1 /**********************************************************
2 * Copyright 2008-2015 VMware, Inc. All rights reserved.
3 *
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 **********************************************************/
25
26 #include "pipe/p_state.h"
27 #include "pipe/p_context.h"
28
29 #include "util/u_bitmask.h"
30 #include "util/u_memory.h"
31
32 #include "svga_cmd.h"
33 #include "svga_context.h"
34 #include "svga_screen.h"
35 #include "svga_resource_buffer.h"
36 #include "svga_winsys.h"
37 #include "svga_debug.h"
38
39
40 /* Fixme: want a public base class for all pipe structs, even if there
41 * isn't much in them.
42 */
43 struct pipe_query {
44 int dummy;
45 };
46
47 struct svga_query {
48 struct pipe_query base;
49 unsigned type; /**< PIPE_QUERY_x or SVGA_QUERY_x */
50 SVGA3dQueryType svga_type; /**< SVGA3D_QUERYTYPE_x or unused */
51
52 unsigned id; /** Per-context query identifier */
53
54 struct pipe_fence_handle *fence;
55
56 /** For PIPE_QUERY_OCCLUSION_COUNTER / SVGA3D_QUERYTYPE_OCCLUSION */
57
58 /* For VGPU9 */
59 struct svga_winsys_buffer *hwbuf;
60 volatile SVGA3dQueryResult *queryResult;
61
62 /** For VGPU10 */
63 struct svga_winsys_gb_query *gb_query;
64 SVGA3dDXQueryFlags flags;
65 unsigned offset; /**< offset to the gb_query memory */
66 struct pipe_query *predicate; /** The associated query that can be used for predicate */
67
68 /** For non-GPU SVGA_QUERY_x queries */
69 uint64_t begin_count, end_count;
70 };
71
72
73 /** cast wrapper */
74 static inline struct svga_query *
svga_query(struct pipe_query * q)75 svga_query(struct pipe_query *q)
76 {
77 return (struct svga_query *)q;
78 }
79
80 /**
81 * VGPU9
82 */
83
84 static boolean
85 svga_get_query_result(struct pipe_context *pipe,
86 struct pipe_query *q,
87 boolean wait,
88 union pipe_query_result *result);
89
90 static enum pipe_error
define_query_vgpu9(struct svga_context * svga,struct svga_query * sq)91 define_query_vgpu9(struct svga_context *svga,
92 struct svga_query *sq)
93 {
94 struct svga_winsys_screen *sws = svga_screen(svga->pipe.screen)->sws;
95
96 sq->hwbuf = svga_winsys_buffer_create(svga, 1,
97 SVGA_BUFFER_USAGE_PINNED,
98 sizeof *sq->queryResult);
99 if (!sq->hwbuf)
100 return PIPE_ERROR_OUT_OF_MEMORY;
101
102 sq->queryResult = (SVGA3dQueryResult *)
103 sws->buffer_map(sws, sq->hwbuf, PIPE_TRANSFER_WRITE);
104 if (!sq->queryResult) {
105 sws->buffer_destroy(sws, sq->hwbuf);
106 return PIPE_ERROR_OUT_OF_MEMORY;
107 }
108
109 sq->queryResult->totalSize = sizeof *sq->queryResult;
110 sq->queryResult->state = SVGA3D_QUERYSTATE_NEW;
111
112 /* We request the buffer to be pinned and assume it is always mapped.
113 * The reason is that we don't want to wait for fences when checking the
114 * query status.
115 */
116 sws->buffer_unmap(sws, sq->hwbuf);
117
118 return PIPE_OK;
119 }
120
121 static enum pipe_error
begin_query_vgpu9(struct svga_context * svga,struct svga_query * sq)122 begin_query_vgpu9(struct svga_context *svga, struct svga_query *sq)
123 {
124 struct svga_winsys_screen *sws = svga_screen(svga->pipe.screen)->sws;
125 enum pipe_error ret = PIPE_OK;
126
127 if (sq->queryResult->state == SVGA3D_QUERYSTATE_PENDING) {
128 /* The application doesn't care for the pending query result.
129 * We cannot let go of the existing buffer and just get a new one
130 * because its storage may be reused for other purposes and clobbered
131 * by the host when it determines the query result. So the only
132 * option here is to wait for the existing query's result -- not a
133 * big deal, given that no sane application would do this.
134 */
135 uint64_t result;
136 svga_get_query_result(&svga->pipe, &sq->base, TRUE, (void*)&result);
137 assert(sq->queryResult->state != SVGA3D_QUERYSTATE_PENDING);
138 }
139
140 sq->queryResult->state = SVGA3D_QUERYSTATE_NEW;
141 sws->fence_reference(sws, &sq->fence, NULL);
142
143 ret = SVGA3D_BeginQuery(svga->swc, sq->svga_type);
144 if (ret != PIPE_OK) {
145 svga_context_flush(svga, NULL);
146 ret = SVGA3D_BeginQuery(svga->swc, sq->svga_type);
147 }
148 return ret;
149 }
150
151 static enum pipe_error
end_query_vgpu9(struct svga_context * svga,struct svga_query * sq)152 end_query_vgpu9(struct svga_context *svga, struct svga_query *sq)
153 {
154 enum pipe_error ret = PIPE_OK;
155
156 /* Set to PENDING before sending EndQuery. */
157 sq->queryResult->state = SVGA3D_QUERYSTATE_PENDING;
158
159 ret = SVGA3D_EndQuery(svga->swc, sq->svga_type, sq->hwbuf);
160 if (ret != PIPE_OK) {
161 svga_context_flush(svga, NULL);
162 ret = SVGA3D_EndQuery(svga->swc, sq->svga_type, sq->hwbuf);
163 }
164 return ret;
165 }
166
167 static boolean
get_query_result_vgpu9(struct svga_context * svga,struct svga_query * sq,boolean wait,uint64_t * result)168 get_query_result_vgpu9(struct svga_context *svga, struct svga_query *sq,
169 boolean wait, uint64_t *result)
170 {
171 struct svga_winsys_screen *sws = svga_screen(svga->pipe.screen)->sws;
172 enum pipe_error ret;
173 SVGA3dQueryState state;
174
175 if (!sq->fence) {
176 /* The query status won't be updated by the host unless
177 * SVGA_3D_CMD_WAIT_FOR_QUERY is emitted. Unfortunately this will cause
178 * a synchronous wait on the host.
179 */
180 ret = SVGA3D_WaitForQuery(svga->swc, sq->svga_type, sq->hwbuf);
181 if (ret != PIPE_OK) {
182 svga_context_flush(svga, NULL);
183 ret = SVGA3D_WaitForQuery(svga->swc, sq->svga_type, sq->hwbuf);
184 }
185 assert (ret == PIPE_OK);
186 svga_context_flush(svga, &sq->fence);
187 assert(sq->fence);
188 }
189
190 state = sq->queryResult->state;
191 if (state == SVGA3D_QUERYSTATE_PENDING) {
192 if (!wait)
193 return FALSE;
194 sws->fence_finish(sws, sq->fence, PIPE_TIMEOUT_INFINITE,
195 SVGA_FENCE_FLAG_QUERY);
196 state = sq->queryResult->state;
197 }
198
199 assert(state == SVGA3D_QUERYSTATE_SUCCEEDED ||
200 state == SVGA3D_QUERYSTATE_FAILED);
201
202 *result = (uint64_t)sq->queryResult->result32;
203 return TRUE;
204 }
205
206
207 /**
208 * VGPU10
209 *
210 * There is one query mob allocated for each context to be shared by all
211 * query types. The mob is used to hold queries's state and result. Since
212 * each query result type is of different length, to ease the query allocation
213 * management, the mob is divided into memory blocks. Each memory block
214 * will hold queries of the same type. Multiple memory blocks can be allocated
215 * for a particular query type.
