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
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
|
; RUN: opt -mtriple=amdgcn-- -S -amdgpu-unify-divergent-exit-nodes -verify -structurizecfg -verify -si-annotate-control-flow %s | FileCheck -check-prefix=IR %s
; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
; Add an extra verifier runs. There were some cases where invalid IR
; was produced but happened to be fixed by the later passes.
; Make sure divergent control flow with multiple exits from a region
; is properly handled. UnifyFunctionExitNodes should be run before
; StructurizeCFG.
; IR-LABEL: @multi_divergent_region_exit_ret_ret(
; IR: %1 = call { i1, i64 } @llvm.amdgcn.if(i1 %0)
; IR: %2 = extractvalue { i1, i64 } %1, 0
; IR: %3 = extractvalue { i1, i64 } %1, 1
; IR: br i1 %2, label %LeafBlock1, label %Flow
; IR: Flow:
; IR: %4 = phi i1 [ true, %LeafBlock1 ], [ false, %entry ]
; IR: %5 = phi i1 [ %10, %LeafBlock1 ], [ false, %entry ]
; IR: %6 = call { i1, i64 } @llvm.amdgcn.else(i64 %3)
; IR: %7 = extractvalue { i1, i64 } %6, 0
; IR: %8 = extractvalue { i1, i64 } %6, 1
; IR: br i1 %7, label %LeafBlock, label %Flow1
; IR: LeafBlock:
; IR: br label %Flow1
; IR: LeafBlock1:
; IR: br label %Flow{{$}}
; IR: Flow2:
; IR: %11 = phi i1 [ false, %exit1 ], [ %15, %Flow1 ]
; IR: call void @llvm.amdgcn.end.cf(i64 %19)
; IR: %12 = call { i1, i64 } @llvm.amdgcn.if(i1 %11)
; IR: %13 = extractvalue { i1, i64 } %12, 0
; IR: %14 = extractvalue { i1, i64 } %12, 1
; IR: br i1 %13, label %exit0, label %UnifiedReturnBlock
; IR: exit0:
; IR: store volatile i32 9, i32 addrspace(1)* undef
; IR: br label %UnifiedReturnBlock
; IR: Flow1:
; IR: %15 = phi i1 [ %SwitchLeaf, %LeafBlock ], [ %4, %Flow ]
; IR: %16 = phi i1 [ %9, %LeafBlock ], [ %5, %Flow ]
; IR: call void @llvm.amdgcn.end.cf(i64 %8)
; IR: %17 = call { i1, i64 } @llvm.amdgcn.if(i1 %16)
; IR: %18 = extractvalue { i1, i64 } %17, 0
; IR: %19 = extractvalue { i1, i64 } %17, 1
; IR: br i1 %18, label %exit1, label %Flow2
; IR: exit1:
; IR: store volatile i32 17, i32 addrspace(3)* undef
; IR: br label %Flow2
; IR: UnifiedReturnBlock:
; IR: call void @llvm.amdgcn.end.cf(i64 %14)
; IR: ret void
; GCN-LABEL: {{^}}multi_divergent_region_exit_ret_ret:
; GCN: v_cmp_lt_i32_e32 vcc, 1
; GCN: s_and_saveexec_b64
; GCN: s_xor_b64
; FIXME: Why is this compare essentially repeated?
