aboutsummaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers/svga/svga_screen_cache.c
blob: 1916a64245c09c2a3edf9d675ccf86fbf0a6c060 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
/**********************************************************
 * Copyright 2008-2009 VMware, Inc.  All rights reserved.
 *
 * 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 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.
 *
 **********************************************************/

#include "util/u_math.h"
#include "util/u_memory.h"
#include "util/crc32.h"

#include "svga_debug.h"
#include "svga_format.h"
#include "svga_winsys.h"
#include "svga_screen.h"
#include "svga_screen_cache.h"
#include "svga_context.h"
#include "svga_cmd.h"

#define SVGA_SURFACE_CACHE_ENABLED 1


/**
 * Return the size of the surface described by the key (in bytes).
 */
static unsigned
surface_size(const struct svga_host_surface_cache_key *key)
{
   unsigned bw, bh, bpb, total_size, i;

   assert(key->numMipLevels > 0);
   assert(key->numFaces > 0);
   assert(key->arraySize > 0);

   if (key->format == SVGA3D_BUFFER) {
      /* Special case: we don't want to count vertex/index buffers
       * against the cache size limit, so view them as zero-sized.
       */
      return 0;
   }

   svga_format_size(key->format, &bw, &bh, &bpb);

   total_size = 0;

   for (i = 0; i < key->numMipLevels; i++) {
      unsigned w = u_minify(key->size.width, i);
      unsigned h = u_minify(key->size.height, i);
      unsigned d = u_minify(key->size.depth, i);
      unsigned img_size = ((w + bw - 1) / bw) * ((h + bh - 1) / bh) * d * bpb;
      total_size += img_size;
   }

   total_size *= key->numFaces * key->arraySize * MAX2(1, key->sampleCount);

   return total_size;
}


/**
 * Compute the bucket for this key.
 */
static inline unsigned
svga_screen_cache_bucket(const struct svga_host_surface_cache_key *key)
{
   return util_hash_crc32(key, sizeof *key) % SVGA_HOST_SURFACE_CACHE_BUCKETS;
}


/**
 * Search the cache for a surface that matches the key.  If a match is
 * found, remove it from the cache and return the surface pointer.
 * Return NULL otherwise.
 */
static struct svga_winsys_surface *
svga_screen_cache_lookup(struct svga_screen *svgascreen,
                         const struct svga_host_surface_cache_key *key)
{
   struct svga_host_surface_cache *cache = &svgascreen->cache;
   struct svga_winsys_screen *sws = svgascreen->sws;
   struct svga_host_surface_cache_entry *entry;
   struct svga_winsys_surface *handle = NULL;
   struct list_head *curr, *next;
   unsigned bucket;
   unsigned tries = 0;

   assert(key->cachable);

   bucket = svga_screen_cache_bucket(key);

   mtx_lock(&cache->mutex);

   curr = cache->bucket[bucket].next;
   next = curr->next;
   while (curr != &cache->bucket[bucket]) {
      ++tries;

      entry = LIST_ENTRY(struct svga_host_surface_cache_entry, curr, bucket_head);

      assert(entry->handle);

      /* If the key matches and the fence is signalled (the surface is no
       * longer needed) the lookup was successful.  We found a surface that
       * can be reused.
       * We unlink the surface from the cache entry and we add the entry to
       * the 'empty' list.
       */
      if (memcmp(&entry->key, key, sizeof *key) == 0 &&
          sws->fence_signalled(sws, entry->fence, 0) == 0) {
         unsigned surf_size;

         assert(sws->surface_is_flushed(sws, entry->handle));

         handle = entry->handle; /* Reference is transfered here. */
         entry->handle = NULL;

         /* Remove from hash table */
         LIST_DEL(&entry->bucket_head);

         /* remove from LRU list */
         LIST_DEL(&entry->head);

         /* Add the cache entry (but not the surface!) to the empty list */
         LIST_ADD(&entry->head, &cache->empty);

