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
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
|
; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=bonaire -enable-amdgpu-aa=0 -verify-machineinstrs -enable-misched -enable-aa-sched-mi < %s | FileCheck -enable-var-scope -check-prefixes=GCN,CI %s
; RUN: llc -amdgpu-scalarize-global-loads=false -mtriple=amdgcn -mcpu=gfx900 -enable-amdgpu-aa=0 -verify-machineinstrs -enable-misched -enable-aa-sched-mi < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9 %s
%struct.lds = type { [64 x ptr], [16 x i8] }
@stored_lds_struct = addrspace(3) global %struct.lds undef, align 16
@stored_lds_ptr = addrspace(3) global ptr addrspace(3) undef, align 4
@stored_constant_ptr = addrspace(3) global ptr addrspace(4) undef, align 8
@stored_global_ptr = addrspace(3) global ptr addrspace(1) undef, align 8
; GCN-LABEL: {{^}}no_reorder_flat_load_local_store_local_load:
; GCN: flat_load_dwordx4
; GCN: ds_write_b128 {{v[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:512
; GCN: ds_read2_b32 {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}} offset0:129 offset1:130
define amdgpu_kernel void @no_reorder_flat_load_local_store_local_load(ptr addrspace(3) %out, ptr %fptr) #0 {
%ptr1 = getelementptr %struct.lds, ptr addrspace(3) @stored_lds_struct, i32 0, i32 1
%ptr2 = getelementptr %struct.lds, ptr addrspace(3) @stored_lds_struct, i32 0, i32 1, i32 4
call void @llvm.memcpy.p3.p0(ptr addrspace(3) align 16 %ptr1, ptr align 8 %fptr, i64 16, i1 false)
%vector_load = load <2 x i32>, ptr addrspace(3) %ptr2, align 4
store <2 x i32> %vector_load, ptr addrspace(3) %out, align 4
ret void
}
; GCN-LABEL: {{^}}reorder_local_load_global_store_local_load:
; CI: ds_read2_b32 {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}} offset0:1 offset1:3
; CI: buffer_store_dword
; GFX9: ds_read2_b32 {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}} offset0:1 offset1:3
; GFX9: global_store_dword
; GFX9: global_store_dword
define amdgpu_kernel void @reorder_local_load_global_store_local_load(ptr addrspace(1) %out, ptr addrspace(1) %gptr) #0 {
%ptr0 = load ptr addrspace(3), ptr addrspace(3) @stored_lds_ptr, align 4
%ptr1 = getelementptr inbounds i32, ptr addrspace(3) %ptr0, i32 1
%ptr2 = getelementptr inbounds i32, ptr addrspace(3) %ptr0, i32 3
%tmp1 = load i32, ptr addrspace(3) %ptr1, align 4
store i32 99, ptr addrspace(1) %gptr, align 4
%tmp2 = load i32, ptr addrspace(3) %ptr2, align 4
%add = add nsw i32 %tmp1, %tmp2
store i32 %add, ptr addrspace(1) %out, align 4
ret void
}
; GCN-LABEL: {{^}}no_reorder_local_load_volatile_global_store_local_load:
; CI: ds_read_b32 {{v[0-9]+}}, {{v[0-9]+}} offset:4
; CI: buffer_store_dword
; CI: ds_read_b32 {{v[0-9]+}}, {{v[0-9]+}} offset:12
; GFX9: ds_read_b32 {{v[0-9]+}}, {{v[0-9]+}} offset:4
; GFX9: global_store_dword
; GFX9: ds_read_b32 {{v[0-9]+}}, {{v[0-9]+}} offset:12
define amdgpu_kernel void @no_reorder_local_load_volatile_global_store_local_load(ptr addrspace(1) %out, ptr addrspace(1) %gptr) #0 {
