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.
52550 Commits
32 Branches
5.3GB
Tree: 0c47c9028b
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '0c47c9028b'
${ noResults }
FFmpeg/libswscale/utils.c

2019 lines
73KB
Raw Blame History 

  1. /*

  2.  * Copyright (C) 2001-2003 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 "config.h"

  22. 

  23. #define _SVID_SOURCE // needed for MAP_ANONYMOUS

  24. #define _DARWIN_C_SOURCE // needed for MAP_ANON

  25. #include <inttypes.h>

  26. #include <math.h>

  27. #include <stdio.h>

  28. #include <string.h>

  29. #if HAVE_SYS_MMAN_H

  30. #include <sys/mman.h>

  31. #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)

  32. #define MAP_ANONYMOUS MAP_ANON

  33. #endif

  34. #endif

  35. #if HAVE_VIRTUALALLOC

  36. #define WIN32_LEAN_AND_MEAN

  37. #include <windows.h>

  38. #endif

  39. 

  40. #include "libavutil/attributes.h"

  41. #include "libavutil/avassert.h"

  42. #include "libavutil/avutil.h"

  43. #include "libavutil/bswap.h"

  44. #include "libavutil/cpu.h"

  45. #include "libavutil/intreadwrite.h"

  46. #include "libavutil/mathematics.h"

  47. #include "libavutil/opt.h"

  48. #include "libavutil/pixdesc.h"

  49. #include "libavutil/x86/asm.h"

  50. #include "libavutil/x86/cpu.h"

  51. #include "rgb2rgb.h"

  52. #include "swscale.h"

  53. #include "swscale_internal.h"

  54. 

  55. unsigned swscale_version(void)

  56. {

  57.     av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);

  58.     return LIBSWSCALE_VERSION_INT;

  59. }

  60. 

  61. const char *swscale_configuration(void)

  62. {

  63.     return FFMPEG_CONFIGURATION;

  64. }

  65. 

  66. const char *swscale_license(void)

  67. {

  68. #define LICENSE_PREFIX "libswscale license: "

  69.     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;

  70. }

  71. 

  72. #define RET 0xC3 // near return opcode for x86

  73. 

  74. typedef struct FormatEntry {

  75.     int is_supported_in, is_supported_out;

  76. } FormatEntry;

  77. 

