summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/radeon/r600_pipe_common.c
blob: 99172c6c55ff876aaa9aab7a4280af27711c1cb0 (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
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
/*
 * Copyright 2013 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 (including the next
 * paragraph) 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 AUTHORS OR COPYRIGHT HOLDERS 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.
 *
 * Authors: Marek Olšák <maraeo@gmail.com>
 *
 */

#include "r600_pipe_common.h"
#include "r600_cs.h"
#include "tgsi/tgsi_parse.h"
#include "util/list.h"
#include "util/u_draw_quad.h"
#include "util/u_memory.h"
#include "util/u_format_s3tc.h"
#include "util/u_upload_mgr.h"
#include "os/os_time.h"
#include "vl/vl_decoder.h"
#include "vl/vl_video_buffer.h"
#include "radeon/radeon_video.h"
#include <inttypes.h>
#include <sys/utsname.h>

#ifndef HAVE_LLVM
#define HAVE_LLVM 0
#endif

#if HAVE_LLVM
#include <llvm-c/TargetMachine.h>
#endif

#ifndef MESA_LLVM_VERSION_PATCH
#define MESA_LLVM_VERSION_PATCH 0
#endif

struct r600_multi_fence {
	struct pipe_reference reference;
	struct pipe_fence_handle *gfx;
	struct pipe_fence_handle *sdma;

	/* If the context wasn't flushed at fence creation, this is non-NULL. */
	struct {
		struct r600_common_context *ctx;
		unsigned ib_index;
	} gfx_unflushed;
};

/*
 * shader binary helpers.
 */
void radeon_shader_binary_init(struct ac_shader_binary *b)
{
	memset(b, 0, sizeof(*b));
}

void radeon_shader_binary_clean(struct ac_shader_binary *b)
{
	if (!b)
		return;
	FREE(b->code);
	FREE(b->config);
	FREE(b->rodata);
	FREE(b->global_symbol_offsets);
	FREE(b->relocs);
	FREE(b->disasm_string);
	FREE(b->llvm_ir_string);
}

/*
 * pipe_context
 */

/**
 * Write an EOP event.
 *
 * \param event		EVENT_TYPE_*
 * \param event_flags	Optional cache flush flags (TC)
 * \param data_sel	1 = fence, 3 = timestamp
 * \param buf		Buffer
 * \param va		GPU address
 * \param old_value	Previous fence value (for a bug workaround)
 * \param new_value	Fence value to write for this event.
 */
void r600_gfx_write_event_eop(struct r600_common_context *ctx,
			      unsigned event, unsigned event_flags,
			      unsigned data_sel,
			      struct r600_resource *buf, uint64_t va,
			      uint32_t new_fence, unsigned query_type)
{
	struct radeon_winsys_cs *cs = ctx->gfx.cs;
	unsigned op = EVENT_TYPE(event) |
		      EVENT_INDEX(5) |
		      event_flags;
	unsigned sel = EOP_DATA_SEL(data_sel);

	/* Wait for write confirmation before writing data, but don't send
	 * an interrupt. */
	if (ctx->chip_class >= SI && data_sel != EOP_DATA_SEL_DISCARD)
		sel |= EOP_INT_SEL(EOP_INT_SEL_SEND_DATA_AFTER_WR_CONFIRM);

	if (ctx->chip_class >= GFX9) {
		/* A ZPASS_DONE or PIXEL_STAT_DUMP_EVENT (of the DB occlusion
		 * counters) must immediately precede every timestamp event to
		 * prevent a GPU hang on GFX9.
		 *
		 * Occlusion queries don't need to do it here, because they
		 * always do ZPASS_DONE before the timestamp.
		 */
		if (ctx->chip_class == GFX9 &&
		    query_type != PIPE_QUERY_OCCLUSION_COUNTER &&
		    query_type != PIPE_QUERY_OCCLUSION_PREDICATE) {
			struct r600_resource *scratch = ctx->eop_bug_scratch;

			assert(16 * ctx->screen->info.num_render_backends <=
			       scratch->b.b.width0);
			radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 2, 0));
			radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1));
			radeon_emit(cs, scratch->gpu_address);
			radeon_emit(cs, scratch->gpu_address >> 32);

			radeon_add_to_buffer_list(ctx, &ctx->gfx, scratch,
						  RADEON_USAGE_WRITE, RADEON_PRIO_QUERY);
		}

		radeon_emit(cs, PKT3(PKT3_RELEASE_MEM, 6, 0));
		radeon_emit(cs, op);
		radeon_emit(cs, sel);
		radeon_emit(cs, va);		/* address lo */
		radeon_emit(cs, va >> 32);	/* address hi */
		radeon_emit(cs, new_fence);	/* immediate data lo */
		radeon_emit(cs, 0); /* immediate data hi */
		radeon_emit(cs, 0); /* unused */
	} else {
		if (ctx->chip_class == CIK ||
		    ctx->chip_class == VI) {
			struct r600_resource *scratch = ctx->eop_bug_scratch;
			uint64_t va = scratch->gpu_address;

			/* Two EOP events are required to make all engines go idle
			 * (and optional cache flushes executed) before the timestamp
			 * is written.
			 */
			radeon_emit(cs, PKT3(PKT3_EVENT_WRITE_EOP, 4, 0));
			radeon_emit(cs, op);
			radeon_emit(cs, va);
			radeon_emit(cs, ((va >> 32) & 0xffff) | sel);
			radeon_emit(cs, 0); /* immediate data */
			radeon_emit(cs, 0); /* unused */

			radeon_add_to_buffer_list(ctx, &ctx->gfx, scratch,
						  RADEON_USAGE_WRITE, RADEON_PRIO_QUERY);
		}

		radeon_emit(cs, PKT3(PKT3_EVENT_WRITE_EOP, 4, 0));
		radeon_emit(cs, op);
		radeon_emit(cs, va);
		radeon_emit(cs, ((va >> 32) & 0xffff) | sel);
		radeon_emit(cs, new_fence); /* immediate data */
		radeon_emit(cs, 0); /* unused */
	}

	if (buf)
		r600_emit_reloc(ctx, &ctx->gfx, buf, RADEON_USAGE_WRITE,
				RADEON_PRIO_QUERY);
}

unsigned r600_gfx_write_fence_dwords(struct r600_common_screen *screen)
{
	unsigned dwords = 6;

	if (screen->chip_class == CIK ||
	    screen->chip_class == VI)
		dwords *= 2;

	if (!screen->info.has_virtual_memory)
		dwords += 2;

	return dwords;
}

void r600_gfx_wait_fence(struct r600_common_context *ctx,
			 uint64_t va, uint32_t ref, uint32_t mask)
{
	struct radeon_winsys_cs *cs = ctx->gfx.cs;

	radeon_emit(cs, PKT3(PKT3_WAIT_REG_MEM, 5, 0));
	radeon_emit(cs, WAIT_REG_MEM_EQUAL | WAIT_REG_MEM_MEM_SPACE(1));
	radeon_emit(cs, va);
	radeon_emit(cs, va >> 32);
	radeon_emit(cs, ref); /* reference value */
	radeon_emit(cs, mask); /* mask */
	radeon_emit(cs, 4); /* poll interval */
}

void r600_draw_rectangle(struct blitter_context *blitter,
			 int x1, int y1, int x2, int y2, float depth,
			 enum blitter_attrib_type type,
			 const union pipe_color_union *attrib)
{
	struct r600_common_context *rctx =
		(struct r600_common_context*)util_blitter_get_pipe(blitter);
	struct pipe_viewport_state viewport;
	struct pipe_resource *buf = NULL;
	unsigned offset = 0;
	float *vb;

	if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) {
		util_blitter_draw_rectangle(blitter, x1, y1, x2, y2, depth, type, attrib);
		return;
	}

