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
| ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
declare float @llvm.amdgcn.rcp.f32(float) #0
declare double @llvm.amdgcn.rcp.f64(double) #0
declare double @llvm.sqrt.f64(double) #0
declare float @llvm.sqrt.f32(float) #0
; FUNC-LABEL: {{^}}rcp_undef_f32:
; SI-NOT: v_rcp_f32
define amdgpu_kernel void @rcp_undef_f32(float addrspace(1)* %out) #1 {
%rcp = call float @llvm.amdgcn.rcp.f32(float undef)
store float %rcp, float addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: {{^}}rcp_2_f32:
; SI-NOT: v_rcp_f32
; SI: v_mov_b32_e32 v{{[0-9]+}}, 0.5
define amdgpu_kernel void @rcp_2_f32(float addrspace(1)* %out) #1 {
%rcp = call float @llvm.amdgcn.rcp.f32(float 2.0)
store float %rcp, float addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: {{^}}rcp_10_f32:
; SI-NOT: v_rcp_f32
; SI: v_mov_b32_e32 v{{[0-9]+}}, 0x3dcccccd
define amdgpu_kernel void @rcp_10_f32(float addrspace(1)* %out) #1 {
%rcp = call float @llvm.amdgcn.rcp.f32(float 10.0)
store float %rcp, float addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: {{^}}safe_no_fp32_denormals_rcp_f32:
; SI: v_rcp_f32_e32 [[RESULT:v[0-9]+]], s{{[0-9]+}}
; SI-NOT: [[RESULT]]
; SI: buffer_store_dword [[RESULT]]
define amdgpu_kernel void @safe_no_fp32_denormals_rcp_f32(float addrspace(1)* %out, float %src) #1 {
%rcp = fdiv float 1.0, %src
store float %rcp, float addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: {{^}}safe_f32_denormals_rcp_pat_f32:
; SI: v_rcp_f32_e32 [[RESULT:v[0-9]+]], s{{[0-9]+}}
; SI-NOT: [[RESULT]]
; SI: buffer_store_dword [[RESULT]]
define amdgpu_kernel void @safe_f32_denormals_rcp_pat_f32(float addrspace(1)* %out, float %src) #4 {
%rcp = fdiv float 1.0, %src
store float %rcp, float addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: {{^}}unsafe_f32_denormals_rcp_pat_f32:
; SI: v_div_scale_f32
define amdgpu_kernel void @unsafe_f32_denormals_rcp_pat_f32(float addrspace(1)* %out, float %src) #3 {
%rcp = fdiv float 1.0, %src
store float %rcp, float addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: {{^}}safe_rsq_rcp_pat_f32:
; SI: v_sqrt_f32_e32
; SI: v_rcp_f32_e32
define amdgpu_kernel void @safe_rsq_rcp_pat_f32(float addrspace(1)* %out, float %src) #1 {
%sqrt = call float @llvm.sqrt.f32(float %src)
%rcp = call float @llvm.amdgcn.rcp.f32(float %sqrt)
store float %rcp, float addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: {{^}}unsafe_rsq_rcp_pat_f32:
; SI: v_rsq_f32_e32
define amdgpu_kernel void @unsafe_rsq_rcp_pat_f32(float addrspace(1)* %out, float %src) #2 {
%sqrt = call float @llvm.sqrt.f32(float %src)
%rcp = call float @llvm.amdgcn.rcp.f32(float %sqrt)
store float %rcp, float addrspace(1)* %out, align 4
ret void
}
; FUNC-LABEL: {{^}}rcp_f64:
; SI: v_rcp_f64_e32 [[RESULT:v\[[0-9]+:[0-9]+\]]], s{{\[[0-9]+:[0-9]+\]}}
; SI-NOT: [[RESULT]]
; SI: buffer_store_dwordx2 [[RESULT]]
define amdgpu_kernel void @rcp_f64(double addrspace(1)* %out, double %src) #1 {
%rcp = call double @llvm.amdgcn.rcp.f64(double %src)
store double %rcp, double addrspace(1)* %out, align 8
ret void
}
; FUNC-LABEL: {{^}}unsafe_rcp_f64:
; SI: v_rcp_f64_e32 [[RESULT:v\[[0-9]+:[0-9]+\]]], s{{\[[0-9]+:[0-9]+\]}}
; SI-NOT: [[RESULT]]
; SI: buffer_store_dwordx2 [[RESULT]]
define amdgpu_kernel void @unsafe_rcp_f64(double addrspace(1)* %out, double %src) #2 {
%rcp = call double @llvm.amdgcn.rcp.f64(double %src)
store double %rcp, double addrspace(1)* %out, align 8
ret void
}
; FUNC-LABEL: {{^}}rcp_pat_f64:
; SI: v_div_scale_f64
define amdgpu_kernel void @rcp_pat_f64(double addrspace(1)* %out, double %src) #1 {
%rcp = fdiv double 1.0, %src
store double %rcp, double addrspace(1)* %out, align 8
ret void
}
; FUNC-LABEL: {{^}}unsafe_rcp_pat_f64:
; SI: v_rcp_f64_e32 [[RESULT:v\[[0-9]+:[0-9]+\]]], s{{\[[0-9]+:[0-9]+\]}}
; SI-NOT: [[RESULT]]
; SI: buffer_store_dwordx2 [[RESULT]]
define amdgpu_kernel void @unsafe_rcp_pat_f64(double addrspace(1)* %out, double %src) #2 {
%rcp = fdiv double 1.0, %src
store double %rcp, double addrspace(1)* %out, align 8
ret void
}
; FUNC-LABEL: {{^}}safe_rsq_rcp_pat_f64:
; SI-NOT: v_rsq_f64_e32
; SI: v_sqrt_f64
; SI: v_rcp_f64
define amdgpu_kernel void @safe_rsq_rcp_pat_f64(double addrspace(1)* %out, double %src) #1 {
%sqrt = call double @llvm.sqrt.f64(double %src)
%rcp = call double @llvm.amdgcn.rcp.f64(double %sqrt)
store double %rcp, double addrspace(1)* %out, align 8
ret void
}
; FUNC-LABEL: {{^}}unsafe_rsq_rcp_pat_f64:
; SI: v_rsq_f64_e32 [[RESULT:v\[[0-9]+:[0-9]+\]]], s{{\[[0-9]+:[0-9]+\]}}
; SI-NOT: [[RESULT]]
; SI: buffer_store_dwordx2 [[RESULT]]
define amdgpu_kernel void @unsafe_rsq_rcp_pat_f64(double addrspace(1)* %out, double %src) #2 {
%sqrt = call double @llvm.sqrt.f64(double %src)
%rcp = call double @llvm.amdgcn.rcp.f64(double %sqrt)
store double %rcp, double addrspace(1)* %out, align 8
ret void
}
attributes #0 = { nounwind readnone }
attributes #1 = { nounwind "unsafe-fp-math"="false" "target-features"="-fp32-denormals" }
attributes #2 = { nounwind "unsafe-fp-math"="true" "target-features"="-fp32-denormals" }
attributes #3 = { nounwind "unsafe-fp-math"="false" "target-features"="+fp32-denormals" }
attributes #4 = { nounwind "unsafe-fp-math"="true" "target-features"="+fp32-denormals" }
|