llvm.org GIT mirror llvm / stable test / CodeGen / AMDGPU / reqd-work-group-size.ll
stable

Tree @stable (Download .tar.gz)

reqd-work-group-size.ll @stableraw · history · blame

  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
; RUN: opt -mtriple=amdgcn-amd-amdhsa -S -amdgpu-lower-kernel-attributes -instcombine %s | FileCheck -enable-var-scope %s

; CHECK-LABEL: @invalid_reqd_work_group_size(
; CHECK: load i16,
define amdgpu_kernel void @invalid_reqd_work_group_size(i16 addrspace(1)* %out) #0 !reqd_work_group_size !1 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  store i16 %group.size.x, i16 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @volatile_load_group_size_x(
; CHECK: load volatile i16,
define amdgpu_kernel void @volatile_load_group_size_x(i16 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load volatile i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  store i16 %group.size.x, i16 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @load_group_size_x(
; CHECK-NEXT: store i16 8,
define amdgpu_kernel void @load_group_size_x(i16 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  store i16 %group.size.x, i16 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @load_group_size_y(
; CHECK-NEXT: store i16 16,
define amdgpu_kernel void @load_group_size_y(i16 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.y = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 6
  %gep.group.size.y.bc = bitcast i8 addrspace(4)* %gep.group.size.y to i16 addrspace(4)*
  %group.size.y = load i16, i16 addrspace(4)* %gep.group.size.y.bc, align 4
  store i16 %group.size.y, i16 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @load_group_size_z(
; CHECK-NEXT: store i16 2,
define amdgpu_kernel void @load_group_size_z(i16 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.z = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 8
  %gep.group.size.z.bc = bitcast i8 addrspace(4)* %gep.group.size.z to i16 addrspace(4)*
  %group.size.z = load i16, i16 addrspace(4)* %gep.group.size.z.bc, align 4
  store i16 %group.size.z, i16 addrspace(1)* %out
  ret void
}

; Metadata uses i64 instead of i32
; CHECK-LABEL: @load_group_size_x_reqd_work_group_size_i64(
; CHECK-NEXT: store i16 8,
define amdgpu_kernel void @load_group_size_x_reqd_work_group_size_i64(i16 addrspace(1)* %out) #0 !reqd_work_group_size !2 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  store i16 %group.size.x, i16 addrspace(1)* %out
  ret void
}

