64-NEXT: vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: and_mask_constant:
; X86-AVX2: ## %bb.0:
; X86-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; X86-AVX2-NEXT: vpcmpeqd %ymm1, %ymm0, %ymm0
; X86-AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: and_mask_constant:
; X64-AVX2: ## %bb.0:
; X64-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; X64-AVX2-NEXT: vpcmpeqd %ymm1, %ymm0, %ymm0
; X64-AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: and_mask_constant:
; X86-AVX512: ## %bb.0:
; X86-AVX512-NEXT: movb $105, %al
; X86-AVX512-NEXT: kmovw %eax, %k1
; X86-AVX512-NEXT: vptestnmd %ymm0, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: vpsrld $31, %ymm0, %ymm0
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: and_mask_constant:
; X64-AVX512: ## %bb.0:
; X64-AVX512-NEXT: movb $105, %al
; X64-AVX512-NEXT: kmovw %eax, %k1
; X64-AVX512-NEXT: vptestnmd %ymm0, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: vpsrld $31, %ymm0, %ymm0
; X64-AVX512-NEXT: retq
%m = icmp eq <8 x i32> %v0, zeroinitializer
%mand = and <8 x i1> %m, <i1 true, i1 false, i1 false, i1 true, i1 false, i1 true, i1 true, i1 false>
%r = zext <8 x i1> %mand to <8 x i32>
ret <8 x i32> %r
}
define <8 x i32> @two_ands(<8 x float> %x) local_unnamed_addr #0 {
; X86-LABEL: two_ands:
; X86: ## %bb.0: ## %entry
; X86-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: vandps %ymm0, %ymm1, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: two_ands:
; X64: ## %bb.0: ## %entry
; X64-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: vandps %ymm0, %ymm1, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: two_ands:
; X86-AVX2: ## %bb.0: ## %entry
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X86-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm0
; X86-AVX2-NEXT: vandps %ymm0, %ymm1, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: two_ands:
; X64-AVX2: ## %bb.0: ## %entry
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X64-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm0
; X64-AVX2-NEXT: vandps %ymm0, %ymm1, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: two_ands:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: two_ands:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: retq
entry:
%cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
%cmp1 = fcmp olt <8 x float> %x, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
%and4 = and <8 x i1> %cmp, %cmp1
%and = sext <8 x i1> %and4 to <8 x i32>
ret <8 x i32> %and
}
define <8 x i32> @three_ands(<8 x float> %x) {
; X86-LABEL: three_ands:
; X86: ## %bb.0: ## %entry
; X86-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X86-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X86-NEXT: vandps %ymm0, %ymm2, %ymm0
; X86-NEXT: vandps %ymm0, %ymm1, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: three_ands:
; X64: ## %bb.0: ## %entry
; X64-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X64-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X64-NEXT: vandps %ymm0, %ymm2, %ymm0
; X64-NEXT: vandps %ymm0, %ymm1, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: three_ands:
; X86-AVX2: ## %bb.0: ## %entry
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X86-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vandps %ymm2, %ymm1, %ymm1
; X86-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X86-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm0
; X86-AVX2-NEXT: vandps %ymm0, %ymm1, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: three_ands:
; X64-AVX2: ## %bb.0: ## %entry
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X64-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vandps %ymm2, %ymm1, %ymm1
; X64-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X64-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm0
; X64-AVX2-NEXT: vandps %ymm0, %ymm1, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: three_ands:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X86-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: three_ands:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X64-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: retq
entry:
%cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
%cmp1 = fcmp olt <8 x float> %x, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
%and8 = and <8 x i1> %cmp, %cmp1
%cmp3 = fcmp une <8 x float> %x, zeroinitializer
%and59 = and <8 x i1> %and8, %cmp3
%and5 = sext <8 x i1> %and59 to <8 x i32>
ret <8 x i32> %and5
}
define <8 x i32> @four_ands(<8 x float> %x) {
; X86-LABEL: four_ands:
; X86: ## %bb.0: ## %entry
; X86-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X86-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X86-NEXT: vandps %ymm3, %ymm2, %ymm2
; X86-NEXT: vandps %ymm2, %ymm1, %ymm1
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: vandps %ymm0, %ymm1, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: four_ands:
