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
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
|
; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=GCN %s
; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=VI -check-prefix=GCN %s
; GCN-LABEL: {{^}}mac_vvv:
; GCN: buffer_load_dword [[A:v[0-9]+]], off, s[{{[0-9]+:[0-9]+}}], 0{{$}}
; GCN: buffer_load_dword [[B:v[0-9]+]], off, s[{{[0-9]+:[0-9]+}}], 0 offset:4
; GCN: buffer_load_dword [[C:v[0-9]+]], off, s[{{[0-9]+:[0-9]+}}], 0 offset:8
; GCN: v_mac_f32_e32 [[C]], [[B]], [[A]]
; GCN: buffer_store_dword [[C]]
define void @mac_vvv(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
entry:
%b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
%c_ptr = getelementptr float, float addrspace(1)* %in, i32 2
%a = load volatile float, float addrspace(1)* %in
%b = load volatile float, float addrspace(1)* %b_ptr
%c = load volatile float, float addrspace(1)* %c_ptr
%tmp0 = fmul float %a, %b
%tmp1 = fadd float %tmp0, %c
store float %tmp1, float addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}mad_inline_sgpr_inline:
; GCN-NOT: v_mac_f32
; GCN: v_mad_f32 v{{[0-9]}}, s{{[0-9]+}}, 0.5, 0.5
define void @mad_inline_sgpr_inline(float addrspace(1)* %out, float %in) #0 {
entry:
%tmp0 = fmul float 0.5, %in
%tmp1 = fadd float %tmp0, 0.5
store float %tmp1, float addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}mad_vvs:
; GCN-NOT: v_mac_f32
; GCN: v_mad_f32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, s{{[0-9]+}}
define void @mad_vvs(float addrspace(1)* %out, float addrspace(1)* %in, float %c) #0 {
entry:
%b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
%a = load float, float addrspace(1)* %in
%b = load float, float addrspace(1)* %b_ptr
%tmp0 = fmul float %a, %b
%tmp1 = fadd float %tmp0, %c
store float %tmp1, float addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}mac_ssv:
; GCN: v_mac_f32_e64 v{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
define void @mac_ssv(float addrspace(1)* %out, float addrspace(1)* %in, float %a) #0 {
entry:
%c = load float, float addrspace(1)* %in
%tmp0 = fmul float %a, %a
%tmp1 = fadd float %tmp0, %c
store float %tmp1, float addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}mac_mad_same_add:
; GCN: v_mad_f32 v{{[0-9]}}, v{{[0-9]+}}, v{{[0-9]+}}, [[ADD:v[0-9]+]]
; GCN: v_mac_f32_e32 [[ADD]], v{{[0-9]+}}, v{{[0-9]+}}
define void @mac_mad_same_add(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
entry:
%b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
%c_ptr = getelementptr float, float addrspace(1)* %in, i32 2
%d_ptr = getelementptr float, float addrspace(1)* %in, i32 3
%e_ptr = getelementptr float, float addrspace(1)* %in, i32 4
%a = load volatile float, float addrspace(1)* %in
%b = load volatile float, float addrspace(1)* %b_ptr
%c = load volatile float, float addrspace(1)* %c_ptr
%d = load volatile float, float addrspace(1)* %d_ptr
%e = load volatile float, float addrspace(1)* %e_ptr
%tmp0 = fmul float %a, %b
%tmp1 = fadd float %tmp0, %c
%tmp2 = fmul float %d, %e
%tmp3 = fadd float %tmp2, %c
%out1 = getelementptr float, float addrspace(1)* %out, i32 1
store float %tmp1, float addrspace(1)* %out
store float %tmp3, float addrspace(1)* %out1
ret void
}
; There is no advantage to using v_mac when one of the operands is negated
; and v_mad accepts more operand types.
