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
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -misched-cluster=0 < %s | FileCheck -check-prefix=GCN %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -misched-cluster=0 -amdgpu-igrouplp-exact-solver-max-branches=250000 < %s | FileCheck -check-prefix=EXACTCUTOFF %s
define amdgpu_kernel void @test_sched_group_barrier_pipeline_WMMA_cluster(ptr addrspace(3) noalias %in, ptr addrspace(3) noalias %out) #0 {
; GCN-LABEL: test_sched_group_barrier_pipeline_WMMA_cluster:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
; GCN-NEXT: v_lshlrev_b32_e32 v0, 5, v0
; GCN-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GCN-NEXT: v_and_b32_e32 v40, 0x7fe0, v0
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: v_add_nc_u32_e32 v32, s0, v40
; GCN-NEXT: v_dual_mov_b32 v81, s1 :: v_dual_add_nc_u32 v80, s1, v40
; GCN-NEXT: ds_load_b128 v[4:7], v32 offset:16
; GCN-NEXT: ds_load_b128 v[12:15], v32 offset:2064
; GCN-NEXT: ds_load_b128 v[20:23], v32 offset:6160
; GCN-NEXT: ds_load_b128 v[28:31], v32 offset:12304
; GCN-NEXT: ds_load_b128 v[36:39], v32 offset:20496
; GCN-NEXT: ds_load_b128 v[0:3], v32
; GCN-NEXT: ds_load_b128 v[8:11], v32 offset:2048
; GCN-NEXT: ds_load_b128 v[16:19], v32 offset:6144
; GCN-NEXT: ds_load_b128 v[24:27], v32 offset:12288
; GCN-NEXT: ds_load_b128 v[32:35], v32 offset:20480
; GCN-NEXT: ; sched_group_barrier mask(0x00000100) size(10) SyncID(0)
; GCN-NEXT: s_waitcnt lgkmcnt(4)
; GCN-NEXT: v_mov_b32_e32 v47, v7
; GCN-NEXT: s_waitcnt lgkmcnt(3)
; GCN-NEXT: v_mov_b32_e32 v55, v15
; GCN-NEXT: s_waitcnt lgkmcnt(2)
; GCN-NEXT: v_mov_b32_e32 v63, v23
; GCN-NEXT: s_waitcnt lgkmcnt(1)
; GCN-NEXT: v_mov_b32_e32 v71, v31
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: v_dual_mov_b32 v79, v39 :: v_dual_mov_b32 v46, v6
; GCN-NEXT: v_dual_mov_b32 v45, v5 :: v_dual_mov_b32 v44, v4
; GCN-NEXT: v_dual_mov_b32 v43, v3 :: v_dual_mov_b32 v42, v2
; GCN-NEXT: v_dual_mov_b32 v41, v1 :: v_dual_mov_b32 v40, v0
; GCN-NEXT: v_dual_mov_b32 v54, v14 :: v_dual_mov_b32 v53, v13
; GCN-NEXT: v_dual_mov_b32 v52, v12 :: v_dual_mov_b32 v51, v11
; GCN-NEXT: v_dual_mov_b32 v50, v10 :: v_dual_mov_b32 v49, v9
; GCN-NEXT: v_mov_b32_e32 v48, v8
; GCN-NEXT: v_dual_mov_b32 v62, v22 :: v_dual_mov_b32 v61, v21
; GCN-NEXT: v_dual_mov_b32 v60, v20 :: v_dual_mov_b32 v59, v19
; GCN-NEXT: v_dual_mov_b32 v58, v18 :: v_dual_mov_b32 v57, v17
; GCN-NEXT: v_mov_b32_e32 v56, v16
; GCN-NEXT: v_dual_mov_b32 v70, v30 :: v_dual_mov_b32 v69, v29
; GCN-NEXT: v_dual_mov_b32 v68, v28 :: v_dual_mov_b32 v67, v27
; GCN-NEXT: v_dual_mov_b32 v66, v26 :: v_dual_mov_b32 v65, v25
; GCN-NEXT: v_mov_b32_e32 v64, v24
; GCN-NEXT: v_dual_mov_b32 v78, v38 :: v_dual_mov_b32 v77, v37
; GCN-NEXT: v_dual_mov_b32 v76, v36 :: v_dual_mov_b32 v75, v35
; GCN-NEXT: v_dual_mov_b32 v74, v34 :: v_dual_mov_b32 v73, v33
; GCN-NEXT: v_mov_b32_e32 v72, v32
; GCN-NEXT: v_wmma_f16_16x16x16_f16 v[40:47], v[0:7], v[0:7], v[40:47]
; GCN-NEXT: v_wmma_f16_16x16x16_f16 v[48:55], v[8:15], v[8:15], v[48:55]
; GCN-NEXT: v_wmma_f16_16x16x16_f16 v[56:63], v[16:23], v[16:23], v[56:63]
; GCN-NEXT: v_wmma_f16_16x16x16_f16 v[64:71], v[24:31], v[24:31], v[64:71]
; GCN-NEXT: v_wmma_f16_16x16x16_f16 v[72:79], v[32:39], v[32:39], v[72:79]
; GCN-NEXT: ds_store_b128 v80, v[44:47] offset:16
; GCN-NEXT: ds_store_b128 v80, v[40:43]
; GCN-NEXT: ds_store_b128 v81, v[52:55] offset:2064
; GCN-NEXT: ds_store_b128 v81, v[48:51] offset:2048
