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.
49020 Commits
32 Branches
817MB
Tree: 8045c0167d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '8045c0167d'
${ noResults }
FFmpeg/libswscale/swscale.c

967 lines
38KB
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 <assert.h>

  22. #include <inttypes.h>

  23. #include <math.h>

  24. #include <stdio.h>

  25. #include <string.h>

  26. 

  27. #include "libavutil/avassert.h"

  28. #include "libavutil/avutil.h"

  29. #include "libavutil/bswap.h"

  30. #include "libavutil/cpu.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, dither_8x8_128)[8][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. };

  49. 

  50. DECLARE_ALIGNED(8, const uint8_t, ff_sws_pb_64)[8] = {

  51.     64, 64, 64, 64, 64, 64, 64, 64

  52. };

  53. 

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

  55.                                        int height, int y, uint8_t val)

  56. {

  57.     int i;

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

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

  60.         memset(ptr, val, width);

  61.         ptr += stride;

  62.     }

  63. }

  64. 

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

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

  67.                            const int32_t *filterPos, int filterSize)

  68. {

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

  70.     int i;

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

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

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

  74.     int sh              = bits - 4;

  75. 

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

  77.         sh= 9;

  78. 

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

  80.         int j;

  81.         int srcPos = filterPos[i];

  82.         int val    = 0;

  83. 

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

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

  86.         }

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

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

  89.     }

  90. }

  91. 

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

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

  94.                            const int32_t *filterPos, int filterSize)

  95. {

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

  97.     int i;

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

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

  100. 

  101.     if(sh<15)

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

  103. 

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

  105.         int j;

  106.         int srcPos = filterPos[i];

  107.         int val    = 0;

  108. 

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

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

  111.         }

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

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

  114.     }

  115. }

  116. 

  117. // bilinear / bicubic scaling

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

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

  120.                           const int32_t *filterPos, int filterSize)

  121. {

  122.     int i;

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

  124.         int j;

  125.         int srcPos = filterPos[i];

  126.         int val    = 0;

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

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

  129.         }

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

  131.     }

  132. }

  133. 

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

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

  136.                           const int32_t *filterPos, int filterSize)

  137. {

  138.     int i;

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

  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 >> 3, (1 << 19) - 1); // the cubic equation does overflow ...

  148.     }

  149. }

  150. 

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

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

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

  154. {

  155.     int i;

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

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

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

  159.     }

  160. }

  161. 

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

  163. {

  164.     int i;

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

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

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

  168.     }

  169. }

  170. 

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

  172. {

  173.     int i;

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

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

  176. }

  177. 

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

  179. {

  180.     int i;

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

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

  183. }

  184. 

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

  186. {

  187.     int i;

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

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

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

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

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

  193.     }

  194. }

  195. 

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

  197. {

  198.     int i;

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

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

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

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

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

  204.     }

  205. }

  206. 

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

  208. {

  209.     int i;

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

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

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

  213. }

  214. 

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

  216. {

  217.     int i;

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

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

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

  221. }

  222. 

  223. static void hyscale_fast_c(SwsContext *c, int16_t *dst, int dstWidth,

  224.                            const uint8_t *src, int srcW, int xInc)

  225. {

  226.     int i;

  227.     unsigned int xpos = 0;

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

  229.         register unsigned int xx     = xpos >> 16;

  230.         register unsigned int xalpha = (xpos & 0xFFFF) >> 9;

  231.         dst[i] = (src[xx] << 7) + (src[xx + 1] - src[xx]) * xalpha;

  232.         xpos  += xInc;

  233.     }

  234.     for (i=dstWidth-1; (i*xInc)>>16 >=srcW-1; i--)

  235.         dst[i] = src[srcW-1]*128;

  236. }

  237. 

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

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

  240.                                      const uint8_t *src_in[4],

  241.                                      int srcW, int xInc,

  242.                                      const int16_t *hLumFilter,

  243.                                      const int32_t *hLumFilterPos,

  244.                                      int hLumFilterSize,

  245.                                      uint8_t *formatConvBuffer,

  246.                                      uint32_t *pal, int isAlpha)

  247. {

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

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

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

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

  252. 