%ptr0 = load ptr addrspace(3), ptr addrspace(3) @stored_lds_ptr, align 4
%ptr1 = getelementptr inbounds i32, ptr addrspace(3) %ptr0, i32 1
%ptr2 = getelementptr inbounds i32, ptr addrspace(3) %ptr0, i32 3
%tmp1 = load i32, ptr addrspace(3) %ptr1, align 4
store volatile i32 99, ptr addrspace(1) %gptr, align 4
%tmp2 = load i32, ptr addrspace(3) %ptr2, align 4
%add = add nsw i32 %tmp1, %tmp2
store i32 %add, ptr addrspace(1) %out, align 4
ret void
}
; GCN-LABEL: {{^}}no_reorder_barrier_local_load_global_store_local_load:
; CI: ds_read_b32 {{v[0-9]+}}, {{v[0-9]+}} offset:4
; CI: ds_read_b32 {{v[0-9]+}}, {{v[0-9]+}} offset:12
; CI: buffer_store_dword
; GFX9-DAG: global_store_dword
; GFX9-DAG: ds_read_b32 {{v[0-9]+}}, {{v[0-9]+}} offset:4
; GFX9: s_barrier
; GFX9-DAG: ds_read_b32 {{v[0-9]+}}, {{v[0-9]+}} offset:12
; GFX9-DAG: global_store_dword
define amdgpu_kernel void @no_reorder_barrier_local_load_global_store_local_load(ptr addrspace(1) %out, ptr addrspace(1) %gptr) #0 {
%ptr0 = load ptr addrspace(3), ptr addrspace(3) @stored_lds_ptr, align 4
%ptr1 = getelementptr inbounds i32, ptr addrspace(3) %ptr0, i32 1
%ptr2 = getelementptr inbounds i32, ptr addrspace(3) %ptr0, i32 3
%tmp1 = load i32, ptr addrspace(3) %ptr1, align 4
store i32 99, ptr addrspace(1) %gptr, align 4
call void @llvm.amdgcn.s.barrier() #1
%tmp2 = load i32, ptr addrspace(3) %ptr2, align 4
%add = add nsw i32 %tmp1, %tmp2
store i32 %add, ptr addrspace(1) %out, align 4
ret void
}
; GCN-LABEL: {{^}}reorder_constant_load_global_store_constant_load:
; GCN-DAG: v_readfirstlane_b32 s[[PTR_LO:[0-9]+]], v{{[0-9]+}}
; GCN: v_readfirstlane_b32 s[[PTR_HI:[0-9]+]], v{{[0-9]+}}
; CI: s_load_dword s{{[0-9]+}}, s[[[PTR_LO]]:[[PTR_HI]]], 0x1
; CI: buffer_store_dword
; CI: s_load_dword s{{[0-9]+}}, s[[[PTR_LO]]:[[PTR_HI]]], 0x3
; GFX9: s_load_dword s{{[0-9]+}}, s[[[PTR_LO]]:[[PTR_HI]]], 0x4
; GFX9: global_store_dword
; GFX9: s_load_dword s{{[0-9]+}}, s[[[PTR_LO]]:[[PTR_HI]]], 0xc
; CI: buffer_store_dword
; GFX9: global_store_dword
define amdgpu_kernel void @reorder_constant_load_global_store_constant_load(ptr addrspace(1) %out, ptr addrspace(1) %gptr) #0 {
%ptr0 = load ptr addrspace(4), ptr addrspace(3) @stored_constant_ptr, align 8
%ptr1 = getelementptr inbounds i32, ptr addrspace(4) %ptr0, i64 1
%ptr2 = getelementptr inbounds i32, ptr addrspace(4) %ptr0, i64 3
%tmp1 = load i32, ptr addrspace(4) %ptr1, align 4
store i32 99, ptr addrspace(1) %gptr, align 4
%tmp2 = load i32, ptr addrspace(4) %ptr2, align 4
%add = add nsw i32 %tmp1, %tmp2
store i32 %add, ptr addrspace(1) %out, align 4
ret void
}
; GCN-LABEL: {{^}}reorder_constant_load_local_store_constant_load:
; GCN: v_readfirstlane_b32 s[[PTR_LO:[0-9]+]], v{{[0-9]+}}
; GCN: v_readfirstlane_b32 s[[PTR_HI:[0-9]+]], v{{[0-9]+}}
; CI-DAG: s_load_dword s{{[0-9]+}}, s[[[PTR_LO]]:[[PTR_HI]]], 0x1
; CI-DAG: s_load_dword s{{[0-9]+}}, s[[[PTR_LO]]:[[PTR_HI]]], 0x3
; GFX9-DAG: s_load_dword s{{[0-9]+}}, s[[[PTR_LO]]:[[PTR_HI]]], 0x4
; GFX9-DAG: s_load_dword s{{[0-9]+}}, s[[[PTR_LO]]:[[PTR_HI]]], 0xc
; GCN-DAG: ds_write_b32
