aboutsummaryrefslogtreecommitdiffstats
path: root/src/mesa/drivers/dri/i965/intel_batchbuffer.c
blob: e522e4e9c1de942086f48d753c72d57c50f4616b (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
/**************************************************************************
 *
 * Copyright 2006 VMware, Inc.
 * All Rights Reserved.
 *
 * 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, sub license, 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 NON-INFRINGEMENT.
 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS 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.
 *
 **************************************************************************/

#include "intel_batchbuffer.h"
#include "intel_buffer_objects.h"
#include "intel_reg.h"
#include "intel_bufmgr.h"
#include "intel_buffers.h"
#include "intel_fbo.h"
#include "brw_context.h"

#include <xf86drm.h>
#include <i915_drm.h>

static void
intel_batchbuffer_reset(struct brw_context *brw);

void
intel_batchbuffer_init(struct brw_context *brw)
{
   intel_batchbuffer_reset(brw);

   if (brw->gen >= 6) {
      /* We can't just use brw_state_batch to get a chunk of space for
       * the gen6 workaround because it involves actually writing to
       * the buffer, and the kernel doesn't let us write to the batch.
       */
      brw->batch.workaround_bo = drm_intel_bo_alloc(brw->bufmgr,
						      "pipe_control workaround",
						      4096, 4096);
   }

   if (!brw->has_llc) {
      brw->batch.cpu_map = malloc(BATCH_SZ);
      brw->batch.map = brw->batch.cpu_map;
   }
}

static void
intel_batchbuffer_reset(struct brw_context *brw)
{
   if (brw->batch.last_bo != NULL) {
      drm_intel_bo_unreference(brw->batch.last_bo);
      brw->batch.last_bo = NULL;
   }
   brw->batch.last_bo = brw->batch.bo;

   brw_render_cache_set_clear(brw);

   brw->batch.bo = drm_intel_bo_alloc(brw->bufmgr, "batchbuffer",
					BATCH_SZ, 4096);
   if (brw->has_llc) {
      drm_intel_bo_map(brw->batch.bo, true);
      brw->batch.map = brw->batch.bo->virtual;
   }

   brw->batch.reserved_space = BATCH_RESERVED;
   brw->batch.state_batch_offset = brw->batch.bo->size;
   brw->batch.used = 0;
   brw->batch.needs_sol_reset = false;
   brw->batch.pipe_controls_since_last_cs_stall = 0;

   /* We don't know what ring the new batch will be sent to until we see the
    * first BEGIN_BATCH or BEGIN_BATCH_BLT.  Mark it as unknown.
    */
   brw->batch.ring = UNKNOWN_RING;
}

void
intel_batchbuffer_save_state(struct brw_context *brw)
{
   brw->batch.saved.used = brw->batch.used;
   brw->batch.saved.reloc_count =
      drm_intel_gem_bo_get_reloc_count(brw->batch.bo);
}

void
intel_batchbuffer_reset_to_saved(struct brw_context *brw)
{
   drm_intel_gem_bo_clear_relocs(brw->batch.bo, brw->batch.saved.reloc_count);

   brw->batch.used = brw->batch.saved.used;
   if (brw->batch.used == 0)
      brw->batch.ring = UNKNOWN_RING;
}

void
intel_batchbuffer_free(struct brw_context *brw)
{
   free(brw->batch.cpu_map);
   drm_intel_bo_unreference(brw->batch.last_bo);
   drm_intel_bo_unreference(brw->batch.bo);
   drm_intel_bo_unreference(brw->batch.workaround_bo);
}

static void
do_batch_dump(struct brw_context *brw)
{
   struct drm_intel_decode *decode;
   struct intel_batchbuffer *batch = &brw->batch;
   int ret;

   decode = drm_intel_decode_context_alloc(brw->intelScreen->deviceID);
   if (!decode)
      return;

   ret = drm_intel_bo_map(batch->bo, false);
   if (ret == 0) {
      drm_intel_decode_set_batch_pointer(decode,
					 batch->bo->virtual,
					 batch->bo->offset64,
					 batch->used);
   } else {
      fprintf(stderr,
	      "WARNING: failed to map batchbuffer (%s), "
	      "dumping uploaded data instead.\n", strerror(ret));