; GCN: v_cmp_eq_u32_e32 vcc, 1, [[REG:v[0-9]+]]
; GCN-NEXT: v_cmp_ne_u32_e64 s{{\[[0-9]+:[0-9]+\]}}, 1, [[REG]]
; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 0, -1, vcc
; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 0, -1
; GCN: ; %Flow1
; GCN-NEXT: s_or_b64 exec, exec
; GCN: v_cmp_ne_u32_e32 vcc, 0
; GCN: ; %exit1
; GCN: ds_write_b32
; GCN: %Flow2
; GCN-NEXT: s_or_b64 exec, exec
; GCN: v_cmp_ne_u32_e32 vcc, 0
; GCN-NEXT: s_and_saveexec_b64
; GCN-NEXT: s_xor_b64
; GCN: ; %exit0
; GCN: buffer_store_dword
; GCN: ; %UnifiedReturnBlock
; GCN-NEXT: s_or_b64 exec, exec
; GCN-NEXT: s_endpgm
define amdgpu_kernel void @multi_divergent_region_exit_ret_ret(i32 addrspace(1)* nocapture %arg0, i32 addrspace(1)* nocapture %arg1, i32 addrspace(1)* nocapture %arg2) #0 {
entry:
%tmp = tail call i32 @llvm.amdgcn.workitem.id.x() #1
%tmp1 = add i32 0, %tmp
%tmp2 = zext i32 %tmp1 to i64
%tmp3 = add i64 0, %tmp2
%tmp4 = shl i64 %tmp3, 32
%tmp5 = ashr exact i64 %tmp4, 32
%tmp6 = getelementptr inbounds i32, i32 addrspace(1)* %arg0, i64 %tmp5
%tmp7 = load i32, i32 addrspace(1)* %tmp6, align 4
%tmp8 = sext i32 %tmp7 to i64
%tmp9 = getelementptr inbounds i32, i32 addrspace(1)* %arg1, i64 %tmp8
%tmp10 = load i32, i32 addrspace(1)* %tmp9, align 4
%tmp13 = zext i32 %tmp10 to i64
%tmp14 = getelementptr inbounds i32, i32 addrspace(1)* %arg2, i64 %tmp13
%tmp16 = load i32, i32 addrspace(1)* %tmp14, align 16
%Pivot = icmp slt i32 %tmp16, 2
br i1 %Pivot, label %LeafBlock, label %LeafBlock1
LeafBlock: ; preds = %entry
%SwitchLeaf = icmp eq i32 %tmp16, 1
br i1 %SwitchLeaf, label %exit0, label %exit1
LeafBlock1: ; preds = %entry
%SwitchLeaf2 = icmp eq i32 %tmp16, 2
br i1 %SwitchLeaf2, label %exit0, label %exit1
exit0: ; preds = %LeafBlock, %LeafBlock1
store volatile i32 9, i32 addrspace(1)* undef
ret void
exit1: ; preds = %LeafBlock, %LeafBlock1
store volatile i32 17, i32 addrspace(3)* undef
ret void
}
; IR-LABEL: @multi_divergent_region_exit_unreachable_unreachable(
; IR: %1 = call { i1, i64 } @llvm.amdgcn.if(i1 %0)
; IR: %6 = call { i1, i64 } @llvm.amdgcn.else(i64 %3)
; IR: %11 = phi i1 [ false, %exit1 ], [ %15, %Flow1 ]
; IR: call void @llvm.amdgcn.end.cf(i64 %19)
; IR: %12 = call { i1, i64 } @llvm.amdgcn.if(i1 %11)
; IR: br i1 %13, label %exit0, label %UnifiedUnreachableBlock
; IR: UnifiedUnreachableBlock:
; IR-NEXT: unreachable
; FIXME: Probably should insert an s_endpgm anyway.
; GCN-LABEL: {{^}}multi_divergent_region_exit_unreachable_unreachable:
; GCN: ; %UnifiedUnreachableBlock
; GCN-NEXT: .Lfunc_end
define amdgpu_kernel void @multi_divergent_region_exit_unreachable_unreachable(i32 addrspace(1)* nocapture %arg0, i32 addrspace(1)* nocapture %arg1, i32 addrspace(1)* nocapture %arg2) #0 {
entry:
%tmp = tail call i32 @llvm.amdgcn.workitem.id.x() #1
%tmp1 = add i32 0, %tmp
%tmp2 = zext i32 %tmp1 to i64
%tmp3 = add i64 0, %tmp2
%tmp4 = shl i64 %tmp3, 32
%tmp5 = ashr exact i64 %tmp4, 32
%tmp6 = getelementptr inbounds i32, i32 addrspace(1)* %arg0, i64 %tmp5
%tmp7 = load i32, i32 addrspace(1)* %tmp6, align 4