; GCN-LABEL: {{^}}mad_neg_src0:
; GCN-NOT: v_mac_f32
; GCN: v_mad_f32 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, v{{[-0-9]}}
define void @mad_neg_src0(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
entry:
%b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
%c_ptr = getelementptr float, float addrspace(1)* %in, i32 2
%a = load float, float addrspace(1)* %in
%b = load float, float addrspace(1)* %b_ptr
%c = load float, float addrspace(1)* %c_ptr
%neg_a = fsub float -0.0, %a
%tmp0 = fmul float %neg_a, %b
%tmp1 = fadd float %tmp0, %c
store float %tmp1, float addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}unsafe_mad_sub0_src0:
; GCN-NOT: v_mac_f32
; GCN: v_mad_f32 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, v{{[-0-9]}}
define void @unsafe_mad_sub0_src0(float addrspace(1)* %out, float addrspace(1)* %in) #1 {
entry:
%b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
%c_ptr = getelementptr float, float addrspace(1)* %in, i32 2
%a = load float, float addrspace(1)* %in
%b = load float, float addrspace(1)* %b_ptr
%c = load float, float addrspace(1)* %c_ptr
%neg_a = fsub float 0.0, %a
%tmp0 = fmul float %neg_a, %b
%tmp1 = fadd float %tmp0, %c
store float %tmp1, float addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}safe_mad_sub0_src0:
; GCN: v_sub_f32_e32 [[SUB0:v[0-9]+]], 0,
; GCN: v_mac_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, [[SUB0]]
define void @safe_mad_sub0_src0(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
entry:
%b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
%c_ptr = getelementptr float, float addrspace(1)* %in, i32 2
%a = load float, float addrspace(1)* %in
%b = load float, float addrspace(1)* %b_ptr
%c = load float, float addrspace(1)* %c_ptr
%neg_a = fsub float 0.0, %a
%tmp0 = fmul float %neg_a, %b
%tmp1 = fadd float %tmp0, %c
store float %tmp1, float addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}mad_neg_src1:
; GCN-NOT: v_mac_f32
; GCN: v_mad_f32 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, v{{[-0-9]}}
define void @mad_neg_src1(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
entry:
%b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
%c_ptr = getelementptr float, float addrspace(1)* %in, i32 2
%a = load float, float addrspace(1)* %in
%b = load float, float addrspace(1)* %b_ptr
%c = load float, float addrspace(1)* %c_ptr
%neg_b = fsub float -0.0, %b
%tmp0 = fmul float %a, %neg_b
%tmp1 = fadd float %tmp0, %c
store float %tmp1, float addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}unsafe_mad_sub0_src1:
; GCN-NOT: v_mac_f32
; GCN: v_mad_f32 v{{[0-9]+}}, -v{{[0-9]+}}, v{{[0-9]+}}, v{{[-0-9]}}
define void @unsafe_mad_sub0_src1(float addrspace(1)* %out, float addrspace(1)* %in) #1 {
entry:
%b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
%c_ptr = getelementptr float, float addrspace(1)* %in, i32 2
%a = load float, float addrspace(1)* %in
%b = load float, float addrspace(1)* %b_ptr
%c = load float, float addrspace(1)* %c_ptr
%neg_b = fsub float 0.0, %b
%tmp0 = fmul float %a, %neg_b
%tmp1 = fadd float %tmp0, %c
store float %tmp1, float addrspace(1)* %out
ret void
}
; GCN-LABEL: {{^}}mad_neg_src2:
; GCN-NOT: v_mac
; GCN: v_mad_f32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, -v{{[-0-9]}}
define void @mad_neg_src2(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
entry:
%b_ptr = getelementptr float, float addrspace(1)* %in, i32 1
%c_ptr = getelementptr float, float addrspace(1)* %in, i32 2
%a = load float, float addrspace(1)* %in
%b = load float, float addrspace(1)* %b_ptr
%c = load float, float addrspace(1)* %c_ptr
%neg_c = fsub float -0.0, %c
%tmp0 = fmul float %a, %b
%tmp1 = fadd float %tmp0, %neg_c
store float %tmp1, float addrspace(1)* %out
ret void
}
; Without special casing the inline constant check for v_mac_f32's
; src2, this fails to fold the 1.0 into a mad.