; GCN-NEXT: ds_store_b128 v81, v[60:63] offset:4112
; GCN-NEXT: ds_store_b128 v81, v[56:59] offset:4096
; GCN-NEXT: ds_store_b128 v81, v[68:71] offset:6160
; GCN-NEXT: ds_store_b128 v81, v[64:67] offset:6144
; GCN-NEXT: ds_store_b128 v81, v[76:79] offset:8208
; GCN-NEXT: ds_store_b128 v81, v[72:75] offset:8192
; GCN-NEXT: ; sched_group_barrier mask(0x00000008) size(5) SyncID(0)
; GCN-NEXT: ; sched_group_barrier mask(0x00000200) size(10) SyncID(0)
; GCN-NEXT: s_endpgm
;
; EXACTCUTOFF-LABEL: test_sched_group_barrier_pipeline_WMMA_cluster:
; EXACTCUTOFF: ; %bb.0: ; %entry
; EXACTCUTOFF-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
; EXACTCUTOFF-NEXT: v_lshlrev_b32_e32 v0, 5, v0
; EXACTCUTOFF-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; EXACTCUTOFF-NEXT: v_and_b32_e32 v40, 0x7fe0, v0
; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(0)
; EXACTCUTOFF-NEXT: v_add_nc_u32_e32 v32, s0, v40
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v81, s1 :: v_dual_add_nc_u32 v80, s1, v40
; EXACTCUTOFF-NEXT: ds_load_b128 v[4:7], v32 offset:16
; EXACTCUTOFF-NEXT: ds_load_b128 v[12:15], v32 offset:2064
; EXACTCUTOFF-NEXT: ds_load_b128 v[20:23], v32 offset:6160
; EXACTCUTOFF-NEXT: ds_load_b128 v[28:31], v32 offset:12304
; EXACTCUTOFF-NEXT: ds_load_b128 v[36:39], v32 offset:20496
; EXACTCUTOFF-NEXT: ds_load_b128 v[0:3], v32
; EXACTCUTOFF-NEXT: ds_load_b128 v[8:11], v32 offset:2048
; EXACTCUTOFF-NEXT: ds_load_b128 v[16:19], v32 offset:6144
; EXACTCUTOFF-NEXT: ds_load_b128 v[24:27], v32 offset:12288
; EXACTCUTOFF-NEXT: ds_load_b128 v[32:35], v32 offset:20480
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000100) size(10) SyncID(0)
; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(4)
; EXACTCUTOFF-NEXT: v_mov_b32_e32 v47, v7
; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(3)
; EXACTCUTOFF-NEXT: v_mov_b32_e32 v55, v15
; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(2)
; EXACTCUTOFF-NEXT: v_mov_b32_e32 v63, v23
; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(1)
; EXACTCUTOFF-NEXT: v_mov_b32_e32 v71, v31
; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(0)
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v79, v39 :: v_dual_mov_b32 v46, v6
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v45, v5 :: v_dual_mov_b32 v44, v4
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v43, v3 :: v_dual_mov_b32 v42, v2
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v41, v1 :: v_dual_mov_b32 v40, v0
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v54, v14 :: v_dual_mov_b32 v53, v13
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v52, v12 :: v_dual_mov_b32 v51, v11
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v50, v10 :: v_dual_mov_b32 v49, v9
; EXACTCUTOFF-NEXT: v_mov_b32_e32 v48, v8
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v62, v22 :: v_dual_mov_b32 v61, v21
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v60, v20 :: v_dual_mov_b32 v59, v19
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v58, v18 :: v_dual_mov_b32 v57, v17
; EXACTCUTOFF-NEXT: v_mov_b32_e32 v56, v16
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v70, v30 :: v_dual_mov_b32 v69, v29
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v68, v28 :: v_dual_mov_b32 v67, v27
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v66, v26 :: v_dual_mov_b32 v65, v25
; EXACTCUTOFF-NEXT: v_mov_b32_e32 v64, v24
