llvm.org GIT mirror llvm / 60ccceb test / CodeGen / AMDGPU / load-hi16.ll
60ccceb

Tree @60ccceb (Download .tar.gz)

load-hi16.ll @60cccebraw · 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
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
; RUN: llc -march=amdgcn -mcpu=gfx900 -amdgpu-sroa=0 -mattr=-promote-alloca -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX900 %s
; RUN: llc -march=amdgcn -mcpu=gfx906 -amdgpu-sroa=0 -mattr=-promote-alloca -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX906,NO-D16-HI %s
; RUN: llc -march=amdgcn -mcpu=fiji -amdgpu-sroa=0 -mattr=-promote-alloca -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX803,NO-D16-HI %s

; GCN-LABEL: {{^}}load_local_hi_v2i16_undeflo:
; GCN: s_waitcnt
; GFX900-NEXT: ds_read_u16_d16_hi v0, v0
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: ds_read_u16 v
define <2 x i16> @load_local_hi_v2i16_undeflo(i16 addrspace(3)* %in) #0 {
entry:
  %load = load i16, i16 addrspace(3)* %in
  %build = insertelement <2 x i16> undef, i16 %load, i32 1
  ret <2 x i16> %build
}

; GCN-LABEL: {{^}}load_local_hi_v2i16_reglo:
; GCN: s_waitcnt
; GFX900-NEXT: ds_read_u16_d16_hi v1, v0
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: v_mov_b32_e32 v0, v1
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: ds_read_u16 v
define <2 x i16> @load_local_hi_v2i16_reglo(i16 addrspace(3)* %in, i16 %reg) #0 {
entry:
  %load = load i16, i16 addrspace(3)* %in
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %load, i32 1
  ret <2 x i16> %build1
}

; Show that we get reasonable regalloc without physreg constraints.
; GCN-LABEL: {{^}}load_local_hi_v2i16_reglo_vreg:
; GCN: s_waitcnt
; GFX900-NEXT: ds_read_u16_d16_hi v1, v0
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v[0:1], v1, off{{$}}
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: ds_read_u16 v
define void @load_local_hi_v2i16_reglo_vreg(i16 addrspace(3)* %in, i16 %reg) #0 {
entry:
  %load = load i16, i16 addrspace(3)* %in
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %load, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_local_hi_v2i16_zerolo:
; GCN: s_waitcnt
; GFX900-NEXT: v_mov_b32_e32 v1, 0
; GFX900-NEXT: ds_read_u16_d16_hi v1, v0
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: v_mov_b32_e32 v0, v1
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: ds_read_u16 v
define <2 x i16> @load_local_hi_v2i16_zerolo(i16 addrspace(3)* %in) #0 {
entry:
  %load = load i16, i16 addrspace(3)* %in
  %build = insertelement <2 x i16> zeroinitializer, i16 %load, i32 1
  ret <2 x i16> %build
}

; FIXME: Remove m0 initialization
; GCN-LABEL: {{^}}load_local_hi_v2i16_zerolo_shift:
; GCN: s_waitcnt
; GFX900-NEXT: ds_read_u16 v0, v0
; GFX900-NEXT: s_waitcnt lgkmcnt(0)
; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: ds_read_u16 v
; NO-D16-HI: v_lshlrev_b32_e32 v0, 16, v0
define i32 @load_local_hi_v2i16_zerolo_shift(i16 addrspace(3)* %in) #0 {
entry:
  %load = load i16, i16 addrspace(3)* %in
  %zext = zext i16 %load to i32
  %shift = shl i32 %zext, 16
  ret i32 %shift
}

