summaryrefslogtreecommitdiff
path: root/src/mesa/drivers/dri/i965/brw_program.c
blob: 446aaee0673624b63dd0461c52dd9dcd417e4f4a (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
/*
 Copyright (C) Intel Corp.  2006.  All Rights Reserved.
 Intel funded Tungsten Graphics to
 develop this 3D driver.

 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 COPYRIGHT OWNER(S) 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.

 **********************************************************************/
 /*
  * Authors:
  *   Keith Whitwell <keithw@vmware.com>
  */

#include <pthread.h>
#include "main/imports.h"
#include "main/glspirv.h"
#include "program/prog_parameter.h"
#include "program/prog_print.h"
#include "program/prog_to_nir.h"
#include "program/program.h"
#include "program/programopt.h"
#include "tnl/tnl.h"
#include "util/ralloc.h"
#include "compiler/glsl/ir.h"
#include "compiler/glsl/glsl_to_nir.h"
#include "compiler/nir/nir_serialize.h"

#include "brw_program.h"
#include "brw_context.h"
#include "compiler/brw_nir.h"
#include "brw_defines.h"
#include "intel_batchbuffer.h"

static bool
brw_nir_lower_uniforms(nir_shader *nir, bool is_scalar)
{
   if (is_scalar) {
      nir_assign_var_locations(&nir->uniforms, &nir->num_uniforms,
                               type_size_scalar_bytes);
      return nir_lower_io(nir, nir_var_uniform, type_size_scalar_bytes, 0);
   } else {
      nir_assign_var_locations(&nir->uniforms, &nir->num_uniforms,
                               type_size_vec4_bytes);
      return nir_lower_io(nir, nir_var_uniform, type_size_vec4_bytes, 0);
   }
}

nir_shader *
brw_create_nir(struct brw_context *brw,
               const struct gl_shader_program *shader_prog,
               struct gl_program *prog,
               gl_shader_stage stage,
               bool is_scalar)
{
   struct gl_context *ctx = &brw->ctx;
   const nir_shader_compiler_options *options =
      ctx->Const.ShaderCompilerOptions[stage].NirOptions;
   nir_shader *nir;

   /* First, lower the GLSL/Mesa IR or SPIR-V to NIR */
   if (shader_prog) {
      if (shader_prog->_LinkedShaders[stage]->spirv_data)
         nir = _mesa_spirv_to_nir(ctx, shader_prog, stage, options);
      else
         nir = glsl_to_nir(shader_prog, stage, options);
      assert (nir);

      nir_remove_dead_variables(nir, nir_var_shader_in | nir_var_shader_out);
      nir_lower_returns(nir);
      nir_validate_shader(nir);
      NIR_PASS_V(nir, nir_lower_io_to_temporaries,
                 nir_shader_get_entrypoint(nir), true, false);
   } else {
      nir = prog_to_nir(prog, options);
      NIR_PASS_V(nir, nir_lower_regs_to_ssa); /* turn registers into SSA */
   }
   nir_validate_shader(nir);

   /* Lower PatchVerticesIn from system value to uniform. This needs to
    * happen before brw_preprocess_nir, since that will lower system values
    * to intrinsics.
    *
    * We only do this for TES if no TCS is present, since otherwise we know
    * the number of vertices in the patch at link time and we can lower it
    * directly to a constant. We do this in nir_lower_patch_vertices, which
    * needs to run after brw_nir_preprocess has turned the system values
    * into intrinsics.
    */
   const bool lower_patch_vertices_in_to_uniform =
      (stage == MESA_SHADER_TESS_CTRL && brw->screen->devinfo.gen >= 8) ||
      (stage == MESA_SHADER_TESS_EVAL &&
       !shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL]);

   if (lower_patch_vertices_in_to_uniform)
      brw_nir_lower_patch_vertices_in_to_uniform(nir);

   nir = brw_preprocess_nir(brw->screen->compiler, nir);

   if (stage == MESA_SHADER_TESS_EVAL && !lower_patch_vertices_in_to_uniform) {
      assert(shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL]);
      struct gl_linked_shader *linked_tcs =
         shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL];
      uint32_t patch_vertices = linked_tcs->Program->info.tess.tcs_vertices_out;
      nir_lower_tes_patch_vertices(nir, patch_vertices);
   }

   if (stage == MESA_SHADER_FRAGMENT) {
      static const struct nir_lower_wpos_ytransform_options wpos_options = {
         .state_tokens = {STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM, 0, 0, 0},
         .fs_coord_pixel_center_integer = 1,
         .fs_coord_origin_upper_left = 1,
      };

      bool progress = false;
      NIR_PASS(progress, nir, nir_lower_wpos_ytransform, &wpos_options);
      if (progress) {
         _mesa_add_state_reference(prog->Parameters,
                                   wpos_options.state_tokens);
      }
   }

