summaryrefslogtreecommitdiffstats
path: root/drivers/gpu/drm/amd/amdkfd/cwsr_trap_handler_gfx10.asm
blob: 5081f91190b84013a63c52a8c4176ef336323eb1 (plain)
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
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
/*
 * Copyright 2018 Advanced Micro Devices, Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 * OTHER DEALINGS IN THE SOFTWARE.
 */

/* To compile this assembly code:
 *
 * Navi1x:
 *   cpp -DASIC_TARGET_NAVI1X=1 cwsr_trap_handler_gfx10.asm -P -o nv1x.sp3
 *   sp3-nv1x nv1x.sp3 -hex nv1x.hex
 *
 * Others:
 *   cpp -DASIC_TARGET_NAVI1X=0 cwsr_trap_handler_gfx10.asm -P -o gfx10.sp3
 *   sp3-gfx10 gfx10.sp3 -hex gfx10.hex
 */

#define NO_SQC_STORE !ASIC_TARGET_NAVI1X

var SINGLE_STEP_MISSED_WORKAROUND		= 1	//workaround for lost MODE.DEBUG_EN exception when SAVECTX raised

var SQ_WAVE_STATUS_INST_ATC_SHIFT		= 23
var SQ_WAVE_STATUS_INST_ATC_MASK		= 0x00800000
var SQ_WAVE_STATUS_SPI_PRIO_MASK		= 0x00000006
var SQ_WAVE_STATUS_HALT_MASK			= 0x2000

var SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT		= 12
var SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE		= 9
var SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT		= 8
var SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE		= 6
var SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SHIFT		= 24
var SQ_WAVE_GPR_ALLOC_SGPR_SIZE_SIZE		= 4
var SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SHIFT	= 24
var SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SIZE	= 4
var SQ_WAVE_IB_STS2_WAVE64_SHIFT		= 11
var SQ_WAVE_IB_STS2_WAVE64_SIZE			= 1

var SQ_WAVE_TRAPSTS_SAVECTX_MASK		= 0x400
var SQ_WAVE_TRAPSTS_EXCE_MASK			= 0x1FF
var SQ_WAVE_TRAPSTS_SAVECTX_SHIFT		= 10
var SQ_WAVE_TRAPSTS_MEM_VIOL_MASK		= 0x100
var SQ_WAVE_TRAPSTS_MEM_VIOL_SHIFT		= 8
var SQ_WAVE_TRAPSTS_PRE_SAVECTX_MASK		= 0x3FF
var SQ_WAVE_TRAPSTS_PRE_SAVECTX_SHIFT		= 0x0
var SQ_WAVE_TRAPSTS_PRE_SAVECTX_SIZE		= 10
var SQ_WAVE_TRAPSTS_POST_SAVECTX_MASK		= 0xFFFFF800
var SQ_WAVE_TRAPSTS_POST_SAVECTX_SHIFT		= 11
var SQ_WAVE_TRAPSTS_POST_SAVECTX_SIZE		= 21
var SQ_WAVE_TRAPSTS_ILLEGAL_INST_MASK		= 0x800

var SQ_WAVE_IB_STS_RCNT_SHIFT			= 16
var SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT		= 15
var SQ_WAVE_IB_STS_REPLAY_W64H_SHIFT		= 25
var SQ_WAVE_IB_STS_REPLAY_W64H_SIZE		= 1
var SQ_WAVE_IB_STS_REPLAY_W64H_MASK		= 0x02000000
var SQ_WAVE_IB_STS_FIRST_REPLAY_SIZE		= 1
var SQ_WAVE_IB_STS_RCNT_SIZE			= 6
var SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK	= 0x003F8000
var SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK_NEG	= 0x00007FFF

var SQ_WAVE_MODE_DEBUG_EN_MASK			= 0x800

var SQ_BUF_RSRC_WORD1_ATC_SHIFT			= 24
var SQ_BUF_RSRC_WORD3_MTYPE_SHIFT		= 27

// bits [31:24] unused by SPI debug data
var TTMP11_SAVE_REPLAY_W64H_SHIFT		= 31
var TTMP11_SAVE_REPLAY_W64H_MASK		= 0x80000000
var TTMP11_SAVE_RCNT_FIRST_REPLAY_SHIFT		= 24
var TTMP11_SAVE_RCNT_FIRST_REPLAY_MASK		= 0x7F000000

// SQ_SEL_X/Y/Z/W, BUF_NUM_FORMAT_FLOAT, (0 for MUBUF stride[17:14]
// when ADD_TID_ENABLE and BUF_DATA_FORMAT_32 for MTBUF), ADD_TID_ENABLE
var S_SAVE_BUF_RSRC_WORD1_STRIDE		= 0x00040000
var S_SAVE_BUF_RSRC_WORD3_MISC			= 0x10807FAC

var S_SAVE_SPI_INIT_ATC_MASK			= 0x08000000
var S_SAVE_SPI_INIT_ATC_SHIFT			= 27
var S_SAVE_SPI_INIT_MTYPE_MASK			= 0x70000000
var S_SAVE_SPI_INIT_MTYPE_SHIFT			= 28
var S_SAVE_SPI_INIT_FIRST_WAVE_MASK		= 0x04000000
var S_SAVE_SPI_INIT_FIRST_WAVE_SHIFT		= 26

var S_SAVE_PC_HI_RCNT_SHIFT			= 26
var S_SAVE_PC_HI_RCNT_MASK			= 0xFC000000
var S_SAVE_PC_HI_FIRST_REPLAY_SHIFT		= 25
var S_SAVE_PC_HI_FIRST_REPLAY_MASK		= 0x02000000
var S_SAVE_PC_HI_REPLAY_W64H_SHIFT		= 24
var S_SAVE_PC_HI_REPLAY_W64H_MASK		= 0x01000000

var s_sgpr_save_num				= 108

var s_save_spi_init_lo				= exec_lo
var s_save_spi_init_hi				= exec_hi
var s_save_pc_lo				= ttmp0
var s_save_pc_hi				= ttmp1
var s_save_exec_lo				= ttmp2
var s_save_exec_hi				= ttmp3
var s_save_status				= ttmp12
var s_save_trapsts				= ttmp15
var s_save_xnack_mask				= s_save_trapsts
var s_wave_size					= ttmp7
var s_save_buf_rsrc0				= ttmp8
var s_save_buf_rsrc1				= ttmp9
var s_save_buf_rsrc2				= ttmp10
var s_save_buf_rsrc3				= ttmp11
var s_save_mem_offset				= ttmp4
var s_save_alloc_size				= s_save_trapsts
var s_save_tmp					= ttmp14
var s_save_m0					= ttmp5
var s_save_ttmps_lo				= s_save_tmp
var s_save_ttmps_hi				= s_save_trapsts

var S_RESTORE_BUF_RSRC_WORD1_STRIDE		= S_SAVE_BUF_RSRC_WORD1_STRIDE
var S_RESTORE_BUF_RSRC_WORD3_MISC		= S_SAVE_BUF_RSRC_WORD3_MISC

var S_RESTORE_SPI_INIT_ATC_MASK			= 0x08000000
var S_RESTORE_SPI_INIT_ATC_SHIFT		= 27
var S_RESTORE_SPI_INIT_MTYPE_MASK		= 0x70000000
var S_RESTORE_SPI_INIT_MTYPE_SHIFT		= 28
var S_RESTORE_SPI_INIT_FIRST_WAVE_MASK		= 0x04000000
var S_RESTORE_SPI_INIT_FIRST_WAVE_SHIFT		= 26
var S_WAVE_SIZE					= 25

var S_RESTORE_PC_HI_RCNT_SHIFT			= S_SAVE_PC_HI_RCNT_SHIFT
var S_RESTORE_PC_HI_RCNT_MASK			= S_SAVE_PC_HI_RCNT_MASK
var S_RESTORE_PC_HI_FIRST_REPLAY_SHIFT		= S_SAVE_PC_HI_FIRST_REPLAY_SHIFT
var S_RESTORE_PC_HI_FIRST_REPLAY_MASK		= S_SAVE_PC_HI_FIRST_REPLAY_MASK