; Metadata uses i16 instead of i32
; CHECK-LABEL: @load_group_size_x_reqd_work_group_size_i16(
; CHECK-NEXT: store i16 8,
define amdgpu_kernel void @load_group_size_x_reqd_work_group_size_i16(i16 addrspace(1)* %out) #0 !reqd_work_group_size !3 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  store i16 %group.size.x, i16 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @use_local_size_x_8_16_2(
; CHECK-NEXT: store i64 8,
define amdgpu_kernel void @use_local_size_x_8_16_2(i64 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  %gep.grid.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 12
  %gep.grid.size.x.bc = bitcast i8 addrspace(4)* %gep.grid.size.x to i32 addrspace(4)*
  %grid.size.x = load i32, i32 addrspace(4)* %gep.grid.size.x.bc, align 4
  %group.id = tail call i32 @llvm.amdgcn.workgroup.id.x()
  %group.size.x.zext = zext i16 %group.size.x to i32
  %group.id_x_group.size.x = mul i32 %group.id, %group.size.x.zext
  %sub = sub i32 %grid.size.x, %group.id_x_group.size.x
  %cmp = icmp ult i32 %sub, %group.size.x.zext
  %select = select i1 %cmp, i32 %sub, i32 %group.size.x.zext
  %zext = zext i32 %select to i64
  store i64 %zext, i64 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @use_local_size_y_8_16_2(
; CHECK-NEXT: store i64 16,
define amdgpu_kernel void @use_local_size_y_8_16_2(i64 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.y = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 6
  %gep.group.size.y.bc = bitcast i8 addrspace(4)* %gep.group.size.y to i16 addrspace(4)*
  %group.size.y = load i16, i16 addrspace(4)* %gep.group.size.y.bc, align 4
  %gep.grid.size.y = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 16
  %gep.grid.size.y.bc = bitcast i8 addrspace(4)* %gep.grid.size.y to i32 addrspace(4)*
  %grid.size.y = load i32, i32 addrspace(4)* %gep.grid.size.y.bc, align 4
  %group.id = tail call i32 @llvm.amdgcn.workgroup.id.y()
  %group.size.y.zext = zext i16 %group.size.y to i32
  %group.id_x_group.size.y = mul i32 %group.id, %group.size.y.zext
  %sub = sub i32 %grid.size.y, %group.id_x_group.size.y
  %cmp = icmp ult i32 %sub, %group.size.y.zext
  %select = select i1 %cmp, i32 %sub, i32 %group.size.y.zext
  %zext = zext i32 %select to i64
  store i64 %zext, i64 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @use_local_size_z_8_16_2(
; CHECK-NEXT: store i64 2,
define amdgpu_kernel void @use_local_size_z_8_16_2(i64 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.z = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 8
  %gep.group.size.z.bc = bitcast i8 addrspace(4)* %gep.group.size.z to i16 addrspace(4)*
  %group.size.z = load i16, i16 addrspace(4)* %gep.group.size.z.bc, align 4
  %gep.grid.size.z = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 20
  %gep.grid.size.z.bc = bitcast i8 addrspace(4)* %gep.grid.size.z to i32 addrspace(4)*
  %grid.size.z = load i32, i32 addrspace(4)* %gep.grid.size.z.bc, align 4
  %group.id = tail call i32 @llvm.amdgcn.workgroup.id.z()
  %group.size.z.zext = zext i16 %group.size.z to i32
  %group.id_x_group.size.z = mul i32 %group.id, %group.size.z.zext
  %sub = sub i32 %grid.size.z, %group.id_x_group.size.z
  %cmp = icmp ult i32 %sub, %group.size.z.zext
  %select = select i1 %cmp, i32 %sub, i32 %group.size.z.zext
  %zext = zext i32 %select to i64
  store i64 %zext, i64 addrspace(1)* %out
  ret void
}

; Simplification on select is invalid, but we can still eliminate the
; load of the group size.