   NIR_PASS_V(nir, brw_nir_lower_uniforms, is_scalar);

   return nir;
}

void
brw_shader_gather_info(nir_shader *nir, struct gl_program *prog)
{
   nir_shader_gather_info(nir, nir_shader_get_entrypoint(nir));

   /* Copy the info we just generated back into the gl_program */
   const char *prog_name = prog->info.name;
   const char *prog_label = prog->info.label;
   prog->info = nir->info;
   prog->info.name = prog_name;
   prog->info.label = prog_label;
}

static unsigned
get_new_program_id(struct intel_screen *screen)
{
   return p_atomic_inc_return(&screen->program_id);
}

static struct gl_program *brwNewProgram(struct gl_context *ctx, GLenum target,
                                        GLuint id, bool is_arb_asm)
{
   struct brw_context *brw = brw_context(ctx);
   struct brw_program *prog = rzalloc(NULL, struct brw_program);

   if (prog) {
      prog->id = get_new_program_id(brw->screen);

      return _mesa_init_gl_program(&prog->program, target, id, is_arb_asm);
   }

   return NULL;
}

static void brwDeleteProgram( struct gl_context *ctx,
			      struct gl_program *prog )
{
   struct brw_context *brw = brw_context(ctx);

   /* Beware!  prog's refcount has reached zero, and it's about to be freed.
    *
    * In brw_upload_pipeline_state(), we compare brw->programs[i] to
    * ctx->FooProgram._Current, and flag BRW_NEW_FOO_PROGRAM if the
    * pointer has changed.
    *
    * We cannot leave brw->programs[i] as a dangling pointer to the dead
    * program.  malloc() may allocate the same memory for a new gl_program,
    * causing us to see matching pointers...but totally different programs.
    *
    * We cannot set brw->programs[i] to NULL, either.  If we've deleted the
    * active program, Mesa may set ctx->FooProgram._Current to NULL.  That
    * would cause us to see matching pointers (NULL == NULL), and fail to
    * detect that a program has changed since our last draw.
    *
    * So, set it to a bogus gl_program pointer that will never match,
    * causing us to properly reevaluate the state on our next draw.
    *
    * Getting this wrong causes heisenbugs which are very hard to catch,
    * as you need a very specific allocation pattern to hit the problem.
    */
   static const struct gl_program deleted_program;

   for (int i = 0; i < MESA_SHADER_STAGES; i++) {
      if (brw->programs[i] == prog)
         brw->programs[i] = (struct gl_program *) &deleted_program;
   }

   _mesa_delete_program( ctx, prog );
}


static GLboolean
brwProgramStringNotify(struct gl_context *ctx,
		       GLenum target,
		       struct gl_program *prog)
{
   assert(target == GL_VERTEX_PROGRAM_ARB || !prog->arb.IsPositionInvariant);

   struct brw_context *brw = brw_context(ctx);
   const struct brw_compiler *compiler = brw->screen->compiler;

   switch (target) {
   case GL_FRAGMENT_PROGRAM_ARB: {
      struct brw_program *newFP = brw_program(prog);
      const struct brw_program *curFP =
         brw_program_const(brw->programs[MESA_SHADER_FRAGMENT]);

      if (newFP == curFP)
	 brw->ctx.NewDriverState |= BRW_NEW_FRAGMENT_PROGRAM;
      newFP->id = get_new_program_id(brw->screen);

      prog->nir = brw_create_nir(brw, NULL, prog, MESA_SHADER_FRAGMENT, true);

      brw_shader_gather_info(prog->nir, prog);

      brw_fs_precompile(ctx, prog);
      break;
   }
   case GL_VERTEX_PROGRAM_ARB: {
      struct brw_program *newVP = brw_program(prog);
      const struct brw_program *curVP =
         brw_program_const(brw->programs[MESA_SHADER_VERTEX]);

      if (newVP == curVP)
	 brw->ctx.NewDriverState |= BRW_NEW_VERTEX_PROGRAM;
      if (newVP->program.arb.IsPositionInvariant) {
	 _mesa_insert_mvp_code(ctx, &newVP->program);
      }
      newVP->id = get_new_program_id(brw->screen);

