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
|
/*
* Copyright © 2014-2015 Broadcom
*
* 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.
*/
/** @file vc4_job.c
*
* Functions for submitting VC4 render jobs to the kernel.
*/
#include <xf86drm.h>
#include "vc4_context.h"
#include "util/hash_table.h"
static void
remove_from_ht(struct hash_table *ht, void *key)
{
struct hash_entry *entry = _mesa_hash_table_search(ht, key);
_mesa_hash_table_remove(ht, entry);
}
static void
vc4_job_free(struct vc4_context *vc4, struct vc4_job *job)
{
struct vc4_bo **referenced_bos = job->bo_pointers.base;
for (int i = 0; i < cl_offset(&job->bo_handles) / 4; i++) {
vc4_bo_unreference(&referenced_bos[i]);
}
remove_from_ht(vc4->jobs, &job->key);
if (job->color_write) {
remove_from_ht(vc4->write_jobs, job->color_write->texture);
pipe_surface_reference(&job->color_write, NULL);
}
if (job->msaa_color_write) {
remove_from_ht(vc4->write_jobs, job->msaa_color_write->texture);
pipe_surface_reference(&job->msaa_color_write, NULL);
}
if (job->zs_write) {
remove_from_ht(vc4->write_jobs, job->zs_write->texture);
pipe_surface_reference(&job->zs_write, NULL);
}
if (job->msaa_zs_write) {
remove_from_ht(vc4->write_jobs, job->msaa_zs_write->texture);
pipe_surface_reference(&job->msaa_zs_write, NULL);
}
pipe_surface_reference(&job->color_read, NULL);
pipe_surface_reference(&job->zs_read, NULL);
if (vc4->job == job)
vc4->job = NULL;
ralloc_free(job);
}
static struct vc4_job *
vc4_job_create(struct vc4_context *vc4)
{
struct vc4_job *job = rzalloc(vc4, struct vc4_job);
vc4_init_cl(job, &job->bcl);
vc4_init_cl(job, &job->shader_rec);
vc4_init_cl(job, &job->uniforms);
vc4_init_cl(job, &job->bo_handles);
vc4_init_cl(job, &job->bo_pointers);
job->draw_min_x = ~0;
job->draw_min_y = ~0;
job->draw_max_x = 0;
job->draw_max_y = 0;
return job;
}
void
vc4_flush_jobs_writing_resource(struct vc4_context *vc4,
struct pipe_resource *prsc)
{
struct hash_entry *entry = _mesa_hash_table_search(vc4->write_jobs,
prsc);
if (entry) {
struct vc4_job *job = entry->data;
vc4_job_submit(vc4, job);
}
}
void
vc4_flush_jobs_reading_resource(struct vc4_context *vc4,
struct pipe_resource *prsc)
{
struct vc4_resource *rsc = vc4_resource(prsc);
vc4_flush_jobs_writing_resource(vc4, prsc);
struct hash_entry *entry;
hash_table_foreach(vc4->jobs, entry) {
struct vc4_job *job = entry->data;
struct vc4_bo **referenced_bos = job->bo_pointers.base;
for (int i = 0; i < cl_offset(&job->bo_handles) / 4; i++) {
if (referenced_bos[i] == rsc->bo) {
vc4_job_submit(vc4, job);
continue;
}
}
/* Also check for the Z/color buffers, since the references to
* those are only added immediately before submit.
*/
if (job->color_read && !(job->cleared & PIPE_CLEAR_COLOR)) {
struct vc4_resource *ctex =
vc4_resource(job->color_read->texture);
if (ctex->bo == rsc->bo) {
vc4_job_submit(vc4, job);
continue;
}
}
if (job->zs_read && !(job->cleared &
(PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL))) {
struct vc4_resource *ztex =
vc4_resource(job->zs_read->texture);
if (ztex->bo == rsc->bo) {
vc4_job_submit(vc4, job);
continue;
}
}
}
}
/**
* Returns a vc4_job struture for tracking V3D rendering to a particular FBO.
*
* If we've already started rendering to this FBO, then return old same job,
* otherwise make a new one. If we're beginning rendering to an FBO, make
* sure that any previous reads of the FBO (or writes to its color/Z surfaces)
* have been flushed.