var s_restore_spi_init_lo			= exec_lo
var s_restore_spi_init_hi			= exec_hi
var s_restore_mem_offset			= ttmp12
var s_restore_alloc_size			= ttmp3
var s_restore_tmp				= ttmp2
var s_restore_mem_offset_save			= s_restore_tmp
var s_restore_m0				= s_restore_alloc_size
var s_restore_mode				= ttmp7
var s_restore_flat_scratch			= s_restore_tmp
var s_restore_pc_lo				= ttmp0
var s_restore_pc_hi				= ttmp1
var s_restore_exec_lo				= ttmp4
var s_restore_exec_hi				= ttmp5
var s_restore_status				= ttmp14
var s_restore_trapsts				= ttmp15
var s_restore_xnack_mask			= ttmp13
var s_restore_buf_rsrc0				= ttmp8
var s_restore_buf_rsrc1				= ttmp9
var s_restore_buf_rsrc2				= ttmp10
var s_restore_buf_rsrc3				= ttmp11
var s_restore_size				= ttmp6
var s_restore_ttmps_lo				= s_restore_tmp
var s_restore_ttmps_hi				= s_restore_alloc_size

shader main
	asic(DEFAULT)
	type(CS)
	wave_size(32)

	s_branch	L_SKIP_RESTORE						//NOT restore. might be a regular trap or save

L_JUMP_TO_RESTORE:
	s_branch	L_RESTORE

L_SKIP_RESTORE:
	s_getreg_b32	s_save_status, hwreg(HW_REG_STATUS)			//save STATUS since we will change SCC
	s_andn2_b32	s_save_status, s_save_status, SQ_WAVE_STATUS_SPI_PRIO_MASK

if SINGLE_STEP_MISSED_WORKAROUND
	// No single step exceptions if MODE.DEBUG_EN=0.
	s_getreg_b32    ttmp2, hwreg(HW_REG_MODE)
	s_and_b32       ttmp2, ttmp2, SQ_WAVE_MODE_DEBUG_EN_MASK
	s_cbranch_scc0  L_NO_SINGLE_STEP_WORKAROUND

	// Second-level trap already handled exception if STATUS.HALT=1.
	s_and_b32       ttmp2, s_save_status, SQ_WAVE_STATUS_HALT_MASK

	// Prioritize single step exception over context save.
	// Second-level trap will halt wave and RFE, re-entering for SAVECTX.
	s_cbranch_scc0  L_FETCH_2ND_TRAP

L_NO_SINGLE_STEP_WORKAROUND:
end


	s_getreg_b32	s_save_trapsts, hwreg(HW_REG_TRAPSTS)
	s_and_b32	ttmp2, s_save_trapsts, SQ_WAVE_TRAPSTS_SAVECTX_MASK	//check whether this is for save
	s_cbranch_scc1	L_SAVE

	// If STATUS.MEM_VIOL is asserted then halt the wave to prevent
	// the exception raising again and blocking context save.
	s_and_b32	ttmp2, s_save_trapsts, SQ_WAVE_TRAPSTS_MEM_VIOL_MASK
	s_cbranch_scc0	L_FETCH_2ND_TRAP
	s_or_b32	s_save_status, s_save_status, SQ_WAVE_STATUS_HALT_MASK

L_FETCH_2ND_TRAP:

#if ASIC_TARGET_NAVI1X
	// Preserve and clear scalar XNACK state before issuing scalar loads.
	// Save IB_STS.REPLAY_W64H[25], RCNT[21:16], FIRST_REPLAY[15] into
	// unused space ttmp11[31:24].
	s_andn2_b32	ttmp11, ttmp11, (TTMP11_SAVE_REPLAY_W64H_MASK | TTMP11_SAVE_RCNT_FIRST_REPLAY_MASK)
	s_getreg_b32	ttmp2, hwreg(HW_REG_IB_STS)
	s_and_b32	ttmp3, ttmp2, SQ_WAVE_IB_STS_REPLAY_W64H_MASK
	s_lshl_b32	ttmp3, ttmp3, (TTMP11_SAVE_REPLAY_W64H_SHIFT - SQ_WAVE_IB_STS_REPLAY_W64H_SHIFT)
	s_or_b32	ttmp11, ttmp11, ttmp3
	s_and_b32	ttmp3, ttmp2, SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK
	s_lshl_b32	ttmp3, ttmp3, (TTMP11_SAVE_RCNT_FIRST_REPLAY_SHIFT - SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT)
	s_or_b32	ttmp11, ttmp11, ttmp3
	s_andn2_b32	ttmp2, ttmp2, (SQ_WAVE_IB_STS_REPLAY_W64H_MASK | SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK)
	s_setreg_b32	hwreg(HW_REG_IB_STS), ttmp2
#endif

	// Read second-level TBA/TMA from first-level TMA and jump if available.
	// ttmp[2:5] and ttmp12 can be used (others hold SPI-initialized debug data)
	// ttmp12 holds SQ_WAVE_STATUS
	s_getreg_b32	ttmp14, hwreg(HW_REG_SHADER_TMA_LO)
	s_getreg_b32	ttmp15, hwreg(HW_REG_SHADER_TMA_HI)
	s_lshl_b64	[ttmp14, ttmp15], [ttmp14, ttmp15], 0x8
	s_load_dwordx2	[ttmp2, ttmp3], [ttmp14, ttmp15], 0x0 glc:1		// second-level TBA
	s_waitcnt	lgkmcnt(0)
	s_load_dwordx2	[ttmp14, ttmp15], [ttmp14, ttmp15], 0x8 glc:1		// second-level TMA
	s_waitcnt	lgkmcnt(0)
	s_and_b64	[ttmp2, ttmp3], [ttmp2, ttmp3], [ttmp2, ttmp3]
	s_cbranch_scc0	L_NO_NEXT_TRAP						// second-level trap handler not been set
	s_setpc_b64	[ttmp2, ttmp3]						// jump to second-level trap handler

L_NO_NEXT_TRAP:
	s_getreg_b32	s_save_trapsts, hwreg(HW_REG_TRAPSTS)
	s_and_b32	s_save_trapsts, s_save_trapsts, SQ_WAVE_TRAPSTS_EXCE_MASK
	s_cbranch_scc1	L_EXCP_CASE						// Exception, jump back to the shader program directly.
	s_add_u32	ttmp0, ttmp0, 4						// S_TRAP case, add 4 to ttmp0
	s_addc_u32	ttmp1, ttmp1, 0
L_EXCP_CASE:
	s_and_b32	ttmp1, ttmp1, 0xFFFF

#if ASIC_TARGET_NAVI1X
	// Restore SQ_WAVE_IB_STS.
	s_lshr_b32	ttmp2, ttmp11, (TTMP11_SAVE_RCNT_FIRST_REPLAY_SHIFT - SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT)
	s_and_b32	ttmp3, ttmp2, SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK
	s_lshr_b32	ttmp2, ttmp11, (TTMP11_SAVE_REPLAY_W64H_SHIFT - SQ_WAVE_IB_STS_REPLAY_W64H_SHIFT)
	s_and_b32	ttmp2, ttmp2, SQ_WAVE_IB_STS_REPLAY_W64H_MASK
	s_or_b32	ttmp2, ttmp2, ttmp3
	s_setreg_b32	hwreg(HW_REG_IB_STS), ttmp2
#endif

	// Restore SQ_WAVE_STATUS.
	s_and_b64	exec, exec, exec					// Restore STATUS.EXECZ, not writable by s_setreg_b32
	s_and_b64	vcc, vcc, vcc						// Restore STATUS.VCCZ, not writable by s_setreg_b32
	s_setreg_b32	hwreg(HW_REG_STATUS), s_save_status

	s_rfe_b64	[ttmp0, ttmp1]

L_SAVE:
	s_and_b32	s_save_pc_hi, s_save_pc_hi, 0x0000ffff			//pc[47:32]
	s_mov_b32	s_save_tmp, 0
	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_SAVECTX_SHIFT, 1), s_save_tmp	//clear saveCtx bit

