aboutsummaryrefslogtreecommitdiffstats
path: root/libavcodec/vvc/thread.c
blob: 2654b40058b580ccd8a406fd22d40e9249aca91a (plain) (blame)
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
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
/*
 * VVC thread logic
 *
 * Copyright (C) 2023 Nuo Mi
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include <stdatomic.h>

#include "libavutil/executor.h"
#include "libavutil/mem.h"
#include "libavutil/thread.h"

#include "thread.h"
#include "ctu.h"
#include "filter.h"
#include "inter.h"
#include "intra.h"
#include "refs.h"

typedef struct ProgressListener {
    VVCProgressListener l;
    struct VVCTask *task;
    VVCContext *s;
} ProgressListener;

typedef enum VVCTaskStage {
    VVC_TASK_STAGE_PARSE,
    VVC_TASK_STAGE_INTER,
    VVC_TASK_STAGE_RECON,
    VVC_TASK_STAGE_LMCS,
    VVC_TASK_STAGE_DEBLOCK_V,
    VVC_TASK_STAGE_DEBLOCK_H,
    VVC_TASK_STAGE_SAO,
    VVC_TASK_STAGE_ALF,
    VVC_TASK_STAGE_LAST
} VVCTaskStage;

typedef struct VVCTask {
    union {
        struct VVCTask *next;                //for executor debug only
        AVTask task;
    } u;

    VVCTaskStage stage;

    // ctu x, y, and raster scan order
    int rx, ry, rs;
    VVCFrameContext *fc;

    ProgressListener col_listener;
    ProgressListener listener[2][VVC_MAX_REF_ENTRIES];

    // for parse task only
    SliceContext *sc;
    EntryPoint *ep;
    int ctu_idx;                    //ctu idx in the current slice

    // tasks with target scores met are ready for scheduling
    atomic_uchar score[VVC_TASK_STAGE_LAST];
    atomic_uchar target_inter_score;
} VVCTask;

typedef struct VVCRowThread {
    atomic_int col_progress[VVC_PROGRESS_LAST];
} VVCRowThread;

typedef struct VVCFrameThread {
    // error return for tasks
    atomic_int ret;

    VVCRowThread *rows;
    VVCTask *tasks;

    int ctu_size;
    int ctu_width;
    int ctu_height;
    int ctu_count;

    //protected by lock
    atomic_int nb_scheduled_tasks;
    atomic_int nb_scheduled_listeners;

    int row_progress[VVC_PROGRESS_LAST];

    AVMutex lock;
    AVCond  cond;
} VVCFrameThread;

static void add_task(VVCContext *s, VVCTask *t)
{
    VVCFrameThread *ft = t->fc->ft;

    atomic_fetch_add(&ft->nb_scheduled_tasks, 1);

    av_executor_execute(s->executor, &t->u.task);
}

static void task_init(VVCTask *t, VVCTaskStage stage, VVCFrameContext *fc, const int rx, const int ry)
{
    memset(t, 0, sizeof(*t));
    t->stage = stage;
    t->fc    = fc;
    t->rx    = rx;
    t->ry    = ry;
    t->rs    = ry * fc->ft->ctu_width + rx;
    for (int i = 0; i < FF_ARRAY_ELEMS(t->score); i++)
        atomic_store(t->score + i, 0);
    atomic_store(&t->target_inter_score, 0);
}

static int task_init_parse(VVCTask *t, SliceContext *sc, EntryPoint *ep, const int ctu_idx)
{
    if (t->sc) {
        // the task already inited, error bitstream
        return AVERROR_INVALIDDATA;
    }
    t->sc      = sc;
    t->ep      = ep;
    t->ctu_idx = ctu_idx;

    return 0;
}

static uint8_t task_add_score(VVCTask *t, const VVCTaskStage stage)
{
    return atomic_fetch_add(&t->score[stage], 1) + 1;
}

static uint8_t task_get_score(VVCTask *t, const VVCTaskStage stage)
{
    return atomic_load(&t->score[stage]);
}

