aboutsummaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers/v3d/v3d_job.c
blob: 58410484d3f93a44350564847f85fae814dd3ec7 (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
/*
 * Copyright © 2014-2017 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 v3d_job.c
 *
 * Functions for submitting VC5 render jobs to the kernel.
 */

#include <xf86drm.h>
#include "v3d_context.h"
/* The OQ/semaphore packets are the same across V3D versions. */
#define V3D_VERSION 33
#include "broadcom/cle/v3dx_pack.h"
#include "broadcom/common/v3d_macros.h"
#include "util/hash_table.h"
#include "util/ralloc.h"
#include "util/set.h"
#include "broadcom/clif/clif_dump.h"

void
v3d_job_free(struct v3d_context *v3d, struct v3d_job *job)
{
        set_foreach(job->bos, entry) {
                struct v3d_bo *bo = (struct v3d_bo *)entry->key;
                v3d_bo_unreference(&bo);
        }

        _mesa_hash_table_remove_key(v3d->jobs, &job->key);

        if (job->write_prscs) {
                set_foreach(job->write_prscs, entry) {
                        const struct pipe_resource *prsc = entry->key;

                        _mesa_hash_table_remove_key(v3d->write_jobs, prsc);
                }
        }

        for (int i = 0; i < V3D_MAX_DRAW_BUFFERS; i++) {
                if (job->cbufs[i]) {
                        _mesa_hash_table_remove_key(v3d->write_jobs,
                                                    job->cbufs[i]->texture);
                        pipe_surface_reference(&job->cbufs[i], NULL);
                }
        }
        if (job->zsbuf) {
                struct v3d_resource *rsc = v3d_resource(job->zsbuf->texture);
                if (rsc->separate_stencil)
                        _mesa_hash_table_remove_key(v3d->write_jobs,
                                                    &rsc->separate_stencil->base);

                _mesa_hash_table_remove_key(v3d->write_jobs,
                                            job->zsbuf->texture);
                pipe_surface_reference(&job->zsbuf, NULL);
        }

        if (v3d->job == job)
                v3d->job = NULL;

        v3d_destroy_cl(&job->bcl);
        v3d_destroy_cl(&job->rcl);
        v3d_destroy_cl(&job->indirect);
        v3d_bo_unreference(&job->tile_alloc);
        v3d_bo_unreference(&job->tile_state);

        ralloc_free(job);
}

struct v3d_job *
v3d_job_create(struct v3d_context *v3d)
{
        struct v3d_job *job = rzalloc(v3d, struct v3d_job);

        job->v3d = v3d;

        v3d_init_cl(job, &job->bcl);
        v3d_init_cl(job, &job->rcl);
        v3d_init_cl(job, &job->indirect);

        job->draw_min_x = ~0;
        job->draw_min_y = ~0;
        job->draw_max_x = 0;
        job->draw_max_y = 0;

        job->bos = _mesa_set_create(job,
                                    _mesa_hash_pointer,
                                    _mesa_key_pointer_equal);
        return job;
}

void
v3d_job_add_bo(struct v3d_job *job, struct v3d_bo *bo)
{
        if (!bo)
                return;

        if (_mesa_set_search(job->bos, bo))
                return;

        v3d_bo_reference(bo);
        _mesa_set_add(job->bos, bo);
        job->referenced_size += bo->size;

        uint32_t *bo_handles = (void *)(uintptr_t)job->submit.bo_handles;

        if (job->submit.bo_handle_count >= job->bo_handles_size) {
                job->bo_handles_size = MAX2(4, job->bo_handles_size * 2);
                bo_handles = reralloc(job, bo_handles,
                                      uint32_t, job->bo_handles_size);
                job->submit.bo_handles = (uintptr_t)(void *)bo_handles;
        }
        bo_handles[job->submit.bo_handle_count++] = bo->handle;
}

void
v3d_job_add_write_resource(struct v3d_job *job, struct pipe_resource *prsc)
{
        struct v3d_context *v3d = job->v3d;

        if (!job->write_prscs) {
                job->write_prscs = _mesa_set_create(job,
                                                    _mesa_hash_pointer,
                                                    _mesa_key_pointer_equal);
        }

        _mesa_set_add(job->write_prscs, prsc);
        _mesa_hash_table_insert(v3d->write_jobs, prsc, job);
}

void
v3d_flush_jobs_using_bo(struct v3d_context *v3d, struct v3d_bo *bo)
{
        hash_table_foreach(v3d->jobs, entry) {
                struct v3d_job *job = entry->data;

                if (_mesa_set_search(job->bos, bo))
                        v3d_job_submit(v3d, job);
        }
}

void
v3d_job_add_tf_write_resource(struct v3d_job *job, struct pipe_resource *prsc)
{
        v3d_job_add_write_resource(job, prsc);

        if (!job->tf_write_prscs)
                job->tf_write_prscs = _mesa_pointer_set_create(job);

