1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s 3 4define <8 x i32> @cmp00(<8 x float> %a, <8 x float> %b) nounwind { 5; CHECK-LABEL: cmp00: 6; CHECK: # %bb.0: 7; CHECK-NEXT: vcmpltps %ymm1, %ymm0, %ymm0 8; CHECK-NEXT: retq 9 %bincmp = fcmp olt <8 x float> %a, %b 10 %s = sext <8 x i1> %bincmp to <8 x i32> 11 ret <8 x i32> %s 12} 13 14define <4 x i64> @cmp01(<4 x double> %a, <4 x double> %b) nounwind { 15; CHECK-LABEL: cmp01: 16; CHECK: # %bb.0: 17; CHECK-NEXT: vcmpltpd %ymm1, %ymm0, %ymm0 18; CHECK-NEXT: retq 19 %bincmp = fcmp olt <4 x double> %a, %b 20 %s = sext <4 x i1> %bincmp to <4 x i64> 21 ret <4 x i64> %s 22} 23 24declare void @scale() nounwind 25 26define void @render() nounwind { 27; CHECK-LABEL: render: 28; CHECK: # %bb.0: # %entry 29; CHECK-NEXT: pushq %rbp 30; CHECK-NEXT: pushq %rbx 31; CHECK-NEXT: pushq %rax 32; CHECK-NEXT: xorl %eax, %eax 33; CHECK-NEXT: testb %al, %al 34; CHECK-NEXT: jne .LBB2_6 35; CHECK-NEXT: # %bb.1: # %for.cond5.preheader 36; CHECK-NEXT: xorl %ebx, %ebx 37; CHECK-NEXT: movb $1, %bpl 38; CHECK-NEXT: jmp .LBB2_2 39; CHECK-NEXT: .p2align 4, 0x90 40; CHECK-NEXT: .LBB2_5: # %if.then 41; CHECK-NEXT: # in Loop: Header=BB2_2 Depth=1 42; CHECK-NEXT: callq scale 43; CHECK-NEXT: .LBB2_2: # %for.cond5 44; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 45; CHECK-NEXT: testb %bl, %bl 46; CHECK-NEXT: jne .LBB2_2 47; CHECK-NEXT: # %bb.3: # %for.cond5 48; CHECK-NEXT: # in Loop: Header=BB2_2 Depth=1 49; CHECK-NEXT: testb %bpl, %bpl 50; CHECK-NEXT: jne .LBB2_2 51; CHECK-NEXT: # %bb.4: # %for.body33 52; CHECK-NEXT: # in Loop: Header=BB2_2 Depth=1 53; CHECK-NEXT: vucomisd {{\.LCPI.*}}, %xmm0 54; CHECK-NEXT: jne .LBB2_5 55; CHECK-NEXT: jp .LBB2_5 56; CHECK-NEXT: jmp .LBB2_2 57; CHECK-NEXT: .LBB2_6: # %for.end52 58; CHECK-NEXT: addq $8, %rsp 59; CHECK-NEXT: popq %rbx 60; CHECK-NEXT: popq %rbp 61; CHECK-NEXT: retq 62entry: 63 br i1 undef, label %for.cond5, label %for.end52 64 65for.cond5: 66 %or.cond = and i1 undef, false 67 br i1 %or.cond, label %for.body33, label %for.cond5 68 69for.cond30: 70 br i1 false, label %for.body33, label %for.cond5 71 72for.body33: 73 %tobool = fcmp une double undef, 0.000000e+00 74 br i1 %tobool, label %if.then, label %for.cond30 75 76if.then: 77 call void @scale() 78 br label %for.cond30 79 80for.end52: 81 ret void 82} 83 84define <8 x i32> @int256_cmp(<8 x i32> %i, <8 x i32> %j) nounwind { 85; CHECK-LABEL: int256_cmp: 86; CHECK: # %bb.0: 87; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm2 88; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm3 89; CHECK-NEXT: vpcmpgtd %xmm2, %xmm3, %xmm2 90; CHECK-NEXT: vpcmpgtd %xmm0, %xmm1, %xmm0 91; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 92; CHECK-NEXT: retq 93 %bincmp = icmp slt <8 x i32> %i, %j 94 %x = sext <8 x i1> %bincmp to <8 x i32> 95 ret <8 x i32> %x 96} 97 98define <4 x i64> @v4i64_cmp(<4 x i64> %i, <4 x i64> %j) nounwind { 99; CHECK-LABEL: v4i64_cmp: 100; CHECK: # %bb.0: 101; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm2 102; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm3 103; CHECK-NEXT: vpcmpgtq %xmm2, %xmm3, %xmm2 104; CHECK-NEXT: vpcmpgtq %xmm0, %xmm1, %xmm0 105; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 106; CHECK-NEXT: retq 107 %bincmp = icmp slt <4 x i64> %i, %j 108 %x = sext <4 x i1> %bincmp to <4 x i64> 109 ret <4 x i64> %x 110} 111 112define <16 x i16> @v16i16_cmp(<16 x i16> %i, <16 x i16> %j) nounwind { 113; CHECK-LABEL: v16i16_cmp: 114; CHECK: # %bb.0: 115; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm2 116; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm3 117; CHECK-NEXT: vpcmpgtw %xmm2, %xmm3, %xmm2 118; CHECK-NEXT: vpcmpgtw %xmm0, %xmm1, %xmm0 119; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 