      /* Also tell tnl about it:
       */
      _tnl_program_string(ctx, target, prog);

      prog->nir = brw_create_nir(brw, NULL, prog, MESA_SHADER_VERTEX,
                                 compiler->scalar_stage[MESA_SHADER_VERTEX]);

      brw_shader_gather_info(prog->nir, prog);

      brw_vs_precompile(ctx, prog);
      break;
   }
   default:
      /*
       * driver->ProgramStringNotify is only called for ARB programs, fixed
       * function vertex programs, and ir_to_mesa (which isn't used by the
       * i965 back-end).  Therefore, even after geometry shaders are added,
       * this function should only ever be called with a target of
       * GL_VERTEX_PROGRAM_ARB or GL_FRAGMENT_PROGRAM_ARB.
       */
      unreachable("Unexpected target in brwProgramStringNotify");
   }

   return true;
}

static void
brw_memory_barrier(struct gl_context *ctx, GLbitfield barriers)
{
   struct brw_context *brw = brw_context(ctx);
   const struct gen_device_info *devinfo = &brw->screen->devinfo;
   unsigned bits = PIPE_CONTROL_DATA_CACHE_FLUSH | PIPE_CONTROL_CS_STALL;
   assert(devinfo->gen >= 7 && devinfo->gen <= 11);

   if (barriers & (GL_VERTEX_ATTRIB_ARRAY_BARRIER_BIT |
                   GL_ELEMENT_ARRAY_BARRIER_BIT |
                   GL_COMMAND_BARRIER_BIT))
      bits |= PIPE_CONTROL_VF_CACHE_INVALIDATE;

   if (barriers & GL_UNIFORM_BARRIER_BIT)
      bits |= (PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE |
               PIPE_CONTROL_CONST_CACHE_INVALIDATE);

   if (barriers & GL_TEXTURE_FETCH_BARRIER_BIT)
      bits |= PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE;

   if (barriers & (GL_TEXTURE_UPDATE_BARRIER_BIT |
                   GL_PIXEL_BUFFER_BARRIER_BIT))
      bits |= (PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE |
               PIPE_CONTROL_RENDER_TARGET_FLUSH);

   if (barriers & GL_FRAMEBUFFER_BARRIER_BIT)
      bits |= (PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE |
               PIPE_CONTROL_RENDER_TARGET_FLUSH);

   /* Typed surface messages are handled by the render cache on IVB, so we
    * need to flush it too.
    */
   if (devinfo->gen == 7 && !devinfo->is_haswell)
      bits |= PIPE_CONTROL_RENDER_TARGET_FLUSH;

   brw_emit_pipe_control_flush(brw, bits);
}

static void
brw_framebuffer_fetch_barrier(struct gl_context *ctx)
{
   struct brw_context *brw = brw_context(ctx);
   const struct gen_device_info *devinfo = &brw->screen->devinfo;

   if (!ctx->Extensions.EXT_shader_framebuffer_fetch) {
      if (devinfo->gen >= 6) {
         brw_emit_pipe_control_flush(brw,
                                     PIPE_CONTROL_RENDER_TARGET_FLUSH |
                                     PIPE_CONTROL_CS_STALL);
         brw_emit_pipe_control_flush(brw,
                                     PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE);
      } else {
         brw_emit_pipe_control_flush(brw,
                                     PIPE_CONTROL_RENDER_TARGET_FLUSH);
      }
   }
}

void
brw_get_scratch_bo(struct brw_context *brw,
		   struct brw_bo **scratch_bo, int size)
{
   struct brw_bo *old_bo = *scratch_bo;

   if (old_bo && old_bo->size < size) {
      brw_bo_unreference(old_bo);
      old_bo = NULL;
   }

   if (!old_bo) {
      *scratch_bo =
         brw_bo_alloc(brw->bufmgr, "scratch bo", size, BRW_MEMZONE_SCRATCH);
   }
}