*/
struct vc4_job *
vc4_get_job(struct vc4_context *vc4,
struct pipe_surface *cbuf, struct pipe_surface *zsbuf)
{
/* Return the existing job for this FBO if we have one */
struct vc4_job_key local_key = {.cbuf = cbuf, .zsbuf = zsbuf};
struct hash_entry *entry = _mesa_hash_table_search(vc4->jobs,
&local_key);
if (entry)
return entry->data;
/* Creating a new job. Make sure that any previous jobs reading or
* writing these buffers are flushed.
*/
if (cbuf)
vc4_flush_jobs_reading_resource(vc4, cbuf->texture);
if (zsbuf)
vc4_flush_jobs_reading_resource(vc4, zsbuf->texture);
struct vc4_job *job = vc4_job_create(vc4);
if (cbuf) {
if (cbuf->texture->nr_samples > 1) {
job->msaa = true;
pipe_surface_reference(&job->msaa_color_write, cbuf);
} else {
pipe_surface_reference(&job->color_write, cbuf);
}
}
if (zsbuf) {
if (zsbuf->texture->nr_samples > 1) {
job->msaa = true;
pipe_surface_reference(&job->msaa_zs_write, zsbuf);
} else {
pipe_surface_reference(&job->zs_write, zsbuf);
}
}
if (job->msaa) {
job->tile_width = 32;
job->tile_height = 32;
} else {
job->tile_width = 64;
job->tile_height = 64;
}
if (cbuf)
_mesa_hash_table_insert(vc4->write_jobs, cbuf->texture, job);
if (zsbuf)
_mesa_hash_table_insert(vc4->write_jobs, zsbuf->texture, job);
job->key.cbuf = cbuf;
job->key.zsbuf = zsbuf;
_mesa_hash_table_insert(vc4->jobs, &job->key, job);
return job;
}
struct vc4_job *
vc4_get_job_for_fbo(struct vc4_context *vc4)
{
if (vc4->job)
return vc4->job;
struct pipe_surface *cbuf = vc4->framebuffer.cbufs[0];
struct pipe_surface *zsbuf = vc4->framebuffer.zsbuf;
struct vc4_job *job = vc4_get_job(vc4, cbuf, zsbuf);
/* The dirty flags are tracking what's been updated while vc4->job has
* been bound, so set them all to ~0 when switching between jobs. We
* also need to reset all state at the start of rendering.
*/
vc4->dirty = ~0;
/* Set up the read surfaces in the job. If they aren't actually
* getting read (due to a clear starting the frame), job->cleared will
* mask out the read.
*/
pipe_surface_reference(&job->color_read, cbuf);
pipe_surface_reference(&job->zs_read, zsbuf);
/* If we're binding to uninitialized buffers, no need to load their
* contents before drawing.
*/
if (cbuf) {
struct vc4_resource *rsc = vc4_resource(cbuf->texture);
if (!rsc->writes)
job->cleared |= PIPE_CLEAR_COLOR0;
}
if (zsbuf) {
struct vc4_resource *rsc = vc4_resource(zsbuf->texture);
if (!rsc->writes)
job->cleared |= PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL;
}
job->draw_tiles_x = DIV_ROUND_UP(vc4->framebuffer.width,
job->tile_width);
job->draw_tiles_y = DIV_ROUND_UP(vc4->framebuffer.height,
job->tile_height);
vc4->job = job;
return job;
}
static void
vc4_submit_setup_rcl_surface(struct vc4_job *job,
struct drm_vc4_submit_rcl_surface *submit_surf,
struct pipe_surface *psurf,
bool is_depth, bool is_write)
{
struct vc4_surface *surf = vc4_surface(psurf);
if (!surf)
return;
struct vc4_resource *rsc = vc4_resource(psurf->texture);
submit_surf->hindex = vc4_gem_hindex(job, rsc->bo);
submit_surf->offset = surf->offset;
if (psurf->texture->nr_samples <= 1) {
if (is_depth) {
submit_surf->bits =
VC4_SET_FIELD(VC4_LOADSTORE_TILE_BUFFER_ZS,
VC4_LOADSTORE_TILE_BUFFER_BUFFER);
} else {
submit_surf->bits =
VC4_SET_FIELD(VC4_LOADSTORE_TILE_BUFFER_COLOR,
VC4_LOADSTORE_TILE_BUFFER_BUFFER) |
VC4_SET_FIELD(vc4_rt_format_is_565(psurf->format) ?