216 *
217 * Currently each memory block is of 184 bytes. We support up to 128
218 * memory blocks. The query memory size is arbitrary right now.
219 * Each occlusion query takes about 8 bytes. One memory block can accomodate
220 * 23 occlusion queries. 128 of those blocks can support up to 2944 occlusion
221 * queries. That seems reasonable for now. If we think this limit is
222 * not enough, we can increase the limit or try to grow the mob in runtime.
223 * Note, SVGA device does not impose one mob per context for queries,
224 * we could allocate multiple mobs for queries; however, wddm KMD does not
225 * currently support that.
226 *
227 * Also note that the GL guest driver does not issue any of the
228 * following commands: DXMoveQuery, DXBindAllQuery & DXReadbackAllQuery.
229 */
230 #define SVGA_QUERY_MEM_BLOCK_SIZE (sizeof(SVGADXQueryResultUnion) * 2)
231 #define SVGA_QUERY_MEM_SIZE (128 * SVGA_QUERY_MEM_BLOCK_SIZE)
232
233 struct svga_qmem_alloc_entry
234 {
235 unsigned start_offset; /* start offset of the memory block */
236 unsigned block_index; /* block index of the memory block */
237 unsigned query_size; /* query size in this memory block */
238 unsigned nquery; /* number of queries allocated */
239 struct util_bitmask *alloc_mask; /* allocation mask */
240 struct svga_qmem_alloc_entry *next; /* next memory block */
241 };
242
243
244 /**
245 * Allocate a memory block from the query object memory
246 * \return -1 if out of memory, else index of the query memory block
247 */
248 static int
allocate_query_block(struct svga_context * svga)249 allocate_query_block(struct svga_context *svga)
250 {
251 int index;
252 unsigned offset;
253
254 /* Find the next available query block */
255 index = util_bitmask_add(svga->gb_query_alloc_mask);
256
257 if (index == UTIL_BITMASK_INVALID_INDEX)
258 return -1;
259
260 offset = index * SVGA_QUERY_MEM_BLOCK_SIZE;
261 if (offset >= svga->gb_query_len) {
262 unsigned i;
263
264 /**
265 * All the memory blocks are allocated, lets see if there is
266 * any empty memory block around that can be freed up.
267 */
268 index = -1;
269 for (i = 0; i < SVGA3D_QUERYTYPE_MAX && index == -1; i++) {
270 struct svga_qmem_alloc_entry *alloc_entry;
271 struct svga_qmem_alloc_entry *prev_alloc_entry = NULL;
272
273 alloc_entry = svga->gb_query_map[i];
274 while (alloc_entry && index == -1) {
275 if (alloc_entry->nquery == 0) {
276 /* This memory block is empty, it can be recycled. */
277 if (prev_alloc_entry) {
278 prev_alloc_entry->next = alloc_entry->next;
279 } else {
280 svga->gb_query_map[i] = alloc_entry->next;
281 }
282 index = alloc_entry->block_index;
283 } else {
284 prev_alloc_entry = alloc_entry;
285 alloc_entry = alloc_entry->next;
286 }
287 }
288 }
289 }
290
291 return index;
292 }
293
294 /**
295 * Allocate a slot in the specified memory block.
296 * All slots in this memory block are of the same size.
297 *
298 * \return -1 if out of memory, else index of the query slot
299 */
300 static int
allocate_query_slot(struct svga_context * svga,struct svga_qmem_alloc_entry * alloc)301 allocate_query_slot(struct svga_context *svga,
302 struct svga_qmem_alloc_entry *alloc)
303 {
304 int index;
305 unsigned offset;
306
307 /* Find the next available slot */
308 index = util_bitmask_add(alloc->alloc_mask);
309
310 if (index == UTIL_BITMASK_INVALID_INDEX)
311 return -1;
312
313 offset = index * alloc->query_size;
314 if (offset >= SVGA_QUERY_MEM_BLOCK_SIZE)
315 return -1;
316
317 alloc->nquery++;
318
319 return index;
320 }
321
322 /**
323 * Deallocate the specified slot in the memory block.
324 * If all slots are freed up, then deallocate the memory block
325 * as well, so it can be allocated for other query type
326 */
327 static void
deallocate_query_slot(struct svga_context * svga,struct svga_qmem_alloc_entry * alloc,unsigned index)328 deallocate_query_slot(struct svga_context *svga,
329 struct svga_qmem_alloc_entry *alloc,
330 unsigned index)
331 {
332 assert(index != UTIL_BITMASK_INVALID_INDEX);
333
334 util_bitmask_clear(alloc->alloc_mask, index);
335 alloc->nquery--;
336
337 /**
338 * Don't worry about deallocating the empty memory block here.
339 * The empty memory block will be recycled when no more memory block
340 * can be allocated.
341 */
342 }
343
344 static struct svga_qmem_alloc_entry *
allocate_query_block_entry(struct svga_context * svga,unsigned len)345 allocate_query_block_entry(struct svga_context *svga,
346 unsigned len)
347 {
348 struct svga_qmem_alloc_entry *alloc_entry;
349 int block_index = -1;
350
351 block_index = allocate_query_block(svga);
352 if (block_index == -1)
353 return NULL;
354 alloc_entry = CALLOC_STRUCT(svga_qmem_alloc_entry);
355 if (!alloc_entry)
356 return NULL;
357
358 alloc_entry->block_index = block_index;
359 alloc_entry->start_offset = block_index * SVGA_QUERY_MEM_BLOCK_SIZE;
360 alloc_entry->nquery = 0;
361 alloc_entry->alloc_mask = util_bitmask_create();
362 alloc_entry->next = NULL;
363 alloc_entry->query_size = len;
364
365 return alloc_entry;
366 }
367
368 /**
369 * Allocate a memory slot for a query of the specified type.
370 * It will first search through the memory blocks that are allocated
371 * for the query type. If no memory slot is available, it will try
372 * to allocate another memory block within the query object memory for
373 * this query type.
