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
|
/**********************************************************
* 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 "svga_cmd.h"
#include "pipe/p_state.h"
#include "pipe/p_defines.h"
#include "util/u_inlines.h"
#include "os/os_thread.h"
#include "util/u_math.h"
#include "util/u_memory.h"
#include "util/u_resource.h"
#include "svga_context.h"
#include "svga_screen.h"
#include "svga_resource_buffer.h"
#include "svga_resource_buffer_upload.h"
#include "svga_winsys.h"
#include "svga_debug.h"
/**
* Determine what buffers eventually need hardware backing.
*
* Vertex- and index buffers need hardware backing. Constant buffers
* do on vgpu10. Staging texture-upload buffers do when they are
* supported.
*/
static inline boolean
svga_buffer_needs_hw_storage(const struct svga_screen *ss,
const struct pipe_resource *template)
{
unsigned bind_mask = (PIPE_BIND_VERTEX_BUFFER | PIPE_BIND_INDEX_BUFFER |
PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_STREAM_OUTPUT);
if (ss->sws->have_vgpu10) {
/*
* Driver-created upload const0- and staging texture upload buffers
* tagged with PIPE_BIND_CUSTOM
*/
bind_mask |= PIPE_BIND_CUSTOM;
/* Uniform buffer objects.
* Make sure we don't create hardware storage for state-tracker
* const0 buffers, because we frequently map them for reading.
* They are distinguished by having PIPE_USAGE_STREAM, but not
* PIPE_BIND_CUSTOM.
*/
if (template->usage != PIPE_USAGE_STREAM)
bind_mask |= PIPE_BIND_CONSTANT_BUFFER;
}
if (template->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT)
return TRUE;
return !!(template->bind & bind_mask);
}
/**
* Create a buffer transfer.
*
* Unlike texture DMAs (which are written immediately to the command buffer and
* therefore inherently serialized with other context operations), for buffers
* we try to coalesce multiple range mappings (i.e, multiple calls to this
* function) into a single DMA command, for better efficiency in command
* processing. This means we need to exercise extra care here to ensure that
* the end result is exactly the same as if one DMA was used for every mapped
* range.
*/
static void *
svga_buffer_transfer_map(struct pipe_context *pipe,
struct pipe_resource *resource,
unsigned level,
unsigned usage,
const struct pipe_box *box,
struct pipe_transfer **ptransfer)
{
struct svga_context *svga = svga_context(pipe);
struct svga_screen *ss = svga_screen(pipe->screen);
struct svga_buffer *sbuf = svga_buffer(resource);
struct pipe_transfer *transfer;
uint8_t *map = NULL;
int64_t begin = svga_get_time(svga);
SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_BUFFERTRANSFERMAP);
assert(box->y == 0);
assert(box->z == 0);
assert(box->height == 1);
assert(box->depth == 1);
transfer = MALLOC_STRUCT(pipe_transfer);
if (!transfer) {
goto done;
}
transfer->resource = resource;
transfer->level = level;
transfer->usage = usage;
transfer->box = *box;
transfer->stride = 0;
transfer->layer_stride = 0;
if (usage & PIPE_TRANSFER_WRITE) {
/* If we write to the buffer for any reason, free any saved translated
* vertices.
*/
pipe_resource_reference(&sbuf->translated_indices.buffer, NULL);
}
if ((usage & PIPE_TRANSFER_READ) && sbuf->dirty &&
!sbuf->key.coherent && !svga->swc->force_coherent) {
enum pipe_error ret;
/* Host-side buffers can only be dirtied with vgpu10 features
* (streamout and buffer copy).
*/
assert(svga_have_vgpu10(svga));
if (!sbuf->user) {
(void) svga_buffer_handle(svga, resource, sbuf->bind_flags);
}
if (sbuf->dma.pending) {
svga_buffer_upload_flush(svga, sbuf);
svga_context_finish(svga);
}
assert(sbuf->handle);
ret = SVGA3D_vgpu10_ReadbackSubResource(svga->swc, sbuf->handle, 0);
if (ret != PIPE_OK) {
svga_context_flush(svga, NULL);
ret = SVGA3D_vgpu10_ReadbackSubResource(svga->swc, sbuf->handle, 0);
assert(ret == PIPE_OK);
}
svga->hud.num_readbacks++;
svga_context_finish(svga);
sbuf->dirty = FALSE;
}
if (usage & PIPE_TRANSFER_WRITE) {
if ((usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE) &&
!(resource->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT)) {
/*
* Flush any pending primitives, finish writing any pending DMA
* commands, and tell the host to discard the buffer contents on
* the next DMA operation.
