aboutsummaryrefslogtreecommitdiffstats
path: root/src/mesa/drivers/dri/i965/brw_program.c
blob: fa59338950a55fa62da3f9b7500dd7c8054b4070 (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
/*
 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/enums.h"
#include "main/shaderobj.h"
#include "program/prog_parameter.h"
#include "program/prog_print.h"
#include "program/program.h"
#include "program/programopt.h"
#include "tnl/tnl.h"
#include "util/ralloc.h"
#include "glsl/ir.h"

#include "brw_context.h"
#include "brw_shader.h"
#include "brw_nir.h"
#include "brw_wm.h"
#include "intel_batchbuffer.h"

static unsigned
get_new_program_id(struct intel_screen *screen)
{
   static pthread_mutex_t m = PTHREAD_MUTEX_INITIALIZER;
   pthread_mutex_lock(&m);
   unsigned id = screen->program_id++;
   pthread_mutex_unlock(&m);
   return id;
}

static struct gl_program *brwNewProgram( struct gl_context *ctx,
				      GLenum target,
				      GLuint id )
{
   struct brw_context *brw = brw_context(ctx);

   switch (target) {
   case GL_VERTEX_PROGRAM_ARB: {
      struct brw_vertex_program *prog = CALLOC_STRUCT(brw_vertex_program);
      if (prog) {
	 prog->id = get_new_program_id(brw->intelScreen);

	 return _mesa_init_vertex_program( ctx, &prog->program,
					     target, id );
      }
      else
	 return NULL;
   }

   case GL_FRAGMENT_PROGRAM_ARB: {
      struct brw_fragment_program *prog = CALLOC_STRUCT(brw_fragment_program);
      if (prog) {
	 prog->id = get_new_program_id(brw->intelScreen);

	 return _mesa_init_fragment_program( ctx, &prog->program,
					     target, id );
      }
      else
	 return NULL;
   }

   case GL_GEOMETRY_PROGRAM_NV: {
      struct brw_geometry_program *prog = CALLOC_STRUCT(brw_geometry_program);
      if (prog) {
         prog->id = get_new_program_id(brw->intelScreen);

         return _mesa_init_geometry_program(ctx, &prog->program, target, id);
      } else {
         return NULL;
      }
   }

   case GL_COMPUTE_PROGRAM_NV: {
      struct brw_compute_program *prog = CALLOC_STRUCT(brw_compute_program);
      if (prog) {
         prog->id = get_new_program_id(brw->intelScreen);

         return _mesa_init_compute_program(ctx, &prog->program, target, id);
      } else {
         return NULL;
      }
   }

   default:
      unreachable("Unsupported target in brwNewProgram()");
   }
}

static void brwDeleteProgram( struct gl_context *ctx,
			      struct gl_program *prog )
{
   _mesa_delete_program( ctx, prog );
}


static GLboolean
brwProgramStringNotify(struct gl_context *ctx,
		       GLenum target,
		       struct gl_program *prog)
{
   struct brw_context *brw = brw_context(ctx);

   switch (target) {
   case GL_FRAGMENT_PROGRAM_ARB: {
      struct gl_fragment_program *fprog = (struct gl_fragment_program *) prog;
      struct brw_fragment_program *newFP = brw_fragment_program(fprog);
      const struct brw_fragment_program *curFP =
         brw_fragment_program_const(brw->fragment_program);

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

      brw_add_texrect_params(prog);

      if (ctx->Const.ShaderCompilerOptions[MESA_SHADER_FRAGMENT].NirOptions) {
         prog->nir = brw_create_nir(brw, NULL, prog, MESA_SHADER_FRAGMENT, true);
      }

      brw_fs_precompile(ctx, NULL, prog);
      break;
   }
   case GL_VERTEX_PROGRAM_ARB: {
      struct gl_vertex_program *vprog = (struct gl_vertex_program *) prog;
      struct brw_vertex_program *newVP = brw_vertex_program(vprog);
      const struct brw_vertex_program *curVP =
         brw_vertex_program_const(brw->vertex_program);

