swscale.c

Go to the documentation of this file.

/*
 * Copyright (C) 2001-2011 Michael Niedermayer <michaelni@gmx.at>
 *
 * 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 <stdint.h>
#include <stdio.h>
#include <string.h>
 
#include "libavutil/avassert.h"
#include "libavutil/bswap.h"
#include "libavutil/common.h"
#include "libavutil/cpu.h"
#include "libavutil/emms.h"
#include "libavutil/intreadwrite.h"
#include "libavutil/mem.h"
#include "libavutil/mem_internal.h"
#include "libavutil/pixdesc.h"
#include "libavutil/hwcontext.h"
#include "config.h"
#include "swscale_internal.h"
#include "swscale.h"
#if CONFIG_SWSCALE_VULKAN
#include "vulkan/ops.h"
#endif
 
DECLARE_ALIGNED(8, const uint8_t, ff_dither_8x8_128)[9][8] = {
    {  36, 68,  60, 92,  34, 66,  58, 90, },
    { 100,  4, 124, 28,  98,  2, 122, 26, },
    {  52, 84,  44, 76,  50, 82,  42, 74, },
    { 116, 20, 108, 12, 114, 18, 106, 10, },
    {  32, 64,  56, 88,  38, 70,  62, 94, },
    {  96,  0, 120, 24, 102,  6, 126, 30, },
    {  48, 80,  40, 72,  54, 86,  46, 78, },
    { 112, 16, 104,  8, 118, 22, 110, 14, },
    {  36, 68,  60, 92,  34, 66,  58, 90, },
};
 
DECLARE_ALIGNED(8, static const uint8_t, sws_pb_64)[8] = {
    64, 64, 64, 64, 64, 64, 64, 64
};
 
static av_always_inline void fillPlane(uint8_t *plane, int stride, int width,
                                       int height, int y, uint8_t val)
{
    int i;
    uint8_t *ptr = plane + stride * y;
    for (i = 0; i < height; i++) {
        memset(ptr, val, width);
        ptr += stride;
    }
}
 
static void hScale16To19_c(SwsInternal *c, int16_t *_dst, int dstW,
                           const uint8_t *_src, const int16_t *filter,
                           const int32_t *filterPos, int filterSize)
{
    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->opts.src_format);
    int i;
    int32_t *dst        = (int32_t *) _dst;
    const uint16_t *src = (const uint16_t *) _src;
    int bits            = desc->comp[0].depth - 1;
    int sh              = bits - 4;
 
    if ((isAnyRGB(c->opts.src_format) || c->opts.src_format==AV_PIX_FMT_PAL8) && desc->comp[0].depth<16) {
        sh = 9;
    } else if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { /* float input are process like uint 16bpc */
        sh = 16 - 1 - 4;
    }
 
    for (i = 0; i < dstW; i++) {
        int j;
        int srcPos = filterPos[i];
        int val    = 0;
 
        for (j = 0; j < filterSize; j++) {
            val += src[srcPos + j] * filter[filterSize * i + j];
        }
        // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit
        dst[i] = FFMIN(val >> sh, (1 << 19) - 1);
    }
}
 
static void hScale16To15_c(SwsInternal *c, int16_t *dst, int dstW,
                           const uint8_t *_src, const int16_t *filter,
                           const int32_t *filterPos, int filterSize)
{
    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->opts.src_format);
    int i;
    const uint16_t *src = (const uint16_t *) _src;
    int sh              = desc->comp[0].depth - 1;
 
    if (sh<15) {
        sh = isAnyRGB(c->opts.src_format) || c->opts.src_format==AV_PIX_FMT_PAL8 ? 13 : (desc->comp[0].depth - 1);
    } else if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { /* float input are process like uint 16bpc */
        sh = 16 - 1;
    }
 
    for (i = 0; i < dstW; i++) {
        int j;
        int srcPos = filterPos[i];
        int val    = 0;
 
        for (j = 0; j < filterSize; j++) {
            val += src[srcPos + j] * filter[filterSize * i + j];
        }
        // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit
        dst[i] = FFMIN(val >> sh, (1 << 15) - 1);
    }
}
 
// bilinear / bicubic scaling
static void hScale8To15_c(SwsInternal *c, int16_t *dst, int dstW,
                          const uint8_t *src, const int16_t *filter,
                          const int32_t *filterPos, int filterSize)
{
    int i;
    for (i = 0; i < dstW; i++) {
        int j;
        int srcPos = filterPos[i];
        int val    = 0;
        for (j = 0; j < filterSize; j++) {
            val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
        }
        dst[i] = FFMIN(val >> 7, (1 << 15) - 1); // the cubic equation does overflow ...
    }
}
 
static void hScale8To19_c(SwsInternal *c, int16_t *_dst, int dstW,
                          const uint8_t *src, const int16_t *filter,
                          const int32_t *filterPos, int filterSize)
{
    int i;
    int32_t *dst = (int32_t *) _dst;
    for (i = 0; i < dstW; i++) {
        int j;
        int srcPos = filterPos[i];
        int val    = 0;
        for (j = 0; j < filterSize; j++) {
            val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
        }
        dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ...
    }
}
 
// FIXME all pal and rgb srcFormats could do this conversion as well
// FIXME all scalers more complex than bilinear could do half of this transform
static void chrRangeToJpeg_c(int16_t *dstU, int16_t *dstV, int width,
                             uint32_t _coeff, int64_t _offset)
{
    uint16_t coeff = _coeff;
    int32_t offset = _offset;
    int i;
    for (i = 0; i < width; i++) {
        int U = (dstU[i] * coeff + offset) >> 14;
        int V = (dstV[i] * coeff + offset) >> 14;
        dstU[i] = FFMIN(U, (1 << 15) - 1);
        dstV[i] = FFMIN(V, (1 << 15) - 1);
    }
}
 
static void chrRangeFromJpeg_c(int16_t *dstU, int16_t *dstV, int width,
                               uint32_t _coeff, int64_t _offset)
{
    uint16_t coeff = _coeff;
    int32_t offset = _offset;
    int i;
    for (i = 0; i < width; i++) {
        dstU[i] = (dstU[i] * coeff + offset) >> 14;
        dstV[i] = (dstV[i] * coeff + offset) >> 14;
    }
}
 
static void lumRangeToJpeg_c(int16_t *dst, int width,
                             uint32_t _coeff, int64_t _offset)
{
    uint16_t coeff = _coeff;
    int32_t offset = _offset;
    int i;
    for (i = 0; i < width; i++) {
        int Y = (dst[i] * coeff + offset) >> 14;
        dst[i] = FFMIN(Y, (1 << 15) - 1);
    }
}
 
static void lumRangeFromJpeg_c(int16_t *dst, int width,
                               uint32_t _coeff, int64_t _offset)
{
    uint16_t coeff = _coeff;
    int32_t offset = _offset;
    int i;
    for (i = 0; i < width; i++)
        dst[i] = (dst[i] * coeff + offset) >> 14;
}
 
static void chrRangeToJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width,
                               uint32_t coeff, int64_t offset)
{
    int i;
    int32_t *dstU = (int32_t *) _dstU;
    int32_t *dstV = (int32_t *) _dstV;
    for (i = 0; i < width; i++) {
        int U = ((int64_t) dstU[i] * coeff + offset) >> 18;
        int V = ((int64_t) dstV[i] * coeff + offset) >> 18;
        dstU[i] = FFMIN(U, (1 << 19) - 1);
        dstV[i] = FFMIN(V, (1 << 19) - 1);
    }
}
 
static void chrRangeFromJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width,
                                 uint32_t coeff, int64_t offset)
{
    int i;
    int32_t *dstU = (int32_t *) _dstU;
    int32_t *dstV = (int32_t *) _dstV;
    for (i = 0; i < width; i++) {
        dstU[i] = ((int64_t) dstU[i] * coeff + offset) >> 18;
        dstV[i] = ((int64_t) dstV[i] * coeff + offset) >> 18;
    }
}
 
static void lumRangeToJpeg16_c(int16_t *_dst, int width,
                               uint32_t coeff, int64_t offset)
{
    int i;
    int32_t *dst = (int32_t *) _dst;
    for (i = 0; i < width; i++) {
        int Y = ((int64_t) dst[i] * coeff + offset) >> 18;
        dst[i] = FFMIN(Y, (1 << 19) - 1);
    }
}
 
static void lumRangeFromJpeg16_c(int16_t *_dst, int width,
                                 uint32_t coeff, int64_t offset)
{
    int i;
    int32_t *dst = (int32_t *) _dst;
    for (i = 0; i < width; i++)
        dst[i] = ((int64_t) dst[i] * coeff + offset) >> 18;
}
 
static void round_xyz12(uint8_t *plane, int stride, int width, int height, int y,
                        int big_endian)
{
    uint8_t *ptr = plane + stride * y;
    for (int i = 0; i < height; i++) {
        for (int j = 0; j < 3 * width; j++) {
            unsigned v = big_endian ? AV_RB16(ptr + 2 * j) : AV_RL16(ptr + 2 * j);
            v = FFMIN(v + 8, UINT16_MAX) & 0xFFF0;
            if (big_endian)
                AV_WB16(ptr + 2 * j, v);
            else
                AV_WL16(ptr + 2 * j, v);
        }
        ptr += stride;
    }
}
 
