1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+sse2 < %s | FileCheck %s --check-prefix=SSE
; RUN: llc -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx < %s | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
; RUN: llc -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512f < %s | FileCheck %s --check-prefix=AVX --check-prefix=AVX512
; Verify that we're folding the load into the math instruction.
; This pattern is generated out of the simplest intrinsics usage:
; _mm_add_ss(a, _mm_load_ss(b));
define <4 x float> @addss(<4 x float> %va, float* %pb) {
; SSE-LABEL: addss:
; SSE: # %bb.0:
; SSE-NEXT: addss (%rdi), %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: addss:
; AVX: # %bb.0:
; AVX-NEXT: vaddss (%rdi), %xmm0, %xmm0
; AVX-NEXT: retq
%a = extractelement <4 x float> %va, i32 0
%b = load float, float* %pb
%r = fadd float %a, %b
%vr = insertelement <4 x float> %va, float %r, i32 0
ret <4 x float> %vr
}
define <2 x double> @addsd(<2 x double> %va, double* %pb) {
; SSE-LABEL: addsd:
; SSE: # %bb.0:
; SSE-NEXT: addsd (%rdi), %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: addsd:
; AVX: # %bb.0:
; AVX-NEXT: vaddsd (%rdi), %xmm0, %xmm0
; AVX-NEXT: retq
%a = extractelement <2 x double> %va, i32 0
%b = load double, double* %pb
%r = fadd double %a, %b
%vr = insertelement <2 x double> %va, double %r, i32 0
ret <2 x double> %vr
}
define <4 x float> @subss(<4 x float> %va, float* %pb) {
; SSE-LABEL: subss:
; SSE: # %bb.0:
; SSE-NEXT: subss (%rdi), %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: subss:
; AVX: # %bb.0:
; AVX-NEXT: vsubss (%rdi), %xmm0, %xmm0
; AVX-NEXT: retq
%a = extractelement <4 x float> %va, i32 0
%b = load float, float* %pb
%r = fsub float %a, %b
%vr = insertelement <4 x float> %va, float %r, i32 0
ret <4 x float> %vr
}
define <2 x double> @subsd(<2 x double> %va, double* %pb) {
; SSE-LABEL: subsd:
; SSE: # %bb.0:
; SSE-NEXT: subsd (%rdi), %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: subsd:
; AVX: # %bb.0:
; AVX-NEXT: vsubsd (%rdi), %xmm0, %xmm0
; AVX-NEXT: retq
%a = extractelement <2 x double> %va, i32 0
%b = load double, double* %pb
%r = fsub double %a, %b
%vr = insertelement <2 x double> %va, double %r, i32 0
ret <2 x double> %vr
}
define <4 x float> @mulss(<4 x float> %va, float* %pb) {
; SSE-LABEL: mulss:
; SSE: # %bb.0:
; SSE-NEXT: mulss (%rdi), %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: mulss:
; AVX: # %bb.0:
; AVX-NEXT: vmulss (%rdi), %xmm0, %xmm0
; AVX-NEXT: retq
%a = extractelement <4 x float> %va, i32 0
%b = load float, float* %pb
%r = fmul float %a, %b
%vr = insertelement <4 x float> %va, float %r, i32 0
ret <4 x float> %vr
}
define <2 x double> @mulsd(<2 x double> %va, double* %pb) {
; SSE-LABEL: mulsd:
; SSE: # %bb.0:
; SSE-NEXT: mulsd (%rdi), %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: mulsd:
; AVX: # %bb.0:
; AVX-NEXT: vmulsd (%rdi), %xmm0, %xmm0
; AVX-NEXT: retq
%a = extractelement <2 x double> %va, i32 0
%b = load double, double* %pb
%r = fmul double %a, %b
%vr = insertelement <2 x double> %va, double %r, i32 0
ret <2 x double> %vr
}
define <4 x float> @divss(<4 x float> %va, float* %pb) {
; SSE-LABEL: divss:
; SSE: # %bb.0:
; SSE-NEXT: divss (%rdi), %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: divss:
; AVX: # %bb.0:
; AVX-NEXT: vdivss (%rdi), %xmm0, %xmm0
; AVX-NEXT: retq
%a = extractelement <4 x float> %va, i32 0
%b = load float, float* %pb
%r = fdiv float %a, %b
%vr = insertelement <4 x float> %va, float %r, i32 0
ret <4 x float> %vr
}
define <2 x double> @divsd(<2 x double> %va, double* %pb) {
; SSE-LABEL: divsd:
; SSE: # %bb.0:
; SSE-NEXT: divsd (%rdi), %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: divsd:
; AVX: # %bb.0:
; AVX-NEXT: vdivsd (%rdi), %xmm0, %xmm0
; AVX-NEXT: retq
%a = extractelement <2 x double> %va, i32 0
%b = load double, double* %pb
%r = fdiv double %a, %b
%vr = insertelement <2 x double> %va, double %r, i32 0
ret <2 x double> %vr
}
|