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
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-xnack -verify-machineinstrs -debug-only=machine-scheduler < %s 2> %t | FileCheck --enable-var-scope --check-prefix=GFX9 %s
; RUN: FileCheck --enable-var-scope --check-prefix=DBG %s < %t
; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs -debug-only=machine-scheduler < %s 2> %t | FileCheck --enable-var-scope --check-prefix=GFX10 %s
; RUN: FileCheck --enable-var-scope --check-prefix=DBG %s < %t
; RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs -debug-only=machine-scheduler < %s 2> %t | FileCheck --enable-var-scope --check-prefix=GFX11 %s
; RUN: FileCheck --enable-var-scope --check-prefixes=DBG,DBG11 %s < %t
; REQUIRES: asserts
; FIXME: Verifier error with xnack enabled.
; DBG-LABEL: cluster_load_cluster_store:
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 16
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
; DBG: Cluster ld/st SU([[L1:[0-9]+]]) - SU([[L2:[0-9]+]])
; DBG: Cluster ld/st SU([[L2]]) - SU([[L3:[0-9]+]])
; DBG: Cluster ld/st SU([[L3]]) - SU([[L4:[0-9]+]])
; DBG11: Cluster ld/st SU([[S1:[0-9]+]]) - SU([[S2:[0-9]+]])
; DBG11: Cluster ld/st SU([[S2]]) - SU([[S3:[0-9]+]])
; DBG11: Cluster ld/st SU([[S3]]) - SU([[S4:[0-9]+]])
; DBG-NOT: Cluster ld/st
define amdgpu_kernel void @cluster_load_cluster_store(ptr noalias %lb, ptr noalias %sb) {
; GFX9-LABEL: cluster_load_cluster_store:
; GFX9: ; %bb.0: ; %bb
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v0, s0
; GFX9-NEXT: v_mov_b32_e32 v1, s1
; GFX9-NEXT: flat_load_dword v2, v[0:1]
; GFX9-NEXT: flat_load_dword v3, v[0:1] offset:8
; GFX9-NEXT: flat_load_dword v4, v[0:1] offset:16
; GFX9-NEXT: flat_load_dword v5, v[0:1] offset:24
; GFX9-NEXT: v_mov_b32_e32 v0, s2
; GFX9-NEXT: v_mov_b32_e32 v1, s3
; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX9-NEXT: flat_store_dword v[0:1], v2
; GFX9-NEXT: flat_store_dword v[0:1], v3 offset:8
; GFX9-NEXT: flat_store_dword v[0:1], v4 offset:16
; GFX9-NEXT: flat_store_dword v[0:1], v5 offset:24
; GFX9-NEXT: s_endpgm
;
; GFX10-LABEL: cluster_load_cluster_store:
; GFX10: ; %bb.0: ; %bb
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
; GFX10-NEXT: s_add_u32 s4, s0, 8
; GFX10-NEXT: s_addc_u32 s5, s1, 0
; GFX10-NEXT: s_add_u32 s6, s0, 16
; GFX10-NEXT: v_mov_b32_e32 v0, s0
; GFX10-NEXT: s_addc_u32 s7, s1, 0
; GFX10-NEXT: v_mov_b32_e32 v1, s1
; GFX10-NEXT: s_add_u32 s0, s0, 24
; GFX10-NEXT: v_mov_b32_e32 v2, s4
; GFX10-NEXT: s_addc_u32 s1, s1, 0
; GFX10-NEXT: v_mov_b32_e32 v3, s5
; GFX10-NEXT: v_mov_b32_e32 v4, s6
; GFX10-NEXT: v_mov_b32_e32 v5, s7
; GFX10-NEXT: v_mov_b32_e32 v7, s1
; GFX10-NEXT: v_mov_b32_e32 v6, s0
; GFX10-NEXT: s_clause 0x3
; GFX10-NEXT: flat_load_dword v8, v[0:1]
; GFX10-NEXT: flat_load_dword v9, v[2:3]
; GFX10-NEXT: flat_load_dword v10, v[4:5]
; GFX10-NEXT: flat_load_dword v11, v[6:7]