; CHECK-LABEL: @local_size_x_8_16_2_wrong_group_id(
; CHECK: %group.id = tail call i32 @llvm.amdgcn.workgroup.id.y()
; CHECK: %group.id_x_group.size.x = shl i32 %group.id, 3
define amdgpu_kernel void @local_size_x_8_16_2_wrong_group_id(i64 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  %gep.grid.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 12
  %gep.grid.size.x.bc = bitcast i8 addrspace(4)* %gep.grid.size.x to i32 addrspace(4)*
  %grid.size.x = load i32, i32 addrspace(4)* %gep.grid.size.x.bc, align 4
  %group.id = tail call i32 @llvm.amdgcn.workgroup.id.y()
  %group.size.x.zext = zext i16 %group.size.x to i32
  %group.id_x_group.size.x = mul i32 %group.id, %group.size.x.zext
  %sub = sub i32 %grid.size.x, %group.id_x_group.size.x
  %cmp = icmp ult i32 %sub, %group.size.x.zext
  %select = select i1 %cmp, i32 %sub, i32 %group.size.x.zext
  %zext = zext i32 %select to i64
  store i64 %zext, i64 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @local_size_x_8_16_2_wrong_grid_size(
; CHECK: %grid.size.x = load i32, i32 addrspace(4)* %gep.grid.size.x.bc, align 4
; CHECK: %group.id = tail call i32 @llvm.amdgcn.workgroup.id.x()
; CHECK: %group.id_x_group.size.x = shl i32 %group.id, 3
; CHECK: %sub = sub i32 %grid.size.x, %group.id_x_group.size.x
  define amdgpu_kernel void @local_size_x_8_16_2_wrong_grid_size(i64 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  %gep.grid.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 16
  %gep.grid.size.x.bc = bitcast i8 addrspace(4)* %gep.grid.size.x to i32 addrspace(4)*
  %grid.size.x = load i32, i32 addrspace(4)* %gep.grid.size.x.bc, align 4
  %group.id = tail call i32 @llvm.amdgcn.workgroup.id.x()
  %group.size.x.zext = zext i16 %group.size.x to i32
  %group.id_x_group.size.x = mul i32 %group.id, %group.size.x.zext
  %sub = sub i32 %grid.size.x, %group.id_x_group.size.x
  %cmp = icmp ult i32 %sub, %group.size.x.zext
  %select = select i1 %cmp, i32 %sub, i32 %group.size.x.zext
  %zext = zext i32 %select to i64
  store i64 %zext, i64 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @local_size_x_8_16_2_wrong_cmp_type(
; CHECK: %grid.size.x = load i32, i32 addrspace(4)* %gep.grid.size.x.bc, align 4
; CHECK: %group.id = tail call i32 @llvm.amdgcn.workgroup.id.x()
; CHECK: %group.id_x_group.size.x = shl i32 %group.id, 3
; CHECK: %sub = sub i32 %grid.size.x, %group.id_x_group.size.x
; CHECK: %cmp = icmp slt i32 %sub, 8
; CHECK: %select = select i1 %cmp, i32 %sub, i32 8
define amdgpu_kernel void @local_size_x_8_16_2_wrong_cmp_type(i64 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  %gep.grid.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 12
  %gep.grid.size.x.bc = bitcast i8 addrspace(4)* %gep.grid.size.x to i32 addrspace(4)*
  %grid.size.x = load i32, i32 addrspace(4)* %gep.grid.size.x.bc, align 4
  %group.id = tail call i32 @llvm.amdgcn.workgroup.id.x()
  %group.size.x.zext = zext i16 %group.size.x to i32
  %group.id_x_group.size.x = mul i32 %group.id, %group.size.x.zext
  %sub = sub i32 %grid.size.x, %group.id_x_group.size.x
  %cmp = icmp slt i32 %sub, %group.size.x.zext
  %select = select i1 %cmp, i32 %sub, i32 %group.size.x.zext
  %zext = zext i32 %select to i64
  store i64 %zext, i64 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @local_size_x_8_16_2_wrong_select(
; CHECK: %group.id_x_group.size.x = shl i32 %group.id, 3
; CHECK: %sub = sub i32 %grid.size.x, %group.id_x_group.size.x
; CHECK: %1 = icmp ugt i32 %sub, 8
; CHECK: %select = select i1 %1, i32 %sub, i32 8
; CHECK: %zext = zext i32 %select to i64
define amdgpu_kernel void @local_size_x_8_16_2_wrong_select(i64 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  %gep.grid.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 12
  %gep.grid.size.x.bc = bitcast i8 addrspace(4)* %gep.grid.size.x to i32 addrspace(4)*
  %grid.size.x = load i32, i32 addrspace(4)* %gep.grid.size.x.bc, align 4
  %group.id = tail call i32 @llvm.amdgcn.workgroup.id.x()
  %group.size.x.zext = zext i16 %group.size.x to i32
  %group.id_x_group.size.x = mul i32 %group.id, %group.size.x.zext
  %sub = sub i32 %grid.size.x, %group.id_x_group.size.x
  %cmp = icmp ult i32 %sub, %group.size.x.zext
  %select = select i1 %cmp, i32 %group.size.x.zext, i32 %sub
  %zext = zext i32 %select to i64
  store i64 %zext, i64 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @use_local_size_x_8_16_2_wrong_grid_load_size(
; CHECK: %grid.size.x = load i16, i16 addrspace(4)* %gep.grid.size.x.bc, align 4
; CHECK: %grid.size.x.zext = zext i16 %grid.size.x to i32
; CHECK: %group.id = tail call i32 @llvm.amdgcn.workgroup.id.x()
; CHECK: %group.id_x_group.size.x = shl i32 %group.id, 3
; CHECK: %sub = sub i32 %grid.size.x.zext, %group.id_x_group.size.x
define amdgpu_kernel void @use_local_size_x_8_16_2_wrong_grid_load_size(i64 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  %gep.grid.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 12
  %gep.grid.size.x.bc = bitcast i8 addrspace(4)* %gep.grid.size.x to i16 addrspace(4)*
  %grid.size.x = load i16, i16 addrspace(4)* %gep.grid.size.x.bc, align 4
  %grid.size.x.zext = zext i16 %grid.size.x to i32
  %group.id = tail call i32 @llvm.amdgcn.workgroup.id.x()
  %group.size.x.zext = zext i16 %group.size.x to i32
  %group.id_x_group.size.x = mul i32 %group.id, %group.size.x.zext
  %sub = sub i32 %grid.size.x.zext, %group.id_x_group.size.x
  %cmp = icmp ult i32 %sub, %group.size.x.zext
  %select = select i1 %cmp, i32 %sub, i32 %group.size.x.zext
  %zext = zext i32 %select to i64
  store i64 %zext, i64 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @func_group_size_x(
; CHECK-NEXT: ret i32 8
define i32 @func_group_size_x(i16 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  %zext = zext i16 %group.size.x to i32
  ret i32 %zext
}

