falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
74503 Commits
32 Branches
944MB
Tree: ed80dec621
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'ed80dec621'
${ noResults }
FFmpeg/libswscale/swscale.c

1257 lines
48KB
Raw Blame History 

  1. /*

  2.  * Copyright (C) 2001-2011 Michael Niedermayer <michaelni@gmx.at>

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

  6.  * FFmpeg is free software; you can redistribute it and/or

  7.  * modify it under the terms of the GNU Lesser General Public

  8.  * License as published by the Free Software Foundation; either

  9.  * version 2.1 of the License, or (at your option) any later version.

  10.  *

  11.  * FFmpeg is distributed in the hope that it will be useful,

  12.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  13.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  14.  * Lesser General Public License for more details.

  15.  *

  16.  * You should have received a copy of the GNU Lesser General Public

  17.  * License along with FFmpeg; if not, write to the Free Software

  18.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  19.  */

  20. 

  21. #include <inttypes.h>

  22. #include <math.h>

  23. #include <stdio.h>

  24. #include <string.h>

  25. 

  26. #include "libavutil/avassert.h"

  27. #include "libavutil/avutil.h"

  28. #include "libavutil/bswap.h"

  29. #include "libavutil/cpu.h"

  30. #include "libavutil/imgutils.h"

  31. #include "libavutil/intreadwrite.h"

  32. #include "libavutil/mathematics.h"

  33. #include "libavutil/pixdesc.h"

  34. #include "config.h"

  35. #include "rgb2rgb.h"

  36. #include "swscale_internal.h"

  37. #include "swscale.h"

  38. 

  39. DECLARE_ALIGNED(8, const uint8_t, ff_dither_8x8_128)[9][8] = {

  40.     {  36, 68,  60, 92,  34, 66,  58, 90, },

  41.     { 100,  4, 124, 28,  98,  2, 122, 26, },

  42.     {  52, 84,  44, 76,  50, 82,  42, 74, },

  43.     { 116, 20, 108, 12, 114, 18, 106, 10, },

  44.     {  32, 64,  56, 88,  38, 70,  62, 94, },

  45.     {  96,  0, 120, 24, 102,  6, 126, 30, },

  46.     {  48, 80,  40, 72,  54, 86,  46, 78, },

  47.     { 112, 16, 104,  8, 118, 22, 110, 14, },

  48.     {  36, 68,  60, 92,  34, 66,  58, 90, },

  49. };

  50. 

  51. DECLARE_ALIGNED(8, static const uint8_t, sws_pb_64)[8] = {

  52.     64, 64, 64, 64, 64, 64, 64, 64

  53. };

  54. 

  55. static void gamma_convert(uint8_t * src[], int width, uint16_t *gamma)

  56. {

  57.     int i;

  58.     uint16_t *src1 = (uint16_t*)src[0];

  59. 

  60.     for (i = 0; i < width; ++i) {

  61.         uint16_t r = AV_RL16(src1 + i*4 + 0);

  62.         uint16_t g = AV_RL16(src1 + i*4 + 1);

  63.         uint16_t b = AV_RL16(src1 + i*4 + 2);

  64. 

  65.         AV_WL16(src1 + i*4 + 0, gamma[r]);

  66.         AV_WL16(src1 + i*4 + 1, gamma[g]);

  67.         AV_WL16(src1 + i*4 + 2, gamma[b]);

  68.     }

  69. }

  70. 

  71. static av_always_inline void fillPlane(uint8_t *plane, int stride, int width,

  72.                                        int height, int y, uint8_t val)

  73. {

  74.     int i;

  75.     uint8_t *ptr = plane + stride * y;

  76.     for (i = 0; i < height; i++) {

  77.         memset(ptr, val, width);

  78.         ptr += stride;

  79.     }

  80. }

  81. 

  82. static void hScale16To19_c(SwsContext *c, int16_t *_dst, int dstW,

  83.                            const uint8_t *_src, const int16_t *filter,

  84.                            const int32_t *filterPos, int filterSize)

  85. {

  86.     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);

  87.     int i;

  88.     int32_t *dst        = (int32_t *) _dst;

  89.     const uint16_t *src = (const uint16_t *) _src;

  90.     int bits            = desc->comp[0].depth_minus1;

  91.     int sh              = bits - 4;

  92. 

  93.     if((isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8) && desc->comp[0].depth_minus1<15)

  94.         sh= 9;

  95. 

  96.     for (i = 0; i < dstW; i++) {

  97.         int j;

  98.         int srcPos = filterPos[i];

  99.         int val    = 0;

  100. 

  101.         for (j = 0; j < filterSize; j++) {

  102.             val += src[srcPos + j] * filter[filterSize * i + j];

  103.         }

  104.         // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit

  105.         dst[i] = FFMIN(val >> sh, (1 << 19) - 1);

  106.     }

  107. }

  108. 

  109. static void hScale16To15_c(SwsContext *c, int16_t *dst, int dstW,

  110.                            const uint8_t *_src, const int16_t *filter,

  111.                            const int32_t *filterPos, int filterSize)

  112. {

  113.     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);

  114.     int i;

  115.     const uint16_t *src = (const uint16_t *) _src;

  116.     int sh              = desc->comp[0].depth_minus1;

  117. 

  118.     if(sh<15)

  119.         sh= isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8 ? 13 : desc->comp[0].depth_minus1;

  120. 

  121.     for (i = 0; i < dstW; i++) {

  122.         int j;

  123.         int srcPos = filterPos[i];

  124.         int val    = 0;

  125. 

  126.         for (j = 0; j < filterSize; j++) {

  127.             val += src[srcPos + j] * filter[filterSize * i + j];

  128.         }

  129.         // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit

  130.         dst[i] = FFMIN(val >> sh, (1 << 15) - 1);

  131.     }

  132. }

  133. 

  134. // bilinear / bicubic scaling

  135. static void hScale8To15_c(SwsContext *c, int16_t *dst, int dstW,

  136.                           const uint8_t *src, const int16_t *filter,

  137.                           const int32_t *filterPos, int filterSize)

  138. {

  139.     int i;

  140.     for (i = 0; i < dstW; i++) {

  141.         int j;

  142.         int srcPos = filterPos[i];

  143.         int val    = 0;

  144.         for (j = 0; j < filterSize; j++) {

  145.             val += ((int)src[srcPos + j]) * filter[filterSize * i + j];

  146.         }

  147.         dst[i] = FFMIN(val >> 7, (1 << 15) - 1); // the cubic equation does overflow ...

  148.     }

  149. }

  150. 

  151. static void hScale8To19_c(SwsContext *c, int16_t *_dst, int dstW,

  152.                           const uint8_t *src, const int16_t *filter,

  153.                           const int32_t *filterPos, int filterSize)

  154. {

  155.     int i;

  156.     int32_t *dst = (int32_t *) _dst;

  157.     for (i = 0; i < dstW; i++) {

  158.         int j;

  159.         int srcPos = filterPos[i];

  160.         int val    = 0;

  161.         for (j = 0; j < filterSize; j++) {

  162.             val += ((int)src[srcPos + j]) * filter[filterSize * i + j];

  163.         }

  164.         dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ...

  165.     }

  166. }

  167. 

  168. // FIXME all pal and rgb srcFormats could do this conversion as well

  169. // FIXME all scalers more complex than bilinear could do half of this transform

  170. static void chrRangeToJpeg_c(int16_t *dstU, int16_t *dstV, int width)

  171. {

  172.     int i;

  173.     for (i = 0; i < width; i++) {

  174.         dstU[i] = (FFMIN(dstU[i], 30775) * 4663 - 9289992) >> 12; // -264

  175.         dstV[i] = (FFMIN(dstV[i], 30775) * 4663 - 9289992) >> 12; // -264

  176.     }

  177. }

  178. 

  179. static void chrRangeFromJpeg_c(int16_t *dstU, int16_t *dstV, int width)

  180. {

  181.     int i;

  182.     for (i = 0; i < width; i++) {

  183.         dstU[i] = (dstU[i] * 1799 + 4081085) >> 11; // 1469

  184.         dstV[i] = (dstV[i] * 1799 + 4081085) >> 11; // 1469

  185.     }

  186. }

  187. 

