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
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
|
; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
declare i32 @llvm.amdgcn.workitem.id.x() #0
declare float @llvm.fabs.f32(float) #0
; FUNC-LABEL: {{^}}mad_sub_f32:
; SI: buffer_load_dword [[REGA:v[0-9]+]]
; SI: buffer_load_dword [[REGB:v[0-9]+]]
; SI: buffer_load_dword [[REGC:v[0-9]+]]
; SI: v_mad_f32 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], -[[REGC]]
; SI: buffer_store_dword [[RESULT]]
define void @mad_sub_f32(float addrspace(1)* noalias nocapture %out, float addrspace(1)* noalias nocapture readonly %ptr) #1 {
%tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
%tid.ext = sext i32 %tid to i64
%gep0 = getelementptr float, float addrspace(1)* %ptr, i64 %tid.ext
%add1 = add i64 %tid.ext, 1
%gep1 = getelementptr float, float addrspace(1)* %ptr, i64 %add1
%add2 = add i64 %tid.ext, 2
%gep2 = getelementptr float, float addrspace(1)* %ptr, i64 %add2
%outgep = getelementptr float, float addrspace(1)* %out, i64 %tid.ext
%a = load volatile float, float addrspace(1)* %gep0, align 4
%b = load volatile float, float addrspace(1)* %gep1, align 4
%c = load volatile float, float addrspace(1)* %gep2, align 4
%mul = fmul float %a, %b
%sub = fsub float %mul, %c
store float %sub, float addrspace(1)* %outgep, align 4
ret void
}
; FUNC-LABEL: {{^}}mad_sub_inv_f32:
; SI: buffer_load_dword [[REGA:v[0-9]+]]
; SI: buffer_load_dword [[REGB:v[0-9]+]]
; SI: buffer_load_dword [[REGC:v[0-9]+]]
; SI: v_mad_f32 [[RESULT:v[0-9]+]], -[[REGA]], [[REGB]], [[REGC]]
; SI: buffer_store_dword [[RESULT]]
define void @mad_sub_inv_f32(float addrspace(1)* noalias nocapture %out, float addrspace(1)* noalias nocapture readonly %ptr) #1 {
%tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
%tid.ext = sext i32 %tid to i64
%gep0 = getelementptr float, float addrspace(1)* %ptr, i64 %tid.ext
%add1 = add i64 %tid.ext, 1
%gep1 = getelementptr float, float addrspace(1)* %ptr, i64 %add1
%add2 = add i64 %tid.ext, 2
%gep2 = getelementptr float, float addrspace(1)* %ptr, i64 %add2
%outgep = getelementptr float, float addrspace(1)* %out, i64 %tid.ext
%a = load volatile float, float addrspace(1)* %gep0, align 4
%b = load volatile float, float addrspace(1)* %gep1, align 4
%c = load volatile float, float addrspace(1)* %gep2, align 4
%mul = fmul float %a, %b
%sub = fsub float %c, %mul
store float %sub, float addrspace(1)* %outgep, align 4
ret void
}
; FUNC-LABEL: {{^}}mad_sub_f64:
; SI: v_mul_f64
; SI: v_add_f64
define void @mad_sub_f64(double addrspace(1)* noalias nocapture %out, double addrspace(1)* noalias nocapture readonly %ptr) #1 {
%tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
%tid.ext = sext i32 %tid to i64
%gep0 = getelementptr double, double addrspace(1)* %ptr, i64 %tid.ext
%add1 = add i64 %tid.ext, 1
%gep1 = getelementptr double, double addrspace(1)* %ptr, i64 %add1
%add2 = add i64 %tid.ext, 2
%gep2 = getelementptr double, double addrspace(1)* %ptr, i64 %add2
%outgep = getelementptr double, double addrspace(1)* %out, i64 %tid.ext
%a = load volatile double, double addrspace(1)* %gep0, align 8
%b = load volatile double, double addrspace(1)* %gep1, align 8