; CHECK-LABEL: @__ockl_get_local_size_reqd_size(
; CHECK: %group.size = phi i32 [ 2, %bb17 ], [ 16, %bb9 ], [ 8, %bb1 ], [ 1, %bb ]
define i64 @__ockl_get_local_size_reqd_size(i32 %arg) #1 !reqd_work_group_size !0 {
bb:
  %tmp = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr() #2
  switch i32 %arg, label %bb25 [
    i32 0, label %bb1
    i32 1, label %bb9
    i32 2, label %bb17
  ]

bb1:                                              ; preds = %bb
  %tmp2 = tail call i32 @llvm.amdgcn.workgroup.id.x()
  %tmp3 = getelementptr inbounds i8, i8 addrspace(4)* %tmp, i64 12
  %tmp4 = bitcast i8 addrspace(4)* %tmp3 to i32 addrspace(4)*
  %tmp5 = load i32, i32 addrspace(4)* %tmp4, align 4
  %tmp6 = getelementptr inbounds i8, i8 addrspace(4)* %tmp, i64 4
  %tmp7 = bitcast i8 addrspace(4)* %tmp6 to i16 addrspace(4)*
  %tmp8 = load i16, i16 addrspace(4)* %tmp7, align 4
  br label %bb25

bb9:                                              ; preds = %bb
  %tmp10 = tail call i32 @llvm.amdgcn.workgroup.id.y()
  %tmp11 = getelementptr inbounds i8, i8 addrspace(4)* %tmp, i64 16
  %tmp12 = bitcast i8 addrspace(4)* %tmp11 to i32 addrspace(4)*
  %tmp13 = load i32, i32 addrspace(4)* %tmp12, align 8
  %tmp14 = getelementptr inbounds i8, i8 addrspace(4)* %tmp, i64 6
  %tmp15 = bitcast i8 addrspace(4)* %tmp14 to i16 addrspace(4)*
  %tmp16 = load i16, i16 addrspace(4)* %tmp15, align 2
  br label %bb25