 
#define DEBUG_SWSCALE_BUFFERS 0
#define DEBUG_BUFFERS(...)                      \
    if (DEBUG_SWSCALE_BUFFERS)                  \
        av_log(c, AV_LOG_DEBUG, __VA_ARGS__)
 
int ff_swscale(SwsInternal *c, const uint8_t *const src[], const int srcStride[],
               int srcSliceY, int srcSliceH, uint8_t *const dst[],
               const int dstStride[], int dstSliceY, int dstSliceH)
{
    const int scale_dst = dstSliceY > 0 || dstSliceH < c->opts.dst_h;
 
    /* load a few things into local vars to make the code more readable?
     * and faster */
    const int dstW                   = c->opts.dst_w;
    int dstH                         = c->opts.dst_h;
 
    const enum AVPixelFormat dstFormat = c->opts.dst_format;
    const int flags                  = c->opts.flags;
    int32_t *vLumFilterPos           = c->vLumFilterPos;
    int32_t *vChrFilterPos           = c->vChrFilterPos;
 
    const int vLumFilterSize         = c->vLumFilterSize;
    const int vChrFilterSize         = c->vChrFilterSize;
 
    yuv2planar1_fn yuv2plane1        = c->yuv2plane1;
    yuv2planarX_fn yuv2planeX        = c->yuv2planeX;
    yuv2interleavedX_fn yuv2nv12cX   = c->yuv2nv12cX;
    yuv2packed1_fn yuv2packed1       = c->yuv2packed1;
    yuv2packed2_fn yuv2packed2       = c->yuv2packed2;
    yuv2packedX_fn yuv2packedX       = c->yuv2packedX;
    yuv2anyX_fn yuv2anyX             = c->yuv2anyX;
    const int chrSrcSliceY           =                srcSliceY >> c->chrSrcVSubSample;
    const int chrSrcSliceH           = AV_CEIL_RSHIFT(srcSliceH,   c->chrSrcVSubSample);
    int should_dither                = isNBPS(c->opts.src_format) ||
                                       is16BPS(c->opts.src_format);
    int lastDstY;
 
    /* vars which will change and which we need to store back in the context */
    int dstY         = c->dstY;
    int lastInLumBuf = c->lastInLumBuf;
    int lastInChrBuf = c->lastInChrBuf;
 
    int lumStart = 0;
    int lumEnd = c->descIndex[0];
    int chrStart = lumEnd;
    int chrEnd = c->descIndex[1];
    int vStart = chrEnd;
    int vEnd = c->numDesc;
    SwsSlice *src_slice = &c->slice[lumStart];
    SwsSlice *hout_slice = &c->slice[c->numSlice-2];
    SwsSlice *vout_slice = &c->slice[c->numSlice-1];
    SwsFilterDescriptor *desc = c->desc;
 
    int needAlpha = c->needAlpha;
 
    int hasLumHoles = 1;
    int hasChrHoles = 1;
 
    const uint8_t *src2[4];
    int srcStride2[4];
 
    if (isPacked(c->opts.src_format)) {
        src2[0] =
        src2[1] =
        src2[2] =
        src2[3] = src[0];
        srcStride2[0] =
        srcStride2[1] =
        srcStride2[2] =
        srcStride2[3] = srcStride[0];
    } else {
        memcpy(src2, src, sizeof(src2));
        memcpy(srcStride2, srcStride, sizeof(srcStride2));
    }
 
    srcStride2[1] *= 1 << c->vChrDrop;
    srcStride2[2] *= 1 << c->vChrDrop;
 
    DEBUG_BUFFERS("swscale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",
                  src2[0], srcStride2[0], src2[1], srcStride2[1],
                  src2[2], srcStride2[2], src2[3], srcStride2[3],
                  dst[0], dstStride[0], dst[1], dstStride[1],
                  dst[2], dstStride[2], dst[3], dstStride[3]);
    DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n",
                  srcSliceY, srcSliceH, dstY, dstH);
    DEBUG_BUFFERS("vLumFilterSize: %d vChrFilterSize: %d\n",
                  vLumFilterSize, vChrFilterSize);
 
    if (dstStride[0]&15 || dstStride[1]&15 ||
        dstStride[2]&15 || dstStride[3]&15) {
        SwsInternal *const ctx = c->parent ? sws_internal(c->parent) : c;
        if (flags & SWS_PRINT_INFO &&
            !atomic_exchange_explicit(&ctx->stride_unaligned_warned, 1, memory_order_relaxed)) {
            av_log(c, AV_LOG_WARNING,
                   "Warning: dstStride is not aligned!\n"
                   "         ->cannot do aligned memory accesses anymore\n");
        }
    }
 
#if ARCH_X86
    if (   (uintptr_t) dst[0]&15 || (uintptr_t) dst[1]&15 || (uintptr_t) dst[2]&15
        || (uintptr_t)src2[0]&15 || (uintptr_t)src2[1]&15 || (uintptr_t)src2[2]&15
        || srcStride2[0]&15 || srcStride2[1]&15 || srcStride2[2]&15 || srcStride2[3]&15
    ) {
        SwsInternal *const ctx = c->parent ? sws_internal(c->parent) : c;
        int cpu_flags = av_get_cpu_flags();
        if (flags & SWS_PRINT_INFO && HAVE_MMXEXT && (cpu_flags & AV_CPU_FLAG_SSE2) &&
            !atomic_exchange_explicit(&ctx->stride_unaligned_warned,1, memory_order_relaxed)) {
            av_log(c, AV_LOG_WARNING, "Warning: data is not aligned! This can lead to a speed loss\n");
        }
    }
#endif
 
    if (scale_dst) {
        dstY         = dstSliceY;
        dstH         = dstY + dstSliceH;
        lastInLumBuf = -1;
        lastInChrBuf = -1;
    } else if (srcSliceY == 0) {
        /* Note the user might start scaling the picture in the middle so this
         * will not get executed. This is not really intended but works
         * currently, so people might do it. */
        dstY         = 0;
        lastInLumBuf = -1;
        lastInChrBuf = -1;
    }
 
    if (!should_dither) {
        c->chrDither8 = c->lumDither8 = sws_pb_64;
    }
    lastDstY = dstY;
 
    ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
                   yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, c->use_mmx_vfilter);
 
    ff_init_slice_from_src(src_slice, (uint8_t**)src2, srcStride2, c->opts.src_w,
            srcSliceY, srcSliceH, chrSrcSliceY, chrSrcSliceH, 1);
 
    ff_init_slice_from_src(vout_slice, (uint8_t**)dst, dstStride, c->opts.dst_w,
            dstY, dstSliceH, dstY >> c->chrDstVSubSample,
            AV_CEIL_RSHIFT(dstSliceH, c->chrDstVSubSample), scale_dst);
    if (srcSliceY == 0) {
        hout_slice->plane[0].sliceY = lastInLumBuf + 1;
        hout_slice->plane[1].sliceY = lastInChrBuf + 1;
        hout_slice->plane[2].sliceY = lastInChrBuf + 1;
        hout_slice->plane[3].sliceY = lastInLumBuf + 1;
 
        hout_slice->plane[0].sliceH =
        hout_slice->plane[1].sliceH =
        hout_slice->plane[2].sliceH =
        hout_slice->plane[3].sliceH = 0;
        hout_slice->width = dstW;
    }
 
    for (; dstY < dstH; dstY++) {
        const int chrDstY = dstY >> c->chrDstVSubSample;
        int use_mmx_vfilter= c->use_mmx_vfilter;
 
        // First line needed as input
        const int firstLumSrcY  = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]);
        const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), c->opts.dst_h - 1)]);
        // First line needed as input
        const int firstChrSrcY  = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]);
 
        // Last line needed as input
        int lastLumSrcY  = FFMIN(c->opts.src_h,    firstLumSrcY  + vLumFilterSize) - 1;
        int lastLumSrcY2 = FFMIN(c->opts.src_h,    firstLumSrcY2 + vLumFilterSize) - 1;
        int lastChrSrcY  = FFMIN(c->chrSrcH, firstChrSrcY  + vChrFilterSize) - 1;
        int enough_lines;
 
        int i;
        int posY, cPosY, firstPosY, lastPosY, firstCPosY, lastCPosY;
 