  78. static const FormatEntry format_entries[AV_PIX_FMT_NB] = {

  79.     [AV_PIX_FMT_YUV420P]     = { 1, 1 },

  80.     [AV_PIX_FMT_YUYV422]     = { 1, 1 },

  81.     [AV_PIX_FMT_RGB24]       = { 1, 1 },

  82.     [AV_PIX_FMT_BGR24]       = { 1, 1 },

  83.     [AV_PIX_FMT_YUV422P]     = { 1, 1 },

  84.     [AV_PIX_FMT_YUV444P]     = { 1, 1 },

  85.     [AV_PIX_FMT_YUV410P]     = { 1, 1 },

  86.     [AV_PIX_FMT_YUV411P]     = { 1, 1 },

  87.     [AV_PIX_FMT_GRAY8]       = { 1, 1 },

  88.     [AV_PIX_FMT_MONOWHITE]   = { 1, 1 },

  89.     [AV_PIX_FMT_MONOBLACK]   = { 1, 1 },

  90.     [AV_PIX_FMT_PAL8]        = { 1, 0 },

  91.     [AV_PIX_FMT_YUVJ420P]    = { 1, 1 },

  92.     [AV_PIX_FMT_YUVJ422P]    = { 1, 1 },

  93.     [AV_PIX_FMT_YUVJ444P]    = { 1, 1 },

  94.     [AV_PIX_FMT_UYVY422]     = { 1, 1 },

  95.     [AV_PIX_FMT_UYYVYY411]   = { 0, 0 },

  96.     [AV_PIX_FMT_BGR8]        = { 1, 1 },

  97.     [AV_PIX_FMT_BGR4]        = { 0, 1 },

  98.     [AV_PIX_FMT_BGR4_BYTE]   = { 1, 1 },

  99.     [AV_PIX_FMT_RGB8]        = { 1, 1 },

  100.     [AV_PIX_FMT_RGB4]        = { 0, 1 },

  101.     [AV_PIX_FMT_RGB4_BYTE]   = { 1, 1 },

  102.     [AV_PIX_FMT_NV12]        = { 1, 1 },

  103.     [AV_PIX_FMT_NV21]        = { 1, 1 },

  104.     [AV_PIX_FMT_ARGB]        = { 1, 1 },

  105.     [AV_PIX_FMT_RGBA]        = { 1, 1 },

  106.     [AV_PIX_FMT_ABGR]        = { 1, 1 },

  107.     [AV_PIX_FMT_BGRA]        = { 1, 1 },

  108.     [AV_PIX_FMT_0RGB]        = { 1, 1 },

  109.     [AV_PIX_FMT_RGB0]        = { 1, 1 },

  110.     [AV_PIX_FMT_0BGR]        = { 1, 1 },

  111.     [AV_PIX_FMT_BGR0]        = { 1, 1 },

  112.     [AV_PIX_FMT_GRAY16BE]    = { 1, 1 },

  113.     [AV_PIX_FMT_GRAY16LE]    = { 1, 1 },

  114.     [AV_PIX_FMT_YUV440P]     = { 1, 1 },

  115.     [AV_PIX_FMT_YUVJ440P]    = { 1, 1 },

  116.     [AV_PIX_FMT_YUVA420P]    = { 1, 1 },

  117.     [AV_PIX_FMT_YUVA422P]    = { 1, 1 },

  118.     [AV_PIX_FMT_YUVA444P]    = { 1, 1 },

  119.     [AV_PIX_FMT_YUVA420P9BE] = { 1, 1 },

  120.     [AV_PIX_FMT_YUVA420P9LE] = { 1, 1 },

  121.     [AV_PIX_FMT_YUVA422P9BE] = { 1, 1 },

  122.     [AV_PIX_FMT_YUVA422P9LE] = { 1, 1 },

  123.     [AV_PIX_FMT_YUVA444P9BE] = { 1, 1 },

  124.     [AV_PIX_FMT_YUVA444P9LE] = { 1, 1 },

  125.     [AV_PIX_FMT_YUVA420P10BE]= { 1, 1 },

  126.     [AV_PIX_FMT_YUVA420P10LE]= { 1, 1 },

  127.     [AV_PIX_FMT_YUVA422P10BE]= { 1, 1 },

  128.     [AV_PIX_FMT_YUVA422P10LE]= { 1, 1 },

  129.     [AV_PIX_FMT_YUVA444P10BE]= { 1, 1 },

  130.     [AV_PIX_FMT_YUVA444P10LE]= { 1, 1 },

  131.     [AV_PIX_FMT_YUVA420P16BE]= { 1, 1 },

  132.     [AV_PIX_FMT_YUVA420P16LE]= { 1, 1 },

  133.     [AV_PIX_FMT_YUVA422P16BE]= { 1, 1 },

  134.     [AV_PIX_FMT_YUVA422P16LE]= { 1, 1 },

  135.     [AV_PIX_FMT_YUVA444P16BE]= { 1, 1 },

  136.     [AV_PIX_FMT_YUVA444P16LE]= { 1, 1 },

  137.     [AV_PIX_FMT_RGB48BE]     = { 1, 1 },

  138.     [AV_PIX_FMT_RGB48LE]     = { 1, 1 },

  139.     [AV_PIX_FMT_RGBA64BE]    = { 1, 0 },

  140.     [AV_PIX_FMT_RGBA64LE]    = { 1, 0 },

  141.     [AV_PIX_FMT_RGB565BE]    = { 1, 1 },

  142.     [AV_PIX_FMT_RGB565LE]    = { 1, 1 },

  143.     [AV_PIX_FMT_RGB555BE]    = { 1, 1 },

  144.     [AV_PIX_FMT_RGB555LE]    = { 1, 1 },

  145.     [AV_PIX_FMT_BGR565BE]    = { 1, 1 },

  146.     [AV_PIX_FMT_BGR565LE]    = { 1, 1 },

  147.     [AV_PIX_FMT_BGR555BE]    = { 1, 1 },

  148.     [AV_PIX_FMT_BGR555LE]    = { 1, 1 },

  149.     [AV_PIX_FMT_YUV420P16LE] = { 1, 1 },

  150.     [AV_PIX_FMT_YUV420P16BE] = { 1, 1 },

  151.     [AV_PIX_FMT_YUV422P16LE] = { 1, 1 },

  152.     [AV_PIX_FMT_YUV422P16BE] = { 1, 1 },

  153.     [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },

  154.     [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },

  155.     [AV_PIX_FMT_RGB444LE]    = { 1, 1 },

  156.     [AV_PIX_FMT_RGB444BE]    = { 1, 1 },

  157.     [AV_PIX_FMT_BGR444LE]    = { 1, 1 },

  158.     [AV_PIX_FMT_BGR444BE]    = { 1, 1 },

  159.     [AV_PIX_FMT_Y400A]       = { 1, 0 },

  160.     [AV_PIX_FMT_BGR48BE]     = { 1, 1 },

  161.     [AV_PIX_FMT_BGR48LE]     = { 1, 1 },

  162.     [AV_PIX_FMT_BGRA64BE]    = { 0, 0 },

  163.     [AV_PIX_FMT_BGRA64LE]    = { 0, 0 },