#if ASIC_TARGET_NAVI1X
	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS, SQ_WAVE_IB_STS_RCNT_SHIFT, SQ_WAVE_IB_STS_RCNT_SIZE)
	s_lshl_b32	s_save_tmp, s_save_tmp, S_SAVE_PC_HI_RCNT_SHIFT
	s_or_b32	s_save_pc_hi, s_save_pc_hi, s_save_tmp
	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS, SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT, SQ_WAVE_IB_STS_FIRST_REPLAY_SIZE)
	s_lshl_b32	s_save_tmp, s_save_tmp, S_SAVE_PC_HI_FIRST_REPLAY_SHIFT
	s_or_b32	s_save_pc_hi, s_save_pc_hi, s_save_tmp
	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS, SQ_WAVE_IB_STS_REPLAY_W64H_SHIFT, SQ_WAVE_IB_STS_REPLAY_W64H_SIZE)
	s_lshl_b32	s_save_tmp, s_save_tmp, S_SAVE_PC_HI_REPLAY_W64H_SHIFT
	s_or_b32	s_save_pc_hi, s_save_pc_hi, s_save_tmp
	s_getreg_b32	s_save_tmp, hwreg(HW_REG_IB_STS)			//clear RCNT and FIRST_REPLAY and REPLAY_W64H in IB_STS
	s_and_b32	s_save_tmp, s_save_tmp, SQ_WAVE_IB_STS_RCNT_FIRST_REPLAY_MASK_NEG

	s_setreg_b32	hwreg(HW_REG_IB_STS), s_save_tmp
#endif

	/* inform SPI the readiness and wait for SPI's go signal */
	s_mov_b32	s_save_exec_lo, exec_lo					//save EXEC and use EXEC for the go signal from SPI
	s_mov_b32	s_save_exec_hi, exec_hi
	s_mov_b64	exec, 0x0						//clear EXEC to get ready to receive

	s_sendmsg	sendmsg(MSG_SAVEWAVE)					//send SPI a message and wait for SPI's write to EXEC

#if ASIC_TARGET_NAVI1X
L_SLEEP:
	// sleep 1 (64clk) is not enough for 8 waves per SIMD, which will cause
	// SQ hang, since the 7,8th wave could not get arbit to exec inst, while
	// other waves are stuck into the sleep-loop and waiting for wrexec!=0
	s_sleep		0x2
	s_cbranch_execz	L_SLEEP
#else
	s_waitcnt	lgkmcnt(0)
#endif

	// Save trap temporaries 4-11, 13 initialized by SPI debug dispatch logic
	// ttmp SR memory offset : size(VGPR)+size(SGPR)+0x40
	get_wave_size(s_save_ttmps_hi)
	get_vgpr_size_bytes(s_save_ttmps_lo, s_save_ttmps_hi)
	s_and_b32	s_save_ttmps_hi, s_save_spi_init_hi, 0xFFFF
	s_add_u32	s_save_ttmps_lo, s_save_ttmps_lo, get_sgpr_size_bytes()
	s_add_u32	s_save_ttmps_lo, s_save_ttmps_lo, s_save_spi_init_lo
	s_addc_u32	s_save_ttmps_hi, s_save_ttmps_hi, 0x0

#if ASIC_TARGET_NAVI1X
	s_store_dwordx4	[ttmp4, ttmp5, ttmp6, ttmp7], [s_save_ttmps_lo, s_save_ttmps_hi], 0x50 glc:1
	s_store_dwordx4	[ttmp8, ttmp9, ttmp10, ttmp11], [s_save_ttmps_lo, s_save_ttmps_hi], 0x60 glc:1
	s_store_dword   ttmp13, [s_save_ttmps_lo, s_save_ttmps_hi], 0x74 glc:1
#endif

	/* setup Resource Contants */
	s_mov_b32	s_save_buf_rsrc0, s_save_spi_init_lo			//base_addr_lo
	s_and_b32	s_save_buf_rsrc1, s_save_spi_init_hi, 0x0000FFFF	//base_addr_hi
	s_or_b32	s_save_buf_rsrc1, s_save_buf_rsrc1, S_SAVE_BUF_RSRC_WORD1_STRIDE
	s_mov_b32	s_save_buf_rsrc2, 0					//NUM_RECORDS initial value = 0 (in bytes) although not neccessarily inited
	s_mov_b32	s_save_buf_rsrc3, S_SAVE_BUF_RSRC_WORD3_MISC
	s_and_b32	s_save_tmp, s_save_spi_init_hi, S_SAVE_SPI_INIT_ATC_MASK
	s_lshr_b32	s_save_tmp, s_save_tmp, (S_SAVE_SPI_INIT_ATC_SHIFT-SQ_BUF_RSRC_WORD1_ATC_SHIFT)
	s_or_b32	s_save_buf_rsrc3, s_save_buf_rsrc3, s_save_tmp		//or ATC
	s_and_b32	s_save_tmp, s_save_spi_init_hi, S_SAVE_SPI_INIT_MTYPE_MASK
	s_lshr_b32	s_save_tmp, s_save_tmp, (S_SAVE_SPI_INIT_MTYPE_SHIFT-SQ_BUF_RSRC_WORD3_MTYPE_SHIFT)
	s_or_b32	s_save_buf_rsrc3, s_save_buf_rsrc3, s_save_tmp		//or MTYPE

	s_mov_b32	s_save_m0, m0

	/* global mem offset */
	s_mov_b32	s_save_mem_offset, 0x0
	get_wave_size(s_wave_size)

#if ASIC_TARGET_NAVI1X
	// Save and clear vector XNACK state late to free up SGPRs.
	s_getreg_b32	s_save_xnack_mask, hwreg(HW_REG_SHADER_XNACK_MASK)
	s_setreg_imm32_b32	hwreg(HW_REG_SHADER_XNACK_MASK), 0x0
#endif

	/* save first 4 VGPRs, needed for SGPR save */
	s_mov_b32	exec_lo, 0xFFFFFFFF					//need every thread from now on
	s_lshr_b32	m0, s_wave_size, S_WAVE_SIZE
	s_and_b32	m0, m0, 1
	s_cmp_eq_u32	m0, 1
	s_cbranch_scc1	L_ENABLE_SAVE_4VGPR_EXEC_HI
	s_mov_b32	exec_hi, 0x00000000
	s_branch	L_SAVE_4VGPR_WAVE32
L_ENABLE_SAVE_4VGPR_EXEC_HI:
	s_mov_b32	exec_hi, 0xFFFFFFFF
	s_branch	L_SAVE_4VGPR_WAVE64
L_SAVE_4VGPR_WAVE32:
	s_mov_b32	s_save_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes

	// VGPR Allocated in 4-GPR granularity

	buffer_store_dword	v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
	buffer_store_dword	v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:128
	buffer_store_dword	v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:128*2
	buffer_store_dword	v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:128*3
	s_branch	L_SAVE_HWREG

L_SAVE_4VGPR_WAVE64:
	s_mov_b32	s_save_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes

	// VGPR Allocated in 4-GPR granularity

	buffer_store_dword	v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
	buffer_store_dword	v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256
	buffer_store_dword	v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*2
	buffer_store_dword	v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*3

	/* save HW registers */

L_SAVE_HWREG:
	// HWREG SR memory offset : size(VGPR)+size(SVGPR)+size(SGPR)
	get_vgpr_size_bytes(s_save_mem_offset, s_wave_size)
	get_svgpr_size_bytes(s_save_tmp)
	s_add_u32	s_save_mem_offset, s_save_mem_offset, s_save_tmp
	s_add_u32	s_save_mem_offset, s_save_mem_offset, get_sgpr_size_bytes()

	s_mov_b32	s_save_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes

#if NO_SQC_STORE
	v_mov_b32	v0, 0x0							//Offset[31:0] from buffer resource
	v_mov_b32	v1, 0x0							//Offset[63:32] from buffer resource
	v_mov_b32	v2, 0x0							//Set of SGPRs for TCP store
	s_mov_b32	m0, 0x0							//Next lane of v2 to write to
#endif

	write_hwreg_to_mem(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset)
	write_hwreg_to_mem(s_save_pc_lo, s_save_buf_rsrc0, s_save_mem_offset)
	write_hwreg_to_mem(s_save_pc_hi, s_save_buf_rsrc0, s_save_mem_offset)
	write_hwreg_to_mem(s_save_exec_lo, s_save_buf_rsrc0, s_save_mem_offset)
	write_hwreg_to_mem(s_save_exec_hi, s_save_buf_rsrc0, s_save_mem_offset)
	write_hwreg_to_mem(s_save_status, s_save_buf_rsrc0, s_save_mem_offset)

	s_getreg_b32	s_save_tmp, hwreg(HW_REG_TRAPSTS)
	write_hwreg_to_mem(s_save_tmp, s_save_buf_rsrc0, s_save_mem_offset)

	// Not used on Sienna_Cichlid but keep layout same for debugger.
	write_hwreg_to_mem(s_save_xnack_mask, s_save_buf_rsrc0, s_save_mem_offset)

	s_getreg_b32	s_save_m0, hwreg(HW_REG_MODE)
	write_hwreg_to_mem(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset)

	s_getreg_b32	s_save_m0, hwreg(HW_REG_SHADER_FLAT_SCRATCH_LO)
	write_hwreg_to_mem(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset)

	s_getreg_b32	s_save_m0, hwreg(HW_REG_SHADER_FLAT_SCRATCH_HI)
	write_hwreg_to_mem(s_save_m0, s_save_buf_rsrc0, s_save_mem_offset)

#if NO_SQC_STORE
	// Write HWREG/SGPRs with 32 VGPR lanes, wave32 is common case.
	s_mov_b32	exec_hi, 0x0
	buffer_store_dword	v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
#endif

	/* save SGPRs */
	// Save SGPR before LDS save, then the s0 to s4 can be used during LDS save...

