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
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
|
; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
; Test fcmp pred (fneg x), c -> fcmp (swapped pred) x, -c combine.
; GCN-LABEL: {{^}}multi_use_fneg_src:
; GCN: buffer_load_dword [[A:v[0-9]+]]
; GCN: buffer_load_dword [[B:v[0-9]+]]
; GCN: buffer_load_dword [[C:v[0-9]+]]
; GCN: v_mul_f32_e32 [[MUL:v[0-9]+]], [[A]], [[B]]
; GCN: v_cmp_eq_f32_e32 vcc, -4.0, [[MUL]]
; GCN: buffer_store_dword [[MUL]]
define amdgpu_kernel void @multi_use_fneg_src() #0 {
%a = load volatile float, ptr addrspace(1) undef
%b = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%mul = fmul float %a, %b
%neg.mul = fsub float -0.0, %mul
%cmp = fcmp oeq float %neg.mul, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
store volatile float %mul, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}multi_foldable_use_fneg_src:
; GCN: buffer_load_dword [[A:v[0-9]+]]
; GCN: buffer_load_dword [[B:v[0-9]+]]
; GCN: buffer_load_dword [[C:v[0-9]+]]
; GCN: v_mul_f32_e32 [[MUL:v[0-9]+]], [[A]], [[B]]
; GCN: v_cmp_eq_f32_e32 vcc, -4.0, [[A]]
; GCN: v_mul_f32_e64 [[USE1:v[0-9]+]], [[MUL]], -[[MUL]]
define amdgpu_kernel void @multi_foldable_use_fneg_src() #0 {
%a = load volatile float, ptr addrspace(1) undef
%b = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%mul = fmul float %a, %b
%neg.mul = fsub float -0.0, %mul
%use1 = fmul float %mul, %neg.mul
%cmp = fcmp oeq float %neg.mul, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
store volatile float %use1, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}multi_use_fneg:
; GCN: buffer_load_dword [[A:v[0-9]+]]
; GCN: buffer_load_dword [[B:v[0-9]+]]
; GCN: buffer_load_dword [[C:v[0-9]+]]
; GCN: v_mul_f32_e64 [[MUL:v[0-9]+]], [[A]], -[[B]]
; GCN-NEXT: v_cmp_eq_f32_e32 vcc, 4.0, [[MUL]]
; GCN-NOT: xor
; GCN: buffer_store_dword [[MUL]]
define amdgpu_kernel void @multi_use_fneg() #0 {
%a = load volatile float, ptr addrspace(1) undef
%b = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%mul = fmul float %a, %b
%neg.mul = fsub float -0.0, %mul
%cmp = fcmp oeq float %neg.mul, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
store volatile float %neg.mul, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}multi_foldable_use_fneg:
; GCN: buffer_load_dword [[A:v[0-9]+]]
; GCN: buffer_load_dword [[B:v[0-9]+]]
; GCN: v_mul_f32_e32 [[MUL0:v[0-9]+]], [[A]], [[B]]
; GCN: v_cmp_eq_f32_e32 vcc, -4.0, [[MUL0]]
; GCN: v_mul_f32_e64 [[MUL1:v[0-9]+]], -[[MUL0]], [[MUL0]]
; GCN: buffer_store_dword [[MUL1]]
define amdgpu_kernel void @multi_foldable_use_fneg() #0 {
%a = load volatile float, ptr addrspace(1) undef
%b = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%z = load volatile i32, ptr addrspace(1) undef
%mul = fmul float %a, %b
%neg.mul = fsub float -0.0, %mul
%cmp = fcmp oeq float %neg.mul, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
%use1 = fmul float %neg.mul, %mul
store volatile i32 %select, ptr addrspace(1) undef
store volatile float %use1, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}test_setcc_fneg_oeq_posk_f32:
; GCN: v_cmp_eq_f32_e32 vcc, -4.0, v{{[0-9]+}}
define amdgpu_kernel void @test_setcc_fneg_oeq_posk_f32() #0 {
%a = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%neg.a = fsub float -0.0, %a
%cmp = fcmp oeq float %neg.a, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}test_setcc_fneg_ogt_posk_f32:
; GCN: v_cmp_gt_f32_e32 vcc, -4.0, v{{[0-9]+}}