  188. static void lumRangeToJpeg_c(int16_t *dst, int width)

  189. {

  190.     int i;

  191.     for (i = 0; i < width; i++)

  192.         dst[i] = (FFMIN(dst[i], 30189) * 19077 - 39057361) >> 14;

  193. }

  194. 

  195. static void lumRangeFromJpeg_c(int16_t *dst, int width)

  196. {

  197.     int i;

  198.     for (i = 0; i < width; i++)

  199.         dst[i] = (dst[i] * 14071 + 33561947) >> 14;

  200. }

  201. 

  202. static void chrRangeToJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)

  203. {

  204.     int i;

  205.     int32_t *dstU = (int32_t *) _dstU;

  206.     int32_t *dstV = (int32_t *) _dstV;

  207.     for (i = 0; i < width; i++) {

  208.         dstU[i] = (FFMIN(dstU[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264

  209.         dstV[i] = (FFMIN(dstV[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264

  210.     }

  211. }

  212. 

  213. static void chrRangeFromJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)

  214. {

  215.     int i;

  216.     int32_t *dstU = (int32_t *) _dstU;

  217.     int32_t *dstV = (int32_t *) _dstV;

  218.     for (i = 0; i < width; i++) {

  219.         dstU[i] = (dstU[i] * 1799 + (4081085 << 4)) >> 11; // 1469

  220.         dstV[i] = (dstV[i] * 1799 + (4081085 << 4)) >> 11; // 1469

  221.     }

  222. }

  223. 

  224. static void lumRangeToJpeg16_c(int16_t *_dst, int width)

  225. {

  226.     int i;

  227.     int32_t *dst = (int32_t *) _dst;

  228.     for (i = 0; i < width; i++) {

  229.         dst[i] = ((int)(FFMIN(dst[i], 30189 << 4) * 4769U - (39057361 << 2))) >> 12;

  230.     }

  231. }

  232. 

  233. static void lumRangeFromJpeg16_c(int16_t *_dst, int width)

  234. {

  235.     int i;

  236.     int32_t *dst = (int32_t *) _dst;

  237.     for (i = 0; i < width; i++)

  238.         dst[i] = (dst[i]*(14071/4) + (33561947<<4)/4)>>12;

  239. }

  240. 

  241. // *** horizontal scale Y line to temp buffer

  242. static av_always_inline void hyscale(SwsContext *c, int16_t *dst, int dstWidth,

  243.                                      const uint8_t *src_in[4],

  244.                                      int srcW, int xInc,

  245.                                      const int16_t *hLumFilter,

  246.                                      const int32_t *hLumFilterPos,

  247.                                      int hLumFilterSize,

  248.                                      uint8_t *formatConvBuffer,

  249.                                      uint32_t *pal, int isAlpha)

  250. {

  251.     void (*toYV12)(uint8_t *, const uint8_t *, const uint8_t *, const uint8_t *, int, uint32_t *) =

  252.         isAlpha ? c->alpToYV12 : c->lumToYV12;

  253.     void (*convertRange)(int16_t *, int) = isAlpha ? NULL : c->lumConvertRange;

  254.     const uint8_t *src = src_in[isAlpha ? 3 : 0];

  255. 

  256.     if (toYV12) {

  257.         toYV12(formatConvBuffer, src, src_in[1], src_in[2], srcW, pal);

  258.         src = formatConvBuffer;

  259.     } else if (c->readLumPlanar && !isAlpha) {

  260.         c->readLumPlanar(formatConvBuffer, src_in, srcW, c->input_rgb2yuv_table);

  261.         src = formatConvBuffer;

  262.     } else if (c->readAlpPlanar && isAlpha) {

  263.         c->readAlpPlanar(formatConvBuffer, src_in, srcW, NULL);

  264.         src = formatConvBuffer;

  265.     }

  266. 

  267.     if (!c->hyscale_fast) {

  268.         c->hyScale(c, dst, dstWidth, src, hLumFilter,

  269.                    hLumFilterPos, hLumFilterSize);

  270.     } else { // fast bilinear upscale / crap downscale

  271.         c->hyscale_fast(c, dst, dstWidth, src, srcW, xInc);

  272.     }

  273. 

  274.     if (convertRange)

  275.         convertRange(dst, dstWidth);

  276. }

  277. 

  278. static av_always_inline void hcscale(SwsContext *c, int16_t *dst1,

  279.                                      int16_t *dst2, int dstWidth,

  280.                                      const uint8_t *src_in[4],

  281.                                      int srcW, int xInc,

  282.                                      const int16_t *hChrFilter,

  283.                                      const int32_t *hChrFilterPos,

  284.                                      int hChrFilterSize,

  285.                                      uint8_t *formatConvBuffer, uint32_t *pal)

  286. {

  287.     const uint8_t *src1 = src_in[1], *src2 = src_in[2];

  288.     if (c->chrToYV12) {

  289.         uint8_t *buf2 = formatConvBuffer +

  290.                         FFALIGN(srcW*2+78, 16);

  291.         c->chrToYV12(formatConvBuffer, buf2, src_in[0], src1, src2, srcW, pal);

  292.         src1= formatConvBuffer;

  293.         src2= buf2;

  294.     } else if (c->readChrPlanar) {

  295.         uint8_t *buf2 = formatConvBuffer +

  296.                         FFALIGN(srcW*2+78, 16);

  297.         c->readChrPlanar(formatConvBuffer, buf2, src_in, srcW, c->input_rgb2yuv_table);

  298.         src1 = formatConvBuffer;

  299.         src2 = buf2;

  300.     }

  301. 

  302.     if (!c->hcscale_fast) {

  303.         c->hcScale(c, dst1, dstWidth, src1, hChrFilter, hChrFilterPos, hChrFilterSize);

  304.         c->hcScale(c, dst2, dstWidth, src2, hChrFilter, hChrFilterPos, hChrFilterSize);

  305.     } else { // fast bilinear upscale / crap downscale

  306.         c->hcscale_fast(c, dst1, dst2, dstWidth, src1, src2, srcW, xInc);

  307.     }

  308. 

  309.     if (c->chrConvertRange)

  310.         c->chrConvertRange(dst1, dst2, dstWidth);

  311. }

  312. 

  313. #define DEBUG_SWSCALE_BUFFERS 0

  314. #define DEBUG_BUFFERS(...)                      \

  315.     if (DEBUG_SWSCALE_BUFFERS)                  \

  316.         av_log(c, AV_LOG_DEBUG, __VA_ARGS__)

  317. 

  318. static int swscale(SwsContext *c, const uint8_t *src[],

  319.                    int srcStride[], int srcSliceY,

  320.                    int srcSliceH, uint8_t *dst[], int dstStride[])

  321. {

  322.     /* load a few things into local vars to make the code more readable?