//first row in tile or slice
static int is_first_row(const VVCFrameContext *fc, const int rx, const int ry)
{
    const VVCFrameThread *ft = fc->ft;
    const VVCPPS *pps        = fc->ps.pps;

    if (ry != pps->ctb_to_row_bd[ry]) {
        const int rs = ry * ft->ctu_width + rx;
        return fc->tab.slice_idx[rs] != fc->tab.slice_idx[rs - ft->ctu_width];
    }
    return 1;
}

static int task_has_target_score(VVCTask *t, const VVCTaskStage stage, const uint8_t score)
{
    // l:left, r:right, t: top, b: bottom
    static const uint8_t target_score[] =
    {
        2,          //VVC_TASK_STAGE_RECON,     need l + rt recon
        3,          //VVC_TASK_STAGE_LMCS,      need r + b + rb recon
        1,          //VVC_TASK_STAGE_DEBLOCK_V, need l deblock v
        2,          //VVC_TASK_STAGE_DEBLOCK_H, need r deblock v + t deblock h
        5,          //VVC_TASK_STAGE_SAO,       need l + r + lb + b + rb deblock h
        8,          //VVC_TASK_STAGE_ALF,       need sao around the ctu
    };
    uint8_t target = 0;
    VVCFrameContext *fc = t->fc;

    if (stage == VVC_TASK_STAGE_PARSE) {
        const H266RawSPS *rsps = fc->ps.sps->r;
        const int wpp = rsps->sps_entropy_coding_sync_enabled_flag && !is_first_row(fc, t->rx, t->ry);
        target = 2 + wpp - 1;                           //left parse + colocation + wpp - no previous stage
    } else if (stage == VVC_TASK_STAGE_INTER) {
        target = atomic_load(&t->target_inter_score);
    } else {
        target = target_score[stage - VVC_TASK_STAGE_RECON];
    }

    //+1 for previous stage
    av_assert0(score <= target + 1);
    return score == target + 1;
}

static void frame_thread_add_score(VVCContext *s, VVCFrameThread *ft,
    const int rx, const int ry, const VVCTaskStage stage)
{
    VVCTask *t = ft->tasks + ft->ctu_width * ry + rx;
    uint8_t score;

    if (rx < 0 || rx >= ft->ctu_width || ry < 0 || ry >= ft->ctu_height)
        return;

    score = task_add_score(t, stage);
    if (task_has_target_score(t, stage, score)) {
        av_assert0(s);
        av_assert0(stage == t->stage);
        add_task(s, t);
    }
}

static void sheduled_done(VVCFrameThread *ft, atomic_int *scheduled)
{
    if (atomic_fetch_sub(scheduled, 1) == 1) {
        ff_mutex_lock(&ft->lock);
        ff_cond_signal(&ft->cond);
        ff_mutex_unlock(&ft->lock);
    }
}

static void progress_done(VVCProgressListener *_l, const int type)
{
    const ProgressListener *l = (ProgressListener *)_l;
    const VVCTask *t          = l->task;
    VVCFrameThread *ft        = t->fc->ft;

    frame_thread_add_score(l->s, ft, t->rx, t->ry, type);
    sheduled_done(ft, &ft->nb_scheduled_listeners);
}

static void pixel_done(VVCProgressListener *l)
{
    progress_done(l, VVC_TASK_STAGE_INTER);
}

static void mv_done(VVCProgressListener *l)
{
    progress_done(l, VVC_TASK_STAGE_PARSE);
}

static void listener_init(ProgressListener *l,  VVCTask *t, VVCContext *s, const VVCProgress vp, const int y)
{
    const int is_inter = vp == VVC_PROGRESS_PIXEL;

    l->task = t;
    l->s    = s;
    l->l.vp = vp;
    l->l.y  = y;
    l->l.progress_done = is_inter ? pixel_done : mv_done;
    if (is_inter)
        atomic_fetch_add(&t->target_inter_score, 1);
}

static void add_progress_listener(VVCFrame *ref, ProgressListener *l,
    VVCTask *t, VVCContext *s, const VVCProgress vp, const int y)
{
    VVCFrameThread *ft = t->fc->ft;

    atomic_fetch_add(&ft->nb_scheduled_listeners, 1);
    listener_init(l, t, s, vp, y);
    ff_vvc_add_progress_listener(ref, (VVCProgressListener*)l);
}

static void schedule_next_parse(VVCContext *s, VVCFrameContext *fc, const SliceContext *sc, const VVCTask *t)
{
    VVCFrameThread *ft = fc->ft;
    EntryPoint *ep     = t->ep;
    const VVCSPS *sps  = fc->ps.sps;