/**
 * Reserve enough scratch space for the given stage to hold \p per_thread_size
 * bytes times the given \p thread_count.
 */
void
brw_alloc_stage_scratch(struct brw_context *brw,
                        struct brw_stage_state *stage_state,
                        unsigned per_thread_size)
{
   if (stage_state->per_thread_scratch >= per_thread_size)
      return;

   stage_state->per_thread_scratch = per_thread_size;

   if (stage_state->scratch_bo)
      brw_bo_unreference(stage_state->scratch_bo);

   const struct gen_device_info *devinfo = &brw->screen->devinfo;
   unsigned thread_count;
   switch(stage_state->stage) {
   case MESA_SHADER_VERTEX:
      thread_count = devinfo->max_vs_threads;
      break;
   case MESA_SHADER_TESS_CTRL:
      thread_count = devinfo->max_tcs_threads;
      break;
   case MESA_SHADER_TESS_EVAL:
      thread_count = devinfo->max_tes_threads;
      break;
   case MESA_SHADER_GEOMETRY:
      thread_count = devinfo->max_gs_threads;
      break;
   case MESA_SHADER_FRAGMENT:
      thread_count = devinfo->max_wm_threads;
      break;
   case MESA_SHADER_COMPUTE: {
      unsigned subslices = MAX2(brw->screen->subslice_total, 1);

      /* The documentation for 3DSTATE_PS "Scratch Space Base Pointer" says:
       *
       * "Scratch Space per slice is computed based on 4 sub-slices.  SW must
       *  allocate scratch space enough so that each slice has 4 slices
       *  allowed."
       *
       * According to the other driver team, this applies to compute shaders
       * as well.  This is not currently documented at all.
       *
       * brw->screen->subslice_total is the TOTAL number of subslices
       * and we wish to view that there are 4 subslices per slice
       * instead of the actual number of subslices per slice.
       */
      if (devinfo->gen >= 9)
         subslices = 4 * brw->screen->devinfo.num_slices;

      unsigned scratch_ids_per_subslice;
      if (devinfo->is_haswell) {
         /* WaCSScratchSize:hsw
          *
          * Haswell's scratch space address calculation appears to be sparse
          * rather than tightly packed. The Thread ID has bits indicating
          * which subslice, EU within a subslice, and thread within an EU it
          * is. There's a maximum of two slices and two subslices, so these
          * can be stored with a single bit. Even though there are only 10 EUs
          * per subslice, this is stored in 4 bits, so there's an effective
          * maximum value of 16 EUs. Similarly, although there are only 7
          * threads per EU, this is stored in a 3 bit number, giving an
          * effective maximum value of 8 threads per EU.
          *
          * This means that we need to use 16 * 8 instead of 10 * 7 for the
          * number of threads per subslice.
          */
         scratch_ids_per_subslice = 16 * 8;
      } else if (devinfo->is_cherryview) {
         /* Cherryview devices have either 6 or 8 EUs per subslice, and each
          * EU has 7 threads. The 6 EU devices appear to calculate thread IDs
          * as if it had 8 EUs.
          */
         scratch_ids_per_subslice = 8 * 7;
      } else {
         scratch_ids_per_subslice = devinfo->max_cs_threads;
      }

      thread_count = scratch_ids_per_subslice * subslices;
      break;
   }
   default:
      unreachable("Unsupported stage!");
   }

   stage_state->scratch_bo =
      brw_bo_alloc(brw->bufmgr, "shader scratch space",
                   per_thread_size * thread_count, BRW_MEMZONE_SCRATCH);
}

void brwInitFragProgFuncs( struct dd_function_table *functions )
{
   assert(functions->ProgramStringNotify == _tnl_program_string);

   functions->NewProgram = brwNewProgram;
   functions->DeleteProgram = brwDeleteProgram;
   functions->ProgramStringNotify = brwProgramStringNotify;

   functions->LinkShader = brw_link_shader;

   functions->MemoryBarrier = brw_memory_barrier;
   functions->FramebufferFetchBarrier = brw_framebuffer_fetch_barrier;
}

struct shader_times {
   uint64_t time;
   uint64_t written;
   uint64_t reset;
};

void
brw_init_shader_time(struct brw_context *brw)
{
   const int max_entries = 2048;
   brw->shader_time.bo =
      brw_bo_alloc(brw->bufmgr, "shader time",
                   max_entries * BRW_SHADER_TIME_STRIDE * 3,
                   BRW_MEMZONE_OTHER);
   brw->shader_time.names = rzalloc_array(brw, const char *, max_entries);
   brw->shader_time.ids = rzalloc_array(brw, int, max_entries);
   brw->shader_time.types = rzalloc_array(brw, enum shader_time_shader_type,
                                          max_entries);
   brw->shader_time.cumulative = rzalloc_array(brw, struct shader_times,
                                               max_entries);
   brw->shader_time.max_entries = max_entries;
}

static int
compare_time(const void *a, const void *b)
{
   uint64_t * const *a_val = a;
   uint64_t * const *b_val = b;