  164.     [AV_PIX_FMT_YUV420P9BE]  = { 1, 1 },

  165.     [AV_PIX_FMT_YUV420P9LE]  = { 1, 1 },

  166.     [AV_PIX_FMT_YUV420P10BE] = { 1, 1 },

  167.     [AV_PIX_FMT_YUV420P10LE] = { 1, 1 },

  168.     [AV_PIX_FMT_YUV420P12BE] = { 1, 1 },

  169.     [AV_PIX_FMT_YUV420P12LE] = { 1, 1 },

  170.     [AV_PIX_FMT_YUV420P14BE] = { 1, 1 },

  171.     [AV_PIX_FMT_YUV420P14LE] = { 1, 1 },

  172.     [AV_PIX_FMT_YUV422P9BE]  = { 1, 1 },

  173.     [AV_PIX_FMT_YUV422P9LE]  = { 1, 1 },

  174.     [AV_PIX_FMT_YUV422P10BE] = { 1, 1 },

  175.     [AV_PIX_FMT_YUV422P10LE] = { 1, 1 },

  176.     [AV_PIX_FMT_YUV422P12BE] = { 1, 1 },

  177.     [AV_PIX_FMT_YUV422P12LE] = { 1, 1 },

  178.     [AV_PIX_FMT_YUV422P14BE] = { 1, 1 },

  179.     [AV_PIX_FMT_YUV422P14LE] = { 1, 1 },

  180.     [AV_PIX_FMT_YUV444P9BE]  = { 1, 1 },

  181.     [AV_PIX_FMT_YUV444P9LE]  = { 1, 1 },

  182.     [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },

  183.     [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },

  184.     [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },

  185.     [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },

  186.     [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },

  187.     [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },

  188.     [AV_PIX_FMT_GBRP]        = { 1, 1 },

  189.     [AV_PIX_FMT_GBRP9LE]     = { 1, 1 },

  190.     [AV_PIX_FMT_GBRP9BE]     = { 1, 1 },

  191.     [AV_PIX_FMT_GBRP10LE]    = { 1, 1 },

  192.     [AV_PIX_FMT_GBRP10BE]    = { 1, 1 },

  193.     [AV_PIX_FMT_GBRP12LE]    = { 1, 1 },

  194.     [AV_PIX_FMT_GBRP12BE]    = { 1, 1 },

  195.     [AV_PIX_FMT_GBRP14LE]    = { 1, 1 },

  196.     [AV_PIX_FMT_GBRP14BE]    = { 1, 1 },

  197.     [AV_PIX_FMT_GBRP16LE]    = { 1, 0 },

  198.     [AV_PIX_FMT_GBRP16BE]    = { 1, 0 },

  199.     [AV_PIX_FMT_XYZ12BE]     = { 1, 0 },

  200.     [AV_PIX_FMT_XYZ12LE]     = { 1, 0 },

  201. };

  202. 

  203. int sws_isSupportedInput(enum AVPixelFormat pix_fmt)

  204. {

  205.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  206.            format_entries[pix_fmt].is_supported_in : 0;

  207. }

  208. 

  209. int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)

  210. {

  211.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  212.            format_entries[pix_fmt].is_supported_out : 0;

  213. }

  214. 

  215. extern const int32_t ff_yuv2rgb_coeffs[8][4];

  216. 

  217. #if FF_API_SWS_FORMAT_NAME

  218. const char *sws_format_name(enum AVPixelFormat format)

  219. {

  220.     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);

  221.     if (desc)

  222.         return desc->name;

  223.     else

  224.         return "Unknown format";

  225. }

  226. #endif

  227. 

  228. static double getSplineCoeff(double a, double b, double c, double d,

  229.                              double dist)

  230. {

  231.     if (dist <= 1.0)

  232.         return ((d * dist + c) * dist + b) * dist + a;

  233.     else

  234.         return getSplineCoeff(0.0,

  235.                                b + 2.0 * c + 3.0 * d,

  236.                                c + 3.0 * d,

  237.                               -b - 3.0 * c - 6.0 * d,

  238.                               dist - 1.0);

  239. }

  240. 

  241. static int initFilter(int16_t **outFilter, int32_t **filterPos,

  242.                       int *outFilterSize, int xInc, int srcW, int dstW,

  243.                       int filterAlign, int one, int flags, int cpu_flags,

  244.                       SwsVector *srcFilter, SwsVector *dstFilter,

  245.                       double param[2])

  246. {

  247.     int i;

  248.     int filterSize;

  249.     int filter2Size;

  250.     int minFilterSize;

  251.     int64_t *filter    = NULL;

  252.     int64_t *filter2   = NULL;

  253.     const int64_t fone = 1LL << (54 - FFMIN(av_log2(srcW/dstW), 8));

  254.     int ret            = -1;

  255. 

  256.     emms_c(); // FIXME should not be required but IS (even for non-MMX versions)

  257. 

  258.     // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end

  259.     FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);

  260. 