VC4_LOADSTORE_TILE_BUFFER_BGR565 :
VC4_LOADSTORE_TILE_BUFFER_RGBA8888,
VC4_LOADSTORE_TILE_BUFFER_FORMAT);
}
submit_surf->bits |=
VC4_SET_FIELD(surf->tiling,
VC4_LOADSTORE_TILE_BUFFER_TILING);
} else {
assert(!is_write);
submit_surf->flags |= VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES;
}
if (is_write)
rsc->writes++;
}
static void
vc4_submit_setup_rcl_render_config_surface(struct vc4_job *job,
struct drm_vc4_submit_rcl_surface *submit_surf,
struct pipe_surface *psurf)
{
struct vc4_surface *surf = vc4_surface(psurf);
if (!surf)
return;
struct vc4_resource *rsc = vc4_resource(psurf->texture);
submit_surf->hindex = vc4_gem_hindex(job, rsc->bo);
submit_surf->offset = surf->offset;
if (psurf->texture->nr_samples <= 1) {
submit_surf->bits =
VC4_SET_FIELD(vc4_rt_format_is_565(surf->base.format) ?
VC4_RENDER_CONFIG_FORMAT_BGR565 :
VC4_RENDER_CONFIG_FORMAT_RGBA8888,
VC4_RENDER_CONFIG_FORMAT) |
VC4_SET_FIELD(surf->tiling,
VC4_RENDER_CONFIG_MEMORY_FORMAT);
}
rsc->writes++;
}
static void
vc4_submit_setup_rcl_msaa_surface(struct vc4_job *job,
struct drm_vc4_submit_rcl_surface *submit_surf,
struct pipe_surface *psurf)
{
struct vc4_surface *surf = vc4_surface(psurf);
if (!surf)
return;
struct vc4_resource *rsc = vc4_resource(psurf->texture);
submit_surf->hindex = vc4_gem_hindex(job, rsc->bo);
submit_surf->offset = surf->offset;
submit_surf->bits = 0;
rsc->writes++;
}
/**
* Submits the job to the kernel and then reinitializes it.
*/
void
vc4_job_submit(struct vc4_context *vc4, struct vc4_job *job)
{
if (!job->needs_flush)
goto done;
/* The RCL setup would choke if the draw bounds cause no drawing, so
* just drop the drawing if that's the case.
*/
if (job->draw_max_x <= job->draw_min_x ||
job->draw_max_y <= job->draw_min_y) {
goto done;
}
if (vc4_debug & VC4_DEBUG_CL) {
fprintf(stderr, "BCL:\n");
vc4_dump_cl(job->bcl.base, cl_offset(&job->bcl), false);
}
if (cl_offset(&job->bcl) > 0) {
/* Increment the semaphore indicating that binning is done and
* unblocking the render thread. Note that this doesn't act
* until the FLUSH completes.
*/
cl_ensure_space(&job->bcl, 8);
cl_emit(&job->bcl, INCREMENT_SEMAPHORE, incr);
/* The FLUSH caps all of our bin lists with a
* VC4_PACKET_RETURN.
*/
cl_emit(&job->bcl, FLUSH, flush);
}
struct drm_vc4_submit_cl submit = {
.color_read.hindex = ~0,
.zs_read.hindex = ~0,
.color_write.hindex = ~0,
.msaa_color_write.hindex = ~0,
.zs_write.hindex = ~0,
.msaa_zs_write.hindex = ~0,
};
cl_ensure_space(&job->bo_handles, 6 * sizeof(uint32_t));
cl_ensure_space(&job->bo_pointers, 6 * sizeof(struct vc4_bo *));
if (job->resolve & PIPE_CLEAR_COLOR) {
if (!(job->cleared & PIPE_CLEAR_COLOR)) {
vc4_submit_setup_rcl_surface(job, &submit.color_read,
job->color_read,
false, false);
}
vc4_submit_setup_rcl_render_config_surface(job,
&submit.color_write,
job->color_write);
vc4_submit_setup_rcl_msaa_surface(job,
&submit.msaa_color_write,
job->msaa_color_write);
}
if (job->resolve & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) {
if (!(job->cleared & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL))) {
vc4_submit_setup_rcl_surface(job, &submit.zs_read,
job->zs_read, true, false);
}
vc4_submit_setup_rcl_surface(job, &submit.zs_write,
job->zs_write, true, true);
vc4_submit_setup_rcl_msaa_surface(job, &submit.msaa_zs_write,
job->msaa_zs_write);
}
if (job->msaa) {
/* This bit controls how many pixels the general
* (i.e. subsampled) loads/stores are iterating over
* (multisample loads replicate out to the other samples).
