1 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s 2 3 define <8 x float> @sitofp00(<8 x i32> %a) nounwind { 4 ; CHECK-LABEL: sitofp00: 5 ; CHECK: # BB#0: 6 ; CHECK-NEXT: vcvtdq2ps %ymm0, %ymm0 7 ; CHECK-NEXT: retq 8 %b = sitofp <8 x i32> %a to <8 x float> 9 ret <8 x float> %b 10 } 11 12 define <8 x i32> @fptosi00(<8 x float> %a) nounwind { 13 ; CHECK-LABEL: fptosi00: 14 ; CHECK: # BB#0: 15 ; CHECK-NEXT: vcvttps2dq %ymm0, %ymm0 16 ; CHECK-NEXT: retq 17 %b = fptosi <8 x float> %a to <8 x i32> 18 ret <8 x i32> %b 19 } 20 21 define <4 x double> @sitofp01(<4 x i32> %a) { 22 ; CHECK-LABEL: sitofp01: 23 ; CHECK: # BB#0: 24 ; CHECK-NEXT: vcvtdq2pd %xmm0, %ymm0 25 ; CHECK-NEXT: retq 26 %b = sitofp <4 x i32> %a to <4 x double> 27 ret <4 x double> %b 28 } 29 30 define <8 x float> @sitofp02(<8 x i16> %a) { 31 ; CHECK-LABEL: sitofp02: 32 ; CHECK: # BB#0: 33 ; CHECK-NEXT: vpmovsxwd %xmm0, %xmm1 34 ; CHECK-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1] 35 ; CHECK-NEXT: vpmovsxwd %xmm0, %xmm0 36 ; CHECK-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 37 ; CHECK-NEXT: vcvtdq2ps %ymm0, %ymm0 38 ; CHECK-NEXT: retq 39 %b = sitofp <8 x i16> %a to <8 x float> 40 ret <8 x float> %b 41 } 42 43 define <4 x i32> @fptosi01(<4 x double> %a) { 44 ; CHECK-LABEL: fptosi01: 45 ; CHECK: # BB#0: 46 ; CHECK-NEXT: vcvttpd2dqy %ymm0, %xmm0 47 ; CHECK-NEXT: vzeroupper 48 ; CHECK-NEXT: retq 49 %b = fptosi <4 x double> %a to <4 x i32> 50 ret <4 x i32> %b 51 } 52 53 define <8 x float> @fptrunc00(<8 x double> %b) nounwind { 54 ; CHECK-LABEL: fptrunc00: 55 ; CHECK: # BB#0: 56 ; CHECK-NEXT: vcvtpd2psy %ymm0, %xmm0 57 ; CHECK-NEXT: vcvtpd2psy %ymm1, %xmm1 58 ; CHECK-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 59 ; CHECK-NEXT: retq 60 %a = fptrunc <8 x double> %b to <8 x float> 61 ret <8 x float> %a 62 } 63 64 define <4 x double> @fpext00(<4 x float> %b) nounwind { 65 ; CHECK-LABEL: fpext00: 66 ; CHECK: # BB#0: 67 ; CHECK-NEXT: vcvtps2pd %xmm0, %ymm0 68 ; CHECK-NEXT: retq 69 %a = fpext <4 x float> %b to <4 x double> 70 ret <4 x double> %a 71 } 72 73 define double @funcA(i64* nocapture %e) nounwind uwtable readonly ssp { 74 ; CHECK-LABEL: funcA: 75 ; CHECK: # BB#0: 76 ; CHECK-NEXT: vcvtsi2sdq (%rdi), %xmm0, %xmm0 77 ; CHECK-NEXT: retq 78 %tmp1 = load i64, i64* %e, align 8 79 %conv = sitofp i64 %tmp1 to double 80 ret double %conv 81 } 82 83 define double @funcB(i32* nocapture %e) nounwind uwtable readonly ssp { 84 ; CHECK-LABEL: funcB: 85 ; CHECK: # BB#0: 86 ; CHECK-NEXT: vcvtsi2sdl (%rdi), %xmm0, %xmm0 87 ; CHECK-NEXT: retq 88 %tmp1 = load i32, i32* %e, align 4 89 %conv = sitofp i32 %tmp1 to double 90 ret double %conv 91 } 92 93 define float @funcC(i32* nocapture %e) nounwind uwtable readonly ssp { 94 ; CHECK-LABEL: funcC: 95 ; CHECK: # BB#0: 96 ; CHECK-NEXT: vcvtsi2ssl (%rdi), %xmm0, %xmm0 97 ; CHECK-NEXT: retq 98 %tmp1 = load i32, i32* %e, align 4 99 %conv = sitofp i32 %tmp1 to float 100 ret float %conv 101 } 102 103 define float @funcD(i64* nocapture %e) nounwind uwtable readonly ssp { 104 ; CHECK-LABEL: funcD: 105 ; CHECK: # BB#0: 106 ; CHECK-NEXT: vcvtsi2ssq (%rdi), %xmm0, %xmm0 107 ; CHECK-NEXT: retq 108 %tmp1 = load i64, i64* %e, align 8 109 %conv = sitofp i64 %tmp1 to float 110 ret float %conv 111 } 112 113 define void @fpext() nounwind uwtable { 114 ; CHECK-LABEL: fpext: 115 ; CHECK: # BB#0: 116 ; CHECK-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero 117 ; CHECK-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0 118 ; CHECK-NEXT: vmovsd %xmm0, -{{[0-9]+}}(%rsp) 119 ; CHECK-NEXT: retq 120 %f = alloca float, align 4 121 %d = alloca double, align 8 122 %tmp = load float, float* %f, align 4 123 %conv = fpext float %tmp to double 124 store double %conv, double* %d, align 8 125 ret void 126 } 127 128 define double @nearbyint_f64(double %a) { 129 ; CHECK-LABEL: nearbyint_f64: 130 ; CHECK: # BB#0: 131 ; CHECK-NEXT: vroundsd $12, %xmm0, %xmm0, %xmm0 132 ; CHECK-NEXT: retq 133 %res = call double @llvm.nearbyint.f64(double %a) 134 ret double %res 135 } 136 declare double @llvm.nearbyint.f64(double %p) 137 138 define float @floor_f32(float %a) { 139 ; CHECK-LABEL: floor_f32: 140 ; CHECK: # BB#0: 141 ; CHECK-NEXT: vroundss $1, %xmm0, %xmm0, %xmm0 142 ; CHECK-NEXT: retq 143 %res = call float @llvm.floor.f32(float %a) 144 ret float %res 145 } 146 declare float @llvm.floor.f32(float %p) 147 148 149