         /* update the cache size */
         surf_size = surface_size(&entry->key);
         assert(surf_size <= cache->total_size);
         if (surf_size > cache->total_size)
            cache->total_size = 0; /* should never happen, but be safe */
         else
            cache->total_size -= surf_size;

         break;
      }

      curr = next;
      next = curr->next;
   }

   mtx_unlock(&cache->mutex);

   if (SVGA_DEBUG & DEBUG_DMA)
      debug_printf("%s: cache %s after %u tries (bucket %d)\n", __FUNCTION__,
                   handle ? "hit" : "miss", tries, bucket);

   return handle;
}


/**
 * Free the least recently used entries in the surface cache until the
 * cache size is <= the target size OR there are no unused entries left
 * to discard.  We don't do any flushing to try to free up additional
 * surfaces.
 */
static void
svga_screen_cache_shrink(struct svga_screen *svgascreen,
                         unsigned target_size)
{
   struct svga_host_surface_cache *cache = &svgascreen->cache;
   struct svga_winsys_screen *sws = svgascreen->sws;
   struct svga_host_surface_cache_entry *entry = NULL, *next_entry;

   /* Walk over the list of unused buffers in reverse order: from oldest
    * to newest.
    */
   LIST_FOR_EACH_ENTRY_SAFE_REV(entry, next_entry, &cache->unused, head) {
      if (entry->key.format != SVGA3D_BUFFER) {
         /* we don't want to discard vertex/index buffers */

         cache->total_size -= surface_size(&entry->key);

         assert(entry->handle);
         sws->surface_reference(sws, &entry->handle, NULL);

         LIST_DEL(&entry->bucket_head);
         LIST_DEL(&entry->head);
         LIST_ADD(&entry->head, &cache->empty);

         if (cache->total_size <= target_size) {
            /* all done */
            break;
         }
      }
   }
}


/**
 * Add a surface to the cache.  This is done when the driver deletes
 * the surface.  Note: transfers a handle reference.
 */
static void
svga_screen_cache_add(struct svga_screen *svgascreen,
                      const struct svga_host_surface_cache_key *key,
                      struct svga_winsys_surface **p_handle)
{
   struct svga_host_surface_cache *cache = &svgascreen->cache;
   struct svga_winsys_screen *sws = svgascreen->sws;
   struct svga_host_surface_cache_entry *entry = NULL;
   struct svga_winsys_surface *handle = *p_handle;
   unsigned surf_size;

   assert(key->cachable);

   if (!handle)
      return;

   surf_size = surface_size(key);

   *p_handle = NULL;
   mtx_lock(&cache->mutex);

   if (surf_size >= SVGA_HOST_SURFACE_CACHE_BYTES) {
      /* this surface is too large to cache, just free it */
      sws->surface_reference(sws, &handle, NULL);
      mtx_unlock(&cache->mutex);
      return;
   }

   if (cache->total_size + surf_size > SVGA_HOST_SURFACE_CACHE_BYTES) {
      /* Adding this surface would exceed the cache size.
       * Try to discard least recently used entries until we hit the
       * new target cache size.
       */
      unsigned target_size = SVGA_HOST_SURFACE_CACHE_BYTES - surf_size;

      svga_screen_cache_shrink(svgascreen, target_size);

      if (cache->total_size > target_size) {
         /* we weren't able to shrink the cache as much as we wanted so
          * just discard this surface.
          */
         sws->surface_reference(sws, &handle, NULL);
         mtx_unlock(&cache->mutex);
         return;
      }
   }

   if (!LIST_IS_EMPTY(&cache->empty)) {
      /* An empty entry has no surface associated with it.
       * Use the first empty entry.
       */
      entry = LIST_ENTRY(struct svga_host_surface_cache_entry,
                         cache->empty.next, head);

      /* Remove from LRU list */
      LIST_DEL(&entry->head);
   }
   else if (!LIST_IS_EMPTY(&cache->unused)) {
      /* free the last used buffer and reuse its entry */
      entry = LIST_ENTRY(struct svga_host_surface_cache_entry,
                         cache->unused.prev, head);
      SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
               "unref sid %p (make space)\n", entry->handle);

      cache->total_size -= surface_size(&entry->key);

      sws->surface_reference(sws, &entry->handle, NULL);