; X64: ## %bb.0: ## %entry
; X64-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X64-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X64-NEXT: vandps %ymm3, %ymm2, %ymm2
; X64-NEXT: vandps %ymm2, %ymm1, %ymm1
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: vandps %ymm0, %ymm1, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: four_ands:
; X86-AVX2: ## %bb.0: ## %entry
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X86-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vandps %ymm2, %ymm1, %ymm1
; X86-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X86-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X86-AVX2-NEXT: vandps %ymm0, %ymm2, %ymm0
; X86-AVX2-NEXT: vandps %ymm0, %ymm1, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: four_ands:
; X64-AVX2: ## %bb.0: ## %entry
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X64-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vandps %ymm2, %ymm1, %ymm1
; X64-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X64-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X64-AVX2-NEXT: vandps %ymm0, %ymm2, %ymm0
; X64-AVX2-NEXT: vandps %ymm0, %ymm1, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: four_ands:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X86-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: four_ands:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X64-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: retq
entry:
%cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
%cmp1 = fcmp olt <8 x float> %x, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
%and12 = and <8 x i1> %cmp, %cmp1
%cmp3 = fcmp une <8 x float> %x, zeroinitializer
%and513 = and <8 x i1> %and12, %cmp3
%cmp6 = fcmp une <8 x float> %x, <float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000>
%and814 = and <8 x i1> %and513, %cmp6
%and8 = sext <8 x i1> %and814 to <8 x i32>
ret <8 x i32> %and8
}
define <8 x i32> @five_ands(<8 x float> %x) {
; X86-LABEL: five_ands:
; X86: ## %bb.0: ## %entry
; X86-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X86-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X86-NEXT: vandps %ymm3, %ymm2, %ymm2
; X86-NEXT: vandps %ymm2, %ymm1, %ymm1
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: vandps %ymm0, %ymm2, %ymm0
; X86-NEXT: vandps %ymm0, %ymm1, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: five_ands:
; X64: ## %bb.0: ## %entry
; X64-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X64-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X64-NEXT: vandps %ymm3, %ymm2, %ymm2
; X64-NEXT: vandps %ymm2, %ymm1, %ymm1
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: vandps %ymm0, %ymm2, %ymm0
; X64-NEXT: vandps %ymm0, %ymm1, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: five_ands:
; X86-AVX2: ## %bb.0: ## %entry
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X86-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vandps %ymm2, %ymm1, %ymm1
; X86-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X86-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X86-AVX2-NEXT: vandps %ymm3, %ymm2, %ymm2
; X86-AVX2-NEXT: vandps %ymm2, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm0
; X86-AVX2-NEXT: vandps %ymm0, %ymm1, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: five_ands:
; X64-AVX2: ## %bb.0: ## %entry
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X64-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vandps %ymm2, %ymm1, %ymm1
; X64-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X64-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X64-AVX2-NEXT: vandps %ymm3, %ymm2, %ymm2
; X64-AVX2-NEXT: vandps %ymm2, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm0
; X64-AVX2-NEXT: vandps %ymm0, %ymm1, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: five_ands:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X86-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: five_ands:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X64-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: retq
entry:
%cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
%cmp1 = fcmp olt <8 x float> %x, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
%and16 = and <8 x i1> %cmp, %cmp1
%cmp3 = fcmp une <8 x float> %x, zeroinitializer
%and517 = and <8 x i1> %and16, %cmp3
%cmp6 = fcmp une <8 x float> %x, <float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000>
%and818 = and <8 x i1> %and517, %cmp6
%cmp9 = fcmp une <8 x float> %x, <float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000>
%and1119 = and <8 x i1> %and818, %cmp9
%and11 = sext <8 x i1> %and1119 to <8 x i32>
ret <8 x i32> %and11
}
define <8 x i32> @two_or(<8 x float> %x) {
; X86-LABEL: two_or:
; X86: ## %bb.0: ## %entry
; X86-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: two_or:
; X64: ## %bb.0: ## %entry