  261.     if (FFABS(xInc - 0x10000) < 10) { // unscaled

  262.         int i;

  263.         filterSize = 1;

  264.         FF_ALLOCZ_OR_GOTO(NULL, filter,

  265.                           dstW * sizeof(*filter) * filterSize, fail);

  266. 

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

  268.             filter[i * filterSize] = fone;

  269.             (*filterPos)[i]        = i;

  270.         }

  271.     } else if (flags & SWS_POINT) { // lame looking point sampling mode

  272.         int i;

  273.         int64_t xDstInSrc;

  274.         filterSize = 1;

  275.         FF_ALLOC_OR_GOTO(NULL, filter,

  276.                          dstW * sizeof(*filter) * filterSize, fail);

  277. 

  278.         xDstInSrc = xInc / 2 - 0x8000;

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

  280.             int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;

  281. 

  282.             (*filterPos)[i] = xx;

  283.             filter[i]       = fone;

  284.             xDstInSrc      += xInc;

  285.         }

  286.     } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||

  287.                (flags & SWS_FAST_BILINEAR)) { // bilinear upscale

  288.         int i;

  289.         int64_t xDstInSrc;

  290.         filterSize = 2;

  291.         FF_ALLOC_OR_GOTO(NULL, filter,

  292.                          dstW * sizeof(*filter) * filterSize, fail);

  293. 

  294.         xDstInSrc = xInc / 2 - 0x8000;

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

  296.             int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;

  297.             int j;

  298. 

  299.             (*filterPos)[i] = xx;

  300.             // bilinear upscale / linear interpolate / area averaging

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

  302.                 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);

  303.                 if (coeff < 0)

  304.                     coeff = 0;

  305.                 filter[i * filterSize + j] = coeff;

  306.                 xx++;

  307.             }

  308.             xDstInSrc += xInc;

  309.         }

  310.     } else {

  311.         int64_t xDstInSrc;

  312.         int sizeFactor;

  313. 

  314.         if (flags & SWS_BICUBIC)

  315.             sizeFactor = 4;

  316.         else if (flags & SWS_X)

  317.             sizeFactor = 8;

  318.         else if (flags & SWS_AREA)

  319.             sizeFactor = 1;     // downscale only, for upscale it is bilinear

  320.         else if (flags & SWS_GAUSS)

  321.             sizeFactor = 8;     // infinite ;)

  322.         else if (flags & SWS_LANCZOS)

  323.             sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;

  324.         else if (flags & SWS_SINC)

  325.             sizeFactor = 20;    // infinite ;)

  326.         else if (flags & SWS_SPLINE)

  327.             sizeFactor = 20;    // infinite ;)

  328.         else if (flags & SWS_BILINEAR)

  329.             sizeFactor = 2;

  330.         else {

  331.             av_assert0(0);

  332.         }

  333. 

  334.         if (xInc <= 1 << 16)

  335.             filterSize = 1 + sizeFactor;    // upscale

  336.         else

  337.             filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;

  338. 

  339.         filterSize = FFMIN(filterSize, srcW - 2);

  340.         filterSize = FFMAX(filterSize, 1);

  341. 

  342.         FF_ALLOC_OR_GOTO(NULL, filter,

  343.                          dstW * sizeof(*filter) * filterSize, fail);

  344. 

  345.         xDstInSrc = xInc - 0x10000;

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

  347.             int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);

  348.             int j;

  349.             (*filterPos)[i] = xx;

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

  351.                 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;

  352.                 double floatd;

  353.                 int64_t coeff;

  354. 

  355.                 if (xInc > 1 << 16)

  356.                     d = d * dstW / srcW;

  357.                 floatd = d * (1.0 / (1 << 30));

  358. 

  359.                 if (flags & SWS_BICUBIC) {

  360.                     int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] :   0) * (1 << 24);

  361.                     int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);

  362. 

  363.                     if (d >= 1LL << 31) {

  364.                         coeff = 0.0;

  365.                     } else {

  366.                         int64_t dd  = (d  * d) >> 30;

  367.                         int64_t ddd = (dd * d) >> 30;

  368. 

  369.                         if (d < 1LL << 30)

  370.                             coeff =  (12 * (1 << 24) -  9 * B - 6 * C) * ddd +

  371.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  372.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  373.                         else

  374.                             coeff =      (-B -  6 * C) * ddd +

  375.                                       (6 * B + 30 * C) * dd  +

  376.                                     (-12 * B - 48 * C) * d   +

  377.                                       (8 * B + 24 * C) * (1 << 30);

  378.                     }

  379.                     coeff /= (1LL<<54)/fone;

  380.                 }

  381. #if 0

  382.                 else if (flags & SWS_X) {

  383.                     double p  = param ? param * 0.01 : 0.3;

  384.                     coeff     = d ? sin(d * M_PI) / (d * M_PI) : 1.0;

  385.                     coeff    *= pow(2.0, -p * d * d);

  386.                 }

  387. #endif

  388.                 else if (flags & SWS_X) {

  389.                     double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;

  390.                     double c;

  391. 