%tmp8 = sext i32 %tmp7 to i64
%tmp9 = getelementptr inbounds i32, i32 addrspace(1)* %arg1, i64 %tmp8
%tmp10 = load i32, i32 addrspace(1)* %tmp9, align 4
%tmp13 = zext i32 %tmp10 to i64
%tmp14 = getelementptr inbounds i32, i32 addrspace(1)* %arg2, i64 %tmp13
%tmp16 = load i32, i32 addrspace(1)* %tmp14, align 16
%Pivot = icmp slt i32 %tmp16, 2
br i1 %Pivot, label %LeafBlock, label %LeafBlock1
LeafBlock: ; preds = %entry
%SwitchLeaf = icmp eq i32 %tmp16, 1
br i1 %SwitchLeaf, label %exit0, label %exit1
LeafBlock1: ; preds = %entry
%SwitchLeaf2 = icmp eq i32 %tmp16, 2
br i1 %SwitchLeaf2, label %exit0, label %exit1
exit0: ; preds = %LeafBlock, %LeafBlock1
store volatile i32 9, i32 addrspace(1)* undef
unreachable
exit1: ; preds = %LeafBlock, %LeafBlock1
store volatile i32 17, i32 addrspace(3)* undef
unreachable
}
; IR-LABEL: @multi_exit_region_divergent_ret_uniform_ret(
; IR: %divergent.cond0 = icmp slt i32 %tmp16, 2
; IR: llvm.amdgcn.if
; IR: br i1
; IR: {{^}}Flow:
; IR: %4 = phi i1 [ true, %LeafBlock1 ], [ false, %entry ]
; IR: %5 = phi i1 [ %10, %LeafBlock1 ], [ false, %entry ]
; IR: %6 = call { i1, i64 } @llvm.amdgcn.else(i64 %3)
; IR: br i1 %7, label %LeafBlock, label %Flow1
; IR: {{^}}LeafBlock:
; IR: %divergent.cond1 = icmp eq i32 %tmp16, 1
; IR: %9 = xor i1 %divergent.cond1, true
; IR: br label %Flow1
; IR: LeafBlock1:
; IR: %uniform.cond0 = icmp eq i32 %arg3, 2
; IR: %10 = xor i1 %uniform.cond0, true
; IR: br label %Flow
; IR: Flow2:
; IR: %11 = phi i1 [ false, %exit1 ], [ %15, %Flow1 ]
; IR: call void @llvm.amdgcn.end.cf(i64 %19)
; IR: %12 = call { i1, i64 } @llvm.amdgcn.if(i1 %11)
; IR: br i1 %13, label %exit0, label %UnifiedReturnBlock
; IR: exit0:
; IR: store volatile i32 9, i32 addrspace(1)* undef
; IR: br label %UnifiedReturnBlock
; IR: {{^}}Flow1:
; IR: %15 = phi i1 [ %divergent.cond1, %LeafBlock ], [ %4, %Flow ]
; IR: %16 = phi i1 [ %9, %LeafBlock ], [ %5, %Flow ]
; IR: call void @llvm.amdgcn.end.cf(i64 %8)
; IR: %17 = call { i1, i64 } @llvm.amdgcn.if(i1 %16)
; IR: %18 = extractvalue { i1, i64 } %17, 0
; IR: %19 = extractvalue { i1, i64 } %17, 1
; IR: br i1 %18, label %exit1, label %Flow2
; IR: exit1:
; IR: store volatile i32 17, i32 addrspace(3)* undef
; IR: br label %Flow2
; IR: UnifiedReturnBlock:
; IR: call void @llvm.amdgcn.end.cf(i64 %14)
; IR: ret void
define amdgpu_kernel void @multi_exit_region_divergent_ret_uniform_ret(i32 addrspace(1)* nocapture %arg0, i32 addrspace(1)* nocapture %arg1, i32 addrspace(1)* nocapture %arg2, i32 %arg3) #0 {
entry:
%tmp = tail call i32 @llvm.amdgcn.workitem.id.x() #1
%tmp1 = add i32 0, %tmp
%tmp2 = zext i32 %tmp1 to i64
%tmp3 = add i64 0, %tmp2
%tmp4 = shl i64 %tmp3, 32
%tmp5 = ashr exact i64 %tmp4, 32
%tmp6 = getelementptr inbounds i32, i32 addrspace(1)* %arg0, i64 %tmp5
%tmp7 = load i32, i32 addrspace(1)* %tmp6, align 4
%tmp8 = sext i32 %tmp7 to i64
%tmp9 = getelementptr inbounds i32, i32 addrspace(1)* %arg1, i64 %tmp8
%tmp10 = load i32, i32 addrspace(1)* %tmp9, align 4
%tmp13 = zext i32 %tmp10 to i64