*/
submit.color_write.bits |= VC4_RENDER_CONFIG_MS_MODE_4X;
/* Controls whether color_write's
* VC4_PACKET_STORE_MS_TILE_BUFFER does 4x decimation
*/
submit.color_write.bits |= VC4_RENDER_CONFIG_DECIMATE_MODE_4X;
}
submit.bo_handles = (uintptr_t)job->bo_handles.base;
submit.bo_handle_count = cl_offset(&job->bo_handles) / 4;
submit.bin_cl = (uintptr_t)job->bcl.base;
submit.bin_cl_size = cl_offset(&job->bcl);
submit.shader_rec = (uintptr_t)job->shader_rec.base;
submit.shader_rec_size = cl_offset(&job->shader_rec);
submit.shader_rec_count = job->shader_rec_count;
submit.uniforms = (uintptr_t)job->uniforms.base;
submit.uniforms_size = cl_offset(&job->uniforms);
assert(job->draw_min_x != ~0 && job->draw_min_y != ~0);
submit.min_x_tile = job->draw_min_x / job->tile_width;
submit.min_y_tile = job->draw_min_y / job->tile_height;
submit.max_x_tile = (job->draw_max_x - 1) / job->tile_width;
submit.max_y_tile = (job->draw_max_y - 1) / job->tile_height;
submit.width = job->draw_width;
submit.height = job->draw_height;
if (job->cleared) {
submit.flags |= VC4_SUBMIT_CL_USE_CLEAR_COLOR;
submit.clear_color[0] = job->clear_color[0];
submit.clear_color[1] = job->clear_color[1];
submit.clear_z = job->clear_depth;
submit.clear_s = job->clear_stencil;
}
if (!(vc4_debug & VC4_DEBUG_NORAST)) {
int ret;
#ifndef USE_VC4_SIMULATOR
ret = drmIoctl(vc4->fd, DRM_IOCTL_VC4_SUBMIT_CL, &submit);
#else
ret = vc4_simulator_flush(vc4, &submit, job);
#endif
static bool warned = false;
if (ret && !warned) {
fprintf(stderr, "Draw call returned %s. "
"Expect corruption.\n", strerror(errno));
warned = true;
} else if (!ret) {
vc4->last_emit_seqno = submit.seqno;
}
}
if (vc4->last_emit_seqno - vc4->screen->finished_seqno > 5) {
if (!vc4_wait_seqno(vc4->screen,
vc4->last_emit_seqno - 5,
PIPE_TIMEOUT_INFINITE,
"job throttling")) {
fprintf(stderr, "Job throttling failed\n");
}
}
if (vc4_debug & VC4_DEBUG_ALWAYS_SYNC) {
if (!vc4_wait_seqno(vc4->screen, vc4->last_emit_seqno,
PIPE_TIMEOUT_INFINITE, "sync")) {
fprintf(stderr, "Wait failed.\n");
abort();
}
}
done:
vc4_job_free(vc4, job);
}
static bool
vc4_job_compare(const void *a, const void *b)
{
return memcmp(a, b, sizeof(struct vc4_job_key)) == 0;
}
static uint32_t
vc4_job_hash(const void *key)
{
return _mesa_hash_data(key, sizeof(struct vc4_job_key));
}
void
vc4_job_init(struct vc4_context *vc4)
{
vc4->jobs = _mesa_hash_table_create(vc4,
vc4_job_hash,
vc4_job_compare);
vc4->write_jobs = _mesa_hash_table_create(vc4,
_mesa_hash_pointer,
_mesa_key_pointer_equal);
}
|