libswscale/graph.c - master - githubmirror/FFmpeg

libswscale/graph.c
1 053 строки33 KB
Niklas Haas
swscale/graph: internally ref copied intermediate buffers
09 авг 2026, 22:03
09 авг 2026, 22:035ca18fd
/*
 * Copyright (C) 2024 Niklas Haas
 *
 * 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 "libavutil/avassert.h"
#include "libavutil/cpu.h"
#include "libavutil/error.h"
#include "libavutil/hwcontext.h"
#include "libavutil/imgutils.h"
#include "libavutil/macros.h"
#include "libavutil/mem.h"
#include "libavutil/opt.h"
#include "libavutil/pixdesc.h"
#include "libavutil/refstruct.h"
#include "libavutil/slicethread.h"

#include "libswscale/swscale.h"
#include "libswscale/format.h"

#include "cms.h"
#include "lut3d.h"
#include "swscale_internal.h"
#include "graph.h"
#include "ops.h"
#include "ops_dispatch.h"
#if CONFIG_VULKAN
#include "vulkan/ops.h"
#endif

int ff_sws_pass_aligned_width(const SwsPass *pass, int width)
{
    if (!pass)
        return width;

    size_t aligned_w = width;
    aligned_w = FFALIGN(aligned_w, pass->output->width_align);
    aligned_w += pass->output->width_pad;
    return aligned_w <= INT_MAX ? aligned_w : width;
}

/* Allocates (or refs) one buffer per plane */
static int frame_alloc_planes_ref(AVFrame *dst, const AVFrame *src,
                                  const int plane_copy[4])
{
    int ret = av_image_check_size2(dst->width, dst->height, INT64_MAX,
                                   dst->format, 0, NULL);
    if (ret < 0)
        return ret;

    const int align = av_cpu_max_align();
    const int aligned_w = FFALIGN(dst->width + 1, align); /* add space for over-write */
    ret = av_image_fill_linesizes(dst->linesize, dst->format, aligned_w);
    if (ret < 0)
        return ret;

    ptrdiff_t linesize1[4];
    for (int i = 0; i < 4; i++)
        linesize1[i] = dst->linesize[i] = FFALIGN(dst->linesize[i], align);

    size_t sizes[4];
    ret = av_image_fill_plane_sizes(sizes, dst->format, dst->height, linesize1);
    if (ret < 0)
        return ret;

    for (int i = 0; i < 4; i++) {
        if (!sizes[i])
            break;
        int src_idx = plane_copy[i];
        if (src_idx >= 0 && src) {
            /* Ref the source plane instead of allocating a new buffer */
            dst->buf[i] = av_buffer_ref(src->buf[src_idx]);
            if (!dst->buf[i])
                return AVERROR(ENOMEM);
            dst->data[i]     = src->data[src_idx];
            dst->linesize[i] = src->linesize[src_idx];
            continue;
        }

        AVBufferRef *buf = av_buffer_alloc(sizes[i]);
        if (!buf)
            return AVERROR(ENOMEM);
        dst->data[i] = buf->data;
        dst->buf[i] = buf;
    }

    return 0;
}

#if CONFIG_VULKAN
static int pass_alloc_output_hw(SwsPass *pass, AVFrame *avframe,
                                AVBufferRef *dev_ref)
{
    SwsPassBuffer *buffer = pass->output;
    AVBufferRef *frames_ref = av_hwframe_ctx_alloc(dev_ref);
    if (!frames_ref)
        return AVERROR(ENOMEM);

    AVHWFramesContext *hwfc = (AVHWFramesContext *)frames_ref->data;
    hwfc->format    = AV_PIX_FMT_VULKAN;
    hwfc->sw_format = pass->format;
    hwfc->width     = buffer->width;
    hwfc->height    = buffer->height;

    int ret = av_hwframe_ctx_init(frames_ref);
    if (ret >= 0) {
        avframe->format = AV_PIX_FMT_VULKAN;
        ret = av_hwframe_get_buffer(frames_ref, avframe, 0);
    }
    av_buffer_unref(&frames_ref);
    return ret;
}
#endif

static int pass_alloc_output(SwsPass *pass)
{
    if (!pass || pass->output->avframe)
        return 0;

    SwsPassBuffer *buffer = pass->output;
    AVFrame *avframe = av_frame_alloc();
    if (!avframe)
        return AVERROR(ENOMEM);
    avframe->width  = buffer->width;
    avframe->height = buffer->height;

    int ret;