; GCN-LABEL: {{^}}load_local_hi_v2f16_reglo_vreg:
; GCN: s_waitcnt
; GFX900-NEXT: ds_read_u16_d16_hi v1, v0
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v[0:1], v1, off{{$}}
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: ds_read_u16 v
define void @load_local_hi_v2f16_reglo_vreg(half addrspace(3)* %in, half %reg) #0 {
entry:
  %load = load half, half addrspace(3)* %in
  %build0 = insertelement <2 x half> undef, half %reg, i32 0
  %build1 = insertelement <2 x half> %build0, half %load, i32 1
  store <2 x half> %build1, <2 x half> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_local_hi_v2i16_reglo_vreg_zexti8:
; GCN: s_waitcnt
; GFX900-NEXT: ds_read_u8_d16_hi v1, v0
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v[0:1], v1, off{{$}}
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: ds_read_u8 v
define void @load_local_hi_v2i16_reglo_vreg_zexti8(i8 addrspace(3)* %in, i16 %reg) #0 {
entry:
  %load = load i8, i8 addrspace(3)* %in
  %ext = zext i8 %load to i16
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %ext, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_local_hi_v2i16_reglo_vreg_sexti8:
; GCN: s_waitcnt
; GFX900-NEXT: ds_read_i8_d16_hi v1, v0
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v[0:1], v1, off{{$}}
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: ds_read_i8 v
define void @load_local_hi_v2i16_reglo_vreg_sexti8(i8 addrspace(3)* %in, i16 %reg) #0 {
entry:
  %load = load i8, i8 addrspace(3)* %in
  %ext = sext i8 %load to i16
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %ext, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_global_hi_v2i16_reglo_vreg:
; GCN: s_waitcnt
; GFX900-NEXT: global_load_short_d16_hi v2, v[0:1], off offset:-4094
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64
define void @load_global_hi_v2i16_reglo_vreg(i16 addrspace(1)* %in, i16 %reg) #0 {
entry:
  %gep = getelementptr inbounds i16, i16 addrspace(1)* %in, i64 -2047
  %load = load i16, i16 addrspace(1)* %gep
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %load, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_global_hi_v2f16_reglo_vreg:
; GCN: s_waitcnt
; GFX900-NEXT: global_load_short_d16_hi v2, v[0:1], off offset:-4094
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64
define void @load_global_hi_v2f16_reglo_vreg(half addrspace(1)* %in, half %reg) #0 {
entry:
  %gep = getelementptr inbounds half, half addrspace(1)* %in, i64 -2047
  %load = load half, half addrspace(1)* %gep
  %build0 = insertelement <2 x half> undef, half %reg, i32 0
  %build1 = insertelement <2 x half> %build0, half %load, i32 1
  store <2 x half> %build1, <2 x half> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_global_hi_v2i16_reglo_vreg_zexti8:
; GCN: s_waitcnt
; GFX900-NEXT: global_load_ubyte_d16_hi v2, v[0:1], off offset:-4095
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64
define void @load_global_hi_v2i16_reglo_vreg_zexti8(i8 addrspace(1)* %in, i16 %reg) #0 {
entry:
  %gep = getelementptr inbounds i8, i8 addrspace(1)* %in, i64 -4095
  %load = load i8, i8 addrspace(1)* %gep
  %ext = zext i8 %load to i16
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %ext, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_global_hi_v2i16_reglo_vreg_sexti8:
; GCN: s_waitcnt
; GFX900-NEXT: global_load_sbyte_d16_hi v2, v[0:1], off offset:-4095
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64
define void @load_global_hi_v2i16_reglo_vreg_sexti8(i8 addrspace(1)* %in, i16 %reg) #0 {
entry:
  %gep = getelementptr inbounds i8, i8 addrspace(1)* %in, i64 -4095
  %load = load i8, i8 addrspace(1)* %gep
  %ext = sext i8 %load to i16
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %ext, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: load_flat_hi_v2i16_reglo_vreg:
; GCN: s_waitcnt
; GFX900-NEXT: flat_load_short_d16_hi v2, v[0:1]
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v[0:1], v2
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: flat_load_ushort v{{[0-9]+}}
; GFX803: v_lshlrev_b32_e32 v{{[0-9]+}}, 16,
; GFX803: v_or_b32_sdwa
; GFX906: v_lshl_or_b32 v{{[0-9]+}}, v{{[0-9]+}}, 16,
define void @load_flat_hi_v2i16_reglo_vreg(i16* %in, i16 %reg) #0 {
entry:
  %load = load i16, i16* %in
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %load, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_flat_hi_v2f16_reglo_vreg:
; GCN: s_waitcnt
; GFX900-NEXT: flat_load_short_d16_hi v2, v[0:1]
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v[0:1], v2
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: flat_load_ushort v{{[0-9]+}}
; GFX803: v_lshlrev_b32_e32 v{{[0-9]+}}, 16,
; GFX803: v_or_b32_sdwa
; GFX906: v_lshl_or_b32 v{{[0-9]+}}, v{{[0-9]+}}, 16,
define void @load_flat_hi_v2f16_reglo_vreg(half* %in, half %reg) #0 {
entry:
  %load = load half, half* %in
  %build0 = insertelement <2 x half> undef, half %reg, i32 0
  %build1 = insertelement <2 x half> %build0, half %load, i32 1
  store <2 x half> %build1, <2 x half> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_flat_hi_v2i16_reglo_vreg_zexti8:
; GCN: s_waitcnt
; GFX900-NEXT: flat_load_ubyte_d16_hi v2, v[0:1]
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v[0:1], v2
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: flat_load_ubyte v{{[0-9]+}}
; GFX803: v_lshlrev_b32_e32 v{{[0-9]+}}, 16,
; GFX803: v_or_b32_sdwa
; GFX906: v_lshl_or_b32 v{{[0-9]+}}, v{{[0-9]+}}, 16,
define void @load_flat_hi_v2i16_reglo_vreg_zexti8(i8* %in, i16 %reg) #0 {
entry:
  %load = load i8, i8* %in
  %ext = zext i8 %load to i16
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %ext, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_flat_hi_v2i16_reglo_vreg_sexti8:
; GCN: s_waitcnt
; GFX900-NEXT: flat_load_sbyte_d16_hi v2, v[0:1]
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v[0:1], v2
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: flat_load_sbyte v{{[0-9]+}}
; GFX803: v_lshlrev_b32_e32 v{{[0-9]+}}, 16,
; GFX803: v_or_b32_sdwa
; GFX906: v_lshl_or_b32 v{{[0-9]+}}, v{{[0-9]+}}, 16,
define void @load_flat_hi_v2i16_reglo_vreg_sexti8(i8* %in, i16 %reg) #0 {
entry:
  %load = load i8, i8* %in
  %ext = sext i8 %load to i16
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %ext, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_private_hi_v2i16_reglo_vreg:
; GCN: s_waitcnt
; GFX900: buffer_load_short_d16_hi v0, off, s[0:3], s5 offset:4094{{$}}
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, v0
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: buffer_load_ushort v{{[0-9]+}}, off, s[0:3], s5 offset:4094{{$}}
define void @load_private_hi_v2i16_reglo_vreg(i16 addrspace(5)* byval %in, i16 %reg) #0 {
entry:
  %gep = getelementptr inbounds i16, i16 addrspace(5)* %in, i64 2045
  %load = load i16, i16 addrspace(5)* %gep
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %load, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_private_hi_v2f16_reglo_vreg:
; GCN: s_waitcnt
; GFX900: buffer_load_short_d16_hi v0, off, s[0:3], s5 offset:4094{{$}}
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, v0
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: buffer_load_ushort v{{[0-9]+}}, off, s[0:3], s5 offset:4094{{$}}
define void @load_private_hi_v2f16_reglo_vreg(half addrspace(5)* byval %in, half %reg) #0 {
entry:
  %gep = getelementptr inbounds half, half addrspace(5)* %in, i64 2045
  %load = load half, half addrspace(5)* %gep
  %build0 = insertelement <2 x half> undef, half %reg, i32 0
  %build1 = insertelement <2 x half> %build0, half %load, i32 1
  store <2 x half> %build1, <2 x half> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_private_hi_v2i16_reglo_vreg_nooff:
; GCN: s_waitcnt
; GFX900: buffer_load_short_d16_hi v0, off, s[0:3], s4 offset:4094{{$}}
; GFX900: s_waitcnt
; GFX900-NEXT: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, v0
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: buffer_load_ushort v{{[0-9]+}}, off, s[0:3], s4 offset:4094{{$}}
define void @load_private_hi_v2i16_reglo_vreg_nooff(i16 addrspace(5)* byval %in, i16 %reg) #0 {
entry:
  %load = load volatile i16, i16 addrspace(5)* inttoptr (i32 4094 to i16 addrspace(5)*)
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %load, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_private_hi_v2f16_reglo_vreg_nooff:
; GCN: s_waitcnt
; GFX900-NEXT: buffer_load_short_d16_hi v1, off, s[0:3], s4 offset:4094{{$}}
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, v1
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: buffer_load_ushort v{{[0-9]+}}, off, s[0:3], s4 offset:4094{{$}}
define void @load_private_hi_v2f16_reglo_vreg_nooff(half addrspace(5)* %in, half %reg) #0 {
entry:
  %load = load volatile half, half addrspace(5)* inttoptr (i32 4094 to half addrspace(5)*)