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v78, v38 :: v_dual_mov_b32 v77, v37
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v76, v36 :: v_dual_mov_b32 v75, v35
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v74, v34 :: v_dual_mov_b32 v73, v33
; EXACTCUTOFF-NEXT: v_mov_b32_e32 v72, v32
; EXACTCUTOFF-NEXT: v_wmma_f16_16x16x16_f16 v[40:47], v[0:7], v[0:7], v[40:47]
; EXACTCUTOFF-NEXT: v_wmma_f16_16x16x16_f16 v[48:55], v[8:15], v[8:15], v[48:55]
; EXACTCUTOFF-NEXT: v_wmma_f16_16x16x16_f16 v[56:63], v[16:23], v[16:23], v[56:63]
; EXACTCUTOFF-NEXT: v_wmma_f16_16x16x16_f16 v[64:71], v[24:31], v[24:31], v[64:71]
; EXACTCUTOFF-NEXT: v_wmma_f16_16x16x16_f16 v[72:79], v[32:39], v[32:39], v[72:79]
; EXACTCUTOFF-NEXT: ds_store_b128 v80, v[44:47] offset:16
; EXACTCUTOFF-NEXT: ds_store_b128 v80, v[40:43]
; EXACTCUTOFF-NEXT: ds_store_b128 v81, v[52:55] offset:2064
; EXACTCUTOFF-NEXT: ds_store_b128 v81, v[48:51] offset:2048
; EXACTCUTOFF-NEXT: ds_store_b128 v81, v[60:63] offset:4112
; EXACTCUTOFF-NEXT: ds_store_b128 v81, v[56:59] offset:4096
; EXACTCUTOFF-NEXT: ds_store_b128 v81, v[68:71] offset:6160
; EXACTCUTOFF-NEXT: ds_store_b128 v81, v[64:67] offset:6144
; EXACTCUTOFF-NEXT: ds_store_b128 v81, v[76:79] offset:8208
; EXACTCUTOFF-NEXT: ds_store_b128 v81, v[72:75] offset:8192
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000008) size(5) SyncID(0)
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000200) size(10) SyncID(0)
; EXACTCUTOFF-NEXT: s_endpgm
entry:
%idx = call i32 @llvm.amdgcn.workitem.id.x()
%load.0.addr = getelementptr <16 x half>, ptr addrspace(3) %in, i32 %idx
%load.0 = load <16 x half>, ptr addrspace(3) %load.0.addr
%load.1.addr = getelementptr <16 x half>, ptr addrspace(3) %load.0.addr, i32 64
%load.1 = load <16 x half>, ptr addrspace(3) %load.1.addr
%load.2.addr = getelementptr <16 x half>, ptr addrspace(3) %load.1.addr, i32 128
%load.2 = load <16 x half>, ptr addrspace(3) %load.2.addr
%load.3.addr = getelementptr <16 x half>, ptr addrspace(3) %load.2.addr, i32 192
%load.3 = load <16 x half>, ptr addrspace(3) %load.3.addr
%load.4.addr = getelementptr <16 x half>, ptr addrspace(3) %load.3.addr, i32 256
%load.4 = load <16 x half>, ptr addrspace(3) %load.4.addr
%mai.0 = call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16(<16 x half> %load.0, <16 x half> %load.0, <16 x half> %load.0, i1 0)
%mai.1 = call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16(<16 x half> %load.1, <16 x half> %load.1, <16 x half> %load.1, i1 0)
%mai.2 = call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16(<16 x half> %load.2, <16 x half> %load.2, <16 x half> %load.2, i1 0)
%mai.3 = call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16(<16 x half> %load.3, <16 x half> %load.3, <16 x half> %load.3, i1 0)
%mai.4 = call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16(<16 x half> %load.4, <16 x half> %load.4, <16 x half> %load.4, i1 0)
%store.0.addr = getelementptr <16 x half>, ptr addrspace(3) %out, i32 %idx
store <16 x half> %mai.0, ptr addrspace(3) %store.0.addr
%store.1.addr = getelementptr <16 x half>, ptr addrspace(3) %out, i32 64
store <16 x half> %mai.1, ptr addrspace(3) %store.1.addr
%store.2.addr = getelementptr <16 x half>, ptr addrspace(3) %out, i32 128
store <16 x half> %mai.2, ptr addrspace(3) %store.2.addr
%store.3.addr = getelementptr <16 x half>, ptr addrspace(3) %out, i32 192
store <16 x half> %mai.3, ptr addrspace(3) %store.3.addr
%store.4.addr = getelementptr <16 x half>, ptr addrspace(3) %out, i32 256