; CI: buffer_store_dword
; GFX9: global_store_dword
define amdgpu_kernel void @reorder_constant_load_local_store_constant_load(ptr addrspace(1) %out, ptr addrspace(3) %lptr) #0 {
%ptr0 = load ptr addrspace(4), ptr addrspace(3) @stored_constant_ptr, align 8
%ptr1 = getelementptr inbounds i32, ptr addrspace(4) %ptr0, i64 1
%ptr2 = getelementptr inbounds i32, ptr addrspace(4) %ptr0, i64 3
%tmp1 = load i32, ptr addrspace(4) %ptr1, align 4
store i32 99, ptr addrspace(3) %lptr, align 4
%tmp2 = load i32, ptr addrspace(4) %ptr2, align 4
%add = add nsw i32 %tmp1, %tmp2
store i32 %add, ptr addrspace(1) %out, align 4
ret void
}
; GCN-LABEL: {{^}}reorder_smrd_load_local_store_smrd_load:
; GCN: s_load_dword
; GCN: s_load_dword
; GCN: s_load_dword
; GCN: ds_write_b32
; CI: buffer_store_dword
; GFX9: global_store_dword
define amdgpu_kernel void @reorder_smrd_load_local_store_smrd_load(ptr addrspace(1) %out, ptr addrspace(3) noalias %lptr, ptr addrspace(4) %ptr0) #0 {
%ptr1 = getelementptr inbounds i32, ptr addrspace(4) %ptr0, i64 1
%ptr2 = getelementptr inbounds i32, ptr addrspace(4) %ptr0, i64 2
%tmp1 = load i32, ptr addrspace(4) %ptr1, align 4
store i32 99, ptr addrspace(3) %lptr, align 4
%tmp2 = load i32, ptr addrspace(4) %ptr2, align 4
%add = add nsw i32 %tmp1, %tmp2
store i32 %add, ptr addrspace(1) %out, align 4
ret void
}
; GCN-LABEL: {{^}}reorder_global_load_local_store_global_load:
; CI: buffer_load_dword
; CI: buffer_load_dword
; CI: ds_write_b32
; CI: buffer_store_dword
; GFX9: global_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}} offset:4
; GFX9: global_load_dword v{{[0-9]+}}, v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}} offset:12
; GFX9: ds_write_b32
define amdgpu_kernel void @reorder_global_load_local_store_global_load(ptr addrspace(1) %out, ptr addrspace(3) %lptr, ptr addrspace(1) %ptr0) #0 {
%ptr1 = getelementptr inbounds i32, ptr addrspace(1) %ptr0, i64 1
%ptr2 = getelementptr inbounds i32, ptr addrspace(1) %ptr0, i64 3
%tmp1 = load i32, ptr addrspace(1) %ptr1, align 4
store i32 99, ptr addrspace(3) %lptr, align 4
%tmp2 = load i32, ptr addrspace(1) %ptr2, align 4
%add = add nsw i32 %tmp1, %tmp2
store i32 %add, ptr addrspace(1) %out, align 4
ret void
}
; GCN-LABEL: {{^}}reorder_local_offsets:
; GCN: ds_read2_b32 {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}} offset0:100 offset1:102
; GCN-DAG: ds_write2_b32 {{v[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} offset0:3 offset1:100
; GCN-DAG: ds_write_b32 {{v[0-9]+}}, {{v[0-9]+}} offset:408
; CI: buffer_store_dword
; GFX9: global_store_dword
; GCN: s_endpgm
define amdgpu_kernel void @reorder_local_offsets(ptr addrspace(1) nocapture %out, ptr addrspace(1) noalias nocapture readnone %gptr, ptr addrspace(3) noalias nocapture %ptr0) #0 {
%ptr1 = getelementptr inbounds i32, ptr addrspace(3) %ptr0, i32 3
%ptr2 = getelementptr inbounds i32, ptr addrspace(3) %ptr0, i32 100
%ptr3 = getelementptr inbounds i32, ptr addrspace(3) %ptr0, i32 102
store i32 123, ptr addrspace(3) %ptr1, align 4