  253.     if (toYV12) {

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

  255.         src = formatConvBuffer;

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

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

  258.         src = formatConvBuffer;

  259.     }

  260. 

  261.     if (!c->hyscale_fast) {

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

  263.                    hLumFilterPos, hLumFilterSize);

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

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

  266.     }

  267. 

  268.     if (convertRange)

  269.         convertRange(dst, dstWidth);

  270. }

  271. 

  272. static void hcscale_fast_c(SwsContext *c, int16_t *dst1, int16_t *dst2,

  273.                            int dstWidth, const uint8_t *src1,

  274.                            const uint8_t *src2, int srcW, int xInc)

  275. {

  276.     int i;

  277.     unsigned int xpos = 0;

  278.     for (i = 0; i < dstWidth; i++) {

  279.         register unsigned int xx     = xpos >> 16;

  280.         register unsigned int xalpha = (xpos & 0xFFFF) >> 9;

  281.         dst1[i] = (src1[xx] * (xalpha ^ 127) + src1[xx + 1] * xalpha);

  282.         dst2[i] = (src2[xx] * (xalpha ^ 127) + src2[xx + 1] * xalpha);

  283.         xpos   += xInc;

  284.     }

  285.     for (i=dstWidth-1; (i*xInc)>>16 >=srcW-1; i--) {

  286.         dst1[i] = src1[srcW-1]*128;

  287.         dst2[i] = src2[srcW-1]*128;

  288.     }

  289. }

  290. 

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

  292.                                      int16_t *dst2, int dstWidth,

  293.                                      const uint8_t *src_in[4],

  294.                                      int srcW, int xInc,

  295.                                      const int16_t *hChrFilter,

  296.                                      const int32_t *hChrFilterPos,

  297.                                      int hChrFilterSize,

  298.                                      uint8_t *formatConvBuffer, uint32_t *pal)

  299. {

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

  301.     if (c->chrToYV12) {

  302.         uint8_t *buf2 = formatConvBuffer +

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

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

  305.         src1= formatConvBuffer;

  306.         src2= buf2;

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

  308.         uint8_t *buf2 = formatConvBuffer +

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

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

  311.         src1 = formatConvBuffer;

  312.         src2 = buf2;

  313.     }

  314. 

  315.     if (!c->hcscale_fast) {

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

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

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

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

  320.     }

  321. 

  322.     if (c->chrConvertRange)

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

  324. }

  325. 

  326. #define DEBUG_SWSCALE_BUFFERS 0

  327. #define DEBUG_BUFFERS(...)                      \

  328.     if (DEBUG_SWSCALE_BUFFERS)                  \

  329.         av_log(c, AV_LOG_DEBUG, __VA_ARGS__)

  330. 

  331. static int swScale(SwsContext *c, const uint8_t *src[],

  332.                    int srcStride[], int srcSliceY,

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

  334. {

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

  336.      * and faster */

  337.     const int srcW                   = c->srcW;

  338.     const int dstW                   = c->dstW;

  339.     const int dstH                   = c->dstH;

  340.     const int chrDstW                = c->chrDstW;

  341.     const int chrSrcW                = c->chrSrcW;

  342.     const int lumXInc                = c->lumXInc;

  343.     const int chrXInc                = c->chrXInc;

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

  345.     const int flags                  = c->flags;

  346.     int32_t *vLumFilterPos           = c->vLumFilterPos;

  347.     int32_t *vChrFilterPos           = c->vChrFilterPos;

  348.     int32_t *hLumFilterPos           = c->hLumFilterPos;

  349.     int32_t *hChrFilterPos           = c->hChrFilterPos;

  350.     int16_t *hLumFilter              = c->hLumFilter;

  351.     int16_t *hChrFilter              = c->hChrFilter;

  352.     int32_t *lumMmxFilter            = c->lumMmxFilter;

  353.     int32_t *chrMmxFilter            = c->chrMmxFilter;

  354.     const int vLumFilterSize         = c->vLumFilterSize;

  355.     const int vChrFilterSize         = c->vChrFilterSize;

  356.     const int hLumFilterSize         = c->hLumFilterSize;

  357.     const int hChrFilterSize         = c->hChrFilterSize;

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

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

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

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

  362.     const int vLumBufSize            = c->vLumBufSize;