; GFX10-NEXT: s_add_u32 s0, s2, 8
; GFX10-NEXT: s_addc_u32 s1, s3, 0
; GFX10-NEXT: v_mov_b32_e32 v0, s2
; GFX10-NEXT: v_mov_b32_e32 v3, s1
; GFX10-NEXT: v_mov_b32_e32 v2, s0
; GFX10-NEXT: s_add_u32 s0, s2, 16
; GFX10-NEXT: s_addc_u32 s1, s3, 0
; GFX10-NEXT: v_mov_b32_e32 v1, s3
; GFX10-NEXT: s_add_u32 s2, s2, 24
; GFX10-NEXT: s_addc_u32 s3, s3, 0
; GFX10-NEXT: v_mov_b32_e32 v5, s1
; GFX10-NEXT: v_mov_b32_e32 v4, s0
; GFX10-NEXT: v_mov_b32_e32 v7, s3
; GFX10-NEXT: v_mov_b32_e32 v6, s2
; GFX10-NEXT: s_waitcnt vmcnt(3) lgkmcnt(3)
; GFX10-NEXT: flat_store_dword v[0:1], v8
; GFX10-NEXT: s_waitcnt vmcnt(2) lgkmcnt(3)
; GFX10-NEXT: flat_store_dword v[2:3], v9
; GFX10-NEXT: s_waitcnt vmcnt(1) lgkmcnt(3)
; GFX10-NEXT: flat_store_dword v[4:5], v10
; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(3)
; GFX10-NEXT: flat_store_dword v[6:7], v11
; GFX10-NEXT: s_endpgm
;
; GFX11-LABEL: cluster_load_cluster_store:
; GFX11: ; %bb.0: ; %bb
; GFX11-NEXT: s_load_b128 s[0:3], s[0:1], 0x24
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: s_clause 0x3
; GFX11-NEXT: flat_load_b32 v2, v[0:1]
; GFX11-NEXT: flat_load_b32 v3, v[0:1] offset:8
; GFX11-NEXT: flat_load_b32 v4, v[0:1] offset:16
; GFX11-NEXT: flat_load_b32 v5, v[0:1] offset:24
; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT: s_waitcnt vmcnt(3) lgkmcnt(3)
; GFX11-NEXT: flat_store_b32 v[0:1], v2
; GFX11-NEXT: s_waitcnt vmcnt(2) lgkmcnt(3)
; GFX11-NEXT: flat_store_b32 v[0:1], v3 offset:8
; GFX11-NEXT: s_waitcnt vmcnt(1) lgkmcnt(3)
; GFX11-NEXT: flat_store_b32 v[0:1], v4 offset:16
; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(3)
; GFX11-NEXT: flat_store_b32 v[0:1], v5 offset:24
; GFX11-NEXT: s_endpgm
bb:
%ld0 = load i32, ptr %lb
%la1 = getelementptr inbounds i32, ptr %lb, i32 2
%ld1 = load i32, ptr %la1
%la2 = getelementptr inbounds i32, ptr %lb, i32 4
%ld2 = load i32, ptr %la2
%la3 = getelementptr inbounds i32, ptr %lb, i32 6
%ld3 = load i32, ptr %la3
store i32 %ld0, ptr %sb
%sa1 = getelementptr inbounds i32, ptr %sb, i32 2
store i32 %ld1, ptr %sa1
%sa2 = getelementptr inbounds i32, ptr %sb, i32 4
store i32 %ld2, ptr %sa2
%sa3 = getelementptr inbounds i32, ptr %sb, i32 6
store i32 %ld3, ptr %sa3
ret void
}
; DBG-LABEL: cluster_load_valu_cluster_store:
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 16
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
; DBG: Cluster ld/st SU([[L1:[0-9]+]]) - SU([[L2:[0-9]+]])
; DBG: Cluster ld/st SU([[L2]]) - SU([[L3:[0-9]+]])
; DBG: Cluster ld/st SU([[L3]]) - SU([[L4:[0-9]+]])
; DBG11: Cluster ld/st SU([[S1:[0-9]+]]) - SU([[S2:[0-9]+]])
; DBG11: Cluster ld/st SU([[S2]]) - SU([[S3:[0-9]+]])
; DBG11: Cluster ld/st SU([[S3]]) - SU([[S4:[0-9]+]])
; DBG-NOT: Cluster ld/st
define amdgpu_kernel void @cluster_load_valu_cluster_store(ptr noalias %lb, ptr noalias %sb) {
; GFX9-LABEL: cluster_load_valu_cluster_store:
; GFX9: ; %bb.0: ; %bb