%tmp1 = load i32, ptr addrspace(3) %ptr2, align 4
%tmp2 = load i32, ptr addrspace(3) %ptr3, align 4
store i32 123, ptr addrspace(3) %ptr2, align 4
%tmp3 = load i32, ptr addrspace(3) %ptr1, align 4
store i32 789, ptr addrspace(3) %ptr3, align 4
%add.0 = add nsw i32 %tmp2, %tmp1
%add.1 = add nsw i32 %add.0, %tmp3
store i32 %add.1, ptr addrspace(1) %out, align 4
ret void
}
; GCN-LABEL: {{^}}reorder_global_offsets:
; CI-DAG: buffer_load_dword {{v[0-9]+}}, off, {{s\[[0-9]+:[0-9]+\]}}, 0 offset:400
; CI-DAG: buffer_load_dword {{v[0-9]+}}, off, {{s\[[0-9]+:[0-9]+\]}}, 0 offset:408
; CI-DAG: buffer_store_dword {{v[0-9]+}}, off, {{s\[[0-9]+:[0-9]+\]}}, 0 offset:12
; CI-DAG: buffer_store_dword {{v[0-9]+}}, off, {{s\[[0-9]+:[0-9]+\]}}, 0 offset:400
; CI-DAG: buffer_store_dword {{v[0-9]+}}, off, {{s\[[0-9]+:[0-9]+\]}}, 0 offset:408
; CI: buffer_store_dword
; CI: s_endpgm
; GFX9-DAG: global_load_dword {{v[0-9]+}}, v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}} offset:400
; GFX9-DAG: global_load_dword {{v[0-9]+}}, v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}} offset:408
; GFX9-DAG: global_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}} offset:12
; GFX9-DAG: global_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}} offset:400
; GFX9-DAG: global_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}} offset:408
; GFX9: global_store_dword
; GFX9: s_endpgm
define amdgpu_kernel void @reorder_global_offsets(ptr addrspace(1) nocapture %out, ptr addrspace(1) noalias nocapture readnone %gptr, ptr addrspace(1) noalias nocapture %ptr0) #0 {
%ptr1 = getelementptr inbounds i32, ptr addrspace(1) %ptr0, i32 3
%ptr2 = getelementptr inbounds i32, ptr addrspace(1) %ptr0, i32 100
%ptr3 = getelementptr inbounds i32, ptr addrspace(1) %ptr0, i32 102
store i32 123, ptr addrspace(1) %ptr1, align 4
%tmp1 = load i32, ptr addrspace(1) %ptr2, align 4
%tmp2 = load i32, ptr addrspace(1) %ptr3, align 4
store i32 123, ptr addrspace(1) %ptr2, align 4
%tmp3 = load i32, ptr addrspace(1) %ptr1, align 4
store i32 789, ptr addrspace(1) %ptr3, align 4
%add.0 = add nsw i32 %tmp2, %tmp1
%add.1 = add nsw i32 %add.0, %tmp3
store i32 %add.1, ptr addrspace(1) %out, align 4
ret void
}
; GCN-LABEL: {{^}}reorder_global_offsets_addr64_soffset0:
; CI: buffer_load_dword v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:12{{$}}
; CI-NEXT: buffer_load_dword v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:28{{$}}
; CI-NEXT: buffer_load_dword v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:44{{$}}
; CI: v_mov_b32
; CI: v_mov_b32
; CI-DAG: v_add_i32
; CI-DAG: v_add_i32
; CI-DAG: buffer_store_dword v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
; CI-DAG: buffer_store_dword v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:20{{$}}
; CI-DAG: buffer_store_dword v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:36{{$}}
; CI: buffer_store_dword v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:52{{$}}
; GFX9: global_load_dword {{v[0-9]+}}, v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}} offset:12