      /* Remove from hash table */
      LIST_DEL(&entry->bucket_head);

      /* Remove from LRU list */
      LIST_DEL(&entry->head);
   }

   if (entry) {
      assert(entry->handle == NULL);
      entry->handle = handle;
      memcpy(&entry->key, key, sizeof entry->key);

      SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
               "cache sid %p\n", entry->handle);

      /* If we don't have gb objects, we don't need to invalidate. */
      if (sws->have_gb_objects)
         LIST_ADD(&entry->head, &cache->validated);
      else
         LIST_ADD(&entry->head, &cache->invalidated);

      cache->total_size += surf_size;
   }
   else {
      /* Couldn't cache the buffer -- this really shouldn't happen */
      SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
               "unref sid %p (couldn't find space)\n", handle);
      sws->surface_reference(sws, &handle, NULL);
   }

   mtx_unlock(&cache->mutex);
}


/**
 * Called during the screen flush to move all buffers not in a validate list
 * into the unused list.
 */
void
svga_screen_cache_flush(struct svga_screen *svgascreen,
                        struct svga_context *svga,
                        struct pipe_fence_handle *fence)
{
   struct svga_host_surface_cache *cache = &svgascreen->cache;
   struct svga_winsys_screen *sws = svgascreen->sws;
   struct svga_host_surface_cache_entry *entry;
   struct list_head *curr, *next;
   unsigned bucket;

   mtx_lock(&cache->mutex);

   /* Loop over entries in the invalidated list */
   curr = cache->invalidated.next;
   next = curr->next;
   while (curr != &cache->invalidated) {
      entry = LIST_ENTRY(struct svga_host_surface_cache_entry, curr, head);

      assert(entry->handle);

      if (sws->surface_is_flushed(sws, entry->handle)) {
         /* remove entry from the invalidated list */
         LIST_DEL(&entry->head);

         sws->fence_reference(sws, &entry->fence, fence);

         /* Add entry to the unused list */
         LIST_ADD(&entry->head, &cache->unused);

         /* Add entry to the hash table bucket */
         bucket = svga_screen_cache_bucket(&entry->key);
         LIST_ADD(&entry->bucket_head, &cache->bucket[bucket]);
      }

      curr = next;
      next = curr->next;
   }

   curr = cache->validated.next;
   next = curr->next;
   while (curr != &cache->validated) {
      entry = LIST_ENTRY(struct svga_host_surface_cache_entry, curr, head);

      assert(entry->handle);
      assert(svga_have_gb_objects(svga));

      if (sws->surface_is_flushed(sws, entry->handle)) {
         /* remove entry from the validated list */
         LIST_DEL(&entry->head);

         /* It is now safe to invalidate the surface content.
          * It will be done using the current context.
          */
         if (SVGA3D_InvalidateGBSurface(svga->swc, entry->handle) != PIPE_OK) {
            MAYBE_UNUSED enum pipe_error ret;

            /* Even though surface invalidation here is done after the command
             * buffer is flushed, it is still possible that it will
             * fail because there might be just enough of this command that is
             * filling up the command buffer, so in this case we will call
             * the winsys flush directly to flush the buffer.
             * Note, we don't want to call svga_context_flush() here because
             * this function itself is called inside svga_context_flush().
             */
            svga->swc->flush(svga->swc, NULL);
            ret = SVGA3D_InvalidateGBSurface(svga->swc, entry->handle);
            assert(ret == PIPE_OK);
         }

