1 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s
3 define <8 x float> @sitofp00(<8 x i32> %a) nounwind {
4 ; CHECK-LABEL: sitofp00:
6 ; CHECK-NEXT: vcvtdq2ps %ymm0, %ymm0
8 %b = sitofp <8 x i32> %a to <8 x float>
12 define <8 x i32> @fptosi00(<8 x float> %a) nounwind {
13 ; CHECK-LABEL: fptosi00:
15 ; CHECK-NEXT: vcvttps2dq %ymm0, %ymm0
17 %b = fptosi <8 x float> %a to <8 x i32>
21 define <4 x double> @sitofp01(<4 x i32> %a) {
22 ; CHECK-LABEL: sitofp01:
24 ; CHECK-NEXT: vcvtdq2pd %xmm0, %ymm0
26 %b = sitofp <4 x i32> %a to <4 x double>
30 define <8 x float> @sitofp02(<8 x i16> %a) {
31 ; CHECK-LABEL: sitofp02:
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
39 %b = sitofp <8 x i16> %a to <8 x float>
43 define <4 x i32> @fptosi01(<4 x double> %a) {
44 ; CHECK-LABEL: fptosi01:
46 ; CHECK-NEXT: vcvttpd2dqy %ymm0, %xmm0
47 ; CHECK-NEXT: vzeroupper
49 %b = fptosi <4 x double> %a to <4 x i32>
53 define <8 x float> @fptrunc00(<8 x double> %b) nounwind {
54 ; CHECK-LABEL: fptrunc00:
56 ; CHECK-NEXT: vcvtpd2psy %ymm0, %xmm0
57 ; CHECK-NEXT: vcvtpd2psy %ymm1, %xmm1
58 ; CHECK-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
60 %a = fptrunc <8 x double> %b to <8 x float>
64 define <4 x double> @fpext00(<4 x float> %b) nounwind {
65 ; CHECK-LABEL: fpext00:
67 ; CHECK-NEXT: vcvtps2pd %xmm0, %ymm0
69 %a = fpext <4 x float> %b to <4 x double>
73 define double @funcA(i64* nocapture %e) nounwind uwtable readonly ssp {
76 ; CHECK-NEXT: vcvtsi2sdq (%rdi), %xmm0, %xmm0
78 %tmp1 = load i64, i64* %e, align 8
79 %conv = sitofp i64 %tmp1 to double
83 define double @funcB(i32* nocapture %e) nounwind uwtable readonly ssp {
86 ; CHECK-NEXT: vcvtsi2sdl (%rdi), %xmm0, %xmm0
88 %tmp1 = load i32, i32* %e, align 4
89 %conv = sitofp i32 %tmp1 to double
93 define float @funcC(i32* nocapture %e) nounwind uwtable readonly ssp {
96 ; CHECK-NEXT: vcvtsi2ssl (%rdi), %xmm0, %xmm0
98 %tmp1 = load i32, i32* %e, align 4
99 %conv = sitofp i32 %tmp1 to float
103 define float @funcD(i64* nocapture %e) nounwind uwtable readonly ssp {
104 ; CHECK-LABEL: funcD:
106 ; CHECK-NEXT: vcvtsi2ssq (%rdi), %xmm0, %xmm0
108 %tmp1 = load i64, i64* %e, align 8
109 %conv = sitofp i64 %tmp1 to float
113 define void @fpext() nounwind uwtable {
114 ; CHECK-LABEL: fpext:
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)
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
128 define double @nearbyint_f64(double %a) {
129 ; CHECK-LABEL: nearbyint_f64:
131 ; CHECK-NEXT: vroundsd $12, %xmm0, %xmm0, %xmm0
133 %res = call double @llvm.nearbyint.f64(double %a)
136 declare double @llvm.nearbyint.f64(double %p)
138 define float @floor_f32(float %a) {
139 ; CHECK-LABEL: floor_f32:
141 ; CHECK-NEXT: vroundss $1, %xmm0, %xmm0, %xmm0
143 %res = call float @llvm.floor.f32(float %a)
146 declare float @llvm.floor.f32(float %p)