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v0, s0
; GFX9-NEXT: v_mov_b32_e32 v1, s1
; GFX9-NEXT: flat_load_dword v2, v[0:1]
; GFX9-NEXT: flat_load_dword v3, v[0:1] offset:8
; GFX9-NEXT: flat_load_dword v4, v[0:1] offset:16
; GFX9-NEXT: flat_load_dword v5, v[0:1] offset:24
; GFX9-NEXT: v_mov_b32_e32 v0, s2
; GFX9-NEXT: v_mov_b32_e32 v1, s3
; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX9-NEXT: flat_store_dword v[0:1], v2
; GFX9-NEXT: v_add_u32_e32 v2, 1, v3
; GFX9-NEXT: flat_store_dword v[0:1], v4 offset:16
; GFX9-NEXT: flat_store_dword v[0:1], v2 offset:8
; GFX9-NEXT: flat_store_dword v[0:1], v5 offset:24
; GFX9-NEXT: s_endpgm
;
; GFX10-LABEL: cluster_load_valu_cluster_store:
; GFX10: ; %bb.0: ; %bb
; GFX10-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
; GFX10-NEXT: s_add_u32 s4, s0, 8
; GFX10-NEXT: s_addc_u32 s5, s1, 0
; GFX10-NEXT: v_mov_b32_e32 v2, s4
; GFX10-NEXT: s_add_u32 s6, s0, 16
; GFX10-NEXT: v_mov_b32_e32 v3, s5
; GFX10-NEXT: v_mov_b32_e32 v0, s0
; GFX10-NEXT: s_addc_u32 s7, s1, 0
; GFX10-NEXT: v_mov_b32_e32 v1, s1
; GFX10-NEXT: s_add_u32 s0, s0, 24
; GFX10-NEXT: s_addc_u32 s1, s1, 0
; GFX10-NEXT: v_mov_b32_e32 v4, s6
; GFX10-NEXT: v_mov_b32_e32 v5, s7
; GFX10-NEXT: flat_load_dword v6, v[2:3]
; GFX10-NEXT: v_mov_b32_e32 v3, s1
; GFX10-NEXT: v_mov_b32_e32 v2, s0
; GFX10-NEXT: s_clause 0x2
; GFX10-NEXT: flat_load_dword v8, v[0:1]
; GFX10-NEXT: flat_load_dword v9, v[4:5]
; GFX10-NEXT: flat_load_dword v10, v[2:3]
; GFX10-NEXT: s_add_u32 s0, s2, 8
; GFX10-NEXT: s_addc_u32 s1, s3, 0
; GFX10-NEXT: s_add_u32 s4, s2, 16
; GFX10-NEXT: v_mov_b32_e32 v3, s1
; GFX10-NEXT: s_addc_u32 s5, s3, 0
; GFX10-NEXT: v_mov_b32_e32 v0, s2
; GFX10-NEXT: v_mov_b32_e32 v2, s0
; GFX10-NEXT: s_add_u32 s0, s2, 24
; GFX10-NEXT: v_mov_b32_e32 v1, s3
; GFX10-NEXT: v_mov_b32_e32 v4, s4
; GFX10-NEXT: s_addc_u32 s1, s3, 0
; GFX10-NEXT: v_mov_b32_e32 v5, s5
; GFX10-NEXT: s_waitcnt vmcnt(3) lgkmcnt(3)
; GFX10-NEXT: v_add_nc_u32_e32 v11, 1, v6
; GFX10-NEXT: v_mov_b32_e32 v7, s1
; GFX10-NEXT: v_mov_b32_e32 v6, s0
; GFX10-NEXT: s_waitcnt vmcnt(2) lgkmcnt(2)
; GFX10-NEXT: flat_store_dword v[0:1], v8
; GFX10-NEXT: s_waitcnt vmcnt(1) lgkmcnt(2)
; GFX10-NEXT: flat_store_dword v[4:5], v9
; GFX10-NEXT: flat_store_dword v[2:3], v11
; GFX10-NEXT: s_waitcnt vmcnt(0) lgkmcnt(3)
; GFX10-NEXT: flat_store_dword v[6:7], v10
; GFX10-NEXT: s_endpgm
;
; GFX11-LABEL: cluster_load_valu_cluster_store:
; GFX11: ; %bb.0: ; %bb
; GFX11-NEXT: s_load_b128 s[0:3], s[0:1], 0x24
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
; GFX11-NEXT: s_clause 0x3
; GFX11-NEXT: flat_load_b32 v2, v[0:1] offset:8
; GFX11-NEXT: flat_load_b32 v3, v[0:1]
; GFX11-NEXT: flat_load_b32 v4, v[0:1] offset:16
; GFX11-NEXT: flat_load_b32 v5, v[0:1] offset:24
; GFX11-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3
; GFX11-NEXT: s_waitcnt vmcnt(3) lgkmcnt(3)