    if (sps->r->sps_entropy_coding_sync_enabled_flag) {
        if (t->rx == fc->ps.pps->ctb_to_col_bd[t->rx]) {
            EntryPoint *next = ep + 1;
            if (next < sc->eps + sc->nb_eps && !is_first_row(fc, t->rx, t->ry + 1)) {
                memcpy(next->cabac_state, ep->cabac_state, sizeof(next->cabac_state));
                ff_vvc_ep_init_stat_coeff(next, sps->bit_depth, sps->r->sps_persistent_rice_adaptation_enabled_flag);
            }
        }
        if (t->ry + 1 < ft->ctu_height && !is_first_row(fc, t->rx, t->ry + 1))
            frame_thread_add_score(s, ft, t->rx, t->ry + 1, VVC_TASK_STAGE_PARSE);
    }

    if (t->ctu_idx + 1 < t->ep->ctu_end) {
        const int next_rs = sc->sh.ctb_addr_in_curr_slice[t->ctu_idx + 1];
        const int next_rx = next_rs % ft->ctu_width;
        const int next_ry = next_rs / ft->ctu_width;
        frame_thread_add_score(s, ft, next_rx, next_ry, VVC_TASK_STAGE_PARSE);
    }
}

static void schedule_inter(VVCContext *s, VVCFrameContext *fc, const SliceContext *sc, VVCTask *t, const int rs)
{
    const VVCSH *sh = &sc->sh;

    if (!IS_I(sh->r)) {
        CTU *ctu = fc->tab.ctus + rs;
        for (int lx = 0; lx < 2; lx++) {
            for (int i = 0; i < sh->r->num_ref_idx_active[lx]; i++) {
                const int y = ctu->max_y[lx][i];
                VVCFrame *ref = sc->rpl[lx].refs[i].ref;
                if (ref && y >= 0)
                    add_progress_listener(ref, &t->listener[lx][i], t, s, VVC_PROGRESS_PIXEL, y + LUMA_EXTRA_AFTER);
            }
        }
    }
}

static void parse_task_done(VVCContext *s, VVCFrameContext *fc, const int rx, const int ry)
{
    VVCFrameThread *ft  = fc->ft;
    const int rs        = ry * ft->ctu_width + rx;
    const int slice_idx = fc->tab.slice_idx[rs];
    VVCTask *t          = ft->tasks + rs;
    const SliceContext *sc = fc->slices[slice_idx];

    schedule_next_parse(s, fc, sc, t);
    schedule_inter(s, fc, sc, t, rs);
}

static void task_stage_done(const VVCTask *t, VVCContext *s)
{
    VVCFrameContext *fc      = t->fc;
    VVCFrameThread *ft       = fc->ft;
    const VVCTaskStage stage = t->stage;

#define ADD(dx, dy, stage) frame_thread_add_score(s, ft, t->rx + (dx), t->ry + (dy), stage)

    //this is a reserve map of ready_score, ordered by zigzag
    if (stage == VVC_TASK_STAGE_PARSE) {
        parse_task_done(s, fc, t->rx, t->ry);
    } else if (stage == VVC_TASK_STAGE_RECON) {
        ADD(-1,  1, VVC_TASK_STAGE_RECON);
        ADD( 1,  0, VVC_TASK_STAGE_RECON);
        ADD(-1, -1, VVC_TASK_STAGE_LMCS);
        ADD( 0, -1, VVC_TASK_STAGE_LMCS);
        ADD(-1,  0, VVC_TASK_STAGE_LMCS);
    } else if (stage == VVC_TASK_STAGE_DEBLOCK_V) {
        ADD( 1,  0,  VVC_TASK_STAGE_DEBLOCK_V);
        ADD(-1,  0,  VVC_TASK_STAGE_DEBLOCK_H);
    } else if (stage == VVC_TASK_STAGE_DEBLOCK_H) {
        ADD( 0,  1,  VVC_TASK_STAGE_DEBLOCK_H);
        ADD(-1, -1,  VVC_TASK_STAGE_SAO);
        ADD( 0, -1,  VVC_TASK_STAGE_SAO);
        ADD(-1,  0,  VVC_TASK_STAGE_SAO);
        ADD( 1, -1,  VVC_TASK_STAGE_SAO);
        ADD( 1,  0,  VVC_TASK_STAGE_SAO);
    } else if (stage == VVC_TASK_STAGE_SAO) {
        ADD(-1, -1,  VVC_TASK_STAGE_ALF);
        ADD( 0, -1,  VVC_TASK_STAGE_ALF);
        ADD(-1,  0,  VVC_TASK_STAGE_ALF);
        ADD( 1, -1,  VVC_TASK_STAGE_ALF);
        ADD(-1,  1,  VVC_TASK_STAGE_ALF);
        ADD( 1,  0,  VVC_TASK_STAGE_ALF);
        ADD( 0,  1,  VVC_TASK_STAGE_ALF);
        ADD( 1,  1,  VVC_TASK_STAGE_ALF);
    }
}

