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.
34093 Commits
32 Branches
283MB
Tree: 5ed032fb8b
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '5ed032fb8b'
${ noResults }
FFmpeg/libswscale/swscale.c

741 lines
29KB
Raw Blame History 

  1. /*

  2.  * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>

  3.  *

  4.  * This file is part of Libav.

  5.  *

  6.  * Libav 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.  * Libav 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 Libav; 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/avutil.h"

  28. #include "libavutil/bswap.h"

  29. #include "libavutil/cpu.h"

  30. #include "libavutil/intreadwrite.h"

  31. #include "libavutil/mathematics.h"

  32. #include "libavutil/pixdesc.h"

  33. #include "config.h"

  34. #include "rgb2rgb.h"

  35. #include "swscale_internal.h"

  36. #include "swscale.h"

  37. 

  38. DECLARE_ALIGNED(8, const uint8_t, dither_8x8_128)[8][8] = {

  39.     {  36, 68,  60, 92,  34, 66,  58, 90, },

  40.     { 100,  4, 124, 28,  98,  2, 122, 26, },

  41.     {  52, 84,  44, 76,  50, 82,  42, 74, },

  42.     { 116, 20, 108, 12, 114, 18, 106, 10, },

  43.     {  32, 64,  56, 88,  38, 70,  62, 94, },

  44.     {  96,  0, 120, 24, 102,  6, 126, 30, },

  45.     {  48, 80,  40, 72,  54, 86,  46, 78, },

  46.     { 112, 16, 104,  8, 118, 22, 110, 14, },

  47. };

  48. 

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

  50.     64, 64, 64, 64, 64, 64, 64, 64

  51. };

  52. 

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

  54.                                        int height, int y, uint8_t val)

  55. {

  56.     int i;

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

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

  59.         memset(ptr, val, width);

  60.         ptr += stride;

  61.     }

  62. }

  63. 

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

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

  66.                            const int32_t *filterPos, int filterSize)

  67. {

  68.     int i;

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

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

  71.     int bits            = av_pix_fmt_descriptors[c->srcFormat].comp[0].depth_minus1;

  72.     int sh              = bits - 4;

  73. 

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

  75.         int j;

  76.         int srcPos = filterPos[i];

  77.         int val    = 0;

  78. 

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

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

  81.         }

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

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

  84.     }

  85. }

  86. 

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

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

  89.                            const int32_t *filterPos, int filterSize)

  90. {

  91.     int i;

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

  93.     int sh              = av_pix_fmt_descriptors[c->srcFormat].comp[0].depth_minus1;

  94. 

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

  96.         int j;

  97.         int srcPos = filterPos[i];

  98.         int val    = 0;

  99. 

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

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

  102.         }

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

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

  105.     }

  106. }

  107. 

  108. // bilinear / bicubic scaling

  109. static void hScale8To15_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.     int i;

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

  115.         int j;

  116.         int srcPos = filterPos[i];

  117.         int val    = 0;

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

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

  120.         }

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

  122.     }

  123. }

  124. 

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

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

  127.                           const int32_t *filterPos, int filterSize)

  128. {

  129.     int i;

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

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

  132.         int j;

  133.         int srcPos = filterPos[i];

  134.         int val    = 0;

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

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

  137.         }

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

  139.     }

  140. }

  141. 

  142. // FIXME all pal and rgb srcFormats could do this convertion as well

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

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

  145. {

  146.     int i;

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

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

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

  150.     }

  151. }

  152. 

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

  154. {

  155.     int i;

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

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

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

  159.     }

  160. }

  161. 

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

  163. {

  164.     int i;

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

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

  167. }

  168. 

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

  170. {

  171.     int i;

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

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

  174. }

  175. 

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

  177. {

  178.     int i;

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

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

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

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

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

  184.     }

  185. }

  186. 

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

  188. {

  189.     int i;

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

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

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

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

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

  195.     }

  196. }

  197. 

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

  199. {

  200.     int i;

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

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

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

  204. }

  205. 

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

  207. {

  208.     int i;

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

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

  211.         dst[i] = (dst[i] * 14071 + (33561947 << 4)) >> 14;

  212. }

  213. 

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

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

  216. {

  217.     int i;

  218.     unsigned int xpos = 0;

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

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

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

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

  223.         xpos  += xInc;

  224.     }

  225. }

  226. 