  392.                     if (floatd < 1.0)

  393.                         c = cos(floatd * M_PI);

  394.                     else

  395.                         c = -1.0;

  396.                     if (c < 0.0)

  397.                         c = -pow(-c, A);

  398.                     else

  399.                         c = pow(c, A);

  400.                     coeff = (c * 0.5 + 0.5) * fone;

  401.                 } else if (flags & SWS_AREA) {

  402.                     int64_t d2 = d - (1 << 29);

  403.                     if (d2 * xInc < -(1LL << (29 + 16)))

  404.                         coeff = 1.0 * (1LL << (30 + 16));

  405.                     else if (d2 * xInc < (1LL << (29 + 16)))

  406.                         coeff = -d2 * xInc + (1LL << (29 + 16));

  407.                     else

  408.                         coeff = 0.0;

  409.                     coeff *= fone >> (30 + 16);

  410.                 } else if (flags & SWS_GAUSS) {

  411.                     double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;

  412.                     coeff = (pow(2.0, -p * floatd * floatd)) * fone;

  413.                 } else if (flags & SWS_SINC) {

  414.                     coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;

  415.                 } else if (flags & SWS_LANCZOS) {

  416.                     double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;

  417.                     coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /

  418.                              (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;

  419.                     if (floatd > p)

  420.                         coeff = 0;

  421.                 } else if (flags & SWS_BILINEAR) {

  422.                     coeff = (1 << 30) - d;

  423.                     if (coeff < 0)

  424.                         coeff = 0;

  425.                     coeff *= fone >> 30;

  426.                 } else if (flags & SWS_SPLINE) {

  427.                     double p = -2.196152422706632;

  428.                     coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;

  429.                 } else {

  430.                     av_assert0(0);

  431.                 }

  432. 

  433.                 filter[i * filterSize + j] = coeff;

  434.                 xx++;

  435.             }

  436.             xDstInSrc += 2 * xInc;

  437.         }

  438.     }

  439. 

  440.     /* apply src & dst Filter to filter -> filter2

  441.      * av_free(filter);

  442.      */

  443.     av_assert0(filterSize > 0);

  444.     filter2Size = filterSize;

  445.     if (srcFilter)

  446.         filter2Size += srcFilter->length - 1;

  447.     if (dstFilter)

  448.         filter2Size += dstFilter->length - 1;

  449.     av_assert0(filter2Size > 0);

  450.     FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);

  451. 

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

  453.         int j, k;

  454. 

  455.         if (srcFilter) {

  456.             for (k = 0; k < srcFilter->length; k++) {

  457.                 for (j = 0; j < filterSize; j++)

  458.                     filter2[i * filter2Size + k + j] +=

  459.                         srcFilter->coeff[k] * filter[i * filterSize + j];

  460.             }

  461.         } else {

  462.             for (j = 0; j < filterSize; j++)

  463.                 filter2[i * filter2Size + j] = filter[i * filterSize + j];

  464.         }

  465.         // FIXME dstFilter

  466. 

  467.         (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;

  468.     }

  469.     av_freep(&filter);

  470. 

  471.     /* try to reduce the filter-size (step1 find size and shift left) */

  472.     // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).

  473.     minFilterSize = 0;

  474.     for (i = dstW - 1; i >= 0; i--) {

  475.         int min = filter2Size;

  476.         int j;

  477.         int64_t cutOff = 0.0;

  478. 

  479.         /* get rid of near zero elements on the left by shifting left */

  480.         for (j = 0; j < filter2Size; j++) {

  481.             int k;

  482.             cutOff += FFABS(filter2[i * filter2Size]);

  483. 

  484.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  485.                 break;

  486. 

  487.             /* preserve monotonicity because the core can't handle the

  488.              * filter otherwise */

  489.             if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])

  490.                 break;

  491. 

  492.             // move filter coefficients left

  493.             for (k = 1; k < filter2Size; k++)

  494.                 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];

  495.             filter2[i * filter2Size + k - 1] = 0;

  496.             (*filterPos)[i]++;

  497.         }

  498. 

  499.         cutOff = 0;

  500.         /* count near zeros on the right */

  501.         for (j = filter2Size - 1; j > 0; j--) {

  502.             cutOff += FFABS(filter2[i * filter2Size + j]);

  503. 

  504.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  505.                 break;

  506.             min--;

  507.         }

  508. 

  509.         if (min > minFilterSize)

  510.             minFilterSize = min;

  511.     }

  512. 

  513.     if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {

  514.         // we can handle the special case 4, so we don't want to go the full 8

  515.         if (minFilterSize < 5)

  516.             filterAlign = 4;

  517. 

  518.         /* We really don't want to waste our time doing useless computation, so

  519.          * fall back on the scalar C code for very small filters.