static int task_is_stage_ready(VVCTask *t, int add)
{
    const VVCTaskStage stage = t->stage;
    uint8_t score;
    if (stage > VVC_TASK_STAGE_ALF)
        return 0;
    score = task_get_score(t, stage) + add;
    return task_has_target_score(t, stage, score);
}

static int task_ready(const AVTask *_t, void *user_data)
{
    VVCTask *t = (VVCTask*)_t;

    return task_is_stage_ready(t, 0);
}

#define CHECK(a, b)                         \
    do {                                    \
        if ((a) != (b))                     \
            return (a) < (b);               \
    } while (0)

static int task_priority_higher(const AVTask *_a, const AVTask *_b)
{
    const VVCTask *a = (const VVCTask*)_a;
    const VVCTask *b = (const VVCTask*)_b;

    CHECK(a->fc->decode_order, b->fc->decode_order);             //decode order

    if (a->stage == VVC_TASK_STAGE_PARSE || b->stage == VVC_TASK_STAGE_PARSE) {
        CHECK(a->stage, b->stage);
        CHECK(a->ry, b->ry);
        return a->rx < b->rx;
    }

    CHECK(a->rx + a->ry + a->stage, b->rx + b->ry + b->stage);    //zigzag with type
    CHECK(a->rx + a->ry, b->rx + b->ry);                          //zigzag
    return a->ry < b->ry;
}

static void report_frame_progress(VVCFrameContext *fc,
   const int ry, const VVCProgress idx)
{
    VVCFrameThread *ft = fc->ft;
    const int ctu_size = ft->ctu_size;
    int old;

    if (atomic_fetch_add(&ft->rows[ry].col_progress[idx], 1) == ft->ctu_width - 1) {
        int y;
        ff_mutex_lock(&ft->lock);
        y = old = ft->row_progress[idx];
        while (y < ft->ctu_height && atomic_load(&ft->rows[y].col_progress[idx]) == ft->ctu_width)
            y++;
        if (old != y) {
            const int progress = y == ft->ctu_height ? INT_MAX : y * ctu_size;
            ft->row_progress[idx] = y;
            ff_vvc_report_progress(fc->ref, idx, progress);
        }
        ff_mutex_unlock(&ft->lock);
    }
}

static int run_parse(VVCContext *s, VVCLocalContext *lc, VVCTask *t)
{
    int ret;
    VVCFrameContext *fc = lc->fc;
    const int rs        = t->rs;
    const CTU *ctu      = fc->tab.ctus + rs;

    lc->ep = t->ep;

    ret = ff_vvc_coding_tree_unit(lc, t->ctu_idx, rs, t->rx, t->ry);
    if (ret < 0)
        return ret;

    if (!ctu->has_dmvr)
        report_frame_progress(lc->fc, t->ry, VVC_PROGRESS_MV);

    return 0;
}

static int run_inter(VVCContext *s, VVCLocalContext *lc, VVCTask *t)
{
    VVCFrameContext *fc = lc->fc;
    const CTU *ctu      = fc->tab.ctus + t->rs;

    ff_vvc_predict_inter(lc, t->rs);

    if (ctu->has_dmvr)
        report_frame_progress(fc, t->ry, VVC_PROGRESS_MV);

    return 0;
}

static int run_recon(VVCContext *s, VVCLocalContext *lc, VVCTask *t)
{
    ff_vvc_reconstruct(lc, t->rs, t->rx, t->ry);