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

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

  229.                                      const uint8_t *src_in[4],

  230.                                      int srcW, int xInc,

  231.                                      const int16_t *hLumFilter,

  232.                                      const int32_t *hLumFilterPos,

  233.                                      int hLumFilterSize,

  234.                                      uint8_t *formatConvBuffer,

  235.                                      uint32_t *pal, int isAlpha)

  236. {

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

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

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

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

  241. 

  242.     if (toYV12) {

  243.         toYV12(formatConvBuffer, src, srcW, pal);

  244.         src = formatConvBuffer;

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

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

  247.         src = formatConvBuffer;

  248.     }

  249. 

  250.     if (!c->hyscale_fast) {

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

  252.                    hLumFilterPos, hLumFilterSize);

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

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

  255.     }

  256. 

  257.     if (convertRange)

  258.         convertRange(dst, dstWidth);

  259. }

  260. 

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

  262.                            int dstWidth, const uint8_t *src1,

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

  264. {

  265.     int i;

  266.     unsigned int xpos = 0;

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

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

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

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

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

  272.         xpos   += xInc;

  273.     }

  274. }

  275. 

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

  277.                                      int16_t *dst2, int dstWidth,

  278.                                      const uint8_t *src_in[4],

  279.                                      int srcW, int xInc,

  280.                                      const int16_t *hChrFilter,

  281.                                      const int32_t *hChrFilterPos,

  282.                                      int hChrFilterSize,

  283.                                      uint8_t *formatConvBuffer, uint32_t *pal)

  284. {

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

  286.     if (c->chrToYV12) {

  287.         uint8_t *buf2 = formatConvBuffer +

  288.                         FFALIGN(srcW * FFALIGN(c->srcBpc, 8) >> 3, 16);

  289.         c->chrToYV12(formatConvBuffer, buf2, src1, src2, srcW, pal);

  290.         src1 = formatConvBuffer;

  291.         src2 = buf2;

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

  293.         uint8_t *buf2 = formatConvBuffer +

  294.                         FFALIGN(srcW * FFALIGN(c->srcBpc, 8) >> 3, 16);

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

  296.         src1 = formatConvBuffer;

  297.         src2 = buf2;

  298.     }

  299. 

  300.     if (!c->hcscale_fast) {

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

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

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

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

  305.     }

  306. 

  307.     if (c->chrConvertRange)

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

  309. }

  310. 

  311. #define DEBUG_SWSCALE_BUFFERS 0

  312. #define DEBUG_BUFFERS(...)                      \

  313.     if (DEBUG_SWSCALE_BUFFERS)                  \

  314.         av_log(c, AV_LOG_DEBUG, __VA_ARGS__)

  315. 

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

  317.                    int srcStride[], int srcSliceY,

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

  319. {

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

  321.      * and faster */

  322.     const int srcW                   = c->srcW;

  323.     const int dstW                   = c->dstW;

  324.     const int dstH                   = c->dstH;

  325.     const int chrDstW                = c->chrDstW;

  326.     const int chrSrcW                = c->chrSrcW;

  327.     const int lumXInc                = c->lumXInc;

  328.     const int chrXInc                = c->chrXInc;

  329.     const enum PixelFormat dstFormat = c->dstFormat;

  330.     const int flags                  = c->flags;

  331.     int32_t *vLumFilterPos           = c->vLumFilterPos;

  332.     int32_t *vChrFilterPos           = c->vChrFilterPos;

  333.     int32_t *hLumFilterPos           = c->hLumFilterPos;

  334.     int32_t *hChrFilterPos           = c->hChrFilterPos;

  335.     int16_t *vLumFilter              = c->vLumFilter;

  336.     int16_t *vChrFilter              = c->vChrFilter;

  337.     int16_t *hLumFilter              = c->hLumFilter;

  338.     int16_t *hChrFilter              = c->hChrFilter;

  339.     int32_t *lumMmxFilter            = c->lumMmxFilter;

  340.     int32_t *chrMmxFilter            = c->chrMmxFilter;

  341.     const int vLumFilterSize         = c->vLumFilterSize;

  342.     const int vChrFilterSize         = c->vChrFilterSize;

  343.     const int hLumFilterSize         = c->hLumFilterSize;

  344.     const int hChrFilterSize         = c->hChrFilterSize;

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

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

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

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

  349.     const int vLumBufSize            = c->vLumBufSize;

  350.     const int vChrBufSize            = c->vChrBufSize;

  351.     uint8_t *formatConvBuffer        = c->formatConvBuffer;

  352.     uint32_t *pal                    = c->pal_yuv;

  353.     yuv2planar1_fn yuv2plane1        = c->yuv2plane1;

  354.     yuv2planarX_fn yuv2planeX        = c->yuv2planeX;

  355.     yuv2interleavedX_fn yuv2nv12cX   = c->yuv2nv12cX;

  356.     yuv2packed1_fn yuv2packed1       = c->yuv2packed1;

  357.     yuv2packed2_fn yuv2packed2       = c->yuv2packed2;

  358.     yuv2packedX_fn yuv2packedX       = c->yuv2packedX;

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

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

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

  362.                                        is16BPS(c->srcFormat);

  363.     int lastDstY;

  364. 