*/
svga_hwtnl_flush_buffer(svga, resource);
if (sbuf->dma.pending) {
svga_buffer_upload_flush(svga, sbuf);
/*
* Instead of flushing the context command buffer, simply discard
* the current hwbuf, and start a new one.
* With GB objects, the map operation takes care of this
* if passed the PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE flag,
* and the old backing store is busy.
*/
if (!svga_have_gb_objects(svga))
svga_buffer_destroy_hw_storage(ss, sbuf);
}
sbuf->map.num_ranges = 0;
sbuf->dma.flags.discard = TRUE;
}
if (usage & PIPE_TRANSFER_UNSYNCHRONIZED) {
if (!sbuf->map.num_ranges) {
/*
* No pending ranges to upload so far, so we can tell the host to
* not synchronize on the next DMA command.
*/
sbuf->dma.flags.unsynchronized = TRUE;
}
} else {
/*
* Synchronizing, so flush any pending primitives, finish writing any
* pending DMA command, and ensure the next DMA will be done in order.
*/
svga_hwtnl_flush_buffer(svga, resource);
if (sbuf->dma.pending) {
svga_buffer_upload_flush(svga, sbuf);
if (svga_buffer_has_hw_storage(sbuf)) {
/*
* We have a pending DMA upload from a hardware buffer, therefore
* we need to ensure that the host finishes processing that DMA
* command before the state tracker can start overwriting the
* hardware buffer.
*
* XXX: This could be avoided by tying the hardware buffer to
* the transfer (just as done with textures), which would allow
* overlapping DMAs commands to be queued on the same context
* buffer. However, due to the likelihood of software vertex
* processing, it is more convenient to hold on to the hardware
* buffer, allowing to quickly access the contents from the CPU
* without having to do a DMA download from the host.
*/
if (usage & PIPE_TRANSFER_DONTBLOCK) {
/*
* Flushing the command buffer here will most likely cause
* the map of the hwbuf below to block, so preemptively
* return NULL here if DONTBLOCK is set to prevent unnecessary
* command buffer flushes.
*/
FREE(transfer);
goto done;
}
svga_context_flush(svga, NULL);
}
}
sbuf->dma.flags.unsynchronized = FALSE;
}
}
if (!sbuf->swbuf && !svga_buffer_has_hw_storage(sbuf)) {
if (svga_buffer_create_hw_storage(ss, sbuf, sbuf->bind_flags) != PIPE_OK) {
/*
* We can't create a hardware buffer big enough, so create a malloc
* buffer instead.
*/
if (0) {
debug_printf("%s: failed to allocate %u KB of DMA, "
"splitting DMA transfers\n",
__FUNCTION__,
(sbuf->b.b.width0 + 1023)/1024);
}
sbuf->swbuf = align_malloc(sbuf->b.b.width0, 16);
if (!sbuf->swbuf) {
FREE(transfer);
goto done;
}
}
}
if (sbuf->swbuf) {
/* User/malloc buffer */
map = sbuf->swbuf;
}
else if (svga_buffer_has_hw_storage(sbuf)) {
boolean retry;
map = svga_buffer_hw_storage_map(svga, sbuf, transfer->usage, &retry);
if (map == NULL && retry) {
/*
* At this point, svga_buffer_get_transfer() has already
* hit the DISCARD_WHOLE_RESOURCE path and flushed HWTNL
* for this buffer.
