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
| ; RUN: llc -march=amdgcn -mcpu=gfx908 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX908 %s
; GCN-LABEL: {{^}}accvgpr_write_read:
; GFX908: v_accvgpr_write [[AREG:a[0-9]+]], 1
; GFX908: v_accvgpr_read [[VREG:v[0-9]+]], [[AREG]]
; GFX908: global_store_dword {{[^,]+}}, [[VREG]], off
define amdgpu_kernel void @accvgpr_write_read(float addrspace(1)* %arg) {
bb:
%in.1 = load float, float addrspace(1)* %arg
%init = tail call float asm "v_accvgpr_write $0, 1", "=a"()
%read = tail call float asm "v_accvgpr_read $0, $1", "=v,a"(float %init)
store float %read, float addrspace(1)* %arg
ret void
}
; GCN-LABEL: {{^}}v_mfma_f32_4x4x1f32_avva
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_mfma_f32_4x4x1f32 a[{{[0-9:]+}}], v{{[0-9]+}}, v{{[0-9]+}}, a[{{[0-9:]+}}]
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
define amdgpu_kernel void @v_mfma_f32_4x4x1f32_avva(<4 x float> addrspace(1)* %arg) {
bb:
%in.1 = load <4 x float>, <4 x float> addrspace(1)* %arg
%mai.1 = tail call <4 x float> asm "v_mfma_f32_4x4x1f32 $0, $1, $2, $3", "=a,v,v,a"(float 1.0, float 2.0, <4 x float> %in.1)
store <4 x float> %mai.1, <4 x float> addrspace(1)* %arg
ret void
}
; GCN-LABEL: {{^}}v_mfma_f32_4x4x1f32_aaaa
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_mfma_f32_4x4x1f32 a[{{[0-9:]+}}], a{{[0-9]+}}, a{{[0-9]+}}, a[{{[0-9:]+}}]
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
define amdgpu_kernel void @v_mfma_f32_4x4x1f32_aaaa(<4 x float> addrspace(1)* %arg) {
bb:
%in.1 = load <4 x float>, <4 x float> addrspace(1)* %arg
%mai.1 = tail call <4 x float> asm "v_mfma_f32_4x4x1f32 $0, $1, $2, $3", "=a,a,a,a"(float 1.0, float 2.0, <4 x float> %in.1)
store <4 x float> %mai.1, <4 x float> addrspace(1)* %arg
ret void
}
; GCN-LABEL: {{^}}v_mfma_f32_4x4x4f16_aaaa
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_mfma_f32_4x4x4f16 a[{{[0-9:]+}}], a[{{[0-9]+:[0-9]+}}], a[{{[0-9]+:[0-9]+}}], a[{{[0-9:]+}}]
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
define amdgpu_kernel void @v_mfma_f32_4x4x4f16_aaaa(<4 x float> addrspace(1)* %arg) {
bb:
%in.1 = load <4 x float>, <4 x float> addrspace(1)* %arg
%mai.1 = tail call <4 x float> asm "v_mfma_f32_4x4x4f16 $0, $1, $2, $3", "=a,a,a,a"(<4 x half> <half 0xH3800, half 0xH3800, half 0xH3800, half 0xH3800>, <4 x half> <half 0xH03FF, half 0xH03FF, half 0xH03FF, half 0xH03FF>, <4 x float> %in.1)
store <4 x float> %mai.1, <4 x float> addrspace(1)* %arg
ret void
}
; GCN-LABEL: {{^}}v_mfma_f32_16x16x1f32_avaa
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_mfma_f32_16x16x1f32 a[{{[0-9:]+}}], v{{[0-9]+}}, a{{[0-9]+}}, a[{{[0-9:]+}}]
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
define amdgpu_kernel void @v_mfma_f32_16x16x1f32_avaa(<16 x float> addrspace(1)* %arg) {
bb:
%in.1 = load <16 x float>, <16 x float> addrspace(1)* %arg
%mai.1 = tail call <16 x float> asm "v_mfma_f32_16x16x1f32 $0, $1, $2, $3", "=a,v,a,a"(float 1.0, float 2.0, <16 x float> %in.1)
store <16 x float> %mai.1, <16 x float> addrspace(1)* %arg
ret void
}
; GCN-LABEL: {{^}}v_mfma_f32_32x32x1f32_avaa
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_accvgpr_write_b32
; GFX908: v_mfma_f32_32x32x1f32 a[{{[0-9:]+}}], v{{[0-9]+}}, a{{[0-9]+}}, a[{{[0-9:]+}}]
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
; GFX908: v_accvgpr_read_b32
define amdgpu_kernel void @v_mfma_f32_32x32x1f32_avaa(<32 x i32> addrspace(1)* %arg) {
bb:
%in.1 = load <32 x i32>, <32 x i32> addrspace(1)* %arg
%mai.1 = tail call <32 x i32> asm "v_mfma_f32_32x32x1f32 $0, $1, $2, $3", "=a,v,a,a"(float 1.0, float 2.0, <32 x i32> %in.1)
store <32 x i32> %mai.1, <32 x i32> addrspace(1)* %arg
ret void
}
|