aboutsummaryrefslogtreecommitdiffstats
path: root/src/mesa/drivers/dri/i965/brw_queryobj.c
blob: ff873aea9dfd55dbe557dd590f8f70bbdedda6b9 (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
/*
 * Copyright © 2008 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 *
 * Authors:
 *    Eric Anholt <eric@anholt.net>
 *
 */

/** @file brw_queryobj.c
 *
 * Support for query objects (GL_ARB_occlusion_query, GL_ARB_timer_query,
 * GL_EXT_transform_feedback, and friends).
 *
 * The hardware provides a PIPE_CONTROL command that can report the number of
 * fragments that passed the depth test, or the hardware timer.  They are
 * appropriately synced with the stage of the pipeline for our extensions'
 * needs.
 *
 * To avoid getting samples from another context's rendering in our results,
 * we capture the counts at the start and end of every batchbuffer while the
 * query is active, and sum up the differences.  (We should do so for
 * GL_TIME_ELAPSED as well, but don't).
 */
#include "main/imports.h"

#include "brw_context.h"
#include "brw_defines.h"
#include "brw_state.h"
#include "intel_batchbuffer.h"
#include "intel_reg.h"

static void
write_timestamp(struct intel_context *intel, drm_intel_bo *query_bo, int idx)
{
   if (intel->gen >= 6) {
      /* Emit workaround flushes: */
      if (intel->gen == 6) {
         /* The timestamp write below is a non-zero post-sync op, which on
          * Gen6 necessitates a CS stall.  CS stalls need stall at scoreboard
          * set.  See the comments for intel_emit_post_sync_nonzero_flush().
          */
         BEGIN_BATCH(4);
         OUT_BATCH(_3DSTATE_PIPE_CONTROL | (4 - 2));
         OUT_BATCH(PIPE_CONTROL_CS_STALL | PIPE_CONTROL_STALL_AT_SCOREBOARD);
         OUT_BATCH(0);
         OUT_BATCH(0);
         ADVANCE_BATCH();
      }

      BEGIN_BATCH(5);
      OUT_BATCH(_3DSTATE_PIPE_CONTROL | (5 - 2));
      OUT_BATCH(PIPE_CONTROL_WRITE_TIMESTAMP);
      OUT_RELOC(query_bo,
                I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
                PIPE_CONTROL_GLOBAL_GTT_WRITE |
                idx * sizeof(uint64_t));
      OUT_BATCH(0);
      OUT_BATCH(0);
      ADVANCE_BATCH();
   } else {
      BEGIN_BATCH(4);
      OUT_BATCH(_3DSTATE_PIPE_CONTROL | (4 - 2) |
                PIPE_CONTROL_WRITE_TIMESTAMP);
      OUT_RELOC(query_bo,
                I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
                PIPE_CONTROL_GLOBAL_GTT_WRITE |
                idx * sizeof(uint64_t));
      OUT_BATCH(0);
      OUT_BATCH(0);
      ADVANCE_BATCH();
   }
}

static void
write_depth_count(struct intel_context *intel, drm_intel_bo *query_bo, int idx)
{
   if (intel->gen >= 6) {
      /* Emit Sandybridge workaround flush: */
      if (intel->gen == 6)
         intel_emit_post_sync_nonzero_flush(intel);