*/
svga_context_flush(svga, NULL);
map = svga_buffer_hw_storage_map(svga, sbuf, transfer->usage, &retry);
}
}
else {
map = NULL;
}
if (map) {
++sbuf->map.count;
map += transfer->box.x;
*ptransfer = transfer;
} else {
FREE(transfer);
}
svga->hud.map_buffer_time += (svga_get_time(svga) - begin);
done:
SVGA_STATS_TIME_POP(svga_sws(svga));
return map;
}
static void
svga_buffer_transfer_flush_region(struct pipe_context *pipe,
struct pipe_transfer *transfer,
const struct pipe_box *box)
{
struct svga_screen *ss = svga_screen(pipe->screen);
struct svga_buffer *sbuf = svga_buffer(transfer->resource);
struct svga_context *svga = svga_context(pipe);
unsigned offset = transfer->box.x + box->x;
unsigned length = box->width;
assert(transfer->usage & PIPE_TRANSFER_WRITE);
assert(transfer->usage & PIPE_TRANSFER_FLUSH_EXPLICIT);
if (!(svga->swc->force_coherent || sbuf->key.coherent) || sbuf->swbuf) {
mtx_lock(&ss->swc_mutex);
svga_buffer_add_range(sbuf, offset, offset + length);
mtx_unlock(&ss->swc_mutex);
}
}
static void
svga_buffer_transfer_unmap(struct pipe_context *pipe,
struct pipe_transfer *transfer)
{
struct svga_screen *ss = svga_screen(pipe->screen);
struct svga_context *svga = svga_context(pipe);
struct svga_buffer *sbuf = svga_buffer(transfer->resource);
SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_BUFFERTRANSFERUNMAP);
mtx_lock(&ss->swc_mutex);
assert(sbuf->map.count);
if (sbuf->map.count) {
--sbuf->map.count;
}
if (svga_buffer_has_hw_storage(sbuf)) {
/* Note: we may wind up flushing here and unmapping other buffers
* which leads to recursively locking ss->swc_mutex.
*/
svga_buffer_hw_storage_unmap(svga, sbuf);
}
if (transfer->usage & PIPE_TRANSFER_WRITE) {
if (!(transfer->usage & PIPE_TRANSFER_FLUSH_EXPLICIT)) {
/*
* Mapped range not flushed explicitly, so flush the whole buffer,
* and tell the host to discard the contents when processing the DMA
* command.
*/
SVGA_DBG(DEBUG_DMA, "flushing the whole buffer\n");
sbuf->dma.flags.discard = TRUE;
if (!(svga->swc->force_coherent || sbuf->key.coherent) || sbuf->swbuf)
svga_buffer_add_range(sbuf, 0, sbuf->b.b.width0);
}
}
mtx_unlock(&ss->swc_mutex);
FREE(transfer);
SVGA_STATS_TIME_POP(svga_sws(svga));
}
static void
svga_buffer_destroy(struct pipe_screen *screen,
struct pipe_resource *buf)
{
struct svga_screen *ss = svga_screen(screen);
struct svga_buffer *sbuf = svga_buffer(buf);
assert(!p_atomic_read(&buf->reference.count));
assert(!sbuf->dma.pending);
if (sbuf->handle)
svga_buffer_destroy_host_surface(ss, sbuf);
if (sbuf->uploaded.buffer)
pipe_resource_reference(&sbuf->uploaded.buffer, NULL);
if (sbuf->hwbuf)
svga_buffer_destroy_hw_storage(ss, sbuf);
if (sbuf->swbuf && !sbuf->user)
align_free(sbuf->swbuf);
pipe_resource_reference(&sbuf->translated_indices.buffer, NULL);
ss->hud.total_resource_bytes -= sbuf->size;
assert(ss->hud.num_resources > 0);
if (ss->hud.num_resources > 0)
ss->hud.num_resources--;
FREE(sbuf);
}
struct u_resource_vtbl svga_buffer_vtbl =
{
u_default_resource_get_handle, /* get_handle */
svga_buffer_destroy, /* resource_destroy */
svga_buffer_transfer_map, /* transfer_map */
svga_buffer_transfer_flush_region, /* transfer_flush_region */
svga_buffer_transfer_unmap, /* transfer_unmap */
};
struct pipe_resource *
svga_buffer_create(struct pipe_screen *screen,
const struct pipe_resource *template)
{
struct svga_screen *ss = svga_screen(screen);
struct svga_buffer *sbuf;
unsigned bind_flags;
SVGA_STATS_TIME_PUSH(ss->sws, SVGA_STATS_TIME_CREATEBUFFER);
sbuf = CALLOC_STRUCT(svga_buffer);
if (!sbuf)
goto error1;
sbuf->b.b = *template;
sbuf->b.vtbl = &svga_buffer_vtbl;
pipe_reference_init(&sbuf->b.b.reference, 1);
sbuf->b.b.screen = screen;
bind_flags = template->bind & ~PIPE_BIND_CUSTOM;
list_inithead(&sbuf->surfaces);
if (bind_flags & PIPE_BIND_CONSTANT_BUFFER) {
/* Constant buffers can only have the PIPE_BIND_CONSTANT_BUFFER
* flag set.