374 */
375 static int
allocate_query(struct svga_context * svga,SVGA3dQueryType type,unsigned len)376 allocate_query(struct svga_context *svga,
377 SVGA3dQueryType type,
378 unsigned len)
379 {
380 struct svga_qmem_alloc_entry *alloc_entry;
381 int slot_index = -1;
382 unsigned offset;
383
384 assert(type < SVGA3D_QUERYTYPE_MAX);
385
386 alloc_entry = svga->gb_query_map[type];
387
388 if (!alloc_entry) {
389 /**
390 * No query memory block has been allocated for this query type,
391 * allocate one now
392 */
393 alloc_entry = allocate_query_block_entry(svga, len);
394 if (!alloc_entry)
395 return -1;
396 svga->gb_query_map[type] = alloc_entry;
397 }
398
399 /* Allocate a slot within the memory block allocated for this query type */
400 slot_index = allocate_query_slot(svga, alloc_entry);
401
402 if (slot_index == -1) {
403 /* This query memory block is full, allocate another one */
404 alloc_entry = allocate_query_block_entry(svga, len);
405 if (!alloc_entry)
406 return -1;
407 alloc_entry->next = svga->gb_query_map[type];
408 svga->gb_query_map[type] = alloc_entry;
409 slot_index = allocate_query_slot(svga, alloc_entry);
410 }
411
412 assert(slot_index != -1);
413 offset = slot_index * len + alloc_entry->start_offset;
414
415 return offset;
416 }
417
418
419 /**
420 * Deallocate memory slot allocated for the specified query
421 */
422 static void
deallocate_query(struct svga_context * svga,struct svga_query * sq)423 deallocate_query(struct svga_context *svga,
424 struct svga_query *sq)
425 {
426 struct svga_qmem_alloc_entry *alloc_entry;
427 unsigned slot_index;
428 unsigned offset = sq->offset;
429
430 alloc_entry = svga->gb_query_map[sq->svga_type];
431
432 while (alloc_entry) {
433 if (offset >= alloc_entry->start_offset &&
434 offset < alloc_entry->start_offset + SVGA_QUERY_MEM_BLOCK_SIZE) {
435
436 /* The slot belongs to this memory block, deallocate it */
437 slot_index = (offset - alloc_entry->start_offset) /
438 alloc_entry->query_size;
439 deallocate_query_slot(svga, alloc_entry, slot_index);
440 alloc_entry = NULL;
441 } else {
442 alloc_entry = alloc_entry->next;
443 }
444 }
445 }
446
447
448 /**
449 * Destroy the gb query object and all the related query structures
450 */
451 static void
destroy_gb_query_obj(struct svga_context * svga)452 destroy_gb_query_obj(struct svga_context *svga)
453 {
454 struct svga_winsys_screen *sws = svga_screen(svga->pipe.screen)->sws;
455 unsigned i;
456
457 for (i = 0; i < SVGA3D_QUERYTYPE_MAX; i++) {
458 struct svga_qmem_alloc_entry *alloc_entry, *next;
459 alloc_entry = svga->gb_query_map[i];
460 while (alloc_entry) {
461 next = alloc_entry->next;
462 util_bitmask_destroy(alloc_entry->alloc_mask);
463 FREE(alloc_entry);
464 alloc_entry = next;
465 }
466 svga->gb_query_map[i] = NULL;
467 }
468
469 if (svga->gb_query)
470 sws->query_destroy(sws, svga->gb_query);
471 svga->gb_query = NULL;
472
473 util_bitmask_destroy(svga->gb_query_alloc_mask);
474 }
475
476 /**
477 * Define query and create the gb query object if it is not already created.
478 * There is only one gb query object per context which will be shared by
479 * queries of all types.
480 */
481 static enum pipe_error
define_query_vgpu10(struct svga_context * svga,struct svga_query * sq,int resultLen)482 define_query_vgpu10(struct svga_context *svga,
483 struct svga_query *sq, int resultLen)
484 {
485 struct svga_winsys_screen *sws = svga_screen(svga->pipe.screen)->sws;
486 int qlen;
487 enum pipe_error ret = PIPE_OK;
488
489 SVGA_DBG(DEBUG_QUERY, "%s\n", __FUNCTION__);
490
491 if (svga->gb_query == NULL) {
492 /* Create a gb query object */
493 svga->gb_query = sws->query_create(sws, SVGA_QUERY_MEM_SIZE);
494 if (!svga->gb_query)
495 return PIPE_ERROR_OUT_OF_MEMORY;
496 svga->gb_query_len = SVGA_QUERY_MEM_SIZE;
497 memset (svga->gb_query_map, 0, sizeof(svga->gb_query_map));
498 svga->gb_query_alloc_mask = util_bitmask_create();
499
500 /* Bind the query object to the context */
501 if (svga->swc->query_bind(svga->swc, svga->gb_query,
502 SVGA_QUERY_FLAG_SET) != PIPE_OK) {
503 svga_context_flush(svga, NULL);
504 svga->swc->query_bind(svga->swc, svga->gb_query,
505 SVGA_QUERY_FLAG_SET);
506 }
507 }
508
509 sq->gb_query = svga->gb_query;
510
511 /* Allocate an integer ID for this query */
512 sq->id = util_bitmask_add(svga->query_id_bm);
513 if (sq->id == UTIL_BITMASK_INVALID_INDEX)
514 return PIPE_ERROR_OUT_OF_MEMORY;
515
516 /* Find a slot for this query in the gb object */
517 qlen = resultLen + sizeof(SVGA3dQueryState);
518 sq->offset = allocate_query(svga, sq->svga_type, qlen);
519 if (sq->offset == -1)
520 return PIPE_ERROR_OUT_OF_MEMORY;
521
522 SVGA_DBG(DEBUG_QUERY, " query type=%d qid=0x%x offset=%d\n",
523 sq->svga_type, sq->id, sq->offset);
524
525 /**
526 * Send SVGA3D commands to define the query
527 */
528 ret = SVGA3D_vgpu10_DefineQuery(svga->swc, sq->id, sq->svga_type, sq->flags);
529 if (ret != PIPE_OK) {
530 svga_context_flush(svga, NULL);
531 ret = SVGA3D_vgpu10_DefineQuery(svga->swc, sq->id, sq->svga_type, sq->flags);
532 }
533 if (ret != PIPE_OK)
534 return PIPE_ERROR_OUT_OF_MEMORY;
535
536 ret = SVGA3D_vgpu10_BindQuery(svga->swc, sq->gb_query, sq->id);
537 if (ret != PIPE_OK) {
538 svga_context_flush(svga, NULL);
539 ret = SVGA3D_vgpu10_BindQuery(svga->swc, sq->gb_query, sq->id);
540 }
541 assert(ret == PIPE_OK);
542
543 ret = SVGA3D_vgpu10_SetQueryOffset(svga->swc, sq->id, sq->offset);
544 if (ret != PIPE_OK) {
545 svga_context_flush(svga, NULL);
546 ret = SVGA3D_vgpu10_SetQueryOffset(svga->swc, sq->id, sq->offset);
547 }
548 assert(ret == PIPE_OK);
549
550 return PIPE_OK;
551 }
552
553 static enum pipe_error
destroy_query_vgpu10(struct svga_context * svga,struct svga_query * sq)554 destroy_query_vgpu10(struct svga_context *svga, struct svga_query *sq)
555 {
556 enum pipe_error ret;
557
558 ret = SVGA3D_vgpu10_DestroyQuery(svga->swc, sq->id);
559
560 /* Deallocate the memory slot allocated for this query */
561 deallocate_query(svga, sq);
562
563 return ret;
564 }
565
566
567 /**
568 * Rebind queryies to the context.