	/* Some operations (like color resolve on r6xx) don't work
	 * with the conventional primitive types.
	 * One that works is PT_RECTLIST, which we use here. */

	/* setup viewport */
	viewport.scale[0] = 1.0f;
	viewport.scale[1] = 1.0f;
	viewport.scale[2] = 1.0f;
	viewport.translate[0] = 0.0f;
	viewport.translate[1] = 0.0f;
	viewport.translate[2] = 0.0f;
	rctx->b.set_viewport_states(&rctx->b, 0, 1, &viewport);

	/* Upload vertices. The hw rectangle has only 3 vertices,
	 * I guess the 4th one is derived from the first 3.
	 * The vertex specification should match u_blitter's vertex element state. */
	u_upload_alloc(rctx->b.stream_uploader, 0, sizeof(float) * 24,
		       rctx->screen->info.tcc_cache_line_size,
                       &offset, &buf, (void**)&vb);
	if (!buf)
		return;

	vb[0] = x1;
	vb[1] = y1;
	vb[2] = depth;
	vb[3] = 1;

	vb[8] = x1;
	vb[9] = y2;
	vb[10] = depth;
	vb[11] = 1;

	vb[16] = x2;
	vb[17] = y1;
	vb[18] = depth;
	vb[19] = 1;

	if (attrib) {
		memcpy(vb+4, attrib->f, sizeof(float)*4);
		memcpy(vb+12, attrib->f, sizeof(float)*4);
		memcpy(vb+20, attrib->f, sizeof(float)*4);
	}

	/* draw */
	util_draw_vertex_buffer(&rctx->b, NULL, buf, blitter->vb_slot, offset,
				R600_PRIM_RECTANGLE_LIST, 3, 2);
	pipe_resource_reference(&buf, NULL);
}

static void r600_dma_emit_wait_idle(struct r600_common_context *rctx)
{
	struct radeon_winsys_cs *cs = rctx->dma.cs;

	/* NOP waits for idle on Evergreen and later. */
	if (rctx->chip_class >= CIK)
		radeon_emit(cs, 0x00000000); /* NOP */
	else if (rctx->chip_class >= EVERGREEN)
		radeon_emit(cs, 0xf0000000); /* NOP */
	else {
		/* TODO: R600-R700 should use the FENCE packet.
		 * CS checker support is required. */
	}
}

void r600_need_dma_space(struct r600_common_context *ctx, unsigned num_dw,
                         struct r600_resource *dst, struct r600_resource *src)
{
	uint64_t vram = ctx->dma.cs->used_vram;
	uint64_t gtt = ctx->dma.cs->used_gart;

	if (dst) {
		vram += dst->vram_usage;
		gtt += dst->gart_usage;
	}
	if (src) {
		vram += src->vram_usage;
		gtt += src->gart_usage;
	}

	/* Flush the GFX IB if DMA depends on it. */
	if (radeon_emitted(ctx->gfx.cs, ctx->initial_gfx_cs_size) &&
	    ((dst &&
	      ctx->ws->cs_is_buffer_referenced(ctx->gfx.cs, dst->buf,
					       RADEON_USAGE_READWRITE)) ||
	     (src &&
	      ctx->ws->cs_is_buffer_referenced(ctx->gfx.cs, src->buf,
					       RADEON_USAGE_WRITE))))
		ctx->gfx.flush(ctx, RADEON_FLUSH_ASYNC, NULL);

	/* Flush if there's not enough space, or if the memory usage per IB
	 * is too large.
	 *
	 * IBs using too little memory are limited by the IB submission overhead.
	 * IBs using too much memory are limited by the kernel/TTM overhead.
	 * Too long IBs create CPU-GPU pipeline bubbles and add latency.
	 *
	 * This heuristic makes sure that DMA requests are executed
	 * very soon after the call is made and lowers memory usage.
	 * It improves texture upload performance by keeping the DMA
	 * engine busy while uploads are being submitted.
	 */
	num_dw++; /* for emit_wait_idle below */
	if (!ctx->ws->cs_check_space(ctx->dma.cs, num_dw) ||
	    ctx->dma.cs->used_vram + ctx->dma.cs->used_gart > 64 * 1024 * 1024 ||
	    !radeon_cs_memory_below_limit(ctx->screen, ctx->dma.cs, vram, gtt)) {
		ctx->dma.flush(ctx, RADEON_FLUSH_ASYNC, NULL);
		assert((num_dw + ctx->dma.cs->current.cdw) <= ctx->dma.cs->current.max_dw);
	}

	/* Wait for idle if either buffer has been used in the IB before to
	 * prevent read-after-write hazards.
	 */
	if ((dst &&
	     ctx->ws->cs_is_buffer_referenced(ctx->dma.cs, dst->buf,
					      RADEON_USAGE_READWRITE)) ||
	    (src &&
	     ctx->ws->cs_is_buffer_referenced(ctx->dma.cs, src->buf,
					      RADEON_USAGE_WRITE)))
		r600_dma_emit_wait_idle(ctx);

	/* If GPUVM is not supported, the CS checker needs 2 entries
	 * in the buffer list per packet, which has to be done manually.
	 */
	if (ctx->screen->info.has_virtual_memory) {
		if (dst)
			radeon_add_to_buffer_list(ctx, &ctx->dma, dst,
						  RADEON_USAGE_WRITE,
						  RADEON_PRIO_SDMA_BUFFER);
		if (src)
			radeon_add_to_buffer_list(ctx, &ctx->dma, src,
						  RADEON_USAGE_READ,
						  RADEON_PRIO_SDMA_BUFFER);
	}

	/* this function is called before all DMA calls, so increment this. */
	ctx->num_dma_calls++;
}

static void r600_memory_barrier(struct pipe_context *ctx, unsigned flags)
{
}

void r600_preflush_suspend_features(struct r600_common_context *ctx)
{
	/* suspend queries */
	if (!LIST_IS_EMPTY(&ctx->active_queries))
		r600_suspend_queries(ctx);

	ctx->streamout.suspended = false;
	if (ctx->streamout.begin_emitted) {
		r600_emit_streamout_end(ctx);
		ctx->streamout.suspended = true;
	}
}

void r600_postflush_resume_features(struct r600_common_context *ctx)
{
	if (ctx->streamout.suspended) {
		ctx->streamout.append_bitmask = ctx->streamout.enabled_mask;
		r600_streamout_buffers_dirty(ctx);
	}

	/* resume queries */
	if (!LIST_IS_EMPTY(&ctx->active_queries))
		r600_resume_queries(ctx);
}

static void r600_flush_from_st(struct pipe_context *ctx,
			       struct pipe_fence_handle **fence,
			       unsigned flags)
{
	struct pipe_screen *screen = ctx->screen;
	struct r600_common_context *rctx = (struct r600_common_context *)ctx;
	struct radeon_winsys *ws = rctx->ws;
	struct pipe_fence_handle *gfx_fence = NULL;
	struct pipe_fence_handle *sdma_fence = NULL;
	bool deferred_fence = false;
	unsigned rflags = RADEON_FLUSH_ASYNC;

	if (flags & PIPE_FLUSH_END_OF_FRAME)
		rflags |= RADEON_FLUSH_END_OF_FRAME;