        // handle holes (FAST_BILINEAR & weird filters)
        if (firstLumSrcY > lastInLumBuf) {
 
            hasLumHoles = lastInLumBuf != firstLumSrcY - 1;
            if (hasLumHoles) {
                hout_slice->plane[0].sliceY = firstLumSrcY;
                hout_slice->plane[3].sliceY = firstLumSrcY;
                hout_slice->plane[0].sliceH =
                hout_slice->plane[3].sliceH = 0;
            }
 
            lastInLumBuf = firstLumSrcY - 1;
        }
        if (firstChrSrcY > lastInChrBuf) {
 
            hasChrHoles = lastInChrBuf != firstChrSrcY - 1;
            if (hasChrHoles) {
                hout_slice->plane[1].sliceY = firstChrSrcY;
                hout_slice->plane[2].sliceY = firstChrSrcY;
                hout_slice->plane[1].sliceH =
                hout_slice->plane[2].sliceH = 0;
            }
 
            lastInChrBuf = firstChrSrcY - 1;
        }
 
        DEBUG_BUFFERS("dstY: %d\n", dstY);
        DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",
                      firstLumSrcY, lastLumSrcY, lastInLumBuf);
        DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",
                      firstChrSrcY, lastChrSrcY, lastInChrBuf);
 
        // Do we have enough lines in this slice to output the dstY line
        enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&
                       lastChrSrcY < AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample);
 
        if (!enough_lines) {
            lastLumSrcY = srcSliceY + srcSliceH - 1;
            lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
            DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",
                          lastLumSrcY, lastChrSrcY);
        }
 
        av_assert0((lastLumSrcY - firstLumSrcY + 1) <= hout_slice->plane[0].available_lines);
        av_assert0((lastChrSrcY - firstChrSrcY + 1) <= hout_slice->plane[1].available_lines);
 
 
        posY = hout_slice->plane[0].sliceY + hout_slice->plane[0].sliceH;
        if (posY <= lastLumSrcY && !hasLumHoles) {
            firstPosY = FFMAX(firstLumSrcY, posY);
            lastPosY = FFMIN(firstLumSrcY + hout_slice->plane[0].available_lines - 1, srcSliceY + srcSliceH - 1);
        } else {
            firstPosY = posY;
            lastPosY = lastLumSrcY;
        }
 
        cPosY = hout_slice->plane[1].sliceY + hout_slice->plane[1].sliceH;
        if (cPosY <= lastChrSrcY && !hasChrHoles) {
            firstCPosY = FFMAX(firstChrSrcY, cPosY);
            lastCPosY = FFMIN(firstChrSrcY + hout_slice->plane[1].available_lines - 1, AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample) - 1);
        } else {
            firstCPosY = cPosY;
            lastCPosY = lastChrSrcY;
        }
 
        ff_rotate_slice(hout_slice, lastPosY, lastCPosY);
 
        if (posY < lastLumSrcY + 1) {
            for (i = lumStart; i < lumEnd; ++i)
                desc[i].process(c, &desc[i], firstPosY, lastPosY - firstPosY + 1);
        }
 
        lastInLumBuf = lastLumSrcY;
 
        if (cPosY < lastChrSrcY + 1) {
            for (i = chrStart; i < chrEnd; ++i)
                desc[i].process(c, &desc[i], firstCPosY, lastCPosY - firstCPosY + 1);
        }
 
        lastInChrBuf = lastChrSrcY;
 
        if (!enough_lines)
            break;  // we can't output a dstY line so let's try with the next slice
 
#if ARCH_X86 && HAVE_MMX
        ff_updateMMXDitherTables(c, dstY);
        c->dstW_mmx = c->opts.dst_w;
#endif
        if (should_dither) {
            c->chrDither8 = ff_dither_8x8_128[chrDstY & 7];
            c->lumDither8 = ff_dither_8x8_128[dstY    & 7];
        }
        if (dstY >= c->opts.dst_h - 2) {
            /* hmm looks like we can't use MMX here without overwriting
             * this array's tail */
            ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX,
                                     &yuv2packed1, &yuv2packed2, &yuv2packedX, &yuv2anyX);
            use_mmx_vfilter= 0;
            ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX,
                           yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, use_mmx_vfilter);
        }
 
        for (i = vStart; i < vEnd; ++i)
            desc[i].process(c, &desc[i], dstY, 1);
    }
    if (isPlanar(dstFormat) && isALPHA(dstFormat) && !needAlpha) {
        int offset = lastDstY - dstSliceY;
        int length = dstW;
        int height = dstY - lastDstY;
 
        if (is16BPS(dstFormat) || isNBPS(dstFormat)) {
            const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
            fillPlane16(dst[3], dstStride[3], length, height, offset,
                    1, desc->comp[3].depth,
                    isBE(dstFormat));
        } else if (is32BPS(dstFormat)) {
            const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
            fillPlane32(dst[3], dstStride[3], length, height, offset,
                    1, desc->comp[3].depth,
                    isBE(dstFormat), desc->flags & AV_PIX_FMT_FLAG_FLOAT);
        } else
            fillPlane(dst[3], dstStride[3], length, height, offset, 255);
    }
 
    /* XYZ is scaled on its codes like 16-bit RGB */
    if (c->srcXYZ && c->dstXYZ)
        round_xyz12(dst[0], dstStride[0], dstW, dstY - lastDstY, lastDstY - dstSliceY,
                    isBE(dstFormat));
 
#if HAVE_MMXEXT_INLINE
    if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT)
        __asm__ volatile ("sfence" ::: "memory");
#endif
    emms_c();
 
    /* store changed local vars back in the context */
    c->dstY         = dstY;
    c->lastInLumBuf = lastInLumBuf;
    c->lastInChrBuf = lastInChrBuf;
 
    return dstY - lastDstY;
}
 
/*
 * Solve for coeff and offset:
 * dst = ((src << src_shift) * coeff + offset) >> (mult_shift + src_shift)
 *
 * If SwsInternal->dstBpc is > 14, coeff is uint16_t and offset is int32_t,
 * otherwise (SwsInternal->dstBpc is <= 14) coeff is uint32_t and offset is
 * int64_t.
 */
static void solve_range_convert(uint16_t src_min, uint16_t src_max,
                                uint16_t dst_min, uint16_t dst_max,
                                uint16_t src_ref, uint16_t dst_ref,
                                int src_bits, int src_shift, int mult_shift,
                                uint32_t *coeff, int64_t *offset)
{
    uint16_t src_range = src_max - src_min;
    uint16_t dst_range = dst_max - dst_min;
    int total_shift = mult_shift + src_shift;
    *coeff = AV_CEIL_RSHIFT(((uint64_t) dst_range << total_shift) / src_range, src_shift);
    *offset = ((int64_t) dst_ref << total_shift) -
              ((int64_t) src_ref << src_shift) * *coeff +
              (1U << (mult_shift - 1));
}
 
static void init_range_convert_constants(SwsInternal *c)
{
    const int bit_depth = c->dstBpc ? FFMIN(c->dstBpc, 16) : 8;
    const int src_bits = bit_depth <= 14 ? 15 : 19;
    const int src_shift = src_bits - bit_depth;
    const int mult_shift = bit_depth <= 14 ? 14 : 18;
    const uint16_t mpeg_min = 16U << (bit_depth - 8);
    const uint16_t mpeg_max_lum = 235U << (bit_depth - 8);
    const uint16_t mpeg_max_chr = 240U << (bit_depth - 8);
    const uint16_t chr_center = 128U << (bit_depth - 8);
    const uint16_t jpeg_max = (1U << bit_depth) - 1;
    uint16_t src_min, src_max_lum, src_max_chr;
    uint16_t dst_min, dst_max_lum, dst_max_chr;
    if (c->opts.src_range) {
        src_min     = 0;
        src_max_lum = jpeg_max;
        src_max_chr = jpeg_max;
        dst_min     = mpeg_min;
        dst_max_lum = mpeg_max_lum;
        dst_max_chr = mpeg_max_chr;
    } else {
        src_min     = mpeg_min;
        src_max_lum = mpeg_max_lum;
        src_max_chr = mpeg_max_chr;
        dst_min     = 0;
        dst_max_lum = jpeg_max;
        dst_max_chr = jpeg_max;
    }
    solve_range_convert(src_min, src_max_lum, dst_min, dst_max_lum,
                        src_max_lum, dst_max_lum,
                        src_bits, src_shift, mult_shift,
                        &c->lumConvertRange_coeff, &c->lumConvertRange_offset);
    solve_range_convert(src_min, src_max_chr, dst_min, dst_max_chr,
                        chr_center, chr_center,
                        src_bits, src_shift, mult_shift,
                        &c->chrConvertRange_coeff, &c->chrConvertRange_offset);
}
 
av_cold void ff_sws_init_range_convert(SwsInternal *c)
{
    c->lumConvertRange = NULL;
    c->chrConvertRange = NULL;
    if (c->opts.src_range != c->opts.dst_range && !isAnyRGB(c->opts.dst_format)) {
        init_range_convert_constants(c);
        if (c->dstBpc <= 14) {
            if (c->opts.src_range) {
                c->lumConvertRange = lumRangeFromJpeg_c;
                c->chrConvertRange = chrRangeFromJpeg_c;
            } else {
                c->lumConvertRange = lumRangeToJpeg_c;
                c->chrConvertRange = chrRangeToJpeg_c;
            }
        } else {
            if (c->opts.src_range) {
                c->lumConvertRange = lumRangeFromJpeg16_c;
                c->chrConvertRange = chrRangeFromJpeg16_c;
            } else {
                c->lumConvertRange = lumRangeToJpeg16_c;
                c->chrConvertRange = chrRangeToJpeg16_c;
            }
        }
 