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

  366.     int dstY         = c->dstY;

  367.     int lumBufIndex  = c->lumBufIndex;

  368.     int chrBufIndex  = c->chrBufIndex;

  369.     int lastInLumBuf = c->lastInLumBuf;

  370.     int lastInChrBuf = c->lastInChrBuf;

  371. 

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

  373.         src[0] =

  374.         src[1] =

  375.         src[2] =

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

  377.         srcStride[0] =

  378.         srcStride[1] =

  379.         srcStride[2] =

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

  381.     }

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

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

  384. 

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

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

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

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

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

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

  391.                   srcSliceY, srcSliceH, dstY, dstH);

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

  393.                   vLumFilterSize, vLumBufSize, vChrFilterSize, vChrBufSize);

  394. 

  395.     if (dstStride[0] % 8 != 0 || dstStride[1] % 8 != 0 ||

  396.         dstStride[2] % 8 != 0 || dstStride[3] % 8 != 0) {

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

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

  399.             av_log(c, AV_LOG_WARNING,

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

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

  402.             warnedAlready = 1;

  403.         }

  404.     }

  405. 

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

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

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

  409.     if (srcSliceY == 0) {

  410.         lumBufIndex  = -1;

  411.         chrBufIndex  = -1;

  412.         dstY         = 0;

  413.         lastInLumBuf = -1;

  414.         lastInChrBuf = -1;

  415.     }

  416. 

  417.     if (!should_dither) {

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

  419.     }

  420.     lastDstY = dstY;

  421. 

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

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

  424.         uint8_t *dest[4]  = {

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

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

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

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

  429.         };

  430. 

  431.         // First line needed as input

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

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

  434.         // First line needed as input

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

  436. 

  437.         // Last line needed as input

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

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

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

  441.         int enough_lines;

  442. 

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

  444.         if (firstLumSrcY > lastInLumBuf)

  445.             lastInLumBuf = firstLumSrcY - 1;

  446.         if (firstChrSrcY > lastInChrBuf)

  447.             lastInChrBuf = firstChrSrcY - 1;

  448.         assert(firstLumSrcY >= lastInLumBuf - vLumBufSize + 1);

  449.         assert(firstChrSrcY >= lastInChrBuf - vChrBufSize + 1);

  450. 

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

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

  453.                       firstLumSrcY, lastLumSrcY, lastInLumBuf);

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

  455.                       firstChrSrcY, lastChrSrcY, lastInChrBuf);

  456. 

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

  458.         enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&

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

  460. 

  461.         if (!enough_lines) {

  462.             lastLumSrcY = srcSliceY + srcSliceH - 1;

  463.             lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;

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

  465.                           lastLumSrcY, lastChrSrcY);

  466.         }

  467. 

  468.         // Do horizontal scaling

  469.         while (lastInLumBuf < lastLumSrcY) {

  470.             const uint8_t *src1[4] = {

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

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

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

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

  475.             };

  476.             lumBufIndex++;

  477.             assert(lumBufIndex < 2 * vLumBufSize);

  478.             assert(lastInLumBuf + 1 - srcSliceY < srcSliceH);

  479.             assert(lastInLumBuf + 1 - srcSliceY >= 0);

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

  481.                     hLumFilter, hLumFilterPos, hLumFilterSize,

  482.                     formatConvBuffer, pal, 0);

  483.             if (CONFIG_SWSCALE_ALPHA && alpPixBuf)

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

  485.                         lumXInc, hLumFilter, hLumFilterPos, hLumFilterSize,

  486.                         formatConvBuffer, pal, 1);

  487.             lastInLumBuf++;

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

  489.                           lumBufIndex, lastInLumBuf);

  490.         }

  491.         while (lastInChrBuf < lastChrSrcY) {

  492.             const uint8_t *src1[4] = {

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

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

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

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

  497.             };

  498.             chrBufIndex++;

  499.             assert(chrBufIndex < 2 * vChrBufSize);

  500.             assert(lastInChrBuf + 1 - chrSrcSliceY < (chrSrcSliceH));

  501.             assert(lastInChrBuf + 1 - chrSrcSliceY >= 0);

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

  503. 