	/* DMA IBs are preambles to gfx IBs, therefore must be flushed first. */
	if (rctx->dma.cs)
		rctx->dma.flush(rctx, rflags, fence ? &sdma_fence : NULL);

	if (!radeon_emitted(rctx->gfx.cs, rctx->initial_gfx_cs_size)) {
		if (fence)
			ws->fence_reference(&gfx_fence, rctx->last_gfx_fence);
		if (!(flags & PIPE_FLUSH_DEFERRED))
			ws->cs_sync_flush(rctx->gfx.cs);
	} else {
		/* Instead of flushing, create a deferred fence. Constraints:
		 * - The state tracker must allow a deferred flush.
		 * - The state tracker must request a fence.
		 * Thread safety in fence_finish must be ensured by the state tracker.
		 */
		if (flags & PIPE_FLUSH_DEFERRED && fence) {
			gfx_fence = rctx->ws->cs_get_next_fence(rctx->gfx.cs);
			deferred_fence = true;
		} else {
			rctx->gfx.flush(rctx, rflags, fence ? &gfx_fence : NULL);
		}
	}

	/* Both engines can signal out of order, so we need to keep both fences. */
	if (fence) {
		struct r600_multi_fence *multi_fence =
			CALLOC_STRUCT(r600_multi_fence);
		if (!multi_fence)
			return;

		multi_fence->reference.count = 1;
		/* If both fences are NULL, fence_finish will always return true. */
		multi_fence->gfx = gfx_fence;
		multi_fence->sdma = sdma_fence;

		if (deferred_fence) {
			multi_fence->gfx_unflushed.ctx = rctx;
			multi_fence->gfx_unflushed.ib_index = rctx->num_gfx_cs_flushes;
		}

		screen->fence_reference(screen, fence, NULL);
		*fence = (struct pipe_fence_handle*)multi_fence;
	}

	if (!(flags & PIPE_FLUSH_DEFERRED)) {
		if (rctx->dma.cs)
			ws->cs_sync_flush(rctx->dma.cs);
		ws->cs_sync_flush(rctx->gfx.cs);
	}
}

static void r600_flush_dma_ring(void *ctx, unsigned flags,
				struct pipe_fence_handle **fence)
{
	struct r600_common_context *rctx = (struct r600_common_context *)ctx;
	struct radeon_winsys_cs *cs = rctx->dma.cs;
	struct radeon_saved_cs saved;
	bool check_vm =
		(rctx->screen->debug_flags & DBG_CHECK_VM) &&
		rctx->check_vm_faults;

	if (!radeon_emitted(cs, 0)) {
		if (fence)
			rctx->ws->fence_reference(fence, rctx->last_sdma_fence);
		return;
	}

	if (check_vm)
		radeon_save_cs(rctx->ws, cs, &saved, true);

	rctx->ws->cs_flush(cs, flags, &rctx->last_sdma_fence);
	if (fence)
		rctx->ws->fence_reference(fence, rctx->last_sdma_fence);

	if (check_vm) {
		/* Use conservative timeout 800ms, after which we won't wait any
		 * longer and assume the GPU is hung.
		 */
		rctx->ws->fence_wait(rctx->ws, rctx->last_sdma_fence, 800*1000*1000);

		rctx->check_vm_faults(rctx, &saved, RING_DMA);
		radeon_clear_saved_cs(&saved);
	}
}

/**
 * Store a linearized copy of all chunks of \p cs together with the buffer
 * list in \p saved.
 */
void radeon_save_cs(struct radeon_winsys *ws, struct radeon_winsys_cs *cs,
		    struct radeon_saved_cs *saved, bool get_buffer_list)
{
	uint32_t *buf;
	unsigned i;

	/* Save the IB chunks. */
	saved->num_dw = cs->prev_dw + cs->current.cdw;
	saved->ib = MALLOC(4 * saved->num_dw);
	if (!saved->ib)
		goto oom;

	buf = saved->ib;
	for (i = 0; i < cs->num_prev; ++i) {
		memcpy(buf, cs->prev[i].buf, cs->prev[i].cdw * 4);
		buf += cs->prev[i].cdw;
	}
	memcpy(buf, cs->current.buf, cs->current.cdw * 4);

	if (!get_buffer_list)
		return;

	/* Save the buffer list. */
	saved->bo_count = ws->cs_get_buffer_list(cs, NULL);
	saved->bo_list = CALLOC(saved->bo_count,
				sizeof(saved->bo_list[0]));
	if (!saved->bo_list) {
		FREE(saved->ib);
		goto oom;
	}
	ws->cs_get_buffer_list(cs, saved->bo_list);

	return;

oom:
	fprintf(stderr, "%s: out of memory\n", __func__);
	memset(saved, 0, sizeof(*saved));
}

void radeon_clear_saved_cs(struct radeon_saved_cs *saved)
{
	FREE(saved->ib);
	FREE(saved->bo_list);

	memset(saved, 0, sizeof(*saved));
}

static enum pipe_reset_status r600_get_reset_status(struct pipe_context *ctx)
{
	struct r600_common_context *rctx = (struct r600_common_context *)ctx;
	unsigned latest = rctx->ws->query_value(rctx->ws,
						RADEON_GPU_RESET_COUNTER);

	if (rctx->gpu_reset_counter == latest)
		return PIPE_NO_RESET;

	rctx->gpu_reset_counter = latest;
	return PIPE_UNKNOWN_CONTEXT_RESET;
}

static void r600_set_debug_callback(struct pipe_context *ctx,
				    const struct pipe_debug_callback *cb)
{
	struct r600_common_context *rctx = (struct r600_common_context *)ctx;

	if (cb)
		rctx->debug = *cb;
	else
		memset(&rctx->debug, 0, sizeof(rctx->debug));
}

static void r600_set_device_reset_callback(struct pipe_context *ctx,
					   const struct pipe_device_reset_callback *cb)
{
	struct r600_common_context *rctx = (struct r600_common_context *)ctx;

	if (cb)
		rctx->device_reset_callback = *cb;
	else
		memset(&rctx->device_reset_callback, 0,
		       sizeof(rctx->device_reset_callback));
}

bool r600_check_device_reset(struct r600_common_context *rctx)
{
	enum pipe_reset_status status;

	if (!rctx->device_reset_callback.reset)
		return false;

	if (!rctx->b.get_device_reset_status)
		return false;

	status = rctx->b.get_device_reset_status(&rctx->b);
	if (status == PIPE_NO_RESET)
		return false;

	rctx->device_reset_callback.reset(rctx->device_reset_callback.data, status);
	return true;
}

static void r600_dma_clear_buffer_fallback(struct pipe_context *ctx,
					   struct pipe_resource *dst,
					   uint64_t offset, uint64_t size,
					   unsigned value)
{
	struct r600_common_context *rctx = (struct r600_common_context *)ctx;

	rctx->clear_buffer(ctx, dst, offset, size, value, R600_COHERENCY_NONE);
}

static bool r600_resource_commit(struct pipe_context *pctx,
				 struct pipe_resource *resource,
				 unsigned level, struct pipe_box *box,
				 bool commit)
{
	struct r600_common_context *ctx = (struct r600_common_context *)pctx;
	struct r600_resource *res = r600_resource(resource);