  520.          * Vectorizing is worth it only if you have a decent-sized vector. */

  521.         if (minFilterSize < 3)

  522.             filterAlign = 1;

  523.     }

  524. 

  525.     if (INLINE_MMX(cpu_flags)) {

  526.         // special case for unscaled vertical filtering

  527.         if (minFilterSize == 1 && filterAlign == 2)

  528.             filterAlign = 1;

  529.     }

  530. 

  531.     av_assert0(minFilterSize > 0);

  532.     filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));

  533.     av_assert0(filterSize > 0);

  534.     filter = av_malloc(filterSize * dstW * sizeof(*filter));

  535.     if (filterSize >= MAX_FILTER_SIZE * 16 /

  536.                       ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter) {

  537.         av_log(NULL, AV_LOG_ERROR, "sws: filterSize %d is too large, try less extreem scaling or increase MAX_FILTER_SIZE and recompile\n", filterSize);

  538.         goto fail;

  539.     }

  540.     *outFilterSize = filterSize;

  541. 

  542.     if (flags & SWS_PRINT_INFO)

  543.         av_log(NULL, AV_LOG_VERBOSE,

  544.                "SwScaler: reducing / aligning filtersize %d -> %d\n",

  545.                filter2Size, filterSize);

  546.     /* try to reduce the filter-size (step2 reduce it) */

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

  548.         int j;

  549. 

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

  551.             if (j >= filter2Size)

  552.                 filter[i * filterSize + j] = 0;

  553.             else

  554.                 filter[i * filterSize + j] = filter2[i * filter2Size + j];

  555.             if ((flags & SWS_BITEXACT) && j >= minFilterSize)

  556.                 filter[i * filterSize + j] = 0;

  557.         }

  558.     }

  559. 

  560.     // FIXME try to align filterPos if possible

  561. 

  562.     // fix borders

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

  564.         int j;

  565.         if ((*filterPos)[i] < 0) {

  566.             // move filter coefficients left to compensate for filterPos

  567.             for (j = 1; j < filterSize; j++) {

  568.                 int left = FFMAX(j + (*filterPos)[i], 0);

  569.                 filter[i * filterSize + left] += filter[i * filterSize + j];

  570.                 filter[i * filterSize + j]     = 0;

  571.             }

  572.             (*filterPos)[i]= 0;

  573.         }

  574. 

  575.         if ((*filterPos)[i] + filterSize > srcW) {

  576.             int shift = (*filterPos)[i] + filterSize - srcW;

  577.             // move filter coefficients right to compensate for filterPos

  578.             for (j = filterSize - 2; j >= 0; j--) {

  579.                 int right = FFMIN(j + shift, filterSize - 1);

  580.                 filter[i * filterSize + right] += filter[i * filterSize + j];

  581.                 filter[i * filterSize + j]      = 0;

  582.             }

  583.             (*filterPos)[i]= srcW - filterSize;

  584.         }

  585.     }

  586. 

  587.     // Note the +1 is for the MMX scaler which reads over the end

  588.     /* align at 16 for AltiVec (needed by hScale_altivec_real) */

  589.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

  590.                       *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);

  591. 

  592.     /* normalize & store in outFilter */

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

  594.         int j;

  595.         int64_t error = 0;

  596.         int64_t sum   = 0;

  597. 

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

  599.             sum += filter[i * filterSize + j];

  600.         }

  601.         sum = (sum + one / 2) / one;

  602.         for (j = 0; j < *outFilterSize; j++) {

  603.             int64_t v = filter[i * filterSize + j] + error;

  604.             int intV  = ROUNDED_DIV(v, sum);

  605.             (*outFilter)[i * (*outFilterSize) + j] = intV;

  606.             error                                  = v - intV * sum;

  607.         }

  608.     }

  609. 

  610.     (*filterPos)[dstW + 0] =

  611.     (*filterPos)[dstW + 1] =

  612.     (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will

  613.                                                       * read over the end */

  614.     for (i = 0; i < *outFilterSize; i++) {

  615.         int k = (dstW - 1) * (*outFilterSize) + i;

  616.         (*outFilter)[k + 1 * (*outFilterSize)] =

  617.         (*outFilter)[k + 2 * (*outFilterSize)] =

  618.         (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];

  619.     }

  620. 

  621.     ret = 0;

  622. 

  623. fail:

  624.     if(ret < 0)

  625.         av_log(NULL, AV_LOG_ERROR, "sws: initFilter failed\n");

  626.     av_free(filter);

  627.     av_free(filter2);

  628.     return ret;

  629. }

  630. 

