1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2 ; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+avx | FileCheck %s --check-prefix=CHECK --check-prefix=X32 3 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx | FileCheck %s --check-prefix=CHECK --check-prefix=X64 4 ; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=X32-AVX2 5 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=X64-AVX2 6 7 define void @and_masks(<8 x float>* %a, <8 x float>* %b, <8 x float>* %c) nounwind uwtable noinline ssp { 8 ; X32-LABEL: and_masks: 9 ; X32: ## %bb.0: 10 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 11 ; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx 12 ; X32-NEXT: movl {{[0-9]+}}(%esp), %edx 13 ; X32-NEXT: vmovups (%edx), %ymm0 14 ; X32-NEXT: vmovups (%ecx), %ymm1 15 ; X32-NEXT: vcmpltps %ymm0, %ymm1, %ymm1 16 ; X32-NEXT: vmovups (%eax), %ymm2 17 ; X32-NEXT: vcmpltps %ymm0, %ymm2, %ymm0 18 ; X32-NEXT: vandps %ymm1, %ymm0, %ymm0 19 ; X32-NEXT: vandps LCPI0_0, %ymm0, %ymm0 20 ; X32-NEXT: vmovaps %ymm0, (%eax) 21 ; X32-NEXT: vzeroupper 22 ; X32-NEXT: retl 23 ; 24 ; X64-LABEL: and_masks: 25 ; X64: ## %bb.0: 26 ; X64-NEXT: vmovups (%rdi), %ymm0 27 ; X64-NEXT: vmovups (%rsi), %ymm1 28 ; X64-NEXT: vcmpltps %ymm0, %ymm1, %ymm1 29 ; X64-NEXT: vmovups (%rdx), %ymm2 30 ; X64-NEXT: vcmpltps %ymm0, %ymm2, %ymm0 31 ; X64-NEXT: vandps %ymm1, %ymm0, %ymm0 32 ; X64-NEXT: vandps {{.*}}(%rip), %ymm0, %ymm0 33 ; X64-NEXT: vmovaps %ymm0, (%rax) 34 ; X64-NEXT: vzeroupper 35 ; X64-NEXT: retq 36 ; 37 ; X32-AVX2-LABEL: and_masks: 38 ; X32-AVX2: ## %bb.0: 39 ; X32-AVX2-NEXT: movl {{[0-9]+}}(%esp), %eax 40 ; X32-AVX2-NEXT: movl {{[0-9]+}}(%esp), %ecx 41 ; X32-AVX2-NEXT: movl {{[0-9]+}}(%esp), %edx 42 ; X32-AVX2-NEXT: vmovups (%edx), %ymm0 43 ; X32-AVX2-NEXT: vmovups (%ecx), %ymm1 44 ; X32-AVX2-NEXT: vcmpltps %ymm0, %ymm1, %ymm1 45 ; X32-AVX2-NEXT: vmovups (%eax), %ymm2 46 ; X32-AVX2-NEXT: vcmpltps %ymm0, %ymm2, %ymm0 47 ; X32-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1,1,1,1,1,1,1,1] 48 ; X32-AVX2-NEXT: vandps %ymm2, %ymm1, %ymm1 49 ; X32-AVX2-NEXT: vandps %ymm1, %ymm0, %ymm0 50 ; X32-AVX2-NEXT: vmovaps %ymm0, (%eax) 51 ; X32-AVX2-NEXT: vzeroupper 52 ; X32-AVX2-NEXT: retl 53 ; 54 ; X64-AVX2-LABEL: and_masks: 55 ; X64-AVX2: ## %bb.0: 56 ; X64-AVX2-NEXT: vmovups (%rdi), %ymm0 57 ; X64-AVX2-NEXT: vmovups (%rsi), %ymm1 58 ; X64-AVX2-NEXT: vcmpltps %ymm0, %ymm1, %ymm1 59 ; X64-AVX2-NEXT: vmovups (%rdx), %ymm2 60 ; X64-AVX2-NEXT: vcmpltps %ymm0, %ymm2, %ymm0 61 ; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm2 = [1,1,1,1,1,1,1,1] 62 ; X64-AVX2-NEXT: vandps %ymm2, %ymm1, %ymm1 63 ; X64-AVX2-NEXT: vandps %ymm1, %ymm0, %ymm0 64 ; X64-AVX2-NEXT: vmovaps %ymm0, (%rax) 65 ; X64-AVX2-NEXT: vzeroupper 66 ; X64-AVX2-NEXT: retq 67 %v0 = load <8 x float>, <8 x float>* %a, align 16 68 %v1 = load <8 x float>, <8 x float>* %b, align 16 69 %m0 = fcmp olt <8 x float> %v1, %v0 70 %v2 = load <8 x float>, <8 x float>* %c, align 16 71 %m1 = fcmp olt <8 x float> %v2, %v0 72 %mand = and <8 x i1> %m1, %m0 73 %r = zext <8 x i1> %mand to <8 x i32> 74 store <8 x i32> %r, <8 x i32>* undef, align 32 75 ret void 76 } 77 78 define void @neg_masks(<8 x float>* %a, <8 x float>* %b, <8 x float>* %c) nounwind uwtable noinline ssp { 79 ; X32-LABEL: neg_masks: 80 ; X32: ## %bb.0: 81 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 82 ; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx 83 ; X32-NEXT: vmovups (%ecx), %ymm0 84 ; X32-NEXT: vcmpnltps (%eax), %ymm0, %ymm0 85 ; X32-NEXT: vandps LCPI1_0, %ymm0, %ymm0 86 ; X32-NEXT: vmovaps %ymm0, (%eax) 87 ; X32-NEXT: vzeroupper 88 ; X32-NEXT: retl 89 ; 90 ; X64-LABEL: neg_masks: 91 ; X64: ## %bb.0: 92 ; X64-NEXT: vmovups (%rsi), %ymm0 93 ; X64-NEXT: vcmpnltps (%rdi), %ymm0, %ymm0 94 ; X64-NEXT: vandps {{.