#if CONFIG_VULKAN
    const SwsGraph *graph = pass->graph;
    if (graph->src.hw_format == AV_PIX_FMT_VULKAN &&
        graph->dst.hw_format == AV_PIX_FMT_VULKAN) {
        AVBufferRef *dev_ref = ff_sws_vk_device_ref(graph->ctx);
        if (dev_ref) {
            ret = pass_alloc_output_hw(pass, avframe, dev_ref);
            if (ret >= 0)
                goto done;
            av_frame_unref(avframe);
        }
    }
#endif

    const AVFrame *src = NULL;
    if (pass->input)
        src = pass->input->output->avframe;

    avframe->format = pass->format;
    ret = frame_alloc_planes_ref(avframe, src, buffer->plane_copy);
    if (ret < 0) {
        av_frame_free(&avframe);
        return ret;
    }

#if CONFIG_VULKAN
done:
#endif
    buffer->avframe = avframe;
    ff_sws_frame_from_avframe(&buffer->frame, avframe);
    return 0;
}

static void free_buffer(AVRefStructOpaque opaque, void *obj)
{
    SwsPassBuffer *buffer = obj;
    av_frame_free(&buffer->avframe);
}

static void pass_free(SwsPass *pass)
{
    if (pass->free)
        pass->free(pass->priv);
    av_refstruct_unref(&pass->output);
    av_free(pass);
}

int ff_sws_graph_add_pass(SwsGraph *graph, enum AVPixelFormat fmt,
                          int width, int height, SwsPass *input,
                          int lines, int align,
                          SwsPassFunc run, SwsPassSetup setup,
                          void *priv, void (*free_cb)(void *priv),
                          SwsPass **out_pass)
{
    int ret;
    SwsPass *pass = av_mallocz(sizeof(*pass));
    if (!pass) {
        if (free_cb)
            free_cb(priv);
        return AVERROR(ENOMEM);
    }

    if (!lines)
        lines = height;

    pass->graph  = graph;
    pass->run    = run;
    pass->setup  = setup;
    pass->priv   = priv;
    pass->free   = free_cb;
    pass->format = fmt;
    pass->lines  = lines;
    pass->input  = input;
    pass->output = av_refstruct_alloc_ext(sizeof(*pass->output), 0, NULL, free_buffer);
    if (!pass->output) {
        ret = AVERROR(ENOMEM);
        goto fail;
    }

    pass->output->height = height;
    pass->output->width  = width;
    pass->output->width_align = 1;
    memset(pass->output->plane_copy, -1, sizeof(pass->output->plane_copy));

    if (!align) {
        pass->slice_h = pass->lines;
        pass->num_slices = 1;
    } else {
        pass->slice_h = (pass->lines + graph->num_threads - 1) / graph->num_threads;
        pass->slice_h = FFALIGN(pass->slice_h, align);
        pass->num_slices = (pass->lines + pass->slice_h - 1) / pass->slice_h;
    }

    ret = av_dynarray_add_nofree(&graph->passes, &graph->num_passes, pass);
    if (ret < 0)
        goto fail;

    *out_pass = pass;
    return 0;

fail:
    pass_free(pass);
    return ret;
}

void ff_sws_pass_link_output(SwsPass *dst, const SwsPass *src)
{
    if (!dst || !src || dst == src)
        return;

    av_assert0(dst->format == src->format);
    SwsPassBuffer *keep = src->output, *drop = dst->output;

    av_assert1(keep->width  == drop->width);
    av_assert1(keep->height == drop->height);
    keep->width_align = FFMAX(keep->width_align, drop->width_align);
    keep->width_pad   = FFMAX(keep->width_pad,   drop->width_pad);

    for (int i = 0; i < FF_ARRAY_ELEMS(keep->plane_copy); i++) {
        if (keep->plane_copy[i] < 0)
            keep->plane_copy[i] = drop->plane_copy[i];
        else if (drop->plane_copy[i] >= 0)
            av_assert1(keep->plane_copy[i] == drop->plane_copy[i]);
    }

    av_refstruct_replace(&dst->output, src->output);
}

static void frame_shift(const SwsFrame *f, const int y, uint8_t *data[4])
{
    for (int i = 0; i < 4; i++) {
        if (f->data[i])
            data[i] = f->data[i] + (y >> ff_fmt_vshift(f->format, i)) * f->linesize[i];
        else
            data[i] = NULL;
    }
}

static void run_copy(const SwsFrame *out, const SwsFrame *in, int y, int h,
                     const SwsPass *pass)
{
    uint8_t *in_data[4], *out_data[4];
    frame_shift(in,  y, in_data);
    frame_shift(out, y, out_data);

    for (int i = 0; i < 4 && out_data[i]; i++) {
        const int lines = h >> ff_fmt_vshift(in->format, i);
        av_assert1(in_data[i]);