; X64-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: two_or:
; X86-AVX2: ## %bb.0: ## %entry
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X86-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm0
; X86-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: two_or:
; X64-AVX2: ## %bb.0: ## %entry
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X64-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm0
; X64-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: two_or:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k0
; X86-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: korw %k1, %k0, %k1
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: two_or:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k0
; X64-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: korw %k1, %k0, %k1
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: retq
entry:
%cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
%cmp1 = fcmp olt <8 x float> %x, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
%or4 = or <8 x i1> %cmp, %cmp1
%or = sext <8 x i1> %or4 to <8 x i32>
ret <8 x i32> %or
}
define <8 x i32> @three_or(<8 x float> %x) {
; X86-LABEL: three_or:
; X86: ## %bb.0: ## %entry
; X86-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X86-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X86-NEXT: vorps %ymm0, %ymm2, %ymm0
; X86-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: three_or:
; X64: ## %bb.0: ## %entry
; X64-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X64-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X64-NEXT: vorps %ymm0, %ymm2, %ymm0
; X64-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: three_or:
; X86-AVX2: ## %bb.0: ## %entry
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X86-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vorps %ymm2, %ymm1, %ymm1
; X86-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X86-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm0
; X86-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: three_or:
; X64-AVX2: ## %bb.0: ## %entry
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X64-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vorps %ymm2, %ymm1, %ymm1
; X64-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X64-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm0
; X64-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: three_or:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k0
; X86-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: korw %k1, %k0, %k0
; X86-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X86-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X86-AVX512-NEXT: korw %k1, %k0, %k1
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: three_or:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k0
; X64-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: korw %k1, %k0, %k0
; X64-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X64-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X64-AVX512-NEXT: korw %k1, %k0, %k1
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: retq
entry:
%cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
%cmp1 = fcmp olt <8 x float> %x, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
%or8 = or <8 x i1> %cmp, %cmp1
%cmp3 = fcmp une <8 x float> %x, zeroinitializer
%or59 = or <8 x i1> %or8, %cmp3
%or5 = sext <8 x i1> %or59 to <8 x i32>
ret <8 x i32> %or5
}
; Function Attrs: norecurse nounwind readnone ssp uwtable
define <8 x i32> @four_or(<8 x float> %x) {
; X86-LABEL: four_or:
; X86: ## %bb.0: ## %entry
; X86-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X86-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X86-NEXT: vorps %ymm3, %ymm2, %ymm2
; X86-NEXT: vorps %ymm2, %ymm1, %ymm1
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: four_or:
; X64: ## %bb.0: ## %entry
; X64-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X64-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X64-NEXT: vorps %ymm3, %ymm2, %ymm2
; X64-NEXT: vorps %ymm2, %ymm1, %ymm1
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: four_or:
; X86-AVX2: ## %bb.0: ## %entry
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X86-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vorps %ymm2, %ymm1, %ymm1
; X86-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X86-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X86-AVX2-NEXT: vorps %ymm0, %ymm2, %ymm0
; X86-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: four_or:
; X64-AVX2: ## %bb.0: ## %entry
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X64-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vorps %ymm2, %ymm1, %ymm1
; X64-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X64-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X64-AVX2-NEXT: vorps %ymm0, %ymm2, %ymm0
; X64-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: four_or:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k0