	/*
	 * Since buffer commitment changes cannot be pipelined, we need to
	 * (a) flush any pending commands that refer to the buffer we're about
	 *     to change, and
	 * (b) wait for threaded submit to finish, including those that were
	 *     triggered by some other, earlier operation.
	 */
	if (radeon_emitted(ctx->gfx.cs, ctx->initial_gfx_cs_size) &&
	    ctx->ws->cs_is_buffer_referenced(ctx->gfx.cs,
					     res->buf, RADEON_USAGE_READWRITE)) {
		ctx->gfx.flush(ctx, RADEON_FLUSH_ASYNC, NULL);
	}
	if (radeon_emitted(ctx->dma.cs, 0) &&
	    ctx->ws->cs_is_buffer_referenced(ctx->dma.cs,
					     res->buf, RADEON_USAGE_READWRITE)) {
		ctx->dma.flush(ctx, RADEON_FLUSH_ASYNC, NULL);
	}

	ctx->ws->cs_sync_flush(ctx->dma.cs);
	ctx->ws->cs_sync_flush(ctx->gfx.cs);

	assert(resource->target == PIPE_BUFFER);

	return ctx->ws->buffer_commit(res->buf, box->x, box->width, commit);
}

bool r600_common_context_init(struct r600_common_context *rctx,
			      struct r600_common_screen *rscreen,
			      unsigned context_flags)
{
	slab_create_child(&rctx->pool_transfers, &rscreen->pool_transfers);
	slab_create_child(&rctx->pool_transfers_unsync, &rscreen->pool_transfers);

	rctx->screen = rscreen;
	rctx->ws = rscreen->ws;
	rctx->family = rscreen->family;
	rctx->chip_class = rscreen->chip_class;

	rctx->b.invalidate_resource = r600_invalidate_resource;
	rctx->b.resource_commit = r600_resource_commit;
	rctx->b.transfer_map = u_transfer_map_vtbl;
	rctx->b.transfer_flush_region = u_transfer_flush_region_vtbl;
	rctx->b.transfer_unmap = u_transfer_unmap_vtbl;
	rctx->b.texture_subdata = u_default_texture_subdata;
	rctx->b.memory_barrier = r600_memory_barrier;
	rctx->b.flush = r600_flush_from_st;
	rctx->b.set_debug_callback = r600_set_debug_callback;
	rctx->dma_clear_buffer = r600_dma_clear_buffer_fallback;

	/* evergreen_compute.c has a special codepath for global buffers.
	 * Everything else can use the direct path.
	 */
	if ((rscreen->chip_class == EVERGREEN || rscreen->chip_class == CAYMAN) &&
	    (context_flags & PIPE_CONTEXT_COMPUTE_ONLY))
		rctx->b.buffer_subdata = u_default_buffer_subdata;
	else
		rctx->b.buffer_subdata = r600_buffer_subdata;

	if (rscreen->info.drm_major == 2 && rscreen->info.drm_minor >= 43) {
		rctx->b.get_device_reset_status = r600_get_reset_status;
		rctx->gpu_reset_counter =
			rctx->ws->query_value(rctx->ws,
					      RADEON_GPU_RESET_COUNTER);
	}

	rctx->b.set_device_reset_callback = r600_set_device_reset_callback;

	r600_init_context_texture_functions(rctx);
	r600_init_viewport_functions(rctx);
	r600_streamout_init(rctx);
	r600_query_init(rctx);
	cayman_init_msaa(&rctx->b);

	if (rctx->chip_class == CIK ||
	    rctx->chip_class == VI ||
	    rctx->chip_class == GFX9) {
		rctx->eop_bug_scratch = (struct r600_resource*)
			pipe_buffer_create(&rscreen->b, 0, PIPE_USAGE_DEFAULT,
					   16 * rscreen->info.num_render_backends);
		if (!rctx->eop_bug_scratch)
			return false;
	}

	rctx->allocator_zeroed_memory =
		u_suballocator_create(&rctx->b, rscreen->info.gart_page_size,
				      0, PIPE_USAGE_DEFAULT, 0, true);
	if (!rctx->allocator_zeroed_memory)
		return false;

	rctx->b.stream_uploader = u_upload_create(&rctx->b, 1024 * 1024,
						  0, PIPE_USAGE_STREAM);
	if (!rctx->b.stream_uploader)
		return false;

	rctx->b.const_uploader = u_upload_create(&rctx->b, 128 * 1024,
						 0, PIPE_USAGE_DEFAULT);
	if (!rctx->b.const_uploader)
		return false;

	rctx->ctx = rctx->ws->ctx_create(rctx->ws);
	if (!rctx->ctx)
		return false;

	if (rscreen->info.num_sdma_rings && !(rscreen->debug_flags & DBG_NO_ASYNC_DMA)) {
		rctx->dma.cs = rctx->ws->cs_create(rctx->ctx, RING_DMA,
						   r600_flush_dma_ring,
						   rctx);
		rctx->dma.flush = r600_flush_dma_ring;
	}

	return true;
}

void r600_common_context_cleanup(struct r600_common_context *rctx)
{
	unsigned i,j;

	/* Release DCC stats. */
	for (i = 0; i < ARRAY_SIZE(rctx->dcc_stats); i++) {
		assert(!rctx->dcc_stats[i].query_active);

		for (j = 0; j < ARRAY_SIZE(rctx->dcc_stats[i].ps_stats); j++)
			if (rctx->dcc_stats[i].ps_stats[j])
				rctx->b.destroy_query(&rctx->b,
						      rctx->dcc_stats[i].ps_stats[j]);

		r600_texture_reference(&rctx->dcc_stats[i].tex, NULL);
	}

	if (rctx->query_result_shader)
		rctx->b.delete_compute_state(&rctx->b, rctx->query_result_shader);

	if (rctx->gfx.cs)
		rctx->ws->cs_destroy(rctx->gfx.cs);
	if (rctx->dma.cs)
		rctx->ws->cs_destroy(rctx->dma.cs);
	if (rctx->ctx)
		rctx->ws->ctx_destroy(rctx->ctx);

	if (rctx->b.stream_uploader)
		u_upload_destroy(rctx->b.stream_uploader);
	if (rctx->b.const_uploader)
		u_upload_destroy(rctx->b.const_uploader);

	slab_destroy_child(&rctx->pool_transfers);
	slab_destroy_child(&rctx->pool_transfers_unsync);

	if (rctx->allocator_zeroed_memory) {
		u_suballocator_destroy(rctx->allocator_zeroed_memory);
	}
	rctx->ws->fence_reference(&rctx->last_gfx_fence, NULL);
	rctx->ws->fence_reference(&rctx->last_sdma_fence, NULL);
	r600_resource_reference(&rctx->eop_bug_scratch, NULL);
}

/*
 * pipe_screen
 */

static const struct debug_named_value common_debug_options[] = {
	/* logging */
	{ "tex", DBG_TEX, "Print texture info" },
	{ "nir", DBG_NIR, "Enable experimental NIR shaders" },
	{ "compute", DBG_COMPUTE, "Print compute info" },
	{ "vm", DBG_VM, "Print virtual addresses when creating resources" },
	{ "info", DBG_INFO, "Print driver information" },

	/* shaders */
	{ "fs", DBG_FS, "Print fetch shaders" },
	{ "vs", DBG_VS, "Print vertex shaders" },
	{ "gs", DBG_GS, "Print geometry shaders" },
	{ "ps", DBG_PS, "Print pixel shaders" },
	{ "cs", DBG_CS, "Print compute shaders" },
	{ "tcs", DBG_TCS, "Print tessellation control shaders" },
	{ "tes", DBG_TES, "Print tessellation evaluation shaders" },
	{ "noir", DBG_NO_IR, "Don't print the LLVM IR"},
	{ "notgsi", DBG_NO_TGSI, "Don't print the TGSI"},
	{ "noasm", DBG_NO_ASM, "Don't print disassembled shaders"},
	{ "preoptir", DBG_PREOPT_IR, "Print the LLVM IR before initial optimizations" },
	{ "checkir", DBG_CHECK_IR, "Enable additional sanity checks on shader IR" },
	{ "nooptvariant", DBG_NO_OPT_VARIANT, "Disable compiling optimized shader variants." },

	{ "testdma", DBG_TEST_DMA, "Invoke SDMA tests and exit." },
	{ "testvmfaultcp", DBG_TEST_VMFAULT_CP, "Invoke a CP VM fault test and exit." },
	{ "testvmfaultsdma", DBG_TEST_VMFAULT_SDMA, "Invoke a SDMA VM fault test and exit." },
	{ "testvmfaultshader", DBG_TEST_VMFAULT_SHADER, "Invoke a shader VM fault test and exit." },