      BEGIN_BATCH(5);
      OUT_BATCH(_3DSTATE_PIPE_CONTROL | (5 - 2));
      OUT_BATCH(PIPE_CONTROL_DEPTH_STALL |
                PIPE_CONTROL_WRITE_DEPTH_COUNT);
      OUT_RELOC(query_bo,
                I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
                PIPE_CONTROL_GLOBAL_GTT_WRITE |
                (idx * sizeof(uint64_t)));
      OUT_BATCH(0);
      OUT_BATCH(0);
      ADVANCE_BATCH();
   } else {
      BEGIN_BATCH(4);
      OUT_BATCH(_3DSTATE_PIPE_CONTROL | (4 - 2) |
                PIPE_CONTROL_DEPTH_STALL |
                PIPE_CONTROL_WRITE_DEPTH_COUNT);
      /* This object could be mapped cacheable, but we don't have an exposed
       * mechanism to support that.  Since it's going uncached, tell GEM that
       * we're writing to it.  The usual clflush should be all that's required
       * to pick up the results.
       */
      OUT_RELOC(query_bo,
                I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
                PIPE_CONTROL_GLOBAL_GTT_WRITE |
                (idx * sizeof(uint64_t)));
      OUT_BATCH(0);
      OUT_BATCH(0);
      ADVANCE_BATCH();
   }
}

/** Waits on the query object's BO and totals the results for this query */
static void
brw_queryobj_get_results(struct gl_context *ctx,
			 struct brw_query_object *query)
{
   struct intel_context *intel = intel_context(ctx);

   int i;
   uint64_t *results;

   if (query->bo == NULL)
      return;

   if (drm_intel_bo_references(intel->batch.bo, query->bo))
      intel_batchbuffer_flush(intel);

   if (unlikely(INTEL_DEBUG & DEBUG_PERF)) {
      if (drm_intel_bo_busy(query->bo)) {
         perf_debug("Stalling on the GPU waiting for a query object.\n");
      }
   }

   drm_intel_bo_map(query->bo, false);
   results = query->bo->virtual;
   switch (query->Base.Target) {
   case GL_TIME_ELAPSED_EXT:
      if (intel->gen >= 6)
	 query->Base.Result += 80 * (results[1] - results[0]);
      else
	 query->Base.Result += 1000 * ((results[1] >> 32) - (results[0] >> 32));
      break;

   case GL_TIMESTAMP:
      if (intel->gen >= 6) {
         /* Our timer is a clock that increments every 80ns (regardless of
          * other clock scaling in the system).  The timestamp register we can
          * read for glGetTimestamp() masks out the top 32 bits, so we do that
          * here too to let the two counters be compared against each other.
          *
          * If we just multiplied that 32 bits of data by 80, it would roll
          * over at a non-power-of-two, so an application couldn't use
          * GL_QUERY_COUNTER_BITS to handle rollover correctly.  Instead, we
          * report 36 bits and truncate at that (rolling over 5 times as often
          * as the HW counter), and when the 32-bit counter rolls over, it
          * happens to also be at a rollover in the reported value from near
          * (1<<36) to 0.
          *
          * The low 32 bits rolls over in ~343 seconds.  Our 36-bit result
          * rolls over every ~69 seconds.
          */
	 query->Base.Result = 80 * (results[1] & 0xffffffff);
         query->Base.Result &= (1ull << 36) - 1;
      } else {
	 query->Base.Result = 1000 * (results[1] >> 32);
      }

      break;

   case GL_SAMPLES_PASSED_ARB:
      /* Map and count the pixels from the current query BO */
      for (i = query->first_index; i <= query->last_index; i++) {
	 query->Base.Result += results[i * 2 + 1] - results[i * 2];
      }
      break;

   case GL_ANY_SAMPLES_PASSED:
      /* Set true if any of the sub-queries passed. */
      for (i = query->first_index; i <= query->last_index; i++) {
	 if (results[i * 2 + 1] != results[i * 2]) {
            query->Base.Result = GL_TRUE;
            break;
         }
      }
      break;

   case GL_PRIMITIVES_GENERATED:
   case GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN:
      /* We don't actually query the hardware for this value, so query->bo
       * should always be NULL and execution should never reach here.
       */
      assert(!"Unreachable");
      break;

   default:
      assert(!"Unrecognized query target in brw_queryobj_get_results()");
      break;
   }
   drm_intel_bo_unmap(query->bo);

   drm_intel_bo_unreference(query->bo);
   query->bo = NULL;
}

static struct gl_query_object *
brw_new_query_object(struct gl_context *ctx, GLuint id)
{
   struct brw_query_object *query;