%tmp14 = getelementptr inbounds i32, i32 addrspace(1)* %arg2, i64 %tmp13
%tmp16 = load i32, i32 addrspace(1)* %tmp14, align 16
%divergent.cond0 = icmp slt i32 %tmp16, 2
br i1 %divergent.cond0, label %LeafBlock, label %LeafBlock1
LeafBlock: ; preds = %entry
%divergent.cond1 = icmp eq i32 %tmp16, 1
br i1 %divergent.cond1, label %exit0, label %exit1
LeafBlock1: ; preds = %entry
%uniform.cond0 = icmp eq i32 %arg3, 2
br i1 %uniform.cond0, label %exit0, label %exit1
exit0: ; preds = %LeafBlock, %LeafBlock1
store volatile i32 9, i32 addrspace(1)* undef
ret void
exit1: ; preds = %LeafBlock, %LeafBlock1
store volatile i32 17, i32 addrspace(3)* undef
ret void
}
; IR-LABEL: @multi_exit_region_uniform_ret_divergent_ret(
; IR: %1 = call { i1, i64 } @llvm.amdgcn.if(i1 %0)
; IR: br i1 %2, label %LeafBlock1, label %Flow
; IR: Flow:
; IR: %4 = phi i1 [ true, %LeafBlock1 ], [ false, %entry ]
; IR: %5 = phi i1 [ %10, %LeafBlock1 ], [ false, %entry ]
; IR: %6 = call { i1, i64 } @llvm.amdgcn.else(i64 %3)
; IR: %11 = phi i1 [ false, %exit1 ], [ %15, %Flow1 ]
; IR: call void @llvm.amdgcn.end.cf(i64 %19)
; IR: %12 = call { i1, i64 } @llvm.amdgcn.if(i1 %11)
define amdgpu_kernel void @multi_exit_region_uniform_ret_divergent_ret(i32 addrspace(1)* nocapture %arg0, i32 addrspace(1)* nocapture %arg1, i32 addrspace(1)* nocapture %arg2, i32 %arg3) #0 {
entry:
%tmp = tail call i32 @llvm.amdgcn.workitem.id.x() #1
%tmp1 = add i32 0, %tmp
%tmp2 = zext i32 %tmp1 to i64
%tmp3 = add i64 0, %tmp2
%tmp4 = shl i64 %tmp3, 32
%tmp5 = ashr exact i64 %tmp4, 32
%tmp6 = getelementptr inbounds i32, i32 addrspace(1)* %arg0, i64 %tmp5
%tmp7 = load i32, i32 addrspace(1)* %tmp6, align 4
%tmp8 = sext i32 %tmp7 to i64
%tmp9 = getelementptr inbounds i32, i32 addrspace(1)* %arg1, i64 %tmp8
%tmp10 = load i32, i32 addrspace(1)* %tmp9, align 4
%tmp13 = zext i32 %tmp10 to i64
%tmp14 = getelementptr inbounds i32, i32 addrspace(1)* %arg2, i64 %tmp13
%tmp16 = load i32, i32 addrspace(1)* %tmp14, align 16
%Pivot = icmp slt i32 %tmp16, 2
br i1 %Pivot, label %LeafBlock, label %LeafBlock1
LeafBlock: ; preds = %entry
%SwitchLeaf = icmp eq i32 %arg3, 1
br i1 %SwitchLeaf, label %exit0, label %exit1
LeafBlock1: ; preds = %entry
%SwitchLeaf2 = icmp eq i32 %tmp16, 2
br i1 %SwitchLeaf2, label %exit0, label %exit1
exit0: ; preds = %LeafBlock, %LeafBlock1
store volatile i32 9, i32 addrspace(1)* undef
ret void
exit1: ; preds = %LeafBlock, %LeafBlock1
store volatile i32 17, i32 addrspace(3)* undef
ret void
}
; IR-LABEL: @multi_divergent_region_exit_ret_ret_return_value(
; IR: Flow2:
; IR: %11 = phi float [ 2.000000e+00, %exit1 ], [ undef, %Flow1 ]
; IR: %12 = phi i1 [ false, %exit1 ], [ %16, %Flow1 ]
; IR: call void @llvm.amdgcn.end.cf(i64 %20)
; IR: UnifiedReturnBlock:
; IR: %UnifiedRetVal = phi float [ %11, %Flow2 ], [ 1.000000e+00, %exit0 ]
; IR: call void @llvm.amdgcn.end.cf(i64 %15)
; IR: ret float %UnifiedRetVal
define amdgpu_ps float @multi_divergent_region_exit_ret_ret_return_value(i32 %vgpr) #0 {
entry:
%Pivot = icmp slt i32 %vgpr, 2
