aboutsummaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers/nouveau/nvc0/nve4_compute.c
blob: 1fe60267ac17246cca07c7413b237bca78239a9a (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
/*
 * Copyright 2012 Nouveau Project
 *
 * 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 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: Christoph Bumiller
 */

#include "nvc0/nvc0_context.h"
#include "nvc0/nve4_compute.h"

#include "codegen/nv50_ir_driver.h"

#ifdef DEBUG
static void nve4_compute_dump_launch_desc(const struct nve4_cp_launch_desc *);
#endif


int
nve4_screen_compute_setup(struct nvc0_screen *screen,
                          struct nouveau_pushbuf *push)
{
   struct nouveau_device *dev = screen->base.device;
   struct nouveau_object *chan = screen->base.channel;
   int i;
   int ret;
   uint32_t obj_class;
   uint64_t address;

   switch (dev->chipset & ~0xf) {
   case 0x100:
   case 0xf0:
      obj_class = NVF0_COMPUTE_CLASS; /* GK110 */
      break;
   case 0xe0:
      obj_class = NVE4_COMPUTE_CLASS; /* GK104 */
      break;
   case 0x110:
      obj_class = GM107_COMPUTE_CLASS;
      break;
   case 0x120:
      obj_class = GM200_COMPUTE_CLASS;
      break;
   default:
      NOUVEAU_ERR("unsupported chipset: NV%02x\n", dev->chipset);
      return -1;
   }

   ret = nouveau_object_new(chan, 0xbeef00c0, obj_class, NULL, 0,
                            &screen->compute);
   if (ret) {
      NOUVEAU_ERR("Failed to allocate compute object: %d\n", ret);
      return ret;
   }

   BEGIN_NVC0(push, SUBC_CP(NV01_SUBCHAN_OBJECT), 1);
   PUSH_DATA (push, screen->compute->oclass);

   BEGIN_NVC0(push, NVE4_CP(TEMP_ADDRESS_HIGH), 2);
   PUSH_DATAh(push, screen->tls->offset);
   PUSH_DATA (push, screen->tls->offset);
   /* No idea why there are 2. Divide size by 2 to be safe.
    * Actually this might be per-MP TEMP size and looks like I'm only using
    * 2 MPs instead of all 8.
    */
   BEGIN_NVC0(push, NVE4_CP(MP_TEMP_SIZE_HIGH(0)), 3);
   PUSH_DATAh(push, screen->tls->size / screen->mp_count);
   PUSH_DATA (push, (screen->tls->size / screen->mp_count) & ~0x7fff);
   PUSH_DATA (push, 0xff);
   BEGIN_NVC0(push, NVE4_CP(MP_TEMP_SIZE_HIGH(1)), 3);
   PUSH_DATAh(push, screen->tls->size / screen->mp_count);
   PUSH_DATA (push, (screen->tls->size / screen->mp_count) & ~0x7fff);
   PUSH_DATA (push, 0xff);

   /* Unified address space ? Who needs that ? Certainly not OpenCL.
    *
    * FATAL: Buffers with addresses inside [0x1000000, 0x3000000] will NOT be
    *  accessible. We cannot prevent that at the moment, so expect failure.
    */
   BEGIN_NVC0(push, NVE4_CP(LOCAL_BASE), 1);
   PUSH_DATA (push, 0xff << 24);
   BEGIN_NVC0(push, NVE4_CP(SHARED_BASE), 1);
   PUSH_DATA (push, 0xfe << 24);

   BEGIN_NVC0(push, NVE4_CP(CODE_ADDRESS_HIGH), 2);
   PUSH_DATAh(push, screen->text->offset);
   PUSH_DATA (push, screen->text->offset);

   BEGIN_NVC0(push, SUBC_CP(0x0310), 1);
   PUSH_DATA (push, (obj_class >= NVF0_COMPUTE_CLASS) ? 0x400 : 0x300);

   /* NOTE: these do not affect the state used by the 3D object */
   BEGIN_NVC0(push, NVE4_CP(TIC_ADDRESS_HIGH), 3);
   PUSH_DATAh(push, screen->txc->offset);
   PUSH_DATA (push, screen->txc->offset);
   PUSH_DATA (push, NVC0_TIC_MAX_ENTRIES - 1);
   BEGIN_NVC0(push, NVE4_CP(TSC_ADDRESS_HIGH), 3);
   PUSH_DATAh(push, screen->txc->offset + 65536);
   PUSH_DATA (push, screen->txc->offset + 65536);
   PUSH_DATA (push, NVC0_TSC_MAX_ENTRIES - 1);

   if (obj_class >= NVF0_COMPUTE_CLASS) {
      /* The blob calls GK110_COMPUTE.FIRMWARE[0x6], along with the args (0x1)
       * passed with GK110_COMPUTE.GRAPH.SCRATCH[0x2]. This is currently
       * disabled because our firmware doesn't support these commands and the
       * GPU hangs if they are used. */
      BEGIN_NIC0(push, SUBC_CP(0x0248), 64);
      for (i = 63; i >= 0; i--)
         PUSH_DATA(push, 0x38000 | i);
      IMMED_NVC0(push, SUBC_CP(NV50_GRAPH_SERIALIZE), 0);
   }

