• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=-avx,+sse2 -fast-isel -fast-isel-abort=1 | FileCheck %s --check-prefix=SSE
2; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=-avx2,+avx -fast-isel -fast-isel-abort=1 | FileCheck %s --check-prefix=AVX
3
4define i32 @cvt_test1(float %a) {
5; SSE-LABEL: cvt_test1
6; SSE:       cvttss2si %xmm0, %eax
7; AVX-LABEL: cvt_test1
8; AVX:       vcvttss2si %xmm0, %eax
9  %1 = insertelement <4 x float> undef, float %a, i32 0
10  %2 = insertelement <4 x float> %1, float 0.000000e+00, i32 1
11  %3 = insertelement <4 x float> %2, float 0.000000e+00, i32 2
12  %4 = insertelement <4 x float> %3, float 0.000000e+00, i32 3
13  %5 = call i32 @llvm.x86.sse.cvttss2si(<4 x float> %4)
14  ret i32 %5
15}
16declare i32 @llvm.x86.sse.cvttss2si(<4 x float>) nounwind readnone
17
18define i64 @cvt_test2(float %a) {
19; SSE-LABEL: cvt_test2
20; SSE:       cvttss2si %xmm0, %rax
21; AVX-LABEL: cvt_test2
22; AVX:       vcvttss2si %xmm0, %rax
23  %1 = insertelement <4 x float> undef, float %a, i32 0
24  %2 = insertelement <4 x float> %1, float 0.000000e+00, i32 1
25  %3 = insertelement <4 x float> %2, float 0.000000e+00, i32 2
26  %4 = insertelement <4 x float> %3, float 0.000000e+00, i32 3
27  %5 = call i64 @llvm.x86.sse.cvttss2si64(<4 x float> %4)
28  ret i64 %5
29}
30declare i64 @llvm.x86.sse.cvttss2si64(<4 x float>) nounwind readnone
31
32define i32 @cvt_test3(double %a) {
33; SSE-LABEL: cvt_test3
34; SSE:       cvttsd2si %xmm0, %eax
35; AVX-LABEL: cvt_test3
36; AVX:       vcvttsd2si %xmm0, %eax
37  %1 = insertelement <2 x double> undef, double %a, i32 0
38  %2 = insertelement <2 x double> %1, double 0.000000e+00, i32 1
39  %3 = call i32 @llvm.x86.sse2.cvttsd2si(<2 x double> %2)
40  ret i32 %3
41}
42declare i32 @llvm.x86.sse2.cvttsd2si(<2 x double>) nounwind readnone
43
44define i64 @cvt_test4(double %a) {
45; SSE-LABEL: cvt_test4
46; SSE:       cvttsd2si %xmm0, %rax
47; AVX-LABEL: cvt_test4
48; AVX:       vcvttsd2si %xmm0, %rax
49  %1 = insertelement <2 x double> undef, double %a, i32 0
50  %2 = insertelement <2 x double> %1, double 0.000000e+00, i32 1
51  %3 = call i64 @llvm.x86.sse2.cvttsd2si64(<2 x double> %2)
52  ret i64 %3
53}
54declare i64 @llvm.x86.sse2.cvttsd2si64(<2 x double>) nounwind readnone
55