   /* We don't just subtract because we're turning the value to an int. */
   if (**a_val < **b_val)
      return -1;
   else if (**a_val == **b_val)
      return 0;
   else
      return 1;
}

static void
print_shader_time_line(const char *stage, const char *name,
                       int shader_num, uint64_t time, uint64_t total)
{
   fprintf(stderr, "%-6s%-18s", stage, name);

   if (shader_num != 0)
      fprintf(stderr, "%4d: ", shader_num);
   else
      fprintf(stderr, "    : ");

   fprintf(stderr, "%16lld (%7.2f Gcycles)      %4.1f%%\n",
           (long long)time,
           (double)time / 1000000000.0,
           (double)time / total * 100.0);
}

static void
brw_report_shader_time(struct brw_context *brw)
{
   if (!brw->shader_time.bo || !brw->shader_time.num_entries)
      return;

   uint64_t scaled[brw->shader_time.num_entries];
   uint64_t *sorted[brw->shader_time.num_entries];
   uint64_t total_by_type[ST_CS + 1];
   memset(total_by_type, 0, sizeof(total_by_type));
   double total = 0;
   for (int i = 0; i < brw->shader_time.num_entries; i++) {
      uint64_t written = 0, reset = 0;
      enum shader_time_shader_type type = brw->shader_time.types[i];

      sorted[i] = &scaled[i];

      switch (type) {
      case ST_VS:
      case ST_TCS:
      case ST_TES:
      case ST_GS:
      case ST_FS8:
      case ST_FS16:
      case ST_CS:
         written = brw->shader_time.cumulative[i].written;
         reset = brw->shader_time.cumulative[i].reset;
         break;

      default:
         /* I sometimes want to print things that aren't the 3 shader times.
          * Just print the sum in that case.
          */
         written = 1;
         reset = 0;
         break;
      }

      uint64_t time = brw->shader_time.cumulative[i].time;
      if (written) {
         scaled[i] = time / written * (written + reset);
      } else {
         scaled[i] = time;
      }

      switch (type) {
      case ST_VS:
      case ST_TCS:
      case ST_TES:
      case ST_GS:
      case ST_FS8:
      case ST_FS16:
      case ST_CS:
         total_by_type[type] += scaled[i];
         break;
      default:
         break;
      }

      total += scaled[i];
   }

   if (total == 0) {
      fprintf(stderr, "No shader time collected yet\n");
      return;
   }

   qsort(sorted, brw->shader_time.num_entries, sizeof(sorted[0]), compare_time);

   fprintf(stderr, "\n");
   fprintf(stderr, "type          ID                  cycles spent                   %% of total\n");
   for (int s = 0; s < brw->shader_time.num_entries; s++) {
      const char *stage;
      /* Work back from the sorted pointers times to a time to print. */
      int i = sorted[s] - scaled;

      if (scaled[i] == 0)
         continue;

      int shader_num = brw->shader_time.ids[i];
      const char *shader_name = brw->shader_time.names[i];

      switch (brw->shader_time.types[i]) {
      case ST_VS:
         stage = "vs";
         break;
      case ST_TCS:
         stage = "tcs";
         break;
      case ST_TES:
         stage = "tes";
         break;
      case ST_GS:
         stage = "gs";
         break;
      case ST_FS8:
         stage = "fs8";
         break;
      case ST_FS16:
         stage = "fs16";
         break;
      case ST_CS:
         stage = "cs";
         break;
      default:
         stage = "other";
         break;
      }

      print_shader_time_line(stage, shader_name, shader_num,
                             scaled[i], total);
   }

   fprintf(stderr, "\n");
   print_shader_time_line("total", "vs", 0, total_by_type[ST_VS], total);
   print_shader_time_line("total", "tcs", 0, total_by_type[ST_TCS], total);
   print_shader_time_line("total", "tes", 0, total_by_type[ST_TES], total);
   print_shader_time_line("total", "gs", 0, total_by_type[ST_GS], total);
   print_shader_time_line("total", "fs8", 0, total_by_type[ST_FS8], total);
   print_shader_time_line("total", "fs16", 0, total_by_type[ST_FS16], total);
   print_shader_time_line("total", "cs", 0, total_by_type[ST_CS], total);
}

static void
brw_collect_shader_time(struct brw_context *brw)
{
   if (!brw->shader_time.bo)
      return;