bb17:                                             ; preds = %bb
  %tmp18 = tail call i32 @llvm.amdgcn.workgroup.id.z()
  %tmp19 = getelementptr inbounds i8, i8 addrspace(4)* %tmp, i64 20
  %tmp20 = bitcast i8 addrspace(4)* %tmp19 to i32 addrspace(4)*
  %tmp21 = load i32, i32 addrspace(4)* %tmp20, align 4
  %tmp22 = getelementptr inbounds i8, i8 addrspace(4)* %tmp, i64 8
  %tmp23 = bitcast i8 addrspace(4)* %tmp22 to i16 addrspace(4)*
  %tmp24 = load i16, i16 addrspace(4)* %tmp23, align 8
  br label %bb25

bb25:                                             ; preds = %bb17, %bb9, %bb1, %bb
  %tmp26 = phi i32 [ %tmp21, %bb17 ], [ %tmp13, %bb9 ], [ %tmp5, %bb1 ], [ 0, %bb ]
  %group.size = phi i16 [ %tmp24, %bb17 ], [ %tmp16, %bb9 ], [ %tmp8, %bb1 ], [ 1, %bb ]
  %tmp28 = phi i32 [ %tmp18, %bb17 ], [ %tmp10, %bb9 ], [ %tmp2, %bb1 ], [ 0, %bb ]
  %tmp29 = zext i16 %group.size to i32
  %tmp30 = mul i32 %tmp28, %tmp29
  %tmp31 = sub i32 %tmp26, %tmp30
  %tmp32 = icmp ult i32 %tmp31, %tmp29
  %tmp33 = select i1 %tmp32, i32 %tmp31, i32 %tmp29
  %tmp34 = zext i32 %tmp33 to i64
  ret i64 %tmp34
}

; CHECK-LABEL: @all_local_size(
; CHECK-NEXT: store volatile i64 8, i64 addrspace(1)* %out, align 4
; CHECK-NEXT: store volatile i64 16, i64 addrspace(1)* %out, align 4
; CHECK-NEXT: store volatile i64 2, i64 addrspace(1)* %out, align 4
define amdgpu_kernel void @all_local_size(i64 addrspace(1)* nocapture readnone %out) #0 !reqd_work_group_size !0 {
  %tmp.i = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr() #0
  %tmp2.i = tail call i32 @llvm.amdgcn.workgroup.id.x() #0
  %tmp3.i = getelementptr inbounds i8, i8 addrspace(4)* %tmp.i, i64 12
  %tmp4.i = bitcast i8 addrspace(4)* %tmp3.i to i32 addrspace(4)*
  %tmp5.i = load i32, i32 addrspace(4)* %tmp4.i, align 4
  %tmp6.i = getelementptr inbounds i8, i8 addrspace(4)* %tmp.i, i64 4
  %tmp7.i = bitcast i8 addrspace(4)* %tmp6.i to i16 addrspace(4)*
  %tmp8.i = load i16, i16 addrspace(4)* %tmp7.i, align 4
  %tmp29.i = zext i16 %tmp8.i to i32
  %tmp30.i = mul i32 %tmp2.i, %tmp29.i
  %tmp31.i = sub i32 %tmp5.i, %tmp30.i
  %tmp32.i = icmp ult i32 %tmp31.i, %tmp29.i
  %tmp33.i = select i1 %tmp32.i, i32 %tmp31.i, i32 %tmp29.i
  %tmp34.i = zext i32 %tmp33.i to i64
  %tmp10.i = tail call i32 @llvm.amdgcn.workgroup.id.y() #0
  %tmp11.i = getelementptr inbounds i8, i8 addrspace(4)* %tmp.i, i64 16
  %tmp12.i = bitcast i8 addrspace(4)* %tmp11.i to i32 addrspace(4)*
  %tmp13.i = load i32, i32 addrspace(4)* %tmp12.i, align 8
  %tmp14.i = getelementptr inbounds i8, i8 addrspace(4)* %tmp.i, i64 6
  %tmp15.i = bitcast i8 addrspace(4)* %tmp14.i to i16 addrspace(4)*
  %tmp16.i = load i16, i16 addrspace(4)* %tmp15.i, align 2
  %tmp29.i9 = zext i16 %tmp16.i to i32
  %tmp30.i10 = mul i32 %tmp10.i, %tmp29.i9
  %tmp31.i11 = sub i32 %tmp13.i, %tmp30.i10
  %tmp32.i12 = icmp ult i32 %tmp31.i11, %tmp29.i9
  %tmp33.i13 = select i1 %tmp32.i12, i32 %tmp31.i11, i32 %tmp29.i9
  %tmp34.i14 = zext i32 %tmp33.i13 to i64
  %tmp18.i = tail call i32 @llvm.amdgcn.workgroup.id.z() #0
  %tmp19.i = getelementptr inbounds i8, i8 addrspace(4)* %tmp.i, i64 20
  %tmp20.i = bitcast i8 addrspace(4)* %tmp19.i to i32 addrspace(4)*
  %tmp21.i = load i32, i32 addrspace(4)* %tmp20.i, align 4
  %tmp22.i = getelementptr inbounds i8, i8 addrspace(4)* %tmp.i, i64 8
  %tmp23.i = bitcast i8 addrspace(4)* %tmp22.i to i16 addrspace(4)*
  %tmp24.i = load i16, i16 addrspace(4)* %tmp23.i, align 8
  %tmp29.i2 = zext i16 %tmp24.i to i32
  %tmp30.i3 = mul i32 %tmp18.i, %tmp29.i2
  %tmp31.i4 = sub i32 %tmp21.i, %tmp30.i3
  %tmp32.i5 = icmp ult i32 %tmp31.i4, %tmp29.i2
  %tmp33.i6 = select i1 %tmp32.i5, i32 %tmp31.i4, i32 %tmp29.i2
  %tmp34.i7 = zext i32 %tmp33.i6 to i64
  store volatile i64 %tmp34.i, i64 addrspace(1)* %out, align 4
  store volatile i64 %tmp34.i14, i64 addrspace(1)* %out, align 4
  store volatile i64 %tmp34.i7, i64 addrspace(1)* %out, align 4
  ret void
}