  323.      * and faster */

  324. #ifndef NEW_FILTER

  325.     const int srcW                   = c->srcW;

  326. #endif

  327.     const int dstW                   = c->dstW;

  328.     const int dstH                   = c->dstH;

  329.     const int chrDstW                = c->chrDstW;

  330. #ifndef NEW_FILTER

  331.     const int chrSrcW                = c->chrSrcW;

  332.     const int lumXInc                = c->lumXInc;

  333.     const int chrXInc                = c->chrXInc;

  334. #endif

  335.     const enum AVPixelFormat dstFormat = c->dstFormat;

  336.     const int flags                  = c->flags;

  337.     int32_t *vLumFilterPos           = c->vLumFilterPos;

  338.     int32_t *vChrFilterPos           = c->vChrFilterPos;

  339. #ifndef NEW_FILTER

  340.     int32_t *hLumFilterPos           = c->hLumFilterPos;

  341.     int32_t *hChrFilterPos           = c->hChrFilterPos;

  342.     int16_t *hLumFilter              = c->hLumFilter;

  343.     int16_t *hChrFilter              = c->hChrFilter;

  344. #endif

  345.     int32_t *lumMmxFilter            = c->lumMmxFilter;

  346.     int32_t *chrMmxFilter            = c->chrMmxFilter;

  347.     const int vLumFilterSize         = c->vLumFilterSize;

  348.     const int vChrFilterSize         = c->vChrFilterSize;

  349. #ifndef NEW_FILTER

  350.     const int hLumFilterSize         = c->hLumFilterSize;

  351.     const int hChrFilterSize         = c->hChrFilterSize;

  352.     int16_t **lumPixBuf              = c->lumPixBuf;

  353.     int16_t **chrUPixBuf             = c->chrUPixBuf;

  354.     int16_t **chrVPixBuf             = c->chrVPixBuf;

  355. #endif

  356.     int16_t **alpPixBuf              = c->alpPixBuf;

  357.     const int vLumBufSize            = c->vLumBufSize;

  358.     const int vChrBufSize            = c->vChrBufSize;

  359. #ifndef NEW_FILTER

  360.     uint8_t *formatConvBuffer        = c->formatConvBuffer;

  361.     uint32_t *pal                    = c->pal_yuv;

  362. #endif

  363.     yuv2planar1_fn yuv2plane1        = c->yuv2plane1;

  364.     yuv2planarX_fn yuv2planeX        = c->yuv2planeX;

  365.     yuv2interleavedX_fn yuv2nv12cX   = c->yuv2nv12cX;

  366.     yuv2packed1_fn yuv2packed1       = c->yuv2packed1;

  367.     yuv2packed2_fn yuv2packed2       = c->yuv2packed2;

  368.     yuv2packedX_fn yuv2packedX       = c->yuv2packedX;

  369.     yuv2anyX_fn yuv2anyX             = c->yuv2anyX;

  370.     const int chrSrcSliceY           =                srcSliceY >> c->chrSrcVSubSample;

  371.     const int chrSrcSliceH           = FF_CEIL_RSHIFT(srcSliceH,   c->chrSrcVSubSample);

  372.     int should_dither                = is9_OR_10BPS(c->srcFormat) ||

  373.                                        is16BPS(c->srcFormat);

  374.     int lastDstY;

  375. 

  376.     /* vars which will change and which we need to store back in the context */

  377.     int dstY         = c->dstY;

  378.     int lumBufIndex  = c->lumBufIndex;

  379.     int chrBufIndex  = c->chrBufIndex;

  380.     int lastInLumBuf = c->lastInLumBuf;

  381.     int lastInChrBuf = c->lastInChrBuf;

  382. //    int perform_gamma = c->is_internal_gamma;

  383. 

  384.     int lumStart = 0;

  385.     int lumEnd = c->descIndex[0];

  386.     int chrStart = lumEnd;

  387.     int chrEnd = c->descIndex[1];

  388.     SwsSlice *src_slice = &c->slice[lumStart];

  389.     SwsSlice *dst_slice = &c->slice[c->numSlice-1];

  390.     SwsFilterDescriptor *desc = c->desc;

  391.     int hasLumHoles = 1;

  392.     int hasChrHoles = 1;

  393. 

  394. #ifndef NEW_FILTER

  395.     if (!usePal(c->srcFormat)) {

  396.         pal = c->input_rgb2yuv_table;

  397.     }

  398. #endif

  399. 

  400.     if (isPacked(c->srcFormat)) {

  401.         src[0] =

  402.         src[1] =

  403.         src[2] =

  404.         src[3] = src[0];

  405.         srcStride[0] =

  406.         srcStride[1] =

  407.         srcStride[2] =

  408.         srcStride[3] = srcStride[0];

  409.     }

  410.     srcStride[1] <<= c->vChrDrop;

  411.     srcStride[2] <<= c->vChrDrop;

  412. 

  413.     DEBUG_BUFFERS("swscale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",

  414.                   src[0], srcStride[0], src[1], srcStride[1],

  415.                   src[2], srcStride[2], src[3], srcStride[3],

  416.                   dst[0], dstStride[0], dst[1], dstStride[1],

  417.                   dst[2], dstStride[2], dst[3], dstStride[3]);

  418.     DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n",

  419.                   srcSliceY, srcSliceH, dstY, dstH);

  420.     DEBUG_BUFFERS("vLumFilterSize: %d vLumBufSize: %d vChrFilterSize: %d vChrBufSize: %d\n",

  421.                   vLumFilterSize, vLumBufSize, vChrFilterSize, vChrBufSize);

  422. 

  423.     if (dstStride[0]&15 || dstStride[1]&15 ||

  424.         dstStride[2]&15 || dstStride[3]&15) {

  425.         static int warnedAlready = 0; // FIXME maybe move this into the context

  426.         if (flags & SWS_PRINT_INFO && !warnedAlready) {

  427.             av_log(c, AV_LOG_WARNING,

  428.                    "Warning: dstStride is not aligned!\n"

  429.                    "         ->cannot do aligned memory accesses anymore\n");

  430.             warnedAlready = 1;

  431.         }

  432.     }

  433. 

  434.     if (   (uintptr_t)dst[0]&15 || (uintptr_t)dst[1]&15 || (uintptr_t)dst[2]&15

  435.         || (uintptr_t)src[0]&15 || (uintptr_t)src[1]&15 || (uintptr_t)src[2]&15

  436.         || dstStride[0]&15 || dstStride[1]&15 || dstStride[2]&15 || dstStride[3]&15

  437.         || srcStride[0]&15 || srcStride[1]&15 || srcStride[2]&15 || srcStride[3]&15

  438.     ) {

  439.         static int warnedAlready=0;

  440.         int cpu_flags = av_get_cpu_flags();

  441.         if (HAVE_MMXEXT && (cpu_flags & AV_CPU_FLAG_SSE2) && !warnedAlready){

  442.             av_log(c, AV_LOG_WARNING, "Warning: data is not aligned! This can lead to a speedloss\n");

  443.             warnedAlready=1;

  444.         }

  445.     }

  446. 

  447.     /* Note the user might start scaling the picture in the middle so this

  448.      * will not get executed. This is not really intended but works

  449.      * currently, so people might do it. */

  450.     if (srcSliceY == 0) {

  451.         lumBufIndex  = -1;

  452.         chrBufIndex  = -1;

  453.         dstY         = 0;

  454.         lastInLumBuf = -1;

  455.         lastInChrBuf = -1;

  456.     }

  457. 

  458.     if (!should_dither) {

  459.         c->chrDither8 = c->lumDither8 = sws_pb_64;

  460.     }

  461.     lastDstY = dstY;

  462. 

  463.     ff_init_slice_from_src(src_slice, (uint8_t**)src, srcStride, c->srcW,

  464.             srcSliceY, srcSliceH,

  465.             chrSrcSliceY, chrSrcSliceH);

  466. 

  467.     dst_slice->plane[0].sliceY = lastInLumBuf + 1;

  468.     dst_slice->plane[1].sliceY = lastInChrBuf + 1;

  469.     dst_slice->plane[2].sliceY = lastInChrBuf + 1;

  470.     dst_slice->plane[3].sliceY = lastInLumBuf + 1;

  471. 

  472.     dst_slice->plane[0].sliceH =

  473.     dst_slice->plane[1].sliceH =

  474.     dst_slice->plane[2].sliceH =

  475.     dst_slice->plane[3].sliceH = 0;

  476.     dst_slice->width = dstW;

  477. 

  478.     for (; dstY < dstH; dstY++) {

  479.         const int chrDstY = dstY >> c->chrDstVSubSample;

  480.         uint8_t *dest[4]  = {

  481.             dst[0] + dstStride[0] * dstY,

  482.             dst[1] + dstStride[1] * chrDstY,

  483.             dst[2] + dstStride[2] * chrDstY,

  484.             (CONFIG_SWSCALE_ALPHA && alpPixBuf) ? dst[3] + dstStride[3] * dstY : NULL,

  485.         };

  486.         int use_mmx_vfilter= c->use_mmx_vfilter;

  487. 