*}}(%rip), %ymm0, %ymm0 95 ; X64-NEXT: vmovaps %ymm0, (%rax) 96 ; X64-NEXT: vzeroupper 97 ; X64-NEXT: retq 98 ; 99 ; X32-AVX2-LABEL: neg_masks: 100 ; X32-AVX2: ## %bb.0: 101 ; X32-AVX2-NEXT: movl {{[0-9]+}}(%esp), %eax 102 ; X32-AVX2-NEXT: movl {{[0-9]+}}(%esp), %ecx 103 ; X32-AVX2-NEXT: vmovups (%ecx), %ymm0 104 ; X32-AVX2-NEXT: vcmpnltps (%eax), %ymm0, %ymm0 105 ; X32-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [1,1,1,1,1,1,1,1] 106 ; X32-AVX2-NEXT: vandps %ymm1, %ymm0, %ymm0 107 ; X32-AVX2-NEXT: vmovaps %ymm0, (%eax) 108 ; X32-AVX2-NEXT: vzeroupper 109 ; X32-AVX2-NEXT: retl 110 ; 111 ; X64-AVX2-LABEL: neg_masks: 112 ; X64-AVX2: ## %bb.0: 113 ; X64-AVX2-NEXT: vmovups (%rsi), %ymm0 114 ; X64-AVX2-NEXT: vcmpnltps (%rdi), %ymm0, %ymm0 115 ; X64-AVX2-NEXT: vbroadcastss {{.*#+}} ymm1 = [1,1,1,1,1,1,1,1] 116 ; X64-AVX2-NEXT: vandps %ymm1, %ymm0, %ymm0 117 ; X64-AVX2-NEXT: vmovaps %ymm0, (%rax) 118 ; X64-AVX2-NEXT: vzeroupper 119 ; X64-AVX2-NEXT: retq 120 %v0 = load <8 x float>, <8 x float>* %a, align 16 121 %v1 = load <8 x float>, <8 x float>* %b, align 16 122 %m0 = fcmp olt <8 x float> %v1, %v0 123 %mand = xor <8 x i1> %m0, <i1 1, i1 1, i1 1, i1 1, i1 1, i1 1, i1 1, i1 1> 124 %r = zext <8 x i1> %mand to <8 x i32> 125 store <8 x i32> %r, <8 x i32>* undef, align 32 126 ret void 127 } 128 129 define <8 x i32> @and_mask_constant(<8 x i32> %v0, <8 x i32> %v1) { 130 ; X32-LABEL: and_mask_constant: 131 ; X32: ## %bb.0: 132 ; X32-NEXT: vextractf128 $1, %ymm0, %xmm1 133 ; X32-NEXT: vpxor %xmm2, %xmm2, %xmm2 134 ; X32-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm1 135 ; X32-NEXT: vpcmpeqd %xmm2, %xmm0, %xmm0 136 ; X32-NEXT: vpackssdw %xmm1, %xmm0, %xmm0 137 ; X32-NEXT: vpand LCPI2_0, %xmm0, %xmm0 138 ; X32-NEXT: vpmovzxwd {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero 139 ; X32-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1] 140 ; X32-NEXT: vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero 141 ; X32-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 142 ; X32-NEXT: retl 143 ; 144 ; X64-LABEL: and_mask_constant: 145 ; X64: ## %bb.0: 146 ; X64-NEXT: vextractf128 $1, %ymm0, %xmm1 147 ; X64-NEXT: vpxor %xmm2, %xmm2, %xmm2 148 ; X64-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm1 149 ; X64-NEXT: vpcmpeqd %xmm2, %xmm0, %xmm0 150 ; X64-NEXT: vpackssdw %xmm1, %xmm0, %xmm0 151 ; X64-NEXT: vpand {{.*}}(%rip), %xmm0, %xmm0 152 ; X64-NEXT: vpmovzxwd {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero 153 ; X64-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1] 154 ; X64-NEXT: vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero 155 ; X64-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 156 ; X64-NEXT: retq 157 ; 158 ; X32-AVX2-LABEL: and_mask_constant: 159 ; X32-AVX2: ## %bb.0: 160 ; X32-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1 161 ; X32-AVX2-NEXT: vpcmpeqd %ymm1, %ymm0, %ymm0 162 ; X32-AVX2-NEXT: vpand LCPI2_0, %ymm0, %ymm0 163 ; X32-AVX2-NEXT: retl 164 ; 165 ; X64-AVX2-LABEL: and_mask_constant: 166 ; X64-AVX2: ## %bb.0: 167 ; X64-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1 168 ; X64-AVX2-NEXT: vpcmpeqd %ymm1, %ymm0, %ymm0 169 ; X64-AVX2-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0 170 ; X64-AVX2-NEXT: retq 171 %m = icmp eq <8 x i32> %v0, zeroinitializer 172 %mand = and <8 x i1> %m, <i1 true, i1 false, i1 false, i1 true, i1 false, i1 true, i1 true, i1 false> 173 %r = zext <8 x i1> %mand to <8 x i32> 174 ret <8 x i32> %r 175 } 176