120; CHECK-NEXT: retq 121 %bincmp = icmp slt <16 x i16> %i, %j 122 %x = sext <16 x i1> %bincmp to <16 x i16> 123 ret <16 x i16> %x 124} 125 126define <32 x i8> @v32i8_cmp(<32 x i8> %i, <32 x i8> %j) nounwind { 127; CHECK-LABEL: v32i8_cmp: 128; CHECK: # %bb.0: 129; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm2 130; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm3 131; CHECK-NEXT: vpcmpgtb %xmm2, %xmm3, %xmm2 132; CHECK-NEXT: vpcmpgtb %xmm0, %xmm1, %xmm0 133; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 134; CHECK-NEXT: retq 135 %bincmp = icmp slt <32 x i8> %i, %j 136 %x = sext <32 x i1> %bincmp to <32 x i8> 137 ret <32 x i8> %x 138} 139 140define <8 x i32> @int256_cmpeq(<8 x i32> %i, <8 x i32> %j) nounwind { 141; CHECK-LABEL: int256_cmpeq: 142; CHECK: # %bb.0: 143; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2 144; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3 145; CHECK-NEXT: vpcmpeqd %xmm2, %xmm3, %xmm2 146; CHECK-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0 147; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 148; CHECK-NEXT: retq 149 %bincmp = icmp eq <8 x i32> %i, %j 150 %x = sext <8 x i1> %bincmp to <8 x i32> 151 ret <8 x i32> %x 152} 153 154define <4 x i64> @v4i64_cmpeq(<4 x i64> %i, <4 x i64> %j) nounwind { 155; CHECK-LABEL: v4i64_cmpeq: 156; CHECK: # %bb.0: 157; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2 158; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3 159; CHECK-NEXT: vpcmpeqq %xmm2, %xmm3, %xmm2 160; CHECK-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0 161; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 162; CHECK-NEXT: retq 163 %bincmp = icmp eq <4 x i64> %i, %j 164 %x = sext <4 x i1> %bincmp to <4 x i64> 165 ret <4 x i64> %x 166} 167 168define <16 x i16> @v16i16_cmpeq(<16 x i16> %i, <16 x i16> %j) nounwind { 169; CHECK-LABEL: v16i16_cmpeq: 170; CHECK: # %bb.0: 171; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2 172; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3 173; CHECK-NEXT: vpcmpeqw %xmm2, %xmm3, %xmm2 174; CHECK-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0 175; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 176; CHECK-NEXT: retq 177 %bincmp = icmp eq <16 x i16> %i, %j 178 %x = sext <16 x i1> %bincmp to <16 x i16> 179 ret <16 x i16> %x 180} 181 182define <32 x i8> @v32i8_cmpeq(<32 x i8> %i, <32 x i8> %j) nounwind { 183; CHECK-LABEL: v32i8_cmpeq: 184; CHECK: # %bb.0: 185; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2 186; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3 187; CHECK-NEXT: vpcmpeqb %xmm2, %xmm3, %xmm2 188; CHECK-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0 189; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 190; CHECK-NEXT: retq 191 %bincmp = icmp eq <32 x i8> %i, %j 192 %x = sext <32 x i1> %bincmp to <32 x i8> 193 ret <32 x i8> %x 194} 195 196;; Scalar comparison 197 198define i32 @scalarcmpA() uwtable ssp { 199; CHECK-LABEL: scalarcmpA: 200; CHECK: # %bb.0: 201; CHECK-NEXT: vxorpd %xmm0, %xmm0, %xmm0 202; CHECK-NEXT: vcmpeqsd %xmm0, %xmm0, %xmm0 203; CHECK-NEXT: vmovq %xmm0, %rax 204; CHECK-NEXT: andl $1, %eax 205; CHECK-NEXT: # kill: def $eax killed $eax killed $rax 206; CHECK-NEXT: retq 207 %cmp29 = fcmp oeq double undef, 0.000000e+00 208 %res = zext i1 %cmp29 to i32 209 ret i32 %res 210} 211 212define i32 @scalarcmpB() uwtable ssp { 213; CHECK-LABEL: scalarcmpB: 214; CHECK: # %bb.0: 215; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0 216; CHECK-NEXT: vcmpeqss %xmm0, %xmm0, %xmm0 217; CHECK-NEXT: vmovd %xmm0, %eax 218; CHECK-NEXT: andl $1, %eax 219; CHECK-NEXT: retq 220 %cmp29 = fcmp oeq float undef, 0.000000e+00 221 %res = zext i1 %cmp29 to i32 222 ret i32 %res 223} 224 225