  504.             if (c->needs_hcscale)

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

  506.                         chrDstW, src1, chrSrcW, chrXInc,

  507.                         hChrFilter, hChrFilterPos, hChrFilterSize,

  508.                         formatConvBuffer, pal);

  509.             lastInChrBuf++;

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

  511.                           chrBufIndex, lastInChrBuf);

  512.         }

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

  514.         if (lumBufIndex >= vLumBufSize)

  515.             lumBufIndex -= vLumBufSize;

  516.         if (chrBufIndex >= vChrBufSize)

  517.             chrBufIndex -= vChrBufSize;

  518.         if (!enough_lines)

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

  520. 

  521. #if HAVE_MMX

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

  523.                               lastInLumBuf, lastInChrBuf);

  524. #endif

  525.         if (should_dither) {

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

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

  528.         }

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

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

  531.              * this array's tail */

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

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

  534.         }

  535. 

  536.         {

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

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

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

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

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

  542. 

  543.             if (firstLumSrcY < 0 || firstLumSrcY + vLumFilterSize > c->srcH) {

  544.                 const int16_t **tmpY = (const int16_t **)lumPixBuf +

  545.                                        2 * vLumBufSize;

  546.                 int neg = -firstLumSrcY, i;

  547.                 int end = FFMIN(c->srcH - firstLumSrcY, vLumFilterSize);

  548.                 for (i = 0; i < neg; i++)

  549.                     tmpY[i] = lumSrcPtr[neg];

  550.                 for (; i < end; i++)

  551.                     tmpY[i] = lumSrcPtr[i];

  552.                 for (; i < vLumFilterSize; i++)

  553.                     tmpY[i] = tmpY[i - 1];

  554.                 lumSrcPtr = tmpY;

  555. 

  556.                 if (alpSrcPtr) {

  557.                     const int16_t **tmpA = (const int16_t **)alpPixBuf +

  558.                                            2 * vLumBufSize;

  559.                     for (i = 0; i < neg; i++)

  560.                         tmpA[i] = alpSrcPtr[neg];

  561.                     for (; i < end; i++)

  562.                         tmpA[i] = alpSrcPtr[i];

  563.                     for (; i < vLumFilterSize; i++)

  564.                         tmpA[i] = tmpA[i - 1];

  565.                     alpSrcPtr = tmpA;

  566.                 }

  567.             }

  568.             if (firstChrSrcY < 0 ||

  569.                 firstChrSrcY + vChrFilterSize > c->chrSrcH) {

  570.                 const int16_t **tmpU = (const int16_t **)chrUPixBuf + 2 * vChrBufSize,

  571.                 **tmpV               = (const int16_t **)chrVPixBuf + 2 * vChrBufSize;

  572.                 int neg = -firstChrSrcY, i;

  573.                 int end = FFMIN(c->chrSrcH - firstChrSrcY, vChrFilterSize);

  574.                 for (i = 0; i < neg; i++) {

  575.                     tmpU[i] = chrUSrcPtr[neg];

  576.                     tmpV[i] = chrVSrcPtr[neg];

  577.                 }

  578.                 for (; i < end; i++) {

  579.                     tmpU[i] = chrUSrcPtr[i];

  580.                     tmpV[i] = chrVSrcPtr[i];

  581.                 }

  582.                 for (; i < vChrFilterSize; i++) {

  583.                     tmpU[i] = tmpU[i - 1];

  584.                     tmpV[i] = tmpV[i - 1];

  585.                 }

  586.                 chrUSrcPtr = tmpU;

  587.                 chrVSrcPtr = tmpV;

  588.             }

  589. 

  590.             if (isPlanarYUV(dstFormat) ||

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

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

  593. 

  594.                 if (vLumFilterSize == 1) {

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

  596.                 } else {

  597.                     yuv2planeX(vLumFilter + dstY * vLumFilterSize,

  598.                                vLumFilterSize, lumSrcPtr, dest[0],

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

  600.                 }

  601. 

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

  603.                     if (yuv2nv12cX) {

  604.                         yuv2nv12cX(c, vChrFilter + chrDstY * vChrFilterSize,

  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 + chrDstY * vChrFilterSize,

  612.                                    vChrFilterSize, chrUSrcPtr, dest[1],

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

  614.                         yuv2planeX(vChrFilter + chrDstY * vChrFilterSize,

  615.                                    vChrFilterSize, chrVSrcPtr, dest[2],

  616.                                    chrDstW, c->chrDither8, 3);

  617.                     }

  618.                 }

  619. 

