local-64.ll
8.06 KB
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
; RUN: llc -march=amdgcn -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,SI,SICIVI %s
; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,CI,SICIVI,CIPLUS %s
; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI,SICIVI,CIPLUS %s
; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9,CIPLUS %s
; GCN-LABEL: {{^}}local_i32_load
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN: ds_read_b32 [[REG:v[0-9]+]], v{{[0-9]+}} offset:28
; GCN: buffer_store_dword [[REG]],
define amdgpu_kernel void @local_i32_load(i32 addrspace(1)* %out, i32 addrspace(3)* %in) nounwind {
%gep = getelementptr i32, i32 addrspace(3)* %in, i32 7
%val = load i32, i32 addrspace(3)* %gep, align 4
store i32 %val, i32 addrspace(1)* %out, align 4
ret void
}
; GCN-LABEL: {{^}}local_i32_load_0_offset
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN: ds_read_b32 [[REG:v[0-9]+]], v{{[0-9]+}}
; GCN: buffer_store_dword [[REG]],
define amdgpu_kernel void @local_i32_load_0_offset(i32 addrspace(1)* %out, i32 addrspace(3)* %in) nounwind {
%val = load i32, i32 addrspace(3)* %in, align 4
store i32 %val, i32 addrspace(1)* %out, align 4
ret void
}
; GCN-LABEL: {{^}}local_i8_load_i16_max_offset:
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN-NOT: add
; GCN: ds_read_u8 [[REG:v[0-9]+]], {{v[0-9]+}} offset:65535
; GCN: buffer_store_byte [[REG]],
define amdgpu_kernel void @local_i8_load_i16_max_offset(i8 addrspace(1)* %out, i8 addrspace(3)* %in) nounwind {
%gep = getelementptr i8, i8 addrspace(3)* %in, i32 65535
%val = load i8, i8 addrspace(3)* %gep, align 4
store i8 %val, i8 addrspace(1)* %out, align 4
ret void
}
; GCN-LABEL: {{^}}local_i8_load_over_i16_max_offset:
; SICIVI-DAG: s_mov_b32 m0
; GFX9-NOT: m0
; The LDS offset will be 65536 bytes, which is larger than the size of LDS on
; SI, which is why it is being OR'd with the base pointer.
; SI-DAG: s_bitset1_b32 [[ADDR:s[0-9]+]], 16
; CI-DAG: s_add_i32 [[ADDR:s[0-9]+]], s{{[0-9]+}}, 0x10000
; VI-DAG: s_add_i32 [[ADDR:s[0-9]+]], s{{[0-9]+}}, 0x10000
; GFX9-DAG: s_add_i32 [[ADDR:s[0-9]+]], s{{[0-9]+}}, 0x10000
; GCN-DAG: v_mov_b32_e32 [[VREGADDR:v[0-9]+]], [[ADDR]]
; GCN: ds_read_u8 [[REG:v[0-9]+]], [[VREGADDR]]
; GCN: buffer_store_byte [[REG]],
define amdgpu_kernel void @local_i8_load_over_i16_max_offset(i8 addrspace(1)* %out, i8 addrspace(3)* %in) nounwind {
%gep = getelementptr i8, i8 addrspace(3)* %in, i32 65536
%val = load i8, i8 addrspace(3)* %gep, align 4
store i8 %val, i8 addrspace(1)* %out, align 4
ret void
}
; GCN-LABEL: {{^}}local_i64_load:
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN-NOT: add
; GCN: ds_read_b64 [[REG:v[[0-9]+:[0-9]+]]], v{{[0-9]+}} offset:56
; GCN: buffer_store_dwordx2 [[REG]],
define amdgpu_kernel void @local_i64_load(i64 addrspace(1)* %out, i64 addrspace(3)* %in) nounwind {
%gep = getelementptr i64, i64 addrspace(3)* %in, i32 7
%val = load i64, i64 addrspace(3)* %gep, align 8
store i64 %val, i64 addrspace(1)* %out, align 8
ret void
}
; GCN-LABEL: {{^}}local_i64_load_0_offset
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN: ds_read_b64 [[REG:v\[[0-9]+:[0-9]+\]]], v{{[0-9]+}}
; GCN: buffer_store_dwordx2 [[REG]],
define amdgpu_kernel void @local_i64_load_0_offset(i64 addrspace(1)* %out, i64 addrspace(3)* %in) nounwind {
%val = load i64, i64 addrspace(3)* %in, align 8
store i64 %val, i64 addrspace(1)* %out, align 8
ret void
}
; GCN-LABEL: {{^}}local_f64_load:
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN-NOT: add
; GCN: ds_read_b64 [[REG:v[[0-9]+:[0-9]+]]], v{{[0-9]+}} offset:56
; GCN: buffer_store_dwordx2 [[REG]],
define amdgpu_kernel void @local_f64_load(double addrspace(1)* %out, double addrspace(3)* %in) nounwind {
%gep = getelementptr double, double addrspace(3)* %in, i32 7
%val = load double, double addrspace(3)* %gep, align 8
store double %val, double addrspace(1)* %out, align 8
ret void
}
; GCN-LABEL: {{^}}local_f64_load_0_offset
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN: ds_read_b64 [[REG:v\[[0-9]+:[0-9]+\]]], v{{[0-9]+}}