  488.         // First line needed as input

  489.         const int firstLumSrcY  = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]);

  490.         const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), dstH - 1)]);

  491.         // First line needed as input

  492.         const int firstChrSrcY  = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]);

  493. 

  494.         // Last line needed as input

  495.         int lastLumSrcY  = FFMIN(c->srcH,    firstLumSrcY  + vLumFilterSize) - 1;

  496.         int lastLumSrcY2 = FFMIN(c->srcH,    firstLumSrcY2 + vLumFilterSize) - 1;

  497.         int lastChrSrcY  = FFMIN(c->chrSrcH, firstChrSrcY  + vChrFilterSize) - 1;

  498.         int enough_lines;

  499.         int i;

  500.         int posY, cPosY, firstPosY, lastPosY, firstCPosY, lastCPosY;

  501. 

  502.         // handle holes (FAST_BILINEAR & weird filters)

  503.         if (firstLumSrcY > lastInLumBuf) {

  504.             hasLumHoles = lastInLumBuf != firstLumSrcY - 1;

  505.             lastInLumBuf = firstLumSrcY - 1;

  506.             if (hasLumHoles) {

  507.                 dst_slice->plane[0].sliceY = lastInLumBuf + 1;

  508.                 dst_slice->plane[3].sliceY = lastInLumBuf + 1;

  509.                 dst_slice->plane[0].sliceH =

  510.                 dst_slice->plane[3].sliceH = 0;

  511.             }

  512.         }

  513.         if (firstChrSrcY > lastInChrBuf) {

  514.             hasChrHoles = lastInChrBuf != firstChrSrcY - 1;

  515.             lastInChrBuf = firstChrSrcY - 1;

  516.             if (hasChrHoles) {

  517.                 dst_slice->plane[1].sliceY = lastInChrBuf + 1;

  518.                 dst_slice->plane[2].sliceY = lastInChrBuf + 1;

  519.                 dst_slice->plane[1].sliceH =

  520.                 dst_slice->plane[2].sliceH = 0;

  521.             }

  522.         }

  523.         av_assert0(firstLumSrcY >= lastInLumBuf - vLumBufSize + 1);

  524.         av_assert0(firstChrSrcY >= lastInChrBuf - vChrBufSize + 1);

  525. 

  526.         DEBUG_BUFFERS("dstY: %d\n", dstY);

  527.         DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",

  528.                       firstLumSrcY, lastLumSrcY, lastInLumBuf);

  529.         DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",

  530.                       firstChrSrcY, lastChrSrcY, lastInChrBuf);

  531. 

  532.         // Do we have enough lines in this slice to output the dstY line

  533.         enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&

  534.                        lastChrSrcY < FF_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample);

  535. 

  536.         if (!enough_lines) {

  537.             lastLumSrcY = srcSliceY + srcSliceH - 1;

  538.             lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;

  539.             DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",

  540.                           lastLumSrcY, lastChrSrcY);

  541.         }

  542. 

  543. #ifdef NEW_FILTER

  544.         posY = dst_slice->plane[0].sliceY + dst_slice->plane[0].sliceH;

  545.         if (posY <= lastLumSrcY && !hasLumHoles) {

  546.             firstPosY = FFMAX(firstLumSrcY, posY);

  547.             lastPosY = FFMIN(lastLumSrcY + MAX_LINES_AHEAD, srcSliceY + srcSliceH - 1);

  548.         } else {

  549.             firstPosY = lastInLumBuf + 1;

  550.             lastPosY = lastLumSrcY;

  551.         }

  552. 

  553.         cPosY = dst_slice->plane[1].sliceY + dst_slice->plane[1].sliceH;

  554.         if (cPosY <= lastChrSrcY && !hasChrHoles) {

  555.             firstCPosY = FFMAX(firstChrSrcY, cPosY);

  556.             lastCPosY = FFMIN(lastChrSrcY + MAX_LINES_AHEAD, FF_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample) - 1);

  557.         } else {

  558.             firstCPosY = lastInChrBuf + 1;

  559.             lastCPosY = lastChrSrcY;

  560.         }

  561. 

  562.         ff_rotate_slice(dst_slice, lastPosY, lastCPosY);

  563. 

  564.         if (posY < lastLumSrcY + 1) {

  565.             for (i = lumStart; i < lumEnd; ++i)

  566.                 desc[i].process(c, &desc[i], firstPosY, lastPosY - firstPosY + 1);

  567.         }

  568. 

  569.         lumBufIndex += lastLumSrcY - lastInLumBuf;

  570.         lastInLumBuf = lastLumSrcY;

  571. 

  572.         if (cPosY < lastChrSrcY + 1) {

  573.             for (i = chrStart; i < chrEnd; ++i)

  574.                 desc[i].process(c, &desc[i], firstCPosY, lastCPosY - firstCPosY + 1);

  575.         }

  576. 

  577.         chrBufIndex += lastChrSrcY - lastInChrBuf;

  578.         lastInChrBuf = lastChrSrcY;

  579. 

  580. #else

  581.         // Do horizontal scaling

  582.         while (lastInLumBuf < lastLumSrcY) {

  583.             const uint8_t *src1[4] = {

  584.                 src[0] + (lastInLumBuf + 1 - srcSliceY) * srcStride[0],

  585.                 src[1] + (lastInLumBuf + 1 - srcSliceY) * srcStride[1],

  586.                 src[2] + (lastInLumBuf + 1 - srcSliceY) * srcStride[2],

  587.                 src[3] + (lastInLumBuf + 1 - srcSliceY) * srcStride[3],

  588.             };

  589.             lumBufIndex++;

  590.             av_assert0(lumBufIndex < 2 * vLumBufSize);

  591.             av_assert0(lastInLumBuf + 1 - srcSliceY < srcSliceH);

  592.             av_assert0(lastInLumBuf + 1 - srcSliceY >= 0);

  593. 

  594.             //if (perform_gamma)

  595.             //    gamma_convert((uint8_t **)src1, srcW, c->inv_gamma);

  596. 

  597.             hyscale(c, lumPixBuf[lumBufIndex], dstW, src1, srcW, lumXInc,

  598.                     hLumFilter, hLumFilterPos, hLumFilterSize,

  599.                     formatConvBuffer, pal, 0);

  600.             if (CONFIG_SWSCALE_ALPHA && alpPixBuf)

  601.                 hyscale(c, alpPixBuf[lumBufIndex], dstW, src1, srcW,

  602.                         lumXInc, hLumFilter, hLumFilterPos, hLumFilterSize,

  603.                         formatConvBuffer, pal, 1);

  604.             lastInLumBuf++;

  605.             DEBUG_BUFFERS("\t\tlumBufIndex %d: lastInLumBuf: %d\n",

  606.                           lumBufIndex, lastInLumBuf);

  607.         }

  608.         while (lastInChrBuf < lastChrSrcY) {

  609.             const uint8_t *src1[4] = {

  610.                 src[0] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[0],

  611.                 src[1] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[1],

  612.                 src[2] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[2],

  613.                 src[3] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[3],

  614.             };

  615.             chrBufIndex++;

  616.             av_assert0(chrBufIndex < 2 * vChrBufSize);

  617.             av_assert0(lastInChrBuf + 1 - chrSrcSliceY < (chrSrcSliceH));

  618.             av_assert0(lastInChrBuf + 1 - chrSrcSliceY >= 0);

  619.             // FIXME replace parameters through context struct (some at least)

  620. 

  621.             if (c->needs_hcscale)

  622.                 hcscale(c, chrUPixBuf[chrBufIndex], chrVPixBuf[chrBufIndex],

  623.                         chrDstW, src1, chrSrcW, chrXInc,

  624.                         hChrFilter, hChrFilterPos, hChrFilterSize,

  625.                         formatConvBuffer, pal);

  626.             lastInChrBuf++;

  627.             DEBUG_BUFFERS("\t\tchrBufIndex %d: lastInChrBuf: %d\n",

  628.                           chrBufIndex, lastInChrBuf);

  629.         }

  630. #endif

  631.         // wrap buf index around to stay inside the ring buffer

  632.         if (lumBufIndex >= vLumBufSize)

  633.             lumBufIndex -= vLumBufSize;

  634.         if (chrBufIndex >= vChrBufSize)

  635.             chrBufIndex -= vChrBufSize;

  636.         if (!enough_lines)

  637.             break;  // we can't output a dstY line so let's try with the next slice

  638. 