  363.     const int vChrBufSize            = c->vChrBufSize;

  364.     uint8_t *formatConvBuffer        = c->formatConvBuffer;

  365.     uint32_t *pal                    = c->pal_yuv;

  366.     yuv2planar1_fn yuv2plane1        = c->yuv2plane1;

  367.     yuv2planarX_fn yuv2planeX        = c->yuv2planeX;

  368.     yuv2interleavedX_fn yuv2nv12cX   = c->yuv2nv12cX;

  369.     yuv2packed1_fn yuv2packed1       = c->yuv2packed1;

  370.     yuv2packed2_fn yuv2packed2       = c->yuv2packed2;

  371.     yuv2packedX_fn yuv2packedX       = c->yuv2packedX;

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

  373.     const int chrSrcSliceH           = -((-srcSliceH) >> c->chrSrcVSubSample);

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

  375.                                        is16BPS(c->srcFormat);

  376.     int lastDstY;

  377. 

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

  379.     int dstY         = c->dstY;

  380.     int lumBufIndex  = c->lumBufIndex;

  381.     int chrBufIndex  = c->chrBufIndex;

  382.     int lastInLumBuf = c->lastInLumBuf;

  383.     int lastInChrBuf = c->lastInChrBuf;

  384. 

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

  386.         src[0] =

  387.         src[1] =

  388.         src[2] =

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

  390.         srcStride[0] =

  391.         srcStride[1] =

  392.         srcStride[2] =

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

  394.     }

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

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

  397. 

  398.     DEBUG_BUFFERS("swScale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",

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

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

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

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

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

  404.                   srcSliceY, srcSliceH, dstY, dstH);

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

  406.                   vLumFilterSize, vLumBufSize, vChrFilterSize, vChrBufSize);

  407. 

  408.     if (dstStride[0]%16 !=0 || dstStride[1]%16 !=0 ||

  409.         dstStride[2]%16 !=0 || dstStride[3]%16 != 0) {

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

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

  412.             av_log(c, AV_LOG_WARNING,

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

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

  415.             warnedAlready = 1;

  416.         }

  417.     }

  418. 

  419.     if (   (uintptr_t)dst[0]%16 || (uintptr_t)dst[1]%16 || (uintptr_t)dst[2]%16

  420.         || (uintptr_t)src[0]%16 || (uintptr_t)src[1]%16 || (uintptr_t)src[2]%16

  421.         || dstStride[0]%16 || dstStride[1]%16 || dstStride[2]%16 || dstStride[3]%16

  422.         || srcStride[0]%16 || srcStride[1]%16 || srcStride[2]%16 || srcStride[3]%16

  423.     ) {

  424.         static int warnedAlready=0;

  425.         int cpu_flags = av_get_cpu_flags();

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

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

  428.             warnedAlready=1;

  429.         }

  430.     }

  431. 

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

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

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

  435.     if (srcSliceY == 0) {

  436.         lumBufIndex  = -1;

  437.         chrBufIndex  = -1;

  438.         dstY         = 0;

  439.         lastInLumBuf = -1;

  440.         lastInChrBuf = -1;

  441.     }

  442. 

  443.     if (!should_dither) {

  444.         c->chrDither8 = c->lumDither8 = ff_sws_pb_64;

  445.     }

  446.     lastDstY = dstY;

  447. 

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

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

  450.         uint8_t *dest[4]  = {

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

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

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

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

  455.         };

  456.         int use_mmx_vfilter= c->use_mmx_vfilter;

  457. 

  458.         // First line needed as input

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

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

  461.         // First line needed as input

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

  463. 

  464.         // Last line needed as input

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

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

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

  468.         int enough_lines;

  469. 

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

  471.         if (firstLumSrcY > lastInLumBuf)

  472.             lastInLumBuf = firstLumSrcY - 1;

  473.         if (firstChrSrcY > lastInChrBuf)

  474.             lastInChrBuf = firstChrSrcY - 1;

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

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

  477. 

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

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

  480.                       firstLumSrcY, lastLumSrcY, lastInLumBuf);

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

  482.                       firstChrSrcY, lastChrSrcY, lastInChrBuf);

  483. 

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

  485.         enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&

  486.                        lastChrSrcY < -((-srcSliceY - srcSliceH) >> c->chrSrcVSubSample);

  487. 