    return 0;
}

static int run_lmcs(VVCContext *s, VVCLocalContext *lc, VVCTask *t)
{
    VVCFrameContext *fc = lc->fc;
    VVCFrameThread *ft  = fc->ft;
    const int ctu_size  = ft->ctu_size;
    const int x0        = t->rx * ctu_size;
    const int y0        = t->ry * ctu_size;

    ff_vvc_lmcs_filter(lc, x0, y0);

    return 0;
}

static int run_deblock_v(VVCContext *s, VVCLocalContext *lc, VVCTask *t)
{
    VVCFrameContext *fc = lc->fc;
    VVCFrameThread *ft  = fc->ft;
    const int ctb_size  = ft->ctu_size;
    const int x0        = t->rx * ctb_size;
    const int y0        = t->ry * ctb_size;

    if (!lc->sc->sh.r->sh_deblocking_filter_disabled_flag) {
        ff_vvc_decode_neighbour(lc, x0, y0, t->rx, t->ry, t->rs);
        ff_vvc_deblock_vertical(lc, x0, y0, t->rs);
    }

    return 0;
}

static int run_deblock_h(VVCContext *s, VVCLocalContext *lc, VVCTask *t)
{
    VVCFrameContext *fc = lc->fc;
    VVCFrameThread *ft  = fc->ft;
    const int ctb_size  = ft->ctu_size;
    const int x0        = t->rx * ctb_size;
    const int y0        = t->ry * ctb_size;

    if (!lc->sc->sh.r->sh_deblocking_filter_disabled_flag) {
        ff_vvc_decode_neighbour(lc, x0, y0, t->rx, t->ry, t->rs);
        ff_vvc_deblock_horizontal(lc, x0, y0, t->rs);
    }
    if (fc->ps.sps->r->sps_sao_enabled_flag)
        ff_vvc_sao_copy_ctb_to_hv(lc, t->rx, t->ry, t->ry == ft->ctu_height - 1);

    return 0;
}

static int run_sao(VVCContext *s, VVCLocalContext *lc, VVCTask *t)
{
    VVCFrameContext *fc = lc->fc;
    VVCFrameThread *ft  = fc->ft;
    const int ctb_size  = ft->ctu_size;
    const int x0        = t->rx * ctb_size;
    const int y0        = t->ry * ctb_size;

    if (fc->ps.sps->r->sps_sao_enabled_flag) {
        ff_vvc_decode_neighbour(lc, x0, y0, t->rx, t->ry, t->rs);
        ff_vvc_sao_filter(lc, x0, y0);
    }

    if (fc->ps.sps->r->sps_alf_enabled_flag)
        ff_vvc_alf_copy_ctu_to_hv(lc, x0, y0);

    return 0;
}

static int run_alf(VVCContext *s, VVCLocalContext *lc, VVCTask *t)
{
    VVCFrameContext *fc = lc->fc;
    VVCFrameThread *ft  = fc->ft;
    const int ctu_size  = ft->ctu_size;
    const int x0        = t->rx * ctu_size;
    const int y0        = t->ry * ctu_size;

    if (fc->ps.sps->r->sps_alf_enabled_flag) {
        ff_vvc_decode_neighbour(lc, x0, y0, t->rx, t->ry, t->rs);
        ff_vvc_alf_filter(lc, x0, y0);
    }
    report_frame_progress(fc, t->ry, VVC_PROGRESS_PIXEL);

    return 0;
}

#define VVC_THREAD_DEBUG
#ifdef VVC_THREAD_DEBUG
const static char* task_name[] = {
    "P",
    "I",
    "R",
    "L",
    "V",
    "H",
    "S",
    "A"
};
#endif

typedef int (*run_func)(VVCContext *s, VVCLocalContext *lc, VVCTask *t);

static void task_run_stage(VVCTask *t, VVCContext *s, VVCLocalContext *lc)
{
    int ret;
    VVCFrameContext *fc      = t->fc;
    VVCFrameThread *ft       = fc->ft;
    const VVCTaskStage stage = t->stage;
    run_func run[] = {
        run_parse,
        run_inter,
        run_recon,
        run_lmcs,
        run_deblock_v,
        run_deblock_h,
        run_sao,
        run_alf,
    };