; GCN: buffer_store_dwordx2 [[REG]],
define amdgpu_kernel void @local_f64_load_0_offset(double addrspace(1)* %out, double addrspace(3)* %in) nounwind {
%val = load double, double addrspace(3)* %in, align 8
store double %val, double addrspace(1)* %out, align 8
ret void
}
; GCN-LABEL: {{^}}local_i64_store:
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN-NOT: add
; GCN: ds_write_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:56
define amdgpu_kernel void @local_i64_store(i64 addrspace(3)* %out) nounwind {
%gep = getelementptr i64, i64 addrspace(3)* %out, i32 7
store i64 5678, i64 addrspace(3)* %gep, align 8
ret void
}
; GCN-LABEL: {{^}}local_i64_store_0_offset:
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN-NOT: add
; GCN: ds_write_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}
define amdgpu_kernel void @local_i64_store_0_offset(i64 addrspace(3)* %out) nounwind {
store i64 1234, i64 addrspace(3)* %out, align 8
ret void
}
; GCN-LABEL: {{^}}local_f64_store:
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN-NOT: add
; GCN: ds_write_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:56
define amdgpu_kernel void @local_f64_store(double addrspace(3)* %out) nounwind {
%gep = getelementptr double, double addrspace(3)* %out, i32 7
store double 16.0, double addrspace(3)* %gep, align 8
ret void
}
; GCN-LABEL: {{^}}local_f64_store_0_offset
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN: ds_write_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}
define amdgpu_kernel void @local_f64_store_0_offset(double addrspace(3)* %out) nounwind {
store double 20.0, double addrspace(3)* %out, align 8
ret void
}
; GCN-LABEL: {{^}}local_v2i64_store:
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN-NOT: add
; SI: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:14 offset1:15
; CIPLUS: ds_write_b128 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:112
; GCN: s_endpgm
define amdgpu_kernel void @local_v2i64_store(<2 x i64> addrspace(3)* %out) nounwind {
%gep = getelementptr <2 x i64>, <2 x i64> addrspace(3)* %out, i32 7
store <2 x i64> <i64 5678, i64 5678>, <2 x i64> addrspace(3)* %gep, align 16
ret void
}
; GCN-LABEL: {{^}}local_v2i64_store_0_offset:
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN-NOT: add
; SI: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset1:1{{$}}
; CIPLUS: ds_write_b128 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]$}}
; GCN: s_endpgm
define amdgpu_kernel void @local_v2i64_store_0_offset(<2 x i64> addrspace(3)* %out) nounwind {
store <2 x i64> <i64 1234, i64 1234>, <2 x i64> addrspace(3)* %out, align 16
ret void
}
; GCN-LABEL: {{^}}local_v4i64_store:
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN-NOT: add
; SI-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:30 offset1:31
; SI-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:28 offset1:29
; CIPLUS-DAG: ds_write_b128 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:224{{$}}
; CIPLUS-DAG: ds_write_b128 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:240{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @local_v4i64_store(<4 x i64> addrspace(3)* %out) nounwind {
%gep = getelementptr <4 x i64>, <4 x i64> addrspace(3)* %out, i32 7
store <4 x i64> <i64 5678, i64 5678, i64 5678, i64 5678>, <4 x i64> addrspace(3)* %gep, align 16
ret void
}
; GCN-LABEL: {{^}}local_v4i64_store_0_offset:
; SICIVI: s_mov_b32 m0
; GFX9-NOT: m0
; GCN-NOT: add
; SI-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:2 offset1:3
; SI-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset1:1
; CIPLUS-DAG: ds_write_b128 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]$}}
; CIPLUS-DAG: ds_write_b128 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:16{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @local_v4i64_store_0_offset(<4 x i64> addrspace(3)* %out) nounwind {
store <4 x i64> <i64 1234, i64 1234, i64 1234, i64 1234>, <4 x i64> addrspace(3)* %out, align 16
ret void
}