  488.         if (!enough_lines) {

  489.             lastLumSrcY = srcSliceY + srcSliceH - 1;

  490.             lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;

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

  492.                           lastLumSrcY, lastChrSrcY);

  493.         }

  494. 

  495.         // Do horizontal scaling

  496.         while (lastInLumBuf < lastLumSrcY) {

  497.             const uint8_t *src1[4] = {

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

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

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

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

  502.             };

  503.             lumBufIndex++;

  504.             av_assert0(lumBufIndex < 2 * vLumBufSize);

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

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

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

  508.                     hLumFilter, hLumFilterPos, hLumFilterSize,

  509.                     formatConvBuffer, pal, 0);

  510.             if (CONFIG_SWSCALE_ALPHA && alpPixBuf)

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

  512.                         lumXInc, hLumFilter, hLumFilterPos, hLumFilterSize,

  513.                         formatConvBuffer, pal, 1);

  514.             lastInLumBuf++;

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

  516.                           lumBufIndex, lastInLumBuf);

  517.         }

  518.         while (lastInChrBuf < lastChrSrcY) {

  519.             const uint8_t *src1[4] = {

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

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

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

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

  524.             };

  525.             chrBufIndex++;

  526.             av_assert0(chrBufIndex < 2 * vChrBufSize);

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

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

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

  530. 

  531.             if (c->needs_hcscale)

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

  533.                         chrDstW, src1, chrSrcW, chrXInc,

  534.                         hChrFilter, hChrFilterPos, hChrFilterSize,

  535.                         formatConvBuffer, pal);

  536.             lastInChrBuf++;

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

  538.                           chrBufIndex, lastInChrBuf);

  539.         }

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

  541.         if (lumBufIndex >= vLumBufSize)

  542.             lumBufIndex -= vLumBufSize;

  543.         if (chrBufIndex >= vChrBufSize)

  544.             chrBufIndex -= vChrBufSize;

  545.         if (!enough_lines)

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

  547. 

  548. #if HAVE_MMX_INLINE

  549.         updateMMXDitherTables(c, dstY, lumBufIndex, chrBufIndex,

  550.                               lastInLumBuf, lastInChrBuf);

  551. #endif

  552.         if (should_dither) {

  553.             c->chrDither8 = dither_8x8_128[chrDstY & 7];

  554.             c->lumDither8 = dither_8x8_128[dstY    & 7];

  555.         }

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

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

  558.              * this array's tail */

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

  560.                                      &yuv2packed1, &yuv2packed2, &yuv2packedX);

  561.             use_mmx_vfilter= 0;

  562.         }

  563. 

  564.         {

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

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

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

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

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

  570.             int16_t *vLumFilter = c->vLumFilter;

  571.             int16_t *vChrFilter = c->vChrFilter;

  572. 

  573.             if (isPlanarYUV(dstFormat) ||

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

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

  576. 

  577.                 vLumFilter +=    dstY * vLumFilterSize;

  578.                 vChrFilter += chrDstY * vChrFilterSize;

  579. 

  580. //                 av_assert0(use_mmx_vfilter != (

  581. //                                yuv2planeX == yuv2planeX_10BE_c

  582. //                             || yuv2planeX == yuv2planeX_10LE_c

  583. //                             || yuv2planeX == yuv2planeX_9BE_c

  584. //                             || yuv2planeX == yuv2planeX_9LE_c

  585. //                             || yuv2planeX == yuv2planeX_16BE_c

  586. //                             || yuv2planeX == yuv2planeX_16LE_c

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

  588. 

  589.                 if(use_mmx_vfilter){

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

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

  592.                 }

  593. 

  594.                 if (vLumFilterSize == 1) {

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

  596.                 } else {

  597.                     yuv2planeX(vLumFilter, vLumFilterSize,

  598.                                lumSrcPtr, dest[0],

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

  600.                 }

  601. 