#ifdef VVC_THREAD_DEBUG
    av_log(s->avctx, AV_LOG_DEBUG, "frame %5d, %s(%3d, %3d)\r\n", (int)t->fc->decode_order, task_name[stage], t->rx, t->ry);
#endif

    lc->sc = t->sc;

    if (!atomic_load(&ft->ret)) {
        if ((ret = run[stage](s, lc, t)) < 0) {
#ifdef COMPAT_ATOMICS_WIN32_STDATOMIC_H
            intptr_t zero = 0;
#else
            int zero = 0;
#endif
            atomic_compare_exchange_strong(&ft->ret, &zero, ret);
            av_log(s->avctx, AV_LOG_ERROR,
                "frame %5d, %s(%3d, %3d) failed with %d\r\n",
                (int)fc->decode_order, task_name[stage], t->rx, t->ry, ret);
        }
    }

    task_stage_done(t, s);
    return;
}

static int task_run(AVTask *_t, void *local_context, void *user_data)
{
    VVCTask *t          = (VVCTask*)_t;
    VVCContext *s       = (VVCContext *)user_data;
    VVCLocalContext *lc = local_context;
    VVCFrameThread *ft  = t->fc->ft;

    lc->fc = t->fc;

    do {
        task_run_stage(t, s, lc);
        t->stage++;
    } while (task_is_stage_ready(t, 1));

    if (t->stage != VVC_TASK_STAGE_LAST)
        frame_thread_add_score(s, ft, t->rx, t->ry, t->stage);

    sheduled_done(ft, &ft->nb_scheduled_tasks);

    return 0;
}

AVExecutor* ff_vvc_executor_alloc(VVCContext *s, const int thread_count)
{
    AVTaskCallbacks callbacks = {
        s,
        sizeof(VVCLocalContext),
        task_priority_higher,
        task_ready,
        task_run,
    };
    return av_executor_alloc(&callbacks, thread_count);
}

void ff_vvc_executor_free(AVExecutor **e)
{
    av_executor_free(e);
}

void ff_vvc_frame_thread_free(VVCFrameContext *fc)
{
    VVCFrameThread *ft = fc->ft;

    if (!ft)
        return;

    ff_mutex_destroy(&ft->lock);
    ff_cond_destroy(&ft->cond);
    av_freep(&ft->rows);
    av_freep(&ft->tasks);
    av_freep(&ft);
}

static void frame_thread_init_score(VVCFrameContext *fc)
{
    const VVCFrameThread *ft = fc->ft;
    VVCTask task;

    task_init(&task, VVC_TASK_STAGE_RECON, fc, 0, 0);

    for (int i = VVC_TASK_STAGE_RECON; i < VVC_TASK_STAGE_LAST; i++) {
        task.stage = i;

        for (task.rx = -1; task.rx <= ft->ctu_width; task.rx++) {
            task.ry = -1;                           //top
            task_stage_done(&task, NULL);
            task.ry = ft->ctu_height;               //bottom
            task_stage_done(&task, NULL);
        }

        for (task.ry = 0; task.ry < ft->ctu_height; task.ry++) {
            task.rx = -1;                           //left
            task_stage_done(&task, NULL);
            task.rx = ft->ctu_width;                //right
            task_stage_done(&task, NULL);
        }
    }
}

int ff_vvc_frame_thread_init(VVCFrameContext *fc)
{
    const VVCSPS *sps  = fc->ps.sps;
    const VVCPPS *pps  = fc->ps.pps;
    VVCFrameThread *ft = fc->ft;
    int ret;

    if (!ft || ft->ctu_width != pps->ctb_width ||
        ft->ctu_height != pps->ctb_height ||
        ft->ctu_size != sps->ctb_size_y) {

        ff_vvc_frame_thread_free(fc);
        ft = av_calloc(1, sizeof(*fc->ft));
        if (!ft)
            return AVERROR(ENOMEM);

        ft->ctu_width  = fc->ps.pps->ctb_width;
        ft->ctu_height = fc->ps.pps->ctb_height;
        ft->ctu_count  = fc->ps.pps->ctb_count;
        ft->ctu_size   = fc->ps.sps->ctb_size_y;

        ft->rows = av_calloc(ft->ctu_height, sizeof(*ft->rows));
        if (!ft->rows)
            goto fail;