	/* features */
	{ "nodma", DBG_NO_ASYNC_DMA, "Disable asynchronous DMA" },
	{ "nohyperz", DBG_NO_HYPERZ, "Disable Hyper-Z" },
	/* GL uses the word INVALIDATE, gallium uses the word DISCARD */
	{ "noinvalrange", DBG_NO_DISCARD_RANGE, "Disable handling of INVALIDATE_RANGE map flags" },
	{ "no2d", DBG_NO_2D_TILING, "Disable 2D tiling" },
	{ "notiling", DBG_NO_TILING, "Disable tiling" },
	{ "switch_on_eop", DBG_SWITCH_ON_EOP, "Program WD/IA to switch on end-of-packet." },
	{ "forcedma", DBG_FORCE_DMA, "Use asynchronous DMA for all operations when possible." },
	{ "precompile", DBG_PRECOMPILE, "Compile one shader variant at shader creation." },
	{ "nowc", DBG_NO_WC, "Disable GTT write combining" },
	{ "check_vm", DBG_CHECK_VM, "Check VM faults and dump debug info." },
	{ "nodcc", DBG_NO_DCC, "Disable DCC." },
	{ "nodccclear", DBG_NO_DCC_CLEAR, "Disable DCC fast clear." },
	{ "norbplus", DBG_NO_RB_PLUS, "Disable RB+." },
	{ "sisched", DBG_SI_SCHED, "Enable LLVM SI Machine Instruction Scheduler." },
	{ "mono", DBG_MONOLITHIC_SHADERS, "Use old-style monolithic shaders compiled on demand" },
	{ "unsafemath", DBG_UNSAFE_MATH, "Enable unsafe math shader optimizations" },
	{ "nodccfb", DBG_NO_DCC_FB, "Disable separate DCC on the main framebuffer" },

	DEBUG_NAMED_VALUE_END /* must be last */
};

static const char* r600_get_vendor(struct pipe_screen* pscreen)
{
	return "X.Org";
}

static const char* r600_get_device_vendor(struct pipe_screen* pscreen)
{
	return "AMD";
}

static const char *r600_get_marketing_name(struct radeon_winsys *ws)
{
	if (!ws->get_chip_name)
		return NULL;
	return ws->get_chip_name(ws);
}

static const char *r600_get_family_name(const struct r600_common_screen *rscreen)
{
	switch (rscreen->info.family) {
	case CHIP_R600: return "AMD R600";
	case CHIP_RV610: return "AMD RV610";
	case CHIP_RV630: return "AMD RV630";
	case CHIP_RV670: return "AMD RV670";
	case CHIP_RV620: return "AMD RV620";
	case CHIP_RV635: return "AMD RV635";
	case CHIP_RS780: return "AMD RS780";
	case CHIP_RS880: return "AMD RS880";
	case CHIP_RV770: return "AMD RV770";
	case CHIP_RV730: return "AMD RV730";
	case CHIP_RV710: return "AMD RV710";
	case CHIP_RV740: return "AMD RV740";
	case CHIP_CEDAR: return "AMD CEDAR";
	case CHIP_REDWOOD: return "AMD REDWOOD";
	case CHIP_JUNIPER: return "AMD JUNIPER";
	case CHIP_CYPRESS: return "AMD CYPRESS";
	case CHIP_HEMLOCK: return "AMD HEMLOCK";
	case CHIP_PALM: return "AMD PALM";
	case CHIP_SUMO: return "AMD SUMO";
	case CHIP_SUMO2: return "AMD SUMO2";
	case CHIP_BARTS: return "AMD BARTS";
	case CHIP_TURKS: return "AMD TURKS";
	case CHIP_CAICOS: return "AMD CAICOS";
	case CHIP_CAYMAN: return "AMD CAYMAN";
	case CHIP_ARUBA: return "AMD ARUBA";
	case CHIP_TAHITI: return "AMD TAHITI";
	case CHIP_PITCAIRN: return "AMD PITCAIRN";
	case CHIP_VERDE: return "AMD CAPE VERDE";
	case CHIP_OLAND: return "AMD OLAND";
	case CHIP_HAINAN: return "AMD HAINAN";
	case CHIP_BONAIRE: return "AMD BONAIRE";
	case CHIP_KAVERI: return "AMD KAVERI";
	case CHIP_KABINI: return "AMD KABINI";
	case CHIP_HAWAII: return "AMD HAWAII";
	case CHIP_MULLINS: return "AMD MULLINS";
	case CHIP_TONGA: return "AMD TONGA";
	case CHIP_ICELAND: return "AMD ICELAND";
	case CHIP_CARRIZO: return "AMD CARRIZO";
	case CHIP_FIJI: return "AMD FIJI";
	case CHIP_POLARIS10: return "AMD POLARIS10";
	case CHIP_POLARIS11: return "AMD POLARIS11";
	case CHIP_POLARIS12: return "AMD POLARIS12";
	case CHIP_STONEY: return "AMD STONEY";
	case CHIP_VEGA10: return "AMD VEGA10";
	case CHIP_RAVEN: return "AMD RAVEN";
	default: return "AMD unknown";
	}
}

static void r600_disk_cache_create(struct r600_common_screen *rscreen)
{
	/* Don't use the cache if shader dumping is enabled. */
	if (rscreen->debug_flags & DBG_ALL_SHADERS)
		return;

	uint32_t mesa_timestamp;
	if (disk_cache_get_function_timestamp(r600_disk_cache_create,
					      &mesa_timestamp)) {
		char *timestamp_str;
		int res = -1;
		if (rscreen->chip_class < SI) {
			res = asprintf(&timestamp_str, "%u",mesa_timestamp);
		}
#if HAVE_LLVM
		else {
			uint32_t llvm_timestamp;
			if (disk_cache_get_function_timestamp(LLVMInitializeAMDGPUTargetInfo,
							      &llvm_timestamp)) {
				res = asprintf(&timestamp_str, "%u_%u",
					       mesa_timestamp, llvm_timestamp);
			}
		}
#endif
		if (res != -1) {
			/* These flags affect shader compilation. */
			uint64_t shader_debug_flags =
				rscreen->debug_flags &
				(DBG_FS_CORRECT_DERIVS_AFTER_KILL |
				 DBG_SI_SCHED |
				 DBG_UNSAFE_MATH);

			rscreen->disk_shader_cache =
				disk_cache_create(r600_get_family_name(rscreen),
						  timestamp_str,
						  shader_debug_flags);
			free(timestamp_str);
		}
	}
}

static struct disk_cache *r600_get_disk_shader_cache(struct pipe_screen *pscreen)
{
	struct r600_common_screen *rscreen = (struct r600_common_screen*)pscreen;
	return rscreen->disk_shader_cache;
}

static const char* r600_get_name(struct pipe_screen* pscreen)
{
	struct r600_common_screen *rscreen = (struct r600_common_screen*)pscreen;

	return rscreen->renderer_string;
}

static float r600_get_paramf(struct pipe_screen* pscreen,
			     enum pipe_capf param)
{
	struct r600_common_screen *rscreen = (struct r600_common_screen *)pscreen;

	switch (param) {
	case PIPE_CAPF_MAX_LINE_WIDTH:
	case PIPE_CAPF_MAX_LINE_WIDTH_AA:
	case PIPE_CAPF_MAX_POINT_WIDTH:
	case PIPE_CAPF_MAX_POINT_WIDTH_AA:
		if (rscreen->family >= CHIP_CEDAR)
			return 16384.0f;
		else
			return 8192.0f;
	case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
		return 16.0f;
	case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
		return 16.0f;
	case PIPE_CAPF_GUARD_BAND_LEFT:
	case PIPE_CAPF_GUARD_BAND_TOP:
	case PIPE_CAPF_GUARD_BAND_RIGHT:
	case PIPE_CAPF_GUARD_BAND_BOTTOM:
		return 0.0f;
	}
	return 0.0f;
}