  639. #if HAVE_MMX_INLINE

  640.         ff_updateMMXDitherTables(c, dstY, lumBufIndex, chrBufIndex,

  641.                               lastInLumBuf, lastInChrBuf);

  642. #endif

  643.         if (should_dither) {

  644.             c->chrDither8 = ff_dither_8x8_128[chrDstY & 7];

  645.             c->lumDither8 = ff_dither_8x8_128[dstY    & 7];

  646.         }

  647.         if (dstY >= dstH - 2) {

  648.             /* hmm looks like we can't use MMX here without overwriting

  649.              * this array's tail */

  650.             ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX,

  651.                                      &yuv2packed1, &yuv2packed2, &yuv2packedX, &yuv2anyX);

  652.             use_mmx_vfilter= 0;

  653.         }

  654. 

  655.         {

  656. #ifdef NEW_FILTER

  657.             const int16_t **lumSrcPtr  = (const int16_t **)(void*) dst_slice->plane[0].line + firstLumSrcY - dst_slice->plane[0].sliceY;

  658.             const int16_t **chrUSrcPtr = (const int16_t **)(void*) dst_slice->plane[1].line + firstChrSrcY - dst_slice->plane[1].sliceY;

  659.             const int16_t **chrVSrcPtr = (const int16_t **)(void*) dst_slice->plane[2].line + firstChrSrcY - dst_slice->plane[2].sliceY;

  660.             const int16_t **alpSrcPtr  = (CONFIG_SWSCALE_ALPHA && alpPixBuf) ?

  661.                                          (const int16_t **)(void*) dst_slice->plane[3].line + firstLumSrcY - dst_slice->plane[3].sliceY : NULL;

  662. #else

  663.             const int16_t **lumSrcPtr  = (const int16_t **)(void*) lumPixBuf  + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;

  664.             const int16_t **chrUSrcPtr = (const int16_t **)(void*) chrUPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;

  665.             const int16_t **chrVSrcPtr = (const int16_t **)(void*) chrVPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;

  666.             const int16_t **alpSrcPtr  = (CONFIG_SWSCALE_ALPHA && alpPixBuf) ?

  667.                                          (const int16_t **)(void*) alpPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize : NULL;

  668. #endif

  669.             int16_t *vLumFilter = c->vLumFilter;

  670.             int16_t *vChrFilter = c->vChrFilter;

  671. 

  672.             if (isPlanarYUV(dstFormat) ||

  673.                 (isGray(dstFormat) && !isALPHA(dstFormat))) { // YV12 like

  674.                 const int chrSkipMask = (1 << c->chrDstVSubSample) - 1;

  675. 

  676.                 vLumFilter +=    dstY * vLumFilterSize;

  677.                 vChrFilter += chrDstY * vChrFilterSize;

  678. 

  679. //                 av_assert0(use_mmx_vfilter != (

  680. //                                yuv2planeX == yuv2planeX_10BE_c

  681. //                             || yuv2planeX == yuv2planeX_10LE_c

  682. //                             || yuv2planeX == yuv2planeX_9BE_c

  683. //                             || yuv2planeX == yuv2planeX_9LE_c

  684. //                             || yuv2planeX == yuv2planeX_16BE_c

  685. //                             || yuv2planeX == yuv2planeX_16LE_c

  686. //                             || yuv2planeX == yuv2planeX_8_c) || !ARCH_X86);

  687. 

  688.                 if(use_mmx_vfilter){

  689.                     vLumFilter= (int16_t *)c->lumMmxFilter;

  690.                     vChrFilter= (int16_t *)c->chrMmxFilter;

  691.                 }

  692. 

  693.                 if (vLumFilterSize == 1) {

  694.                     yuv2plane1(lumSrcPtr[0], dest[0], dstW, c->lumDither8, 0);

  695.                 } else {

  696.                     yuv2planeX(vLumFilter, vLumFilterSize,

  697.                                lumSrcPtr, dest[0],

  698.                                dstW, c->lumDither8, 0);

  699.                 }

  700. 

  701.                 if (!((dstY & chrSkipMask) || isGray(dstFormat))) {

  702.                     if (yuv2nv12cX) {

  703.                         yuv2nv12cX(c, vChrFilter,

  704.                                    vChrFilterSize, chrUSrcPtr, chrVSrcPtr,

  705.                                    dest[1], chrDstW);

  706.                     } else if (vChrFilterSize == 1) {

  707.                         yuv2plane1(chrUSrcPtr[0], dest[1], chrDstW, c->chrDither8, 0);

  708.                         yuv2plane1(chrVSrcPtr[0], dest[2], chrDstW, c->chrDither8, 3);

  709.                     } else {

  710.                         yuv2planeX(vChrFilter,

  711.                                    vChrFilterSize, chrUSrcPtr, dest[1],

  712.                                    chrDstW, c->chrDither8, 0);

  713.                         yuv2planeX(vChrFilter,

  714.                                    vChrFilterSize, chrVSrcPtr, dest[2],

  715.                                    chrDstW, c->chrDither8, use_mmx_vfilter ? (c->uv_offx2 >> 1) : 3);

  716.                     }

  717.                 }

  718. 

  719.                 if (CONFIG_SWSCALE_ALPHA && alpPixBuf) {

  720.                     if(use_mmx_vfilter){

  721.                         vLumFilter= (int16_t *)c->alpMmxFilter;

  722.                     }

  723.                     if (vLumFilterSize == 1) {

  724.                         yuv2plane1(alpSrcPtr[0], dest[3], dstW,

  725.                                    c->lumDither8, 0);

  726.                     } else {

  727.                         yuv2planeX(vLumFilter,

  728.                                    vLumFilterSize, alpSrcPtr, dest[3],

  729.                                    dstW, c->lumDither8, 0);

  730.                     }

  731.                 }

  732.             } else if (yuv2packedX) {

  733. #ifndef NEW_FILTER

  734.                 av_assert1(lumSrcPtr  + vLumFilterSize - 1 < (const int16_t **)lumPixBuf  + vLumBufSize * 2);

  735.                 av_assert1(chrUSrcPtr + vChrFilterSize - 1 < (const int16_t **)chrUPixBuf + vChrBufSize * 2);

  736. #endif

  737.                 if (c->yuv2packed1 && vLumFilterSize == 1 &&

  738.                     vChrFilterSize <= 2) { // unscaled RGB

  739.                     int chrAlpha = vChrFilterSize == 1 ? 0 : vChrFilter[2 * dstY + 1];

  740.                     yuv2packed1(c, *lumSrcPtr, chrUSrcPtr, chrVSrcPtr,

  741.                                 alpPixBuf ? *alpSrcPtr : NULL,

  742.                                 dest[0], dstW, chrAlpha, dstY);

  743.                 } else if (c->yuv2packed2 && vLumFilterSize == 2 &&

  744.                            vChrFilterSize == 2) { // bilinear upscale RGB

  745.                     int lumAlpha = vLumFilter[2 * dstY + 1];

  746.                     int chrAlpha = vChrFilter[2 * dstY + 1];

  747.                     lumMmxFilter[2] =

  748.                     lumMmxFilter[3] = vLumFilter[2 * dstY]    * 0x10001;

  749.                     chrMmxFilter[2] =

  750.                     chrMmxFilter[3] = vChrFilter[2 * chrDstY] * 0x10001;

  751.                     yuv2packed2(c, lumSrcPtr, chrUSrcPtr, chrVSrcPtr,

  752.                                 alpPixBuf ? alpSrcPtr : NULL,

  753.                                 dest[0], dstW, lumAlpha, chrAlpha, dstY);

  754.                 } else { // general RGB

  755.                     yuv2packedX(c, vLumFilter + dstY * vLumFilterSize,

  756.                                 lumSrcPtr, vLumFilterSize,

  757.                                 vChrFilter + dstY * vChrFilterSize,

  758.                                 chrUSrcPtr, chrVSrcPtr, vChrFilterSize,

  759.                                 alpSrcPtr, dest[0], dstW, dstY);

  760.                 }

  761.             } else {

  762.                 av_assert1(!yuv2packed1 && !yuv2packed2);

  763.                 yuv2anyX(c, vLumFilter + dstY * vLumFilterSize,

  764.                          lumSrcPtr, vLumFilterSize,

  765.                          vChrFilter + dstY * vChrFilterSize,

  766.                          chrUSrcPtr, chrVSrcPtr, vChrFilterSize,

  767.                          alpSrcPtr, dest, dstW, dstY);

  768.             }

  769.             //if (perform_gamma)

  770.             //    gamma_convert(dest, dstW, c->gamma);

  771.         }

  772.     }

  773.     if (isPlanar(dstFormat) && isALPHA(dstFormat) && !alpPixBuf) {

  774.         int length = dstW;

  775.         int height = dstY - lastDstY;

  776. 