        _mesa_set_add(job->tf_write_prscs, prsc);
}

static bool
v3d_job_writes_resource_from_tf(struct v3d_job *job,
                                struct pipe_resource *prsc)
{
        if (!job->tf_enabled)
                return false;

        if (!job->tf_write_prscs)
                return false;

        return _mesa_set_search(job->tf_write_prscs, prsc) != NULL;
}

void
v3d_flush_jobs_writing_resource(struct v3d_context *v3d,
                                struct pipe_resource *prsc,
                                enum v3d_flush_cond flush_cond,
                                bool is_compute_pipeline)
{
        struct hash_entry *entry = _mesa_hash_table_search(v3d->write_jobs,
                                                           prsc);
        struct v3d_resource *rsc = v3d_resource(prsc);

        /* We need to sync if graphics pipeline reads a resource written
         * by the compute pipeline. The same would be needed for the case of
         * graphics-compute dependency but nowadays all compute jobs
         * are serialized with the previous submitted job.
         */
        if (!is_compute_pipeline && rsc->bo != NULL && rsc->compute_written) {
           v3d->sync_on_last_compute_job = true;
           rsc->compute_written = false;
        }

        if (!entry)
                return;

        struct v3d_job *job = entry->data;

        bool needs_flush;
        switch (flush_cond) {
        case V3D_FLUSH_ALWAYS:
                needs_flush = true;
                break;
        case V3D_FLUSH_NOT_CURRENT_JOB:
                needs_flush = !v3d->job || v3d->job != job;
                break;
        case V3D_FLUSH_DEFAULT:
        default:
                /* For writes from TF in the same job we use the "Wait for TF"
                 * feature provided by the hardware so we don't want to flush.
                 * The exception to this is when the caller is about to map the
                 * resource since in that case we don't have a 'Wait for TF'
                 * command the in command stream. In this scenario the caller
                 * is expected to set 'always_flush' to True.
                 */
                needs_flush = !v3d_job_writes_resource_from_tf(job, prsc);
        }

        if (needs_flush)
                v3d_job_submit(v3d, job);
}

void
v3d_flush_jobs_reading_resource(struct v3d_context *v3d,
                                struct pipe_resource *prsc,
                                enum v3d_flush_cond flush_cond,
                                bool is_compute_pipeline)
{
        struct v3d_resource *rsc = v3d_resource(prsc);

        /* We only need to force the flush on TF writes, which is the only
         * case where we might skip the flush to use the 'Wait for TF'
         * command. Here we are flushing for a read, which means that the
         * caller intends to write to the resource, so we don't care if
         * there was a previous TF write to it.
         */
        v3d_flush_jobs_writing_resource(v3d, prsc, flush_cond,
                                        is_compute_pipeline);

        hash_table_foreach(v3d->jobs, entry) {
                struct v3d_job *job = entry->data;

                if (!_mesa_set_search(job->bos, rsc->bo))
                        continue;

                bool needs_flush;
                switch (flush_cond) {
                case V3D_FLUSH_NOT_CURRENT_JOB:
                        needs_flush = !v3d->job || v3d->job != job;
                        break;
                case V3D_FLUSH_ALWAYS:
                case V3D_FLUSH_DEFAULT:
                default:
                        needs_flush = true;
                }

                if (needs_flush)
                        v3d_job_submit(v3d, job);

                /* Reminder: v3d->jobs is safe to keep iterating even
                 * after deletion of an entry.
                 */
                continue;
        }
}

static void
v3d_job_set_tile_buffer_size(struct v3d_job *job)
{
        static const uint8_t tile_sizes[] = {
                64, 64,
                64, 32,
                32, 32,
                32, 16,
                16, 16,
        };
        int tile_size_index = 0;
        if (job->msaa)
                tile_size_index += 2;

        if (job->cbufs[3] || job->cbufs[2])
                tile_size_index += 2;
        else if (job->cbufs[1])
                tile_size_index++;

        int max_bpp = RENDER_TARGET_MAXIMUM_32BPP;
        for (int i = 0; i < V3D_MAX_DRAW_BUFFERS; i++) {
                if (job->cbufs[i]) {
                        struct v3d_surface *surf = v3d_surface(job->cbufs[i]);
                        max_bpp = MAX2(max_bpp, surf->internal_bpp);
                }
        }
        job->internal_bpp = max_bpp;
        STATIC_ASSERT(RENDER_TARGET_MAXIMUM_32BPP == 0);
        tile_size_index += max_bpp;

        assert(tile_size_index < ARRAY_SIZE(tile_sizes));
        job->tile_width = tile_sizes[tile_size_index * 2 + 0];
        job->tile_height = tile_sizes[tile_size_index * 2 + 1];
}