store <16 x half> %mai.4, ptr addrspace(3) %store.4.addr
; 10 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 10, i32 0)
; 5 WMMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 5, i32 0)
; 10 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 10, i32 0)
ret void
}
define amdgpu_kernel void @test_sched_group_barrier_pipeline_WMMA_interleave(ptr addrspace(3) noalias %in, ptr addrspace(3) noalias %out) #0 {
; GCN-LABEL: test_sched_group_barrier_pipeline_WMMA_interleave:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
; GCN-NEXT: v_lshlrev_b32_e32 v0, 5, v0
; GCN-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; GCN-NEXT: v_and_b32_e32 v16, 0x7fe0, v0
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: v_add_nc_u32_e32 v17, s0, v16
; GCN-NEXT: v_add_nc_u32_e32 v16, s1, v16
; GCN-NEXT: ds_load_b128 v[4:7], v17 offset:16
; GCN-NEXT: ds_load_b128 v[0:3], v17
; GCN-NEXT: ; sched_group_barrier mask(0x00000100) size(2) SyncID(0)
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: v_dual_mov_b32 v15, v7 :: v_dual_mov_b32 v14, v6
; GCN-NEXT: v_dual_mov_b32 v13, v5 :: v_dual_mov_b32 v12, v4
; GCN-NEXT: v_dual_mov_b32 v11, v3 :: v_dual_mov_b32 v10, v2
; GCN-NEXT: v_dual_mov_b32 v9, v1 :: v_dual_mov_b32 v8, v0
; GCN-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GCN-NEXT: v_wmma_f16_16x16x16_f16 v[8:15], v[0:7], v[0:7], v[8:15]
; GCN-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-NEXT: ds_store_b128 v16, v[12:15] offset:16
; GCN-NEXT: ds_store_b128 v16, v[8:11]
; GCN-NEXT: ds_load_b128 v[4:7], v17 offset:2064
; GCN-NEXT: ds_load_b128 v[0:3], v17 offset:2048
; GCN-NEXT: v_mov_b32_e32 v16, s1
; GCN-NEXT: ; sched_group_barrier mask(0x00000200) size(2) SyncID(0)
; GCN-NEXT: ; sched_group_barrier mask(0x00000100) size(2) SyncID(0)
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: v_dual_mov_b32 v15, v7 :: v_dual_mov_b32 v14, v6
; GCN-NEXT: v_dual_mov_b32 v13, v5 :: v_dual_mov_b32 v12, v4
; GCN-NEXT: v_dual_mov_b32 v11, v3 :: v_dual_mov_b32 v10, v2
; GCN-NEXT: v_dual_mov_b32 v9, v1 :: v_dual_mov_b32 v8, v0
; GCN-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GCN-NEXT: v_wmma_f16_16x16x16_f16 v[8:15], v[0:7], v[0:7], v[8:15]
; GCN-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-NEXT: ds_store_b128 v16, v[12:15] offset:2064
; GCN-NEXT: ds_store_b128 v16, v[8:11] offset:2048
; GCN-NEXT: ds_load_b128 v[4:7], v17 offset:6160
; GCN-NEXT: ds_load_b128 v[0:3], v17 offset:6144
; GCN-NEXT: ; sched_group_barrier mask(0x00000200) size(2) SyncID(0)
; GCN-NEXT: ; sched_group_barrier mask(0x00000100) size(2) SyncID(0)
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: v_dual_mov_b32 v15, v7 :: v_dual_mov_b32 v14, v6
; GCN-NEXT: v_dual_mov_b32 v13, v5 :: v_dual_mov_b32 v12, v4
; GCN-NEXT: v_dual_mov_b32 v11, v3 :: v_dual_mov_b32 v10, v2
; GCN-NEXT: v_dual_mov_b32 v9, v1 :: v_dual_mov_b32 v8, v0
; GCN-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GCN-NEXT: v_wmma_f16_16x16x16_f16 v[8:15], v[0:7], v[0:7], v[8:15]
; GCN-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-NEXT: ds_store_b128 v16, v[12:15] offset:4112
; GCN-NEXT: ds_store_b128 v16, v[8:11] offset:4096
; GCN-NEXT: ds_load_b128 v[4:7], v17 offset:12304
; GCN-NEXT: ds_load_b128 v[0:3], v17 offset:12288
; GCN-NEXT: ; sched_group_barrier mask(0x00000200) size(2) SyncID(0)
; GCN-NEXT: ; sched_group_barrier mask(0x00000100) size(2) SyncID(0)