static int r600_get_video_param(struct pipe_screen *screen,
				enum pipe_video_profile profile,
				enum pipe_video_entrypoint entrypoint,
				enum pipe_video_cap param)
{
	switch (param) {
	case PIPE_VIDEO_CAP_SUPPORTED:
		return vl_profile_supported(screen, profile, entrypoint);
	case PIPE_VIDEO_CAP_NPOT_TEXTURES:
		return 1;
	case PIPE_VIDEO_CAP_MAX_WIDTH:
	case PIPE_VIDEO_CAP_MAX_HEIGHT:
		return vl_video_buffer_max_size(screen);
	case PIPE_VIDEO_CAP_PREFERED_FORMAT:
		return PIPE_FORMAT_NV12;
	case PIPE_VIDEO_CAP_PREFERS_INTERLACED:
		return false;
	case PIPE_VIDEO_CAP_SUPPORTS_INTERLACED:
		return false;
	case PIPE_VIDEO_CAP_SUPPORTS_PROGRESSIVE:
		return true;
	case PIPE_VIDEO_CAP_MAX_LEVEL:
		return vl_level_supported(screen, profile);
	default:
		return 0;
	}
}

const char *r600_get_llvm_processor_name(enum radeon_family family)
{
	switch (family) {
	case CHIP_R600:
	case CHIP_RV630:
	case CHIP_RV635:
	case CHIP_RV670:
		return "r600";
	case CHIP_RV610:
	case CHIP_RV620:
	case CHIP_RS780:
	case CHIP_RS880:
		return "rs880";
	case CHIP_RV710:
		return "rv710";
	case CHIP_RV730:
		return "rv730";
	case CHIP_RV740:
	case CHIP_RV770:
		return "rv770";
	case CHIP_PALM:
	case CHIP_CEDAR:
		return "cedar";
	case CHIP_SUMO:
	case CHIP_SUMO2:
		return "sumo";
	case CHIP_REDWOOD:
		return "redwood";
	case CHIP_JUNIPER:
		return "juniper";
	case CHIP_HEMLOCK:
	case CHIP_CYPRESS:
		return "cypress";
	case CHIP_BARTS:
		return "barts";
	case CHIP_TURKS:
		return "turks";
	case CHIP_CAICOS:
		return "caicos";
	case CHIP_CAYMAN:
        case CHIP_ARUBA:
		return "cayman";

	case CHIP_TAHITI: return "tahiti";
	case CHIP_PITCAIRN: return "pitcairn";
	case CHIP_VERDE: return "verde";
	case CHIP_OLAND: return "oland";
	case CHIP_HAINAN: return "hainan";
	case CHIP_BONAIRE: return "bonaire";
	case CHIP_KABINI: return "kabini";
	case CHIP_KAVERI: return "kaveri";
	case CHIP_HAWAII: return "hawaii";
	case CHIP_MULLINS:
		return "mullins";
	case CHIP_TONGA: return "tonga";
	case CHIP_ICELAND: return "iceland";
	case CHIP_CARRIZO: return "carrizo";
	case CHIP_FIJI:
		return "fiji";
	case CHIP_STONEY:
		return "stoney";
	case CHIP_POLARIS10:
		return "polaris10";
	case CHIP_POLARIS11:
	case CHIP_POLARIS12: /* same as polaris11 */
		return "polaris11";
	case CHIP_VEGA10:
	case CHIP_RAVEN:
		return "gfx900";
	default:
		return "";
	}
}

static unsigned get_max_threads_per_block(struct r600_common_screen *screen,
					  enum pipe_shader_ir ir_type)
{
	if (ir_type != PIPE_SHADER_IR_TGSI)
		return 256;

	/* Only 16 waves per thread-group on gfx9. */
	if (screen->chip_class >= GFX9)
		return 1024;

	/* Up to 40 waves per thread-group on GCN < gfx9. Expose a nice
	 * round number.
	 */
	if (screen->chip_class >= SI)
		return 2048;

	return 256;
}

static int r600_get_compute_param(struct pipe_screen *screen,
        enum pipe_shader_ir ir_type,
        enum pipe_compute_cap param,
        void *ret)
{
	struct r600_common_screen *rscreen = (struct r600_common_screen *)screen;

	//TODO: select these params by asic
	switch (param) {
	case PIPE_COMPUTE_CAP_IR_TARGET: {
		const char *gpu;
		const char *triple;
		if (rscreen->family <= CHIP_ARUBA) {
			triple = "r600--";
		} else {
			if (HAVE_LLVM < 0x0400) {
				triple = "amdgcn--";
			} else {
				triple = "amdgcn-mesa-mesa3d";
			}
		}
		switch(rscreen->family) {
		/* Clang < 3.6 is missing Hainan in its list of
		 * GPUs, so we need to use the name of a similar GPU.
		 */
		default:
			gpu = r600_get_llvm_processor_name(rscreen->family);
			break;
		}
		if (ret) {
			sprintf(ret, "%s-%s", gpu, triple);
		}
		/* +2 for dash and terminating NIL byte */
		return (strlen(triple) + strlen(gpu) + 2) * sizeof(char);
	}
	case PIPE_COMPUTE_CAP_GRID_DIMENSION:
		if (ret) {
			uint64_t *grid_dimension = ret;
			grid_dimension[0] = 3;
		}
		return 1 * sizeof(uint64_t);

	case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
		if (ret) {
			uint64_t *grid_size = ret;
			grid_size[0] = 65535;
			grid_size[1] = 65535;
			grid_size[2] = 65535;
		}
		return 3 * sizeof(uint64_t) ;

	case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
		if (ret) {
			uint64_t *block_size = ret;
			unsigned threads_per_block = get_max_threads_per_block(rscreen, ir_type);
			block_size[0] = threads_per_block;
			block_size[1] = threads_per_block;
			block_size[2] = threads_per_block;
		}
		return 3 * sizeof(uint64_t);

	case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
		if (ret) {
			uint64_t *max_threads_per_block = ret;
			*max_threads_per_block = get_max_threads_per_block(rscreen, ir_type);
		}
		return sizeof(uint64_t);
	case PIPE_COMPUTE_CAP_ADDRESS_BITS:
		if (ret) {
			uint32_t *address_bits = ret;
			address_bits[0] = 32;
			if (rscreen->chip_class >= SI)
				address_bits[0] = 64;
		}
		return 1 * sizeof(uint32_t);

	case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE:
		if (ret) {
			uint64_t *max_global_size = ret;
			uint64_t max_mem_alloc_size;

			r600_get_compute_param(screen, ir_type,
				PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE,
				&max_mem_alloc_size);

			/* In OpenCL, the MAX_MEM_ALLOC_SIZE must be at least
			 * 1/4 of the MAX_GLOBAL_SIZE.  Since the
			 * MAX_MEM_ALLOC_SIZE is fixed for older kernels,
			 * make sure we never report more than
			 * 4 * MAX_MEM_ALLOC_SIZE.
			 */
			*max_global_size = MIN2(4 * max_mem_alloc_size,
						MAX2(rscreen->info.gart_size,
						     rscreen->info.vram_size));
		}
		return sizeof(uint64_t);

	case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE:
		if (ret) {
			uint64_t *max_local_size = ret;
			/* Value reported by the closed source driver. */
			*max_local_size = 32768;
		}
		return sizeof(uint64_t);

	case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE:
		if (ret) {
			uint64_t *max_input_size = ret;
			/* Value reported by the closed source driver. */
			*max_input_size = 1024;
		}
		return sizeof(uint64_t);

	case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
		if (ret) {
			uint64_t *max_mem_alloc_size = ret;