	// SGPR SR memory offset : size(VGPR)+size(SVGPR)
	get_vgpr_size_bytes(s_save_mem_offset, s_wave_size)
	get_svgpr_size_bytes(s_save_tmp)
	s_add_u32	s_save_mem_offset, s_save_mem_offset, s_save_tmp
	s_mov_b32	s_save_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes

#if NO_SQC_STORE
	s_mov_b32	ttmp13, 0x0						//next VGPR lane to copy SGPR into
#else
	// backup s_save_buf_rsrc0,1 to s_save_pc_lo/hi, since write_16sgpr_to_mem function will change the rsrc0
	s_mov_b32	s_save_xnack_mask, s_save_buf_rsrc0
	s_add_u32	s_save_buf_rsrc0, s_save_buf_rsrc0, s_save_mem_offset
	s_addc_u32	s_save_buf_rsrc1, s_save_buf_rsrc1, 0
#endif

	s_mov_b32	m0, 0x0							//SGPR initial index value =0
	s_nop		0x0							//Manually inserted wait states
L_SAVE_SGPR_LOOP:
	// SGPR is allocated in 16 SGPR granularity
	s_movrels_b64	s0, s0							//s0 = s[0+m0], s1 = s[1+m0]
	s_movrels_b64	s2, s2							//s2 = s[2+m0], s3 = s[3+m0]
	s_movrels_b64	s4, s4							//s4 = s[4+m0], s5 = s[5+m0]
	s_movrels_b64	s6, s6							//s6 = s[6+m0], s7 = s[7+m0]
	s_movrels_b64	s8, s8							//s8 = s[8+m0], s9 = s[9+m0]
	s_movrels_b64	s10, s10						//s10 = s[10+m0], s11 = s[11+m0]
	s_movrels_b64	s12, s12						//s12 = s[12+m0], s13 = s[13+m0]
	s_movrels_b64	s14, s14						//s14 = s[14+m0], s15 = s[15+m0]

	write_16sgpr_to_mem(s0, s_save_buf_rsrc0, s_save_mem_offset)

#if NO_SQC_STORE
	s_cmp_eq_u32	ttmp13, 0x20						//have 32 VGPR lanes filled?
	s_cbranch_scc0	L_SAVE_SGPR_SKIP_TCP_STORE

	buffer_store_dword	v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
	s_add_u32	s_save_mem_offset, s_save_mem_offset, 0x80
	s_mov_b32	ttmp13, 0x0
	v_mov_b32	v2, 0x0
L_SAVE_SGPR_SKIP_TCP_STORE:
#endif

	s_add_u32	m0, m0, 16						//next sgpr index
	s_cmp_lt_u32	m0, 96							//scc = (m0 < first 96 SGPR) ? 1 : 0
	s_cbranch_scc1	L_SAVE_SGPR_LOOP					//first 96 SGPR save is complete?

	//save the rest 12 SGPR
	s_movrels_b64	s0, s0							//s0 = s[0+m0], s1 = s[1+m0]
	s_movrels_b64	s2, s2							//s2 = s[2+m0], s3 = s[3+m0]
	s_movrels_b64	s4, s4							//s4 = s[4+m0], s5 = s[5+m0]
	s_movrels_b64	s6, s6							//s6 = s[6+m0], s7 = s[7+m0]
	s_movrels_b64	s8, s8							//s8 = s[8+m0], s9 = s[9+m0]
	s_movrels_b64	s10, s10						//s10 = s[10+m0], s11 = s[11+m0]
	write_12sgpr_to_mem(s0, s_save_buf_rsrc0, s_save_mem_offset)

#if NO_SQC_STORE
	buffer_store_dword	v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
#else
	// restore s_save_buf_rsrc0,1
	s_mov_b32	s_save_buf_rsrc0, s_save_xnack_mask
#endif

	/* save LDS */

L_SAVE_LDS:
	// Change EXEC to all threads...
	s_mov_b32	exec_lo, 0xFFFFFFFF					//need every thread from now on
	s_lshr_b32	m0, s_wave_size, S_WAVE_SIZE
	s_and_b32	m0, m0, 1
	s_cmp_eq_u32	m0, 1
	s_cbranch_scc1	L_ENABLE_SAVE_LDS_EXEC_HI
	s_mov_b32	exec_hi, 0x00000000
	s_branch	L_SAVE_LDS_NORMAL
L_ENABLE_SAVE_LDS_EXEC_HI:
	s_mov_b32	exec_hi, 0xFFFFFFFF
L_SAVE_LDS_NORMAL:
	s_getreg_b32	s_save_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE)
	s_and_b32	s_save_alloc_size, s_save_alloc_size, 0xFFFFFFFF	//lds_size is zero?
	s_cbranch_scc0	L_SAVE_LDS_DONE						//no lds used? jump to L_SAVE_DONE

	s_barrier								//LDS is used? wait for other waves in the same TG
	s_and_b32	s_save_tmp, s_wave_size, S_SAVE_SPI_INIT_FIRST_WAVE_MASK
	s_cbranch_scc0	L_SAVE_LDS_DONE

	// first wave do LDS save;

	s_lshl_b32	s_save_alloc_size, s_save_alloc_size, 6			//LDS size in dwords = lds_size * 64dw
	s_lshl_b32	s_save_alloc_size, s_save_alloc_size, 2			//LDS size in bytes
	s_mov_b32	s_save_buf_rsrc2, s_save_alloc_size			//NUM_RECORDS in bytes

	// LDS at offset: size(VGPR)+size(SVGPR)+SIZE(SGPR)+SIZE(HWREG)
	//
	get_vgpr_size_bytes(s_save_mem_offset, s_wave_size)
	get_svgpr_size_bytes(s_save_tmp)
	s_add_u32	s_save_mem_offset, s_save_mem_offset, s_save_tmp
	s_add_u32	s_save_mem_offset, s_save_mem_offset, get_sgpr_size_bytes()
	s_add_u32	s_save_mem_offset, s_save_mem_offset, get_hwreg_size_bytes()

	s_mov_b32	s_save_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes

	//load 0~63*4(byte address) to vgpr v0
	v_mbcnt_lo_u32_b32	v0, -1, 0
	v_mbcnt_hi_u32_b32	v0, -1, v0
	v_mul_u32_u24	v0, 4, v0

	s_lshr_b32	m0, s_wave_size, S_WAVE_SIZE
	s_and_b32	m0, m0, 1
	s_cmp_eq_u32	m0, 1
	s_mov_b32	m0, 0x0
	s_cbranch_scc1	L_SAVE_LDS_W64

L_SAVE_LDS_W32:
	s_mov_b32	s3, 128
	s_nop		0
	s_nop		0
	s_nop		0
L_SAVE_LDS_LOOP_W32:
	ds_read_b32	v1, v0
	s_waitcnt	0
	buffer_store_dword	v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1

	s_add_u32	m0, m0, s3						//every buffer_store_lds does 256 bytes
	s_add_u32	s_save_mem_offset, s_save_mem_offset, s3
	v_add_nc_u32	v0, v0, 128						//mem offset increased by 128 bytes
	s_cmp_lt_u32	m0, s_save_alloc_size					//scc=(m0 < s_save_alloc_size) ? 1 : 0
	s_cbranch_scc1	L_SAVE_LDS_LOOP_W32					//LDS save is complete?

	s_branch	L_SAVE_LDS_DONE

L_SAVE_LDS_W64:
	s_mov_b32	s3, 256
	s_nop		0
	s_nop		0
	s_nop		0
L_SAVE_LDS_LOOP_W64:
	ds_read_b32	v1, v0
	s_waitcnt	0
	buffer_store_dword	v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1

	s_add_u32	m0, m0, s3						//every buffer_store_lds does 256 bytes
	s_add_u32	s_save_mem_offset, s_save_mem_offset, s3
	v_add_nc_u32	v0, v0, 256						//mem offset increased by 256 bytes
	s_cmp_lt_u32	m0, s_save_alloc_size					//scc=(m0 < s_save_alloc_size) ? 1 : 0
	s_cbranch_scc1	L_SAVE_LDS_LOOP_W64					//LDS save is complete?