br i1 %Pivot, label %LeafBlock, label %LeafBlock1
LeafBlock: ; preds = %entry
%SwitchLeaf = icmp eq i32 %vgpr, 1
br i1 %SwitchLeaf, label %exit0, label %exit1
LeafBlock1: ; preds = %entry
%SwitchLeaf2 = icmp eq i32 %vgpr, 2
br i1 %SwitchLeaf2, label %exit0, label %exit1
exit0: ; preds = %LeafBlock, %LeafBlock1
store i32 9, i32 addrspace(1)* undef
ret float 1.0
exit1: ; preds = %LeafBlock, %LeafBlock1
store i32 17, i32 addrspace(3)* undef
ret float 2.0
}
; IR-LABEL: @uniform_branch_to_multi_divergent_region_exit_ret_ret_return_value(
; GCN-LABEL: {{^}}uniform_branch_to_multi_divergent_region_exit_ret_ret_return_value:
; GCN: s_cmp_gt_i32 s0, 1
; GCN: s_cbranch_scc0 [[FLOW:BB[0-9]+_[0-9]+]]
; GCN: v_cmp_ne_u32_e32 vcc, 7, v0
; GCN: {{^}}[[FLOW]]:
; GCN: s_cbranch_vccnz [[FLOW1:BB[0-9]+]]
; GCN: v_mov_b32_e32 v0, 2.0
; GCN: s_or_b64 exec, exec
; GCN: s_and_b64 exec, exec
; GCN: v_mov_b32_e32 v0, 1.0
; GCN: {{^BB[0-9]+_[0-9]+}}: ; %UnifiedReturnBlock
; GCN-NEXT: s_or_b64 exec, exec
; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN-NEXT: ; return
define amdgpu_ps float @uniform_branch_to_multi_divergent_region_exit_ret_ret_return_value(i32 inreg %sgpr, i32 %vgpr) #0 {
entry:
%uniform.cond = icmp slt i32 %sgpr, 2
br i1 %uniform.cond, label %LeafBlock, label %LeafBlock1
LeafBlock: ; preds = %entry
%divergent.cond0 = icmp eq i32 %vgpr, 3
br i1 %divergent.cond0, label %exit0, label %exit1
LeafBlock1: ; preds = %entry
%divergent.cond1 = icmp eq i32 %vgpr, 7
br i1 %divergent.cond1, label %exit0, label %exit1
exit0: ; preds = %LeafBlock, %LeafBlock1
store i32 9, i32 addrspace(1)* undef
ret float 1.0
exit1: ; preds = %LeafBlock, %LeafBlock1
store i32 17, i32 addrspace(3)* undef
ret float 2.0
}
; IR-LABEL: @multi_divergent_region_exit_ret_unreachable(
; IR: %1 = call { i1, i64 } @llvm.amdgcn.if(i1 %0)
; IR: Flow:
; IR: %4 = phi i1 [ true, %LeafBlock1 ], [ false, %entry ]
; IR: %5 = phi i1 [ %10, %LeafBlock1 ], [ false, %entry ]
; IR: %6 = call { i1, i64 } @llvm.amdgcn.else(i64 %3)
; IR: Flow2:
; IR: %11 = phi i1 [ false, %exit1 ], [ %15, %Flow1 ]
; IR: call void @llvm.amdgcn.end.cf(i64 %19)
; IR: %12 = call { i1, i64 } @llvm.amdgcn.if(i1 %11)
; IR: br i1 %13, label %exit0, label %UnifiedReturnBlock
; IR: exit0:
; IR-NEXT: store volatile i32 17, i32 addrspace(3)* undef
; IR-NEXT: br label %UnifiedReturnBlock
; IR: Flow1:
; IR: %15 = phi i1 [ %SwitchLeaf, %LeafBlock ], [ %4, %Flow ]
; IR: %16 = phi i1 [ %9, %LeafBlock ], [ %5, %Flow ]
; IR: call void @llvm.amdgcn.end.cf(i64 %8)
; IR: %17 = call { i1, i64 } @llvm.amdgcn.if(i1 %16)
; IR: %18 = extractvalue { i1, i64 } %17, 0
; IR: %19 = extractvalue { i1, i64 } %17, 1
; IR: br i1 %18, label %exit1, label %Flow2
; IR: exit1:
; IR-NEXT: store volatile i32 9, i32 addrspace(1)* undef
; IR-NEXT: call void @llvm.amdgcn.unreachable()
; IR-NEXT: br label %Flow2
; IR: UnifiedReturnBlock:
; IR-NEXT: call void @llvm.amdgcn.end.cf(i64 %14)
; IR-NEXT: ret void