/**
 * Returns a v3d_job struture for tracking V3D rendering to a particular FBO.
 *
 * If we've already started rendering to this FBO, then return the 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 v3d_job *
v3d_get_job(struct v3d_context *v3d,
            struct pipe_surface **cbufs, struct pipe_surface *zsbuf)
{
        /* Return the existing job for this FBO if we have one */
        struct v3d_job_key local_key = {
                .cbufs = {
                        cbufs[0],
                        cbufs[1],
                        cbufs[2],
                        cbufs[3],
                },
                .zsbuf = zsbuf,
        };
        struct hash_entry *entry = _mesa_hash_table_search(v3d->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.
         */
        struct v3d_job *job = v3d_job_create(v3d);

        for (int i = 0; i < V3D_MAX_DRAW_BUFFERS; i++) {
                if (cbufs[i]) {
                        v3d_flush_jobs_reading_resource(v3d, cbufs[i]->texture,
                                                        V3D_FLUSH_DEFAULT,
                                                        false);
                        pipe_surface_reference(&job->cbufs[i], cbufs[i]);

                        if (cbufs[i]->texture->nr_samples > 1)
                                job->msaa = true;
                }
        }
        if (zsbuf) {
                v3d_flush_jobs_reading_resource(v3d, zsbuf->texture,
                                                V3D_FLUSH_DEFAULT,
                                                false);
                pipe_surface_reference(&job->zsbuf, zsbuf);
                if (zsbuf->texture->nr_samples > 1)
                        job->msaa = true;
        }

        for (int i = 0; i < V3D_MAX_DRAW_BUFFERS; i++) {
                if (cbufs[i])
                        _mesa_hash_table_insert(v3d->write_jobs,
                                                cbufs[i]->texture, job);
        }
        if (zsbuf) {
                _mesa_hash_table_insert(v3d->write_jobs, zsbuf->texture, job);

                struct v3d_resource *rsc = v3d_resource(zsbuf->texture);
                if (rsc->separate_stencil) {
                        v3d_flush_jobs_reading_resource(v3d,
                                                        &rsc->separate_stencil->base,
                                                        V3D_FLUSH_DEFAULT,
                                                        false);
                        _mesa_hash_table_insert(v3d->write_jobs,
                                                &rsc->separate_stencil->base,
                                                job);
                }
        }

        memcpy(&job->key, &local_key, sizeof(local_key));
        _mesa_hash_table_insert(v3d->jobs, &job->key, job);

        return job;
}

struct v3d_job *
v3d_get_job_for_fbo(struct v3d_context *v3d)
{
        if (v3d->job)
                return v3d->job;

        struct pipe_surface **cbufs = v3d->framebuffer.cbufs;
        struct pipe_surface *zsbuf = v3d->framebuffer.zsbuf;
        struct v3d_job *job = v3d_get_job(v3d, cbufs, zsbuf);

        if (v3d->framebuffer.samples >= 1)
                job->msaa = true;

        v3d_job_set_tile_buffer_size(job);

        /* The dirty flags are tracking what's been updated while v3d->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.
         */
        v3d->dirty = ~0;

        /* If we're binding to uninitialized buffers, no need to load their
         * contents before drawing.
         */
        for (int i = 0; i < 4; i++) {
                if (cbufs[i]) {
                        struct v3d_resource *rsc = v3d_resource(cbufs[i]->texture);
                        if (!rsc->writes)
                                job->clear |= PIPE_CLEAR_COLOR0 << i;
                }
        }

        if (zsbuf) {
                struct v3d_resource *rsc = v3d_resource(zsbuf->texture);
                if (!rsc->writes)
                        job->clear |= PIPE_CLEAR_DEPTH;

                if (rsc->separate_stencil)
                        rsc = rsc->separate_stencil;

                if (!rsc->writes)
                        job->clear |= PIPE_CLEAR_STENCIL;
        }

        job->draw_tiles_x = DIV_ROUND_UP(v3d->framebuffer.width,
                                         job->tile_width);
        job->draw_tiles_y = DIV_ROUND_UP(v3d->framebuffer.height,
                                         job->tile_height);

        v3d->job = job;

        return job;
}

static void
v3d_clif_dump(struct v3d_context *v3d, struct v3d_job *job)
{
        if (!(V3D_DEBUG & (V3D_DEBUG_CL | V3D_DEBUG_CLIF)))
                return;

        struct clif_dump *clif = clif_dump_init(&v3d->screen->devinfo,
                                                stderr,
                                                V3D_DEBUG & V3D_DEBUG_CL);

        set_foreach(job->bos, entry) {
                struct v3d_bo *bo = (void *)entry->key;
                char *name = ralloc_asprintf(NULL, "%s_0x%x",
                                             bo->name, bo->offset);

