• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+sse2 < %s | FileCheck %s --check-prefix=SSE
3; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+avx < %s | FileCheck %s --check-prefix=AVX
4
5; Verify that we're folding the load into the math instruction.
6
7define float @rcpss(float* %a) {
8; SSE-LABEL: rcpss:
9; SSE:       # BB#0:
10; SSE-NEXT:    rcpss (%rdi), %xmm0
11; SSE-NEXT:    retq
12;
13; AVX-LABEL: rcpss:
14; AVX:       # BB#0:
15; AVX-NEXT:    vrcpss (%rdi), %xmm0, %xmm0
16; AVX-NEXT:    retq
17    %ld = load float, float* %a
18    %ins = insertelement <4 x float> undef, float %ld, i32 0
19    %res = tail call <4 x float> @llvm.x86.sse.rcp.ss(<4 x float> %ins)
20    %ext = extractelement <4 x float> %res, i32 0
21    ret float %ext
22}
23
24define float @rsqrtss(float* %a) {
25; SSE-LABEL: rsqrtss:
26; SSE:       # BB#0:
27; SSE-NEXT:    rsqrtss (%rdi), %xmm0
28; SSE-NEXT:    retq
29;
30; AVX-LABEL: rsqrtss:
31; AVX:       # BB#0:
32; AVX-NEXT:    vrsqrtss (%rdi), %xmm0, %xmm0
33; AVX-NEXT:    retq
34    %ld = load float, float* %a
35    %ins = insertelement <4 x float> undef, float %ld, i32 0
36    %res = tail call <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float> %ins)
37    %ext = extractelement <4 x float> %res, i32 0
38    ret float %ext
39}
40
41define float @sqrtss(float* %a) {
42; SSE-LABEL: sqrtss:
43; SSE:       # BB#0:
44; SSE-NEXT:    sqrtss (%rdi), %xmm0
45; SSE-NEXT:    retq
46;
47; AVX-LABEL: sqrtss:
48; AVX:       # BB#0:
49; AVX-NEXT:    vsqrtss (%rdi), %xmm0, %xmm0
50; AVX-NEXT:    retq
51    %ld = load float, float* %a
52    %ins = insertelement <4 x float> undef, float %ld, i32 0
53    %res = tail call <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float> %ins)
54    %ext = extractelement <4 x float> %res, i32 0
55    ret float %ext
56}
57
58define double @sqrtsd(double* %a) {
59; SSE-LABEL: sqrtsd:
60; SSE:       # BB#0:
61; SSE-NEXT:    sqrtsd (%rdi), %xmm0
62; SSE-NEXT:    retq
63;
64; AVX-LABEL: sqrtsd:
65; AVX:       # BB#0:
66; AVX-NEXT:    vsqrtsd (%rdi), %xmm0, %xmm0
67; AVX-NEXT:    retq
68    %ld = load double, double* %a
69    %ins = insertelement <2 x double> undef, double %ld, i32 0
70    %res = tail call <2 x double> @llvm.x86.sse2.sqrt.sd(<2 x double> %ins)
71    %ext = extractelement <2 x double> %res, i32 0
72    ret double %ext
73}
74
75
76declare <4 x float> @llvm.x86.sse.rcp.ss(<4 x float>) nounwind readnone
77declare <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float>) nounwind readnone
78declare <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float>) nounwind readnone
79declare <2 x double> @llvm.x86.sse2.sqrt.sd(<2 x double>) nounwind readnone
80
81