  777.         if (is16BPS(dstFormat) || isNBPS(dstFormat)) {

  778.             const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);

  779.             fillPlane16(dst[3], dstStride[3], length, height, lastDstY,

  780.                     1, desc->comp[3].depth_minus1,

  781.                     isBE(dstFormat));

  782.         } else

  783.             fillPlane(dst[3], dstStride[3], length, height, lastDstY, 255);

  784.     }

  785. 

  786. #if HAVE_MMXEXT_INLINE

  787.     if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT)

  788.         __asm__ volatile ("sfence" ::: "memory");

  789. #endif

  790.     emms_c();

  791. 

  792.     /* store changed local vars back in the context */

  793.     c->dstY         = dstY;

  794.     c->lumBufIndex  = lumBufIndex;

  795.     c->chrBufIndex  = chrBufIndex;

  796.     c->lastInLumBuf = lastInLumBuf;

  797.     c->lastInChrBuf = lastInChrBuf;

  798. 

  799.     return dstY - lastDstY;

  800. }

  801. 

  802. av_cold void ff_sws_init_range_convert(SwsContext *c)

  803. {

  804.     c->lumConvertRange = NULL;

  805.     c->chrConvertRange = NULL;

  806.     if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {

  807.         if (c->dstBpc <= 14) {

  808.             if (c->srcRange) {

  809.                 c->lumConvertRange = lumRangeFromJpeg_c;

  810.                 c->chrConvertRange = chrRangeFromJpeg_c;

  811.             } else {

  812.                 c->lumConvertRange = lumRangeToJpeg_c;

  813.                 c->chrConvertRange = chrRangeToJpeg_c;

  814.             }

  815.         } else {

  816.             if (c->srcRange) {

  817.                 c->lumConvertRange = lumRangeFromJpeg16_c;

  818.                 c->chrConvertRange = chrRangeFromJpeg16_c;

  819.             } else {

  820.                 c->lumConvertRange = lumRangeToJpeg16_c;

  821.                 c->chrConvertRange = chrRangeToJpeg16_c;

  822.             }

  823.         }

  824.     }

  825. }

  826. 

  827. static av_cold void sws_init_swscale(SwsContext *c)

  828. {

  829.     enum AVPixelFormat srcFormat = c->srcFormat;

  830. 

  831.     ff_sws_init_output_funcs(c, &c->yuv2plane1, &c->yuv2planeX,

  832.                              &c->yuv2nv12cX, &c->yuv2packed1,

  833.                              &c->yuv2packed2, &c->yuv2packedX, &c->yuv2anyX);

  834. 

  835.     ff_sws_init_input_funcs(c);

  836. 

  837. 

  838.     if (c->srcBpc == 8) {

  839.         if (c->dstBpc <= 14) {

  840.             c->hyScale = c->hcScale = hScale8To15_c;

  841.             if (c->flags & SWS_FAST_BILINEAR) {

  842.                 c->hyscale_fast = ff_hyscale_fast_c;

  843.                 c->hcscale_fast = ff_hcscale_fast_c;

  844.             }

  845.         } else {

  846.             c->hyScale = c->hcScale = hScale8To19_c;

  847.         }

  848.     } else {

  849.         c->hyScale = c->hcScale = c->dstBpc > 14 ? hScale16To19_c

  850.                                                  : hScale16To15_c;

  851.     }

  852. 

  853.     ff_sws_init_range_convert(c);

  854. 

  855.     if (!(isGray(srcFormat) || isGray(c->dstFormat) ||

  856.           srcFormat == AV_PIX_FMT_MONOBLACK || srcFormat == AV_PIX_FMT_MONOWHITE))

  857.         c->needs_hcscale = 1;

  858. }

  859. 

  860. SwsFunc ff_getSwsFunc(SwsContext *c)

  861. {

  862.     sws_init_swscale(c);

  863. 

  864.     if (ARCH_PPC)

  865.         ff_sws_init_swscale_ppc(c);

  866.     if (ARCH_X86)

  867.         ff_sws_init_swscale_x86(c);

  868. 

  869.     return swscale;

  870. }

  871. 

  872. static void reset_ptr(const uint8_t *src[], enum AVPixelFormat format)

  873. {

  874.     if (!isALPHA(format))

  875.         src[3] = NULL;

  876.     if (!isPlanar(format)) {

  877.         src[3] = src[2] = NULL;

  878. 

  879.         if (!usePal(format))

  880.             src[1] = NULL;

  881.     }

  882. }

  883. 

  884. static int check_image_pointers(const uint8_t * const data[4], enum AVPixelFormat pix_fmt,

  885.                                 const int linesizes[4])

  886. {

  887.     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt);

  888.     int i;

  889. 

  890.     av_assert2(desc);

  891. 

  892.     for (i = 0; i < 4; i++) {

  893.         int plane = desc->comp[i].plane;

  894.         if (!data[plane] || !linesizes[plane])

  895.             return 0;

  896.     }

  897. 

  898.     return 1;

  899. }

  900. 

  901. static void xyz12Torgb48(struct SwsContext *c, uint16_t *dst,

  902.                          const uint16_t *src, int stride, int h)

  903. {

  904.     int xp,yp;

  905.     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);

  906. 

  907.     for (yp=0; yp<h; yp++) {

  908.         for (xp=0; xp+2<stride; xp+=3) {

  909.             int x, y, z, r, g, b;

  910. 

  911.             if (desc->flags & AV_PIX_FMT_FLAG_BE) {

  912.                 x = AV_RB16(src + xp + 0);

  913.                 y = AV_RB16(src + xp + 1);

  914.                 z = AV_RB16(src + xp + 2);

  915.             } else {

  916.                 x = AV_RL16(src + xp + 0);

  917.                 y = AV_RL16(src + xp + 1);

  918.                 z = AV_RL16(src + xp + 2);

  919.             }

  920. 

  921.             x = c->xyzgamma[x>>4];

  922.             y = c->xyzgamma[y>>4];

  923.             z = c->xyzgamma[z>>4];

  924. 

  925.             // convert from XYZlinear to sRGBlinear

  926.             r = c->xyz2rgb_matrix[0][0] * x +

  927.                 c->xyz2rgb_matrix[0][1] * y +

  928.                 c->xyz2rgb_matrix[0][2] * z >> 12;

  929.             g = c->xyz2rgb_matrix[1][0] * x +

  930.                 c->xyz2rgb_matrix[1][1] * y +

  931.                 c->xyz2rgb_matrix[1][2] * z >> 12;

  932.             b = c->xyz2rgb_matrix[2][0] * x +

  933.                 c->xyz2rgb_matrix[2][1] * y +

  934.                 c->xyz2rgb_matrix[2][2] * z >> 12;

  935. 

  936.             // limit values to 12-bit depth

  937.             r = av_clip_uintp2(r, 12);

  938.             g = av_clip_uintp2(g, 12);

  939.             b = av_clip_uintp2(b, 12);

  940. 