        if (in_data[i] == out_data[i]) {
            av_assert0(in->linesize[i] == out->linesize[i]);
        } else if (in->linesize[i] == out->linesize[i]) {
            memcpy(out_data[i], in_data[i], lines * out->linesize[i]);
        } else {
            const int linesize = FFMIN(out->linesize[i], in->linesize[i]);
            for (int j = 0; j < lines; j++) {
                memcpy(out_data[i], in_data[i], linesize);
                in_data[i]  += in->linesize[i];
                out_data[i] += out->linesize[i];
            }
        }
    }
}

static void run_rgb0(const SwsFrame *out, const SwsFrame *in, int y, int h,
                     const SwsPass *pass)
{
    SwsInternal *c = pass->priv;
    const int x0 = c->src0Alpha - 1;
    const int w4 = 4 * out->width;
    const int src_stride = in->linesize[0];
    const int dst_stride = out->linesize[0];
    const uint8_t *src = in->data[0] + y * src_stride;
    uint8_t *dst = out->data[0] + y * dst_stride;

    for (int y = 0; y < h; y++) {
        memcpy(dst, src, w4 * sizeof(*dst));
        for (int x = x0; x < w4; x += 4)
            dst[x] = 0xFF;

        src += src_stride;
        dst += dst_stride;
    }
}

static void run_xyz2rgb(const SwsFrame *out, const SwsFrame *in, int y, int h,
                        const SwsPass *pass)
{
    const SwsInternal *c = pass->priv;
    c->xyz12Torgb48(c, out->data[0] + y * out->linesize[0], out->linesize[0],
                    in->data[0] + y * in->linesize[0], in->linesize[0],
                    out->width, h);
}

static void run_rgb2xyz(const SwsFrame *out, const SwsFrame *in, int y, int h,
                        const SwsPass *pass)
{
    const SwsInternal *c = pass->priv;
    c->rgb48Toxyz12(c, out->data[0] + y * out->linesize[0], out->linesize[0],
                    in->data[0] + y * in->linesize[0], in->linesize[0],
                    out->width, h);
}

/***********************************************************************
 * Internal ff_swscale() wrapper. This reuses the legacy scaling API. *
 * This is considered fully deprecated, and will be replaced by a full *
 * reimplementation ASAP.                                              *
 ***********************************************************************/

static void free_legacy_swscale(void *priv)
{
    SwsContext *sws = priv;
    sws_free_context(&sws);
}

static int setup_legacy_swscale(const SwsFrame *out, const SwsFrame *in,
                                const SwsPass *pass)
{
    SwsContext *sws = pass->priv;
    SwsInternal *c = sws_internal(sws);
    if (sws->flags & SWS_BITEXACT && sws->dither == SWS_DITHER_ED && c->dither_error[0]) {
        for (int i = 0; i < 4; i++)
            memset(c->dither_error[i], 0, sizeof(c->dither_error[0][0]) * (sws->dst_w + 2));
    }

    if (usePal(sws->src_format))
        ff_update_palette(c, (const uint32_t *) in->data[1]);

    return 0;
}

static inline SwsContext *slice_ctx(const SwsPass *pass, int y)
{
    SwsContext *sws = pass->priv;
    SwsInternal *parent = sws_internal(sws);
    if (pass->num_slices == 1)
        return sws;

    av_assert1(parent->nb_slice_ctx == pass->num_slices);
    sws = parent->slice_ctx[y / pass->slice_h];

    if (usePal(sws->src_format)) {
        SwsInternal *sub = sws_internal(sws);
        memcpy(sub->pal_yuv, parent->pal_yuv, sizeof(sub->pal_yuv));
        memcpy(sub->pal_rgb, parent->pal_rgb, sizeof(sub->pal_rgb));
    }

    return sws;
}

static void run_legacy_unscaled(const SwsFrame *out, const SwsFrame *in,
                                int y, int h, const SwsPass *pass)
{
    SwsContext *sws = slice_ctx(pass, y);
    SwsInternal *c = sws_internal(sws);
    uint8_t *in_data[4];
    frame_shift(in, y, in_data);

    c->convert_unscaled(c, (const uint8_t *const *) in_data, in->linesize, y, h,
                        out->data, out->linesize);
}

static void run_legacy_swscale(const SwsFrame *out, const SwsFrame *in,
                               int y, int h, const SwsPass *pass)
{
    SwsContext *sws = slice_ctx(pass, y);
    SwsInternal *c = sws_internal(sws);
    uint8_t *out_data[4];
    frame_shift(out, y, out_data);

    ff_swscale(c, (const uint8_t *const *) in->data, in->linesize, 0,
               sws->src_h, out_data, out->linesize, y, h);
}