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

  603.                     if (yuv2nv12cX) {

  604.                         yuv2nv12cX(c, vChrFilter,

  605.                                    vChrFilterSize, chrUSrcPtr, chrVSrcPtr,

  606.                                    dest[1], chrDstW);

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

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

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

  610.                     } else {

  611.                         yuv2planeX(vChrFilter,

  612.                                    vChrFilterSize, chrUSrcPtr, dest[1],

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

  614.                         yuv2planeX(vChrFilter,

  615.                                    vChrFilterSize, chrVSrcPtr, dest[2],

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

  617.                     }

  618.                 }

  619. 

  620.                 if (CONFIG_SWSCALE_ALPHA && alpPixBuf) {

  621.                     if(use_mmx_vfilter){

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

  623.                     }

  624.                     if (vLumFilterSize == 1) {

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

  626.                                    c->lumDither8, 0);

  627.                     } else {

  628.                         yuv2planeX(vLumFilter,

  629.                                    vLumFilterSize, alpSrcPtr, dest[3],

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

  631.                     }

  632.                 }

  633.             } else {

  634.                 av_assert1(lumSrcPtr  + vLumFilterSize - 1 < lumPixBuf  + vLumBufSize * 2);

  635.                 av_assert1(chrUSrcPtr + vChrFilterSize - 1 < chrUPixBuf + vChrBufSize * 2);

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

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

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

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

  640.                                 alpPixBuf ? *alpSrcPtr : NULL,

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

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

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

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

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

  646.                     lumMmxFilter[2] =

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

  648.                     chrMmxFilter[2] =

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

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

  651.                                 alpPixBuf ? alpSrcPtr : NULL,

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

  653.                 } else { // general RGB

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

  655.                                 lumSrcPtr, vLumFilterSize,

  656.                                 vChrFilter + dstY * vChrFilterSize,

  657.                                 chrUSrcPtr, chrVSrcPtr, vChrFilterSize,

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

  659.                 }

  660.             }

  661.         }

  662.     }

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

  664.         int length = dstW;

  665.         int height = dstY - lastDstY;

  666. 

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

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

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

  670.                     1, desc->comp[3].depth_minus1,

  671.                     isBE(dstFormat));

  672.         } else

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

  674.     }

  675. 

  676. #if HAVE_MMXEXT_INLINE

  677.     if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT)

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

  679. #endif

  680.     emms_c();

  681. 

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

  683.     c->dstY         = dstY;

  684.     c->lumBufIndex  = lumBufIndex;

  685.     c->chrBufIndex  = chrBufIndex;

  686.     c->lastInLumBuf = lastInLumBuf;

  687.     c->lastInChrBuf = lastInChrBuf;

  688. 

  689.     return dstY - lastDstY;

  690. }

  691. 

  692. static av_cold void sws_init_swScale_c(SwsContext *c)

  693. {

  694.     enum AVPixelFormat srcFormat = c->srcFormat;

  695. 

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

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

  698.                              &c->yuv2packed2, &c->yuv2packedX);

  699. 

  700.     ff_sws_init_input_funcs(c);

  701. 

  702. 

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

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

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

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

  707.                 c->hyscale_fast = hyscale_fast_c;

  708.                 c->hcscale_fast = hcscale_fast_c;

  709.             }

  710.         } else {

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

  712.         }

  713.     } else {

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

  715.                                                  : hScale16To15_c;

  716.     }

  717. 

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

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

  720.             if (c->srcRange) {

  721.                 c->lumConvertRange = lumRangeFromJpeg_c;

  722.                 c->chrConvertRange = chrRangeFromJpeg_c;

  723.             } else {

  724.                 c->lumConvertRange = lumRangeToJpeg_c;

  725.                 c->chrConvertRange = chrRangeToJpeg_c;

  726.             }

  727.         } else {

  728.             if (c->srcRange) {

  729.                 c->lumConvertRange = lumRangeFromJpeg16_c;

  730.                 c->chrConvertRange = chrRangeFromJpeg16_c;

  731.             } else {

  732.                 c->lumConvertRange = lumRangeToJpeg16_c;

  733.                 c->chrConvertRange = chrRangeToJpeg16_c;

  734.             }

  735.         }

  736.     }

  737. 

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

  739.           srcFormat == AV_PIX_FMT_MONOBLACK || srcFormat == AV_PIX_FMT_MONOWHITE))

  740.         c->needs_hcscale = 1;

  741. }

  742. 