569 */
570 static void
rebind_vgpu10_query(struct svga_context * svga)571 rebind_vgpu10_query(struct svga_context *svga)
572 {
573 if (svga->swc->query_bind(svga->swc, svga->gb_query,
574 SVGA_QUERY_FLAG_REF) != PIPE_OK) {
575 svga_context_flush(svga, NULL);
576 svga->swc->query_bind(svga->swc, svga->gb_query,
577 SVGA_QUERY_FLAG_REF);
578 }
579
580 svga->rebind.flags.query = FALSE;
581 }
582
583
584 static enum pipe_error
begin_query_vgpu10(struct svga_context * svga,struct svga_query * sq)585 begin_query_vgpu10(struct svga_context *svga, struct svga_query *sq)
586 {
587 struct svga_winsys_screen *sws = svga_screen(svga->pipe.screen)->sws;
588 enum pipe_error ret = PIPE_OK;
589 int status = 0;
590
591 sws->fence_reference(sws, &sq->fence, NULL);
592
593 /* Initialize the query state to NEW */
594 status = sws->query_init(sws, sq->gb_query, sq->offset, SVGA3D_QUERYSTATE_NEW);
595 if (status)
596 return PIPE_ERROR;
597
598 if (svga->rebind.flags.query) {
599 rebind_vgpu10_query(svga);
600 }
601
602 /* Send the BeginQuery command to the device */
603 ret = SVGA3D_vgpu10_BeginQuery(svga->swc, sq->id);
604 if (ret != PIPE_OK) {
605 svga_context_flush(svga, NULL);
606 ret = SVGA3D_vgpu10_BeginQuery(svga->swc, sq->id);
607 }
608 return ret;
609 }
610
611 static enum pipe_error
end_query_vgpu10(struct svga_context * svga,struct svga_query * sq)612 end_query_vgpu10(struct svga_context *svga, struct svga_query *sq)
613 {
614 enum pipe_error ret = PIPE_OK;
615
616 if (svga->rebind.flags.query) {
617 rebind_vgpu10_query(svga);
618 }
619
620 ret = SVGA3D_vgpu10_EndQuery(svga->swc, sq->id);
621 if (ret != PIPE_OK) {
622 svga_context_flush(svga, NULL);
623 ret = SVGA3D_vgpu10_EndQuery(svga->swc, sq->id);
624 }
625
626 return ret;
627 }
628
629 static boolean
get_query_result_vgpu10(struct svga_context * svga,struct svga_query * sq,boolean wait,void * result,int resultLen)630 get_query_result_vgpu10(struct svga_context *svga, struct svga_query *sq,
631 boolean wait, void *result, int resultLen)
632 {
633 struct svga_winsys_screen *sws = svga_screen(svga->pipe.screen)->sws;
634 SVGA3dQueryState queryState;
635
636 if (svga->rebind.flags.query) {
637 rebind_vgpu10_query(svga);
638 }
639
640 sws->query_get_result(sws, sq->gb_query, sq->offset, &queryState, result, resultLen);
641
642 if (queryState != SVGA3D_QUERYSTATE_SUCCEEDED && !sq->fence) {
643 /* We don't have the query result yet, and the query hasn't been
644 * submitted. We need to submit it now since the GL spec says
645 * "Querying the state for a given occlusion query forces that
646 * occlusion query to complete within a finite amount of time."
647 */
648 svga_context_flush(svga, &sq->fence);
649 }
650
651 if (queryState == SVGA3D_QUERYSTATE_PENDING ||
652 queryState == SVGA3D_QUERYSTATE_NEW) {
653 if (!wait)
654 return FALSE;
655 sws->fence_finish(sws, sq->fence, PIPE_TIMEOUT_INFINITE,
656 SVGA_FENCE_FLAG_QUERY);
657 sws->query_get_result(sws, sq->gb_query, sq->offset, &queryState, result, resultLen);
658 }
659
660 assert(queryState == SVGA3D_QUERYSTATE_SUCCEEDED ||
661 queryState == SVGA3D_QUERYSTATE_FAILED);
662
663 return TRUE;
664 }
665
666 static struct pipe_query *
svga_create_query(struct pipe_context * pipe,unsigned query_type,unsigned index)667 svga_create_query(struct pipe_context *pipe,
668 unsigned query_type,
669 unsigned index)
670 {
671 struct svga_context *svga = svga_context(pipe);
672 struct svga_query *sq;
673
674 assert(query_type < SVGA_QUERY_MAX);
675
676 sq = CALLOC_STRUCT(svga_query);
677 if (!sq)
678 goto fail;
679
680 /* Allocate an integer ID for the query */
681 sq->id = util_bitmask_add(svga->query_id_bm);
682 if (sq->id == UTIL_BITMASK_INVALID_INDEX)
683 goto fail;
684
685 SVGA_DBG(DEBUG_QUERY, "%s type=%d sq=0x%x id=%d\n", __FUNCTION__,
686 query_type, sq, sq->id);
687
688 switch (query_type) {
689 case PIPE_QUERY_OCCLUSION_COUNTER:
690 sq->svga_type = SVGA3D_QUERYTYPE_OCCLUSION;
691 if (svga_have_vgpu10(svga)) {
692 define_query_vgpu10(svga, sq, sizeof(SVGADXOcclusionQueryResult));
693
694 /**
695 * In OpenGL, occlusion counter query can be used in conditional
696 * rendering; however, in DX10, only OCCLUSION_PREDICATE query can
697 * be used for predication. Hence, we need to create an occlusion
698 * predicate query along with the occlusion counter query. So when
699 * the occlusion counter query is used for predication, the associated
700 * query of occlusion predicate type will be used
701 * in the SetPredication command.