      drm_intel_decode_set_batch_pointer(decode,
					 batch->map,
					 batch->bo->offset64,
					 batch->used);
   }

   drm_intel_decode_set_output_file(decode, stderr);
   drm_intel_decode(decode);

   drm_intel_decode_context_free(decode);

   if (ret == 0) {
      drm_intel_bo_unmap(batch->bo);

      brw_debug_batch(brw);
   }
}

void
intel_batchbuffer_emit_render_ring_prelude(struct brw_context *brw)
{
   /* We may need to enable and snapshot OA counters. */
   brw_perf_monitor_new_batch(brw);
}

/**
 * Called when starting a new batch buffer.
 */
static void
brw_new_batch(struct brw_context *brw)
{
   /* Create a new batchbuffer and reset the associated state: */
   drm_intel_gem_bo_clear_relocs(brw->batch.bo, 0);
   intel_batchbuffer_reset(brw);

   /* If the kernel supports hardware contexts, then most hardware state is
    * preserved between batches; we only need to re-emit state that is required
    * to be in every batch.  Otherwise we need to re-emit all the state that
    * would otherwise be stored in the context (which for all intents and
    * purposes means everything).
    */
   if (brw->hw_ctx == NULL)
      brw->ctx.NewDriverState |= BRW_NEW_CONTEXT;

   brw->ctx.NewDriverState |= BRW_NEW_BATCH;

   brw->state_batch_count = 0;

   brw->ib.type = -1;

   /* We need to periodically reap the shader time results, because rollover
    * happens every few seconds.  We also want to see results every once in a
    * while, because many programs won't cleanly destroy our context, so the
    * end-of-run printout may not happen.
    */
   if (INTEL_DEBUG & DEBUG_SHADER_TIME)
      brw_collect_and_report_shader_time(brw);

   if (INTEL_DEBUG & DEBUG_PERFMON)
      brw_dump_perf_monitors(brw);
}

/**
 * Called from intel_batchbuffer_flush before emitting MI_BATCHBUFFER_END and
 * sending it off.
 *
 * This function can emit state (say, to preserve registers that aren't saved
 * between batches).  All of this state MUST fit in the reserved space at the
 * end of the batchbuffer.  If you add more GPU state, increase the reserved
 * space by updating the BATCH_RESERVED macro.
 */
static void
brw_finish_batch(struct brw_context *brw)
{
   /* Capture the closing pipeline statistics register values necessary to
    * support query objects (in the non-hardware context world).
    */
   brw_emit_query_end(brw);

   /* We may also need to snapshot and disable OA counters. */
   if (brw->batch.ring == RENDER_RING)
      brw_perf_monitor_finish_batch(brw);

   /* Mark that the current program cache BO has been used by the GPU.
    * It will be reallocated if we need to put new programs in for the
    * next batch.
    */
   brw->cache.bo_used_by_gpu = true;
}

static void
throttle(struct brw_context *brw)
{
   /* Wait for the swapbuffers before the one we just emitted, so we
    * don't get too many swaps outstanding for apps that are GPU-heavy
    * but not CPU-heavy.
    *
    * We're using intelDRI2Flush (called from the loader before
    * swapbuffer) and glFlush (for front buffer rendering) as the
    * indicator that a frame is done and then throttle when we get
    * here as we prepare to render the next frame.  At this point for
    * round trips for swap/copy and getting new buffers are done and
    * we'll spend less time waiting on the GPU.
    *
    * Unfortunately, we don't have a handle to the batch containing
    * the swap, and getting our hands on that doesn't seem worth it,
    * so we just use the first batch we emitted after the last swap.
    */
   if (brw->need_swap_throttle && brw->throttle_batch[0]) {
      if (brw->throttle_batch[1]) {
         if (!brw->disable_throttling)
            drm_intel_bo_wait_rendering(brw->throttle_batch[1]);
         drm_intel_bo_unreference(brw->throttle_batch[1]);
      }
      brw->throttle_batch[1] = brw->throttle_batch[0];
      brw->throttle_batch[0] = NULL;
      brw->need_swap_throttle = false;
      /* Throttling here is more precise than the throttle ioctl, so skip it */
      brw->need_flush_throttle = false;
   }

   if (brw->need_flush_throttle) {
      __DRIscreen *psp = brw->intelScreen->driScrnPriv;
      drmCommandNone(psp->fd, DRM_I915_GEM_THROTTLE);
      brw->need_flush_throttle = false;
   }
}