   query = calloc(1, sizeof(struct brw_query_object));

   query->Base.Id = id;
   query->Base.Result = 0;
   query->Base.Active = false;
   query->Base.Ready = true;

   return &query->Base;
}

static void
brw_delete_query(struct gl_context *ctx, struct gl_query_object *q)
{
   struct brw_query_object *query = (struct brw_query_object *)q;

   drm_intel_bo_unreference(query->bo);
   free(query);
}

static void
brw_begin_query(struct gl_context *ctx, struct gl_query_object *q)
{
   struct brw_context *brw = brw_context(ctx);
   struct intel_context *intel = intel_context(ctx);
   struct brw_query_object *query = (struct brw_query_object *)q;

   switch (query->Base.Target) {
   case GL_TIME_ELAPSED_EXT:
      drm_intel_bo_unreference(query->bo);
      query->bo = drm_intel_bo_alloc(intel->bufmgr, "timer query", 4096, 4096);
      write_timestamp(intel, query->bo, 0);
      break;

   case GL_ANY_SAMPLES_PASSED:
   case GL_SAMPLES_PASSED_ARB:
      /* Reset our driver's tracking of query state. */
      drm_intel_bo_unreference(query->bo);
      query->bo = NULL;
      query->first_index = -1;
      query->last_index = -1;

      brw->query.obj = query;
      intel->stats_wm++;
      break;

   case GL_PRIMITIVES_GENERATED:
      /* We don't actually query the hardware for this value; we keep track of
       * it a software counter.  So just reset the counter.
       */
      brw->sol.primitives_generated = 0;
      brw->sol.counting_primitives_generated = true;
      break;

   case GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN:
      /* We don't actually query the hardware for this value; we keep track of
       * it a software counter.  So just reset the counter.
       */
      brw->sol.primitives_written = 0;
      brw->sol.counting_primitives_written = true;
      break;

   default:
      assert(!"Unrecognized query target in brw_begin_query()");
      break;
   }
}

/**
 * Begin the ARB_occlusion_query query on a query object.
 */
static void
brw_end_query(struct gl_context *ctx, struct gl_query_object *q)
{
   struct brw_context *brw = brw_context(ctx);
   struct intel_context *intel = intel_context(ctx);
   struct brw_query_object *query = (struct brw_query_object *)q;

   switch (query->Base.Target) {
   case GL_TIMESTAMP:
      drm_intel_bo_unreference(query->bo);
      query->bo = drm_intel_bo_alloc(intel->bufmgr, "timer query",
				     4096, 4096);
      /* FALLTHROUGH */

   case GL_TIME_ELAPSED_EXT:
      write_timestamp(intel, query->bo, 1);
      break;

   case GL_ANY_SAMPLES_PASSED:
   case GL_SAMPLES_PASSED_ARB:
      if (query->bo) {
	 brw_emit_query_end(brw);

	 drm_intel_bo_unreference(brw->query.bo);
	 brw->query.bo = NULL;
      }

      brw->query.obj = NULL;

      intel->stats_wm--;
      break;

   case GL_PRIMITIVES_GENERATED:
      /* We don't actually query the hardware for this value; we keep track of
       * it in a software counter.  So just read the counter and store it in
       * the query object.
       */
      query->Base.Result = brw->sol.primitives_generated;
      brw->sol.counting_primitives_generated = false;

      /* And set brw->query.obj to NULL so that this query won't try to wait
       * for any rendering to complete.
       */
      query->bo = NULL;
      break;

   case GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN:
      /* We don't actually query the hardware for this value; we keep track of
       * it in a software counter.  So just read the counter and store it in
       * the query object.
       */
      query->Base.Result = brw->sol.primitives_written;
      brw->sol.counting_primitives_written = false;