        ft->tasks = av_malloc(ft->ctu_count * sizeof(*ft->tasks));
        if (!ft->tasks)
            goto fail;

        if ((ret = ff_cond_init(&ft->cond, NULL)))
            goto fail;

        if ((ret = ff_mutex_init(&ft->lock, NULL))) {
            ff_cond_destroy(&ft->cond);
            goto fail;
        }
    }
    fc->ft = ft;
    ft->ret = 0;
    for (int y = 0; y < ft->ctu_height; y++) {
        VVCRowThread *row = ft->rows + y;
        memset(row->col_progress, 0, sizeof(row->col_progress));
    }

    for (int rs = 0; rs < ft->ctu_count; rs++) {
        VVCTask *t = ft->tasks + rs;
        task_init(t, VVC_TASK_STAGE_PARSE, fc, rs % ft->ctu_width, rs / ft->ctu_width);
    }

    memset(&ft->row_progress[0], 0, sizeof(ft->row_progress));

    frame_thread_init_score(fc);

    return 0;

fail:
    if (ft) {
        av_freep(&ft->rows);
        av_freep(&ft->tasks);
        av_freep(&ft);
    }

    return AVERROR(ENOMEM);
}

static void check_colocation(VVCContext *s, VVCTask *t)
{
    const VVCFrameContext *fc = t->fc;

    if (fc->ps.ph.r->ph_temporal_mvp_enabled_flag || fc->ps.sps->r->sps_sbtmvp_enabled_flag) {
        VVCFrame *col       = fc->ref->collocated_ref;
        const int first_col = t->rx == fc->ps.pps->ctb_to_col_bd[t->rx];
        if (col && first_col) {
            //we depend on bottom and right boundary, do not - 1 for y
            const int y = (t->ry << fc->ps.sps->ctb_log2_size_y);
            add_progress_listener(col, &t->col_listener, t, s, VVC_PROGRESS_MV, y);
            return;
        }
    }
    frame_thread_add_score(s, fc->ft, t->rx, t->ry, VVC_TASK_STAGE_PARSE);
}

static void submit_entry_point(VVCContext *s, VVCFrameThread *ft, SliceContext *sc, EntryPoint *ep)
{
    const int rs = sc->sh.ctb_addr_in_curr_slice[ep->ctu_start];
    VVCTask *t   = ft->tasks + rs;

    frame_thread_add_score(s, ft, t->rx, t->ry, VVC_TASK_STAGE_PARSE);
}

int ff_vvc_frame_submit(VVCContext *s, VVCFrameContext *fc)
{
    VVCFrameThread *ft = fc->ft;

    // We'll handle this in two passes:
    // Pass 0 to initialize tasks with parser, this will help detect bit stream error
    // Pass 1 to shedule location check and submit the entry point
    for (int pass = 0; pass < 2; pass++) {
        for (int i = 0; i < fc->nb_slices; i++) {
            SliceContext *sc = fc->slices[i];
            for (int j = 0; j < sc->nb_eps; j++) {
                EntryPoint *ep = sc->eps + j;
                for (int k = ep->ctu_start; k < ep->ctu_end; k++) {
                    const int rs = sc->sh.ctb_addr_in_curr_slice[k];
                    VVCTask *t   = ft->tasks + rs;
                    if (pass) {
                        check_colocation(s, t);
                    } else {
                        const int ret = task_init_parse(t, sc, ep, k);
                        if (ret < 0)
                            return ret;
                    }
                }
                if (pass)
                    submit_entry_point(s, ft, sc, ep);
            }
        }
    }
    return 0;
}

int ff_vvc_frame_wait(VVCContext *s, VVCFrameContext *fc)
{
    VVCFrameThread *ft = fc->ft;

    ff_mutex_lock(&ft->lock);

    while (atomic_load(&ft->nb_scheduled_tasks) || atomic_load(&ft->nb_scheduled_listeners))
        ff_cond_wait(&ft->cond, &ft->lock);

    ff_mutex_unlock(&ft->lock);
    ff_vvc_report_frame_finished(fc->ref);

#ifdef VVC_THREAD_DEBUG
    av_log(s->avctx, AV_LOG_DEBUG, "frame %5d done\r\n", (int)fc->decode_order);
#endif
    return ft->ret;
}