/* TODO: Push this whole function into bufmgr.
 */
static int
do_flush_locked(struct brw_context *brw)
{
   struct intel_batchbuffer *batch = &brw->batch;
   int ret = 0;

   if (brw->has_llc) {
      drm_intel_bo_unmap(batch->bo);
   } else {
      ret = drm_intel_bo_subdata(batch->bo, 0, 4*batch->used, batch->map);
      if (ret == 0 && batch->state_batch_offset != batch->bo->size) {
	 ret = drm_intel_bo_subdata(batch->bo,
				    batch->state_batch_offset,
				    batch->bo->size - batch->state_batch_offset,
				    (char *)batch->map + batch->state_batch_offset);
      }
   }

   if (!brw->intelScreen->no_hw) {
      int flags;

      if (brw->gen >= 6 && batch->ring == BLT_RING) {
         flags = I915_EXEC_BLT;
      } else {
         flags = I915_EXEC_RENDER;
      }
      if (batch->needs_sol_reset)
	 flags |= I915_EXEC_GEN7_SOL_RESET;

      if (ret == 0) {
         if (unlikely(INTEL_DEBUG & DEBUG_AUB))
            brw_annotate_aub(brw);

	 if (brw->hw_ctx == NULL || batch->ring != RENDER_RING) {
	    ret = drm_intel_bo_mrb_exec(batch->bo, 4 * batch->used, NULL, 0, 0,
					flags);
	 } else {
	    ret = drm_intel_gem_bo_context_exec(batch->bo, brw->hw_ctx,
						4 * batch->used, flags);
	 }
      }

      throttle(brw);
   }

   if (unlikely(INTEL_DEBUG & DEBUG_BATCH))
      do_batch_dump(brw);

   if (ret != 0) {
      fprintf(stderr, "intel_do_flush_locked failed: %s\n", strerror(-ret));
      exit(1);
   }

   return ret;
}

int
_intel_batchbuffer_flush(struct brw_context *brw,
			 const char *file, int line)
{
   int ret;

   if (brw->batch.used == 0)
      return 0;

   if (brw->throttle_batch[0] == NULL) {
      brw->throttle_batch[0] = brw->batch.bo;
      drm_intel_bo_reference(brw->throttle_batch[0]);
   }

   if (unlikely(INTEL_DEBUG & DEBUG_BATCH)) {
      int bytes_for_commands = 4 * brw->batch.used;
      int bytes_for_state = brw->batch.bo->size - brw->batch.state_batch_offset;
      int total_bytes = bytes_for_commands + bytes_for_state;
      fprintf(stderr, "%s:%d: Batchbuffer flush with %4db (pkt) + "
              "%4db (state) = %4db (%0.1f%%)\n", file, line,
              bytes_for_commands, bytes_for_state,
              total_bytes,
              100.0f * total_bytes / BATCH_SZ);
   }

   brw->batch.reserved_space = 0;

   brw_finish_batch(brw);

   /* Mark the end of the buffer. */
   intel_batchbuffer_emit_dword(brw, MI_BATCH_BUFFER_END);
   if (brw->batch.used & 1) {
      /* Round batchbuffer usage to 2 DWORDs. */
      intel_batchbuffer_emit_dword(brw, MI_NOOP);
   }

   intel_upload_finish(brw);