define amdgpu_kernel void @test_setcc_fneg_ogt_posk_f32() #0 {
%a = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%neg.a = fsub float -0.0, %a
%cmp = fcmp ogt float %neg.a, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}test_setcc_fneg_oge_posk_f32:
; GCN: v_cmp_ge_f32_e32 vcc, -4.0, v{{[0-9]+}}
define amdgpu_kernel void @test_setcc_fneg_oge_posk_f32() #0 {
%a = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%neg.a = fsub float -0.0, %a
%cmp = fcmp oge float %neg.a, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}test_setcc_fneg_olt_posk_f32:
; GCN: v_cmp_lt_f32_e32 vcc, -4.0, v{{[0-9]+}}
define amdgpu_kernel void @test_setcc_fneg_olt_posk_f32() #0 {
%a = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%neg.a = fsub float -0.0, %a
%cmp = fcmp olt float %neg.a, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}test_setcc_fneg_ole_posk_f32:
; GCN: v_cmp_le_f32_e32 vcc, -4.0, v{{[0-9]+}}
define amdgpu_kernel void @test_setcc_fneg_ole_posk_f32() #0 {
%a = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%neg.a = fsub float -0.0, %a
%cmp = fcmp ole float %neg.a, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}test_setcc_fneg_one_posk_f32:
; GCN: v_cmp_lg_f32_e32 vcc, -4.0, v{{[0-9]+}}
define amdgpu_kernel void @test_setcc_fneg_one_posk_f32() #0 {
%a = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%neg.a = fsub float -0.0, %a
%cmp = fcmp one float %neg.a, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}test_setcc_fneg_ueq_posk_f32:
; GCN: v_cmp_nlg_f32_e32 vcc, -4.0, v{{[0-9]+}}
define amdgpu_kernel void @test_setcc_fneg_ueq_posk_f32() #0 {
%a = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%neg.a = fsub float -0.0, %a
%cmp = fcmp ueq float %neg.a, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}test_setcc_fneg_ugt_posk_f32:
; GCN: v_cmp_nle_f32_e32 vcc, -4.0, v{{[0-9]+}}
define amdgpu_kernel void @test_setcc_fneg_ugt_posk_f32() #0 {
%a = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%neg.a = fsub float -0.0, %a
%cmp = fcmp ugt float %neg.a, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}test_setcc_fneg_uge_posk_f32:
; GCN: v_cmp_nlt_f32_e32 vcc, -4.0, v{{[0-9]+}}
define amdgpu_kernel void @test_setcc_fneg_uge_posk_f32() #0 {
%a = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%neg.a = fsub float -0.0, %a
%cmp = fcmp uge float %neg.a, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}test_setcc_fneg_ult_posk_f32:
; GCN: v_cmp_nge_f32_e32 vcc, -4.0, v{{[0-9]+}}
define amdgpu_kernel void @test_setcc_fneg_ult_posk_f32() #0 {
%a = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%neg.a = fsub float -0.0, %a
%cmp = fcmp ult float %neg.a, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}test_setcc_fneg_ule_posk_f32:
; GCN: v_cmp_ngt_f32_e32 vcc, -4.0, v{{[0-9]+}}
define amdgpu_kernel void @test_setcc_fneg_ule_posk_f32() #0 {
%a = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%neg.a = fsub float -0.0, %a
%cmp = fcmp ule float %neg.a, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
ret void
}
; GCN-LABEL: {{^}}test_setcc_fneg_une_posk_f32:
; GCN: v_cmp_neq_f32_e32 vcc, -4.0, v{{[0-9]+}}
define amdgpu_kernel void @test_setcc_fneg_une_posk_f32() #0 {
%a = load volatile float, ptr addrspace(1) undef
%x = load volatile i32, ptr addrspace(1) undef
%y = load volatile i32, ptr addrspace(1) undef
%neg.a = fsub float -0.0, %a
%cmp = fcmp une float %neg.a, 4.0
%select = select i1 %cmp, i32 %x, i32 %y
store volatile i32 %select, ptr addrspace(1) undef
ret void
}
attributes #0 = { nounwind }
|