; X86-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: korw %k1, %k0, %k0
; X86-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X86-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X86-AVX512-NEXT: korw %k1, %k0, %k0
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: korw %k1, %k0, %k1
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: four_or:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k0
; X64-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: korw %k1, %k0, %k0
; X64-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X64-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X64-AVX512-NEXT: korw %k1, %k0, %k0
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: korw %k1, %k0, %k1
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: retq
entry:
%cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
%cmp1 = fcmp olt <8 x float> %x, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
%or12 = or <8 x i1> %cmp, %cmp1
%cmp3 = fcmp une <8 x float> %x, zeroinitializer
%or513 = or <8 x i1> %or12, %cmp3
%cmp6 = fcmp une <8 x float> %x, <float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000>
%or814 = or <8 x i1> %or513, %cmp6
%or8 = sext <8 x i1> %or814 to <8 x i32>
ret <8 x i32> %or8
}
; Function Attrs: norecurse nounwind readnone ssp uwtable
define <8 x i32> @five_or(<8 x float> %x) {
; X86-LABEL: five_or:
; X86: ## %bb.0: ## %entry
; X86-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X86-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X86-NEXT: vorps %ymm3, %ymm2, %ymm2
; X86-NEXT: vorps %ymm2, %ymm1, %ymm1
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: vorps %ymm0, %ymm2, %ymm0
; X86-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: five_or:
; X64: ## %bb.0: ## %entry
; X64-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X64-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X64-NEXT: vorps %ymm3, %ymm2, %ymm2
; X64-NEXT: vorps %ymm2, %ymm1, %ymm1
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: vorps %ymm0, %ymm2, %ymm0
; X64-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: five_or:
; X86-AVX2: ## %bb.0: ## %entry
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X86-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vorps %ymm2, %ymm1, %ymm1
; X86-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X86-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X86-AVX2-NEXT: vorps %ymm3, %ymm2, %ymm2
; X86-AVX2-NEXT: vorps %ymm2, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm0
; X86-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: five_or:
; X64-AVX2: ## %bb.0: ## %entry
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X64-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vorps %ymm2, %ymm1, %ymm1
; X64-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X64-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X64-AVX2-NEXT: vorps %ymm3, %ymm2, %ymm2
; X64-AVX2-NEXT: vorps %ymm2, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm0
; X64-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: five_or:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k0
; X86-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: korw %k1, %k0, %k0
; X86-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X86-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X86-AVX512-NEXT: korw %k1, %k0, %k0
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: korw %k1, %k0, %k0
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: korw %k1, %k0, %k1
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: five_or:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k0
; X64-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: korw %k1, %k0, %k0
; X64-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X64-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X64-AVX512-NEXT: korw %k1, %k0, %k0
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: korw %k1, %k0, %k0
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: korw %k1, %k0, %k1
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: retq
entry:
%cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
%cmp1 = fcmp olt <8 x float> %x, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
%or16 = or <8 x i1> %cmp, %cmp1
%cmp3 = fcmp une <8 x float> %x, zeroinitializer
%or517 = or <8 x i1> %or16, %cmp3
%cmp6 = fcmp une <8 x float> %x, <float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000>
%or818 = or <8 x i1> %or517, %cmp6
%cmp9 = fcmp une <8 x float> %x, <float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000>
%or1119 = or <8 x i1> %or818, %cmp9
%or11 = sext <8 x i1> %or1119 to <8 x i32>
ret <8 x i32> %or11
}
define <8 x i32> @three_or_and(<8 x float> %x) {
; X86-LABEL: three_or_and:
; X86: ## %bb.0: ## %entry
; X86-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X86-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X86-NEXT: vandps %ymm0, %ymm2, %ymm0