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

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

      brw_add_texrect_params(prog);

      if (ctx->Const.ShaderCompilerOptions[MESA_SHADER_VERTEX].NirOptions) {
         prog->nir = brw_create_nir(brw, NULL, prog, MESA_SHADER_VERTEX,
                                    brw->intelScreen->compiler->scalar_vs);
      }

      brw_vs_precompile(ctx, NULL, 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);
   unsigned bits = (PIPE_CONTROL_DATA_CACHE_INVALIDATE |
                    PIPE_CONTROL_NO_WRITE |
                    PIPE_CONTROL_CS_STALL);
   assert(brw->gen >= 7 && brw->gen <= 9);

   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)
      bits |= PIPE_CONTROL_RENDER_TARGET_FLUSH;

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

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

   brw_emit_pipe_control_flush(brw, bits);
}

void
brw_add_texrect_params(struct gl_program *prog)
{
   for (int texunit = 0; texunit < BRW_MAX_TEX_UNIT; texunit++) {
      if (!(prog->TexturesUsed[texunit] & (1 << TEXTURE_RECT_INDEX)))
         continue;

      int tokens[STATE_LENGTH] = {
         STATE_INTERNAL,
         STATE_TEXRECT_SCALE,
         texunit,
         0,
         0
      };

      _mesa_add_state_reference(prog->Parameters, (gl_state_index *)tokens);
   }
}

/* Per-thread scratch space is a power-of-two multiple of 1KB. */
int
brw_get_scratch_size(int size)
{
   int i;

   for (i = 1024; i < size; i *= 2)
      ;

   return i;
}

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

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

   if (!old_bo) {
      *scratch_bo = drm_intel_bo_alloc(brw->bufmgr, "scratch bo", size, 4096);
   }
}

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

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

   functions->NewShader = brw_new_shader;
   functions->LinkShader = brw_link_shader;

   functions->MemoryBarrier = brw_memory_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 =
      drm_intel_bo_alloc(brw->bufmgr, "shader time",
                         max_entries * SHADER_TIME_STRIDE * 3, 4096);
   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_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_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_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", "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.
    */
   drm_intel_bo_map(brw->shader_time.bo, true);
   void *bo_map = brw->shader_time.bo->virtual;

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

      brw->shader_time.cumulative[i].time += times[SHADER_TIME_STRIDE * 0 / 4];
      brw->shader_time.cumulative[i].written += times[SHADER_TIME_STRIDE * 1 / 4];
      brw->shader_time.cumulative[i].reset += times[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);
   drm_intel_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_shader_program *shader_prog,
                          struct gl_program *prog,
                          enum shader_time_shader_type type)
{
   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;

   int id = shader_prog ? shader_prog->Name : prog->Id;
   const char *name;
   if (id == 0) {
      name = "ff";
   } else if (!shader_prog) {
      name = "prog";
   } else if (shader_prog->Label) {
      name = ralloc_strdup(brw->shader_time.names, shader_prog->Label);
   } else {
      name = "glsl";
   }

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

   return shader_time_index;
}

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

void
brw_mark_surface_used(struct brw_stage_prog_data *prog_data,
                      unsigned surf_index)
{
   assert(surf_index < BRW_MAX_SURFACES);

   prog_data->binding_table.size_bytes =
      MAX2(prog_data->binding_table.size_bytes, (surf_index + 1) * 4);
}

bool
brw_stage_prog_data_compare(const struct brw_stage_prog_data *a,
                            const struct brw_stage_prog_data *b)
{
   /* Compare all the struct up to the pointers. */
   if (memcmp(a, b, offsetof(struct brw_stage_prog_data, param)))
      return false;

   if (memcmp(a->param, b->param, a->nr_params * sizeof(void *)))
      return false;

   if (memcmp(a->pull_param, b->pull_param, a->nr_pull_params * sizeof(void *)))
      return false;

   return true;
}

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_ir(const char *stage, struct gl_shader_program *shader_prog,
            struct gl_shader *shader, struct gl_program *prog)
{
   if (shader_prog) {
      if (shader->ir) {
         fprintf(stderr,
                 "GLSL IR for native %s shader %d:\n",
                 stage, shader_prog->Name);
         _mesa_print_ir(stderr, shader->ir, NULL);
         fprintf(stderr, "\n\n");
      }
   } else {
      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 bool has_shader_channel_select = brw->is_haswell || brw->gen >= 8;
   unsigned sampler_count = _mesa_fls(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;
      }
   }
}