702 */
703 sq->predicate = svga_create_query(pipe, PIPE_QUERY_OCCLUSION_PREDICATE, index);
704
705 } else {
706 define_query_vgpu9(svga, sq);
707 }
708 break;
709 case PIPE_QUERY_OCCLUSION_PREDICATE:
710 case PIPE_QUERY_OCCLUSION_PREDICATE_CONSERVATIVE:
711 if (svga_have_vgpu10(svga)) {
712 sq->svga_type = SVGA3D_QUERYTYPE_OCCLUSIONPREDICATE;
713 define_query_vgpu10(svga, sq, sizeof(SVGADXOcclusionPredicateQueryResult));
714 } else {
715 sq->svga_type = SVGA3D_QUERYTYPE_OCCLUSION;
716 define_query_vgpu9(svga, sq);
717 }
718 break;
719 case PIPE_QUERY_PRIMITIVES_GENERATED:
720 case PIPE_QUERY_PRIMITIVES_EMITTED:
721 case PIPE_QUERY_SO_STATISTICS:
722 assert(svga_have_vgpu10(svga));
723 sq->svga_type = SVGA3D_QUERYTYPE_STREAMOUTPUTSTATS;
724 define_query_vgpu10(svga, sq,
725 sizeof(SVGADXStreamOutStatisticsQueryResult));
726 break;
727 case PIPE_QUERY_TIMESTAMP:
728 assert(svga_have_vgpu10(svga));
729 sq->svga_type = SVGA3D_QUERYTYPE_TIMESTAMP;
730 define_query_vgpu10(svga, sq,
731 sizeof(SVGADXTimestampQueryResult));
732 break;
733 case SVGA_QUERY_NUM_DRAW_CALLS:
734 case SVGA_QUERY_NUM_FALLBACKS:
735 case SVGA_QUERY_NUM_FLUSHES:
736 case SVGA_QUERY_NUM_VALIDATIONS:
737 case SVGA_QUERY_NUM_BUFFERS_MAPPED:
738 case SVGA_QUERY_NUM_TEXTURES_MAPPED:
739 case SVGA_QUERY_NUM_BYTES_UPLOADED:
740 case SVGA_QUERY_COMMAND_BUFFER_SIZE:
741 case SVGA_QUERY_SURFACE_WRITE_FLUSHES:
742 case SVGA_QUERY_MEMORY_USED:
743 case SVGA_QUERY_NUM_SHADERS:
744 case SVGA_QUERY_NUM_RESOURCES:
745 case SVGA_QUERY_NUM_STATE_OBJECTS:
746 case SVGA_QUERY_NUM_SURFACE_VIEWS:
747 case SVGA_QUERY_NUM_GENERATE_MIPMAP:
748 case SVGA_QUERY_NUM_READBACKS:
749 case SVGA_QUERY_NUM_RESOURCE_UPDATES:
750 case SVGA_QUERY_NUM_BUFFER_UPLOADS:
751 case SVGA_QUERY_NUM_CONST_BUF_UPDATES:
752 case SVGA_QUERY_NUM_CONST_UPDATES:
753 case SVGA_QUERY_NUM_FAILED_ALLOCATIONS:
754 case SVGA_QUERY_NUM_COMMANDS_PER_DRAW:
755 break;
756 case SVGA_QUERY_FLUSH_TIME:
757 case SVGA_QUERY_MAP_BUFFER_TIME:
758 /* These queries need os_time_get() */
759 svga->hud.uses_time = TRUE;
760 break;
761 default:
762 assert(!"unexpected query type in svga_create_query()");
763 }
764
765 sq->type = query_type;
766
767 return &sq->base;
768
769 fail:
770 FREE(sq);
771 return NULL;
772 }
773
774 static void
svga_destroy_query(struct pipe_context * pipe,struct pipe_query * q)775 svga_destroy_query(struct pipe_context *pipe, struct pipe_query *q)
776 {
777 struct svga_context *svga = svga_context(pipe);
778 struct svga_winsys_screen *sws = svga_screen(svga->pipe.screen)->sws;
779 struct svga_query *sq;
780
781 if (!q) {
782 destroy_gb_query_obj(svga);
783 return;
784 }
785
786 sq = svga_query(q);
787
788 SVGA_DBG(DEBUG_QUERY, "%s sq=0x%x id=%d\n", __FUNCTION__,
789 sq, sq->id);
790
791 switch (sq->type) {
792 case PIPE_QUERY_OCCLUSION_COUNTER:
793 case PIPE_QUERY_OCCLUSION_PREDICATE:
794 case PIPE_QUERY_OCCLUSION_PREDICATE_CONSERVATIVE:
795 if (svga_have_vgpu10(svga)) {
796 /* make sure to also destroy any associated predicate query */
797 if (sq->predicate)
798 svga_destroy_query(pipe, sq->predicate);
799 destroy_query_vgpu10(svga, sq);
800 } else {
801 sws->buffer_destroy(sws, sq->hwbuf);
802 }
803 sws->fence_reference(sws, &sq->fence, NULL);
804 break;
805 case PIPE_QUERY_PRIMITIVES_GENERATED:
806 case PIPE_QUERY_PRIMITIVES_EMITTED:
807 case PIPE_QUERY_SO_STATISTICS:
808 case PIPE_QUERY_TIMESTAMP:
809 assert(svga_have_vgpu10(svga));
810 destroy_query_vgpu10(svga, sq);
811 sws->fence_reference(sws, &sq->fence, NULL);
812 break;
813 case SVGA_QUERY_NUM_DRAW_CALLS:
814 case SVGA_QUERY_NUM_FALLBACKS:
815 case SVGA_QUERY_NUM_FLUSHES:
816 case SVGA_QUERY_NUM_VALIDATIONS:
817 case SVGA_QUERY_MAP_BUFFER_TIME:
818 case SVGA_QUERY_NUM_BUFFERS_MAPPED:
819 case SVGA_QUERY_NUM_TEXTURES_MAPPED:
820 case SVGA_QUERY_NUM_BYTES_UPLOADED:
821 case SVGA_QUERY_COMMAND_BUFFER_SIZE:
822 case SVGA_QUERY_FLUSH_TIME:
823 case SVGA_QUERY_SURFACE_WRITE_FLUSHES:
824 case SVGA_QUERY_MEMORY_USED:
825 case SVGA_QUERY_NUM_SHADERS:
826 case SVGA_QUERY_NUM_RESOURCES:
827 case SVGA_QUERY_NUM_STATE_OBJECTS:
828 case SVGA_QUERY_NUM_SURFACE_VIEWS:
829 case SVGA_QUERY_NUM_GENERATE_MIPMAP:
830 case SVGA_QUERY_NUM_READBACKS:
831 case SVGA_QUERY_NUM_RESOURCE_UPDATES:
832 case SVGA_QUERY_NUM_BUFFER_UPLOADS:
833 case SVGA_QUERY_NUM_CONST_BUF_UPDATES:
834 case SVGA_QUERY_NUM_CONST_UPDATES:
835 case SVGA_QUERY_NUM_FAILED_ALLOCATIONS:
836 case SVGA_QUERY_NUM_COMMANDS_PER_DRAW:
837 /* nothing */
838 break;
839 default:
840 assert(!"svga: unexpected query type in svga_destroy_query()");
841 }
842
843 /* Free the query id */
844 util_bitmask_clear(svga->query_id_bm, sq->id);
845
846 FREE(sq);
847 }
848
849
850 static boolean
svga_begin_query(struct pipe_context * pipe,struct pipe_query * q)851 svga_begin_query(struct pipe_context *pipe, struct pipe_query *q)
852 {
853 struct svga_context *svga = svga_context(pipe);
854 struct svga_query *sq = svga_query(q);
855 enum pipe_error ret;
856
857 assert(sq);
858 assert(sq->type < SVGA_QUERY_MAX);
859
860 SVGA_DBG(DEBUG_QUERY, "%s sq=0x%x id=%d\n", __FUNCTION__,
861 sq, sq->id);
862
863 /* Need to flush out buffered drawing commands so that they don't
864 * get counted in the query results.