  941.             // convert from sRGBlinear to RGB and scale from 12bit to 16bit

  942.             if (desc->flags & AV_PIX_FMT_FLAG_BE) {

  943.                 AV_WB16(dst + xp + 0, c->rgbgamma[r] << 4);

  944.                 AV_WB16(dst + xp + 1, c->rgbgamma[g] << 4);

  945.                 AV_WB16(dst + xp + 2, c->rgbgamma[b] << 4);

  946.             } else {

  947.                 AV_WL16(dst + xp + 0, c->rgbgamma[r] << 4);

  948.                 AV_WL16(dst + xp + 1, c->rgbgamma[g] << 4);

  949.                 AV_WL16(dst + xp + 2, c->rgbgamma[b] << 4);

  950.             }

  951.         }

  952.         src += stride;

  953.         dst += stride;

  954.     }

  955. }

  956. 

  957. static void rgb48Toxyz12(struct SwsContext *c, uint16_t *dst,

  958.                          const uint16_t *src, int stride, int h)

  959. {

  960.     int xp,yp;

  961.     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->dstFormat);

  962. 

  963.     for (yp=0; yp<h; yp++) {

  964.         for (xp=0; xp+2<stride; xp+=3) {

  965.             int x, y, z, r, g, b;

  966. 

  967.             if (desc->flags & AV_PIX_FMT_FLAG_BE) {

  968.                 r = AV_RB16(src + xp + 0);

  969.                 g = AV_RB16(src + xp + 1);

  970.                 b = AV_RB16(src + xp + 2);

  971.             } else {

  972.                 r = AV_RL16(src + xp + 0);

  973.                 g = AV_RL16(src + xp + 1);

  974.                 b = AV_RL16(src + xp + 2);

  975.             }

  976. 

  977.             r = c->rgbgammainv[r>>4];

  978.             g = c->rgbgammainv[g>>4];

  979.             b = c->rgbgammainv[b>>4];

  980. 

  981.             // convert from sRGBlinear to XYZlinear

  982.             x = c->rgb2xyz_matrix[0][0] * r +

  983.                 c->rgb2xyz_matrix[0][1] * g +

  984.                 c->rgb2xyz_matrix[0][2] * b >> 12;

  985.             y = c->rgb2xyz_matrix[1][0] * r +

  986.                 c->rgb2xyz_matrix[1][1] * g +

  987.                 c->rgb2xyz_matrix[1][2] * b >> 12;

  988.             z = c->rgb2xyz_matrix[2][0] * r +

  989.                 c->rgb2xyz_matrix[2][1] * g +

  990.                 c->rgb2xyz_matrix[2][2] * b >> 12;

  991. 

  992.             // limit values to 12-bit depth

  993.             x = av_clip_uintp2(x, 12);

  994.             y = av_clip_uintp2(y, 12);

  995.             z = av_clip_uintp2(z, 12);

  996. 

  997.             // convert from XYZlinear to X'Y'Z' and scale from 12bit to 16bit

  998.             if (desc->flags & AV_PIX_FMT_FLAG_BE) {

  999.                 AV_WB16(dst + xp + 0, c->xyzgammainv[x] << 4);

  1000.                 AV_WB16(dst + xp + 1, c->xyzgammainv[y] << 4);

  1001.                 AV_WB16(dst + xp + 2, c->xyzgammainv[z] << 4);

  1002.             } else {

  1003.                 AV_WL16(dst + xp + 0, c->xyzgammainv[x] << 4);

  1004.                 AV_WL16(dst + xp + 1, c->xyzgammainv[y] << 4);

  1005.                 AV_WL16(dst + xp + 2, c->xyzgammainv[z] << 4);

  1006.             }

  1007.         }

  1008.         src += stride;

  1009.         dst += stride;

  1010.     }

  1011. }

  1012. 

  1013. /**

  1014.  * swscale wrapper, so we don't need to export the SwsContext.

  1015.  * Assumes planar YUV to be in YUV order instead of YVU.

  1016.  */

  1017. int attribute_align_arg sws_scale(struct SwsContext *c,

  1018.                                   const uint8_t * const srcSlice[],

  1019.                                   const int srcStride[], int srcSliceY,

  1020.                                   int srcSliceH, uint8_t *const dst[],

  1021.                                   const int dstStride[])

  1022. {

  1023.     int i, ret;

  1024.     const uint8_t *src2[4];

  1025.     uint8_t *dst2[4];

  1026.     uint8_t *rgb0_tmp = NULL;

  1027. 

  1028.     if (!srcStride || !dstStride || !dst || !srcSlice) {

  1029.         av_log(c, AV_LOG_ERROR, "One of the input parameters to sws_scale() is NULL, please check the calling code\n");

  1030.         return 0;

  1031.     }

  1032. 

  1033.     if (c->gamma_flag && c->cascaded_context[0]) {

  1034. 

  1035. 

  1036.         ret = sws_scale(c->cascaded_context[0],

  1037.                     srcSlice, srcStride, srcSliceY, srcSliceH,

  1038.                     c->cascaded_tmp, c->cascaded_tmpStride);

  1039. 

  1040.         if (ret < 0)

  1041.             return ret;

  1042. 

  1043.         if (c->cascaded_context[2])

  1044.             ret = sws_scale(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp, c->cascaded_tmpStride, srcSliceY, srcSliceH, c->cascaded1_tmp, c->cascaded1_tmpStride);

  1045.         else

  1046.             ret = sws_scale(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp, c->cascaded_tmpStride, srcSliceY, srcSliceH, dst, dstStride);

  1047. 

  1048.         if (ret < 0)

  1049.             return ret;

  1050. 

  1051.         if (c->cascaded_context[2]) {

  1052.             ret = sws_scale(c->cascaded_context[2],

  1053.                         (const uint8_t * const *)c->cascaded1_tmp, c->cascaded1_tmpStride, c->cascaded_context[1]->dstY - ret, c->cascaded_context[1]->dstY,

  1054.                         dst, dstStride);

  1055.         }

  1056.         return ret;

  1057.     }

  1058. 

  1059.     if (c->cascaded_context[0] && srcSliceY == 0 && srcSliceH == c->cascaded_context[0]->srcH) {

  1060.         ret = sws_scale(c->cascaded_context[0],

  1061.                         srcSlice, srcStride, srcSliceY, srcSliceH,

  1062.                         c->cascaded_tmp, c->cascaded_tmpStride);

  1063.         if (ret < 0)

  1064.             return ret;

  1065.         ret = sws_scale(c->cascaded_context[1],

  1066.                         (const uint8_t * const * )c->cascaded_tmp, c->cascaded_tmpStride, 0, c->cascaded_context[0]->dstH,

  1067.                         dst, dstStride);

  1068.         return ret;

  1069.     }

  1070. 

  1071.     memcpy(src2, srcSlice, sizeof(src2));

  1072.     memcpy(dst2, dst, sizeof(dst2));

  1073. 

  1074.     // do not mess up sliceDir if we have a "trailing" 0-size slice

  1075.     if (srcSliceH == 0)

  1076.         return 0;

  1077. 

  1078.     if (!check_image_pointers(srcSlice, c->srcFormat, srcStride)) {

  1079.         av_log(c, AV_LOG_ERROR, "bad src image pointers\n");

  1080.         return 0;

  1081.     }

  1082.     if (!check_image_pointers((const uint8_t* const*)dst, c->dstFormat, dstStride)) {

  1083.         av_log(c, AV_LOG_ERROR, "bad dst image pointers\n");

  1084.         return 0;

  1085.     }

  1086. 

  1087.     if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {

  1088.         av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");

  1089.         return 0;

  1090.     }

  1091.     if (c->sliceDir == 0) {

  1092.         if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;

  1093.     }

  1094. 

