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
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-- -mattr=sse2 | FileCheck %s --check-prefix=SSE
; RUN: llc < %s -mtriple=x86_64-- -mattr=avx | FileCheck %s --check-prefix=AVX
define <2 x double> @splat_fdiv_v2f64(<2 x double> %x, double %y) {
; SSE-LABEL: splat_fdiv_v2f64:
; SSE: # %bb.0:
; SSE-NEXT: movsd {{.*#+}} xmm2 = mem[0],zero
; SSE-NEXT: divsd %xmm1, %xmm2
; SSE-NEXT: unpcklpd {{.*#+}} xmm2 = xmm2[0,0]
; SSE-NEXT: mulpd %xmm2, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: splat_fdiv_v2f64:
; AVX: # %bb.0:
; AVX-NEXT: vmovsd {{.*#+}} xmm2 = mem[0],zero
; AVX-NEXT: vdivsd %xmm1, %xmm2, %xmm1
; AVX-NEXT: vmovddup {{.*#+}} xmm1 = xmm1[0,0]
; AVX-NEXT: vmulpd %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
%vy = insertelement <2 x double> undef, double %y, i32 0
%splaty = shufflevector <2 x double> %vy, <2 x double> undef, <2 x i32> zeroinitializer
%r = fdiv fast <2 x double> %x, %splaty
ret <2 x double> %r
}
define <4 x double> @splat_fdiv_v4f64(<4 x double> %x, double %y) {
; SSE-LABEL: splat_fdiv_v4f64:
; SSE: # %bb.0:
; SSE-NEXT: movsd {{.*#+}} xmm3 = mem[0],zero
; SSE-NEXT: divsd %xmm2, %xmm3
; SSE-NEXT: unpcklpd {{.*#+}} xmm3 = xmm3[0,0]
; SSE-NEXT: mulpd %xmm3, %xmm0
; SSE-NEXT: mulpd %xmm3, %xmm1
; SSE-NEXT: retq
;
; AVX-LABEL: splat_fdiv_v4f64:
; AVX: # %bb.0:
; AVX-NEXT: vmovsd {{.*#+}} xmm2 = mem[0],zero
; AVX-NEXT: vdivsd %xmm1, %xmm2, %xmm1
; AVX-NEXT: vmovddup {{.*#+}} xmm1 = xmm1[0,0]
; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm1, %ymm1
; AVX-NEXT: vmulpd %ymm1, %ymm0, %ymm0
; AVX-NEXT: retq
%vy = insertelement <4 x double> undef, double %y, i32 0
%splaty = shufflevector <4 x double> %vy, <4 x double> undef, <4 x i32> zeroinitializer
%r = fdiv arcp <4 x double> %x, %splaty
ret <4 x double> %r
}
define <4 x float> @splat_fdiv_v4f32(<4 x float> %x, float %y) {
; SSE-LABEL: splat_fdiv_v4f32:
; SSE: # %bb.0:
; SSE-NEXT: movss {{.*#+}} xmm2 = mem[0],zero,zero,zero
; SSE-NEXT: divss %xmm1, %xmm2
; SSE-NEXT: shufps {{.*#+}} xmm2 = xmm2[0,0,0,0]
; SSE-NEXT: mulps %xmm2, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: splat_fdiv_v4f32:
; AVX: # %bb.0:
; AVX-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero
; AVX-NEXT: vdivss %xmm1, %xmm2, %xmm1
; AVX-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[0,0,0,0]
; AVX-NEXT: vmulps %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
%vy = insertelement <4 x float> undef, float %y, i32 0
%splaty = shufflevector <4 x float> %vy, <4 x float> undef, <4 x i32> zeroinitializer
%r = fdiv arcp reassoc <4 x float> %x, %splaty
ret <4 x float> %r
}
define <8 x float> @splat_fdiv_v8f32(<8 x float> %x, float %y) {
; SSE-LABEL: splat_fdiv_v8f32:
; SSE: # %bb.0:
; SSE-NEXT: movss {{.*#+}} xmm3 = mem[0],zero,zero,zero
; SSE-NEXT: divss %xmm2, %xmm3
; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,0,0,0]
; SSE-NEXT: mulps %xmm3, %xmm0