; TODO: Should be able to handle this, but not much reason to.
; CHECK-LABEL: @partial_load_group_size_x(
; CHECK-NEXT: %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
; CHECK-NEXT: %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
; CHECK-NEXT: %group.size.x.lo = load i8, i8 addrspace(4)* %gep.group.size.x, align 1
; CHECK-NEXT: store i8 %group.size.x.lo, i8 addrspace(1)* %out, align 1
define amdgpu_kernel void @partial_load_group_size_x(i8 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %group.size.x.lo = load i8, i8 addrspace(4)* %gep.group.size.x, align 1
  store i8 %group.size.x.lo, i8 addrspace(1)* %out
  ret void
}

; TODO: Should be able to handle this
; CHECK-LABEL: @load_group_size_xy_i32(
; CHECK: %group.size.xy = load i32,
; CHECK: store i32 %group.size.xy
define amdgpu_kernel void @load_group_size_xy_i32(i32 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i32 addrspace(4)*
  %group.size.xy = load i32, i32 addrspace(4)* %gep.group.size.x.bc, align 4
  store i32 %group.size.xy, i32 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @load_group_size_x_y_multiple_dispatch_ptr(
; CHECK-NEXT: store volatile i16 8, i16 addrspace(1)* %out, align 2
; CHECK-NEXT: store volatile i16 16, i16 addrspace(1)* %out, align 2
define amdgpu_kernel void @load_group_size_x_y_multiple_dispatch_ptr(i16 addrspace(1)* %out) #0 !reqd_work_group_size !0 {
  %dispatch.ptr0 = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr0, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  store volatile i16 %group.size.x, i16 addrspace(1)* %out