  631. #if HAVE_MMXEXT_INLINE

  632. static int init_hscaler_mmxext(int dstW, int xInc, uint8_t *filterCode,

  633.                                int16_t *filter, int32_t *filterPos,

  634.                                int numSplits)

  635. {

  636.     uint8_t *fragmentA;

  637.     x86_reg imm8OfPShufW1A;

  638.     x86_reg imm8OfPShufW2A;

  639.     x86_reg fragmentLengthA;

  640.     uint8_t *fragmentB;

  641.     x86_reg imm8OfPShufW1B;

  642.     x86_reg imm8OfPShufW2B;

  643.     x86_reg fragmentLengthB;

  644.     int fragmentPos;

  645. 

  646.     int xpos, i;

  647. 

  648.     // create an optimized horizontal scaling routine

  649.     /* This scaler is made of runtime-generated MMXEXT code using specially tuned

  650.      * pshufw instructions. For every four output pixels, if four input pixels

  651.      * are enough for the fast bilinear scaling, then a chunk of fragmentB is

  652.      * used. If five input pixels are needed, then a chunk of fragmentA is used.

  653.      */

  654. 

  655.     // code fragment

  656. 

  657.     __asm__ volatile (

  658.         "jmp                         9f                 \n\t"

  659.         // Begin

  660.         "0:                                             \n\t"

  661.         "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"

  662.         "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"

  663.         "movd   1(%%"REG_c", %%"REG_S"), %%mm1          \n\t"

  664.         "punpcklbw                %%mm7, %%mm1          \n\t"

  665.         "punpcklbw                %%mm7, %%mm0          \n\t"

  666.         "pshufw                   $0xFF, %%mm1, %%mm1   \n\t"

  667.         "1:                                             \n\t"

  668.         "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"

  669.         "2:                                             \n\t"

  670.         "psubw                    %%mm1, %%mm0          \n\t"

  671.         "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"

  672.         "pmullw                   %%mm3, %%mm0          \n\t"

  673.         "psllw                       $7, %%mm1          \n\t"

  674.         "paddw                    %%mm1, %%mm0          \n\t"

  675. 

  676.         "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"

  677. 

  678.         "add                         $8, %%"REG_a"      \n\t"

  679.         // End

  680.         "9:                                             \n\t"

  681.         // "int $3                                         \n\t"

  682.         "lea       " LOCAL_MANGLE(0b) ", %0             \n\t"

  683.         "lea       " LOCAL_MANGLE(1b) ", %1             \n\t"

  684.         "lea       " LOCAL_MANGLE(2b) ", %2             \n\t"

  685.         "dec                         %1                 \n\t"

  686.         "dec                         %2                 \n\t"

  687.         "sub                         %0, %1             \n\t"

  688.         "sub                         %0, %2             \n\t"

  689.         "lea       " LOCAL_MANGLE(9b) ", %3             \n\t"

  690.         "sub                         %0, %3             \n\t"

  691. 

  692. 

  693.         : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),

  694.           "=r" (fragmentLengthA)

  695.         );

  696. 

  697.     __asm__ volatile (

  698.         "jmp                         9f                 \n\t"

  699.         // Begin

  700.         "0:                                             \n\t"

  701.         "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"

  702.         "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"

  703.         "punpcklbw                %%mm7, %%mm0          \n\t"

  704.         "pshufw                   $0xFF, %%mm0, %%mm1   \n\t"

  705.         "1:                                             \n\t"

  706.         "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"

  707.         "2:                                             \n\t"

  708.         "psubw                    %%mm1, %%mm0          \n\t"

  709.         "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"

  710.         "pmullw                   %%mm3, %%mm0          \n\t"

  711.         "psllw                       $7, %%mm1          \n\t"

  712.         "paddw                    %%mm1, %%mm0          \n\t"

  713. 

  714.         "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"

  715. 

  716.         "add                         $8, %%"REG_a"      \n\t"

  717.         // End

  718.         "9:                                             \n\t"

  719.         // "int                       $3                   \n\t"

  720.         "lea       " LOCAL_MANGLE(0b) ", %0             \n\t"

  721.         "lea       " LOCAL_MANGLE(1b) ", %1             \n\t"

  722.         "lea       " LOCAL_MANGLE(2b) ", %2             \n\t"

  723.         "dec                         %1                 \n\t"

  724.         "dec                         %2                 \n\t"

  725.         "sub                         %0, %1             \n\t"

  726.         "sub                         %0, %2             \n\t"

  727.         "lea       " LOCAL_MANGLE(9b) ", %3             \n\t"

  728.         "sub                         %0, %3             \n\t"

  729. 

  730. 

  731.         : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),

  732.           "=r" (fragmentLengthB)

  733.         );

  734. 

  735.     xpos        = 0; // lumXInc/2 - 0x8000; // difference between pixel centers

  736.     fragmentPos = 0;

  737. 

  738.     for (i = 0; i < dstW / numSplits; i++) {

  739.         int xx = xpos >> 16;

  740. 

