1 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7-avx -mattr=+avx | FileCheck %s 2 3 ; CHECK: vcmpltps %ymm 4 ; CHECK-NOT: vucomiss 5 define <8 x i32> @cmp00(<8 x float> %a, <8 x float> %b) nounwind readnone { 6 %bincmp = fcmp olt <8 x float> %a, %b 7 %s = sext <8 x i1> %bincmp to <8 x i32> 8 ret <8 x i32> %s 9 } 10 11 ; CHECK: vcmpltpd %ymm 12 ; CHECK-NOT: vucomisd 13 define <4 x i64> @cmp01(<4 x double> %a, <4 x double> %b) nounwind readnone { 14 %bincmp = fcmp olt <4 x double> %a, %b 15 %s = sext <4 x i1> %bincmp to <4 x i64> 16 ret <4 x i64> %s 17 } 18 19 declare void @scale() nounwind uwtable 20 21 ; CHECK: vucomisd 22 define void @render() nounwind uwtable { 23 entry: 24 br i1 undef, label %for.cond5, label %for.end52 25 26 for.cond5: 27 %or.cond = and i1 undef, false 28 br i1 %or.cond, label %for.body33, label %for.cond5 29 30 for.cond30: 31 br i1 false, label %for.body33, label %for.cond5 32 33 for.body33: 34 %tobool = fcmp une double undef, 0.000000e+00 35 br i1 %tobool, label %if.then, label %for.cond30 36 37 if.then: 38 call void @scale() 39 br label %for.cond30 40 41 for.end52: 42 ret void 43 } 44 45 ; CHECK: vextractf128 $1 46 ; CHECK: vextractf128 $1 47 ; CHECK-NEXT: vpcmpgtd %xmm 48 ; CHECK-NEXT: vpcmpgtd %xmm 49 ; CHECK-NEXT: vinsertf128 $1 50 define <8 x i32> @int256-cmp(<8 x i32> %i, <8 x i32> %j) nounwind readnone { 51 %bincmp = icmp slt <8 x i32> %i, %j 52 %x = sext <8 x i1> %bincmp to <8 x i32> 53 ret <8 x i32> %x 54 } 55 56 ; CHECK: vextractf128 $1 57 ; CHECK: vextractf128 $1 58 ; CHECK-NEXT: vpcmpgtq %xmm 59 ; CHECK-NEXT: vpcmpgtq %xmm 60 ; CHECK-NEXT: vinsertf128 $1 61 define <4 x i64> @v4i64-cmp(<4 x i64> %i, <4 x i64> %j) nounwind readnone { 62 %bincmp = icmp slt <4 x i64> %i, %j 63 %x = sext <4 x i1> %bincmp to <4 x i64> 64 ret <4 x i64> %x 65 } 66 67 ; CHECK: vextractf128 $1 68 ; CHECK: vextractf128 $1 69 ; CHECK-NEXT: vpcmpgtw %xmm 70 ; CHECK-NEXT: vpcmpgtw %xmm 71 ; CHECK-NEXT: vinsertf128 $1 72 define <16 x i16> @v16i16-cmp(<16 x i16> %i, <16 x i16> %j) nounwind readnone { 73 %bincmp = icmp slt <16 x i16> %i, %j 74 %x = sext <16 x i1> %bincmp to <16 x i16> 75 ret <16 x i16> %x 76 } 77 78 ; CHECK: vextractf128 $1 79 ; CHECK: vextractf128 $1 80 ; CHECK-NEXT: vpcmpgtb %xmm 81 ; CHECK-NEXT: vpcmpgtb %xmm 82 ; CHECK-NEXT: vinsertf128 $1 83 define <32 x i8> @v32i8-cmp(<32 x i8> %i, <32 x i8> %j) nounwind readnone { 84 %bincmp = icmp slt <32 x i8> %i, %j 85 %x = sext <32 x i1> %bincmp to <32 x i8> 86 ret <32 x i8> %x 87 } 88 89 ; CHECK: vextractf128 $1 90 ; CHECK: vextractf128 $1 91 ; CHECK-NEXT: vpcmpeqd %xmm 92 ; CHECK-NEXT: vpcmpeqd %xmm 93 ; CHECK-NEXT: vinsertf128 $1 94 define <8 x i32> @int256-cmpeq(<8 x i32> %i, <8 x i32> %j) nounwind readnone { 95 %bincmp = icmp eq <8 x i32> %i, %j 96 %x = sext <8 x i1> %bincmp to <8 x i32> 97 ret <8 x i32> %x 98 } 99 100 ; CHECK: vextractf128 $1 101 ; CHECK: vextractf128 $1 102 ; CHECK-NEXT: vpcmpeqq %xmm 103 ; CHECK-NEXT: vpcmpeqq %xmm 104 ; CHECK-NEXT: vinsertf128 $1 105 define <4 x i64> @v4i64-cmpeq(<4 x i64> %i, <4 x i64> %j) nounwind readnone { 106 %bincmp = icmp eq <4 x i64> %i, %j 107 %x = sext <4 x i1> %bincmp to <4 x i64> 108 ret <4 x i64> %x 109 } 110 111 ; CHECK: vextractf128 $1 112 ; CHECK: vextractf128 $1 113 ; CHECK-NEXT: vpcmpeqw %xmm 114 ; CHECK-NEXT: vpcmpeqw %xmm 115 ; CHECK-NEXT: vinsertf128 $1 116 define <16 x i16> @v16i16-cmpeq(<16 x i16> %i, <16 x i16> %j) nounwind readnone { 117 %bincmp = icmp eq <16 x i16> %i, %j 118 %x = sext <16 x i1> %bincmp to <16 x i16> 119 ret <16 x i16> %x 120 } 121 122 ; CHECK: vextractf128 $1 123 ; CHECK: vextractf128 $1 124 ; CHECK-NEXT: vpcmpeqb %xmm 125 ; CHECK-NEXT: vpcmpeqb %xmm 126 ; CHECK-NEXT: vinsertf128 $1 127 define <32 x i8> @v32i8-cmpeq(<32 x i8> %i, <32 x i8> %j) nounwind readnone { 128 %bincmp = icmp eq <32 x i8> %i, %j 129 %x = sext <32 x i1> %bincmp to <32 x i8> 130 ret <32 x i8> %x 131 } 132 133 ;; Scalar comparison 134 135 ; CHECK: scalarcmpA 136 ; CHECK: vcmpeqsd 137 define i32 @scalarcmpA() uwtable ssp { 138 %cmp29 = fcmp oeq double undef, 0.000000e+00 139 %res = zext i1 %cmp29 to i32 140 ret i32 %res 141 } 142 143 ; CHECK: scalarcmpB 144 ; CHECK: vcmpeqss 145 define i32 @scalarcmpB() uwtable ssp { 146 %cmp29 = fcmp oeq float undef, 0.000000e+00 147 %res = zext i1 %cmp29 to i32 148 ret i32 %res 149 } 150 151