         /* add the entry to the invalidated list */
         LIST_ADD(&entry->head, &cache->invalidated);
      }

      curr = next;
      next = curr->next;
   }

   mtx_unlock(&cache->mutex);
}


/**
 * Free all the surfaces in the cache.
 * Called when destroying the svga screen object.
 */
void
svga_screen_cache_cleanup(struct svga_screen *svgascreen)
{
   struct svga_host_surface_cache *cache = &svgascreen->cache;
   struct svga_winsys_screen *sws = svgascreen->sws;
   unsigned i;

   for (i = 0; i < SVGA_HOST_SURFACE_CACHE_SIZE; ++i) {
      if (cache->entries[i].handle) {
	 SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
                  "unref sid %p (shutdown)\n", cache->entries[i].handle);
	 sws->surface_reference(sws, &cache->entries[i].handle, NULL);

         cache->total_size -= surface_size(&cache->entries[i].key);
      }

      if (cache->entries[i].fence)
         sws->fence_reference(sws, &cache->entries[i].fence, NULL);
   }

   mtx_destroy(&cache->mutex);
}


enum pipe_error
svga_screen_cache_init(struct svga_screen *svgascreen)
{
   struct svga_host_surface_cache *cache = &svgascreen->cache;
   unsigned i;

   assert(cache->total_size == 0);

   (void) mtx_init(&cache->mutex, mtx_plain);

   for (i = 0; i < SVGA_HOST_SURFACE_CACHE_BUCKETS; ++i)
      LIST_INITHEAD(&cache->bucket[i]);

   LIST_INITHEAD(&cache->unused);

   LIST_INITHEAD(&cache->validated);

   LIST_INITHEAD(&cache->invalidated);

   LIST_INITHEAD(&cache->empty);
   for (i = 0; i < SVGA_HOST_SURFACE_CACHE_SIZE; ++i)
      LIST_ADDTAIL(&cache->entries[i].head, &cache->empty);

   return PIPE_OK;
}


/**
 * Allocate a new host-side surface.  If the surface is marked as cachable,
 * first try re-using a surface in the cache of freed surfaces.  Otherwise,
 * allocate a new surface.
 * \param bind_flags  bitmask of PIPE_BIND_x flags
 * \param usage  one of PIPE_USAGE_x values
 * \param validated return True if the surface is a reused surface
 */
struct svga_winsys_surface *
svga_screen_surface_create(struct svga_screen *svgascreen,
                           unsigned bind_flags, enum pipe_resource_usage usage,
                           boolean *validated,
                           struct svga_host_surface_cache_key *key)
{
   struct svga_winsys_screen *sws = svgascreen->sws;
   struct svga_winsys_surface *handle = NULL;
   boolean cachable = SVGA_SURFACE_CACHE_ENABLED && key->cachable;

   SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
            "%s sz %dx%dx%d mips %d faces %d arraySize %d cachable %d\n",
            __FUNCTION__,
            key->size.width,
            key->size.height,
            key->size.depth,
            key->numMipLevels,
            key->numFaces,
            key->arraySize,
            key->cachable);

   if (cachable) {
      /* Try to re-cycle a previously freed, cached surface */
      if (key->format == SVGA3D_BUFFER) {
         SVGA3dSurfaceAllFlags hint_flag;

         /* For buffers, round the buffer size up to the nearest power
          * of two to increase the probability of cache hits.  Keep
          * texture surface dimensions unchanged.
          */
         uint32_t size = 1;
         while (size < key->size.width)
            size <<= 1;
         key->size.width = size;

         /* Determine whether the buffer is static or dynamic.
          * This is a bit of a heuristic which can be tuned as needed.
          */
         if (usage == PIPE_USAGE_DEFAULT ||
             usage == PIPE_USAGE_IMMUTABLE) {
            hint_flag = SVGA3D_SURFACE_HINT_STATIC;
         }
         else if (bind_flags & PIPE_BIND_INDEX_BUFFER) {
            /* Index buffers don't change too often.  Mark them as static.
             */
            hint_flag = SVGA3D_SURFACE_HINT_STATIC;
         }
         else {
            /* Since we're reusing buffers we're effectively transforming all
             * of them into dynamic buffers.
             *
             * It would be nice to not cache long lived static buffers. But there
             * is no way to detect the long lived from short lived ones yet. A
             * good heuristic would be buffer size.
             */
            hint_flag = SVGA3D_SURFACE_HINT_DYNAMIC;
         }