   BEGIN_NVC0(push, NVE4_CP(TEX_CB_INDEX), 1);
   PUSH_DATA (push, 7); /* does not interfere with 3D */

   /* Disabling this UNK command avoid a read fault when using texelFetch()
    * from a compute shader for weird reasons.
   if (obj_class == NVF0_COMPUTE_CLASS)
      IMMED_NVC0(push, SUBC_CP(0x02c4), 1);
   */

   address = screen->uniform_bo->offset + NVC0_CB_AUX_INFO(5);

   /* MS sample coordinate offsets: these do not work with _ALT modes ! */
   BEGIN_NVC0(push, NVE4_CP(UPLOAD_DST_ADDRESS_HIGH), 2);
   PUSH_DATAh(push, address + NVC0_CB_AUX_MS_INFO);
   PUSH_DATA (push, address + NVC0_CB_AUX_MS_INFO);
   BEGIN_NVC0(push, NVE4_CP(UPLOAD_LINE_LENGTH_IN), 2);
   PUSH_DATA (push, 64);
   PUSH_DATA (push, 1);
   BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 17);
   PUSH_DATA (push, NVE4_COMPUTE_UPLOAD_EXEC_LINEAR | (0x20 << 1));
   PUSH_DATA (push, 0); /* 0 */
   PUSH_DATA (push, 0);
   PUSH_DATA (push, 1); /* 1 */
   PUSH_DATA (push, 0);
   PUSH_DATA (push, 0); /* 2 */
   PUSH_DATA (push, 1);
   PUSH_DATA (push, 1); /* 3 */
   PUSH_DATA (push, 1);
   PUSH_DATA (push, 2); /* 4 */
   PUSH_DATA (push, 0);
   PUSH_DATA (push, 3); /* 5 */
   PUSH_DATA (push, 0);
   PUSH_DATA (push, 2); /* 6 */
   PUSH_DATA (push, 1);
   PUSH_DATA (push, 3); /* 7 */
   PUSH_DATA (push, 1);

#ifdef NOUVEAU_NVE4_MP_TRAP_HANDLER
   BEGIN_NVC0(push, NVE4_CP(UPLOAD_DST_ADDRESS_HIGH), 2);
   PUSH_DATAh(push, screen->parm->offset + NVE4_CP_INPUT_TRAP_INFO_PTR);
   PUSH_DATA (push, screen->parm->offset + NVE4_CP_INPUT_TRAP_INFO_PTR);
   BEGIN_NVC0(push, NVE4_CP(UPLOAD_LINE_LENGTH_IN), 2);
   PUSH_DATA (push, 28);
   PUSH_DATA (push, 1);
   BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 8);
   PUSH_DATA (push, 1);
   PUSH_DATA (push, screen->parm->offset + NVE4_CP_PARAM_TRAP_INFO);
   PUSH_DATAh(push, screen->parm->offset + NVE4_CP_PARAM_TRAP_INFO);
   PUSH_DATA (push, screen->tls->offset);
   PUSH_DATAh(push, screen->tls->offset);
   PUSH_DATA (push, screen->tls->size / 2); /* MP TEMP block size */
   PUSH_DATA (push, screen->tls->size / 2 / 64); /* warp TEMP block size */
   PUSH_DATA (push, 0); /* warp cfstack size */
#endif

   BEGIN_NVC0(push, NVE4_CP(FLUSH), 1);
   PUSH_DATA (push, NVE4_COMPUTE_FLUSH_CB);

   return 0;
}

static void
nve4_compute_validate_surfaces(struct nvc0_context *nvc0)
{
   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
   uint64_t address;
   const int s = 5;
   int i, j;

   if (!nvc0->images_dirty[s])
      return;

   address = nvc0->screen->uniform_bo->offset + NVC0_CB_AUX_INFO(s);

   BEGIN_NVC0(push, NVE4_CP(UPLOAD_DST_ADDRESS_HIGH), 2);
   PUSH_DATAh(push, address + NVC0_CB_AUX_SU_INFO(0));
   PUSH_DATA (push, address + NVC0_CB_AUX_SU_INFO(0));
   BEGIN_NVC0(push, NVE4_CP(UPLOAD_LINE_LENGTH_IN), 2);
   PUSH_DATA (push, 16 * NVC0_MAX_IMAGES * 4);
   PUSH_DATA (push, 0x1);
   BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 1 + 16 * NVC0_MAX_IMAGES);
   PUSH_DATA (push, NVE4_COMPUTE_UPLOAD_EXEC_LINEAR | (0x20 << 1));

   for (i = 0; i < NVC0_MAX_IMAGES; ++i) {
      struct pipe_image_view *view = &nvc0->images[s][i];
      if (view->resource) {
         struct nv04_resource *res = nv04_resource(view->resource);

         nve4_set_surface_info(push, view, nvc0);
         BCTX_REFN(nvc0->bufctx_cp, CP_SUF, res, RDWR);
      } else {
         for (j = 0; j < 16; j++)
            PUSH_DATA(push, 0);
      }
   }
}