      /* And set brw->query.obj to NULL so that this query won't try to wait
       * for any rendering to complete.
       */
      query->bo = NULL;
      break;

   default:
      assert(!"Unrecognized query target in brw_end_query()");
      break;
   }
}

static void brw_wait_query(struct gl_context *ctx, struct gl_query_object *q)
{
   struct brw_query_object *query = (struct brw_query_object *)q;

   brw_queryobj_get_results(ctx, query);
   query->Base.Ready = true;
}

static void brw_check_query(struct gl_context *ctx, struct gl_query_object *q)
{
   struct intel_context *intel = intel_context(ctx);
   struct brw_query_object *query = (struct brw_query_object *)q;

   /* From the GL_ARB_occlusion_query spec:
    *
    *     "Instead of allowing for an infinite loop, performing a
    *      QUERY_RESULT_AVAILABLE_ARB will perform a flush if the result is
    *      not ready yet on the first time it is queried.  This ensures that
    *      the async query will return true in finite time.
    */
   if (query->bo && drm_intel_bo_references(intel->batch.bo, query->bo))
      intel_batchbuffer_flush(intel);

   if (query->bo == NULL || !drm_intel_bo_busy(query->bo)) {
      brw_queryobj_get_results(ctx, query);
      query->Base.Ready = true;
   }
}

/** Called just before primitive drawing to get a beginning PS_DEPTH_COUNT. */
void
brw_emit_query_begin(struct brw_context *brw)
{
   struct intel_context *intel = &brw->intel;
   struct gl_context *ctx = &intel->ctx;
   struct brw_query_object *query = brw->query.obj;

   /* Skip if we're not doing any queries, or we've emitted the start. */
   if (!query || brw->query.begin_emitted)
      return;

   /* Get a new query BO if we're going to need it. */
   if (brw->query.bo == NULL ||
       brw->query.index * 2 + 1 >= 4096 / sizeof(uint64_t)) {
      drm_intel_bo_unreference(brw->query.bo);
      brw->query.bo = NULL;

      brw->query.bo = drm_intel_bo_alloc(intel->bufmgr, "query", 4096, 1);

      /* clear target buffer */
      drm_intel_bo_map(brw->query.bo, true);
      memset((char *)brw->query.bo->virtual, 0, 4096);
      drm_intel_bo_unmap(brw->query.bo);

      brw->query.index = 0;
   }

   write_depth_count(intel, brw->query.bo, brw->query.index * 2);

   if (query->bo != brw->query.bo) {
      if (query->bo != NULL)
	 brw_queryobj_get_results(ctx, query);
      drm_intel_bo_reference(brw->query.bo);
      query->bo = brw->query.bo;
      query->first_index = brw->query.index;
   }
   query->last_index = brw->query.index;
   brw->query.begin_emitted = true;
}

/** Called at batchbuffer flush to get an ending PS_DEPTH_COUNT */
void
brw_emit_query_end(struct brw_context *brw)
{
   struct intel_context *intel = &brw->intel;

   if (!brw->query.begin_emitted)
      return;

   write_depth_count(intel, brw->query.bo, brw->query.index * 2 + 1);

   brw->query.begin_emitted = false;
   brw->query.index++;
}

static uint64_t
brw_get_timestamp(struct gl_context *ctx)
{
   struct intel_context *intel = intel_context(ctx);
   uint64_t result = 0;

   drm_intel_reg_read(intel->bufmgr, TIMESTAMP, &result);

   /* See logic in brw_queryobj_get_results() */
   result = result >> 32;
   result *= 80;
   result &= (1ull << 36) - 1;

   return result;
}

void brw_init_queryobj_functions(struct dd_function_table *functions)
{
   functions->NewQueryObject = brw_new_query_object;
   functions->DeleteQuery = brw_delete_query;
   functions->BeginQuery = brw_begin_query;
   functions->EndQuery = brw_end_query;
   functions->CheckQuery = brw_check_query;
   functions->WaitQuery = brw_wait_query;
   functions->GetTimestamp = brw_get_timestamp;
}