  620.                 if (CONFIG_SWSCALE_ALPHA && alpPixBuf) {

  621.                     if (vLumFilterSize == 1) {

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

  623.                                    c->lumDither8, 0);

  624.                     } else {

  625.                         yuv2planeX(vLumFilter + dstY * vLumFilterSize,

  626.                                    vLumFilterSize, alpSrcPtr, dest[3],

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

  628.                     }

  629.                 }

  630.             } else {

  631.                 assert(lumSrcPtr  + vLumFilterSize - 1 < lumPixBuf  + vLumBufSize * 2);

  632.                 assert(chrUSrcPtr + vChrFilterSize - 1 < chrUPixBuf + vChrBufSize * 2);

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

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

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

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

  637.                                 alpPixBuf ? *alpSrcPtr : NULL,

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

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

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

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

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

  643.                     lumMmxFilter[2] =

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

  645.                     chrMmxFilter[2] =

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

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

  648.                                 alpPixBuf ? alpSrcPtr : NULL,

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

  650.                 } else { // general RGB

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

  652.                                 lumSrcPtr, vLumFilterSize,

  653.                                 vChrFilter + dstY * vChrFilterSize,

  654.                                 chrUSrcPtr, chrVSrcPtr, vChrFilterSize,

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

  656.                 }

  657.             }

  658.         }

  659.     }

  660. 

  661.     if (isPlanar(dstFormat) && isALPHA(dstFormat) && !alpPixBuf)

  662.         fillPlane(dst[3], dstStride[3], dstW, dstY - lastDstY, lastDstY, 255);

  663. 

  664. #if HAVE_MMX2

  665.     if (av_get_cpu_flags() & AV_CPU_FLAG_MMX2)

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

  667. #endif

  668.     emms_c();

  669. 

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

  671.     c->dstY         = dstY;

  672.     c->lumBufIndex  = lumBufIndex;

  673.     c->chrBufIndex  = chrBufIndex;

  674.     c->lastInLumBuf = lastInLumBuf;

  675.     c->lastInChrBuf = lastInChrBuf;

  676. 

  677.     return dstY - lastDstY;

  678. }

  679. 

  680. static av_cold void sws_init_swScale_c(SwsContext *c)

  681. {

  682.     enum PixelFormat srcFormat = c->srcFormat;

  683. 

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

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

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

  687. 

  688.     ff_sws_init_input_funcs(c);

  689. 

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

  691.         if (c->dstBpc <= 10) {

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

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

  694.                 c->hyscale_fast = hyscale_fast_c;

  695.                 c->hcscale_fast = hcscale_fast_c;

  696.             }

  697.         } else {

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

  699.         }

  700.     } else {

  701.         c->hyScale = c->hcScale = c->dstBpc > 10 ? hScale16To19_c

  702.                                                  : hScale16To15_c;

  703.     }

  704. 

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

  706.         if (c->dstBpc <= 10) {

  707.             if (c->srcRange) {

  708.                 c->lumConvertRange = lumRangeFromJpeg_c;

  709.                 c->chrConvertRange = chrRangeFromJpeg_c;

  710.             } else {

  711.                 c->lumConvertRange = lumRangeToJpeg_c;

  712.                 c->chrConvertRange = chrRangeToJpeg_c;

  713.             }

  714.         } else {

  715.             if (c->srcRange) {

  716.                 c->lumConvertRange = lumRangeFromJpeg16_c;

  717.                 c->chrConvertRange = chrRangeFromJpeg16_c;

  718.             } else {

  719.                 c->lumConvertRange = lumRangeToJpeg16_c;

  720.                 c->chrConvertRange = chrRangeToJpeg16_c;

  721.             }

  722.         }

  723.     }

  724. 

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

  726.           srcFormat == PIX_FMT_MONOBLACK || srcFormat == PIX_FMT_MONOWHITE))

  727.         c->needs_hcscale = 1;

  728. }

  729. 

  730. SwsFunc ff_getSwsFunc(SwsContext *c)

  731. {

  732.     sws_init_swScale_c(c);

  733. 

  734.     if (HAVE_MMX)

  735.         ff_sws_init_swScale_mmx(c);

  736.     if (HAVE_ALTIVEC)

  737.         ff_sws_init_swScale_altivec(c);

  738. 

  739.     return swScale;

  740. }