%c = load volatile double, double addrspace(1)* %gep2, align 8
%mul = fmul double %a, %b
%sub = fsub double %mul, %c
store double %sub, double addrspace(1)* %outgep, align 8
ret void
}
; FUNC-LABEL: {{^}}mad_sub_fabs_f32:
; SI: buffer_load_dword [[REGA:v[0-9]+]]
; SI: buffer_load_dword [[REGB:v[0-9]+]]
; SI: buffer_load_dword [[REGC:v[0-9]+]]
; SI: v_mad_f32 [[RESULT:v[0-9]+]], [[REGA]], [[REGB]], -|[[REGC]]|
; SI: buffer_store_dword [[RESULT]]
define void @mad_sub_fabs_f32(float addrspace(1)* noalias nocapture %out, float addrspace(1)* noalias nocapture readonly %ptr) #1 {
%tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
%tid.ext = sext i32 %tid to i64
%gep0 = getelementptr float, float addrspace(1)* %ptr, i64 %tid.ext
%add1 = add i64 %tid.ext, 1
%gep1 = getelementptr float, float addrspace(1)* %ptr, i64 %add1
%add2 = add i64 %tid.ext, 2
%gep2 = getelementptr float, float addrspace(1)* %ptr, i64 %add2
%outgep = getelementptr float, float addrspace(1)* %out, i64 %tid.ext
%a = load volatile float, float addrspace(1)* %gep0, align 4
%b = load volatile float, float addrspace(1)* %gep1, align 4
%c = load volatile float, float addrspace(1)* %gep2, align 4
%c.abs = call float @llvm.fabs.f32(float %c) #0
%mul = fmul float %a, %b
%sub = fsub float %mul, %c.abs
store float %sub, float addrspace(1)* %outgep, align 4
ret void
}
; FUNC-LABEL: {{^}}mad_sub_fabs_inv_f32:
; SI: buffer_load_dword [[REGA:v[0-9]+]]
; SI: buffer_load_dword [[REGB:v[0-9]+]]
; SI: buffer_load_dword [[REGC:v[0-9]+]]
; SI: v_mad_f32 [[RESULT:v[0-9]+]], -[[REGA]], [[REGB]], |[[REGC]]|
; SI: buffer_store_dword [[RESULT]]
define void @mad_sub_fabs_inv_f32(float addrspace(1)* noalias nocapture %out, float addrspace(1)* noalias nocapture readonly %ptr) #1 {
%tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
%tid.ext = sext i32 %tid to i64
%gep0 = getelementptr float, float addrspace(1)* %ptr, i64 %tid.ext
%add1 = add i64 %tid.ext, 1
%gep1 = getelementptr float, float addrspace(1)* %ptr, i64 %add1
%add2 = add i64 %tid.ext, 2
%gep2 = getelementptr float, float addrspace(1)* %ptr, i64 %add2
%outgep = getelementptr float, float addrspace(1)* %out, i64 %tid.ext
%a = load volatile float, float addrspace(1)* %gep0, align 4
%b = load volatile float, float addrspace(1)* %gep1, align 4
%c = load volatile float, float addrspace(1)* %gep2, align 4
%c.abs = call float @llvm.fabs.f32(float %c) #0
%mul = fmul float %a, %b
%sub = fsub float %c.abs, %mul
store float %sub, float addrspace(1)* %outgep, align 4
ret void
}
; FUNC-LABEL: {{^}}neg_neg_mad_f32:
; SI: v_mac_f32_e32 {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}}
define void @neg_neg_mad_f32(float addrspace(1)* noalias nocapture %out, float addrspace(1)* noalias nocapture readonly %ptr) #1 {
%tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
%tid.ext = sext i32 %tid to i64
%gep0 = getelementptr float, float addrspace(1)* %ptr, i64 %tid.ext
%add1 = add i64 %tid.ext, 1
%gep1 = getelementptr float, float addrspace(1)* %ptr, i64 %add1
%add2 = add i64 %tid.ext, 2
%gep2 = getelementptr float, float addrspace(1)* %ptr, i64 %add2
%outgep = getelementptr float, float addrspace(1)* %out, i64 %tid.ext
%a = load volatile float, float addrspace(1)* %gep0, align 4