   /* This probably stalls on the last rendering.  We could fix that by
    * delaying reading the reports, but it doesn't look like it's a big
    * overhead compared to the cost of tracking the time in the first place.
    */
   void *bo_map = brw_bo_map(brw, brw->shader_time.bo, MAP_READ | MAP_WRITE);

   for (int i = 0; i < brw->shader_time.num_entries; i++) {
      uint32_t *times = bo_map + i * 3 * BRW_SHADER_TIME_STRIDE;

      brw->shader_time.cumulative[i].time += times[BRW_SHADER_TIME_STRIDE * 0 / 4];
      brw->shader_time.cumulative[i].written += times[BRW_SHADER_TIME_STRIDE * 1 / 4];
      brw->shader_time.cumulative[i].reset += times[BRW_SHADER_TIME_STRIDE * 2 / 4];
   }

   /* Zero the BO out to clear it out for our next collection.
    */
   memset(bo_map, 0, brw->shader_time.bo->size);
   brw_bo_unmap(brw->shader_time.bo);
}

void
brw_collect_and_report_shader_time(struct brw_context *brw)
{
   brw_collect_shader_time(brw);

   if (brw->shader_time.report_time == 0 ||
       get_time() - brw->shader_time.report_time >= 1.0) {
      brw_report_shader_time(brw);
      brw->shader_time.report_time = get_time();
   }
}

/**
 * Chooses an index in the shader_time buffer and sets up tracking information
 * for our printouts.
 *
 * Note that this holds on to references to the underlying programs, which may
 * change their lifetimes compared to normal operation.
 */
int
brw_get_shader_time_index(struct brw_context *brw, struct gl_program *prog,
                          enum shader_time_shader_type type, bool is_glsl_sh)
{
   int shader_time_index = brw->shader_time.num_entries++;
   assert(shader_time_index < brw->shader_time.max_entries);
   brw->shader_time.types[shader_time_index] = type;

   const char *name;
   if (prog->Id == 0) {
      name = "ff";
   } else if (is_glsl_sh) {
      name = prog->info.label ?
         ralloc_strdup(brw->shader_time.names, prog->info.label) : "glsl";
   } else {
      name = "prog";
   }

   brw->shader_time.names[shader_time_index] = name;
   brw->shader_time.ids[shader_time_index] = prog->Id;

   return shader_time_index;
}

void
brw_destroy_shader_time(struct brw_context *brw)
{
   brw_bo_unreference(brw->shader_time.bo);
   brw->shader_time.bo = NULL;
}

void
brw_stage_prog_data_free(const void *p)
{
   struct brw_stage_prog_data *prog_data = (struct brw_stage_prog_data *)p;

   ralloc_free(prog_data->param);
   ralloc_free(prog_data->pull_param);
}

void
brw_dump_arb_asm(const char *stage, struct gl_program *prog)
{
   fprintf(stderr, "ARB_%s_program %d ir for native %s shader\n",
           stage, prog->Id, stage);
   _mesa_print_program(prog);
}

void
brw_setup_tex_for_precompile(struct brw_context *brw,
                             struct brw_sampler_prog_key_data *tex,
                             struct gl_program *prog)
{
   const struct gen_device_info *devinfo = &brw->screen->devinfo;
   const bool has_shader_channel_select = devinfo->is_haswell || devinfo->gen >= 8;
   unsigned sampler_count = util_last_bit(prog->SamplersUsed);
   for (unsigned i = 0; i < sampler_count; i++) {
      if (!has_shader_channel_select && (prog->ShadowSamplers & (1 << i))) {
         /* Assume DEPTH_TEXTURE_MODE is the default: X, X, X, 1 */
         tex->swizzles[i] =
            MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_ONE);
      } else {
         /* Color sampler: assume no swizzling. */
         tex->swizzles[i] = SWIZZLE_XYZW;
      }
   }
}