                v3d_bo_map(bo);
                clif_dump_add_bo(clif, name, bo->offset, bo->size, bo->map);

                ralloc_free(name);
        }

        clif_dump(clif, &job->submit);

        clif_dump_destroy(clif);
}

static void
v3d_read_and_accumulate_primitive_counters(struct v3d_context *v3d)
{
        assert(v3d->prim_counts);

        perf_debug("stalling on TF counts readback\n");
        struct v3d_resource *rsc = v3d_resource(v3d->prim_counts);
        if (v3d_bo_wait(rsc->bo, PIPE_TIMEOUT_INFINITE, "prim-counts")) {
                uint32_t *map = v3d_bo_map(rsc->bo) + v3d->prim_counts_offset;
                v3d->tf_prims_generated += map[V3D_PRIM_COUNTS_TF_WRITTEN];
                /* When we only have a vertex shader we determine the primitive
                 * count in the CPU so don't update it here again.
                 */
                if (v3d->prog.gs)
                        v3d->prims_generated += map[V3D_PRIM_COUNTS_WRITTEN];
        }
}

/**
 * Submits the job to the kernel and then reinitializes it.
 */
void
v3d_job_submit(struct v3d_context *v3d, struct v3d_job *job)
{
        struct v3d_screen *screen = v3d->screen;

        if (!job->needs_flush)
                goto done;

        if (screen->devinfo.ver >= 41)
                v3d41_emit_rcl(job);
        else
                v3d33_emit_rcl(job);

        if (cl_offset(&job->bcl) > 0) {
                if (screen->devinfo.ver >= 41)
                        v3d41_bcl_epilogue(v3d, job);
                else
                        v3d33_bcl_epilogue(v3d, job);
        }

        /* While the RCL will implicitly depend on the last RCL to have
         * finished, we also need to block on any previous TFU job we may have
         * dispatched.
         */
        job->submit.in_sync_rcl = v3d->out_sync;

        /* Update the sync object for the last rendering by our context. */
        job->submit.out_sync = v3d->out_sync;

        job->submit.bcl_end = job->bcl.bo->offset + cl_offset(&job->bcl);
        job->submit.rcl_end = job->rcl.bo->offset + cl_offset(&job->rcl);

        job->submit.flags = 0;
        if (job->tmu_dirty_rcl && screen->has_cache_flush)
                job->submit.flags |= DRM_V3D_SUBMIT_CL_FLUSH_CACHE;

        /* On V3D 4.1, the tile alloc/state setup moved to register writes
         * instead of binner packets.
         */
        if (screen->devinfo.ver >= 41) {
                v3d_job_add_bo(job, job->tile_alloc);
                job->submit.qma = job->tile_alloc->offset;
                job->submit.qms = job->tile_alloc->size;

                v3d_job_add_bo(job, job->tile_state);
                job->submit.qts = job->tile_state->offset;
        }

        v3d_clif_dump(v3d, job);

        if (!(V3D_DEBUG & V3D_DEBUG_NORAST)) {
                int ret;

                ret = v3d_ioctl(v3d->fd, DRM_IOCTL_V3D_SUBMIT_CL, &job->submit);
                static bool warned = false;
                if (ret && !warned) {
                        fprintf(stderr, "Draw call returned %s.  "
                                        "Expect corruption.\n", strerror(errno));
                        warned = true;
                }

                /* If we are submitting a job in the middle of transform
                 * feedback we need to read the primitive counts and accumulate
                 * them, otherwise they will be reset at the start of the next
                 * draw when we emit the Tile Binning Mode Configuration packet.
                 *
                 * If the job doesn't have any TF draw calls, then we know
                 * the primitive count must be zero and we can skip stalling
                 * for this. This also fixes a problem because it seems that
                 * in this scenario the counters are not reset with the Tile
                 * Binning Mode Configuration packet, which would translate
                 * to us reading an obsolete (possibly non-zero) value from
                 * the GPU counters.
                 */
                if (v3d->streamout.num_targets && job->tf_draw_calls_queued > 0)
                        v3d_read_and_accumulate_primitive_counters(v3d);
        }

done:
        v3d_job_free(v3d, job);
}

static bool
v3d_job_compare(const void *a, const void *b)
{
        return memcmp(a, b, sizeof(struct v3d_job_key)) == 0;
}

static uint32_t
v3d_job_hash(const void *key)
{
        return _mesa_hash_data(key, sizeof(struct v3d_job_key));
}

void
v3d_job_init(struct v3d_context *v3d)
{
        v3d->jobs = _mesa_hash_table_create(v3d,
                                            v3d_job_hash,
                                            v3d_job_compare);
        v3d->write_jobs = _mesa_hash_table_create(v3d,
                                                  _mesa_hash_pointer,
                                                  _mesa_key_pointer_equal);
}