#if ARCH_AARCH64
        ff_sws_init_range_convert_aarch64(c);
#elif ARCH_LOONGARCH64
        ff_sws_init_range_convert_loongarch(c);
#elif ARCH_RISCV
        ff_sws_init_range_convert_riscv(c);
#elif ARCH_X86
        ff_sws_init_range_convert_x86(c);
#endif
    }
}
 
static av_cold void sws_init_swscale(SwsInternal *c)
{
    enum AVPixelFormat srcFormat = c->opts.src_format;
 
    ff_sws_init_xyzdsp(c);
 
    ff_sws_init_output_funcs(c, &c->yuv2plane1, &c->yuv2planeX,
                             &c->yuv2nv12cX, &c->yuv2packed1,
                             &c->yuv2packed2, &c->yuv2packedX, &c->yuv2anyX);
 
    ff_sws_init_input_funcs(c, &c->lumToYV12, &c->alpToYV12, &c->chrToYV12,
                            &c->readLumPlanar, &c->readAlpPlanar, &c->readChrPlanar);
 
    if (c->srcBpc == 8) {
        if (c->dstBpc <= 14) {
            c->hyScale = c->hcScale = hScale8To15_c;
            if (c->opts.flags & SWS_FAST_BILINEAR) {
                c->hyscale_fast = ff_hyscale_fast_c;
                c->hcscale_fast = ff_hcscale_fast_c;
            }
        } else {
            c->hyScale = c->hcScale = hScale8To19_c;
        }
    } else {
        c->hyScale = c->hcScale = c->dstBpc > 14 ? hScale16To19_c
                                                 : hScale16To15_c;
    }
 
    ff_sws_init_range_convert(c);
 
    if (!(isGray(srcFormat) || isGray(c->opts.dst_format) ||
          srcFormat == AV_PIX_FMT_MONOBLACK || srcFormat == AV_PIX_FMT_MONOWHITE))
        c->needs_hcscale = 1;
}
 
void ff_sws_init_scale(SwsInternal *c)
{
    sws_init_swscale(c);
 
#if ARCH_PPC
    ff_sws_init_swscale_ppc(c);
#elif ARCH_X86
    ff_sws_init_swscale_x86(c);
#elif ARCH_AARCH64
    ff_sws_init_swscale_aarch64(c);
#elif ARCH_ARM
    ff_sws_init_swscale_arm(c);
#elif ARCH_LOONGARCH64
    ff_sws_init_swscale_loongarch(c);
#elif ARCH_RISCV
    ff_sws_init_swscale_riscv(c);
#endif
}
 
static void reset_ptr(const uint8_t *src[], enum AVPixelFormat format)
{
    if (!isALPHA(format))
        src[3] = NULL;
    if (!isPlanar(format)) {
        src[3] = src[2] = NULL;
 
        if (!usePal(format))
            src[1] = NULL;
    }
}
 
static int check_image_pointers(const uint8_t * const data[4], enum AVPixelFormat pix_fmt,
                                const int linesizes[4], int h)
{
    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt);
    int i;
 
    av_assert2(desc);
 
    for (i = 0; i < 4; i++) {
        int plane = desc->comp[i].plane;
        if (!data[plane] || !linesizes[plane] || FFABS((int64_t)linesizes[plane]) * h > INT_MAX)
            return 0;
    }
 
    return 1;
}
 
static void xyz12Torgb48_c(const SwsInternal *c, uint8_t *dst, int dst_stride,
                           const uint8_t *src, int src_stride, int w, int h)
{
    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->opts.src_format);
 
    for (int yp = 0; yp < h; yp++) {
        const uint16_t *src16 = (const uint16_t *) src;
        uint16_t *dst16 = (uint16_t *) dst;
 
        for (int xp = 0; xp < 3 * w; xp += 3) {
            int x, y, z, r, g, b;
 
            if (desc->flags & AV_PIX_FMT_FLAG_BE) {
                x = AV_RB16(src16 + xp + 0);
                y = AV_RB16(src16 + xp + 1);
                z = AV_RB16(src16 + xp + 2);
            } else {
                x = AV_RL16(src16 + xp + 0);
                y = AV_RL16(src16 + xp + 1);
                z = AV_RL16(src16 + xp + 2);
            }
 
            x = c->xyz2rgb.gamma.in[x >> 4];
            y = c->xyz2rgb.gamma.in[y >> 4];
            z = c->xyz2rgb.gamma.in[z >> 4];
 
            // convert from XYZlinear to sRGBlinear
            r = c->xyz2rgb.mat[0][0] * x +
                c->xyz2rgb.mat[0][1] * y +
                c->xyz2rgb.mat[0][2] * z >> 12;
            g = c->xyz2rgb.mat[1][0] * x +
                c->xyz2rgb.mat[1][1] * y +
                c->xyz2rgb.mat[1][2] * z >> 12;
            b = c->xyz2rgb.mat[2][0] * x +
                c->xyz2rgb.mat[2][1] * y +
                c->xyz2rgb.mat[2][2] * z >> 12;
 
            // limit values to 16-bit depth
            r = av_clip_uint16(r);
            g = av_clip_uint16(g);
            b = av_clip_uint16(b);
 
            // convert from sRGBlinear to RGB and scale from 12bit to 16bit
            if (desc->flags & AV_PIX_FMT_FLAG_BE) {
                AV_WB16(dst16 + xp + 0, c->xyz2rgb.gamma.out[r] << 4);
                AV_WB16(dst16 + xp + 1, c->xyz2rgb.gamma.out[g] << 4);
                AV_WB16(dst16 + xp + 2, c->xyz2rgb.gamma.out[b] << 4);
            } else {
                AV_WL16(dst16 + xp + 0, c->xyz2rgb.gamma.out[r] << 4);
                AV_WL16(dst16 + xp + 1, c->xyz2rgb.gamma.out[g] << 4);
                AV_WL16(dst16 + xp + 2, c->xyz2rgb.gamma.out[b] << 4);
            }
        }
 
        src += src_stride;
        dst += dst_stride;
    }
}
 
static void rgb48Toxyz12_c(const SwsInternal *c, uint8_t *dst, int dst_stride,
                           const uint8_t *src, int src_stride, int w, int h)
{
    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->opts.dst_format);
 
    for (int yp = 0; yp < h; yp++) {
        uint16_t *src16 = (uint16_t *) src;
        uint16_t *dst16 = (uint16_t *) dst;
 
        for (int xp = 0; xp < 3 * w; xp += 3) {
            int x, y, z, r, g, b;
 
            if (desc->flags & AV_PIX_FMT_FLAG_BE) {
                r = AV_RB16(src16 + xp + 0);
                g = AV_RB16(src16 + xp + 1);
                b = AV_RB16(src16 + xp + 2);
            } else {
                r = AV_RL16(src16 + xp + 0);
                g = AV_RL16(src16 + xp + 1);
                b = AV_RL16(src16 + xp + 2);
            }
 
            r = c->rgb2xyz.gamma.in[r >> 4];
            g = c->rgb2xyz.gamma.in[g >> 4];
            b = c->rgb2xyz.gamma.in[b >> 4];
 
            // convert from sRGBlinear to XYZlinear
            x = c->rgb2xyz.mat[0][0] * r +
                c->rgb2xyz.mat[0][1] * g +
                c->rgb2xyz.mat[0][2] * b >> 12;
            y = c->rgb2xyz.mat[1][0] * r +
                c->rgb2xyz.mat[1][1] * g +
                c->rgb2xyz.mat[1][2] * b >> 12;
            z = c->rgb2xyz.mat[2][0] * r +
                c->rgb2xyz.mat[2][1] * g +
                c->rgb2xyz.mat[2][2] * b >> 12;
 
            // limit values to 16-bit depth
            x = av_clip_uint16(x);
            y = av_clip_uint16(y);
            z = av_clip_uint16(z);
 