  1095.     if (usePal(c->srcFormat)) {

  1096.         for (i = 0; i < 256; i++) {

  1097.             int r, g, b, y, u, v, a = 0xff;

  1098.             if (c->srcFormat == AV_PIX_FMT_PAL8) {

  1099.                 uint32_t p = ((const uint32_t *)(srcSlice[1]))[i];

  1100.                 a = (p >> 24) & 0xFF;

  1101.                 r = (p >> 16) & 0xFF;

  1102.                 g = (p >>  8) & 0xFF;

  1103.                 b =  p        & 0xFF;

  1104.             } else if (c->srcFormat == AV_PIX_FMT_RGB8) {

  1105.                 r = ( i >> 5     ) * 36;

  1106.                 g = ((i >> 2) & 7) * 36;

  1107.                 b = ( i       & 3) * 85;

  1108.             } else if (c->srcFormat == AV_PIX_FMT_BGR8) {

  1109.                 b = ( i >> 6     ) * 85;

  1110.                 g = ((i >> 3) & 7) * 36;

  1111.                 r = ( i       & 7) * 36;

  1112.             } else if (c->srcFormat == AV_PIX_FMT_RGB4_BYTE) {

  1113.                 r = ( i >> 3     ) * 255;

  1114.                 g = ((i >> 1) & 3) * 85;

  1115.                 b = ( i       & 1) * 255;

  1116.             } else if (c->srcFormat == AV_PIX_FMT_GRAY8 || c->srcFormat == AV_PIX_FMT_GRAY8A) {

  1117.                 r = g = b = i;

  1118.             } else {

  1119.                 av_assert1(c->srcFormat == AV_PIX_FMT_BGR4_BYTE);

  1120.                 b = ( i >> 3     ) * 255;

  1121.                 g = ((i >> 1) & 3) * 85;

  1122.                 r = ( i       & 1) * 255;

  1123.             }

  1124. #define RGB2YUV_SHIFT 15

  1125. #define BY ( (int) (0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  1126. #define BV (-(int) (0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  1127. #define BU ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  1128. #define GY ( (int) (0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  1129. #define GV (-(int) (0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  1130. #define GU (-(int) (0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  1131. #define RY ( (int) (0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  1132. #define RV ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  1133. #define RU (-(int) (0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  1134. 

  1135.             y = av_clip_uint8((RY * r + GY * g + BY * b + ( 33 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);

  1136.             u = av_clip_uint8((RU * r + GU * g + BU * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);

  1137.             v = av_clip_uint8((RV * r + GV * g + BV * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);

  1138.             c->pal_yuv[i]= y + (u<<8) + (v<<16) + ((unsigned)a<<24);

  1139. 

  1140.             switch (c->dstFormat) {

  1141.             case AV_PIX_FMT_BGR32:

  1142. #if !HAVE_BIGENDIAN

  1143.             case AV_PIX_FMT_RGB24:

  1144. #endif

  1145.                 c->pal_rgb[i]=  r + (g<<8) + (b<<16) + ((unsigned)a<<24);

  1146.                 break;

  1147.             case AV_PIX_FMT_BGR32_1:

  1148. #if HAVE_BIGENDIAN

  1149.             case AV_PIX_FMT_BGR24:

  1150. #endif

  1151.                 c->pal_rgb[i]= a + (r<<8) + (g<<16) + ((unsigned)b<<24);

  1152.                 break;

  1153.             case AV_PIX_FMT_RGB32_1:

  1154. #if HAVE_BIGENDIAN

  1155.             case AV_PIX_FMT_RGB24:

  1156. #endif

  1157.                 c->pal_rgb[i]= a + (b<<8) + (g<<16) + ((unsigned)r<<24);

  1158.                 break;

  1159.             case AV_PIX_FMT_RGB32:

  1160. #if !HAVE_BIGENDIAN

  1161.             case AV_PIX_FMT_BGR24:

  1162. #endif

  1163.             default:

  1164.                 c->pal_rgb[i]=  b + (g<<8) + (r<<16) + ((unsigned)a<<24);

  1165.             }

  1166.         }

  1167.     }

  1168. 

  1169.     if (c->src0Alpha && !c->dst0Alpha && isALPHA(c->dstFormat)) {

  1170.         uint8_t *base;

  1171.         int x,y;

  1172.         rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);

  1173.         if (!rgb0_tmp)

  1174.             return AVERROR(ENOMEM);

  1175. 

  1176.         base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;

  1177.         for (y=0; y<srcSliceH; y++){

  1178.             memcpy(base + srcStride[0]*y, src2[0] + srcStride[0]*y, 4*c->srcW);

  1179.             for (x=c->src0Alpha-1; x<4*c->srcW; x+=4) {

  1180.                 base[ srcStride[0]*y + x] = 0xFF;

  1181.             }

  1182.         }

  1183.         src2[0] = base;

  1184.     }

  1185. 

  1186.     if (c->srcXYZ && !(c->dstXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {

  1187.         uint8_t *base;

  1188.         rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);

  1189.         if (!rgb0_tmp)

  1190.             return AVERROR(ENOMEM);

  1191. 

  1192.         base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;

  1193. 

  1194.         xyz12Torgb48(c, (uint16_t*)base, (const uint16_t*)src2[0], srcStride[0]/2, srcSliceH);

  1195.         src2[0] = base;

  1196.     }

  1197. 

  1198.     if (!srcSliceY && (c->flags & SWS_BITEXACT) && c->dither == SWS_DITHER_ED && c->dither_error[0])

  1199.         for (i = 0; i < 4; i++)

  1200.             memset(c->dither_error[i], 0, sizeof(c->dither_error[0][0]) * (c->dstW+2));

  1201. 

  1202. 

  1203.     // copy strides, so they can safely be modified

  1204.     if (c->sliceDir == 1) {

  1205.         // slices go from top to bottom

  1206.         int srcStride2[4] = { srcStride[0], srcStride[1], srcStride[2],

  1207.                               srcStride[3] };

  1208.         int dstStride2[4] = { dstStride[0], dstStride[1], dstStride[2],

  1209.                               dstStride[3] };

  1210. 

  1211.         reset_ptr(src2, c->srcFormat);

  1212.         reset_ptr((void*)dst2, c->dstFormat);

  1213. 

  1214.         /* reset slice direction at end of frame */

  1215.         if (srcSliceY + srcSliceH == c->srcH)

  1216.             c->sliceDir = 0;

  1217. 

  1218.         ret = c->swscale(c, src2, srcStride2, srcSliceY, srcSliceH, dst2,

  1219.                           dstStride2);

  1220.     } else {

  1221.         // slices go from bottom to top => we flip the image internally

  1222.         int srcStride2[4] = { -srcStride[0], -srcStride[1], -srcStride[2],

  1223.                               -srcStride[3] };

  1224.         int dstStride2[4] = { -dstStride[0], -dstStride[1], -dstStride[2],

  1225.                               -dstStride[3] };

  1226. 

  1227.         src2[0] += (srcSliceH - 1) * srcStride[0];

  1228.         if (!usePal(c->srcFormat))

  1229.             src2[1] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[1];

  1230.         src2[2] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[2];

  1231.         src2[3] += (srcSliceH - 1) * srcStride[3];

  1232.         dst2[0] += ( c->dstH                         - 1) * dstStride[0];

  1233.         dst2[1] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[1];

  1234.         dst2[2] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[2];

  1235.         dst2[3] += ( c->dstH                         - 1) * dstStride[3];

  1236. 

  1237.         reset_ptr(src2, c->srcFormat);

  1238.         reset_ptr((void*)dst2, c->dstFormat);

  1239. 

  1240.         /* reset slice direction at end of frame */

  1241.         if (!srcSliceY)

  1242.             c->sliceDir = 0;

  1243. 

  1244.         ret = c->swscale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH,

  1245.                           srcSliceH, dst2, dstStride2);

  1246.     }

  1247. 

  1248. 

  1249.     if (c->dstXYZ && !(c->srcXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {

  1250.         /* replace on the same data */

  1251.         rgb48Toxyz12(c, (uint16_t*)dst2[0], (const uint16_t*)dst2[0], dstStride[0]/2, ret);

  1252.     }

  1253. 

  1254.     av_free(rgb0_tmp);

  1255.     return ret;

  1256. }