  743. SwsFunc ff_getSwsFunc(SwsContext *c)

  744. {

  745.     sws_init_swScale_c(c);

  746. 

  747.     if (HAVE_MMX)

  748.         ff_sws_init_swScale_mmx(c);

  749.     if (HAVE_ALTIVEC)

  750.         ff_sws_init_swScale_altivec(c);

  751. 

  752.     return swScale;

  753. }

  754. 

  755. static void reset_ptr(const uint8_t *src[], int format)

  756. {

  757.     if (!isALPHA(format))

  758.         src[3] = NULL;

  759.     if (!isPlanar(format)) {

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

  761. 

  762.         if (!usePal(format))

  763.             src[1] = NULL;

  764.     }

  765. }

  766. 

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

  768.                                 const int linesizes[4])

  769. {

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

  771.     int i;

  772. 

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

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

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

  776.             return 0;

  777.     }

  778. 

  779.     return 1;

  780. }

  781. 

  782. /**

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

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

  785.  */

  786. int attribute_align_arg sws_scale(struct SwsContext *c,

  787.                                   const uint8_t * const srcSlice[],

  788.                                   const int srcStride[], int srcSliceY,

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

  790.                                   const int dstStride[])

  791. {

  792.     int i, ret;

  793.     const uint8_t *src2[4];

  794.     uint8_t *dst2[4];

  795.     uint8_t *rgb0_tmp = NULL;

  796. 

  797.     if (!srcSlice || !dstStride || !dst || !srcSlice) {

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

  799.         return 0;

  800.     }

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

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

  803. 

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

  805.     if (srcSliceH == 0)

  806.         return 0;

  807. 

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

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

  810.         return 0;

  811.     }

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

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

  814.         return 0;

  815.     }

  816. 

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

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

  819.         return 0;

  820.     }

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

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

  823.     }

  824. 

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

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

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

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

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

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

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

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

  833.                 b =  p        & 0xFF;

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

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

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

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

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

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

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

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

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

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

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

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

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

  847.                 r = g = b = i;

  848.             } else {

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

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

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

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

  853.             }

  854. #define RGB2YUV_SHIFT 15

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

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

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

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

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

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

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

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

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

  864. 

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

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

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

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

  869. 

  870.             switch (c->dstFormat) {

  871.             case AV_PIX_FMT_BGR32:

  872. #if !HAVE_BIGENDIAN

  873.             case AV_PIX_FMT_RGB24:

  874. #endif

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

  876.                 break;

  877.             case AV_PIX_FMT_BGR32_1:

  878. #if HAVE_BIGENDIAN

  879.             case AV_PIX_FMT_BGR24:

  880. #endif

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

  882.                 break;

  883.             case AV_PIX_FMT_RGB32_1:

  884. #if HAVE_BIGENDIAN

  885.             case AV_PIX_FMT_RGB24:

  886. #endif

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

  888.                 break;

  889.             case AV_PIX_FMT_RGB32:

  890. #if !HAVE_BIGENDIAN

  891.             case AV_PIX_FMT_BGR24:

  892. #endif

  893.             default:

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

  895.             }

  896.         }

  897.     }

  898. 

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

  900.         uint8_t *base;

  901.         int x,y;

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

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

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

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

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

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

  908.             }

  909.         }

  910.         src2[0] = base;

  911.     }

  912. 

  913.     if (!srcSliceY && (c->flags & SWS_BITEXACT) && (c->flags & SWS_ERROR_DIFFUSION))

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

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

  916. 

  917. 

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

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

  920.         // slices go from top to bottom

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

  922.                               srcStride[3] };

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

  924.                               dstStride[3] };

  925. 

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

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

  928. 

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

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

  931.             c->sliceDir = 0;

  932. 

  933.         ret = c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst2,

  934.                           dstStride2);

  935.     } else {

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

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

  938.                               -srcStride[3] };

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

  940.                               -dstStride[3] };

  941. 

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

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

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

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

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

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

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

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

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

  951. 

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

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

  954. 

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

  956.         if (!srcSliceY)

  957.             c->sliceDir = 0;

  958. 

  959.         ret = c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH,

  960.                           srcSliceH, dst2, dstStride2);

  961.     }

  962. 

  963.     av_free(rgb0_tmp);

  964.     return ret;

  965. }