L_SAVE_LDS_DONE:
	/* save VGPRs  - set the Rest VGPRs */
L_SAVE_VGPR:
	// VGPR SR memory offset: 0
	s_mov_b32	exec_lo, 0xFFFFFFFF					//need every thread from now on
	s_lshr_b32	m0, s_wave_size, S_WAVE_SIZE
	s_and_b32	m0, m0, 1
	s_cmp_eq_u32	m0, 1
	s_cbranch_scc1	L_ENABLE_SAVE_VGPR_EXEC_HI
	s_mov_b32	s_save_mem_offset, (0+128*4)				// for the rest VGPRs
	s_mov_b32	exec_hi, 0x00000000
	s_branch	L_SAVE_VGPR_NORMAL
L_ENABLE_SAVE_VGPR_EXEC_HI:
	s_mov_b32	s_save_mem_offset, (0+256*4)				// for the rest VGPRs
	s_mov_b32	exec_hi, 0xFFFFFFFF
L_SAVE_VGPR_NORMAL:
	s_getreg_b32	s_save_alloc_size, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE)
	s_add_u32	s_save_alloc_size, s_save_alloc_size, 1
	s_lshl_b32	s_save_alloc_size, s_save_alloc_size, 2			//Number of VGPRs = (vgpr_size + 1) * 4    (non-zero value)
	//determine it is wave32 or wave64
	s_lshr_b32	m0, s_wave_size, S_WAVE_SIZE
	s_and_b32	m0, m0, 1
	s_cmp_eq_u32	m0, 1
	s_cbranch_scc1	L_SAVE_VGPR_WAVE64

	s_mov_b32	s_save_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes

	// VGPR Allocated in 4-GPR granularity

	// VGPR store using dw burst
	s_mov_b32	m0, 0x4							//VGPR initial index value =4
	s_cmp_lt_u32	m0, s_save_alloc_size
	s_cbranch_scc0	L_SAVE_VGPR_END

L_SAVE_VGPR_W32_LOOP:
	v_movrels_b32	v0, v0							//v0 = v[0+m0]
	v_movrels_b32	v1, v1							//v1 = v[1+m0]
	v_movrels_b32	v2, v2							//v2 = v[2+m0]
	v_movrels_b32	v3, v3							//v3 = v[3+m0]

	buffer_store_dword	v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
	buffer_store_dword	v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:128
	buffer_store_dword	v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:128*2
	buffer_store_dword	v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:128*3

	s_add_u32	m0, m0, 4						//next vgpr index
	s_add_u32	s_save_mem_offset, s_save_mem_offset, 128*4		//every buffer_store_dword does 128 bytes
	s_cmp_lt_u32	m0, s_save_alloc_size					//scc = (m0 < s_save_alloc_size) ? 1 : 0
	s_cbranch_scc1	L_SAVE_VGPR_W32_LOOP					//VGPR save is complete?

	s_branch	L_SAVE_VGPR_END

L_SAVE_VGPR_WAVE64:
	s_mov_b32	s_save_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes

	// VGPR store using dw burst
	s_mov_b32	m0, 0x4							//VGPR initial index value =4
	s_cmp_lt_u32	m0, s_save_alloc_size
	s_cbranch_scc0	L_SAVE_VGPR_END

L_SAVE_VGPR_W64_LOOP:
	v_movrels_b32	v0, v0							//v0 = v[0+m0]
	v_movrels_b32	v1, v1							//v1 = v[1+m0]
	v_movrels_b32	v2, v2							//v2 = v[2+m0]
	v_movrels_b32	v3, v3							//v3 = v[3+m0]

	buffer_store_dword	v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
	buffer_store_dword	v1, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256
	buffer_store_dword	v2, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*2
	buffer_store_dword	v3, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1 offset:256*3

	s_add_u32	m0, m0, 4						//next vgpr index
	s_add_u32	s_save_mem_offset, s_save_mem_offset, 256*4		//every buffer_store_dword does 256 bytes
	s_cmp_lt_u32	m0, s_save_alloc_size					//scc = (m0 < s_save_alloc_size) ? 1 : 0
	s_cbranch_scc1	L_SAVE_VGPR_W64_LOOP					//VGPR save is complete?

	//Below part will be the save shared vgpr part (new for gfx10)
	s_getreg_b32	s_save_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SIZE)
	s_and_b32	s_save_alloc_size, s_save_alloc_size, 0xFFFFFFFF	//shared_vgpr_size is zero?
	s_cbranch_scc0	L_SAVE_VGPR_END						//no shared_vgpr used? jump to L_SAVE_LDS
	s_lshl_b32	s_save_alloc_size, s_save_alloc_size, 3			//Number of SHARED_VGPRs = shared_vgpr_size * 8    (non-zero value)
	//m0 now has the value of normal vgpr count, just add the m0 with shared_vgpr count to get the total count.
	//save shared_vgpr will start from the index of m0
	s_add_u32	s_save_alloc_size, s_save_alloc_size, m0
	s_mov_b32	exec_lo, 0xFFFFFFFF
	s_mov_b32	exec_hi, 0x00000000
L_SAVE_SHARED_VGPR_WAVE64_LOOP:
	v_movrels_b32	v0, v0							//v0 = v[0+m0]
	buffer_store_dword	v0, v0, s_save_buf_rsrc0, s_save_mem_offset slc:1 glc:1
	s_add_u32	m0, m0, 1						//next vgpr index
	s_add_u32	s_save_mem_offset, s_save_mem_offset, 128
	s_cmp_lt_u32	m0, s_save_alloc_size					//scc = (m0 < s_save_alloc_size) ? 1 : 0
	s_cbranch_scc1	L_SAVE_SHARED_VGPR_WAVE64_LOOP				//SHARED_VGPR save is complete?

L_SAVE_VGPR_END:
	s_branch	L_END_PGM

L_RESTORE:
	/* Setup Resource Contants */
	s_mov_b32	s_restore_buf_rsrc0, s_restore_spi_init_lo		//base_addr_lo
	s_and_b32	s_restore_buf_rsrc1, s_restore_spi_init_hi, 0x0000FFFF	//base_addr_hi
	s_or_b32	s_restore_buf_rsrc1, s_restore_buf_rsrc1, S_RESTORE_BUF_RSRC_WORD1_STRIDE
	s_mov_b32	s_restore_buf_rsrc2, 0					//NUM_RECORDS initial value = 0 (in bytes)
	s_mov_b32	s_restore_buf_rsrc3, S_RESTORE_BUF_RSRC_WORD3_MISC
	s_and_b32	s_restore_tmp, s_restore_spi_init_hi, S_RESTORE_SPI_INIT_ATC_MASK
	s_lshr_b32	s_restore_tmp, s_restore_tmp, (S_RESTORE_SPI_INIT_ATC_SHIFT-SQ_BUF_RSRC_WORD1_ATC_SHIFT)
	s_or_b32	s_restore_buf_rsrc3, s_restore_buf_rsrc3, s_restore_tmp	//or ATC
	s_and_b32	s_restore_tmp, s_restore_spi_init_hi, S_RESTORE_SPI_INIT_MTYPE_MASK
	s_lshr_b32	s_restore_tmp, s_restore_tmp, (S_RESTORE_SPI_INIT_MTYPE_SHIFT-SQ_BUF_RSRC_WORD3_MTYPE_SHIFT)
	s_or_b32	s_restore_buf_rsrc3, s_restore_buf_rsrc3, s_restore_tmp	//or MTYPE
	//determine it is wave32 or wave64
	get_wave_size(s_restore_size)

	s_and_b32	s_restore_tmp, s_restore_spi_init_hi, S_RESTORE_SPI_INIT_FIRST_WAVE_MASK
	s_cbranch_scc0	L_RESTORE_VGPR