  %build0 = insertelement <2 x half> undef, half %reg, i32 0
  %build1 = insertelement <2 x half> %build0, half %load, i32 1
  store <2 x half> %build1, <2 x half> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_private_hi_v2i16_reglo_vreg_zexti8:
; GCN: s_waitcnt
; GFX900: buffer_load_ubyte_d16_hi v0, off, s[0:3], s5 offset:4095{{$}}
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, v0
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: buffer_load_ubyte v{{[0-9]+}}, off, s[0:3], s5 offset:4095{{$}}
define void @load_private_hi_v2i16_reglo_vreg_zexti8(i8 addrspace(5)* byval %in, i16 %reg) #0 {
entry:
  %gep = getelementptr inbounds i8, i8 addrspace(5)* %in, i64 4091
  %load = load i8, i8 addrspace(5)* %gep
  %ext = zext i8 %load to i16
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %ext, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_private_hi_v2i16_reglo_vreg_sexti8:
; GCN: s_waitcnt
; GFX900: buffer_load_sbyte_d16_hi v0, off, s[0:3], s5 offset:4095{{$}}
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, v0
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: buffer_load_sbyte v{{[0-9]+}}, off, s[0:3], s5 offset:4095{{$}}
define void @load_private_hi_v2i16_reglo_vreg_sexti8(i8 addrspace(5)* byval %in, i16 %reg) #0 {
entry:
  %gep = getelementptr inbounds i8, i8 addrspace(5)* %in, i64 4091
  %load = load i8, i8 addrspace(5)* %gep
  %ext = sext i8 %load to i16
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %ext, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_private_hi_v2i16_reglo_vreg_nooff_zexti8:
; GCN: s_waitcnt
; GFX900-NEXT: buffer_load_ubyte_d16_hi v1, off, s[0:3], s4 offset:4094{{$}}
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, v1
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: buffer_load_ubyte v0, off, s[0:3], s4 offset:4094{{$}}
define void @load_private_hi_v2i16_reglo_vreg_nooff_zexti8(i8 addrspace(5)* %in, i16 %reg) #0 {
entry:
  %load = load volatile i8, i8 addrspace(5)* inttoptr (i32 4094 to i8 addrspace(5)*)
  %ext = zext i8 %load to i16
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %ext, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_private_hi_v2i16_reglo_vreg_nooff_sexti8:
; GCN: s_waitcnt
; GFX900-NEXT: buffer_load_sbyte_d16_hi v1, off, s[0:3], s4 offset:4094{{$}}
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, v1
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: buffer_load_sbyte v0, off, s[0:3], s4 offset:4094{{$}}
define void @load_private_hi_v2i16_reglo_vreg_nooff_sexti8(i8 addrspace(5)* %in, i16 %reg) #0 {
entry:
  %load = load volatile i8, i8 addrspace(5)* inttoptr (i32 4094 to i8 addrspace(5)*)
  %ext = sext i8 %load to i16
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %ext, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_private_hi_v2f16_reglo_vreg_nooff_zexti8:
; GCN: s_waitcnt
; GFX900-NEXT: buffer_load_ubyte_d16_hi v1, off, s[0:3], s4 offset:4094{{$}}
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, v1
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; NO-D16-HI: buffer_load_ubyte v0, off, s[0:3], s4 offset:4094{{$}}
define void @load_private_hi_v2f16_reglo_vreg_nooff_zexti8(i8 addrspace(5)* %in, half %reg) #0 {
entry:
  %load = load volatile i8, i8 addrspace(5)* inttoptr (i32 4094 to i8 addrspace(5)*)
  %ext = zext i8 %load to i16
  %bc.ext = bitcast i16 %ext to half
  %build0 = insertelement <2 x half> undef, half %reg, i32 0
  %build1 = insertelement <2 x half> %build0, half %bc.ext, i32 1
  store <2 x half> %build1, <2 x half> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_constant_hi_v2i16_reglo_vreg:
; GCN: s_waitcnt
; GFX900-NEXT: global_load_short_d16_hi v2, v[0:1], off offset:-4094
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; GFX803: flat_load_ushort
; GFX906: global_load_ushort
define void @load_constant_hi_v2i16_reglo_vreg(i16 addrspace(4)* %in, i16 %reg) #0 {
entry:
  %gep = getelementptr inbounds i16, i16 addrspace(4)* %in, i64 -2047
  %load = load i16, i16 addrspace(4)* %gep
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %load, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: load_constant_hi_v2f16_reglo_vreg
; GCN: s_waitcnt
; GFX900-NEXT: global_load_short_d16_hi v2, v[0:1], off offset:-4094
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_store_dword
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64