define amdgpu_kernel void @multi_divergent_region_exit_ret_unreachable(i32 addrspace(1)* nocapture %arg0, i32 addrspace(1)* nocapture %arg1, i32 addrspace(1)* nocapture %arg2) #0 {
entry:
%tmp = tail call i32 @llvm.amdgcn.workitem.id.x() #1
%tmp1 = add i32 0, %tmp
%tmp2 = zext i32 %tmp1 to i64
%tmp3 = add i64 0, %tmp2
%tmp4 = shl i64 %tmp3, 32
%tmp5 = ashr exact i64 %tmp4, 32
%tmp6 = getelementptr inbounds i32, i32 addrspace(1)* %arg0, i64 %tmp5
%tmp7 = load i32, i32 addrspace(1)* %tmp6, align 4
%tmp8 = sext i32 %tmp7 to i64
%tmp9 = getelementptr inbounds i32, i32 addrspace(1)* %arg1, i64 %tmp8
%tmp10 = load i32, i32 addrspace(1)* %tmp9, align 4
%tmp13 = zext i32 %tmp10 to i64
%tmp14 = getelementptr inbounds i32, i32 addrspace(1)* %arg2, i64 %tmp13
%tmp16 = load i32, i32 addrspace(1)* %tmp14, align 16
%Pivot = icmp slt i32 %tmp16, 2
br i1 %Pivot, label %LeafBlock, label %LeafBlock1
LeafBlock: ; preds = %entry
%SwitchLeaf = icmp eq i32 %tmp16, 1
br i1 %SwitchLeaf, label %exit0, label %exit1
LeafBlock1: ; preds = %entry
%SwitchLeaf2 = icmp eq i32 %tmp16, 2
br i1 %SwitchLeaf2, label %exit0, label %exit1
exit0: ; preds = %LeafBlock, %LeafBlock1
store volatile i32 17, i32 addrspace(3)* undef
ret void
exit1: ; preds = %LeafBlock, %LeafBlock1
store volatile i32 9, i32 addrspace(1)* undef
unreachable
}
; The non-uniformity of the branch to the exiting blocks requires
; looking at transitive predecessors.
; IR-LABEL: @indirect_multi_divergent_region_exit_ret_unreachable(
; IR: exit0: ; preds = %Flow2
; IR-NEXT: store volatile i32 17, i32 addrspace(3)* undef
; IR-NEXT: br label %UnifiedReturnBlock
; IR: indirect.exit1:
; IR: %load = load volatile i32, i32 addrspace(1)* undef
; IR: store volatile i32 %load, i32 addrspace(1)* undef
; IR: store volatile i32 9, i32 addrspace(1)* undef
; IR: call void @llvm.amdgcn.unreachable()
; IR-NEXT: br label %Flow2
; IR: UnifiedReturnBlock: ; preds = %exit0, %Flow2
; IR-NEXT: call void @llvm.amdgcn.end.cf(i64 %14)
; IR-NEXT: ret void
define amdgpu_kernel void @indirect_multi_divergent_region_exit_ret_unreachable(i32 addrspace(1)* nocapture %arg0, i32 addrspace(1)* nocapture %arg1, i32 addrspace(1)* nocapture %arg2) #0 {
entry:
%tmp = tail call i32 @llvm.amdgcn.workitem.id.x() #1
%tmp1 = add i32 0, %tmp
%tmp2 = zext i32 %tmp1 to i64
%tmp3 = add i64 0, %tmp2
%tmp4 = shl i64 %tmp3, 32
%tmp5 = ashr exact i64 %tmp4, 32
%tmp6 = getelementptr inbounds i32, i32 addrspace(1)* %arg0, i64 %tmp5
%tmp7 = load i32, i32 addrspace(1)* %tmp6, align 4
%tmp8 = sext i32 %tmp7 to i64
%tmp9 = getelementptr inbounds i32, i32 addrspace(1)* %arg1, i64 %tmp8
%tmp10 = load i32, i32 addrspace(1)* %tmp9, align 4
%tmp13 = zext i32 %tmp10 to i64
%tmp14 = getelementptr inbounds i32, i32 addrspace(1)* %arg2, i64 %tmp13
%tmp16 = load i32, i32 addrspace(1)* %tmp14, align 16
%Pivot = icmp slt i32 %tmp16, 2
br i1 %Pivot, label %LeafBlock, label %LeafBlock1
LeafBlock: ; preds = %entry
%SwitchLeaf = icmp eq i32 %tmp16, 1
br i1 %SwitchLeaf, label %exit0, label %indirect.exit1
LeafBlock1: ; preds = %entry
%SwitchLeaf2 = icmp eq i32 %tmp16, 2