/* Will be removed once images are completely done. */
#if 0
static void
nve4_compute_validate_surfaces(struct nvc0_context *nvc0)
{
   struct nvc0_screen *screen = nvc0->screen;
   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
   struct nv50_surface *sf;
   struct nv04_resource *res;
   uint32_t mask;
   unsigned i;
   const unsigned t = 1;
   uint64_t address;

   address = screen->uniform_bo->offset + NVC0_CB_AUX_INFO(5);

   mask = nvc0->surfaces_dirty[t];
   while (mask) {
      i = ffs(mask) - 1;
      mask &= ~(1 << i);

      /*
       * NVE4's surface load/store instructions receive all the information
       * directly instead of via binding points, so we have to supply them.
       */
      BEGIN_NVC0(push, NVE4_CP(UPLOAD_DST_ADDRESS_HIGH), 2);
      PUSH_DATAh(push, address + NVC0_CB_AUX_BUF_INFO(i));
      PUSH_DATA (push, address + NVC0_CB_AUX_BUF_INFO(i));
      BEGIN_NVC0(push, NVE4_CP(UPLOAD_LINE_LENGTH_IN), 2);
      PUSH_DATA (push, 64);
      PUSH_DATA (push, 1);
      BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 17);
      PUSH_DATA (push, NVE4_COMPUTE_UPLOAD_EXEC_LINEAR | (0x20 << 1));

      nve4_set_surface_info(push, nvc0->surfaces[t][i], screen);

      sf = nv50_surface(nvc0->surfaces[t][i]);
      if (sf) {
         res = nv04_resource(sf->base.texture);

         if (sf->base.writable)
            BCTX_REFN(nvc0->bufctx_cp, CP_SUF, res, RDWR);
         else
            BCTX_REFN(nvc0->bufctx_cp, CP_SUF, res, RD);
      }
   }
   if (nvc0->surfaces_dirty[t]) {
      BEGIN_NVC0(push, NVE4_CP(FLUSH), 1);
      PUSH_DATA (push, NVE4_COMPUTE_FLUSH_CB);
   }

   /* re-reference non-dirty surfaces */
   mask = nvc0->surfaces_valid[t] & ~nvc0->surfaces_dirty[t];
   while (mask) {
      i = ffs(mask) - 1;
      mask &= ~(1 << i);

      sf = nv50_surface(nvc0->surfaces[t][i]);
      res = nv04_resource(sf->base.texture);

      if (sf->base.writable)
         BCTX_REFN(nvc0->bufctx_cp, CP_SUF, res, RDWR);
      else
         BCTX_REFN(nvc0->bufctx_cp, CP_SUF, res, RD);
   }

   nvc0->surfaces_dirty[t] = 0;
}
#endif

/* Thankfully, textures with samplers follow the normal rules. */
static void
nve4_compute_validate_samplers(struct nvc0_context *nvc0)
{
   bool need_flush = nve4_validate_tsc(nvc0, 5);
   if (need_flush) {
      BEGIN_NVC0(nvc0->base.pushbuf, NVE4_CP(TSC_FLUSH), 1);
      PUSH_DATA (nvc0->base.pushbuf, 0);
   }
}
/* (Code duplicated at bottom for various non-convincing reasons.
 *  E.g. we might want to use the COMPUTE subchannel to upload TIC/TSC
 *  entries to avoid a subchannel switch.
 *  Same for texture cache flushes.
 *  Also, the bufctx differs, and more IFs in the 3D version looks ugly.)
 */
static void nve4_compute_validate_textures(struct nvc0_context *);

static void
nve4_compute_set_tex_handles(struct nvc0_context *nvc0)
{
   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
   struct nvc0_screen *screen = nvc0->screen;
   uint64_t address;
   const unsigned s = nvc0_shader_stage(PIPE_SHADER_COMPUTE);
   unsigned i, n;
   uint32_t dirty = nvc0->textures_dirty[s] | nvc0->samplers_dirty[s];

   if (!dirty)
      return;
   i = ffs(dirty) - 1;
   n = util_logbase2(dirty) + 1 - i;
   assert(n);

   address = screen->uniform_bo->offset + NVC0_CB_AUX_INFO(s);