; GFX11-NEXT: v_add_nc_u32_e32 v2, 1, v2
; GFX11-NEXT: s_waitcnt vmcnt(2) lgkmcnt(2)
; GFX11-NEXT: s_clause 0x1
; GFX11-NEXT: flat_store_b32 v[0:1], v3
; GFX11-NEXT: flat_store_b32 v[0:1], v2 offset:8
; GFX11-NEXT: s_waitcnt vmcnt(1) lgkmcnt(3)
; GFX11-NEXT: flat_store_b32 v[0:1], v4 offset:16
; GFX11-NEXT: s_waitcnt vmcnt(0) lgkmcnt(3)
; GFX11-NEXT: flat_store_b32 v[0:1], v5 offset:24
; GFX11-NEXT: s_endpgm
bb:
%ld0 = load i32, ptr %lb
%la1 = getelementptr inbounds i32, ptr %lb, i32 2
%ld1 = load i32, ptr %la1
%la2 = getelementptr inbounds i32, ptr %lb, i32 4
%ld2 = load i32, ptr %la2
%la3 = getelementptr inbounds i32, ptr %lb, i32 6
%ld3 = load i32, ptr %la3
store i32 %ld0, ptr %sb
%sa1 = getelementptr inbounds i32, ptr %sb, i32 2
%add = add i32 %ld1, 1
store i32 %add, ptr %sa1
%sa2 = getelementptr inbounds i32, ptr %sb, i32 4
store i32 %ld2, ptr %sa2
%sa3 = getelementptr inbounds i32, ptr %sb, i32 6
store i32 %ld3, ptr %sa3
ret void
}
; Cluster loads from the same texture with different coordinates
; DBG-LABEL: cluster_image_load:
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 16
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 16
; DBG: {{^}}Cluster ld/st [[SU1:SU\([0-9]+\)]] - [[SU2:SU\([0-9]+\)]]
; DBG: {{^}}[[SU1]]: {{.*}} IMAGE_LOAD
; DBG: {{^}}[[SU2]]: {{.*}} IMAGE_LOAD
define amdgpu_ps void @cluster_image_load(<8 x i32> inreg %src, <8 x i32> inreg %dst, i32 %x, i32 %y) {
; GFX9-LABEL: cluster_image_load:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: v_add_u32_e32 v2, 1, v0
; GFX9-NEXT: v_add_u32_e32 v3, 1, v1
; GFX9-NEXT: v_add_u32_e32 v6, 2, v0
; GFX9-NEXT: v_add_u32_e32 v7, 2, v1
; GFX9-NEXT: image_load v[2:5], v[2:3], s[0:7] dmask:0xf unorm
; GFX9-NEXT: image_load v[6:9], v[6:7], s[0:7] dmask:0xf unorm
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_add_f32_e32 v5, v5, v9
; GFX9-NEXT: v_add_f32_e32 v4, v4, v8
; GFX9-NEXT: v_add_f32_e32 v3, v3, v7
; GFX9-NEXT: v_add_f32_e32 v2, v2, v6
; GFX9-NEXT: image_store v[2:5], v[0:1], s[8:15] dmask:0xf unorm
; GFX9-NEXT: s_endpgm
;
; GFX10-LABEL: cluster_image_load:
; GFX10: ; %bb.0: ; %entry
; GFX10-NEXT: v_add_nc_u32_e32 v10, 1, v0
; GFX10-NEXT: v_add_nc_u32_e32 v11, 1, v1
; GFX10-NEXT: v_add_nc_u32_e32 v12, 2, v0
; GFX10-NEXT: v_add_nc_u32_e32 v13, 2, v1
; GFX10-NEXT: s_clause 0x1
; GFX10-NEXT: image_load v[2:5], v[10:11], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_2D unorm
; GFX10-NEXT: image_load v[6:9], v[12:13], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_2D unorm
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: v_add_f32_e32 v5, v5, v9
; GFX10-NEXT: v_add_f32_e32 v4, v4, v8
; GFX10-NEXT: v_add_f32_e32 v3, v3, v7
; GFX10-NEXT: v_add_f32_e32 v2, v2, v6
; GFX10-NEXT: image_store v[2:5], v[0:1], s[8:15] dmask:0xf dim:SQ_RSRC_IMG_2D unorm
; GFX10-NEXT: s_endpgm
;
; GFX11-LABEL: cluster_image_load:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: v_add_nc_u32_e32 v2, 1, v0