   /* Check that we didn't just wrap our batchbuffer at a bad time. */
   assert(!brw->no_batch_wrap);

   ret = do_flush_locked(brw);

   if (unlikely(INTEL_DEBUG & DEBUG_SYNC)) {
      fprintf(stderr, "waiting for idle\n");
      drm_intel_bo_wait_rendering(brw->batch.bo);
   }

   /* Start a new batch buffer. */
   brw_new_batch(brw);

   return ret;
}


/*  This is the only way buffers get added to the validate list.
 */
bool
intel_batchbuffer_emit_reloc(struct brw_context *brw,
                             drm_intel_bo *buffer,
                             uint32_t read_domains, uint32_t write_domain,
			     uint32_t delta)
{
   int ret;

   ret = drm_intel_bo_emit_reloc(brw->batch.bo, 4*brw->batch.used,
				 buffer, delta,
				 read_domains, write_domain);
   assert(ret == 0);
   (void)ret;

   /* Using the old buffer offset, write in what the right data would be, in
    * case the buffer doesn't move and we can short-circuit the relocation
    * processing in the kernel
    */
   intel_batchbuffer_emit_dword(brw, buffer->offset64 + delta);

   return true;
}

bool
intel_batchbuffer_emit_reloc64(struct brw_context *brw,
                               drm_intel_bo *buffer,
                               uint32_t read_domains, uint32_t write_domain,
			       uint32_t delta)
{
   int ret = drm_intel_bo_emit_reloc(brw->batch.bo, 4*brw->batch.used,
                                     buffer, delta,
                                     read_domains, write_domain);
   assert(ret == 0);
   (void) ret;

   /* Using the old buffer offset, write in what the right data would be, in
    * case the buffer doesn't move and we can short-circuit the relocation
    * processing in the kernel
    */
   uint64_t offset = buffer->offset64 + delta;
   intel_batchbuffer_emit_dword(brw, offset);
   intel_batchbuffer_emit_dword(brw, offset >> 32);

   return true;
}


void
intel_batchbuffer_data(struct brw_context *brw,
                       const void *data, GLuint bytes, enum brw_gpu_ring ring)
{
   assert((bytes & 3) == 0);
   intel_batchbuffer_require_space(brw, bytes, ring);
   memcpy(brw->batch.map + brw->batch.used, data, bytes);
   brw->batch.used += bytes >> 2;
}

/**
 * According to the latest documentation, any PIPE_CONTROL with the
 * "Command Streamer Stall" bit set must also have another bit set,
 * with five different options:
 *
 *  - Render Target Cache Flush
 *  - Depth Cache Flush
 *  - Stall at Pixel Scoreboard
 *  - Post-Sync Operation
 *  - Depth Stall
 *
 * I chose "Stall at Pixel Scoreboard" since we've used it effectively
 * in the past, but the choice is fairly arbitrary.
 */
static void
gen8_add_cs_stall_workaround_bits(uint32_t *flags)
{
   uint32_t wa_bits = PIPE_CONTROL_RENDER_TARGET_FLUSH |
                      PIPE_CONTROL_DEPTH_CACHE_FLUSH |
                      PIPE_CONTROL_WRITE_IMMEDIATE |
                      PIPE_CONTROL_WRITE_DEPTH_COUNT |
                      PIPE_CONTROL_WRITE_TIMESTAMP |
                      PIPE_CONTROL_STALL_AT_SCOREBOARD |
                      PIPE_CONTROL_DEPTH_STALL;

   /* If we're doing a CS stall, and don't already have one of the
    * workaround bits set, add "Stall at Pixel Scoreboard."
    */
   if ((*flags & PIPE_CONTROL_CS_STALL) != 0 && (*flags & wa_bits) == 0)
      *flags |= PIPE_CONTROL_STALL_AT_SCOREBOARD;
}