   BEGIN_NVC0(push, NVE4_CP(UPLOAD_DST_ADDRESS_HIGH), 2);
   PUSH_DATAh(push, address + NVC0_CB_AUX_TEX_INFO(i));
   PUSH_DATA (push, address + NVC0_CB_AUX_TEX_INFO(i));
   BEGIN_NVC0(push, NVE4_CP(UPLOAD_LINE_LENGTH_IN), 2);
   PUSH_DATA (push, n * 4);
   PUSH_DATA (push, 0x1);
   BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 1 + n);
   PUSH_DATA (push, NVE4_COMPUTE_UPLOAD_EXEC_LINEAR | (0x20 << 1));
   PUSH_DATAp(push, &nvc0->tex_handles[s][i], n);

   BEGIN_NVC0(push, NVE4_CP(FLUSH), 1);
   PUSH_DATA (push, NVE4_COMPUTE_FLUSH_CB);

   nvc0->textures_dirty[s] = 0;
   nvc0->samplers_dirty[s] = 0;
}

static void
nve4_compute_validate_constbufs(struct nvc0_context *nvc0)
{
   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
   const int s = 5;

   while (nvc0->constbuf_dirty[s]) {
      int i = ffs(nvc0->constbuf_dirty[s]) - 1;
      nvc0->constbuf_dirty[s] &= ~(1 << i);

      if (nvc0->constbuf[s][i].user) {
         struct nouveau_bo *bo = nvc0->screen->uniform_bo;
         const unsigned base = NVC0_CB_USR_INFO(s);
         const unsigned size = nvc0->constbuf[s][0].size;
         assert(i == 0); /* we really only want OpenGL uniforms here */
         assert(nvc0->constbuf[s][0].u.data);

         BEGIN_NVC0(push, NVE4_CP(UPLOAD_DST_ADDRESS_HIGH), 2);
         PUSH_DATAh(push, bo->offset + base);
         PUSH_DATA (push, bo->offset + base);
         BEGIN_NVC0(push, NVE4_CP(UPLOAD_LINE_LENGTH_IN), 2);
         PUSH_DATA (push, size);
         PUSH_DATA (push, 0x1);
         BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 1 + (size / 4));
         PUSH_DATA (push, NVE4_COMPUTE_UPLOAD_EXEC_LINEAR | (0x20 << 1));
         PUSH_DATAp(push, nvc0->constbuf[s][0].u.data, size / 4);
      }
      else {
         struct nv04_resource *res =
            nv04_resource(nvc0->constbuf[s][i].u.buf);
         if (res) {
            uint64_t address
               = nvc0->screen->uniform_bo->offset + NVC0_CB_AUX_INFO(s);

            assert(i > 0); /* we really only want uniform buffer objects */

            BEGIN_NVC0(push, NVE4_CP(UPLOAD_DST_ADDRESS_HIGH), 2);
            PUSH_DATAh(push, address + NVC0_CB_AUX_UBO_INFO(i - 1));
            PUSH_DATA (push, address + NVC0_CB_AUX_UBO_INFO(i - 1));
            BEGIN_NVC0(push, NVE4_CP(UPLOAD_LINE_LENGTH_IN), 2);
            PUSH_DATA (push, 4 * 4);
            PUSH_DATA (push, 0x1);
            BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 1 + 4);
            PUSH_DATA (push, NVE4_COMPUTE_UPLOAD_EXEC_LINEAR | (0x20 << 1));

            PUSH_DATA (push, res->address + nvc0->constbuf[s][i].offset);
            PUSH_DATAh(push, res->address + nvc0->constbuf[s][i].offset);
            PUSH_DATA (push, nvc0->constbuf[5][i].size);
            PUSH_DATA (push, 0);
            BCTX_REFN(nvc0->bufctx_cp, CP_CB(i), res, RD);

            res->cb_bindings[s] |= 1 << i;
         }
      }
   }

   BEGIN_NVC0(push, NVE4_CP(FLUSH), 1);
   PUSH_DATA (push, NVE4_COMPUTE_FLUSH_CB);
}

static void
nve4_compute_validate_buffers(struct nvc0_context *nvc0)
{
   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
   uint64_t address;
   const int s = 5;
   int i;

   address = nvc0->screen->uniform_bo->offset + NVC0_CB_AUX_INFO(s);

   BEGIN_NVC0(push, NVE4_CP(UPLOAD_DST_ADDRESS_HIGH), 2);
   PUSH_DATAh(push, address + NVC0_CB_AUX_BUF_INFO(0));
   PUSH_DATA (push, address + NVC0_CB_AUX_BUF_INFO(0));
   BEGIN_NVC0(push, NVE4_CP(UPLOAD_LINE_LENGTH_IN), 2);
   PUSH_DATA (push, 4 * NVC0_MAX_BUFFERS * 4);
   PUSH_DATA (push, 0x1);
   BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 1 + 4 * NVC0_MAX_BUFFERS);
   PUSH_DATA (push, NVE4_COMPUTE_UPLOAD_EXEC_LINEAR | (0x20 << 1));