; GFX11-NEXT: v_add_nc_u32_e32 v3, 1, v1
; GFX11-NEXT: v_add_nc_u32_e32 v6, 2, v0
; GFX11-NEXT: v_add_nc_u32_e32 v7, 2, v1
; GFX11-NEXT: s_clause 0x1
; GFX11-NEXT: image_load v[2:5], v[2:3], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_2D unorm
; GFX11-NEXT: image_load v[6:9], v[6:7], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_2D unorm
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_dual_add_f32 v2, v2, v6 :: v_dual_add_f32 v5, v5, v9
; GFX11-NEXT: v_dual_add_f32 v4, v4, v8 :: v_dual_add_f32 v3, v3, v7
; GFX11-NEXT: image_store v[2:5], v[0:1], s[8:15] dmask:0xf dim:SQ_RSRC_IMG_2D unorm
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
entry:
%x1 = add i32 %x, 1
%y1 = add i32 %y, 1
%val1 = call <4 x float> @llvm.amdgcn.image.load.2d.v4f32.i32(i32 15, i32 %x1, i32 %y1, <8 x i32> %src, i32 0, i32 0)
%x2 = add i32 %x, 2
%y2 = add i32 %y, 2
%val2 = call <4 x float> @llvm.amdgcn.image.load.2d.v4f32.i32(i32 15, i32 %x2, i32 %y2, <8 x i32> %src, i32 0, i32 0)
%val = fadd fast <4 x float> %val1, %val2
call void @llvm.amdgcn.image.store.2d.v4f32.i32(<4 x float> %val, i32 15, i32 %x, i32 %y, <8 x i32> %dst, i32 0, i32 0)
ret void
}
; Don't cluster loads from different textures
; DBG-LABEL: no_cluster_image_load:
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 16
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 16
; DBG-NOT: {{^}}Cluster ld/st
define amdgpu_ps void @no_cluster_image_load(<8 x i32> inreg %src1, <8 x i32> inreg %src2, <8 x i32> inreg %dst, i32 %x, i32 %y) {
; GFX9-LABEL: no_cluster_image_load:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: image_load_mip v[3:6], v[0:2], s[0:7] dmask:0xf unorm
; GFX9-NEXT: image_load_mip v[7:10], v[0:2], s[8:15] dmask:0xf unorm
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_add_f32_e32 v6, v6, v10
; GFX9-NEXT: v_add_f32_e32 v5, v5, v9
; GFX9-NEXT: v_add_f32_e32 v4, v4, v8
; GFX9-NEXT: v_add_f32_e32 v3, v3, v7
; GFX9-NEXT: image_store v[3:6], v[0:1], s[16:23] dmask:0xf unorm
; GFX9-NEXT: s_endpgm
;
; GFX10-LABEL: no_cluster_image_load:
; GFX10: ; %bb.0: ; %entry
; GFX10-NEXT: v_mov_b32_e32 v10, 0
; GFX10-NEXT: image_load_mip v[2:5], [v0, v1, v10], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_2D unorm
; GFX10-NEXT: image_load_mip v[6:9], [v0, v1, v10], s[8:15] dmask:0xf dim:SQ_RSRC_IMG_2D unorm
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: v_add_f32_e32 v5, v5, v9
; GFX10-NEXT: v_add_f32_e32 v4, v4, v8
; GFX10-NEXT: v_add_f32_e32 v3, v3, v7
; GFX10-NEXT: v_add_f32_e32 v2, v2, v6
; GFX10-NEXT: image_store v[2:5], v[0:1], s[16:23] dmask:0xf dim:SQ_RSRC_IMG_2D unorm
; GFX10-NEXT: s_endpgm
;
; GFX11-LABEL: no_cluster_image_load:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: v_mov_b32_e32 v6, 0
; GFX11-NEXT: image_load_mip v[2:5], [v0, v1, v6], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_2D unorm