; X86-NEXT: vorps %ymm1, %ymm0, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: three_or_and:
; X64: ## %bb.0: ## %entry
; X64-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X64-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X64-NEXT: vandps %ymm0, %ymm2, %ymm0
; X64-NEXT: vorps %ymm1, %ymm0, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: three_or_and:
; X86-AVX2: ## %bb.0: ## %entry
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X86-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X86-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X86-AVX2-NEXT: vandps %ymm0, %ymm2, %ymm0
; X86-AVX2-NEXT: vorps %ymm1, %ymm0, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: three_or_and:
; X64-AVX2: ## %bb.0: ## %entry
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X64-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X64-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X64-AVX2-NEXT: vandps %ymm0, %ymm2, %ymm0
; X64-AVX2-NEXT: vorps %ymm1, %ymm0, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: three_or_and:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k0
; X86-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X86-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X86-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: korw %k0, %k1, %k1
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: three_or_and:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k0
; X64-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X64-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X64-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: korw %k0, %k1, %k1
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: retq
entry:
%cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
%cmp1 = fcmp olt <8 x float> %x, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
%cmp3 = fcmp une <8 x float> %x, zeroinitializer
%and7 = and <8 x i1> %cmp1, %cmp3
%or8 = or <8 x i1> %and7, %cmp
%or = sext <8 x i1> %or8 to <8 x i32>
ret <8 x i32> %or
}
define <8 x i32> @four_or_and(<8 x float> %x) {
; X86-LABEL: four_or_and:
; X86: ## %bb.0: ## %entry
; X86-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vandps %ymm2, %ymm1, %ymm1
; X86-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X86-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: vandps %ymm0, %ymm2, %ymm0
; X86-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: four_or_and:
; X64: ## %bb.0: ## %entry
; X64-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vandps %ymm2, %ymm1, %ymm1
; X64-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X64-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: vandps %ymm0, %ymm2, %ymm0
; X64-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: four_or_and:
; X86-AVX2: ## %bb.0: ## %entry
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X86-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vandps %ymm2, %ymm1, %ymm1
; X86-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X86-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X86-AVX2-NEXT: vandps %ymm0, %ymm2, %ymm0
; X86-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: four_or_and:
; X64-AVX2: ## %bb.0: ## %entry
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X64-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vandps %ymm2, %ymm1, %ymm1
; X64-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X64-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X64-AVX2-NEXT: vandps %ymm0, %ymm2, %ymm0
; X64-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: four_or_and:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k0 {%k1}
; X86-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X86-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: korw %k1, %k0, %k1
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: four_or_and:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k0 {%k1}
; X64-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X64-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: korw %k1, %k0, %k1
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: retq
entry:
%cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
%cmp1 = fcmp olt <8 x float> %x, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
%and11 = and <8 x i1> %cmp, %cmp1
%cmp3 = fcmp une <8 x float> %x, zeroinitializer
%cmp5 = fcmp une <8 x float> %x, <float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000>
%and712 = and <8 x i1> %cmp3, %cmp5
%or13 = or <8 x i1> %and11, %and712
%or = sext <8 x i1> %or13 to <8 x i32>
ret <8 x i32> %or
}
define <8 x i32> @five_or_and(<8 x float> %x) {
; X86-LABEL: five_or_and:
; X86: ## %bb.0: ## %entry
; X86-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X86-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X86-NEXT: vandps %ymm3, %ymm2, %ymm2
; X86-NEXT: vorps %ymm1, %ymm2, %ymm1
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: vandps %ymm0, %ymm2, %ymm0