; SSE-NEXT: mulps %xmm3, %xmm1
; SSE-NEXT: retq
;
; AVX-LABEL: splat_fdiv_v8f32:
; AVX: # %bb.0:
; AVX-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero
; AVX-NEXT: vdivss %xmm1, %xmm2, %xmm1
; AVX-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[0,0,0,0]
; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm1, %ymm1
; AVX-NEXT: vmulps %ymm1, %ymm0, %ymm0
; AVX-NEXT: retq
%vy = insertelement <8 x float> undef, float %y, i32 0
%splaty = shufflevector <8 x float> %vy, <8 x float> undef, <8 x i32> zeroinitializer
%r = fdiv fast <8 x float> %x, %splaty
ret <8 x float> %r
}
define <4 x float> @splat_fdiv_v4f32_estimate(<4 x float> %x, float %y) #0 {
; SSE-LABEL: splat_fdiv_v4f32_estimate:
; SSE: # %bb.0:
; SSE-NEXT: rcpss %xmm1, %xmm2
; SSE-NEXT: mulss %xmm2, %xmm1
; SSE-NEXT: movss {{.*#+}} xmm3 = mem[0],zero,zero,zero
; SSE-NEXT: subss %xmm1, %xmm3
; SSE-NEXT: mulss %xmm2, %xmm3
; SSE-NEXT: addss %xmm2, %xmm3
; SSE-NEXT: shufps {{.*#+}} xmm3 = xmm3[0,0,0,0]
; SSE-NEXT: mulps %xmm3, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: splat_fdiv_v4f32_estimate:
; AVX: # %bb.0:
; AVX-NEXT: vrcpss %xmm1, %xmm1, %xmm2
; AVX-NEXT: vmulss %xmm2, %xmm1, %xmm1
; AVX-NEXT: vmovss {{.*#+}} xmm3 = mem[0],zero,zero,zero
; AVX-NEXT: vsubss %xmm1, %xmm3, %xmm1
; AVX-NEXT: vmulss %xmm1, %xmm2, %xmm1
; AVX-NEXT: vaddss %xmm1, %xmm2, %xmm1
; AVX-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[0,0,0,0]
; AVX-NEXT: vmulps %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
%vy = insertelement <4 x float> undef, float %y, i32 0
%splaty = shufflevector <4 x float> %vy, <4 x float> undef, <4 x i32> zeroinitializer
%r = fdiv arcp reassoc <4 x float> %x, %splaty
ret <4 x float> %r
}
define <8 x float> @splat_fdiv_v8f32_estimate(<8 x float> %x, float %y) #0 {
; SSE-LABEL: splat_fdiv_v8f32_estimate:
; SSE: # %bb.0:
; SSE-NEXT: rcpss %xmm2, %xmm3
; SSE-NEXT: mulss %xmm3, %xmm2
; SSE-NEXT: movss {{.*#+}} xmm4 = mem[0],zero,zero,zero
; SSE-NEXT: subss %xmm2, %xmm4
; SSE-NEXT: mulss %xmm3, %xmm4
; SSE-NEXT: addss %xmm3, %xmm4
; SSE-NEXT: shufps {{.*#+}} xmm4 = xmm4[0,0,0,0]
; SSE-NEXT: mulps %xmm4, %xmm0
; SSE-NEXT: mulps %xmm4, %xmm1
; SSE-NEXT: retq
;
; AVX-LABEL: splat_fdiv_v8f32_estimate:
; AVX: # %bb.0:
; AVX-NEXT: vrcpss %xmm1, %xmm1, %xmm2
; AVX-NEXT: vmulss %xmm2, %xmm1, %xmm1
; AVX-NEXT: vmovss {{.*#+}} xmm3 = mem[0],zero,zero,zero
; AVX-NEXT: vsubss %xmm1, %xmm3, %xmm1
; AVX-NEXT: vmulss %xmm1, %xmm2, %xmm1
; AVX-NEXT: vaddss %xmm1, %xmm2, %xmm1
; AVX-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[0,0,0,0]
; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm1, %ymm1
; AVX-NEXT: vmulps %ymm1, %ymm0, %ymm0
; AVX-NEXT: retq
%vy = insertelement <8 x float> undef, float %y, i32 0
%splaty = shufflevector <8 x float> %vy, <8 x float> undef, <8 x i32> zeroinitializer
%r = fdiv fast <8 x float> %x, %splaty
ret <8 x float> %r
}
attributes #0 = { "reciprocal-estimates"="divf,vec-divf" }
|