%b = load volatile float, float addrspace(1)* %gep1, align 4
%c = load volatile float, float addrspace(1)* %gep2, align 4
%nega = fsub float -0.000000e+00, %a
%negb = fsub float -0.000000e+00, %b
%mul = fmul float %nega, %negb
%sub = fadd float %mul, %c
store float %sub, float addrspace(1)* %outgep, align 4
ret void
}
; FUNC-LABEL: {{^}}mad_fabs_sub_f32:
; SI: buffer_load_dword [[REGA:v[0-9]+]]
; SI: buffer_load_dword [[REGB:v[0-9]+]]
; SI: buffer_load_dword [[REGC:v[0-9]+]]
; SI: v_mad_f32 [[RESULT:v[0-9]+]], [[REGA]], |[[REGB]]|, -[[REGC]]
; SI: buffer_store_dword [[RESULT]]
define void @mad_fabs_sub_f32(float addrspace(1)* noalias nocapture %out, float addrspace(1)* noalias nocapture readonly %ptr) #1 {
%tid = tail call i32 @llvm.amdgcn.workitem.id.x() #0
%tid.ext = sext i32 %tid to i64
%gep0 = getelementptr float, float addrspace(1)* %ptr, i64 %tid.ext
%add1 = add i64 %tid.ext, 1
%gep1 = getelementptr float, float addrspace(1)* %ptr, i64 %add1
%add2 = add i64 %tid.ext, 2
%gep2 = getelementptr float, float addrspace(1)* %ptr, i64 %add2
%outgep = getelementptr float, float addrspace(1)* %out, i64 %tid.ext
%a = load volatile float, float addrspace(1)* %gep0, align 4
%b = load volatile float, float addrspace(1)* %gep1, align 4
%c = load volatile float, float addrspace(1)* %gep2, align 4
%b.abs = call float @llvm.fabs.f32(float %b) #0
%mul = fmul float %a, %b.abs
%sub = fsub float %mul, %c
store float %sub, float addrspace(1)* %outgep, align 4
ret void
}
; FUNC-LABEL: {{^}}fsub_c_fadd_a_a:
; SI-DAG: buffer_load_dword [[R1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
; SI-DAG: buffer_load_dword [[R2:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
; SI: v_mac_f32_e32 [[R2]], -2.0, [[R1]]
; SI: buffer_store_dword [[R2]]
define void @fsub_c_fadd_a_a(float addrspace(1)* %out, float addrspace(1)* %in) {
%tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
%gep.0 = getelementptr float, float addrspace(1)* %out, i32 %tid
%gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
%gep.out = getelementptr float, float addrspace(1)* %out, i32 %tid
%r1 = load volatile float, float addrspace(1)* %gep.0
%r2 = load volatile float, float addrspace(1)* %gep.1
%add = fadd float %r1, %r1
%r3 = fsub float %r2, %add
store float %r3, float addrspace(1)* %gep.out
ret void
}
; FUNC-LABEL: {{^}}fsub_fadd_a_a_c:
; SI-DAG: buffer_load_dword [[R1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
; SI-DAG: buffer_load_dword [[R2:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
; SI: v_mad_f32 [[RESULT:v[0-9]+]], 2.0, [[R1]], -[[R2]]
; SI: buffer_store_dword [[RESULT]]
define void @fsub_fadd_a_a_c(float addrspace(1)* %out, float addrspace(1)* %in) {
%tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
%gep.0 = getelementptr float, float addrspace(1)* %out, i32 %tid
%gep.1 = getelementptr float, float addrspace(1)* %gep.0, i32 1
%gep.out = getelementptr float, float addrspace(1)* %out, i32 %tid
%r1 = load volatile float, float addrspace(1)* %gep.0
%r2 = load volatile float, float addrspace(1)* %gep.1
%add = fadd float %r1, %r1
%r3 = fsub float %add, %r2
store float %r3, float addrspace(1)* %gep.out
ret void
}
attributes #0 = { nounwind readnone }
attributes #1 = { nounwind }
|