static void run_legacy_lut3d(const SwsFrame *out, const SwsFrame *in,
                             int y, int h, const SwsPass *pass)
{
    const SwsLut3D *lut = pass->graph->lut3d;
    uint8_t *in_data[4], *out_data[4];
    frame_shift(in,  y, in_data);
    frame_shift(out, y, out_data);

    ff_sws_lut3d_apply_rgba64(lut, in_data[0], in->linesize[0], out_data[0],
                              out->linesize[0], out->width, h);
}

static void legacy_chr_pos(SwsGraph *graph, int *chr_pos, int override, int *warned)
{
    if (override == -513 || override == *chr_pos)
        return;

    if (!*warned) {
        av_log(NULL, AV_LOG_WARNING,
               "Setting chroma position directly is deprecated, make sure "
               "the frame is tagged with the correct chroma location.\n");
        *warned = 1;
    }

    *chr_pos = override;
}

/* Takes over ownership of `sws` */
static int init_legacy_subpass(SwsGraph *graph, SwsContext *sws,
                               SwsPass *input, SwsPass **output)
{
    SwsInternal *c = sws_internal(sws);
    const int src_w = sws->src_w, src_h = sws->src_h;
    const int dst_w = sws->dst_w, dst_h = sws->dst_h;
    const int unscaled = src_w == dst_w && src_h == dst_h;
    int align = c->dst_slice_align;
    SwsPass *pass = NULL;
    int ret;

    if (c->cascaded_context[0]) {
        const int num_cascaded = c->cascaded_context[2] ? 3 : 2;
        for (int i = 0; i < num_cascaded; i++) {
            const int is_last = i + 1 == num_cascaded;

            /* Steal cascaded context, so we can manage its lifetime independently */
            SwsContext *sub = c->cascaded_context[i];
            c->cascaded_context[i] = NULL;

            ret = init_legacy_subpass(graph, sub, input, is_last ? output : &input);
            if (ret < 0)
                break;
        }

        sws_free_context(&sws);
        return ret;
    }

    if (sws->dither == SWS_DITHER_ED && !c->convert_unscaled)
        align = 0; /* disable slice threading */

    if (c->src0Alpha && !c->dst0Alpha && isALPHA(sws->dst_format)) {
        ret = ff_sws_graph_add_pass(graph, AV_PIX_FMT_RGBA, src_w, src_h, input,
                                    0, 1, run_rgb0, NULL, c, NULL, &input);
        if (ret < 0) {
            sws_free_context(&sws);
            return ret;
        }
    }

    if (c->srcXYZ && !(c->dstXYZ && unscaled)) {
        ret = ff_sws_graph_add_pass(graph, AV_PIX_FMT_RGB48, src_w, src_h, input,
                                    0, 1, run_xyz2rgb, NULL, c, NULL, &input);
        if (ret < 0) {
            sws_free_context(&sws);
            return ret;
        }
    }

    ret = ff_sws_graph_add_pass(graph, sws->dst_format, dst_w, dst_h, input, 0, align,
                                c->convert_unscaled ? run_legacy_unscaled : run_legacy_swscale,
                                setup_legacy_swscale, sws, free_legacy_swscale, &pass);
    if (ret < 0)
        return ret;
    pass->backend = SWS_BACKEND_LEGACY;

    /**
     * For slice threading, we need to create sub contexts, similar to how
     * swscale normally handles it internally. The most important difference
     * is that we handle cascaded contexts before threaded contexts; whereas
     * context_init_threaded() does it the other way around.
     */

    if (pass->num_slices > 1) {
        c->slice_ctx = av_calloc(pass->num_slices, sizeof(*c->slice_ctx));
        if (!c->slice_ctx)
            return AVERROR(ENOMEM);

        for (int i = 0; i < pass->num_slices; i++) {
            SwsContext *slice;
            SwsInternal *c2;
            slice = c->slice_ctx[i] = sws_alloc_context();
            if (!slice)
                return AVERROR(ENOMEM);
            c->nb_slice_ctx++;

            c2 = sws_internal(slice);
            c2->parent = sws;

            ret = av_opt_copy(slice, sws);
            if (ret < 0)
                return ret;

            ret = ff_sws_init_single_context(slice, NULL, NULL);
            if (ret < 0)
                return ret;

            sws_setColorspaceDetails(slice, c->srcColorspaceTable,
                                     slice->src_range, c->dstColorspaceTable,
                                     slice->dst_range, c->brightness, c->contrast,
                                     c->saturation);

            for (int i = 0; i < FF_ARRAY_ELEMS(c->srcColorspaceTable); i++) {
                c2->srcColorspaceTable[i] = c->srcColorspaceTable[i];
                c2->dstColorspaceTable[i] = c->dstColorspaceTable[i];
            }
        }
    }

    if (c->dstXYZ && !(c->srcXYZ && unscaled)) {
        ret = ff_sws_graph_add_pass(graph, AV_PIX_FMT_RGB48, dst_w, dst_h, pass,
                                    0, 1, run_rgb2xyz, NULL, c, NULL, &pass);
        if (ret < 0)
            return ret;
    }