; GFX11-NEXT: image_load_mip v[6:9], [v0, v1, v6], s[8:15] dmask:0xf dim:SQ_RSRC_IMG_2D unorm
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_dual_add_f32 v5, v5, v9 :: v_dual_add_f32 v4, v4, v8
; GFX11-NEXT: v_dual_add_f32 v3, v3, v7 :: v_dual_add_f32 v2, v2, v6
; GFX11-NEXT: image_store v[2:5], v[0:1], s[16:23] dmask:0xf dim:SQ_RSRC_IMG_2D unorm
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
entry:
%val1 = call <4 x float> @llvm.amdgcn.image.load.mip.2d.v4f32.i32(i32 15, i32 %x, i32 %y, i32 0, <8 x i32> %src1, i32 0, i32 0)
%val2 = call <4 x float> @llvm.amdgcn.image.load.mip.2d.v4f32.i32(i32 15, i32 %x, i32 %y, i32 0, <8 x i32> %src2, i32 0, i32 0)
%val = fadd fast <4 x float> %val1, %val2
call void @llvm.amdgcn.image.store.2d.v4f32.i32(<4 x float> %val, i32 15, i32 %x, i32 %y, <8 x i32> %dst, i32 0, i32 0)
ret void
}
; Cluster loads from the same texture and sampler with different coordinates
; DBG-LABEL: cluster_image_sample:
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 16
; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 16
; DBG: {{^}}Cluster ld/st [[SU1:SU\([0-9]+\)]] - [[SU2:SU\([0-9]+\)]]
; DBG: {{^}}[[SU1]]: {{.*}} IMAGE_SAMPLE
; DBG: {{^}}[[SU2]]: {{.*}} IMAGE_SAMPLE
define amdgpu_ps void @cluster_image_sample(<8 x i32> inreg %src, <4 x i32> inreg %smp, <8 x i32> inreg %dst, i32 %x, i32 %y) {
; GFX9-LABEL: cluster_image_sample:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: v_cvt_f32_i32_e32 v8, v0
; GFX9-NEXT: v_cvt_f32_i32_e32 v9, v1
; GFX9-NEXT: v_mov_b32_e32 v4, 0
; GFX9-NEXT: v_mov_b32_e32 v5, v4
; GFX9-NEXT: v_add_f32_e32 v2, 1.0, v8
; GFX9-NEXT: v_add_f32_e32 v3, 1.0, v9
; GFX9-NEXT: v_mov_b32_e32 v6, v4
; GFX9-NEXT: v_mov_b32_e32 v7, v4
; GFX9-NEXT: v_add_f32_e32 v8, 2.0, v8
; GFX9-NEXT: v_add_f32_e32 v9, 2.0, v9
; GFX9-NEXT: v_mov_b32_e32 v10, 1.0
; GFX9-NEXT: v_mov_b32_e32 v11, v10
; GFX9-NEXT: v_mov_b32_e32 v12, v10
; GFX9-NEXT: v_mov_b32_e32 v13, v10
; GFX9-NEXT: image_sample_d v[2:5], v[2:7], s[0:7], s[8:11] dmask:0xf
; GFX9-NEXT: image_sample_d v[6:9], v[8:13], s[0:7], s[8:11] dmask:0xf
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_add_f32_e32 v5, v5, v9
; GFX9-NEXT: v_add_f32_e32 v4, v4, v8
; GFX9-NEXT: v_add_f32_e32 v3, v3, v7
; GFX9-NEXT: v_add_f32_e32 v2, v2, v6
; GFX9-NEXT: image_store v[2:5], v[0:1], s[12:19] dmask:0xf unorm
; GFX9-NEXT: s_endpgm
;
; GFX10-LABEL: cluster_image_sample:
; GFX10: ; %bb.0: ; %entry
; GFX10-NEXT: v_cvt_f32_i32_e32 v8, v0
; GFX10-NEXT: v_cvt_f32_i32_e32 v9, v1
; GFX10-NEXT: v_mov_b32_e32 v4, 0
; GFX10-NEXT: v_mov_b32_e32 v10, 1.0
; GFX10-NEXT: v_add_f32_e32 v2, 1.0, v8
; GFX10-NEXT: v_add_f32_e32 v3, 1.0, v9
; GFX10-NEXT: v_mov_b32_e32 v5, v4
; GFX10-NEXT: v_mov_b32_e32 v6, v4
; GFX10-NEXT: v_mov_b32_e32 v7, v4
; GFX10-NEXT: v_add_f32_e32 v8, 2.0, v8
; GFX10-NEXT: v_add_f32_e32 v9, 2.0, v9