/* Implement the WaCsStallAtEveryFourthPipecontrol workaround on IVB, BYT:
 *
 * "Every 4th PIPE_CONTROL command, not counting the PIPE_CONTROL with
 *  only read-cache-invalidate bit(s) set, must have a CS_STALL bit set."
 *
 * Note that the kernel does CS stalls between batches, so we only need
 * to count them within a batch.
 */
static uint32_t
gen7_cs_stall_every_four_pipe_controls(struct brw_context *brw, uint32_t flags)
{
   if (brw->gen == 7 && !brw->is_haswell) {
      if (flags & PIPE_CONTROL_CS_STALL) {
         /* If we're doing a CS stall, reset the counter and carry on. */
         brw->batch.pipe_controls_since_last_cs_stall = 0;
         return 0;
      }

      /* If this is the fourth pipe control without a CS stall, do one now. */
      if (++brw->batch.pipe_controls_since_last_cs_stall == 4) {
         brw->batch.pipe_controls_since_last_cs_stall = 0;
         return PIPE_CONTROL_CS_STALL;
      }
   }
   return 0;
}

/**
 * Emit a PIPE_CONTROL with various flushing flags.
 *
 * The caller is responsible for deciding what flags are appropriate for the
 * given generation.
 */
void
brw_emit_pipe_control_flush(struct brw_context *brw, uint32_t flags)
{
   if (brw->gen >= 8) {
      gen8_add_cs_stall_workaround_bits(&flags);

      BEGIN_BATCH(6);
      OUT_BATCH(_3DSTATE_PIPE_CONTROL | (6 - 2));
      OUT_BATCH(flags);
      OUT_BATCH(0);
      OUT_BATCH(0);
      OUT_BATCH(0);
      OUT_BATCH(0);
      ADVANCE_BATCH();
   } else if (brw->gen >= 6) {
      flags |= gen7_cs_stall_every_four_pipe_controls(brw, flags);

      BEGIN_BATCH(5);
      OUT_BATCH(_3DSTATE_PIPE_CONTROL | (5 - 2));
      OUT_BATCH(flags);
      OUT_BATCH(0);
      OUT_BATCH(0);
      OUT_BATCH(0);
      ADVANCE_BATCH();
   } else {
      BEGIN_BATCH(4);
      OUT_BATCH(_3DSTATE_PIPE_CONTROL | flags | (4 - 2));
      OUT_BATCH(0);
      OUT_BATCH(0);
      OUT_BATCH(0);
      ADVANCE_BATCH();
   }
}

/**
 * Emit a PIPE_CONTROL that writes to a buffer object.
 *
 * \p flags should contain one of the following items:
 *  - PIPE_CONTROL_WRITE_IMMEDIATE
 *  - PIPE_CONTROL_WRITE_TIMESTAMP
 *  - PIPE_CONTROL_WRITE_DEPTH_COUNT
 */
void
brw_emit_pipe_control_write(struct brw_context *brw, uint32_t flags,
                            drm_intel_bo *bo, uint32_t offset,
                            uint32_t imm_lower, uint32_t imm_upper)
{
   if (brw->gen >= 8) {
      gen8_add_cs_stall_workaround_bits(&flags);

      BEGIN_BATCH(6);
      OUT_BATCH(_3DSTATE_PIPE_CONTROL | (6 - 2));
      OUT_BATCH(flags);
      OUT_RELOC64(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
                  offset);
      OUT_BATCH(imm_lower);
      OUT_BATCH(imm_upper);
      ADVANCE_BATCH();
   } else if (brw->gen >= 6) {
      flags |= gen7_cs_stall_every_four_pipe_controls(brw, flags);

      /* PPGTT/GGTT is selected by DW2 bit 2 on Sandybridge, but DW1 bit 24
       * on later platforms.  We always use PPGTT on Gen7+.
       */
      unsigned gen6_gtt = brw->gen == 6 ? PIPE_CONTROL_GLOBAL_GTT_WRITE : 0;

      BEGIN_BATCH(5);
      OUT_BATCH(_3DSTATE_PIPE_CONTROL | (5 - 2));
      OUT_BATCH(flags);
      OUT_RELOC(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
                gen6_gtt | offset);
      OUT_BATCH(imm_lower);
      OUT_BATCH(imm_upper);
      ADVANCE_BATCH();
   } else {
      BEGIN_BATCH(4);
      OUT_BATCH(_3DSTATE_PIPE_CONTROL | flags | (4 - 2));
      OUT_RELOC(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
                PIPE_CONTROL_GLOBAL_GTT_WRITE | offset);
      OUT_BATCH(imm_lower);
      OUT_BATCH(imm_upper);
      ADVANCE_BATCH();
   }
}