    *output = pass;
    return 0;
}

static int add_legacy_3dlut_pass(SwsGraph *graph, const SwsFormat *src,
                                 SwsPass *input, SwsPass **output);

static int add_legacy_sws_pass(SwsGraph *graph, const SwsFormat *src,
                               const SwsFormat *dst, const SwsLut3D *lut3d,
                               SwsPass *input, SwsPass **output)
{
    int ret, warned = 0;
    SwsContext *const ctx = graph->ctx;
    const SwsBackend backend = ff_sws_enabled_backends(ctx);
    if (!(backend & SWS_BACKEND_LEGACY))
        return AVERROR(ENOTSUP);
    if (src->hw_format != AV_PIX_FMT_NONE || dst->hw_format != AV_PIX_FMT_NONE)
        return AVERROR(ENOTSUP);

    /* Re-check this here because this might not be excluded if the caller was
     * testing against multiple backends */
    if (!sws_isSupportedInput(src->format) || !sws_isSupportedOutput(dst->format))
        return AVERROR(ENOTSUP);

    /* If we need to apply a 3D LUT, add it as an explicit input prepass */
    if (lut3d) {
        ret = add_legacy_3dlut_pass(graph, src, input, &input);
        if (ret < 0)
            return ret;

        SwsFormat tmp = *src;
        tmp.format = input->format;
        tmp.color  = lut3d->map.dst;
        return add_legacy_sws_pass(graph, &tmp, dst, NULL, input, output);
    }

    SwsContext *sws = sws_alloc_context();
    if (!sws)
        return AVERROR(ENOMEM);

    sws->flags       = ctx->flags;
    sws->dither      = ctx->dither;
    sws->alpha_blend = ctx->alpha_blend;
    sws->gamma_flag  = ctx->gamma_flag;
    sws->scaler      = ctx->scaler;
    sws->scaler_sub  = ctx->scaler_sub;

    sws->src_w       = src->width;
    sws->src_h       = src->height;
    sws->src_format  = src->format;
    sws->src_range   = src->range == AVCOL_RANGE_JPEG;

    sws->dst_w      = dst->width;
    sws->dst_h      = dst->height;
    sws->dst_format = dst->format;
    sws->dst_range  = dst->range == AVCOL_RANGE_JPEG;
    ff_sws_chroma_pos(src, &graph->incomplete, &sws->src_h_chr_pos, &sws->src_v_chr_pos);
    ff_sws_chroma_pos(dst, &graph->incomplete, &sws->dst_h_chr_pos, &sws->dst_v_chr_pos);

    graph->incomplete |= src->range == AVCOL_RANGE_UNSPECIFIED;
    graph->incomplete |= dst->range == AVCOL_RANGE_UNSPECIFIED;

    /* Allow overriding chroma position with the legacy API */
    legacy_chr_pos(graph, &sws->src_h_chr_pos, ctx->src_h_chr_pos, &warned);
    legacy_chr_pos(graph, &sws->src_v_chr_pos, ctx->src_v_chr_pos, &warned);
    legacy_chr_pos(graph, &sws->dst_h_chr_pos, ctx->dst_h_chr_pos, &warned);
    legacy_chr_pos(graph, &sws->dst_v_chr_pos, ctx->dst_v_chr_pos, &warned);

    /* Explicitly strip chroma offsets when not subsampling, because it
     * interferes with the operation of flags like SWS_FULL_CHR_H_INP */
    if (!src->desc->log2_chroma_w)
        sws->src_h_chr_pos = -513;
    if (!src->desc->log2_chroma_h)
        sws->src_v_chr_pos = -513;
    if (!dst->desc->log2_chroma_w)
        sws->dst_h_chr_pos = -513;
    if (!dst->desc->log2_chroma_h)
        sws->dst_v_chr_pos = -513;

    for (int i = 0; i < SWS_NUM_SCALER_PARAMS; i++)
        sws->scaler_params[i] = ctx->scaler_params[i];

    ret = sws_init_context(sws, NULL, NULL);
    if (ret < 0) {
        sws_free_context(&sws);
        return ret;
    }