            // convert from XYZlinear to X'Y'Z' and scale from 12bit to 16bit
            if (desc->flags & AV_PIX_FMT_FLAG_BE) {
                AV_WB16(dst16 + xp + 0, c->rgb2xyz.gamma.out[x] << 4);
                AV_WB16(dst16 + xp + 1, c->rgb2xyz.gamma.out[y] << 4);
                AV_WB16(dst16 + xp + 2, c->rgb2xyz.gamma.out[z] << 4);
            } else {
                AV_WL16(dst16 + xp + 0, c->rgb2xyz.gamma.out[x] << 4);
                AV_WL16(dst16 + xp + 1, c->rgb2xyz.gamma.out[y] << 4);
                AV_WL16(dst16 + xp + 2, c->rgb2xyz.gamma.out[z] << 4);
            }
        }
 
        src += src_stride;
        dst += dst_stride;
    }
}
 
av_cold void ff_sws_init_xyzdsp(SwsInternal *c)
{
    c->xyz12Torgb48 = xyz12Torgb48_c;
    c->rgb48Toxyz12 = rgb48Toxyz12_c;
 
#if ARCH_AARCH64
    ff_sws_init_xyzdsp_aarch64(c);
#endif
}
 
void ff_update_palette(SwsInternal *c, const uint32_t *pal)
{
    uint32_t *rgb2yuv = c->input_rgb2yuv_table;
 
    int32_t ry = rgb2yuv[RY_IDX], gy = rgb2yuv[GY_IDX], by = rgb2yuv[BY_IDX];
    int32_t ru = rgb2yuv[RU_IDX], gu = rgb2yuv[GU_IDX], bu = rgb2yuv[BU_IDX];
    int32_t rv = rgb2yuv[RV_IDX], gv = rgb2yuv[GV_IDX], bv = rgb2yuv[BV_IDX];
 
    for (int i = 0; i < 256; i++) {
        int r, g, b, y, u, v, a = 0xff;
        if (c->opts.src_format == AV_PIX_FMT_PAL8) {
            uint32_t p = pal[i];
            a = (p >> 24) & 0xFF;
            r = (p >> 16) & 0xFF;
            g = (p >>  8) & 0xFF;
            b =  p        & 0xFF;
        } else if (c->opts.src_format == AV_PIX_FMT_RGB8) {
            r = ( i >> 5     ) * 36;
            g = ((i >> 2) & 7) * 36;
            b = ( i       & 3) * 85;
        } else if (c->opts.src_format == AV_PIX_FMT_BGR8) {
            b = ( i >> 6     ) * 85;
            g = ((i >> 3) & 7) * 36;
            r = ( i       & 7) * 36;
        } else if (c->opts.src_format == AV_PIX_FMT_RGB4_BYTE) {
            r = ( i >> 3     ) * 255;
            g = ((i >> 1) & 3) * 85;
            b = ( i       & 1) * 255;
        } else if (c->opts.src_format == AV_PIX_FMT_GRAY8 || c->opts.src_format == AV_PIX_FMT_GRAY8A) {
            r = g = b = i;
        } else {
            av_assert1(c->opts.src_format == AV_PIX_FMT_BGR4_BYTE);
            b = ( i >> 3     ) * 255;
            g = ((i >> 1) & 3) * 85;
            r = ( i       & 1) * 255;
        }
 
        y = av_clip_uint8((ry * r + gy * g + by * b + ( 33 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
        u = av_clip_uint8((ru * r + gu * g + bu * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
        v = av_clip_uint8((rv * r + gv * g + bv * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);
 
        c->pal_yuv[i]= y + (u<<8) + (v<<16) + ((unsigned)a<<24);
 
        switch (c->opts.dst_format) {
        case AV_PIX_FMT_BGR32:
#if !HAVE_BIGENDIAN
        case AV_PIX_FMT_RGB24:
#endif
            c->pal_rgb[i]=  r + (g<<8) + (b<<16) + ((unsigned)a<<24);
            break;
        case AV_PIX_FMT_BGR32_1:
#if HAVE_BIGENDIAN
        case AV_PIX_FMT_BGR24:
#endif
            c->pal_rgb[i]= a + (r<<8) + (g<<16) + ((unsigned)b<<24);
            break;
        case AV_PIX_FMT_RGB32_1:
#if HAVE_BIGENDIAN
        case AV_PIX_FMT_RGB24:
#endif
            c->pal_rgb[i]= a + (b<<8) + (g<<16) + ((unsigned)r<<24);
            break;
        case AV_PIX_FMT_GBRP:
        case AV_PIX_FMT_GBRAP:
#if HAVE_BIGENDIAN
            c->pal_rgb[i]= a + (r<<8) + (b<<16) + ((unsigned)g<<24);
#else
            c->pal_rgb[i]= g + (b<<8) + (r<<16) + ((unsigned)a<<24);
#endif
            break;
        case AV_PIX_FMT_RGB32:
#if !HAVE_BIGENDIAN
        case AV_PIX_FMT_BGR24:
#endif
        default:
            c->pal_rgb[i]=  b + (g<<8) + (r<<16) + ((unsigned)a<<24);
        }
    }
}
 
static int scale_internal(SwsContext *sws,
                          const uint8_t * const srcSlice[], const int srcStride[],
                          int srcSliceY, int srcSliceH,
                          uint8_t *const dstSlice[], const int dstStride[],
                          int dstSliceY, int dstSliceH);
 
static int scale_gamma(SwsInternal *c,
                       const uint8_t * const srcSlice[], const int srcStride[],
                       int srcSliceY, int srcSliceH,
                       uint8_t * const dstSlice[], const int dstStride[],
                       int dstSliceY, int dstSliceH)
{
    int ret = scale_internal(c->cascaded_context[0],
                             srcSlice, srcStride, srcSliceY, srcSliceH,
                             c->cascaded_tmp[0], c->cascaded_tmpStride[0], 0, c->opts.src_h);
 
    if (ret < 0)
        return ret;
 
    if (c->cascaded_context[2])
        ret = scale_internal(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp[0],
                             c->cascaded_tmpStride[0], srcSliceY, srcSliceH,
                             c->cascaded_tmp[1], c->cascaded_tmpStride[1], 0, c->opts.dst_h);
    else
        ret = scale_internal(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp[0],
                             c->cascaded_tmpStride[0], srcSliceY, srcSliceH,
                             dstSlice, dstStride, dstSliceY, dstSliceH);
 
    if (ret < 0)
        return ret;
 
    if (c->cascaded_context[2]) {
        const int dstY1 = sws_internal(c->cascaded_context[1])->dstY;
        ret = scale_internal(c->cascaded_context[2], (const uint8_t * const *)c->cascaded_tmp[1],
                             c->cascaded_tmpStride[1], dstY1 - ret, dstY1,
                             dstSlice, dstStride, dstSliceY, dstSliceH);
    }
    return ret;
}
 
static int scale_cascaded(SwsInternal *c,
                          const uint8_t * const srcSlice[], const int srcStride[],
                          int srcSliceY, int srcSliceH,
                          uint8_t * const dstSlice[], const int dstStride[],
                          int dstSliceY, int dstSliceH)
{
    const int dstH0 = c->cascaded_context[0]->dst_h;
    int ret = scale_internal(c->cascaded_context[0],
                             srcSlice, srcStride, srcSliceY, srcSliceH,
                             c->cascaded_tmp[0], c->cascaded_tmpStride[0],
                             0, dstH0);
    if (ret < 0)
        return ret;
 
    /* The first stage assembles the full intermediate image from the input,
     * one slice at a time (it is itself a regular slice-capable context). The
     * second stage scales that whole intermediate to the output in one step,
     * so it can only run once the entire source has been consumed. The first
     * stage resets its slice direction to 0 at end of frame; until then the
     * intermediate is incomplete and this call produces no output lines. */
    if (sws_internal(c->cascaded_context[0])->sliceDir != 0)
        return 0;
 
    ret = scale_internal(c->cascaded_context[1],
                         (const uint8_t * const * )c->cascaded_tmp[0], c->cascaded_tmpStride[0],
                         0, dstH0, dstSlice, dstStride, dstSliceY, dstSliceH);
    return ret;
}
 
static int scale_internal(SwsContext *sws,
                          const uint8_t * const srcSlice[], const int srcStride[],
                          int srcSliceY, int srcSliceH,
                          uint8_t *const dstSlice[], const int dstStride[],
                          int dstSliceY, int dstSliceH)
{
    SwsInternal *c = sws_internal(sws);
    const int scale_dst = dstSliceY > 0 || dstSliceH < sws->dst_h;
    const int frame_start = scale_dst || !c->sliceDir;
    int i, ret;
    const uint8_t *src2[4];
    uint8_t *dst2[4];
    int macro_height_src = isBayer(sws->src_format) ? 2 : (1 << c->chrSrcVSubSample);
    int macro_height_dst = isBayer(sws->dst_format) ? 2 : (1 << c->chrDstVSubSample);
    // copy strides, so they can safely be modified
    int srcStride2[4];
    int dstStride2[4];
    int srcSliceY_internal = srcSliceY;
 