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: v_dual_mov_b32 v15, v7 :: v_dual_mov_b32 v14, v6
; GCN-NEXT: v_dual_mov_b32 v13, v5 :: v_dual_mov_b32 v12, v4
; GCN-NEXT: v_dual_mov_b32 v11, v3 :: v_dual_mov_b32 v10, v2
; GCN-NEXT: v_dual_mov_b32 v9, v1 :: v_dual_mov_b32 v8, v0
; GCN-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GCN-NEXT: v_wmma_f16_16x16x16_f16 v[8:15], v[0:7], v[0:7], v[8:15]
; GCN-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-NEXT: ds_store_b128 v16, v[12:15] offset:6160
; GCN-NEXT: ds_store_b128 v16, v[8:11] offset:6144
; GCN-NEXT: ds_load_b128 v[4:7], v17 offset:20496
; GCN-NEXT: ds_load_b128 v[0:3], v17 offset:20480
; GCN-NEXT: ; sched_group_barrier mask(0x00000200) size(2) SyncID(0)
; GCN-NEXT: ; sched_group_barrier mask(0x00000100) size(2) SyncID(0)
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: v_dual_mov_b32 v15, v7 :: v_dual_mov_b32 v14, v6
; GCN-NEXT: v_dual_mov_b32 v13, v5 :: v_dual_mov_b32 v12, v4
; GCN-NEXT: v_dual_mov_b32 v11, v3 :: v_dual_mov_b32 v10, v2
; GCN-NEXT: v_dual_mov_b32 v9, v1 :: v_dual_mov_b32 v8, v0
; GCN-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GCN-NEXT: v_wmma_f16_16x16x16_f16 v[8:15], v[0:7], v[0:7], v[8:15]
; GCN-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; GCN-NEXT: ds_store_b128 v16, v[12:15] offset:8208
; GCN-NEXT: ds_store_b128 v16, v[8:11] offset:8192
; GCN-NEXT: ; sched_group_barrier mask(0x00000200) size(2) SyncID(0)
; GCN-NEXT: s_endpgm
;
; EXACTCUTOFF-LABEL: test_sched_group_barrier_pipeline_WMMA_interleave:
; EXACTCUTOFF: ; %bb.0: ; %entry
; EXACTCUTOFF-NEXT: s_load_b64 s[0:1], s[4:5], 0x24
; EXACTCUTOFF-NEXT: v_lshlrev_b32_e32 v0, 5, v0
; EXACTCUTOFF-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
; EXACTCUTOFF-NEXT: v_and_b32_e32 v16, 0x7fe0, v0
; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(0)
; EXACTCUTOFF-NEXT: v_add_nc_u32_e32 v17, s0, v16
; EXACTCUTOFF-NEXT: v_add_nc_u32_e32 v16, s1, v16
; EXACTCUTOFF-NEXT: ds_load_b128 v[4:7], v17 offset:16
; EXACTCUTOFF-NEXT: ds_load_b128 v[0:3], v17
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000100) size(2) SyncID(0)
; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(0)
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v15, v7 :: v_dual_mov_b32 v14, v6
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v13, v5 :: v_dual_mov_b32 v12, v4
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v11, v3 :: v_dual_mov_b32 v10, v2
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v9, v1 :: v_dual_mov_b32 v8, v0
; EXACTCUTOFF-NEXT: s_delay_alu instid0(VALU_DEP_1)
; EXACTCUTOFF-NEXT: v_wmma_f16_16x16x16_f16 v[8:15], v[0:7], v[0:7], v[8:15]
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; EXACTCUTOFF-NEXT: ds_store_b128 v16, v[12:15] offset:16
; EXACTCUTOFF-NEXT: ds_store_b128 v16, v[8:11]
; EXACTCUTOFF-NEXT: ds_load_b128 v[4:7], v17 offset:2064
; EXACTCUTOFF-NEXT: ds_load_b128 v[0:3], v17 offset:2048
; EXACTCUTOFF-NEXT: v_mov_b32_e32 v16, s1
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000200) size(2) SyncID(0)
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000100) size(2) SyncID(0)
; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(0)
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v15, v7 :: v_dual_mov_b32 v14, v6
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v13, v5 :: v_dual_mov_b32 v12, v4
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v11, v3 :: v_dual_mov_b32 v10, v2