    /* Set correct color matrices */
    {
        int in_full, out_full, brightness, contrast, saturation;
        const int *inv_table, *table;
        sws_getColorspaceDetails(sws, (int **)&inv_table, &in_full,
                                (int **)&table, &out_full,
                                &brightness, &contrast, &saturation);

        inv_table = sws_getCoefficients(src->csp);
        table     = sws_getCoefficients(dst->csp);

        graph->incomplete |= src->csp != dst->csp &&
                            (src->csp == AVCOL_SPC_UNSPECIFIED ||
                             dst->csp == AVCOL_SPC_UNSPECIFIED);

        sws_setColorspaceDetails(sws, inv_table, in_full, table, out_full,
                                brightness, contrast, saturation);
    }

    return init_legacy_subpass(graph, sws, input, output);
}

static int add_legacy_3dlut_pass(SwsGraph *graph, const SwsFormat *src,
                                 SwsPass *input, SwsPass **output)
{
    int ret;

    const SwsLut3D *lut3d = graph->lut3d;
    if (!lut3d)
        return 0;

    const enum AVPixelFormat fmt = AV_PIX_FMT_RGBA64;
    if (src->format != fmt) {
        SwsFormat tmp = *src;
        tmp.format = fmt;
        ret = add_legacy_sws_pass(graph, src, &tmp, NULL, input, &input);
        if (ret < 0)
            return ret;
    }

    ret = ff_sws_graph_add_pass(graph, fmt, src->width, src->height,
                                input, 0, 1, run_legacy_lut3d, NULL, NULL, NULL,
                                output);
    if (ret < 0)
        return ret;

    return 0;
}

/*********************************
 * Format conversion and scaling *
 *********************************/

static int add_ops_convert_pass(SwsGraph *graph, const SwsFormat *src,
                                const SwsFormat *dst, const SwsLut3D *lut3d,
                                SwsPass *input, SwsPass **output)
{
#if CONFIG_UNSTABLE
    SwsContext *ctx = graph->ctx;

    /* Preemptively skip the ops list generation if the backend was
     * constrained to the legacy implementation only. This would
     * normally also fail in ff_sws_compile_pass() with the same
     * error, but this way saves a bit of unnecessary overhead */
    const SwsBackend backends = ff_sws_enabled_backends(ctx);
    if (backends == SWS_BACKEND_LEGACY)
        return AVERROR(ENOTSUP);

    SwsOpList *ops;
    int ret = ff_sws_op_list_generate(ctx, src, dst, lut3d, &ops, &graph->incomplete);
    if (ret < 0)
        return ret;

    av_log(ctx, AV_LOG_VERBOSE, "Conversion pass for %s -> %s:\n",
           av_get_pix_fmt_name(src->format), av_get_pix_fmt_name(dst->format));

    av_log(ctx, AV_LOG_DEBUG, "Unoptimized operation list:\n");
    ff_sws_op_list_print(ctx, AV_LOG_DEBUG, AV_LOG_TRACE, ops);

    const int flags = SWS_OP_FLAG_OPTIMIZE | SWS_OP_FLAG_SPLIT_MEMCPY;
    return ff_sws_compile_pass(graph, NULL, &ops, flags, input, output);
#else
    return AVERROR(ENOTSUP);
#endif
}

static bool prefer_ops_backend(SwsContext *ctx, const SwsFormat *src, const SwsFormat *dst)
{
    if (ctx->flags & SWS_UNSTABLE)
        return true;
    if (isFloat(src->format) || isFloat(dst->format))
        return true; /* ops backend has better support for float formats */
    return false; /* default to legacy for stability reasons */
}

static int add_convert_pass(SwsGraph *graph, const SwsFormat *src,
                            const SwsFormat *dst, const SwsLut3D *lut3d,
                            SwsPass *input, SwsPass **output)
{
    SwsContext *ctx = graph->ctx;
    int ret;

    if (prefer_ops_backend(ctx, src, dst)) {
        ret = add_ops_convert_pass(graph, src, dst, lut3d, input, output);
        if (ret == AVERROR(ENOTSUP))
            ret = add_legacy_sws_pass(graph, src, dst, lut3d, input, output);
    } else {
        ret = add_legacy_sws_pass(graph, src, dst, lut3d, input, output);
        if (ret == AVERROR(ENOTSUP))
            ret = add_ops_convert_pass(graph, src, dst, lut3d, input, output);
    }

    return ret;
}

/**************************
 * Gamut and tone mapping *
 **************************/

static int generate_3dlut(SwsGraph *graph, SwsFormat *src, SwsFormat *dst)
{
    SwsColorMap map = {0};

    /**
     * Grayspace does not really have primaries, so just force the use of
     * the equivalent other primary set to avoid a conversion. Technically,
     * this does affect the weights used for the Grayscale conversion, but
     * in practise, that should give the expected results more often than not.
     */
    if (isGray(dst->format)) {
        dst->color = src->color;
    } else if (isGray(src->format)) {
        src->color = dst->color;
    }