; X86-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: five_or_and:
; X64: ## %bb.0: ## %entry
; X64-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X64-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X64-NEXT: vandps %ymm3, %ymm2, %ymm2
; X64-NEXT: vorps %ymm1, %ymm2, %ymm1
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: vandps %ymm0, %ymm2, %ymm0
; X64-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: five_or_and:
; X86-AVX2: ## %bb.0: ## %entry
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X86-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X86-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X86-AVX2-NEXT: vandps %ymm3, %ymm2, %ymm2
; X86-AVX2-NEXT: vorps %ymm1, %ymm2, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X86-AVX2-NEXT: vandps %ymm0, %ymm2, %ymm0
; X86-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: five_or_and:
; X64-AVX2: ## %bb.0: ## %entry
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X64-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X64-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X64-AVX2-NEXT: vandps %ymm3, %ymm2, %ymm2
; X64-AVX2-NEXT: vorps %ymm1, %ymm2, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X64-AVX2-NEXT: vandps %ymm0, %ymm2, %ymm0
; X64-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: five_or_and:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k0
; X86-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X86-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X86-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: korw %k0, %k1, %k0
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: korw %k1, %k0, %k1
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: five_or_and:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k0
; X64-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X64-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X64-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: korw %k0, %k1, %k0
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: korw %k1, %k0, %k1
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: retq
entry:
%cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
%cmp1 = fcmp olt <8 x float> %x, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
%cmp3 = fcmp une <8 x float> %x, zeroinitializer
%and15 = and <8 x i1> %cmp1, %cmp3
%or16 = or <8 x i1> %and15, %cmp
%cmp5 = fcmp une <8 x float> %x, <float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000>
%cmp7 = fcmp une <8 x float> %x, <float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000>
%and917 = and <8 x i1> %cmp5, %cmp7
%or1018 = or <8 x i1> %or16, %and917
%or10 = sext <8 x i1> %or1018 to <8 x i32>
ret <8 x i32> %or10
}
define <8 x i32> @four_or_and_xor(<8 x float> %x) {
; X86-LABEL: four_or_and_xor:
; X86: ## %bb.0: ## %entry
; X86-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vxorps %ymm2, %ymm1, %ymm1
; X86-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X86-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: vandps %ymm0, %ymm2, %ymm0
; X86-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: four_or_and_xor:
; X64: ## %bb.0: ## %entry
; X64-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vxorps %ymm2, %ymm1, %ymm1
; X64-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X64-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: vandps %ymm0, %ymm2, %ymm0
; X64-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: four_or_and_xor:
; X86-AVX2: ## %bb.0: ## %entry
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X86-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vxorps %ymm2, %ymm1, %ymm1
; X86-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X86-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X86-AVX2-NEXT: vandps %ymm0, %ymm2, %ymm0
; X86-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: four_or_and_xor:
; X64-AVX2: ## %bb.0: ## %entry
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X64-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vxorps %ymm2, %ymm1, %ymm1
; X64-AVX2-NEXT: vxorps %xmm2, %xmm2, %xmm2
; X64-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm0
; X64-AVX2-NEXT: vandps %ymm0, %ymm2, %ymm0
; X64-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: four_or_and_xor:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k0
; X86-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: kxorw %k1, %k0, %k0
; X86-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X86-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: korw %k1, %k0, %k1
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: four_or_and_xor:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k0
; X64-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: kxorw %k1, %k0, %k0
; X64-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X64-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: korw %k1, %k0, %k1