  %dispatch.ptr1 = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.y = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr1, i64 6
  %gep.group.size.y.bc = bitcast i8 addrspace(4)* %gep.group.size.y to i16 addrspace(4)*
  %group.size.y = load i16, i16 addrspace(4)* %gep.group.size.y.bc, align 4
  store volatile i16 %group.size.y, i16 addrspace(1)* %out

  ret void
}

; CHECK-LABEL: @use_local_size_x_uniform_work_group_size(
; CHECK-NEXT: %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
; CHECK-NEXT: %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
; CHECK-NEXT: %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
; CHECK-NEXT: %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
; CHECK-NEXT: %zext = zext i16 %group.size.x to i64
; CHECK-NEXT: store i64 %zext, i64 addrspace(1)* %out, align 4
define amdgpu_kernel void @use_local_size_x_uniform_work_group_size(i64 addrspace(1)* %out) #2 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  %gep.grid.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 12
  %gep.grid.size.x.bc = bitcast i8 addrspace(4)* %gep.grid.size.x to i32 addrspace(4)*
  %grid.size.x = load i32, i32 addrspace(4)* %gep.grid.size.x.bc, align 4
  %group.id = tail call i32 @llvm.amdgcn.workgroup.id.x()
  %group.size.x.zext = zext i16 %group.size.x to i32
  %group.id_x_group.size.x = mul i32 %group.id, %group.size.x.zext
  %sub = sub i32 %grid.size.x, %group.id_x_group.size.x
  %cmp = icmp ult i32 %sub, %group.size.x.zext
  %select = select i1 %cmp, i32 %sub, i32 %group.size.x.zext
  %zext = zext i32 %select to i64
  store i64 %zext, i64 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @use_local_size_x_uniform_work_group_size_false(
; CHECK: icmp ult
; CHECK: select
define amdgpu_kernel void @use_local_size_x_uniform_work_group_size_false(i64 addrspace(1)* %out) #3 {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  %gep.group.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 4
  %gep.group.size.x.bc = bitcast i8 addrspace(4)* %gep.group.size.x to i16 addrspace(4)*
  %group.size.x = load i16, i16 addrspace(4)* %gep.group.size.x.bc, align 4
  %gep.grid.size.x = getelementptr inbounds i8, i8 addrspace(4)* %dispatch.ptr, i64 12
  %gep.grid.size.x.bc = bitcast i8 addrspace(4)* %gep.grid.size.x to i32 addrspace(4)*
  %grid.size.x = load i32, i32 addrspace(4)* %gep.grid.size.x.bc, align 4
  %group.id = tail call i32 @llvm.amdgcn.workgroup.id.x()
  %group.size.x.zext = zext i16 %group.size.x to i32
  %group.id_x_group.size.x = mul i32 %group.id, %group.size.x.zext
  %sub = sub i32 %grid.size.x, %group.id_x_group.size.x
  %cmp = icmp ult i32 %sub, %group.size.x.zext
  %select = select i1 %cmp, i32 %sub, i32 %group.size.x.zext
  %zext = zext i32 %select to i64
  store i64 %zext, i64 addrspace(1)* %out
  ret void
}

; CHECK-LABEL: @no_use_dispatch_ptr(
; CHECK-NEXT: ret void
define amdgpu_kernel void @no_use_dispatch_ptr() {
  %dispatch.ptr = tail call i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr()
  ret void
}

declare i8 addrspace(4)* @llvm.amdgcn.dispatch.ptr() #1
declare i32 @llvm.amdgcn.workgroup.id.x() #1
declare i32 @llvm.amdgcn.workgroup.id.y() #1
declare i32 @llvm.amdgcn.workgroup.id.z() #1

attributes #0 = { nounwind "uniform-work-group-size"="true" }
attributes #1 = { nounwind readnone speculatable }
attributes #2 = { nounwind "uniform-work-group-size"="true" }
attributes #3 = { nounwind "uniform-work-group-size"="false" }

!0 = !{i32 8, i32 16, i32 2}
!1 = !{i32 8, i32 16}
!2 = !{i64 8, i64 16, i64 2}
!3 = !{i16 8, i16 16, i16 2}