; GFX9: global_load_dword {{v[0-9]+}}, v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}} offset:28
; GFX9: global_load_dword {{v[0-9]+}}, v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}} offset:44
; GFX9: global_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]$}}
; GFX9: global_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}} offset:20
; GFX9: global_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}} offset:36
; GFX9: global_store_dword v{{[0-9]+}}, v{{[0-9]+}}, s{{\[[0-9]+:[0-9]+\]}} offset:52
define amdgpu_kernel void @reorder_global_offsets_addr64_soffset0(ptr addrspace(1) noalias nocapture %ptr.base) #0 {
%id = call i32 @llvm.amdgcn.workitem.id.x()
%id.ext = sext i32 %id to i64
%ptr0 = getelementptr inbounds i32, ptr addrspace(1) %ptr.base, i64 %id.ext
%ptr1 = getelementptr inbounds i32, ptr addrspace(1) %ptr0, i32 3
%ptr2 = getelementptr inbounds i32, ptr addrspace(1) %ptr0, i32 5
%ptr3 = getelementptr inbounds i32, ptr addrspace(1) %ptr0, i32 7
%ptr4 = getelementptr inbounds i32, ptr addrspace(1) %ptr0, i32 9
%ptr5 = getelementptr inbounds i32, ptr addrspace(1) %ptr0, i32 11
%ptr6 = getelementptr inbounds i32, ptr addrspace(1) %ptr0, i32 13
store i32 789, ptr addrspace(1) %ptr0, align 4
%tmp1 = load i32, ptr addrspace(1) %ptr1, align 4
store i32 123, ptr addrspace(1) %ptr2, align 4
%tmp2 = load i32, ptr addrspace(1) %ptr3, align 4
%add.0 = add nsw i32 %tmp1, %tmp2
store i32 %add.0, ptr addrspace(1) %ptr4, align 4
%tmp3 = load i32, ptr addrspace(1) %ptr5, align 4
%add.1 = add nsw i32 %add.0, %tmp3
store i32 %add.1, ptr addrspace(1) %ptr6, align 4
ret void
}
; GCN-LABEL: {{^}}reorder_local_load_tbuffer_store_local_load:
; GCN: ds_read2_b32 {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}} offset0:1 offset1:2
; GCN: tbuffer_store_format
define amdgpu_vs void @reorder_local_load_tbuffer_store_local_load(ptr addrspace(1) %out, i32 %a1, i32 %vaddr) #0 {
%ptr0 = load ptr addrspace(3), ptr addrspace(3) @stored_lds_ptr, align 4
%ptr1 = getelementptr inbounds i32, ptr addrspace(3) %ptr0, i32 1
%ptr2 = getelementptr inbounds i32, ptr addrspace(3) %ptr0, i32 2
%tmp1 = load i32, ptr addrspace(3) %ptr1, align 4
%vdata = insertelement <4 x i32> undef, i32 %a1, i32 0
%vaddr.add = add i32 %vaddr, 32
call void @llvm.amdgcn.struct.ptr.tbuffer.store.v4i32(<4 x i32> %vdata, ptr addrspace(8) undef, i32 %vaddr.add, i32 0, i32 0, i32 228, i32 3)
%tmp2 = load i32, ptr addrspace(3) %ptr2, align 4
%add = add nsw i32 %tmp1, %tmp2
store i32 %add, ptr addrspace(1) %out, align 4
ret void
}
declare void @llvm.memcpy.p3.p0(ptr addrspace(3), ptr, i64, i1)
declare void @llvm.amdgcn.s.barrier() #1
declare i32 @llvm.amdgcn.workitem.id.x() #2
declare void @llvm.amdgcn.struct.ptr.tbuffer.store.v4i32(<4 x i32>, ptr addrspace(8), i32, i32, i32, i32 immarg, i32 immarg) #3
attributes #0 = { nounwind }
attributes #1 = { convergent nounwind willreturn }
attributes #2 = { nounwind readnone speculatable willreturn }
attributes #3 = { nounwind willreturn writeonly }
|