   for (i = 0; i < NVC0_MAX_BUFFERS; i++) {
      if (nvc0->buffers[s][i].buffer) {
         struct nv04_resource *res =
            nv04_resource(nvc0->buffers[s][i].buffer);
         PUSH_DATA (push, res->address + nvc0->buffers[s][i].buffer_offset);
         PUSH_DATAh(push, res->address + nvc0->buffers[s][i].buffer_offset);
         PUSH_DATA (push, nvc0->buffers[s][i].buffer_size);
         PUSH_DATA (push, 0);
         BCTX_REFN(nvc0->bufctx_cp, CP_BUF, res, RDWR);
      } else {
         PUSH_DATA (push, 0);
         PUSH_DATA (push, 0);
         PUSH_DATA (push, 0);
         PUSH_DATA (push, 0);
      }
   }
}

static struct nvc0_state_validate
validate_list_cp[] = {
   { nvc0_compprog_validate,              NVC0_NEW_CP_PROGRAM     },
   { nve4_compute_validate_textures,      NVC0_NEW_CP_TEXTURES    },
   { nve4_compute_validate_samplers,      NVC0_NEW_CP_SAMPLERS    },
   { nve4_compute_set_tex_handles,        NVC0_NEW_CP_TEXTURES |
                                          NVC0_NEW_CP_SAMPLERS    },
   { nve4_compute_validate_surfaces,      NVC0_NEW_CP_SURFACES    },
   { nvc0_compute_validate_globals,       NVC0_NEW_CP_GLOBALS     },
   { nve4_compute_validate_buffers,       NVC0_NEW_CP_BUFFERS     },
   { nve4_compute_validate_constbufs,     NVC0_NEW_CP_CONSTBUF    },
};

static bool
nve4_state_validate_cp(struct nvc0_context *nvc0, uint32_t mask)
{
   bool ret;

   ret = nvc0_state_validate(nvc0, mask, validate_list_cp,
                             ARRAY_SIZE(validate_list_cp), &nvc0->dirty_cp,
                             nvc0->bufctx_cp);

   if (unlikely(nvc0->state.flushed))
      nvc0_bufctx_fence(nvc0, nvc0->bufctx_cp, true);
   return ret;
}

static void
nve4_compute_upload_input(struct nvc0_context *nvc0,
                          const struct pipe_grid_info *info)
{
   struct nvc0_screen *screen = nvc0->screen;
   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
   struct nvc0_program *cp = nvc0->compprog;
   uint64_t address;

   address = screen->uniform_bo->offset + NVC0_CB_AUX_INFO(5);

   if (cp->parm_size) {
      BEGIN_NVC0(push, NVE4_CP(UPLOAD_DST_ADDRESS_HIGH), 2);
      PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_USR_INFO(5));
      PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_USR_INFO(5));
      BEGIN_NVC0(push, NVE4_CP(UPLOAD_LINE_LENGTH_IN), 2);
      PUSH_DATA (push, cp->parm_size);
      PUSH_DATA (push, 0x1);
      BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 1 + (cp->parm_size / 4));
      PUSH_DATA (push, NVE4_COMPUTE_UPLOAD_EXEC_LINEAR | (0x20 << 1));
      PUSH_DATAp(push, info->input, cp->parm_size / 4);
   }
   BEGIN_NVC0(push, NVE4_CP(UPLOAD_DST_ADDRESS_HIGH), 2);
   PUSH_DATAh(push, address + NVC0_CB_AUX_GRID_INFO);
   PUSH_DATA (push, address + NVC0_CB_AUX_GRID_INFO);
   BEGIN_NVC0(push, NVE4_CP(UPLOAD_LINE_LENGTH_IN), 2);
   PUSH_DATA (push, 7 * 4);
   PUSH_DATA (push, 0x1);

   if (unlikely(info->indirect)) {
      struct nv04_resource *res = nv04_resource(info->indirect);
      uint32_t offset = res->offset + info->indirect_offset;

      nouveau_pushbuf_space(push, 16, 0, 1);
      PUSH_REFN(push, res->bo, NOUVEAU_BO_RD | res->domain);

      BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 1 + 7);
      PUSH_DATA (push, NVE4_COMPUTE_UPLOAD_EXEC_LINEAR | (0x20 << 1));
      PUSH_DATAp(push, info->block, 3);
      nouveau_pushbuf_data(push, res->bo, offset,
                           NVC0_IB_ENTRY_1_NO_PREFETCH | 3 * 4);
   } else {
      BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 1 + 7);
      PUSH_DATA (push, NVE4_COMPUTE_UPLOAD_EXEC_LINEAR | (0x20 << 1));
      PUSH_DATAp(push, info->block, 3);
      PUSH_DATAp(push, info->grid, 3);
   }
   PUSH_DATA (push, 0);