	/* restore LDS */
L_RESTORE_LDS:
	s_mov_b32	exec_lo, 0xFFFFFFFF					//need every thread from now on
	s_lshr_b32	m0, s_restore_size, S_WAVE_SIZE
	s_and_b32	m0, m0, 1
	s_cmp_eq_u32	m0, 1
	s_cbranch_scc1	L_ENABLE_RESTORE_LDS_EXEC_HI
	s_mov_b32	exec_hi, 0x00000000
	s_branch	L_RESTORE_LDS_NORMAL
L_ENABLE_RESTORE_LDS_EXEC_HI:
	s_mov_b32	exec_hi, 0xFFFFFFFF
L_RESTORE_LDS_NORMAL:
	s_getreg_b32	s_restore_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE)
	s_and_b32	s_restore_alloc_size, s_restore_alloc_size, 0xFFFFFFFF	//lds_size is zero?
	s_cbranch_scc0	L_RESTORE_VGPR						//no lds used? jump to L_RESTORE_VGPR
	s_lshl_b32	s_restore_alloc_size, s_restore_alloc_size, 6		//LDS size in dwords = lds_size * 64dw
	s_lshl_b32	s_restore_alloc_size, s_restore_alloc_size, 2		//LDS size in bytes
	s_mov_b32	s_restore_buf_rsrc2, s_restore_alloc_size		//NUM_RECORDS in bytes

	// LDS at offset: size(VGPR)+size(SVGPR)+SIZE(SGPR)+SIZE(HWREG)
	//
	get_vgpr_size_bytes(s_restore_mem_offset, s_restore_size)
	get_svgpr_size_bytes(s_restore_tmp)
	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, s_restore_tmp
	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, get_sgpr_size_bytes()
	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, get_hwreg_size_bytes()

	s_mov_b32	s_restore_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes

	s_lshr_b32	m0, s_restore_size, S_WAVE_SIZE
	s_and_b32	m0, m0, 1
	s_cmp_eq_u32	m0, 1
	s_mov_b32	m0, 0x0
	s_cbranch_scc1	L_RESTORE_LDS_LOOP_W64

L_RESTORE_LDS_LOOP_W32:
	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset lds:1	// first 64DW
	s_add_u32	m0, m0, 128						// 128 DW
	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, 128		//mem offset increased by 128DW
	s_cmp_lt_u32	m0, s_restore_alloc_size				//scc=(m0 < s_restore_alloc_size) ? 1 : 0
	s_cbranch_scc1	L_RESTORE_LDS_LOOP_W32					//LDS restore is complete?
	s_branch	L_RESTORE_VGPR

L_RESTORE_LDS_LOOP_W64:
	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset lds:1	// first 64DW
	s_add_u32	m0, m0, 256						// 256 DW
	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, 256		//mem offset increased by 256DW
	s_cmp_lt_u32	m0, s_restore_alloc_size				//scc=(m0 < s_restore_alloc_size) ? 1 : 0
	s_cbranch_scc1	L_RESTORE_LDS_LOOP_W64					//LDS restore is complete?

	/* restore VGPRs */
L_RESTORE_VGPR:
	// VGPR SR memory offset : 0
	s_mov_b32	s_restore_mem_offset, 0x0
 	s_mov_b32	exec_lo, 0xFFFFFFFF					//need every thread from now on
	s_lshr_b32	m0, s_restore_size, S_WAVE_SIZE
	s_and_b32	m0, m0, 1
	s_cmp_eq_u32	m0, 1
	s_cbranch_scc1	L_ENABLE_RESTORE_VGPR_EXEC_HI
	s_mov_b32	exec_hi, 0x00000000
	s_branch	L_RESTORE_VGPR_NORMAL
L_ENABLE_RESTORE_VGPR_EXEC_HI:
	s_mov_b32	exec_hi, 0xFFFFFFFF
L_RESTORE_VGPR_NORMAL:
	s_getreg_b32	s_restore_alloc_size, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE)
	s_add_u32	s_restore_alloc_size, s_restore_alloc_size, 1
	s_lshl_b32	s_restore_alloc_size, s_restore_alloc_size, 2		//Number of VGPRs = (vgpr_size + 1) * 4    (non-zero value)
	//determine it is wave32 or wave64
	s_lshr_b32	m0, s_restore_size, S_WAVE_SIZE
	s_and_b32	m0, m0, 1
	s_cmp_eq_u32	m0, 1
	s_cbranch_scc1	L_RESTORE_VGPR_WAVE64

	s_mov_b32	s_restore_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes

	// VGPR load using dw burst
	s_mov_b32	s_restore_mem_offset_save, s_restore_mem_offset		// restore start with v1, v0 will be the last
	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, 128*4
	s_mov_b32	m0, 4							//VGPR initial index value = 4

L_RESTORE_VGPR_WAVE32_LOOP:
	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1
	buffer_load_dword	v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:128
	buffer_load_dword	v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:128*2
	buffer_load_dword	v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:128*3
	s_waitcnt	vmcnt(0)
	v_movreld_b32	v0, v0							//v[0+m0] = v0
	v_movreld_b32	v1, v1
	v_movreld_b32	v2, v2
	v_movreld_b32	v3, v3
	s_add_u32	m0, m0, 4						//next vgpr index
	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, 128*4	//every buffer_load_dword does 128 bytes
	s_cmp_lt_u32	m0, s_restore_alloc_size				//scc = (m0 < s_restore_alloc_size) ? 1 : 0
	s_cbranch_scc1	L_RESTORE_VGPR_WAVE32_LOOP				//VGPR restore (except v0) is complete?

	/* VGPR restore on v0 */
	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1
	buffer_load_dword	v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:128
	buffer_load_dword	v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:128*2
	buffer_load_dword	v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:128*3

	s_branch	L_RESTORE_SGPR

L_RESTORE_VGPR_WAVE64:
	s_mov_b32	s_restore_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes

	// VGPR load using dw burst
	s_mov_b32	s_restore_mem_offset_save, s_restore_mem_offset		// restore start with v4, v0 will be the last
	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, 256*4
	s_mov_b32	m0, 4							//VGPR initial index value = 4

L_RESTORE_VGPR_WAVE64_LOOP:
	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1
	buffer_load_dword	v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256
	buffer_load_dword	v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256*2
	buffer_load_dword	v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1 offset:256*3
	s_waitcnt	vmcnt(0)
	v_movreld_b32	v0, v0							//v[0+m0] = v0
	v_movreld_b32	v1, v1
	v_movreld_b32	v2, v2
	v_movreld_b32	v3, v3
	s_add_u32	m0, m0, 4						//next vgpr index
	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, 256*4	//every buffer_load_dword does 256 bytes
	s_cmp_lt_u32	m0, s_restore_alloc_size				//scc = (m0 < s_restore_alloc_size) ? 1 : 0
	s_cbranch_scc1	L_RESTORE_VGPR_WAVE64_LOOP				//VGPR restore (except v0) is complete?

	//Below part will be the restore shared vgpr part (new for gfx10)
	s_getreg_b32	s_restore_alloc_size, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SIZE)	//shared_vgpr_size
	s_and_b32	s_restore_alloc_size, s_restore_alloc_size, 0xFFFFFFFF	//shared_vgpr_size is zero?
	s_cbranch_scc0	L_RESTORE_V0						//no shared_vgpr used?
	s_lshl_b32	s_restore_alloc_size, s_restore_alloc_size, 3		//Number of SHARED_VGPRs = shared_vgpr_size * 8    (non-zero value)
	//m0 now has the value of normal vgpr count, just add the m0 with shared_vgpr count to get the total count.
	//restore shared_vgpr will start from the index of m0
	s_add_u32	s_restore_alloc_size, s_restore_alloc_size, m0
	s_mov_b32	exec_lo, 0xFFFFFFFF
	s_mov_b32	exec_hi, 0x00000000
L_RESTORE_SHARED_VGPR_WAVE64_LOOP:
	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset slc:1 glc:1
	s_waitcnt	vmcnt(0)
	v_movreld_b32	v0, v0							//v[0+m0] = v0
	s_add_u32	m0, m0, 1						//next vgpr index
	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, 128
	s_cmp_lt_u32	m0, s_restore_alloc_size				//scc = (m0 < s_restore_alloc_size) ? 1 : 0
	s_cbranch_scc1	L_RESTORE_SHARED_VGPR_WAVE64_LOOP			//VGPR restore (except v0) is complete?

	s_mov_b32	exec_hi, 0xFFFFFFFF					//restore back exec_hi before restoring V0!!

