summaryrefslogtreecommitdiff
path: root/src/mesa/drivers/dri/i965/gen6_queryobj.c
blob: 498b1877b6d5abf8952d1fa9ba6c0d3ddc387d17 (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
/*
 * 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>
 *    Kenneth Graunke <kenneth@whitecape.org>
 */

/** @file gen6_queryobj.c
 *
 * Support for query objects (GL_ARB_occlusion_query, GL_ARB_timer_query,
 * GL_EXT_transform_feedback, and friends) on platforms that support
 * hardware contexts (Gen6+).
 */
#include "main/imports.h"

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

/**
 * Emit PIPE_CONTROLs to write the current GPU timestamp into a buffer.
 */
static void
write_timestamp(struct brw_context *brw, drm_intel_bo *query_bo, int idx)
{
   /* Emit workaround flushes: */
   if (brw->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();
}

/**
 * Emit PIPE_CONTROLs to write the PS_DEPTH_COUNT register into a buffer.
 */
static void
write_depth_count(struct brw_context *brw, drm_intel_bo *query_bo, int idx)
{
   /* Emit Sandybridge workaround flush: */
   if (brw->gen == 6)
      intel_emit_post_sync_nonzero_flush(brw);

   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();
}

/*
 * Write an arbitrary 64-bit register to a buffer via MI_STORE_REGISTER_MEM.
 *
 * Only TIMESTAMP and PS_DEPTH_COUNT have special PIPE_CONTROL support; other
 * counters have to be read via the generic MI_STORE_REGISTER_MEM.  This
 * function also performs a pipeline flush for proper synchronization.
 */
static void
write_reg(struct brw_context *brw,
          drm_intel_bo *query_bo, uint32_t reg, int idx)
{
   assert(brw->gen >= 6);

   intel_batchbuffer_emit_mi_flush(brw);

   /* MI_STORE_REGISTER_MEM only stores a single 32-bit value, so to
    * read a full 64-bit register, we need to do two of them.
    */
   BEGIN_BATCH(3);
   OUT_BATCH(MI_STORE_REGISTER_MEM | (3 - 2));
   OUT_BATCH(reg);
   OUT_RELOC(query_bo, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
             idx * sizeof(uint64_t));
   ADVANCE_BATCH();

   BEGIN_BATCH(3);
   OUT_BATCH(MI_STORE_REGISTER_MEM | (3 - 2));
   OUT_BATCH(reg + sizeof(uint32_t));
   OUT_RELOC(query_bo, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
             sizeof(uint32_t) + idx * sizeof(uint64_t));
   ADVANCE_BATCH();
}

static void
write_primitives_generated(struct brw_context *brw,
                           drm_intel_bo *query_bo, int idx)
{
   write_reg(brw, query_bo, CL_INVOCATION_COUNT, idx);
}

static void
write_xfb_primitives_written(struct brw_context *brw,
                             drm_intel_bo *query_bo, int idx)
{
   if (brw->gen >= 7) {
      write_reg(brw, query_bo, GEN7_SO_NUM_PRIMS_WRITTEN(0), idx);
   } else {
      write_reg(brw, query_bo, GEN6_SO_NUM_PRIMS_WRITTEN, idx);
   }
}

/**
 * Wait on the query object's BO and calculate the final result.
 */
static void
gen6_queryobj_get_results(struct gl_context *ctx,
                          struct brw_query_object *query)
{
   struct brw_context *brw = brw_context(ctx);

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

   /* If the application has requested the query result, but this batch is
    * still contributing to it, flush it now so the results will be present
    * when mapped.
    */
   if (drm_intel_bo_references(brw->batch.bo, query->bo))
      intel_batchbuffer_flush(brw);

   if (unlikely(brw->perf_debug)) {
      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);
   uint64_t *results = query->bo->virtual;
   switch (query->Base.Target) {
   case GL_TIME_ELAPSED:
      /* The query BO contains the starting and ending timestamps.
       * Subtract the two and convert to nanoseconds.
       */
      query->Base.Result += 80 * (results[1] - results[0]);
      break;

   case GL_TIMESTAMP:
      /* 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.
       *
       * The query BO contains a single timestamp value in results[0].
       */
      query->Base.Result = 80 * (results[0] & 0xffffffff);
      query->Base.Result &= (1ull << 36) - 1;
      break;

   case GL_SAMPLES_PASSED_ARB:
      /* We need to use += rather than = here since some BLT-based operations
       * may have added additional samples to our occlusion query value.
       */
      query->Base.Result += results[1] - results[0];
      break;

   case GL_ANY_SAMPLES_PASSED:
   case GL_ANY_SAMPLES_PASSED_CONSERVATIVE:
      if (results[0] != results[1])
         query->Base.Result = true;
      break;

   case GL_PRIMITIVES_GENERATED:
   case GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN:
      query->Base.Result = results[1] - results[0];
      break;