865 */
866 svga_hwtnl_flush_retry(svga);
867
868 switch (sq->type) {
869 case PIPE_QUERY_OCCLUSION_COUNTER:
870 case PIPE_QUERY_OCCLUSION_PREDICATE:
871 case PIPE_QUERY_OCCLUSION_PREDICATE_CONSERVATIVE:
872 if (svga_have_vgpu10(svga)) {
873 ret = begin_query_vgpu10(svga, sq);
874 /* also need to start the associated occlusion predicate query */
875 if (sq->predicate) {
876 enum pipe_error status;
877 status = begin_query_vgpu10(svga, svga_query(sq->predicate));
878 assert(status == PIPE_OK);
879 (void) status;
880 }
881 } else {
882 ret = begin_query_vgpu9(svga, sq);
883 }
884 assert(ret == PIPE_OK);
885 (void) ret;
886 break;
887 case PIPE_QUERY_PRIMITIVES_GENERATED:
888 case PIPE_QUERY_PRIMITIVES_EMITTED:
889 case PIPE_QUERY_SO_STATISTICS:
890 case PIPE_QUERY_TIMESTAMP:
891 assert(svga_have_vgpu10(svga));
892 ret = begin_query_vgpu10(svga, sq);
893 assert(ret == PIPE_OK);
894 break;
895 case SVGA_QUERY_NUM_DRAW_CALLS:
896 sq->begin_count = svga->hud.num_draw_calls;
897 break;
898 case SVGA_QUERY_NUM_FALLBACKS:
899 sq->begin_count = svga->hud.num_fallbacks;
900 break;
901 case SVGA_QUERY_NUM_FLUSHES:
902 sq->begin_count = svga->hud.num_flushes;
903 break;
904 case SVGA_QUERY_NUM_VALIDATIONS:
905 sq->begin_count = svga->hud.num_validations;
906 break;
907 case SVGA_QUERY_MAP_BUFFER_TIME:
908 sq->begin_count = svga->hud.map_buffer_time;
909 break;
910 case SVGA_QUERY_NUM_BUFFERS_MAPPED:
911 sq->begin_count = svga->hud.num_buffers_mapped;
912 break;
913 case SVGA_QUERY_NUM_TEXTURES_MAPPED:
914 sq->begin_count = svga->hud.num_textures_mapped;
915 break;
916 case SVGA_QUERY_NUM_BYTES_UPLOADED:
917 sq->begin_count = svga->hud.num_bytes_uploaded;
918 break;
919 case SVGA_QUERY_COMMAND_BUFFER_SIZE:
920 sq->begin_count = svga->hud.command_buffer_size;
921 break;
922 case SVGA_QUERY_FLUSH_TIME:
923 sq->begin_count = svga->hud.flush_time;
924 break;
925 case SVGA_QUERY_SURFACE_WRITE_FLUSHES:
926 sq->begin_count = svga->hud.surface_write_flushes;
927 break;
928 case SVGA_QUERY_NUM_READBACKS:
929 sq->begin_count = svga->hud.num_readbacks;
930 break;
931 case SVGA_QUERY_NUM_RESOURCE_UPDATES:
932 sq->begin_count = svga->hud.num_resource_updates;
933 break;
934 case SVGA_QUERY_NUM_BUFFER_UPLOADS:
935 sq->begin_count = svga->hud.num_buffer_uploads;
936 break;
937 case SVGA_QUERY_NUM_CONST_BUF_UPDATES:
938 sq->begin_count = svga->hud.num_const_buf_updates;
939 break;
940 case SVGA_QUERY_NUM_CONST_UPDATES:
941 sq->begin_count = svga->hud.num_const_updates;
942 break;
943 case SVGA_QUERY_MEMORY_USED:
944 case SVGA_QUERY_NUM_SHADERS:
945 case SVGA_QUERY_NUM_RESOURCES:
946 case SVGA_QUERY_NUM_STATE_OBJECTS:
947 case SVGA_QUERY_NUM_SURFACE_VIEWS:
948 case SVGA_QUERY_NUM_GENERATE_MIPMAP:
949 case SVGA_QUERY_NUM_FAILED_ALLOCATIONS:
950 case SVGA_QUERY_NUM_COMMANDS_PER_DRAW:
951 /* nothing */
952 break;
953 default:
954 assert(!"unexpected query type in svga_begin_query()");
955 }
956
957 svga->sq[sq->type] = sq;
958
959 return true;
960 }
961
962
963 static bool
svga_end_query(struct pipe_context * pipe,struct pipe_query * q)964 svga_end_query(struct pipe_context *pipe, struct pipe_query *q)
965 {
966 struct svga_context *svga = svga_context(pipe);
967 struct svga_query *sq = svga_query(q);
968 enum pipe_error ret;
969
970 assert(sq);
971 assert(sq->type < SVGA_QUERY_MAX);
972
973 SVGA_DBG(DEBUG_QUERY, "%s sq=0x%x id=%d\n", __FUNCTION__,
974 sq, sq->id);
975
976 if (sq->type == PIPE_QUERY_TIMESTAMP && svga->sq[sq->type] != sq)
977 svga_begin_query(pipe, q);
978
979 svga_hwtnl_flush_retry(svga);
980
981 assert(svga->sq[sq->type] == sq);
982
983 switch (sq->type) {
984 case PIPE_QUERY_OCCLUSION_COUNTER:
985 case PIPE_QUERY_OCCLUSION_PREDICATE:
986 case PIPE_QUERY_OCCLUSION_PREDICATE_CONSERVATIVE:
987 if (svga_have_vgpu10(svga)) {
988 ret = end_query_vgpu10(svga, sq);
989 /* also need to end the associated occlusion predicate query */
990 if (sq->predicate) {
991 enum pipe_error status;
992 status = end_query_vgpu10(svga, svga_query(sq->predicate));
993 assert(status == PIPE_OK);
994 (void) status;
995 }
996 } else {
997 ret = end_query_vgpu9(svga, sq);
998 }
999 assert(ret == PIPE_OK);
1000 (void) ret;
1001 break;
1002 case PIPE_QUERY_PRIMITIVES_GENERATED:
1003 case PIPE_QUERY_PRIMITIVES_EMITTED:
1004 case PIPE_QUERY_SO_STATISTICS:
1005 case PIPE_QUERY_TIMESTAMP:
1006 assert(svga_have_vgpu10(svga));
1007 ret = end_query_vgpu10(svga, sq);
1008 assert(ret == PIPE_OK);
1009 break;
1010 case SVGA_QUERY_NUM_DRAW_CALLS:
1011 sq->end_count = svga->hud.num_draw_calls;
1012 break;
1013 case SVGA_QUERY_NUM_FALLBACKS:
1014 sq->end_count = svga->hud.num_fallbacks;
1015 break;
1016 case SVGA_QUERY_NUM_FLUSHES:
1017 sq->end_count = svga->hud.num_flushes;
1018 break;
1019 case SVGA_QUERY_NUM_VALIDATIONS:
1020 sq->end_count = svga->hud.num_validations;
1021 break;
1022 case SVGA_QUERY_MAP_BUFFER_TIME:
1023 sq->end_count = svga->hud.map_buffer_time;
1024 break;
1025 case SVGA_QUERY_NUM_BUFFERS_MAPPED:
1026 sq->end_count = svga->hud.num_buffers_mapped;
1027 break;
1028 case SVGA_QUERY_NUM_TEXTURES_MAPPED:
1029 sq->end_count = svga->hud.num_textures_mapped;
1030 break;
1031 case SVGA_QUERY_NUM_BYTES_UPLOADED:
1032 sq->end_count = svga->hud.num_bytes_uploaded;
1033 break;
1034 case SVGA_QUERY_COMMAND_BUFFER_SIZE:
1035 sq->end_count = svga->hud.command_buffer_size;
1036 break;