   BEGIN_NVC0(push, NVE4_CP(FLUSH), 1);
   PUSH_DATA (push, NVE4_COMPUTE_FLUSH_CB);
}

static inline uint8_t
nve4_compute_derive_cache_split(struct nvc0_context *nvc0, uint32_t shared_size)
{
   if (shared_size > (32 << 10))
      return NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1;
   if (shared_size > (16 << 10))
      return NVE4_3D_CACHE_SPLIT_32K_SHARED_32K_L1;
   return NVC1_3D_CACHE_SPLIT_16K_SHARED_48K_L1;
}

static void
nve4_compute_setup_launch_desc(struct nvc0_context *nvc0,
                               struct nve4_cp_launch_desc *desc,
                               const struct pipe_grid_info *info)
{
   const struct nvc0_screen *screen = nvc0->screen;
   const struct nvc0_program *cp = nvc0->compprog;

   nve4_cp_launch_desc_init_default(desc);

   desc->entry = nvc0_program_symbol_offset(cp, info->pc);

   desc->griddim_x = info->grid[0];
   desc->griddim_y = info->grid[1];
   desc->griddim_z = info->grid[2];
   desc->blockdim_x = info->block[0];
   desc->blockdim_y = info->block[1];
   desc->blockdim_z = info->block[2];

   desc->shared_size = align(cp->cp.smem_size, 0x100);
   desc->local_size_p = align(cp->cp.lmem_size, 0x10);
   desc->local_size_n = 0;
   desc->cstack_size = 0x800;
   desc->cache_split = nve4_compute_derive_cache_split(nvc0, cp->cp.smem_size);

   desc->gpr_alloc = cp->num_gprs;
   desc->bar_alloc = cp->num_barriers;

   // Only bind user uniforms and the driver constant buffer through the
   // launch descriptor because UBOs are sticked to the driver cb to avoid the
   // limitation of 8 CBs.
   if (nvc0->constbuf[5][0].user || cp->parm_size) {
      nve4_cp_launch_desc_set_cb(desc, 0, screen->uniform_bo,
                                 NVC0_CB_USR_INFO(5), 1 << 16);
   }
   nve4_cp_launch_desc_set_cb(desc, 7, screen->uniform_bo,
                              NVC0_CB_AUX_INFO(5), 1 << 11);
}

static inline struct nve4_cp_launch_desc *
nve4_compute_alloc_launch_desc(struct nouveau_context *nv,
                               struct nouveau_bo **pbo, uint64_t *pgpuaddr)
{
   uint8_t *ptr = nouveau_scratch_get(nv, 512, pgpuaddr, pbo);
   if (!ptr)
      return NULL;
   if (*pgpuaddr & 255) {
      unsigned adj = 256 - (*pgpuaddr & 255);
      ptr += adj;
      *pgpuaddr += adj;
   }
   return (struct nve4_cp_launch_desc *)ptr;
}

void
nve4_launch_grid(struct pipe_context *pipe, const struct pipe_grid_info *info)
{
   struct nvc0_context *nvc0 = nvc0_context(pipe);
   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
   struct nve4_cp_launch_desc *desc;
   uint64_t desc_gpuaddr;
   struct nouveau_bo *desc_bo;
   int ret;

   desc = nve4_compute_alloc_launch_desc(&nvc0->base, &desc_bo, &desc_gpuaddr);
   if (!desc) {
      ret = -1;
      goto out;
   }
   BCTX_REFN_bo(nvc0->bufctx_cp, CP_DESC, NOUVEAU_BO_GART | NOUVEAU_BO_RD,
                desc_bo);

   ret = !nve4_state_validate_cp(nvc0, ~0);
   if (ret)
      goto out;

   nve4_compute_setup_launch_desc(nvc0, desc, info);

   nve4_compute_upload_input(nvc0, info);

#ifdef DEBUG
   if (debug_get_num_option("NV50_PROG_DEBUG", 0))
      nve4_compute_dump_launch_desc(desc);
#endif

   if (unlikely(info->indirect)) {
      struct nv04_resource *res = nv04_resource(info->indirect);
      uint32_t offset = res->offset + info->indirect_offset;

      /* upload the descriptor */
      BEGIN_NVC0(push, NVE4_CP(UPLOAD_DST_ADDRESS_HIGH), 2);
      PUSH_DATAh(push, desc_gpuaddr);
      PUSH_DATA (push, desc_gpuaddr);
      BEGIN_NVC0(push, NVE4_CP(UPLOAD_LINE_LENGTH_IN), 2);
      PUSH_DATA (push, 256);
      PUSH_DATA (push, 1);
      BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 1 + (256 / 4));
      PUSH_DATA (push, NVE4_COMPUTE_UPLOAD_EXEC_LINEAR | (0x08 << 1));
      PUSH_DATAp(push, (const uint32_t *)desc, 256 / 4);