         key->flags &= ~(SVGA3D_SURFACE_HINT_STATIC |
                         SVGA3D_SURFACE_HINT_DYNAMIC);
         key->flags |= hint_flag;
      }

      handle = svga_screen_cache_lookup(svgascreen, key);
      if (handle) {
         if (key->format == SVGA3D_BUFFER)
            SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
                     "reuse sid %p sz %d (buffer)\n", handle,
                     key->size.width);
         else
            SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
                     "reuse sid %p sz %dx%dx%d mips %d faces %d arraySize %d\n", handle,
                     key->size.width,
                     key->size.height,
                     key->size.depth,
                     key->numMipLevels,
                     key->numFaces,
                     key->arraySize);
         *validated = TRUE;
      }
   }

   if (!handle) {
      /* Unable to recycle surface, allocate a new one */
      unsigned usage = 0;

      if (!key->cachable)
         usage |= SVGA_SURFACE_USAGE_SHARED;
      if (key->scanout)
         usage |= SVGA_SURFACE_USAGE_SCANOUT;
      if (key->coherent)
         usage |= SVGA_SURFACE_USAGE_COHERENT;

      handle = sws->surface_create(sws,
                                   key->flags,
                                   key->format,
                                   usage,
                                   key->size,
                                   key->numFaces * key->arraySize,
                                   key->numMipLevels,
                                   key->sampleCount);
      if (handle)
         SVGA_DBG(DEBUG_CACHE|DEBUG_DMA,
                  "  CREATE sid %p sz %dx%dx%d\n",
                  handle,
                  key->size.width,
                  key->size.height,
                  key->size.depth);

      *validated = FALSE;
   }

   return handle;
}


/**
 * Release a surface.  We don't actually free the surface- we put
 * it into the cache of freed surfaces (if it's cachable).
 */
void
svga_screen_surface_destroy(struct svga_screen *svgascreen,
                            const struct svga_host_surface_cache_key *key,
                            struct svga_winsys_surface **p_handle)
{
   struct svga_winsys_screen *sws = svgascreen->sws;

   /* We only set the cachable flag for surfaces of which we are the
    * exclusive owner.  So just hold onto our existing reference in
    * that case.
    */
   if (SVGA_SURFACE_CACHE_ENABLED && key->cachable) {
      svga_screen_cache_add(svgascreen, key, p_handle);
   }
   else {
      SVGA_DBG(DEBUG_DMA,
               "unref sid %p (uncachable)\n", *p_handle);
      sws->surface_reference(sws, p_handle, NULL);
   }
}


/**
 * Print/dump the contents of the screen cache.  For debugging.
 */
void
svga_screen_cache_dump(const struct svga_screen *svgascreen)
{
   const struct svga_host_surface_cache *cache = &svgascreen->cache;
   unsigned bucket;
   unsigned count = 0;

   debug_printf("svga3d surface cache:\n");
   for (bucket = 0; bucket < SVGA_HOST_SURFACE_CACHE_BUCKETS; bucket++) {
      struct list_head *curr;
      curr = cache->bucket[bucket].next;
      while (curr && curr != &cache->bucket[bucket]) {
         struct svga_host_surface_cache_entry *entry =
            LIST_ENTRY(struct svga_host_surface_cache_entry,
                       curr, bucket_head);
         if (entry->key.format == SVGA3D_BUFFER) {
            debug_printf("  %p: buffer %u bytes\n",
                         entry->handle,
                         entry->key.size.width);
         }
         else {
            debug_printf("  %p: %u x %u x %u format %u\n",
                         entry->handle,
                         entry->key.size.width,
                         entry->key.size.height,
                         entry->key.size.depth,
                         entry->key.format);
         }
         curr = curr->next;
         count++;
      }
   }

   debug_printf("%u surfaces, %u bytes\n", count, cache->total_size);
}