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v9, v1 :: v_dual_mov_b32 v8, v0
; EXACTCUTOFF-NEXT: s_delay_alu instid0(VALU_DEP_1)
; EXACTCUTOFF-NEXT: v_wmma_f16_16x16x16_f16 v[8:15], v[0:7], v[0:7], v[8:15]
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; EXACTCUTOFF-NEXT: ds_store_b128 v16, v[12:15] offset:2064
; EXACTCUTOFF-NEXT: ds_store_b128 v16, v[8:11] offset:2048
; EXACTCUTOFF-NEXT: ds_load_b128 v[4:7], v17 offset:6160
; EXACTCUTOFF-NEXT: ds_load_b128 v[0:3], v17 offset:6144
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000200) size(2) SyncID(0)
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000100) size(2) SyncID(0)
; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(0)
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v15, v7 :: v_dual_mov_b32 v14, v6
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v13, v5 :: v_dual_mov_b32 v12, v4
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v11, v3 :: v_dual_mov_b32 v10, v2
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v9, v1 :: v_dual_mov_b32 v8, v0
; EXACTCUTOFF-NEXT: s_delay_alu instid0(VALU_DEP_1)
; EXACTCUTOFF-NEXT: v_wmma_f16_16x16x16_f16 v[8:15], v[0:7], v[0:7], v[8:15]
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; EXACTCUTOFF-NEXT: ds_store_b128 v16, v[12:15] offset:4112
; EXACTCUTOFF-NEXT: ds_store_b128 v16, v[8:11] offset:4096
; EXACTCUTOFF-NEXT: ds_load_b128 v[4:7], v17 offset:12304
; EXACTCUTOFF-NEXT: ds_load_b128 v[0:3], v17 offset:12288
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000200) size(2) SyncID(0)
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000100) size(2) SyncID(0)
; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(0)
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v15, v7 :: v_dual_mov_b32 v14, v6
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v13, v5 :: v_dual_mov_b32 v12, v4
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v11, v3 :: v_dual_mov_b32 v10, v2
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v9, v1 :: v_dual_mov_b32 v8, v0
; EXACTCUTOFF-NEXT: s_delay_alu instid0(VALU_DEP_1)
; EXACTCUTOFF-NEXT: v_wmma_f16_16x16x16_f16 v[8:15], v[0:7], v[0:7], v[8:15]
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; EXACTCUTOFF-NEXT: ds_store_b128 v16, v[12:15] offset:6160
; EXACTCUTOFF-NEXT: ds_store_b128 v16, v[8:11] offset:6144
; EXACTCUTOFF-NEXT: ds_load_b128 v[4:7], v17 offset:20496
; EXACTCUTOFF-NEXT: ds_load_b128 v[0:3], v17 offset:20480
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000200) size(2) SyncID(0)
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000100) size(2) SyncID(0)
; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(0)
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v15, v7 :: v_dual_mov_b32 v14, v6
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v13, v5 :: v_dual_mov_b32 v12, v4
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v11, v3 :: v_dual_mov_b32 v10, v2
; EXACTCUTOFF-NEXT: v_dual_mov_b32 v9, v1 :: v_dual_mov_b32 v8, v0
; EXACTCUTOFF-NEXT: s_delay_alu instid0(VALU_DEP_1)
; EXACTCUTOFF-NEXT: v_wmma_f16_16x16x16_f16 v[8:15], v[0:7], v[0:7], v[8:15]
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000008) size(1) SyncID(0)
; EXACTCUTOFF-NEXT: ds_store_b128 v16, v[12:15] offset:8208
; EXACTCUTOFF-NEXT: ds_store_b128 v16, v[8:11] offset:8192
; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000200) size(2) SyncID(0)
; EXACTCUTOFF-NEXT: s_endpgm
entry:
%idx = call i32 @llvm.amdgcn.workitem.id.x()
%load.0.addr = getelementptr <16 x half>, ptr addrspace(3) %in, i32 %idx
%load.0 = load <16 x half>, ptr addrspace(3) %load.0.addr
%load.1.addr = getelementptr <16 x half>, ptr addrspace(3) %load.0.addr, i32 64
%load.1 = load <16 x half>, ptr addrspace(3) %load.1.addr
%load.2.addr = getelementptr <16 x half>, ptr addrspace(3) %load.1.addr, i32 128
%load.2 = load <16 x half>, ptr addrspace(3) %load.2.addr
%load.3.addr = getelementptr <16 x half>, ptr addrspace(3) %load.2.addr, i32 192
%load.3 = load <16 x half>, ptr addrspace(3) %load.3.addr
%load.4.addr = getelementptr <16 x half>, ptr addrspace(3) %load.3.addr, i32 256
%load.4 = load <16 x half>, ptr addrspace(3) %load.4.addr
%mai.0 = call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16(<16 x half> %load.0, <16 x half> %load.0, <16 x half> %load.0, i1 0)
%mai.1 = call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16(<16 x half> %load.1, <16 x half> %load.1, <16 x half> %load.1, i1 0)
%mai.2 = call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16(<16 x half> %load.2, <16 x half> %load.2, <16 x half> %load.2, i1 0)
%mai.3 = call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16(<16 x half> %load.3, <16 x half> %load.3, <16 x half> %load.3, i1 0)
%mai.4 = call <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16(<16 x half> %load.4, <16 x half> %load.4, <16 x half> %load.4, i1 0)
%store.0.addr = getelementptr <16 x half>, ptr addrspace(3) %out, i32 %idx
store <16 x half> %mai.0, ptr addrspace(3) %store.0.addr
%store.1.addr = getelementptr <16 x half>, ptr addrspace(3) %out, i32 64
store <16 x half> %mai.1, ptr addrspace(3) %store.1.addr
%store.2.addr = getelementptr <16 x half>, ptr addrspace(3) %out, i32 128
store <16 x half> %mai.2, ptr addrspace(3) %store.2.addr
%store.3.addr = getelementptr <16 x half>, ptr addrspace(3) %out, i32 192
store <16 x half> %mai.3, ptr addrspace(3) %store.3.addr
%store.4.addr = getelementptr <16 x half>, ptr addrspace(3) %out, i32 256
store <16 x half> %mai.4, ptr addrspace(3) %store.4.addr
; 2 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 2, i32 0)
; 1 WMMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 2 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 2, i32 0)
; 2 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 2, i32 0)
; 1 WMMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 2 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 2, i32 0)
; 2 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 2, i32 0)
; 1 WMMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 2 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 2, i32 0)
; 2 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 2, i32 0)
; 1 WMMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 2 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 2, i32 0)
; 2 DS read
call void @llvm.amdgcn.sched.group.barrier(i32 256, i32 2, i32 0)
; 1 WMMA
call void @llvm.amdgcn.sched.group.barrier(i32 8, i32 1, i32 0)
; 2 DS write
call void @llvm.amdgcn.sched.group.barrier(i32 512, i32 2, i32 0)
ret void
}
declare i32 @llvm.amdgcn.workitem.id.x() #2
declare void @llvm.amdgcn.sched.group.barrier(i32, i32, i32) #1
declare <16 x half> @llvm.amdgcn.wmma.f16.16x16x16.f16(<16 x half>, <16 x half> , <16 x half>, i1 immarg) #1
attributes #0 = { nounwind "amdgpu-flat-work-group-size"="1,32" }
attributes #1 = { nounwind }
attributes #2 = { nounwind readnone speculatable }
|