	/* VGPR restore on v0 */
L_RESTORE_V0:
	buffer_load_dword	v0, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1
	buffer_load_dword	v1, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:256
	buffer_load_dword	v2, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:256*2
	buffer_load_dword	v3, v0, s_restore_buf_rsrc0, s_restore_mem_offset_save slc:1 glc:1 offset:256*3
	s_waitcnt	vmcnt(0)

	/* restore SGPRs */
	//will be 2+8+16*6
	// SGPR SR memory offset : size(VGPR)+size(SVGPR)
L_RESTORE_SGPR:
	get_vgpr_size_bytes(s_restore_mem_offset, s_restore_size)
	get_svgpr_size_bytes(s_restore_tmp)
	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, s_restore_tmp
	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, get_sgpr_size_bytes()
	s_sub_u32	s_restore_mem_offset, s_restore_mem_offset, 20*4	//s108~s127 is not saved

	s_mov_b32	s_restore_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes

	s_mov_b32	m0, s_sgpr_save_num

	read_4sgpr_from_mem(s0, s_restore_buf_rsrc0, s_restore_mem_offset)
	s_waitcnt	lgkmcnt(0)

	s_sub_u32	m0, m0, 4						// Restore from S[0] to S[104]
	s_nop		0							// hazard SALU M0=> S_MOVREL

	s_movreld_b64	s0, s0							//s[0+m0] = s0
	s_movreld_b64	s2, s2

	read_8sgpr_from_mem(s0, s_restore_buf_rsrc0, s_restore_mem_offset)
	s_waitcnt	lgkmcnt(0)

	s_sub_u32	m0, m0, 8						// Restore from S[0] to S[96]
	s_nop		0							// hazard SALU M0=> S_MOVREL

	s_movreld_b64	s0, s0							//s[0+m0] = s0
	s_movreld_b64	s2, s2
	s_movreld_b64	s4, s4
	s_movreld_b64	s6, s6

 L_RESTORE_SGPR_LOOP:
	read_16sgpr_from_mem(s0, s_restore_buf_rsrc0, s_restore_mem_offset)
	s_waitcnt	lgkmcnt(0)

	s_sub_u32	m0, m0, 16						// Restore from S[n] to S[0]
	s_nop		0							// hazard SALU M0=> S_MOVREL

	s_movreld_b64	s0, s0							//s[0+m0] = s0
	s_movreld_b64	s2, s2
	s_movreld_b64	s4, s4
	s_movreld_b64	s6, s6
	s_movreld_b64	s8, s8
	s_movreld_b64	s10, s10
	s_movreld_b64	s12, s12
	s_movreld_b64	s14, s14

	s_cmp_eq_u32	m0, 0							//scc = (m0 < s_sgpr_save_num) ? 1 : 0
	s_cbranch_scc0	L_RESTORE_SGPR_LOOP

	// s_barrier with MODE.DEBUG_EN=1, STATUS.PRIV=1 incorrectly asserts debug exception.
	// Clear DEBUG_EN before and restore MODE after the barrier.
	s_setreg_imm32_b32	hwreg(HW_REG_MODE), 0
	s_barrier								//barrier to ensure the readiness of LDS before access attemps from any other wave in the same TG

	/* restore HW registers */
L_RESTORE_HWREG:
	// HWREG SR memory offset : size(VGPR)+size(SVGPR)+size(SGPR)
	get_vgpr_size_bytes(s_restore_mem_offset, s_restore_size)
	get_svgpr_size_bytes(s_restore_tmp)
	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, s_restore_tmp
	s_add_u32	s_restore_mem_offset, s_restore_mem_offset, get_sgpr_size_bytes()

	s_mov_b32	s_restore_buf_rsrc2, 0x1000000				//NUM_RECORDS in bytes

	read_hwreg_from_mem(s_restore_m0, s_restore_buf_rsrc0, s_restore_mem_offset)
	read_hwreg_from_mem(s_restore_pc_lo, s_restore_buf_rsrc0, s_restore_mem_offset)
	read_hwreg_from_mem(s_restore_pc_hi, s_restore_buf_rsrc0, s_restore_mem_offset)
	read_hwreg_from_mem(s_restore_exec_lo, s_restore_buf_rsrc0, s_restore_mem_offset)
	read_hwreg_from_mem(s_restore_exec_hi, s_restore_buf_rsrc0, s_restore_mem_offset)
	read_hwreg_from_mem(s_restore_status, s_restore_buf_rsrc0, s_restore_mem_offset)
	read_hwreg_from_mem(s_restore_trapsts, s_restore_buf_rsrc0, s_restore_mem_offset)
	read_hwreg_from_mem(s_restore_xnack_mask, s_restore_buf_rsrc0, s_restore_mem_offset)
	read_hwreg_from_mem(s_restore_mode, s_restore_buf_rsrc0, s_restore_mem_offset)
	read_hwreg_from_mem(s_restore_flat_scratch, s_restore_buf_rsrc0, s_restore_mem_offset)
	s_waitcnt	lgkmcnt(0)

	s_setreg_b32	hwreg(HW_REG_SHADER_FLAT_SCRATCH_LO), s_restore_flat_scratch

	read_hwreg_from_mem(s_restore_flat_scratch, s_restore_buf_rsrc0, s_restore_mem_offset)
	s_waitcnt	lgkmcnt(0)						//from now on, it is safe to restore STATUS and IB_STS

	s_setreg_b32	hwreg(HW_REG_SHADER_FLAT_SCRATCH_HI), s_restore_flat_scratch

	s_mov_b32	m0, s_restore_m0
	s_mov_b32	exec_lo, s_restore_exec_lo
	s_mov_b32	exec_hi, s_restore_exec_hi

	s_and_b32	s_restore_m0, SQ_WAVE_TRAPSTS_PRE_SAVECTX_MASK, s_restore_trapsts
	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_PRE_SAVECTX_SHIFT, SQ_WAVE_TRAPSTS_PRE_SAVECTX_SIZE), s_restore_m0

#if ASIC_TARGET_NAVI1X
	s_setreg_b32	hwreg(HW_REG_SHADER_XNACK_MASK), s_restore_xnack_mask
#endif

	s_and_b32	s_restore_m0, SQ_WAVE_TRAPSTS_POST_SAVECTX_MASK, s_restore_trapsts
	s_lshr_b32	s_restore_m0, s_restore_m0, SQ_WAVE_TRAPSTS_POST_SAVECTX_SHIFT
	s_setreg_b32	hwreg(HW_REG_TRAPSTS, SQ_WAVE_TRAPSTS_POST_SAVECTX_SHIFT, SQ_WAVE_TRAPSTS_POST_SAVECTX_SIZE), s_restore_m0
	s_setreg_b32	hwreg(HW_REG_MODE), s_restore_mode

	// Restore trap temporaries 4-11, 13 initialized by SPI debug dispatch logic
	// ttmp SR memory offset : size(VGPR)+size(SGPR)+0x40
	get_vgpr_size_bytes(s_restore_ttmps_lo, s_restore_size)
	s_add_u32	s_restore_ttmps_lo, s_restore_ttmps_lo, get_sgpr_size_bytes()
	s_add_u32	s_restore_ttmps_lo, s_restore_ttmps_lo, s_restore_buf_rsrc0
	s_addc_u32	s_restore_ttmps_hi, s_restore_buf_rsrc1, 0x0
	s_and_b32	s_restore_ttmps_hi, s_restore_ttmps_hi, 0xFFFF
	s_load_dwordx4	[ttmp4, ttmp5, ttmp6, ttmp7], [s_restore_ttmps_lo, s_restore_ttmps_hi], 0x50 glc:1
	s_load_dwordx4	[ttmp8, ttmp9, ttmp10, ttmp11], [s_restore_ttmps_lo, s_restore_ttmps_hi], 0x60 glc:1
	s_load_dword	ttmp13, [s_restore_ttmps_lo, s_restore_ttmps_hi], 0x74 glc:1
	s_waitcnt	lgkmcnt(0)