; GFX803: flat_load_ushort
; GFX906: global_load_ushort
define void @load_constant_hi_v2f16_reglo_vreg(half addrspace(4)* %in, half %reg) #0 {
entry:
  %gep = getelementptr inbounds half, half addrspace(4)* %in, i64 -2047
  %load = load half, half addrspace(4)* %gep
  %build0 = insertelement <2 x half> undef, half %reg, i32 0
  %build1 = insertelement <2 x half> %build0, half %load, i32 1
  store <2 x half> %build1, <2 x half> addrspace(1)* undef
  ret void
}

; Local object gives known offset, so requires converting from offen
; to offset variant.

; GCN-LABEL: {{^}}load_private_hi_v2i16_reglo_vreg_to_offset:
; GFX900: buffer_store_dword
; GFX900-NEXT: buffer_load_short_d16_hi v{{[0-9]+}}, off, s[0:3], s5 offset:4094
define void @load_private_hi_v2i16_reglo_vreg_to_offset(i16 %reg) #0 {
entry:
  %obj0 = alloca [10 x i32], align 4, addrspace(5)
  %obj1 = alloca [4096 x i16], align 2, addrspace(5)
  %bc = bitcast [10 x i32] addrspace(5)* %obj0 to i32 addrspace(5)*
  store volatile i32 123, i32 addrspace(5)* %bc
  %gep = getelementptr inbounds [4096 x i16], [4096 x i16] addrspace(5)* %obj1, i32 0, i32 2025
  %load = load i16, i16 addrspace(5)* %gep
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %load, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_private_hi_v2i16_reglo_vreg_sexti8_to_offset:
; GFX900: buffer_store_dword
; GFX900-NEXT: buffer_load_sbyte_d16_hi v{{[0-9]+}}, off, s[0:3], s5 offset:4095
define void @load_private_hi_v2i16_reglo_vreg_sexti8_to_offset(i16 %reg) #0 {
entry:
  %obj0 = alloca [10 x i32], align 4, addrspace(5)
  %obj1 = alloca [4096 x i8], align 2, addrspace(5)
  %bc = bitcast [10 x i32] addrspace(5)* %obj0 to i32 addrspace(5)*
  store volatile i32 123, i32 addrspace(5)* %bc
  %gep = getelementptr inbounds [4096 x i8], [4096 x i8] addrspace(5)* %obj1, i32 0, i32 4051
  %load = load i8, i8 addrspace(5)* %gep
  %ext = sext i8 %load to i16
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %ext, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; GCN-LABEL: {{^}}load_private_hi_v2i16_reglo_vreg_zexti8_to_offset:
; GFX900: buffer_store_dword
; GFX900-NEXT: buffer_load_ubyte_d16_hi v{{[0-9]+}}, off, s[0:3], s5 offset:4095
define void @load_private_hi_v2i16_reglo_vreg_zexti8_to_offset(i16 %reg) #0 {
entry:
  %obj0 = alloca [10 x i32], align 4, addrspace(5)
  %obj1 = alloca [4096 x i8], align 2, addrspace(5)
  %bc = bitcast [10 x i32] addrspace(5)* %obj0 to i32 addrspace(5)*
  store volatile i32 123, i32 addrspace(5)* %bc
  %gep = getelementptr inbounds [4096 x i8], [4096 x i8] addrspace(5)* %obj1, i32 0, i32 4051
  %load = load i8, i8 addrspace(5)* %gep
  %ext = zext i8 %load to i16
  %build0 = insertelement <2 x i16> undef, i16 %reg, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %ext, i32 1
  store <2 x i16> %build1, <2 x i16> addrspace(1)* undef
  ret void
}