1037 case SVGA_QUERY_FLUSH_TIME:
1038 sq->end_count = svga->hud.flush_time;
1039 break;
1040 case SVGA_QUERY_SURFACE_WRITE_FLUSHES:
1041 sq->end_count = svga->hud.surface_write_flushes;
1042 break;
1043 case SVGA_QUERY_NUM_READBACKS:
1044 sq->end_count = svga->hud.num_readbacks;
1045 break;
1046 case SVGA_QUERY_NUM_RESOURCE_UPDATES:
1047 sq->end_count = svga->hud.num_resource_updates;
1048 break;
1049 case SVGA_QUERY_NUM_BUFFER_UPLOADS:
1050 sq->end_count = svga->hud.num_buffer_uploads;
1051 break;
1052 case SVGA_QUERY_NUM_CONST_BUF_UPDATES:
1053 sq->end_count = svga->hud.num_const_buf_updates;
1054 break;
1055 case SVGA_QUERY_NUM_CONST_UPDATES:
1056 sq->end_count = svga->hud.num_const_updates;
1057 break;
1058 case SVGA_QUERY_MEMORY_USED:
1059 case SVGA_QUERY_NUM_SHADERS:
1060 case SVGA_QUERY_NUM_RESOURCES:
1061 case SVGA_QUERY_NUM_STATE_OBJECTS:
1062 case SVGA_QUERY_NUM_SURFACE_VIEWS:
1063 case SVGA_QUERY_NUM_GENERATE_MIPMAP:
1064 case SVGA_QUERY_NUM_FAILED_ALLOCATIONS:
1065 case SVGA_QUERY_NUM_COMMANDS_PER_DRAW:
1066 /* nothing */
1067 break;
1068 default:
1069 assert(!"unexpected query type in svga_end_query()");
1070 }
1071 svga->sq[sq->type] = NULL;
1072 return true;
1073 }
1074
1075
1076 static boolean
svga_get_query_result(struct pipe_context * pipe,struct pipe_query * q,boolean wait,union pipe_query_result * vresult)1077 svga_get_query_result(struct pipe_context *pipe,
1078 struct pipe_query *q,
1079 boolean wait,
1080 union pipe_query_result *vresult)
1081 {
1082 struct svga_screen *svgascreen = svga_screen(pipe->screen);
1083 struct svga_context *svga = svga_context(pipe);
1084 struct svga_query *sq = svga_query(q);
1085 uint64_t *result = (uint64_t *)vresult;
1086 boolean ret = TRUE;
1087
1088 assert(sq);
1089
1090 SVGA_DBG(DEBUG_QUERY, "%s sq=0x%x id=%d wait: %d\n",
1091 __FUNCTION__, sq, sq->id, wait);
1092
1093 switch (sq->type) {
1094 case PIPE_QUERY_OCCLUSION_COUNTER:
1095 if (svga_have_vgpu10(svga)) {
1096 SVGADXOcclusionQueryResult occResult;
1097 ret = get_query_result_vgpu10(svga, sq, wait,
1098 (void *)&occResult, sizeof(occResult));
1099 *result = (uint64_t)occResult.samplesRendered;
1100 } else {
1101 ret = get_query_result_vgpu9(svga, sq, wait, result);
1102 }
1103 break;
1104 case PIPE_QUERY_OCCLUSION_PREDICATE:
1105 case PIPE_QUERY_OCCLUSION_PREDICATE_CONSERVATIVE: {
1106 if (svga_have_vgpu10(svga)) {
1107 SVGADXOcclusionPredicateQueryResult occResult;
1108 ret = get_query_result_vgpu10(svga, sq, wait,
1109 (void *)&occResult, sizeof(occResult));
1110 vresult->b = occResult.anySamplesRendered != 0;
1111 } else {
1112 uint64_t count = 0;
1113 ret = get_query_result_vgpu9(svga, sq, wait, &count);
1114 vresult->b = count != 0;
1115 }
1116 break;
1117 }
1118 case PIPE_QUERY_SO_STATISTICS: {
1119 SVGADXStreamOutStatisticsQueryResult sResult;
1120 struct pipe_query_data_so_statistics *pResult =
1121 (struct pipe_query_data_so_statistics *)vresult;
1122
1123 assert(svga_have_vgpu10(svga));
1124 ret = get_query_result_vgpu10(svga, sq, wait,
1125 (void *)&sResult, sizeof(sResult));
1126 pResult->num_primitives_written = sResult.numPrimitivesWritten;
1127 pResult->primitives_storage_needed = sResult.numPrimitivesRequired;
1128 break;
1129 }
1130 case PIPE_QUERY_TIMESTAMP: {
1131 SVGADXTimestampQueryResult sResult;
1132
1133 assert(svga_have_vgpu10(svga));
1134 ret = get_query_result_vgpu10(svga, sq, wait,
1135 (void *)&sResult, sizeof(sResult));
1136 *result = (uint64_t)sResult.timestamp;
1137 break;
1138 }
1139 case PIPE_QUERY_PRIMITIVES_GENERATED: {
1140 SVGADXStreamOutStatisticsQueryResult sResult;
1141
1142 assert(svga_have_vgpu10(svga));
1143 ret = get_query_result_vgpu10(svga, sq, wait,
1144 (void *)&sResult, sizeof sResult);
1145 *result = (uint64_t)sResult.numPrimitivesRequired;
1146 break;
1147 }
1148 case PIPE_QUERY_PRIMITIVES_EMITTED: {
1149 SVGADXStreamOutStatisticsQueryResult sResult;
1150
1151 assert(svga_have_vgpu10(svga));
1152 ret = get_query_result_vgpu10(svga, sq, wait,
1153 (void *)&sResult, sizeof sResult);
1154 *result = (uint64_t)sResult.numPrimitivesWritten;
1155 break;
1156 }
1157 /* These are per-frame counters */
1158 case SVGA_QUERY_NUM_DRAW_CALLS:
1159 case SVGA_QUERY_NUM_FALLBACKS:
1160 case SVGA_QUERY_NUM_FLUSHES:
1161 case SVGA_QUERY_NUM_VALIDATIONS:
1162 case SVGA_QUERY_MAP_BUFFER_TIME:
1163 case SVGA_QUERY_NUM_BUFFERS_MAPPED:
1164 case SVGA_QUERY_NUM_TEXTURES_MAPPED:
1165 case SVGA_QUERY_NUM_BYTES_UPLOADED:
1166 case SVGA_QUERY_COMMAND_BUFFER_SIZE:
1167 case SVGA_QUERY_FLUSH_TIME:
1168 case SVGA_QUERY_SURFACE_WRITE_FLUSHES:
1169 case SVGA_QUERY_NUM_READBACKS:
1170 case SVGA_QUERY_NUM_RESOURCE_UPDATES:
1171 case SVGA_QUERY_NUM_BUFFER_UPLOADS:
1172 case SVGA_QUERY_NUM_CONST_BUF_UPDATES:
1173 case SVGA_QUERY_NUM_CONST_UPDATES:
1174 vresult->u64 = sq->end_count - sq->begin_count;
1175 break;
1176 /* These are running total counters */
1177 case SVGA_QUERY_MEMORY_USED:
1178 vresult->u64 = svgascreen->hud.total_resource_bytes;
1179 break;
1180 case SVGA_QUERY_NUM_SHADERS:
1181 vresult->u64 = svga->hud.num_shaders;
1182 break;
1183 case SVGA_QUERY_NUM_RESOURCES:
1184 vresult->u64 = svgascreen->hud.num_resources;
1185 break;
1186 case SVGA_QUERY_NUM_STATE_OBJECTS:
1187 vresult->u64 = (svga->hud.num_blend_objects +
1188 svga->hud.num_depthstencil_objects +
1189 svga->hud.num_rasterizer_objects +
1190 svga->hud.num_sampler_objects +