    if (!srcStride || !dstStride || !dstSlice || !srcSlice) {
        av_log(c, AV_LOG_ERROR, "One of the input parameters to sws_scale() is NULL, please check the calling code\n");
        return AVERROR(EINVAL);
    }
 
    if ((srcSliceY  & (macro_height_src - 1)) ||
        ((srcSliceH & (macro_height_src - 1)) && srcSliceY + srcSliceH != sws->src_h) ||
        srcSliceY + srcSliceH > sws->src_h ||
        srcSliceY < 0 ||
        srcSliceH < 0 ||
        (isBayer(sws->src_format) && srcSliceH <= 1)) {
        av_log(c, AV_LOG_ERROR, "Slice parameters %d, %d are invalid\n", srcSliceY, srcSliceH);
        return AVERROR(EINVAL);
    }
 
    if ((dstSliceY  & (macro_height_dst - 1)) ||
        ((dstSliceH & (macro_height_dst - 1)) && dstSliceY + dstSliceH != sws->dst_h) ||
        dstSliceY + dstSliceH > sws->dst_h) {
        av_log(c, AV_LOG_ERROR, "Slice parameters %d, %d are invalid\n", dstSliceY, dstSliceH);
        return AVERROR(EINVAL);
    }
 
    if (!check_image_pointers(srcSlice, sws->src_format, srcStride, sws->src_h)) {
        av_log(c, AV_LOG_ERROR, "bad src image pointers or strides\n");
        return AVERROR(EINVAL);
    }
    if (!check_image_pointers((const uint8_t* const*)dstSlice, sws->dst_format, dstStride, sws->dst_h)) {
        av_log(c, AV_LOG_ERROR, "bad dst image pointers or strides\n");
        return AVERROR(EINVAL);
    }
 
    // do not mess up sliceDir if we have a "trailing" 0-size slice
    if (srcSliceH == 0)
        return 0;
 
    if (sws->gamma_flag && c->cascaded_context[0])
        return scale_gamma(c, srcSlice, srcStride, srcSliceY, srcSliceH,
                           dstSlice, dstStride, dstSliceY, dstSliceH);
 
    if (c->cascaded_context[0])
        return scale_cascaded(c, srcSlice, srcStride, srcSliceY, srcSliceH,
                              dstSlice, dstStride, dstSliceY, dstSliceH);
 
    if (!srcSliceY && (sws->flags & SWS_BITEXACT) && sws->dither == SWS_DITHER_ED && c->dither_error[0])
        for (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 *)srcSlice[1]);
 
    memcpy(src2,       srcSlice,  sizeof(src2));
    memcpy(dst2,       dstSlice,  sizeof(dst2));
    memcpy(srcStride2, srcStride, sizeof(srcStride2));
    memcpy(dstStride2, dstStride, sizeof(dstStride2));
 
    if (frame_start && !scale_dst) {
        if (srcSliceY != 0 && srcSliceY + srcSliceH != sws->src_h) {
            av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
            return AVERROR(EINVAL);
        }
 
        c->sliceDir = (srcSliceY == 0) ? 1 : -1;
    } else if (scale_dst)
        c->sliceDir = 1;
 
    if (c->src0Alpha && !c->dst0Alpha && isALPHA(sws->dst_format)) {
        uint8_t *base;
        int x,y;
 
        av_fast_malloc(&c->rgb0_scratch, &c->rgb0_scratch_allocated,
                       FFABS((int64_t)srcStride[0]) * srcSliceH + 32);
        if (!c->rgb0_scratch)
            return AVERROR(ENOMEM);
 
        base = srcStride[0] < 0 ? c->rgb0_scratch - srcStride[0] * (srcSliceH-1) :
                                  c->rgb0_scratch;
        for (y=0; y<srcSliceH; y++){
            memcpy(base + srcStride[0]*y, src2[0] + srcStride[0]*y, 4*sws->src_w);
            for (x=c->src0Alpha-1; x<4*sws->src_w; x+=4) {
                base[ srcStride[0]*y + x] = 0xFF;
            }
        }
        src2[0] = base;
    }
 
    if (c->srcXYZ && !c->dstXYZ) {
        uint8_t *base;
 
        av_fast_malloc(&c->xyz_scratch, &c->xyz_scratch_allocated,
                       FFABS((int64_t)srcStride[0]) * srcSliceH + 32);
        if (!c->xyz_scratch)
            return AVERROR(ENOMEM);
 
        base = srcStride[0] < 0 ? c->xyz_scratch - srcStride[0] * (srcSliceH-1) :
                                  c->xyz_scratch;
 
        c->xyz12Torgb48(c, base, srcStride[0], src2[0], srcStride[0], sws->src_w, srcSliceH);
        src2[0] = base;
    }
 
    if (c->sliceDir != 1) {
        // slices go from bottom to top => we flip the image internally
        for (i=0; i<4; i++) {
            srcStride2[i] *= -1;
            dstStride2[i] *= -1;
        }
 
        src2[0] += (srcSliceH - 1) * srcStride[0];
        if (!usePal(sws->src_format))
            src2[1] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[1];
        src2[2] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[2];
        src2[3] += (srcSliceH - 1) * srcStride[3];
        dst2[0] += ( sws->dst_h                         - 1) * dstStride[0];
        dst2[1] += ((sws->dst_h >> c->chrDstVSubSample) - 1) * dstStride[1];
        dst2[2] += ((sws->dst_h >> c->chrDstVSubSample) - 1) * dstStride[2];
        dst2[3] += ( sws->dst_h                         - 1) * dstStride[3];
 
        srcSliceY_internal = sws->src_h-srcSliceY-srcSliceH;
    }
    reset_ptr(src2, sws->src_format);
    reset_ptr((void*)dst2, sws->dst_format);
 
    if (c->convert_unscaled) {
        int offset  = srcSliceY_internal;
        int slice_h = srcSliceH;
 
        // for dst slice scaling, offset the pointers to match the unscaled API
        if (scale_dst) {
            av_assert0(offset == 0);
            for (i = 0; i < 4 && src2[i]; i++) {
                if (!src2[i] || (i > 0 && usePal(sws->src_format)))
                    break;
                src2[i] += (dstSliceY >> ((i == 1 || i == 2) ? c->chrSrcVSubSample : 0)) * srcStride2[i];
            }
 
            for (i = 0; i < 4 && dst2[i]; i++) {
                if (!dst2[i] || (i > 0 && usePal(sws->dst_format)))
                    break;
                dst2[i] -= (dstSliceY >> ((i == 1 || i == 2) ? c->chrDstVSubSample : 0)) * dstStride2[i];
            }
            offset  = dstSliceY;
            slice_h = dstSliceH;
        }
 
        ret = c->convert_unscaled(c, src2, srcStride2, offset, slice_h,
                                  dst2, dstStride2);
        if (scale_dst)
            dst2[0] += dstSliceY * dstStride2[0];
    } else {
        ret = ff_swscale(c, src2, srcStride2, srcSliceY_internal, srcSliceH,
                         dst2, dstStride2, dstSliceY, dstSliceH);
    }
 
    if (c->dstXYZ && !c->srcXYZ) {
        uint8_t *dst;
 
        if (scale_dst) {
            dst = dst2[0];
        } else {
            int dstY = c->dstY ? c->dstY : srcSliceY + srcSliceH;
 
            av_assert0(dstY >= ret);
            av_assert0(ret >= 0);
            av_assert0(sws->dst_h >= dstY);
            dst = dst2[0] + (dstY - ret) * dstStride2[0];
        }
 
        /* replace on the same data */
        c->rgb48Toxyz12(c, dst, dstStride2[0], dst, dstStride2[0], sws->dst_w, ret);
    }
 
    /* reset slice direction at end of frame */
    if ((srcSliceY_internal + srcSliceH == sws->src_h) || scale_dst)
        c->sliceDir = 0;
 
    return ret;
}
 
void sws_frame_end(SwsContext *sws)
{
    SwsInternal *c = sws_internal(sws);
    if (!c->is_legacy_init)
        return;
    av_frame_unref(c->frame_src);
    av_frame_unref(c->frame_dst);
    c->src_ranges.nb_ranges = 0;
}
 
static int ptr_in_buf(const uint8_t *ptr, const AVBufferRef *buf)
{
    uintptr_t ptr_val = (uintptr_t) ptr;
    uintptr_t buf_start = (uintptr_t) buf->data;
    return ptr_val >= buf_start && ptr_val < buf_start + buf->size;
}
 