    /* Fully infer color spaces before color mapping logic */
    graph->incomplete |= ff_infer_colors(&src->color, &dst->color);

    map.intent = graph->ctx->intent;
    map.src    = src->color;
    map.dst    = dst->color;

    if (ff_sws_color_map_noop(&map))
        return 0;

    if (src->hw_format != AV_PIX_FMT_NONE || dst->hw_format != AV_PIX_FMT_NONE)
        return AVERROR(ENOTSUP);

    graph->lut3d = ff_sws_lut3d_alloc();
    if (!graph->lut3d)
        return AVERROR(ENOMEM);

    return ff_sws_lut3d_generate(graph->lut3d, &map);
}

/***************************************
 * Main filter graph construction code *
 ***************************************/

static int init_passes(SwsGraph *graph)
{
    SwsFormat src = graph->src;
    SwsFormat dst = graph->dst;
    SwsPass *pass = NULL; /* read from main input image */
    int ret;

    ret = generate_3dlut(graph, &src, &dst);
    if (ret < 0)
        return ret;

    if (!ff_fmt_equal(&src, &dst) || graph->lut3d) {
        ret = add_convert_pass(graph, &src, &dst, graph->lut3d, pass, &pass);
        if (ret < 0)
            return ret;
    }

    if (!pass) {
        /* No passes were added, so no operations were necessary */
        graph->noop = 1;

        const int nb_planes = av_pix_fmt_count_planes(dst.format);
        for (int i = 0; i < nb_planes; i++)
            graph->plane_copy[i] = i;

        /* Add threaded memcpy pass */
        return ff_sws_graph_add_pass(graph, dst.format, dst.width, dst.height,
                                    pass, 0, 1, run_copy, NULL, NULL, NULL, &pass);
    }

    /* Compute end-to-end plane copy map */
    for (int n = 0; n < graph->num_passes; n++) {
        const SwsPass *pass = graph->passes[n];
        /* This pass writes to an output buffer other than the image
         * output, or copies from the output of a different pass */
        if (pass->output->avframe || pass->input)
            continue;
        for (int i = 0; i < FF_ARRAY_ELEMS(graph->plane_copy); i++) {
            const int idx = pass->output->plane_copy[i];
            if (idx < 0)
                continue;
            if (graph->plane_copy[i] < 0) {
                graph->plane_copy[i] = idx;
                av_log(graph->ctx, AV_LOG_DEBUG, "Plane %d passthrough from "
                       "plane %d\n", i, idx);
            } else {
                av_assert0(graph->plane_copy[i] == idx);
            }
        }
    }

    return 0;
}

static void sws_graph_worker(void *priv, int jobnr, int threadnr, int nb_jobs,
                             int nb_threads)
{
    SwsGraph *graph = priv;
    const SwsPass *pass = graph->exec.pass;
    const int slice_y = jobnr * pass->slice_h;
    const int slice_h = FFMIN(pass->slice_h, pass->lines - slice_y);

    pass->run(graph->exec.output, graph->exec.input, slice_y, slice_h, pass);
}

SwsGraph *ff_sws_graph_alloc(void)
{
    return av_mallocz(sizeof(SwsGraph));
}

static void graph_uninit(SwsGraph *graph)
{
    avpriv_slicethread_free(&graph->slicethread);

    for (int i = 0; i < graph->num_passes; i++)
        pass_free(graph->passes[i]);
    av_free(graph->passes);

    av_refstruct_unref(&graph->lut3d);

    memset(graph, 0, sizeof(*graph));
}

int ff_sws_graph_init(SwsGraph *graph, SwsContext *ctx, const SwsFormat *dst,
                      const SwsFormat *src)
{
    int ret;
    if (graph->ctx) {
        av_log(ctx, AV_LOG_ERROR, "Graph is already initialized\n");
        return AVERROR(EINVAL);
    }

    graph->ctx = ctx;
    graph->src = *src;
    graph->dst = *dst;
    graph->opts_copy = *ctx;
    av_assert0(src->interlaced == dst->interlaced);
    av_assert0(src->field      == dst->field);
    memset(graph->plane_copy, -1, sizeof(graph->plane_copy));

    if (ctx->threads == 1) {
        graph->num_threads = 1;
    } else {
        ret = avpriv_slicethread_create(&graph->slicethread, (void *) graph,
                                        sws_graph_worker, NULL, ctx->threads);
        if (ret == AVERROR(ENOSYS)) {
            /* Fall back to single threaded operation */
            graph->num_threads = 1;
        } else if (ret < 0) {
            goto error;
        } else {
            graph->num_threads = ret;
        }
    }

    ret = init_passes(graph);
    if (ret < 0)
        goto error;