#if ASIC_TARGET_NAVI1X
	s_and_b32	s_restore_m0, s_restore_pc_hi, S_SAVE_PC_HI_RCNT_MASK
	s_lshr_b32	s_restore_m0, s_restore_m0, S_SAVE_PC_HI_RCNT_SHIFT
	s_lshl_b32	s_restore_m0, s_restore_m0, SQ_WAVE_IB_STS_RCNT_SHIFT
	s_mov_b32	s_restore_tmp, 0x0
	s_or_b32	s_restore_tmp, s_restore_tmp, s_restore_m0
	s_and_b32	s_restore_m0, s_restore_pc_hi, S_SAVE_PC_HI_FIRST_REPLAY_MASK
	s_lshr_b32	s_restore_m0, s_restore_m0, S_SAVE_PC_HI_FIRST_REPLAY_SHIFT
	s_lshl_b32	s_restore_m0, s_restore_m0, SQ_WAVE_IB_STS_FIRST_REPLAY_SHIFT
	s_or_b32	s_restore_tmp, s_restore_tmp, s_restore_m0
	s_and_b32	s_restore_m0, s_restore_pc_hi, S_SAVE_PC_HI_REPLAY_W64H_MASK
	s_lshr_b32	s_restore_m0, s_restore_m0, S_SAVE_PC_HI_REPLAY_W64H_SHIFT
	s_lshl_b32	s_restore_m0, s_restore_m0, SQ_WAVE_IB_STS_REPLAY_W64H_SHIFT
	s_or_b32	s_restore_tmp, s_restore_tmp, s_restore_m0

	s_and_b32	s_restore_m0, s_restore_status, SQ_WAVE_STATUS_INST_ATC_MASK
	s_lshr_b32	s_restore_m0, s_restore_m0, SQ_WAVE_STATUS_INST_ATC_SHIFT
	s_setreg_b32 	hwreg(HW_REG_IB_STS), s_restore_tmp
#endif

	s_and_b32	s_restore_pc_hi, s_restore_pc_hi, 0x0000ffff		//pc[47:32] //Do it here in order not to affect STATUS
	s_and_b64	exec, exec, exec					// Restore STATUS.EXECZ, not writable by s_setreg_b32
	s_and_b64	vcc, vcc, vcc						// Restore STATUS.VCCZ, not writable by s_setreg_b32
	s_setreg_b32	hwreg(HW_REG_STATUS), s_restore_status			// SCC is included, which is changed by previous salu

	s_rfe_b64	s_restore_pc_lo						//Return to the main shader program and resume execution

L_END_PGM:
	s_endpgm
end

function write_hwreg_to_mem(s, s_rsrc, s_mem_offset)
#if NO_SQC_STORE
	// Copy into VGPR for later TCP store.
	v_writelane_b32	v2, s, m0
	s_add_u32	m0, m0, 0x1
#else
	s_mov_b32	exec_lo, m0
	s_mov_b32	m0, s_mem_offset
	s_buffer_store_dword	s, s_rsrc, m0 glc:1
	s_add_u32	s_mem_offset, s_mem_offset, 4
	s_mov_b32	m0, exec_lo
#endif
end


function write_16sgpr_to_mem(s, s_rsrc, s_mem_offset)
#if NO_SQC_STORE
	// Copy into VGPR for later TCP store.
	for var sgpr_idx = 0; sgpr_idx < 16; sgpr_idx ++
		v_writelane_b32	v2, s[sgpr_idx], ttmp13
		s_add_u32	ttmp13, ttmp13, 0x1
	end
#else
	s_buffer_store_dwordx4	s[0], s_rsrc, 0 glc:1
	s_buffer_store_dwordx4	s[4], s_rsrc, 16 glc:1
	s_buffer_store_dwordx4	s[8], s_rsrc, 32 glc:1
	s_buffer_store_dwordx4	s[12], s_rsrc, 48 glc:1
	s_add_u32	s_rsrc[0], s_rsrc[0], 4*16
	s_addc_u32	s_rsrc[1], s_rsrc[1], 0x0
#endif
end

function write_12sgpr_to_mem(s, s_rsrc, s_mem_offset)
#if NO_SQC_STORE
	// Copy into VGPR for later TCP store.
	for var sgpr_idx = 0; sgpr_idx < 12; sgpr_idx ++
		v_writelane_b32	v2, s[sgpr_idx], ttmp13
		s_add_u32	ttmp13, ttmp13, 0x1
	end
#else
	s_buffer_store_dwordx4	s[0], s_rsrc, 0 glc:1
	s_buffer_store_dwordx4	s[4], s_rsrc, 16 glc:1
	s_buffer_store_dwordx4	s[8], s_rsrc, 32 glc:1
	s_add_u32	s_rsrc[0], s_rsrc[0], 4*12
	s_addc_u32	s_rsrc[1], s_rsrc[1], 0x0
#endif
end

function read_hwreg_from_mem(s, s_rsrc, s_mem_offset)
	s_buffer_load_dword	s, s_rsrc, s_mem_offset glc:1
	s_add_u32	s_mem_offset, s_mem_offset, 4
end

function read_16sgpr_from_mem(s, s_rsrc, s_mem_offset)
	s_sub_u32	s_mem_offset, s_mem_offset, 4*16
	s_buffer_load_dwordx16	s, s_rsrc, s_mem_offset glc:1
end

function read_8sgpr_from_mem(s, s_rsrc, s_mem_offset)
	s_sub_u32	s_mem_offset, s_mem_offset, 4*8
	s_buffer_load_dwordx8	s, s_rsrc, s_mem_offset glc:1
end

function read_4sgpr_from_mem(s, s_rsrc, s_mem_offset)
	s_sub_u32	s_mem_offset, s_mem_offset, 4*4
	s_buffer_load_dwordx4	s, s_rsrc, s_mem_offset glc:1
end


function get_lds_size_bytes(s_lds_size_byte)
	s_getreg_b32	s_lds_size_byte, hwreg(HW_REG_LDS_ALLOC, SQ_WAVE_LDS_ALLOC_LDS_SIZE_SHIFT, SQ_WAVE_LDS_ALLOC_LDS_SIZE_SIZE)
	s_lshl_b32	s_lds_size_byte, s_lds_size_byte, 8			//LDS size in dwords = lds_size * 64 *4Bytes // granularity 64DW
end

function get_vgpr_size_bytes(s_vgpr_size_byte, s_size)
	s_getreg_b32	s_vgpr_size_byte, hwreg(HW_REG_GPR_ALLOC,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SHIFT,SQ_WAVE_GPR_ALLOC_VGPR_SIZE_SIZE)
	s_add_u32	s_vgpr_size_byte, s_vgpr_size_byte, 1
	s_bitcmp1_b32	s_size, S_WAVE_SIZE
	s_cbranch_scc1	L_ENABLE_SHIFT_W64
	s_lshl_b32	s_vgpr_size_byte, s_vgpr_size_byte, (2+7)		//Number of VGPRs = (vgpr_size + 1) * 4 * 32 * 4   (non-zero value)
	s_branch	L_SHIFT_DONE
L_ENABLE_SHIFT_W64:
	s_lshl_b32	s_vgpr_size_byte, s_vgpr_size_byte, (2+8)		//Number of VGPRs = (vgpr_size + 1) * 4 * 64 * 4   (non-zero value)
L_SHIFT_DONE:
end

function get_svgpr_size_bytes(s_svgpr_size_byte)
	s_getreg_b32	s_svgpr_size_byte, hwreg(HW_REG_LDS_ALLOC,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SHIFT,SQ_WAVE_LDS_ALLOC_VGPR_SHARED_SIZE_SIZE)
	s_lshl_b32	s_svgpr_size_byte, s_svgpr_size_byte, (3+7)
end

function get_sgpr_size_bytes
	return 512
end

function get_hwreg_size_bytes
	return 128
end

function get_wave_size(s_reg)
	s_getreg_b32	s_reg, hwreg(HW_REG_IB_STS2,SQ_WAVE_IB_STS2_WAVE64_SHIFT,SQ_WAVE_IB_STS2_WAVE64_SIZE)
	s_lshl_b32	s_reg, s_reg, S_WAVE_SIZE
	s_or_b32	s_reg, s_save_spi_init_hi, s_reg			//share with exec_hi, it's at bit25
end