/* Similar to av_frame_ref() but only references planes in the given map */
static int frame_ref(AVFrame *dst, const AVFrame *src, const int plane_copy[4])
{
    int copied[AV_NUM_DATA_POINTERS] = {0};
    int nb_copied = 0;
 
    for (int i = 0; i < 4; i++) {
        const int idx = plane_copy[i];
        if (idx < 0)
            continue;
        /* Find corresponding source buffer */
        uint8_t *src_data = src->data[idx];
        if (!src_data)
            return AVERROR(EINVAL);
        for (int j = 0; j < FF_ARRAY_ELEMS(src->buf); j++) {
            AVBufferRef *buf = src->buf[j];
            if (!buf)
                break;
            if (!ptr_in_buf(src_data, buf))
                continue;
            if (!copied[j]) {
                AVBufferRef *ref = av_buffer_ref(buf);
                if (!ref)
                    return AVERROR(ENOMEM);
                dst->buf[nb_copied++] = ref;
                copied[j] = 1;
            }
            dst->data[i]     = src_data;
            dst->linesize[i] = src->linesize[idx];
            break;
        }
    }
 
    return 0;
}
 
/* Returns the number of buffers allocated */
static int frame_alloc_buffers(SwsContext *sws, AVFrame *frame)
{
    SwsInternal *c = sws_internal(sws);
    FFFramePool *pool = &c->frame_pool;
    av_assert0(!frame->hw_frames_ctx);
 
    /* Find first free buffer slot */
    int buf_start = 0;
    while (frame->buf[buf_start])
        buf_start++;
    int nb_bufs = 0;
 
    for (int i = 0; i < 4 && pool->pools[i]; i++) {
        if (frame->data[i])
            continue; /* already ref'd by frame_ref */
 
        const int idx = buf_start + nb_bufs++;
        av_assert1(idx < FF_ARRAY_ELEMS(frame->buf));
        frame->buf[idx] = av_buffer_pool_get(pool->pools[i]);
        if (!frame->buf[idx]) {
            av_frame_unref(frame);
            return AVERROR(ENOMEM);
        }
        frame->data[i] = frame->buf[idx]->data;
        frame->linesize[i] = pool->linesize[i];
    }
 
    return nb_bufs;
}
 
int sws_frame_start(SwsContext *sws, AVFrame *dst, const AVFrame *src)
{
    SwsInternal *c = sws_internal(sws);
    int ret, allocated = 0;
    if (!c->is_legacy_init)
        return AVERROR(EINVAL);
 
    ret = av_frame_ref(c->frame_src, src);
    if (ret < 0)
        return ret;
 
    if (!dst->buf[0]) {
        dst->width  = sws->dst_w;
        dst->height = sws->dst_h;
        dst->format = sws->dst_format;
 
        ret = av_frame_get_buffer(dst, 0);
        if (ret < 0)
            return ret;
        allocated = 1;
    }
 
    ret = av_frame_ref(c->frame_dst, dst);
    if (ret < 0) {
        if (allocated)
            av_frame_unref(dst);
 
        return ret;
    }
 
    return 0;
}
 
int sws_send_slice(SwsContext *sws, unsigned int slice_start,
                   unsigned int slice_height)
{
    SwsInternal *c = sws_internal(sws);
    int ret;
    if (!c->is_legacy_init)
        return AVERROR(EINVAL);
 
    ret = ff_range_add(&c->src_ranges, slice_start, slice_height);
    if (ret < 0)
        return ret;
 
    return 0;
}
 
unsigned int sws_receive_slice_alignment(const SwsContext *sws)
{
    SwsInternal *c = sws_internal(sws);
    if (c->slice_ctx)
        return sws_internal(c->slice_ctx[0])->dst_slice_align;
 
    return c->dst_slice_align;
}
 
int sws_receive_slice(SwsContext *sws, unsigned int slice_start,
                      unsigned int slice_height)
{
    SwsInternal *c = sws_internal(sws);
    unsigned int align = sws_receive_slice_alignment(sws);
    uint8_t *dst[4];
    if (!c->is_legacy_init)
        return AVERROR(EINVAL);
 
    /* wait until complete input has been received */
    if (!(c->src_ranges.nb_ranges == 1        &&
          c->src_ranges.ranges[0].start == 0 &&
          c->src_ranges.ranges[0].len == sws->src_h))
        return AVERROR(EAGAIN);
 
    if ((slice_start > 0 || slice_height < sws->dst_h) &&
        (slice_start % align || slice_height % align)) {
        av_log(c, AV_LOG_ERROR,
               "Incorrectly aligned output: %u/%u not multiples of %u\n",
               slice_start, slice_height, align);
        return AVERROR(EINVAL);
    }
 
    if (c->slicethread) {
        int nb_jobs = c->nb_slice_ctx;
        if (c->slice_ctx[0]->dither == SWS_DITHER_ED)
            nb_jobs = 1;
 
        c->dst_slice_start  = slice_start;
        c->dst_slice_height = slice_height;
 
        return avpriv_slicethread_execute2(c->slicethread, nb_jobs, 0);
    }
 
    for (int i = 0; i < FF_ARRAY_ELEMS(dst); i++) {
        ptrdiff_t offset = c->frame_dst->linesize[i] * (ptrdiff_t)(slice_start >> c->chrDstVSubSample);
        dst[i] = FF_PTR_ADD(c->frame_dst->data[i], offset);
    }
 
    return scale_internal(sws, (const uint8_t * const *)c->frame_src->data,
                          c->frame_src->linesize, 0, sws->src_h,
                          dst, c->frame_dst->linesize, slice_start, slice_height);
}
 
static int frame_strides_fit(const AVFrame *frame)
{
    for (int i = 0; i < FF_ARRAY_ELEMS(frame->linesize); i++)
        if (FFABS((int64_t)frame->linesize[i]) * frame->height > INT_MAX)
            return 0;
    return 1;
}
 
int sws_scale_frame(SwsContext *sws, AVFrame *dst, const AVFrame *src)
{
    int ret, allocated = 0;
    SwsInternal *c = sws_internal(sws);
    if (!src || !dst)
        return AVERROR(EINVAL);
 
    if (c->is_legacy_init) {
        /* Context has been initialized with explicit values, fall back to
         * legacy API behavior. */
        ret = sws_frame_start(sws, dst, src);
        if (ret < 0)
            return ret;
 
        ret = sws_send_slice(sws, 0, src->height);
        if (ret >= 0)
            ret = sws_receive_slice(sws, 0, dst->height);
 
        sws_frame_end(sws);
 
        return ret;
    }
 
    ret = sws_frame_setup(sws, dst, src);
    if (ret < 0)
        return ret;
 
    if (!src->data[0])
        return 0;
 
    if (!frame_strides_fit(src) || (dst->data[0] && !frame_strides_fit(dst))) {
        av_log(sws, AV_LOG_ERROR, "Frame planes larger than INT_MAX bytes are not supported\n");
        return AVERROR(EINVAL);
    }
 
    const SwsGraph *top = c->graph[FIELD_TOP];
    const SwsGraph *bot = c->graph[FIELD_BOTTOM];
    if (dst->data[0]) /* user-provided buffers */
        goto process_frame;
 
    /* Sanity */
    memset(dst->buf, 0, sizeof(dst->buf));
    memset(dst->data, 0, sizeof(dst->data));
    memset(dst->linesize, 0, sizeof(dst->linesize));
    dst->extended_data = dst->data;
 
    if (src->buf[0]) {
        /* Determine end-to-end plane copy map */
        int plane_copy[FF_ARRAY_ELEMS(top->plane_copy)];
        memcpy(plane_copy, top->plane_copy, sizeof(plane_copy));
        for (int i = 0; bot && i < FF_ARRAY_ELEMS(plane_copy); i++) {
            if (bot->plane_copy[i] != plane_copy[i])
                plane_copy[i] = -1;
        }
 
        ret = frame_ref(dst, src, plane_copy);
        if (ret < 0)
            return ret;
    }
 
    /* Allocate any missing buffers not yet ref'd */
    ret = frame_alloc_buffers(sws, dst);
    if (ret < 0)
        return ret;
    else if (!ret)
        return 0; /* no buffers allocated, no-op (all ref'd) */
    else
        allocated = 1;
 
process_frame:
    for (int field = 0; field < (bot ? 2 : 1); field++) {
        ret = ff_sws_graph_run(c->graph[field], dst, src);
        if (ret < 0) {
            if (allocated)
                av_frame_unref(dst);
            return ret;
        }
    }
 
    return 0;
}
 
static int validate_params(SwsContext *ctx)
{
#define VALIDATE(field, min, max) \
    if (ctx->field < min || ctx->field > max) { \
        av_log(ctx, AV_LOG_ERROR, "'%s' (%d) out of range [%d, %d]\n", \
               #field, (int) ctx->field, min, max); \
        return AVERROR(EINVAL); \
    }
 