; FIXME: Remove m0 init and waitcnt between reads
; FIXME: Is there a cost to using the extload over not?
; GCN-LABEL: {{^}}load_local_v2i16_split:
; GCN: s_waitcnt
; GFX900-NEXT: ds_read_u16 v1, v0
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: ds_read_u16_d16_hi v1, v0 offset:2
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: v_mov_b32_e32 v0, v1
; GFX900-NEXT: s_setpc_b64
define <2 x i16> @load_local_v2i16_split(i16 addrspace(3)* %in) #0 {
entry:
  %gep = getelementptr inbounds i16, i16 addrspace(3)* %in, i32 1
  %load0 = load volatile i16, i16 addrspace(3)* %in
  %load1 = load volatile i16, i16 addrspace(3)* %gep
  %build0 = insertelement <2 x i16> undef, i16 %load0, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %load1, i32 1
  ret <2 x i16> %build1
}

; FIXME: Remove waitcnt between reads
; GCN-LABEL: {{^}}load_global_v2i16_split:
; GCN: s_waitcnt
; GFX900-NEXT: global_load_ushort v2
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_load_short_d16_hi v2
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: v_mov_b32_e32 v0, v2
; GFX900-NEXT: s_setpc_b64
define <2 x i16> @load_global_v2i16_split(i16 addrspace(1)* %in) #0 {
entry:
  %gep = getelementptr inbounds i16, i16 addrspace(1)* %in, i64 1
  %load0 = load volatile i16, i16 addrspace(1)* %in
  %load1 = load volatile i16, i16 addrspace(1)* %gep
  %build0 = insertelement <2 x i16> undef, i16 %load0, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %load1, i32 1
  ret <2 x i16> %build1
}

; FIXME: Remove waitcnt between reads
; GCN-LABEL: {{^}}load_flat_v2i16_split:
; GCN: s_waitcnt
; GFX900-NEXT: flat_load_ushort v2
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: flat_load_short_d16_hi v2
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: v_mov_b32_e32 v0, v2
; GFX900-NEXT: s_setpc_b64
define <2 x i16> @load_flat_v2i16_split(i16* %in) #0 {
entry:
  %gep = getelementptr inbounds i16, i16* %in, i64 1
  %load0 = load volatile i16, i16* %in
  %load1 = load volatile i16, i16* %gep
  %build0 = insertelement <2 x i16> undef, i16 %load0, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %load1, i32 1
  ret <2 x i16> %build1
}

; FIXME: Remove waitcnt between reads
; GCN-LABEL: {{^}}load_constant_v2i16_split:
; GCN: s_waitcnt
; GFX900-NEXT: global_load_ushort v2
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: global_load_short_d16_hi v2
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: v_mov_b32_e32 v0, v2
; GFX900-NEXT: s_setpc_b64
define <2 x i16> @load_constant_v2i16_split(i16 addrspace(4)* %in) #0 {
entry:
  %gep = getelementptr inbounds i16, i16 addrspace(4)* %in, i64 1
  %load0 = load volatile i16, i16 addrspace(4)* %in
  %load1 = load volatile i16, i16 addrspace(4)* %gep
  %build0 = insertelement <2 x i16> undef, i16 %load0, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %load1, i32 1
  ret <2 x i16> %build1
}

; FIXME: Remove m0 init and waitcnt between reads
; FIXME: Is there a cost to using the extload over not?
; GCN-LABEL: {{^}}load_private_v2i16_split:
; GCN: s_waitcnt
; GFX900: buffer_load_ushort v0, off, s[0:3], s5 offset:4{{$}}
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: buffer_load_short_d16_hi v0, off, s[0:3], s5 offset:6
; GFX900-NEXT: s_waitcnt
; GFX900-NEXT: s_setpc_b64
define <2 x i16> @load_private_v2i16_split(i16 addrspace(5)* byval %in) #0 {
entry:
  %gep = getelementptr inbounds i16, i16 addrspace(5)* %in, i32 1
  %load0 = load volatile i16, i16 addrspace(5)* %in
  %load1 = load volatile i16, i16 addrspace(5)* %gep
  %build0 = insertelement <2 x i16> undef, i16 %load0, i32 0
  %build1 = insertelement <2 x i16> %build0, i16 %load1, i32 1
  ret <2 x i16> %build1
}

attributes #0 = { nounwind }