/**
 * Restriction [DevSNB, DevIVB]:
 *
 * Prior to changing Depth/Stencil Buffer state (i.e. any combination of
 * 3DSTATE_DEPTH_BUFFER, 3DSTATE_CLEAR_PARAMS, 3DSTATE_STENCIL_BUFFER,
 * 3DSTATE_HIER_DEPTH_BUFFER) SW must first issue a pipelined depth stall
 * (PIPE_CONTROL with Depth Stall bit set), followed by a pipelined depth
 * cache flush (PIPE_CONTROL with Depth Flush Bit set), followed by
 * another pipelined depth stall (PIPE_CONTROL with Depth Stall bit set),
 * unless SW can otherwise guarantee that the pipeline from WM onwards is
 * already flushed (e.g., via a preceding MI_FLUSH).
 */
void
intel_emit_depth_stall_flushes(struct brw_context *brw)
{
   assert(brw->gen >= 6 && brw->gen <= 9);

   brw_emit_pipe_control_flush(brw, PIPE_CONTROL_DEPTH_STALL);
   brw_emit_pipe_control_flush(brw, PIPE_CONTROL_DEPTH_CACHE_FLUSH);
   brw_emit_pipe_control_flush(brw, PIPE_CONTROL_DEPTH_STALL);
}

/**
 * From the Ivybridge PRM, Volume 2 Part 1, Section 3.2 (VS Stage Input):
 * "A PIPE_CONTROL with Post-Sync Operation set to 1h and a depth
 *  stall needs to be sent just prior to any 3DSTATE_VS, 3DSTATE_URB_VS,
 *  3DSTATE_CONSTANT_VS, 3DSTATE_BINDING_TABLE_POINTER_VS,
 *  3DSTATE_SAMPLER_STATE_POINTER_VS command.  Only one PIPE_CONTROL needs
 *  to be sent before any combination of VS associated 3DSTATE."
 */
void
gen7_emit_vs_workaround_flush(struct brw_context *brw)
{
   assert(brw->gen == 7);
   brw_emit_pipe_control_write(brw,
                               PIPE_CONTROL_WRITE_IMMEDIATE
                               | PIPE_CONTROL_DEPTH_STALL,
                               brw->batch.workaround_bo, 0,
                               0, 0);
}


/**
 * Emit a PIPE_CONTROL command for gen7 with the CS Stall bit set.
 */
void
gen7_emit_cs_stall_flush(struct brw_context *brw)
{
   brw_emit_pipe_control_write(brw,
                               PIPE_CONTROL_CS_STALL
                               | PIPE_CONTROL_WRITE_IMMEDIATE,
                               brw->batch.workaround_bo, 0,
                               0, 0);
}


/**
 * Emits a PIPE_CONTROL with a non-zero post-sync operation, for
 * implementing two workarounds on gen6.  From section 1.4.7.1
 * "PIPE_CONTROL" of the Sandy Bridge PRM volume 2 part 1:
 *
 * [DevSNB-C+{W/A}] Before any depth stall flush (including those
 * produced by non-pipelined state commands), software needs to first
 * send a PIPE_CONTROL with no bits set except Post-Sync Operation !=
 * 0.
 *
 * [Dev-SNB{W/A}]: Before a PIPE_CONTROL with Write Cache Flush Enable
 * =1, a PIPE_CONTROL with any non-zero post-sync-op is required.
 *
 * And the workaround for these two requires this workaround first:
 *
 * [Dev-SNB{W/A}]: Pipe-control with CS-stall bit set must be sent
 * BEFORE the pipe-control with a post-sync op and no write-cache
 * flushes.
 *
 * And this last workaround is tricky because of the requirements on
 * that bit.  From section 1.4.7.2.3 "Stall" of the Sandy Bridge PRM
 * volume 2 part 1:
 *
 *     "1 of the following must also be set:
 *      - Render Target Cache Flush Enable ([12] of DW1)
 *      - Depth Cache Flush Enable ([0] of DW1)
 *      - Stall at Pixel Scoreboard ([1] of DW1)
 *      - Depth Stall ([13] of DW1)
 *      - Post-Sync Operation ([13] of DW1)
 *      - Notify Enable ([8] of DW1)"
 *
 * The cache flushes require the workaround flush that triggered this
 * one, so we can't use it.  Depth stall would trigger the same.
 * Post-sync nonzero is what triggered this second workaround, so we
 * can't use that one either.  Notify enable is IRQs, which aren't
 * really our business.  That leaves only stall at scoreboard.
 */
void
intel_emit_post_sync_nonzero_flush(struct brw_context *brw)
{
   brw_emit_pipe_control_flush(brw,
                               PIPE_CONTROL_CS_STALL |
                               PIPE_CONTROL_STALL_AT_SCOREBOARD);