			*max_mem_alloc_size = rscreen->info.max_alloc_size;
		}
		return sizeof(uint64_t);

	case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
		if (ret) {
			uint32_t *max_clock_frequency = ret;
			*max_clock_frequency = rscreen->info.max_shader_clock;
		}
		return sizeof(uint32_t);

	case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
		if (ret) {
			uint32_t *max_compute_units = ret;
			*max_compute_units = rscreen->info.num_good_compute_units;
		}
		return sizeof(uint32_t);

	case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
		if (ret) {
			uint32_t *images_supported = ret;
			*images_supported = 0;
		}
		return sizeof(uint32_t);
	case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE:
		break; /* unused */
	case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
		if (ret) {
			uint32_t *subgroup_size = ret;
			*subgroup_size = r600_wavefront_size(rscreen->family);
		}
		return sizeof(uint32_t);
	case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
		if (ret) {
			uint64_t *max_variable_threads_per_block = ret;
			if (rscreen->chip_class >= SI &&
			    ir_type == PIPE_SHADER_IR_TGSI)
				*max_variable_threads_per_block = SI_MAX_VARIABLE_THREADS_PER_BLOCK;
			else
				*max_variable_threads_per_block = 0;
		}
		return sizeof(uint64_t);
	}

        fprintf(stderr, "unknown PIPE_COMPUTE_CAP %d\n", param);
        return 0;
}

static uint64_t r600_get_timestamp(struct pipe_screen *screen)
{
	struct r600_common_screen *rscreen = (struct r600_common_screen*)screen;

	return 1000000 * rscreen->ws->query_value(rscreen->ws, RADEON_TIMESTAMP) /
			rscreen->info.clock_crystal_freq;
}

static void r600_fence_reference(struct pipe_screen *screen,
				 struct pipe_fence_handle **dst,
				 struct pipe_fence_handle *src)
{
	struct radeon_winsys *ws = ((struct r600_common_screen*)screen)->ws;
	struct r600_multi_fence **rdst = (struct r600_multi_fence **)dst;
	struct r600_multi_fence *rsrc = (struct r600_multi_fence *)src;

	if (pipe_reference(&(*rdst)->reference, &rsrc->reference)) {
		ws->fence_reference(&(*rdst)->gfx, NULL);
		ws->fence_reference(&(*rdst)->sdma, NULL);
		FREE(*rdst);
	}
        *rdst = rsrc;
}

static boolean r600_fence_finish(struct pipe_screen *screen,
				 struct pipe_context *ctx,
				 struct pipe_fence_handle *fence,
				 uint64_t timeout)
{
	struct radeon_winsys *rws = ((struct r600_common_screen*)screen)->ws;
	struct r600_multi_fence *rfence = (struct r600_multi_fence *)fence;
	struct r600_common_context *rctx;
	int64_t abs_timeout = os_time_get_absolute_timeout(timeout);

	ctx = threaded_context_unwrap_sync(ctx);
	rctx = ctx ? (struct r600_common_context*)ctx : NULL;

	if (rfence->sdma) {
		if (!rws->fence_wait(rws, rfence->sdma, timeout))
			return false;

		/* Recompute the timeout after waiting. */
		if (timeout && timeout != PIPE_TIMEOUT_INFINITE) {
			int64_t time = os_time_get_nano();
			timeout = abs_timeout > time ? abs_timeout - time : 0;
		}
	}

	if (!rfence->gfx)
		return true;

	/* Flush the gfx IB if it hasn't been flushed yet. */
	if (rctx &&
	    rfence->gfx_unflushed.ctx == rctx &&
	    rfence->gfx_unflushed.ib_index == rctx->num_gfx_cs_flushes) {
		rctx->gfx.flush(rctx, timeout ? 0 : RADEON_FLUSH_ASYNC, NULL);
		rfence->gfx_unflushed.ctx = NULL;

		if (!timeout)
			return false;

		/* Recompute the timeout after all that. */
		if (timeout && timeout != PIPE_TIMEOUT_INFINITE) {
			int64_t time = os_time_get_nano();
			timeout = abs_timeout > time ? abs_timeout - time : 0;
		}
	}

	return rws->fence_wait(rws, rfence->gfx, timeout);
}

static void r600_query_memory_info(struct pipe_screen *screen,
				   struct pipe_memory_info *info)
{
	struct r600_common_screen *rscreen = (struct r600_common_screen*)screen;
	struct radeon_winsys *ws = rscreen->ws;
	unsigned vram_usage, gtt_usage;

	info->total_device_memory = rscreen->info.vram_size / 1024;
	info->total_staging_memory = rscreen->info.gart_size / 1024;

	/* The real TTM memory usage is somewhat random, because:
	 *
	 * 1) TTM delays freeing memory, because it can only free it after
	 *    fences expire.
	 *
	 * 2) The memory usage can be really low if big VRAM evictions are
	 *    taking place, but the real usage is well above the size of VRAM.
	 *
	 * Instead, return statistics of this process.
	 */
	vram_usage = ws->query_value(ws, RADEON_REQUESTED_VRAM_MEMORY) / 1024;
	gtt_usage =  ws->query_value(ws, RADEON_REQUESTED_GTT_MEMORY) / 1024;

	info->avail_device_memory =
		vram_usage <= info->total_device_memory ?
				info->total_device_memory - vram_usage : 0;
	info->avail_staging_memory =
		gtt_usage <= info->total_staging_memory ?
				info->total_staging_memory - gtt_usage : 0;

	info->device_memory_evicted =
		ws->query_value(ws, RADEON_NUM_BYTES_MOVED) / 1024;

	if (rscreen->info.drm_major == 3 && rscreen->info.drm_minor >= 4)
		info->nr_device_memory_evictions =
			ws->query_value(ws, RADEON_NUM_EVICTIONS);
	else
		/* Just return the number of evicted 64KB pages. */
		info->nr_device_memory_evictions = info->device_memory_evicted / 64;
}

struct pipe_resource *r600_resource_create_common(struct pipe_screen *screen,
						  const struct pipe_resource *templ)
{
	if (templ->target == PIPE_BUFFER) {
		return r600_buffer_create(screen, templ, 256);
	} else {
		return r600_texture_create(screen, templ);
	}
}

bool r600_common_screen_init(struct r600_common_screen *rscreen,
			     struct radeon_winsys *ws)
{
	char family_name[32] = {}, llvm_string[32] = {}, kernel_version[128] = {};
	struct utsname uname_data;
	const char *chip_name;

	ws->query_info(ws, &rscreen->info);
	rscreen->ws = ws;

	if ((chip_name = r600_get_marketing_name(ws)))
		snprintf(family_name, sizeof(family_name), "%s / ",
			 r600_get_family_name(rscreen) + 4);
	else
		chip_name = r600_get_family_name(rscreen);

	if (uname(&uname_data) == 0)
		snprintf(kernel_version, sizeof(kernel_version),
			 " / %s", uname_data.release);

	if (HAVE_LLVM > 0) {
		snprintf(llvm_string, sizeof(llvm_string),
			 ", LLVM %i.%i.%i", (HAVE_LLVM >> 8) & 0xff,
			 HAVE_LLVM & 0xff, MESA_LLVM_VERSION_PATCH);
	}

	snprintf(rscreen->renderer_string, sizeof(rscreen->renderer_string),
		 "%s (%sDRM %i.%i.%i%s%s)",
		 chip_name, family_name, rscreen->info.drm_major,
		 rscreen->info.drm_minor, rscreen->info.drm_patchlevel,
		 kernel_version, llvm_string);