      /* overwrite griddim_x and griddim_y as two 32-bits integers even
       * if griddim_y must be a 16-bits integer */
      BEGIN_NVC0(push, NVE4_CP(UPLOAD_DST_ADDRESS_HIGH), 2);
      PUSH_DATAh(push, desc_gpuaddr + 48);
      PUSH_DATA (push, desc_gpuaddr + 48);
      BEGIN_NVC0(push, NVE4_CP(UPLOAD_LINE_LENGTH_IN), 2);
      PUSH_DATA (push, 8);
      PUSH_DATA (push, 1);

      nouveau_pushbuf_space(push, 16, 0, 1);
      PUSH_REFN(push, res->bo, NOUVEAU_BO_RD | res->domain);

      BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 1 + (8 / 4));
      PUSH_DATA (push, NVE4_COMPUTE_UPLOAD_EXEC_LINEAR | (0x08 << 1));
      nouveau_pushbuf_data(push, res->bo, offset,
                           NVC0_IB_ENTRY_1_NO_PREFETCH | 2 * 4);

      /* overwrite the 16 high bits of griddim_y with griddim_z because
       * we need (z << 16) | x */
      BEGIN_NVC0(push, NVE4_CP(UPLOAD_DST_ADDRESS_HIGH), 2);
      PUSH_DATAh(push, desc_gpuaddr + 54);
      PUSH_DATA (push, desc_gpuaddr + 54);
      BEGIN_NVC0(push, NVE4_CP(UPLOAD_LINE_LENGTH_IN), 2);
      PUSH_DATA (push, 4);
      PUSH_DATA (push, 1);
      BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 1 + (4 / 4));
      PUSH_DATA (push, NVE4_COMPUTE_UPLOAD_EXEC_LINEAR | (0x08 << 1));
      nouveau_pushbuf_data(push, res->bo, offset + 8,
                           NVC0_IB_ENTRY_1_NO_PREFETCH | 1 * 4);
   }

   /* upload descriptor and flush */
   BEGIN_NVC0(push, NVE4_CP(LAUNCH_DESC_ADDRESS), 1);
   PUSH_DATA (push, desc_gpuaddr >> 8);
   BEGIN_NVC0(push, NVE4_CP(LAUNCH), 1);
   PUSH_DATA (push, 0x3);
   BEGIN_NVC0(push, SUBC_CP(NV50_GRAPH_SERIALIZE), 1);
   PUSH_DATA (push, 0);

out:
   if (ret)
      NOUVEAU_ERR("Failed to launch grid !\n");
   nouveau_scratch_done(&nvc0->base);
   nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_DESC);
}


#define NVE4_TIC_ENTRY_INVALID 0x000fffff

static void
nve4_compute_validate_textures(struct nvc0_context *nvc0)
{
   struct nouveau_bo *txc = nvc0->screen->txc;
   struct nouveau_pushbuf *push = nvc0->base.pushbuf;
   const unsigned s = 5;
   unsigned i;
   uint32_t commands[2][32];
   unsigned n[2] = { 0, 0 };

   for (i = 0; i < nvc0->num_textures[s]; ++i) {
      struct nv50_tic_entry *tic = nv50_tic_entry(nvc0->textures[s][i]);
      struct nv04_resource *res;
      const bool dirty = !!(nvc0->textures_dirty[s] & (1 << i));

      if (!tic) {
         nvc0->tex_handles[s][i] |= NVE4_TIC_ENTRY_INVALID;
         continue;
      }
      res = nv04_resource(tic->pipe.texture);

      if (tic->id < 0) {
         tic->id = nvc0_screen_tic_alloc(nvc0->screen, tic);

         PUSH_SPACE(push, 16);
         BEGIN_NVC0(push, NVE4_CP(UPLOAD_DST_ADDRESS_HIGH), 2);
         PUSH_DATAh(push, txc->offset + (tic->id * 32));
         PUSH_DATA (push, txc->offset + (tic->id * 32));
         BEGIN_NVC0(push, NVE4_CP(UPLOAD_LINE_LENGTH_IN), 2);
         PUSH_DATA (push, 32);
         PUSH_DATA (push, 1);
         BEGIN_1IC0(push, NVE4_CP(UPLOAD_EXEC), 9);
         PUSH_DATA (push, NVE4_COMPUTE_UPLOAD_EXEC_LINEAR | (0x20 << 1));
         PUSH_DATAp(push, &tic->tic[0], 8);

         commands[0][n[0]++] = (tic->id << 4) | 1;
      } else
      if (res->status & NOUVEAU_BUFFER_STATUS_GPU_WRITING) {
         commands[1][n[1]++] = (tic->id << 4) | 1;
      }
      nvc0->screen->tic.lock[tic->id / 32] |= 1 << (tic->id % 32);

      res->status &= ~NOUVEAU_BUFFER_STATUS_GPU_WRITING;
      res->status |=  NOUVEAU_BUFFER_STATUS_GPU_READING;