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: retq
entry:
%cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
%cmp1 = fcmp olt <8 x float> %x, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
%xor10 = xor <8 x i1> %cmp, %cmp1
%cmp3 = fcmp une <8 x float> %x, zeroinitializer
%cmp5 = fcmp une <8 x float> %x, <float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000>
%and11 = and <8 x i1> %cmp3, %cmp5
%or12 = or <8 x i1> %xor10, %and11
%or = sext <8 x i1> %or12 to <8 x i32>
ret <8 x i32> %or
}
; Function Attrs: norecurse nounwind readnone ssp uwtable
define <8 x i32> @five_or_and_xor(<8 x float> %x) {
; X86-LABEL: five_or_and_xor:
; X86: ## %bb.0: ## %entry
; X86-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X86-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X86-NEXT: vxorps %ymm3, %ymm2, %ymm2
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm3
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: vandps %ymm0, %ymm3, %ymm0
; X86-NEXT: vxorps %ymm0, %ymm2, %ymm0
; X86-NEXT: vorps %ymm1, %ymm0, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: five_or_and_xor:
; X64: ## %bb.0: ## %entry
; X64-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X64-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X64-NEXT: vxorps %ymm3, %ymm2, %ymm2
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm3
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: vandps %ymm0, %ymm3, %ymm0
; X64-NEXT: vxorps %ymm0, %ymm2, %ymm0
; X64-NEXT: vorps %ymm1, %ymm0, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: five_or_and_xor:
; X86-AVX2: ## %bb.0: ## %entry
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X86-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X86-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X86-AVX2-NEXT: vxorps %ymm3, %ymm2, %ymm2
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm4 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm4, %ymm0, %ymm0
; X86-AVX2-NEXT: vandps %ymm0, %ymm3, %ymm0
; X86-AVX2-NEXT: vxorps %ymm0, %ymm2, %ymm0
; X86-AVX2-NEXT: vorps %ymm1, %ymm0, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: five_or_and_xor:
; X64-AVX2: ## %bb.0: ## %entry
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X64-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X64-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X64-AVX2-NEXT: vxorps %ymm3, %ymm2, %ymm2
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm4 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm4, %ymm0, %ymm0
; X64-AVX2-NEXT: vandps %ymm0, %ymm3, %ymm0
; X64-AVX2-NEXT: vxorps %ymm0, %ymm2, %ymm0
; X64-AVX2-NEXT: vorps %ymm1, %ymm0, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: five_or_and_xor:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k0
; X86-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X86-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k2
; X86-AVX512-NEXT: kxorw %k2, %k1, %k1
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k2
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k2 {%k2}
; X86-AVX512-NEXT: kxorw %k2, %k1, %k1
; X86-AVX512-NEXT: korw %k0, %k1, %k1
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: five_or_and_xor:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k0
; X64-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X64-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k2
; X64-AVX512-NEXT: kxorw %k2, %k1, %k1
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k2
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k2 {%k2}
; X64-AVX512-NEXT: kxorw %k2, %k1, %k1
; X64-AVX512-NEXT: korw %k0, %k1, %k1
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: retq
entry:
%cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
%cmp1 = fcmp olt <8 x float> %x, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
%cmp3 = fcmp une <8 x float> %x, zeroinitializer
%xor14 = xor <8 x i1> %cmp1, %cmp3
%cmp5 = fcmp une <8 x float> %x, <float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000>
%cmp7 = fcmp une <8 x float> %x, <float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000>
%and15 = and <8 x i1> %cmp5, %cmp7
%xor916 = xor <8 x i1> %xor14, %and15
%or17 = or <8 x i1> %xor916, %cmp
%or = sext <8 x i1> %or17 to <8 x i32>
ret <8 x i32> %or
}
define <8 x i32> @six_or_and_xor(<8 x float> %x) {
; X86-LABEL: six_or_and_xor:
; X86: ## %bb.0: ## %entry
; X86-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm2
; X86-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X86-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm4
; X86-NEXT: vandps %ymm4, %ymm3, %ymm3
; X86-NEXT: vandps %ymm2, %ymm3, %ymm2
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm3
; X86-NEXT: vxorps %ymm1, %ymm3, %ymm1
; X86-NEXT: vxorps %ymm2, %ymm1, %ymm1
; X86-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
; X86-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-NEXT: retl
;