	rscreen->b.get_name = r600_get_name;
	rscreen->b.get_vendor = r600_get_vendor;
	rscreen->b.get_device_vendor = r600_get_device_vendor;
	rscreen->b.get_disk_shader_cache = r600_get_disk_shader_cache;
	rscreen->b.get_compute_param = r600_get_compute_param;
	rscreen->b.get_paramf = r600_get_paramf;
	rscreen->b.get_timestamp = r600_get_timestamp;
	rscreen->b.fence_finish = r600_fence_finish;
	rscreen->b.fence_reference = r600_fence_reference;
	rscreen->b.resource_destroy = u_resource_destroy_vtbl;
	rscreen->b.resource_from_user_memory = r600_buffer_from_user_memory;
	rscreen->b.query_memory_info = r600_query_memory_info;

	if (rscreen->info.has_hw_decode) {
		rscreen->b.get_video_param = rvid_get_video_param;
		rscreen->b.is_video_format_supported = rvid_is_format_supported;
	} else {
		rscreen->b.get_video_param = r600_get_video_param;
		rscreen->b.is_video_format_supported = vl_video_buffer_is_format_supported;
	}

	r600_init_screen_texture_functions(rscreen);
	r600_init_screen_query_functions(rscreen);

	rscreen->family = rscreen->info.family;
	rscreen->chip_class = rscreen->info.chip_class;
	rscreen->debug_flags |= debug_get_flags_option("R600_DEBUG", common_debug_options, 0);
	rscreen->has_rbplus = false;
	rscreen->rbplus_allowed = false;

	r600_disk_cache_create(rscreen);

	slab_create_parent(&rscreen->pool_transfers, sizeof(struct r600_transfer), 64);

	rscreen->force_aniso = MIN2(16, debug_get_num_option("R600_TEX_ANISO", -1));
	if (rscreen->force_aniso >= 0) {
		printf("radeon: Forcing anisotropy filter to %ix\n",
		       /* round down to a power of two */
		       1 << util_logbase2(rscreen->force_aniso));
	}

	util_format_s3tc_init();
	(void) mtx_init(&rscreen->aux_context_lock, mtx_plain);
	(void) mtx_init(&rscreen->gpu_load_mutex, mtx_plain);

	if (rscreen->debug_flags & DBG_INFO) {
		printf("pci (domain:bus:dev.func): %04x:%02x:%02x.%x\n",
		       rscreen->info.pci_domain, rscreen->info.pci_bus,
		       rscreen->info.pci_dev, rscreen->info.pci_func);
		printf("pci_id = 0x%x\n", rscreen->info.pci_id);
		printf("family = %i (%s)\n", rscreen->info.family,
		       r600_get_family_name(rscreen));
		printf("chip_class = %i\n", rscreen->info.chip_class);
		printf("pte_fragment_size = %u\n", rscreen->info.pte_fragment_size);
		printf("gart_page_size = %u\n", rscreen->info.gart_page_size);
		printf("gart_size = %i MB\n", (int)DIV_ROUND_UP(rscreen->info.gart_size, 1024*1024));
		printf("vram_size = %i MB\n", (int)DIV_ROUND_UP(rscreen->info.vram_size, 1024*1024));
		printf("vram_vis_size = %i MB\n", (int)DIV_ROUND_UP(rscreen->info.vram_vis_size, 1024*1024));
		printf("max_alloc_size = %i MB\n",
		       (int)DIV_ROUND_UP(rscreen->info.max_alloc_size, 1024*1024));
		printf("min_alloc_size = %u\n", rscreen->info.min_alloc_size);
		printf("has_dedicated_vram = %u\n", rscreen->info.has_dedicated_vram);
		printf("has_virtual_memory = %i\n", rscreen->info.has_virtual_memory);
		printf("gfx_ib_pad_with_type2 = %i\n", rscreen->info.gfx_ib_pad_with_type2);
		printf("has_hw_decode = %u\n", rscreen->info.has_hw_decode);
		printf("num_sdma_rings = %i\n", rscreen->info.num_sdma_rings);
		printf("num_compute_rings = %u\n", rscreen->info.num_compute_rings);
		printf("uvd_fw_version = %u\n", rscreen->info.uvd_fw_version);
		printf("vce_fw_version = %u\n", rscreen->info.vce_fw_version);
		printf("me_fw_version = %i\n", rscreen->info.me_fw_version);
		printf("pfp_fw_version = %i\n", rscreen->info.pfp_fw_version);
		printf("ce_fw_version = %i\n", rscreen->info.ce_fw_version);
		printf("vce_harvest_config = %i\n", rscreen->info.vce_harvest_config);
		printf("clock_crystal_freq = %i\n", rscreen->info.clock_crystal_freq);
		printf("tcc_cache_line_size = %u\n", rscreen->info.tcc_cache_line_size);
		printf("drm = %i.%i.%i\n", rscreen->info.drm_major,
		       rscreen->info.drm_minor, rscreen->info.drm_patchlevel);
		printf("has_userptr = %i\n", rscreen->info.has_userptr);
		printf("has_syncobj = %u\n", rscreen->info.has_syncobj);

		printf("r600_max_quad_pipes = %i\n", rscreen->info.r600_max_quad_pipes);
		printf("max_shader_clock = %i\n", rscreen->info.max_shader_clock);
		printf("num_good_compute_units = %i\n", rscreen->info.num_good_compute_units);
		printf("max_se = %i\n", rscreen->info.max_se);
		printf("max_sh_per_se = %i\n", rscreen->info.max_sh_per_se);

		printf("r600_gb_backend_map = %i\n", rscreen->info.r600_gb_backend_map);
		printf("r600_gb_backend_map_valid = %i\n", rscreen->info.r600_gb_backend_map_valid);
		printf("r600_num_banks = %i\n", rscreen->info.r600_num_banks);
		printf("num_render_backends = %i\n", rscreen->info.num_render_backends);
		printf("num_tile_pipes = %i\n", rscreen->info.num_tile_pipes);
		printf("pipe_interleave_bytes = %i\n", rscreen->info.pipe_interleave_bytes);
		printf("enabled_rb_mask = 0x%x\n", rscreen->info.enabled_rb_mask);
		printf("max_alignment = %u\n", (unsigned)rscreen->info.max_alignment);
	}
	return true;
}

void r600_destroy_common_screen(struct r600_common_screen *rscreen)
{
	r600_perfcounters_destroy(rscreen);
	r600_gpu_load_kill_thread(rscreen);

	mtx_destroy(&rscreen->gpu_load_mutex);
	mtx_destroy(&rscreen->aux_context_lock);
	rscreen->aux_context->destroy(rscreen->aux_context);

	slab_destroy_parent(&rscreen->pool_transfers);

	disk_cache_destroy(rscreen->disk_shader_cache);
	rscreen->ws->destroy(rscreen->ws);
	FREE(rscreen);
}

bool r600_can_dump_shader(struct r600_common_screen *rscreen,
			  unsigned processor)
{
	return rscreen->debug_flags & (1 << processor);
}

bool r600_extra_shader_checks(struct r600_common_screen *rscreen, unsigned processor)
{
	return (rscreen->debug_flags & DBG_CHECK_IR) ||
	       r600_can_dump_shader(rscreen, processor);
}

void r600_screen_clear_buffer(struct r600_common_screen *rscreen, struct pipe_resource *dst,
			      uint64_t offset, uint64_t size, unsigned value)
{
	struct r600_common_context *rctx = (struct r600_common_context*)rscreen->aux_context;

	mtx_lock(&rscreen->aux_context_lock);
	rctx->dma_clear_buffer(&rctx->b, dst, offset, size, value);
	rscreen->aux_context->flush(rscreen->aux_context, NULL, 0);
	mtx_unlock(&rscreen->aux_context_lock);
}