br i1 %SwitchLeaf2, label %exit0, label %indirect.exit1
exit0: ; preds = %LeafBlock, %LeafBlock1
store volatile i32 17, i32 addrspace(3)* undef
ret void
indirect.exit1:
%load = load volatile i32, i32 addrspace(1)* undef
store volatile i32 %load, i32 addrspace(1)* undef
br label %exit1
exit1: ; preds = %LeafBlock, %LeafBlock1
store volatile i32 9, i32 addrspace(1)* undef
unreachable
}
; IR-LABEL: @multi_divergent_region_exit_ret_switch(
define amdgpu_kernel void @multi_divergent_region_exit_ret_switch(i32 addrspace(1)* nocapture %arg0, i32 addrspace(1)* nocapture %arg1, i32 addrspace(1)* nocapture %arg2) #0 {
entry:
%tmp = tail call i32 @llvm.amdgcn.workitem.id.x() #1
%tmp1 = add i32 0, %tmp
%tmp2 = zext i32 %tmp1 to i64
%tmp3 = add i64 0, %tmp2
%tmp4 = shl i64 %tmp3, 32
%tmp5 = ashr exact i64 %tmp4, 32
%tmp6 = getelementptr inbounds i32, i32 addrspace(1)* %arg0, i64 %tmp5
%tmp7 = load i32, i32 addrspace(1)* %tmp6, align 4
%tmp8 = sext i32 %tmp7 to i64
%tmp9 = getelementptr inbounds i32, i32 addrspace(1)* %arg1, i64 %tmp8
%tmp10 = load i32, i32 addrspace(1)* %tmp9, align 4
%tmp13 = zext i32 %tmp10 to i64
%tmp14 = getelementptr inbounds i32, i32 addrspace(1)* %arg2, i64 %tmp13
%tmp16 = load i32, i32 addrspace(1)* %tmp14, align 16
switch i32 %tmp16, label %exit1
[ i32 1, label %LeafBlock
i32 2, label %LeafBlock1
i32 3, label %exit0 ]
LeafBlock: ; preds = %entry
%SwitchLeaf = icmp eq i32 %tmp16, 1
br i1 %SwitchLeaf, label %exit0, label %exit1
LeafBlock1: ; preds = %entry
%SwitchLeaf2 = icmp eq i32 %tmp16, 2
br i1 %SwitchLeaf2, label %exit0, label %exit1
exit0: ; preds = %LeafBlock, %LeafBlock1
store volatile i32 17, i32 addrspace(3)* undef
ret void
exit1: ; preds = %LeafBlock, %LeafBlock1
store volatile i32 9, i32 addrspace(1)* undef
unreachable
}
; IR-LABEL: @divergent_multi_ret_nest_in_uniform_triangle(
define amdgpu_kernel void @divergent_multi_ret_nest_in_uniform_triangle(i32 %arg0) #0 {
entry:
%uniform.cond0 = icmp eq i32 %arg0, 4
br i1 %uniform.cond0, label %divergent.multi.exit.region, label %uniform.ret
divergent.multi.exit.region:
%id.x = tail call i32 @llvm.amdgcn.workitem.id.x()
%divergent.cond0 = icmp eq i32 %id.x, 0
br i1 %divergent.cond0, label %divergent.ret0, label %divergent.ret1
divergent.ret0:
store volatile i32 11, i32 addrspace(3)* undef
ret void
divergent.ret1:
store volatile i32 42, i32 addrspace(3)* undef
ret void
uniform.ret:
store volatile i32 9, i32 addrspace(1)* undef
ret void
}
; IR-LABEL: @divergent_complex_multi_ret_nest_in_uniform_triangle(
define amdgpu_kernel void @divergent_complex_multi_ret_nest_in_uniform_triangle(i32 %arg0) #0 {
entry:
%uniform.cond0 = icmp eq i32 %arg0, 4
br i1 %uniform.cond0, label %divergent.multi.exit.region, label %uniform.ret
divergent.multi.exit.region:
%id.x = tail call i32 @llvm.amdgcn.workitem.id.x()
%divergent.cond0 = icmp eq i32 %id.x, 0
br i1 %divergent.cond0, label %divergent.if, label %divergent.ret1
divergent.if:
%vgpr0 = load volatile float, float addrspace(1)* undef
%divergent.cond1 = fcmp ogt float %vgpr0, 1.0