/**
 * Sets up the starting offsets for the groups of binding table entries
 * common to all pipeline stages.
 *
 * Unused groups are initialized to 0xd0d0d0d0 to make it obvious that they're
 * unused but also make sure that addition of small offsets to them will
 * trigger some of our asserts that surface indices are < BRW_MAX_SURFACES.
 */
uint32_t
brw_assign_common_binding_table_offsets(const struct gen_device_info *devinfo,
                                        const struct gl_program *prog,
                                        struct brw_stage_prog_data *stage_prog_data,
                                        uint32_t next_binding_table_offset)
{
   int num_textures = util_last_bit(prog->SamplersUsed);

   stage_prog_data->binding_table.texture_start = next_binding_table_offset;
   next_binding_table_offset += num_textures;

   if (prog->info.num_ubos) {
      assert(prog->info.num_ubos <= BRW_MAX_UBO);
      stage_prog_data->binding_table.ubo_start = next_binding_table_offset;
      next_binding_table_offset += prog->info.num_ubos;
   } else {
      stage_prog_data->binding_table.ubo_start = 0xd0d0d0d0;
   }

   if (prog->info.num_ssbos || prog->info.num_abos) {
      assert(prog->info.num_abos <= BRW_MAX_ABO);
      assert(prog->info.num_ssbos <= BRW_MAX_SSBO);
      stage_prog_data->binding_table.ssbo_start = next_binding_table_offset;
      next_binding_table_offset += prog->info.num_abos + prog->info.num_ssbos;
   } else {
      stage_prog_data->binding_table.ssbo_start = 0xd0d0d0d0;
   }

   if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
      stage_prog_data->binding_table.shader_time_start = next_binding_table_offset;
      next_binding_table_offset++;
   } else {
      stage_prog_data->binding_table.shader_time_start = 0xd0d0d0d0;
   }

   if (prog->info.uses_texture_gather) {
      if (devinfo->gen >= 8) {
         stage_prog_data->binding_table.gather_texture_start =
            stage_prog_data->binding_table.texture_start;
      } else {
         stage_prog_data->binding_table.gather_texture_start = next_binding_table_offset;
         next_binding_table_offset += num_textures;
      }
   } else {
      stage_prog_data->binding_table.gather_texture_start = 0xd0d0d0d0;
   }

   if (prog->info.num_images) {
      stage_prog_data->binding_table.image_start = next_binding_table_offset;
      next_binding_table_offset += prog->info.num_images;
   } else {
      stage_prog_data->binding_table.image_start = 0xd0d0d0d0;
   }

   /* This may or may not be used depending on how the compile goes. */
   stage_prog_data->binding_table.pull_constants_start = next_binding_table_offset;
   next_binding_table_offset++;

   /* Plane 0 is just the regular texture section */
   stage_prog_data->binding_table.plane_start[0] = stage_prog_data->binding_table.texture_start;

   stage_prog_data->binding_table.plane_start[1] = next_binding_table_offset;
   next_binding_table_offset += num_textures;

   stage_prog_data->binding_table.plane_start[2] = next_binding_table_offset;
   next_binding_table_offset += num_textures;

   /* prog_data->base.binding_table.size will be set by brw_mark_surface_used. */

   assert(next_binding_table_offset <= BRW_MAX_SURFACES);
   return next_binding_table_offset;
}

void
brw_program_serialize_nir(struct gl_context *ctx, struct gl_program *prog)
{
   struct blob writer;
   blob_init(&writer);
   nir_serialize(&writer, prog->nir);
   prog->driver_cache_blob = ralloc_size(NULL, writer.size);
   memcpy(prog->driver_cache_blob, writer.data, writer.size);
   prog->driver_cache_blob_size = writer.size;
   blob_finish(&writer);
}

void
brw_program_deserialize_nir(struct gl_context *ctx, struct gl_program *prog,
                            gl_shader_stage stage)
{
   if (!prog->nir) {
      assert(prog->driver_cache_blob && prog->driver_cache_blob_size > 0);
      const struct nir_shader_compiler_options *options =
         ctx->Const.ShaderCompilerOptions[stage].NirOptions;
      struct blob_reader reader;
      blob_reader_init(&reader, prog->driver_cache_blob,
                       prog->driver_cache_blob_size);
      prog->nir = nir_deserialize(NULL, options, &reader);
   }

   if (prog->driver_cache_blob) {
      ralloc_free(prog->driver_cache_blob);
      prog->driver_cache_blob = NULL;
      prog->driver_cache_blob_size = 0;
   }
}