1191 svga->hud.num_samplerview_objects +
1192 svga->hud.num_vertexelement_objects);
1193 break;
1194 case SVGA_QUERY_NUM_SURFACE_VIEWS:
1195 vresult->u64 = svga->hud.num_surface_views;
1196 break;
1197 case SVGA_QUERY_NUM_GENERATE_MIPMAP:
1198 vresult->u64 = svga->hud.num_generate_mipmap;
1199 break;
1200 case SVGA_QUERY_NUM_FAILED_ALLOCATIONS:
1201 vresult->u64 = svgascreen->hud.num_failed_allocations;
1202 break;
1203 case SVGA_QUERY_NUM_COMMANDS_PER_DRAW:
1204 vresult->f = (float) svga->swc->num_commands
1205 / (float) svga->swc->num_draw_commands;
1206 break;
1207 default:
1208 assert(!"unexpected query type in svga_get_query_result");
1209 }
1210
1211 SVGA_DBG(DEBUG_QUERY, "%s result %d\n", __FUNCTION__, *((uint64_t *)vresult));
1212
1213 return ret;
1214 }
1215
1216 static void
svga_render_condition(struct pipe_context * pipe,struct pipe_query * q,boolean condition,enum pipe_render_cond_flag mode)1217 svga_render_condition(struct pipe_context *pipe, struct pipe_query *q,
1218 boolean condition, enum pipe_render_cond_flag mode)
1219 {
1220 struct svga_context *svga = svga_context(pipe);
1221 struct svga_winsys_screen *sws = svga_screen(svga->pipe.screen)->sws;
1222 struct svga_query *sq = svga_query(q);
1223 SVGA3dQueryId queryId;
1224 enum pipe_error ret;
1225
1226 SVGA_DBG(DEBUG_QUERY, "%s\n", __FUNCTION__);
1227
1228 assert(svga_have_vgpu10(svga));
1229 if (sq == NULL) {
1230 queryId = SVGA3D_INVALID_ID;
1231 }
1232 else {
1233 assert(sq->svga_type == SVGA3D_QUERYTYPE_OCCLUSION ||
1234 sq->svga_type == SVGA3D_QUERYTYPE_OCCLUSIONPREDICATE);
1235
1236 if (sq->svga_type == SVGA3D_QUERYTYPE_OCCLUSION) {
1237 assert(sq->predicate);
1238 /**
1239 * For conditional rendering, make sure to use the associated
1240 * predicate query.
1241 */
1242 sq = svga_query(sq->predicate);
1243 }
1244 queryId = sq->id;
1245
1246 if ((mode == PIPE_RENDER_COND_WAIT ||
1247 mode == PIPE_RENDER_COND_BY_REGION_WAIT) && sq->fence) {
1248 sws->fence_finish(sws, sq->fence, PIPE_TIMEOUT_INFINITE,
1249 SVGA_FENCE_FLAG_QUERY);
1250 }
1251 }
1252 /*
1253 * if the kernel module doesn't support the predication command,
1254 * we'll just render unconditionally.
1255 * This is probably acceptable for the typical case of occlusion culling.
1256 */
1257 if (sws->have_set_predication_cmd) {
1258 ret = SVGA3D_vgpu10_SetPredication(svga->swc, queryId,
1259 (uint32) condition);
1260 if (ret != PIPE_OK) {
1261 svga_context_flush(svga, NULL);
1262 ret = SVGA3D_vgpu10_SetPredication(svga->swc, queryId,
1263 (uint32) condition);
1264 }
1265 svga->pred.query_id = queryId;
1266 svga->pred.cond = condition;
1267 }
1268
1269 svga->render_condition = (sq != NULL);
1270 }
1271
1272
1273 /*
1274 * This function is a workaround because we lack the ability to query
1275 * renderer's time synchornously.
1276 */
1277 static uint64_t
svga_get_timestamp(struct pipe_context * pipe)1278 svga_get_timestamp(struct pipe_context *pipe)
1279 {
1280 struct pipe_query *q = svga_create_query(pipe, PIPE_QUERY_TIMESTAMP, 0);
1281 union pipe_query_result result;
1282
1283 svga_begin_query(pipe, q);
1284 svga_end_query(pipe,q);
1285 svga_get_query_result(pipe, q, TRUE, &result);
1286 svga_destroy_query(pipe, q);
1287
1288 return result.u64;
1289 }
1290
1291
1292 static void
svga_set_active_query_state(struct pipe_context * pipe,boolean enable)1293 svga_set_active_query_state(struct pipe_context *pipe, boolean enable)
1294 {
1295 }
1296
1297
1298 /**
1299 * \brief Toggle conditional rendering if already enabled
1300 *
1301 * \param svga[in] The svga context
1302 * \param render_condition_enabled[in] Whether to ignore requests to turn
1303 * conditional rendering off
1304 * \param on[in] Whether to turn conditional rendering on or off
1305 */
1306 void
svga_toggle_render_condition(struct svga_context * svga,boolean render_condition_enabled,boolean on)1307 svga_toggle_render_condition(struct svga_context *svga,
1308 boolean render_condition_enabled,
1309 boolean on)
1310 {
1311 SVGA3dQueryId query_id;
1312 enum pipe_error ret;
1313
1314 if (render_condition_enabled ||
1315 svga->pred.query_id == SVGA3D_INVALID_ID) {
1316 return;
1317 }
1318
1319 /*
1320 * If we get here, it means that the system supports
1321 * conditional rendering since svga->pred.query_id has already been
1322 * modified for this context and thus support has already been
1323 * verified.
1324 */
1325 query_id = on ? svga->pred.query_id : SVGA3D_INVALID_ID;
1326
1327 ret = SVGA3D_vgpu10_SetPredication(svga->swc, query_id,
1328 (uint32) svga->pred.cond);
1329 if (ret == PIPE_ERROR_OUT_OF_MEMORY) {
1330 svga_context_flush(svga, NULL);
1331 ret = SVGA3D_vgpu10_SetPredication(svga->swc, query_id,
1332 (uint32) svga->pred.cond);
1333 assert(ret == PIPE_OK);
1334 }
1335 }
1336
1337
1338 void
svga_init_query_functions(struct svga_context * svga)1339 svga_init_query_functions(struct svga_context *svga)
1340 {
1341 svga->pipe.create_query = svga_create_query;
1342 svga->pipe.destroy_query = svga_destroy_query;
1343 svga->pipe.begin_query = svga_begin_query;
1344 svga->pipe.end_query = svga_end_query;
1345 svga->pipe.get_query_result = svga_get_query_result;
1346 svga->pipe.set_active_query_state = svga_set_active_query_state;
1347 svga->pipe.render_condition = svga_render_condition;
1348 svga->pipe.get_timestamp = svga_get_timestamp;
1349 }
1350