br i1 %divergent.cond1, label %divergent.then, label %divergent.endif
divergent.then:
%vgpr1 = load volatile float, float addrspace(1)* undef
%divergent.cond2 = fcmp olt float %vgpr1, 4.0
store volatile i32 33, i32 addrspace(1)* undef
br i1 %divergent.cond2, label %divergent.ret0, label %divergent.endif
divergent.endif:
store volatile i32 38, i32 addrspace(1)* undef
br label %divergent.ret0
divergent.ret0:
store volatile i32 11, i32 addrspace(3)* undef
ret void
divergent.ret1:
store volatile i32 42, i32 addrspace(3)* undef
ret void
uniform.ret:
store volatile i32 9, i32 addrspace(1)* undef
ret void
}
; IR-LABEL: @uniform_complex_multi_ret_nest_in_divergent_triangle(
; IR: Flow1: ; preds = %uniform.ret1, %uniform.multi.exit.region
; IR: %8 = phi i1 [ false, %uniform.ret1 ], [ true, %uniform.multi.exit.region ]
; IR: br i1 %8, label %uniform.if, label %Flow2
; IR: Flow: ; preds = %uniform.then, %uniform.if
; IR: %11 = phi i1 [ %10, %uniform.then ], [ %9, %uniform.if ]
; IR: br i1 %11, label %uniform.endif, label %uniform.ret0
; IR: UnifiedReturnBlock: ; preds = %Flow3, %Flow2
; IR-NEXT: call void @llvm.amdgcn.end.cf(i64 %6)
; IR-NEXT: ret void
define amdgpu_kernel void @uniform_complex_multi_ret_nest_in_divergent_triangle(i32 %arg0) #0 {
entry:
%id.x = tail call i32 @llvm.amdgcn.workitem.id.x()
%divergent.cond0 = icmp eq i32 %id.x, 0
br i1 %divergent.cond0, label %uniform.multi.exit.region, label %divergent.ret
uniform.multi.exit.region:
%uniform.cond0 = icmp eq i32 %arg0, 4
br i1 %uniform.cond0, label %uniform.if, label %uniform.ret1
uniform.if:
%sgpr0 = load volatile i32, i32 addrspace(2)* undef
%uniform.cond1 = icmp slt i32 %sgpr0, 1
br i1 %uniform.cond1, label %uniform.then, label %uniform.endif
uniform.then:
%sgpr1 = load volatile i32, i32 addrspace(2)* undef
%uniform.cond2 = icmp sge i32 %sgpr1, 4
store volatile i32 33, i32 addrspace(1)* undef
br i1 %uniform.cond2, label %uniform.ret0, label %uniform.endif
uniform.endif:
store volatile i32 38, i32 addrspace(1)* undef
br label %uniform.ret0
uniform.ret0:
store volatile i32 11, i32 addrspace(3)* undef
ret void
uniform.ret1:
store volatile i32 42, i32 addrspace(3)* undef
ret void
divergent.ret:
store volatile i32 9, i32 addrspace(1)* undef
ret void
}
; IR-LABEL: @multi_divergent_unreachable_exit(
; IR: UnifiedUnreachableBlock:
; IR-NEXT: call void @llvm.amdgcn.unreachable()
; IR-NEXT: br label %UnifiedReturnBlock
; IR: UnifiedReturnBlock:
; IR-NEXT: call void @llvm.amdgcn.end.cf(i64
; IR-NEXT: ret void
define amdgpu_kernel void @multi_divergent_unreachable_exit() #0 {
bb:
%tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
switch i32 %tmp, label %bb3 [
i32 2, label %bb1
i32 0, label %bb2
]
bb1: ; preds = %bb
unreachable
bb2: ; preds = %bb
unreachable
bb3: ; preds = %bb
switch i32 undef, label %bb5 [
i32 2, label %bb4
]
bb4: ; preds = %bb3
ret void
bb5: ; preds = %bb3
unreachable
}
declare i32 @llvm.amdgcn.workitem.id.x() #1
attributes #0 = { nounwind }
attributes #1 = { nounwind readnone }
|