      nvc0->tex_handles[s][i] &= ~NVE4_TIC_ENTRY_INVALID;
      nvc0->tex_handles[s][i] |= tic->id;
      if (dirty)
         BCTX_REFN(nvc0->bufctx_cp, CP_TEX(i), res, RD);
   }
   for (; i < nvc0->state.num_textures[s]; ++i)
      nvc0->tex_handles[s][i] |= NVE4_TIC_ENTRY_INVALID;

   if (n[0]) {
      BEGIN_NIC0(push, NVE4_CP(TIC_FLUSH), n[0]);
      PUSH_DATAp(push, commands[0], n[0]);
   }
   if (n[1]) {
      BEGIN_NIC0(push, NVE4_CP(TEX_CACHE_CTL), n[1]);
      PUSH_DATAp(push, commands[1], n[1]);
   }

   nvc0->state.num_textures[s] = nvc0->num_textures[s];
}


#ifdef DEBUG
static const char *nve4_cache_split_name(unsigned value)
{
   switch (value) {
   case NVC1_3D_CACHE_SPLIT_16K_SHARED_48K_L1: return "16K_SHARED_48K_L1";
   case NVE4_3D_CACHE_SPLIT_32K_SHARED_32K_L1: return "32K_SHARED_32K_L1";
   case NVC0_3D_CACHE_SPLIT_48K_SHARED_16K_L1: return "48K_SHARED_16K_L1";
   default:
      return "(invalid)";
   }
}

static void
nve4_compute_dump_launch_desc(const struct nve4_cp_launch_desc *desc)
{
   const uint32_t *data = (const uint32_t *)desc;
   unsigned i;
   bool zero = false;

   debug_printf("COMPUTE LAUNCH DESCRIPTOR:\n");

   for (i = 0; i < sizeof(*desc); i += 4) {
      if (data[i / 4]) {
         debug_printf("[%x]: 0x%08x\n", i, data[i / 4]);
         zero = false;
      } else
      if (!zero) {
         debug_printf("...\n");
         zero = true;
      }
   }

   debug_printf("entry = 0x%x\n", desc->entry);
   debug_printf("grid dimensions = %ux%ux%u\n",
                desc->griddim_x, desc->griddim_y, desc->griddim_z);
   debug_printf("block dimensions = %ux%ux%u\n",
                desc->blockdim_x, desc->blockdim_y, desc->blockdim_z);
   debug_printf("s[] size: 0x%x\n", desc->shared_size);
   debug_printf("l[] size: -0x%x / +0x%x\n",
                desc->local_size_n, desc->local_size_p);
   debug_printf("stack size: 0x%x\n", desc->cstack_size);
   debug_printf("barrier count: %u\n", desc->bar_alloc);
   debug_printf("$r count: %u\n", desc->gpr_alloc);
   debug_printf("cache split: %s\n", nve4_cache_split_name(desc->cache_split));

   for (i = 0; i < 8; ++i) {
      uint64_t address;
      uint32_t size = desc->cb[i].size;
      bool valid = !!(desc->cb_mask & (1 << i));

      address = ((uint64_t)desc->cb[i].address_h << 32) | desc->cb[i].address_l;

      if (!valid && !address && !size)
         continue;
      debug_printf("CB[%u]: address = 0x%"PRIx64", size 0x%x%s\n",
                   i, address, size, valid ? "" : "  (invalid)");
   }
}
#endif

#ifdef NOUVEAU_NVE4_MP_TRAP_HANDLER
static void
nve4_compute_trap_info(struct nvc0_context *nvc0)
{
   struct nvc0_screen *screen = nvc0->screen;
   struct nouveau_bo *bo = screen->parm;
   int ret, i;
   volatile struct nve4_mp_trap_info *info;
   uint8_t *map;

   ret = nouveau_bo_map(bo, NOUVEAU_BO_RDWR, nvc0->base.client);
   if (ret)
      return;
   map = (uint8_t *)bo->map;
   info = (volatile struct nve4_mp_trap_info *)(map + NVE4_CP_PARAM_TRAP_INFO);

   if (info->lock) {
      debug_printf("trapstat = %08x\n", info->trapstat);
      debug_printf("warperr = %08x\n", info->warperr);
      debug_printf("PC = %x\n", info->pc);
      debug_printf("tid = %u %u %u\n",
                   info->tid[0], info->tid[1], info->tid[2]);
      debug_printf("ctaid = %u %u %u\n",
                   info->ctaid[0], info->ctaid[1], info->ctaid[2]);
      for (i = 0; i <= 63; ++i)
         debug_printf("$r%i = %08x\n", i, info->r[i]);
      for (i = 0; i <= 6; ++i)
         debug_printf("$p%i = %i\n", i, (info->flags >> i) & 1);
      debug_printf("$c = %x\n", info->flags >> 12);
   }
   info->lock = 0;
}
#endif