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

   /* Now that we've processed the data stored in the query's buffer object,
    * we can release it.
    */
   drm_intel_bo_unreference(query->bo);
   query->bo = NULL;
}

/**
 * Driver hook for glBeginQuery().
 *
 * Initializes driver structures and emits any GPU commands required to begin
 * recording data for the query.
 */
static void
gen6_begin_query(struct gl_context *ctx, struct gl_query_object *q)
{
   struct brw_context *brw = brw_context(ctx);
   struct brw_query_object *query = (struct brw_query_object *)q;

   /* Since we're starting a new query, we need to throw away old results. */
   drm_intel_bo_unreference(query->bo);
   query->bo = drm_intel_bo_alloc(brw->bufmgr, "query results", 4096, 4096);

   switch (query->Base.Target) {
   case GL_TIME_ELAPSED:
      /* For timestamp queries, we record the starting time right away so that
       * we measure the full time between BeginQuery and EndQuery.  There's
       * some debate about whether this is the right thing to do.  Our decision
       * is based on the following text from the ARB_timer_query extension:
       *
       * "(5) Should the extension measure total time elapsed between the full
       *      completion of the BeginQuery and EndQuery commands, or just time
       *      spent in the graphics library?
       *
       *  RESOLVED:  This extension will measure the total time elapsed
       *  between the full completion of these commands.  Future extensions
       *  may implement a query to determine time elapsed at different stages
       *  of the graphics pipeline."
       *
       * We write a starting timestamp now (at index 0).  At EndQuery() time,
       * we'll write a second timestamp (at index 1), and subtract the two to
       * obtain the time elapsed.  Notably, this includes time elapsed while
       * the system was doing other work, such as running other applications.
       */
      write_timestamp(brw, query->bo, 0);
      break;

   case GL_ANY_SAMPLES_PASSED:
   case GL_ANY_SAMPLES_PASSED_CONSERVATIVE:
   case GL_SAMPLES_PASSED_ARB:
      write_depth_count(brw, query->bo, 0);
      break;

   case GL_PRIMITIVES_GENERATED:
      write_primitives_generated(brw, query->bo, 0);
      break;

   case GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN:
      write_xfb_primitives_written(brw, query->bo, 0);
      break;

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

/**
 * Driver hook for glEndQuery().
 *
 * Emits GPU commands to record a final query value, ending any data capturing.
 * However, the final result isn't necessarily available until the GPU processes
 * those commands.  brw_queryobj_get_results() processes the captured data to
 * produce the final result.
 */
static void
gen6_end_query(struct gl_context *ctx, struct gl_query_object *q)
{
   struct brw_context *brw = brw_context(ctx);
   struct brw_query_object *query = (struct brw_query_object *)q;

   switch (query->Base.Target) {
   case GL_TIME_ELAPSED:
      write_timestamp(brw, query->bo, 1);
      break;

   case GL_ANY_SAMPLES_PASSED:
   case GL_ANY_SAMPLES_PASSED_CONSERVATIVE:
   case GL_SAMPLES_PASSED_ARB:
      write_depth_count(brw, query->bo, 1);
      break;

   case GL_PRIMITIVES_GENERATED:
      write_primitives_generated(brw, query->bo, 1);
      break;

   case GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN:
      write_xfb_primitives_written(brw, query->bo, 1);
      break;

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

/**
 * The WaitQuery() driver hook.
 *
 * Wait for a query result to become available and return it.  This is the
 * backing for glGetQueryObjectiv() with the GL_QUERY_RESULT pname.
 */
static void gen6_wait_query(struct gl_context *ctx, struct gl_query_object *q)
{
   struct brw_query_object *query = (struct brw_query_object *)q;

   gen6_queryobj_get_results(ctx, query);
   query->Base.Ready = true;
}

/**
 * The CheckQuery() driver hook.
 *
 * Checks whether a query result is ready yet.  If not, flushes.
 * This is the backing for glGetQueryObjectiv()'s QUERY_RESULT_AVAILABLE pname.
 */
static void gen6_check_query(struct gl_context *ctx, struct gl_query_object *q)
{
   struct brw_context *brw = brw_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(brw->batch.bo, query->bo))
      intel_batchbuffer_flush(brw);

   if (query->bo == NULL || !drm_intel_bo_busy(query->bo)) {
      gen6_queryobj_get_results(ctx, query);
      query->Base.Ready = true;
   }
}

/* Initialize Gen6+-specific query object functions. */
void gen6_init_queryobj_functions(struct dd_function_table *functions)
{
   functions->BeginQuery = gen6_begin_query;
   functions->EndQuery = gen6_end_query;
   functions->CheckQuery = gen6_check_query;
   functions->WaitQuery = gen6_wait_query;
}