*/
if (ss->sws->have_vgpu10) {
bind_flags = PIPE_BIND_CONSTANT_BUFFER;
}
}
/* Although svga device only requires constant buffer size to be
* in multiples of 16, in order to allow bind_flags promotion,
* we are mandating all buffer size to be in multiples of 16.
*/
sbuf->b.b.width0 = align(sbuf->b.b.width0, 16);
if (svga_buffer_needs_hw_storage(ss, template)) {
/* If the buffer is not used for constant buffer, set
* the vertex/index bind flags as well so that the buffer will be
* accepted for those uses.
* Note that the PIPE_BIND_ flags we get from the state tracker are
* just a hint about how the buffer may be used. And OpenGL buffer
* object may be used for many different things.
* Also note that we do not unconditionally set the streamout
* bind flag since streamout buffer is an output buffer and
* might have performance implication.
*/
if (!(template->bind & PIPE_BIND_CONSTANT_BUFFER) &&
!(template->bind & PIPE_BIND_CUSTOM)) {
/* Not a constant- or staging buffer.
* The buffer may be used for vertex data or indexes.
*/
bind_flags |= (PIPE_BIND_VERTEX_BUFFER |
PIPE_BIND_INDEX_BUFFER);
}
if (svga_buffer_create_host_surface(ss, sbuf, bind_flags) != PIPE_OK)
goto error2;
}
else {
sbuf->swbuf = align_malloc(sbuf->b.b.width0, 64);
if (!sbuf->swbuf)
goto error2;
}
debug_reference(&sbuf->b.b.reference,
(debug_reference_descriptor)debug_describe_resource, 0);
sbuf->bind_flags = bind_flags;
sbuf->size = util_resource_size(&sbuf->b.b);
ss->hud.total_resource_bytes += sbuf->size;
ss->hud.num_resources++;
SVGA_STATS_TIME_POP(ss->sws);
return &sbuf->b.b;
error2:
FREE(sbuf);
error1:
SVGA_STATS_TIME_POP(ss->sws);
return NULL;
}
struct pipe_resource *
svga_user_buffer_create(struct pipe_screen *screen,
void *ptr,
unsigned bytes,
unsigned bind)
{
struct svga_buffer *sbuf;
struct svga_screen *ss = svga_screen(screen);
sbuf = CALLOC_STRUCT(svga_buffer);
if (!sbuf)
goto no_sbuf;
pipe_reference_init(&sbuf->b.b.reference, 1);
sbuf->b.vtbl = &svga_buffer_vtbl;
sbuf->b.b.screen = screen;
sbuf->b.b.format = PIPE_FORMAT_R8_UNORM; /* ?? */
sbuf->b.b.usage = PIPE_USAGE_IMMUTABLE;
sbuf->b.b.bind = bind;
sbuf->b.b.width0 = bytes;
sbuf->b.b.height0 = 1;
sbuf->b.b.depth0 = 1;
sbuf->b.b.array_size = 1;
sbuf->bind_flags = bind;
sbuf->swbuf = ptr;
sbuf->user = TRUE;
debug_reference(&sbuf->b.b.reference,
(debug_reference_descriptor)debug_describe_resource, 0);
ss->hud.num_resources++;
return &sbuf->b.b;
no_sbuf:
return NULL;
}
|