; X64-LABEL: six_or_and_xor:
; X64: ## %bb.0: ## %entry
; X64-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
; X64-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X64-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm4
; X64-NEXT: vandps %ymm4, %ymm3, %ymm3
; X64-NEXT: vandps %ymm2, %ymm3, %ymm2
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm3
; X64-NEXT: vxorps %ymm1, %ymm3, %ymm1
; X64-NEXT: vxorps %ymm2, %ymm1, %ymm1
; X64-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
; X64-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-NEXT: retq
;
; X86-AVX2-LABEL: six_or_and_xor:
; X86-AVX2: ## %bb.0: ## %entry
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X86-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X86-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X86-AVX2-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X86-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm4 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm4, %ymm0, %ymm4
; X86-AVX2-NEXT: vandps %ymm4, %ymm3, %ymm3
; X86-AVX2-NEXT: vandps %ymm2, %ymm3, %ymm2
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X86-AVX2-NEXT: vxorps %ymm1, %ymm3, %ymm1
; X86-AVX2-NEXT: vxorps %ymm2, %ymm1, %ymm1
; X86-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1]
; X86-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm0
; X86-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X86-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: six_or_and_xor:
; X64-AVX2: ## %bb.0: ## %entry
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1,-5.0E-1]
; X64-AVX2-NEXT: vcmpleps %ymm0, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; X64-AVX2-NEXT: vcmpltps %ymm2, %ymm0, %ymm2
; X64-AVX2-NEXT: vxorps %xmm3, %xmm3, %xmm3
; X64-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm4 = [1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1,1.00000001E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm4, %ymm0, %ymm4
; X64-AVX2-NEXT: vandps %ymm4, %ymm3, %ymm3
; X64-AVX2-NEXT: vandps %ymm2, %ymm3, %ymm2
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm3 = [2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1,2.00000003E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm3, %ymm0, %ymm3
; X64-AVX2-NEXT: vxorps %ymm1, %ymm3, %ymm1
; X64-AVX2-NEXT: vxorps %ymm2, %ymm1, %ymm1
; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1,4.00000006E-1]
; X64-AVX2-NEXT: vcmpneqps %ymm2, %ymm0, %ymm0
; X64-AVX2-NEXT: vorps %ymm0, %ymm1, %ymm0
; X64-AVX2-NEXT: retq
;
; X86-AVX512-LABEL: six_or_and_xor:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k0
; X86-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X86-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1 {%k1}
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k2
; X86-AVX512-NEXT: kxorw %k0, %k2, %k0
; X86-AVX512-NEXT: kxorw %k1, %k0, %k0
; X86-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}{1to8}, %ymm0, %k1
; X86-AVX512-NEXT: korw %k1, %k0, %k1
; X86-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X86-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X86-AVX512-NEXT: retl
;
; X64-AVX512-LABEL: six_or_and_xor:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vcmpgeps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k0
; X64-AVX512-NEXT: vxorps %xmm1, %xmm1, %xmm1
; X64-AVX512-NEXT: vcmpneqps %ymm1, %ymm0, %k1
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: vcmpltps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1 {%k1}
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k2
; X64-AVX512-NEXT: kxorw %k0, %k2, %k0
; X64-AVX512-NEXT: kxorw %k1, %k0, %k0
; X64-AVX512-NEXT: vcmpneqps {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %k1
; X64-AVX512-NEXT: korw %k1, %k0, %k1
; X64-AVX512-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
; X64-AVX512-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; X64-AVX512-NEXT: retq
entry:
%cmp = fcmp oge <8 x float> %x, <float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01, float -5.000000e-01>
%cmp1 = fcmp olt <8 x float> %x, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
%cmp3 = fcmp une <8 x float> %x, zeroinitializer
%and18 = and <8 x i1> %cmp1, %cmp3
%cmp5 = fcmp une <8 x float> %x, <float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000, float 0x3FB99999A0000000>
%and719 = and <8 x i1> %and18, %cmp5
%cmp8 = fcmp une <8 x float> %x, <float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000, float 0x3FC99999A0000000>
%xor20 = xor <8 x i1> %cmp8, %cmp
%xor1021 = xor <8 x i1> %xor20, %and719
%cmp11 = fcmp une <8 x float> %x, <float 0x3FD99999A0000000, float 0x3FD99999A0000000, float 0x3FD99999A0000000, float 0x3FD99999A0000000, float 0x3FD99999A0000000, float 0x3FD99999A0000000, float 0x3FD99999A0000000, float 0x3FD99999A0000000>
%or22 = or <8 x i1> %xor1021, %cmp11
%or = sext <8 x i1> %or22 to <8 x i32>
ret <8 x i32> %or
}
Codebase Browser
llvm