    VALIDATE(threads,       0, SWS_MAX_THREADS);
    VALIDATE(dither,        0, SWS_DITHER_NB - 1)
    VALIDATE(alpha_blend,   0, SWS_ALPHA_BLEND_NB - 1)
    VALIDATE(intent,        0, SWS_INTENT_NB - 1);
    VALIDATE(scaler,        0, SWS_SCALE_NB - 1)
    VALIDATE(scaler_sub,    0, SWS_SCALE_NB - 1)
    return 0;
}
 
static int bottom_field_fits(const AVFrame *frame, const SwsFormat *field)
{
    const int vsub = field->desc->log2_chroma_h;
    return 2 * AV_CEIL_RSHIFT(field->height, vsub) <= AV_CEIL_RSHIFT(frame->height, vsub);
}
 
int sws_frame_setup(SwsContext *ctx, const AVFrame *dst, const AVFrame *src)
{
    SwsInternal *s = sws_internal(ctx);
    const char *err_msg;
    int ret;
 
    if (!src || !dst)
        return AVERROR(EINVAL);
    if ((ret = validate_params(ctx)) < 0)
        return ret;
 
    const AVPixFmtDescriptor *src_desc = av_pix_fmt_desc_get(src->format);
    const AVPixFmtDescriptor *dst_desc = av_pix_fmt_desc_get(dst->format);
    av_assert0(src_desc);
    av_assert0(dst_desc);
 
    const int src_is_hwaccel = !!(src_desc->flags & AV_PIX_FMT_FLAG_HWACCEL);
    const int dst_is_hwaccel = !!(dst_desc->flags & AV_PIX_FMT_FLAG_HWACCEL);
    if (src_is_hwaccel && !src->hw_frames_ctx)
        return AVERROR(EINVAL);
    if (dst_is_hwaccel && !dst->hw_frames_ctx)
        return AVERROR(EINVAL);
 
    /* For now, if a single frame has a context, then both need a context */
    if (src_is_hwaccel ^ dst_is_hwaccel) {
        return AVERROR(ENOTSUP);
    } else if (src_is_hwaccel) {
        /* Both hardware frames must already be allocated */
        if (!src->data[0] || !dst->data[0])
            return AVERROR(EINVAL);
 
        AVHWFramesContext *src_hwfc, *dst_hwfc;
        src_hwfc = (AVHWFramesContext *)src->hw_frames_ctx->data;
        dst_hwfc = (AVHWFramesContext *)dst->hw_frames_ctx->data;
 
        /* Both frames must live on the same device */
        if (src_hwfc->device_ref->data != dst_hwfc->device_ref->data)
            return AVERROR(EINVAL);
 
        /* Only Vulkan devices are supported */
        AVHWDeviceContext *dev_ctx;
        dev_ctx = (AVHWDeviceContext *)src_hwfc->device_ref->data;
        if (dev_ctx->type != AV_HWDEVICE_TYPE_VULKAN)
            return AVERROR(ENOTSUP);
 
#if CONFIG_SWSCALE_VULKAN
        ret = ff_sws_vk_init(ctx, src_hwfc->device_ref);
        if (ret < 0)
            return ret;
#endif
    }
 
    int dst_width = dst->width;
    const SwsBackend backends = ff_sws_enabled_backends(ctx);
    for (int field = 0; field < 2; field++) {
        SwsFormat src_fmt = ff_fmt_from_frame(src, field);
        SwsFormat dst_fmt = ff_fmt_from_frame(dst, field);
        int src_ok, dst_ok;
 
        if ((src->flags ^ dst->flags) & AV_FRAME_FLAG_INTERLACED) {
            err_msg = "Cannot convert interlaced to progressive frames or vice versa.\n";
            ret = AVERROR(EINVAL);
            goto fail;
        }
 
        if (src_fmt.height <= 0 || dst_fmt.height <= 0) {
            err_msg = "Interlaced frames need at least two lines";
            ret = AVERROR(EINVAL);
            goto fail;
        }
 
        if (field == FIELD_BOTTOM &&
            (!bottom_field_fits(src, &src_fmt) || !bottom_field_fits(dst, &dst_fmt))) {
            err_msg = "The bottom field has fewer chroma rows than its height requires";
            ret = AVERROR(EINVAL);
            goto fail;
        }
 
        src_ok = ff_test_fmt(backends, &src_fmt, 0);
        dst_ok = ff_test_fmt(backends, &dst_fmt, 1);
        if ((!src_ok || !dst_ok) && !ff_fmt_equal(&src_fmt, &dst_fmt)) {
            err_msg = src_ok ? "Unsupported output" : "Unsupported input";
            ret = AVERROR(ENOTSUP);
            goto fail;
        }
 
        if (!s->graph[field]) {
            s->graph[field] = ff_sws_graph_alloc();
            if (!s->graph[field]) {
                err_msg = "Failed allocating scaling graph";
                ret = AVERROR(ENOMEM);
                goto fail;
            }
        }
 
        ret = ff_sws_graph_reinit(s->graph[field], ctx, &dst_fmt, &src_fmt);
        if (ret < 0) {
            err_msg = "Failed initializing scaling graph";
            goto fail;
        }
 
        const SwsGraph *graph = s->graph[field];
        if (graph->incomplete && ctx->flags & SWS_STRICT) {
            err_msg = "Incomplete scaling graph";
            ret = AVERROR(EINVAL);
            goto fail;
        }
 
        if (!graph->noop) {
            av_assert0(graph->num_passes);
            const SwsPass *last_pass = graph->passes[graph->num_passes - 1];
            const int aligned_w = ff_sws_pass_aligned_width(last_pass, dst->width);
            dst_width = FFMAX(dst_width, aligned_w);
        }
 
        if (!src_fmt.interlaced) {
            ff_sws_graph_free(&s->graph[FIELD_BOTTOM]);
            break;
        }
 
        continue;
 
    fail:
        av_log(ctx, AV_LOG_ERROR, "%s (%s): fmt:%s csp:%s prim:%s trc:%s ->"
                                          " fmt:%s csp:%s prim:%s trc:%s\n",
               err_msg, av_err2str(ret),
               av_get_pix_fmt_name(src_fmt.format), av_color_space_name(src_fmt.csp),
               av_color_primaries_name(src_fmt.color.prim), av_color_transfer_name(src_fmt.color.trc),
               av_get_pix_fmt_name(dst_fmt.format), av_color_space_name(dst_fmt.csp),
               av_color_primaries_name(dst_fmt.color.prim), av_color_transfer_name(dst_fmt.color.trc));
 
        for (int i = 0; i < FF_ARRAY_ELEMS(s->graph); i++)
            ff_sws_graph_free(&s->graph[i]);
 
        return ret;
    }
 
    if (!dst->hw_frames_ctx) {
        ret = ff_frame_pool_video_reinit(&s->frame_pool, dst_width, dst->height,
                                         dst->format, av_cpu_max_align());
        if (ret < 0)
            return ret;
    }
 
    return 0;
}
 
/**
 * swscale wrapper, so we don't need to export the SwsContext.
 * Assumes planar YUV to be in YUV order instead of YVU.
 */
int attribute_align_arg sws_scale(SwsContext *sws,
                                  const uint8_t * const srcSlice[],
                                  const int srcStride[], int srcSliceY,
                                  int srcSliceH, uint8_t *const dst[],
                                  const int dstStride[])
{
    SwsInternal *c = sws_internal(sws);
    if (!c->is_legacy_init)
        return AVERROR(EINVAL);
 
    if (c->nb_slice_ctx) {
        sws = c->slice_ctx[0];
        c = sws_internal(sws);
    }
 
    return scale_internal(sws, srcSlice, srcStride, srcSliceY, srcSliceH,
                          dst, dstStride, 0, sws->dst_h);
}
 
int ff_sws_slice_worker(void *priv, int jobnr, int threadnr,
                        int nb_jobs, int nb_threads)
{
    SwsInternal *parent = priv;
    SwsContext     *sws = parent->slice_ctx[threadnr];
    SwsInternal      *c = sws_internal(sws);
 
    const int slice_height = FFALIGN(FFMAX((parent->dst_slice_height + nb_jobs - 1) / nb_jobs, 1),
                                     c->dst_slice_align);
    const int slice_start  = jobnr * slice_height;
    const int slice_end    = FFMIN((jobnr + 1) * slice_height, parent->dst_slice_height);
    int err = 0;
 
    if (slice_end > slice_start) {
        uint8_t *dst[4] = { NULL };
 
        for (int i = 0; i < FF_ARRAY_ELEMS(dst) && parent->frame_dst->data[i]; i++) {
            const int vshift = (i == 1 || i == 2) ? c->chrDstVSubSample : 0;
            const ptrdiff_t offset = parent->frame_dst->linesize[i] *
                (ptrdiff_t)((slice_start + parent->dst_slice_start) >> vshift);
 
            dst[i] = parent->frame_dst->data[i] + offset;
        }
 
        err = scale_internal(sws, (const uint8_t * const *)parent->frame_src->data,
                             parent->frame_src->linesize, 0, sws->src_h,
                             dst, parent->frame_dst->linesize,
                             parent->dst_slice_start + slice_start, slice_end - slice_start);
    }
 
    if (err < 0)
        return err;
 
    return 0; /* ff_slicethread_execute() aborts on non-zero */
}