    /* Resolve output buffers for all intermediate passes */
    for (int i = 0; i < graph->num_passes; i++) {
        graph->backend |= graph->passes[i]->backend;
        ret = pass_alloc_output(graph->passes[i]->input);
        if (ret < 0)
            goto error;
    }

    return 0;

error:
    graph_uninit(graph);
    return ret;
}

void ff_sws_graph_rollback(SwsGraph *graph, int since_idx)
{
    for (int i = since_idx; i < graph->num_passes; i++)
        pass_free(graph->passes[i]);
    graph->num_passes = since_idx;
}

void ff_sws_graph_free(SwsGraph **pgraph)
{
    SwsGraph *graph = *pgraph;
    if (!graph)
        return;

    graph_uninit(graph);
    av_free(graph);
    *pgraph = NULL;
}

/* Tests only options relevant to SwsGraph */
static int opts_equal(const SwsContext *c1, const SwsContext *c2)
{
    return c1->flags         == c2->flags         &&
           c1->threads       == c2->threads       &&
           c1->dither        == c2->dither        &&
           c1->alpha_blend   == c2->alpha_blend   &&
           c1->gamma_flag    == c2->gamma_flag    &&
           c1->src_h_chr_pos == c2->src_h_chr_pos &&
           c1->src_v_chr_pos == c2->src_v_chr_pos &&
           c1->dst_h_chr_pos == c2->dst_h_chr_pos &&
           c1->dst_v_chr_pos == c2->dst_v_chr_pos &&
           c1->intent        == c2->intent        &&
           c1->scaler        == c2->scaler        &&
           c1->scaler_sub    == c2->scaler_sub    &&
           c1->backends      == c2->backends      &&
           !memcmp(c1->scaler_params, c2->scaler_params, sizeof(c1->scaler_params));

}

int ff_sws_graph_reinit(SwsGraph *graph, SwsContext *ctx, const SwsFormat *dst,
                        const SwsFormat *src)
{
    if (ff_fmt_equal(&graph->src, src) && ff_fmt_equal(&graph->dst, dst) &&
        opts_equal(ctx, &graph->opts_copy))
    {
        ff_sws_graph_update_metadata(graph, &src->color);
        return 0;
    }

    graph_uninit(graph);
    return ff_sws_graph_init(graph, ctx, dst, src);
}

void ff_sws_graph_update_metadata(SwsGraph *graph, const SwsColor *color)
{
    if (!color)
        return;

    ff_color_update_dynamic(&graph->src.color, color);

    if (graph->lut3d)
        ff_sws_lut3d_update(graph->lut3d, &graph->src.color);
}

static void get_field(SwsGraph *graph, const SwsFormat *fmt,
                      const AVFrame *avframe, SwsFrame *frame)
{
    ff_sws_frame_from_avframe(frame, avframe);

    if (!(avframe->flags & AV_FRAME_FLAG_INTERLACED)) {
        av_assert1(!fmt->field);
        return;
    }

    if (fmt->field == FIELD_BOTTOM) {
        /* Odd rows, offset by one line */
        const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(frame->format);
        for (int i = 0; i < 4; i++) {
            if (frame->data[i])
                frame->data[i] += frame->linesize[i];
            if (desc->flags & AV_PIX_FMT_FLAG_PAL)
                break;
        }
    }

    /* Take only every second line */
    for (int i = 0; i < 4; i++)
        frame->linesize[i] <<= 1;

    frame->height = (frame->height + (fmt->field == FIELD_TOP)) >> 1;
}

int ff_sws_graph_run(SwsGraph *graph, const AVFrame *dst, const AVFrame *src)
{
    av_assert0(dst->format == graph->dst.hw_format || dst->format == graph->dst.format);
    av_assert0(src->format == graph->src.hw_format || src->format == graph->src.format);

    SwsFrame src_field, dst_field;
    get_field(graph, &graph->dst, dst, &dst_field);
    get_field(graph, &graph->src, src, &src_field);

    for (int i = 0; i < graph->num_passes; i++) {
        const SwsPass *pass = graph->passes[i];
        graph->exec.pass   = pass;
        graph->exec.input  = pass->input ? &pass->input->output->frame : &src_field;
        graph->exec.output = pass->output->avframe ? &pass->output->frame : &dst_field;
        if (pass->setup) {
            int ret = pass->setup(graph->exec.output, graph->exec.input, pass);
            if (ret < 0)
                return ret;
        }

        if (pass->num_slices == 1) {
            pass->run(graph->exec.output, graph->exec.input, 0, pass->lines, pass);
        } else {
            avpriv_slicethread_execute(graph->slicethread, pass->num_slices, 0);
        }
    }

    return 0;
}