   brw_emit_pipe_control_write(brw, PIPE_CONTROL_WRITE_IMMEDIATE,
                               brw->batch.workaround_bo, 0, 0, 0);
}

/* Emit a pipelined flush to either flush render and texture cache for
 * reading from a FBO-drawn texture, or flush so that frontbuffer
 * render appears on the screen in DRI1.
 *
 * This is also used for the always_flush_cache driconf debug option.
 */
void
intel_batchbuffer_emit_mi_flush(struct brw_context *brw)
{
   if (brw->batch.ring == BLT_RING && brw->gen >= 6) {
      BEGIN_BATCH_BLT(4);
      OUT_BATCH(MI_FLUSH_DW);
      OUT_BATCH(0);
      OUT_BATCH(0);
      OUT_BATCH(0);
      ADVANCE_BATCH();
   } else {
      int flags = PIPE_CONTROL_NO_WRITE | PIPE_CONTROL_RENDER_TARGET_FLUSH;
      if (brw->gen >= 6) {
         if (brw->gen == 9) {
            /* Hardware workaround: SKL
             *
             * Emit Pipe Control with all bits set to zero before emitting
             * a Pipe Control with VF Cache Invalidate set.
             */
            brw_emit_pipe_control_flush(brw, 0);
         }

         flags |= PIPE_CONTROL_INSTRUCTION_INVALIDATE |
                  PIPE_CONTROL_DEPTH_CACHE_FLUSH |
                  PIPE_CONTROL_VF_CACHE_INVALIDATE |
                  PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE |
                  PIPE_CONTROL_CS_STALL;

         if (brw->gen == 6) {
            /* Hardware workaround: SNB B-Spec says:
             *
             * [Dev-SNB{W/A}]: Before a PIPE_CONTROL with Write Cache
             * Flush Enable =1, a PIPE_CONTROL with any non-zero
             * post-sync-op is required.
             */
            intel_emit_post_sync_nonzero_flush(brw);
         }
      }
      brw_emit_pipe_control_flush(brw, flags);
   }

   brw_render_cache_set_clear(brw);
}

void
brw_load_register_mem(struct brw_context *brw,
                      uint32_t reg,
                      drm_intel_bo *bo,
                      uint32_t read_domains, uint32_t write_domain,
                      uint32_t offset)
{
   /* MI_LOAD_REGISTER_MEM only exists on Gen7+. */
   assert(brw->gen >= 7);

   if (brw->gen >= 8) {
      BEGIN_BATCH(4);
      OUT_BATCH(GEN7_MI_LOAD_REGISTER_MEM | (4 - 2));
      OUT_BATCH(reg);
      OUT_RELOC64(bo, read_domains, write_domain, offset);
      ADVANCE_BATCH();
   } else {
      BEGIN_BATCH(3);
      OUT_BATCH(GEN7_MI_LOAD_REGISTER_MEM | (3 - 2));
      OUT_BATCH(reg);
      OUT_RELOC(bo, read_domains, write_domain, offset);
      ADVANCE_BATCH();
   }
}