; GFX10-NEXT: v_mov_b32_e32 v11, v10
; GFX10-NEXT: v_mov_b32_e32 v12, v10
; GFX10-NEXT: v_mov_b32_e32 v13, v10
; GFX10-NEXT: s_clause 0x1
; GFX10-NEXT: image_sample_d v[14:17], v[2:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
; GFX10-NEXT: image_sample_d v[18:21], v[8:13], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: v_add_f32_e32 v5, v17, v21
; GFX10-NEXT: v_add_f32_e32 v4, v16, v20
; GFX10-NEXT: v_add_f32_e32 v3, v15, v19
; GFX10-NEXT: v_add_f32_e32 v2, v14, v18
; GFX10-NEXT: image_store v[2:5], v[0:1], s[12:19] dmask:0xf dim:SQ_RSRC_IMG_2D unorm
; GFX10-NEXT: s_endpgm
;
; GFX11-LABEL: cluster_image_sample:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: v_cvt_f32_i32_e32 v4, v0
; GFX11-NEXT: v_cvt_f32_i32_e32 v5, v1
; GFX11-NEXT: v_mov_b32_e32 v2, 0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-NEXT: v_dual_mov_b32 v6, 1.0 :: v_dual_add_f32 v11, 2.0, v5
; GFX11-NEXT: v_dual_add_f32 v9, 1.0, v5 :: v_dual_add_f32 v8, 1.0, v4
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-NEXT: v_dual_mov_b32 v3, v2 :: v_dual_add_f32 v10, 2.0, v4
; GFX11-NEXT: v_mov_b32_e32 v7, v6
; GFX11-NEXT: s_clause 0x1
; GFX11-NEXT: image_sample_d v[2:5], [v8, v9, v2, v2, v[2:3]], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
; GFX11-NEXT: image_sample_d v[6:9], [v10, v11, v6, v6, v[6:7]], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: v_dual_add_f32 v5, v5, v9 :: v_dual_add_f32 v4, v4, v8
; GFX11-NEXT: v_dual_add_f32 v3, v3, v7 :: v_dual_add_f32 v2, v2, v6
; GFX11-NEXT: image_store v[2:5], v[0:1], s[12:19] dmask:0xf dim:SQ_RSRC_IMG_2D unorm
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
entry:
%s = sitofp i32 %x to float
%t = sitofp i32 %y to float
%s1 = fadd float %s, 1.0
%t1 = fadd float %t, 1.0
%val1 = call <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f32(i32 15, float %s1, float %t1, float 0.0, float 0.0, float 0.0, float 0.0, <8 x i32> %src, <4 x i32> %smp, i1 false, i32 0, i32 0)
%s2 = fadd float %s, 2.0
%t2 = fadd float %t, 2.0
%val2 = call <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f32(i32 15, float %s2, float %t2, float 1.0, float 1.0, float 1.0, float 1.0, <8 x i32> %src, <4 x i32> %smp, i1 false, i32 0, i32 0)
%val = fadd fast <4 x float> %val1, %val2
call void @llvm.amdgcn.image.store.2d.v4f32.i32(<4 x float> %val, i32 15, i32 %x, i32 %y, <8 x i32> %dst, i32 0, i32 0)
ret void
}
declare <4 x float> @llvm.amdgcn.image.load.2d.v4f32.i32(i32 immarg, i32, i32, <8 x i32>, i32 immarg, i32 immarg)
declare <4 x float> @llvm.amdgcn.image.load.mip.2d.v4f32.i32(i32 immarg, i32, i32, i32, <8 x i32>, i32 immarg, i32 immarg)
declare <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f32(i32, float, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32)
declare void @llvm.amdgcn.image.store.2d.v4f32.i32(<4 x float>, i32 immarg, i32, i32, <8 x i32>, i32 immarg, i32 immarg)
|
