ffmpeg/libavcodec/vvc/thread.c

/*
 * 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 "libavcodec/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_INIT,                    // for CTU(0, 0) only
    VVC_TASK_STAGE_PARSE,
    VVC_TASK_STAGE_DEBLOCK_BS,
    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
        FFTask 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;

#define PRIORITY_LOWEST 2
static void add_task(VVCContext *s, VVCTask *t)
{
    VVCFrameThread *ft     = t->fc->ft;
    FFTask *task           = &t->u.task;
    const int priorities[] = {
        0,                  // VVC_TASK_STAGE_INIT,
        0,                  // VVC_TASK_STAGE_PARSE,
        1,                  // VVC_TASK_STAGE_DEBLOCK_BS
        // For an 8K clip, a CTU line completed in the reference frame may trigger 64 and more inter tasks.
        // We assign these tasks the lowest priority to avoid being overwhelmed with inter tasks.
        PRIORITY_LOWEST,    // VVC_TASK_STAGE_INTER
        1,                  // VVC_TASK_STAGE_RECON,
        1,                  // VVC_TASK_STAGE_LMCS,
        1,                  // VVC_TASK_STAGE_DEBLOCK_V,
        1,                  // VVC_TASK_STAGE_DEBLOCK_H,
        1,                  // VVC_TASK_STAGE_SAO,
        1,                  // VVC_TASK_STAGE_ALF,
    };

    atomic_fetch_add(&ft->nb_scheduled_tasks, 1);
    task->priority = priorities[t->stage];
    ff_executor_execute(s->executor, 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_DEBLOCK_BS,need l + t parse
        0,          //VVC_TASK_STAGE_INTER,     not used
        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_INIT)
        return 1;

    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);
        const int no_prev_stage  = t->rs > 0;
        target = 2 + wpp - no_prev_stage;                           //left parse + colocation + wpp - no_prev_stage
    } else if (stage == VVC_TASK_STAGE_INTER) {
        target = atomic_load(&t->target_inter_score);
    } else {
        target = target_score[stage - VVC_TASK_STAGE_DEBLOCK_BS];
    }

    //+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++) {
                int y = ctu->max_y[lx][i];
                VVCRefPic *refp = sc->rpl[lx].refs + i;
                VVCFrame *ref   = refp->ref;
                if (ref && y >= 0) {
                    if (refp->is_scaled)
                        y = y * refp->scale[1] >> 14;
                    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) {
        ADD( 0,  1, VVC_TASK_STAGE_DEBLOCK_BS);
        ADD( 1,  0, VVC_TASK_STAGE_DEBLOCK_BS);
        if (t->rx < 0 || t->rx >= ft->ctu_width || t->ry < 0 || t->ry >= ft->ctu_height)
            return;
        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 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);
}

static int run_init(VVCContext *s, VVCLocalContext *lc, VVCTask *t)
{
    VVCFrameContext *fc = lc->fc;
    VVCFrameThread *ft  = fc->ft;
    const int ret       = ff_vvc_per_frame_init(fc);

    if (ret < 0)
        return ret;

    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;
                check_colocation(s, t);
            }
            submit_entry_point(s, ft, sc, ep);
        }
    }
    return 0;
}

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)
            ft->row_progress[idx] = y;
        // ff_vvc_report_progress will acquire other frames' locks, which could lead to a deadlock
        // We need to unlock ft->lock first
        ff_mutex_unlock(&ft->lock);

        if (old != y) {
            const int progress = y == ft->ctu_height ? INT_MAX : y * ctu_size;
            ff_vvc_report_progress(fc->ref, idx, progress);
        }
    }
}

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_deblock_bs(VVCContext *s, VVCLocalContext *lc, VVCTask *t)
{
    if (!lc->sc->sh.r->sh_deblocking_filter_disabled_flag)
        ff_vvc_deblock_bs(lc, t->rx, t->ry, t->rs);

    return 0;
}

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

    ret = ff_vvc_predict_inter(lc, t->rs);
    if (ret < 0)
        return ret;

    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)
{
    return ff_vvc_reconstruct(lc, t->rs, t->rx, t->ry);
}

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[] = {
    "INIT",
    "P",
    "B",
    "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;
    static const run_func run[] = {
        run_init,
        run_parse,
        run_deblock_bs,
        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);
        }
        if (!ret)
            task_stage_done(t, s);
    }
    return;
}

static int task_run(FFTask *_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;
}

FFExecutor* ff_vvc_executor_alloc(VVCContext *s, const int thread_count)
{
    FFTaskCallbacks callbacks = {
        s,
        sizeof(VVCLocalContext),
        PRIORITY_LOWEST + 1,
        task_run,
    };
    return ff_executor_alloc(&callbacks, thread_count);
}

void ff_vvc_executor_free(FFExecutor **e)
{
    ff_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_PARSE, fc, 0, 0);

    for (int i = VVC_TASK_STAGE_PARSE; 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, rs ? VVC_TASK_STAGE_PARSE : VVC_TASK_STAGE_INIT, 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);
}

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

    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;
                const int ret = task_init_parse(t, sc, ep, k);
                if (ret < 0)
                    return ret;
            }
        }
    }
    for (int rs = 0; rs < ft->ctu_count; rs++) {
        const VVCTask *t = ft->tasks + rs;
        if (!t->sc) {
            av_log(s->avctx, AV_LOG_ERROR, "frame %5d, CTU(%d, %d) not belong to any slice\r\n", (int)fc->decode_order, t->rx, t->ry);
            return AVERROR_INVALIDDATA;
        }
    }
    frame_thread_add_score(s, ft, 0, 0, VVC_TASK_STAGE_INIT);

    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;
}