; GCN-LABEL: {{^}}fold_inline_imm_into_mac_src2_f32:
; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]]
; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]]
; GCN: v_add_f32_e32 [[TMP2:v[0-9]+]], [[A]], [[A]]
; GCN: v_mad_f32 v{{[0-9]+}}, [[TMP2]], -4.0, 1.0
define void @fold_inline_imm_into_mac_src2_f32(float addrspace(1)* %out, float addrspace(1)* %a, float addrspace(1)* %b) #3 {
bb:
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%tid.ext = sext i32 %tid to i64
%gep.a = getelementptr inbounds float, float addrspace(1)* %a, i64 %tid.ext
%gep.b = getelementptr inbounds float, float addrspace(1)* %b, i64 %tid.ext
%gep.out = getelementptr inbounds float, float addrspace(1)* %out, i64 %tid.ext
%tmp = load volatile float, float addrspace(1)* %gep.a
%tmp1 = load volatile float, float addrspace(1)* %gep.b
%tmp2 = fadd float %tmp, %tmp
%tmp3 = fmul float %tmp2, 4.0
%tmp4 = fsub float 1.0, %tmp3
%tmp5 = fadd float %tmp4, %tmp1
%tmp6 = fadd float %tmp1, %tmp1
%tmp7 = fmul float %tmp6, %tmp
%tmp8 = fsub float 1.0, %tmp7
%tmp9 = fmul float %tmp8, 8.0
%tmp10 = fadd float %tmp5, %tmp9
store float %tmp10, float addrspace(1)* %gep.out
ret void
}
; GCN-LABEL: {{^}}fold_inline_imm_into_mac_src2_f16:
; GCN: {{buffer|flat}}_load_ushort [[A:v[0-9]+]]
; GCN: {{buffer|flat}}_load_ushort [[B:v[0-9]+]]
; FIXME: How is this not folded?
; SI: v_cvt_f32_f16_e32 v{{[0-9]+}}, 0x3c00
; VI: v_add_f16_e32 [[TMP2:v[0-9]+]], [[A]], [[A]]
; VI: v_mad_f16 v{{[0-9]+}}, [[TMP2]], -4.0, 1.0
define void @fold_inline_imm_into_mac_src2_f16(half addrspace(1)* %out, half addrspace(1)* %a, half addrspace(1)* %b) #3 {
bb:
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%tid.ext = sext i32 %tid to i64
%gep.a = getelementptr inbounds half, half addrspace(1)* %a, i64 %tid.ext
%gep.b = getelementptr inbounds half, half addrspace(1)* %b, i64 %tid.ext
%gep.out = getelementptr inbounds half, half addrspace(1)* %out, i64 %tid.ext
%tmp = load volatile half, half addrspace(1)* %gep.a
%tmp1 = load volatile half, half addrspace(1)* %gep.b
%tmp2 = fadd half %tmp, %tmp
%tmp3 = fmul half %tmp2, 4.0
%tmp4 = fsub half 1.0, %tmp3
%tmp5 = fadd half %tmp4, %tmp1
%tmp6 = fadd half %tmp1, %tmp1
%tmp7 = fmul half %tmp6, %tmp
%tmp8 = fsub half 1.0, %tmp7
%tmp9 = fmul half %tmp8, 8.0
%tmp10 = fadd half %tmp5, %tmp9
store half %tmp10, half addrspace(1)* %gep.out
ret void
}
declare i32 @llvm.amdgcn.workitem.id.x() #2
attributes #0 = { nounwind "unsafe-fp-math"="false" }
attributes #1 = { nounwind "unsafe-fp-math"="true" }
attributes #2 = { nounwind readnone }
attributes #3 = { nounwind }
|
