xref: /aosp_15_r20/external/llvm/test/CodeGen/X86/fold-load-unops.ll (revision 9880d6810fe72a1726cb53787c6711e909410d58) 
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 we fold loads into unary sse intrinsics only when optimizing for size
6
7define float @rcpss(float* %a) {
8; SSE-LABEL: rcpss:
9; SSE:       # BB#0:
10; SSE-NEXT:    movss (%rdi), %xmm0
11; SSE-NEXT:    rcpss %xmm0, %xmm0
12; SSE-NEXT:    retq
13;
14; AVX-LABEL: rcpss:
15; AVX:       # BB#0:
16; AVX-NEXT:    vmovss (%rdi), %xmm0
17; AVX-NEXT:    vrcpss %xmm0, %xmm0, %xmm0
18; AVX-NEXT:    retq
19    %ld = load float, float* %a
20    %ins = insertelement <4 x float> undef, float %ld, i32 0
21    %res = tail call <4 x float> @llvm.x86.sse.rcp.ss(<4 x float> %ins)
22    %ext = extractelement <4 x float> %res, i32 0
23    ret float %ext
24}
25
26define float @rsqrtss(float* %a) {
27; SSE-LABEL: rsqrtss:
28; SSE:       # BB#0:
29; SSE-NEXT:    movss (%rdi), %xmm0
30; SSE-NEXT:    rsqrtss %xmm0, %xmm0
31; SSE-NEXT:    retq
32;
33; AVX-LABEL: rsqrtss:
34; AVX:       # BB#0:
35; AVX-NEXT:    vmovss (%rdi), %xmm0
36; AVX-NEXT:    vrsqrtss %xmm0, %xmm0, %xmm0
37; AVX-NEXT:    retq
38    %ld = load float, float* %a
39    %ins = insertelement <4 x float> undef, float %ld, i32 0
40    %res = tail call <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float> %ins)
41    %ext = extractelement <4 x float> %res, i32 0
42    ret float %ext
43}
44
45define float @sqrtss(float* %a) {
46; SSE-LABEL: sqrtss:
47; SSE:       # BB#0:
48; SSE-NEXT:    movss (%rdi), %xmm0
49; SSE-NEXT:    sqrtss %xmm0, %xmm0
50; SSE-NEXT:    retq
51;
52; AVX-LABEL: sqrtss:
53; AVX:       # BB#0:
54; AVX-NEXT:    vmovss (%rdi), %xmm0
55; AVX-NEXT:    vsqrtss %xmm0, %xmm0, %xmm0
56; AVX-NEXT:    retq
57    %ld = load float, float* %a
58    %ins = insertelement <4 x float> undef, float %ld, i32 0
59    %res = tail call <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float> %ins)
60    %ext = extractelement <4 x float> %res, i32 0
61    ret float %ext
62}
63
64define double @sqrtsd(double* %a) {
65; SSE-LABEL: sqrtsd:
66; SSE:       # BB#0:
67; SSE-NEXT:    movsd (%rdi), %xmm0
68; SSE-NEXT:    sqrtsd %xmm0, %xmm0
69; SSE-NEXT:    retq
70;
71; AVX-LABEL: sqrtsd:
72; AVX:       # BB#0:
73; AVX-NEXT:    vmovsd (%rdi), %xmm0
74; AVX-NEXT:    vsqrtsd %xmm0, %xmm0, %xmm0
75; AVX-NEXT:    retq
76    %ld = load double, double* %a
77    %ins = insertelement <2 x double> undef, double %ld, i32 0
78    %res = tail call <2 x double> @llvm.x86.sse2.sqrt.sd(<2 x double> %ins)
79    %ext = extractelement <2 x double> %res, i32 0
80    ret double %ext
81}
82
83define float @rcpss_size(float* %a) optsize {
84; SSE-LABEL: rcpss_size:
85; SSE:       # BB#0:
86; SSE-NEXT:    rcpss (%rdi), %xmm0
87; SSE-NEXT:    retq
88;
89; AVX-LABEL: rcpss_size:
90; AVX:       # BB#0:
91; AVX-NEXT:    vrcpss (%rdi), %xmm0, %xmm0
92; AVX-NEXT:    retq
93    %ld = load float, float* %a
94    %ins = insertelement <4 x float> undef, float %ld, i32 0
95    %res = tail call <4 x float> @llvm.x86.sse.rcp.ss(<4 x float> %ins)
96    %ext = extractelement <4 x float> %res, i32 0
97    ret float %ext
98}
99
100define float @rsqrtss_size(float* %a) optsize {
101; SSE-LABEL: rsqrtss_size:
102; SSE:       # BB#0:
103; SSE-NEXT:    rsqrtss (%rdi), %xmm0
104; SSE-NEXT:    retq
105;
106; AVX-LABEL: rsqrtss_size:
107; AVX:       # BB#0:
108; AVX-NEXT:    vrsqrtss (%rdi), %xmm0, %xmm0
109; AVX-NEXT:    retq
110    %ld = load float, float* %a
111    %ins = insertelement <4 x float> undef, float %ld, i32 0
112    %res = tail call <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float> %ins)
113    %ext = extractelement <4 x float> %res, i32 0
114    ret float %ext
115}
116
117define float @sqrtss_size(float* %a) optsize{
118; SSE-LABEL: sqrtss_size:
119; SSE:       # BB#0:
120; SSE-NEXT:    sqrtss (%rdi), %xmm0
121; SSE-NEXT:    retq
122;
123; AVX-LABEL: sqrtss_size:
124; AVX:       # BB#0:
125; AVX-NEXT:    vsqrtss (%rdi), %xmm0, %xmm0
126; AVX-NEXT:    retq
127    %ld = load float, float* %a
128    %ins = insertelement <4 x float> undef, float %ld, i32 0
129    %res = tail call <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float> %ins)
130    %ext = extractelement <4 x float> %res, i32 0
131    ret float %ext
132}
133
134define double @sqrtsd_size(double* %a) optsize {
135; SSE-LABEL: sqrtsd_size:
136; SSE:       # BB#0:
137; SSE-NEXT:    sqrtsd (%rdi), %xmm0
138; SSE-NEXT:    retq
139;
140; AVX-LABEL: sqrtsd_size:
141; AVX:       # BB#0:
142; AVX-NEXT:    vsqrtsd (%rdi), %xmm0, %xmm0
143; AVX-NEXT:    retq
144    %ld = load double, double* %a
145    %ins = insertelement <2 x double> undef, double %ld, i32 0
146    %res = tail call <2 x double> @llvm.x86.sse2.sqrt.sd(<2 x double> %ins)
147    %ext = extractelement <2 x double> %res, i32 0
148    ret double %ext
149}
150
151declare <4 x float> @llvm.x86.sse.rcp.ss(<4 x float>) nounwind readnone
152declare <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float>) nounwind readnone
153declare <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float>) nounwind readnone
154declare <2 x double> @llvm.x86.sse2.sqrt.sd(<2 x double>) nounwind readnone
155