  741.         if ((i & 3) == 0) {

  742.             int a                  = 0;

  743.             int b                  = ((xpos + xInc) >> 16) - xx;

  744.             int c                  = ((xpos + xInc * 2) >> 16) - xx;

  745.             int d                  = ((xpos + xInc * 3) >> 16) - xx;

  746.             int inc                = (d + 1 < 4);

  747.             uint8_t *fragment      = (d + 1 < 4) ? fragmentB : fragmentA;

  748.             x86_reg imm8OfPShufW1  = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;

  749.             x86_reg imm8OfPShufW2  = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;

  750.             x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;

  751.             int maxShift           = 3 - (d + inc);

  752.             int shift              = 0;

  753. 

  754.             if (filterCode) {

  755.                 filter[i]        = ((xpos              & 0xFFFF) ^ 0xFFFF) >> 9;

  756.                 filter[i + 1]    = (((xpos + xInc)     & 0xFFFF) ^ 0xFFFF) >> 9;

  757.                 filter[i + 2]    = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;

  758.                 filter[i + 3]    = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;

  759.                 filterPos[i / 2] = xx;

  760. 

  761.                 memcpy(filterCode + fragmentPos, fragment, fragmentLength);

  762. 

  763.                 filterCode[fragmentPos + imm8OfPShufW1] =  (a + inc)       |

  764.                                                           ((b + inc) << 2) |

  765.                                                           ((c + inc) << 4) |

  766.                                                           ((d + inc) << 6);

  767.                 filterCode[fragmentPos + imm8OfPShufW2] =  a | (b << 2) |

  768.                                                                (c << 4) |

  769.                                                                (d << 6);

  770. 

  771.                 if (i + 4 - inc >= dstW)

  772.                     shift = maxShift;               // avoid overread

  773.                 else if ((filterPos[i / 2] & 3) <= maxShift)

  774.                     shift = filterPos[i / 2] & 3;   // align

  775. 

  776.                 if (shift && i >= shift) {

  777.                     filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;

  778.                     filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;

  779.                     filterPos[i / 2]                        -= shift;

  780.                 }

  781.             }

  782. 

  783.             fragmentPos += fragmentLength;

  784. 

  785.             if (filterCode)

  786.                 filterCode[fragmentPos] = RET;

  787.         }

  788.         xpos += xInc;

  789.     }

  790.     if (filterCode)

  791.         filterPos[((i / 2) + 1) & (~1)] = xpos >> 16;  // needed to jump to the next part

  792. 

  793.     return fragmentPos + 1;

  794. }

  795. #endif /* HAVE_MMXEXT_INLINE */

  796. 

  797. static void getSubSampleFactors(int *h, int *v, enum AVPixelFormat format)

  798. {

  799.     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);

  800.     *h = desc->log2_chroma_w;

  801.     *v = desc->log2_chroma_h;

  802. }

  803. 

  804. static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)

  805. {

  806.     int64_t W, V, Z, Cy, Cu, Cv;

  807.     int64_t vr =  table[0];

  808.     int64_t ub =  table[1];

  809.     int64_t ug = -table[2];

  810.     int64_t vg = -table[3];

  811.     int64_t ONE = 65536;

  812.     int64_t cy = ONE;

  813.     uint8_t *p = (uint8_t*)c->input_rgb2yuv_table;

  814.     int i;

  815.     static const int8_t map[] = {

  816.     BY_IDX, GY_IDX, -1    , BY_IDX, BY_IDX, GY_IDX, -1    , BY_IDX,

  817.     RY_IDX, -1    , GY_IDX, RY_IDX, RY_IDX, -1    , GY_IDX, RY_IDX,

  818.     RY_IDX, GY_IDX, -1    , RY_IDX, RY_IDX, GY_IDX, -1    , RY_IDX,

  819.     BY_IDX, -1    , GY_IDX, BY_IDX, BY_IDX, -1    , GY_IDX, BY_IDX,

  820.     BU_IDX, GU_IDX, -1    , BU_IDX, BU_IDX, GU_IDX, -1    , BU_IDX,

  821.     RU_IDX, -1    , GU_IDX, RU_IDX, RU_IDX, -1    , GU_IDX, RU_IDX,

  822.     RU_IDX, GU_IDX, -1    , RU_IDX, RU_IDX, GU_IDX, -1    , RU_IDX,

  823.     BU_IDX, -1    , GU_IDX, BU_IDX, BU_IDX, -1    , GU_IDX, BU_IDX,

  824.     BV_IDX, GV_IDX, -1    , BV_IDX, BV_IDX, GV_IDX, -1    , BV_IDX,

  825.     RV_IDX, -1    , GV_IDX, RV_IDX, RV_IDX, -1    , GV_IDX, RV_IDX,

  826.     RV_IDX, GV_IDX, -1    , RV_IDX, RV_IDX, GV_IDX, -1    , RV_IDX,

  827.     BV_IDX, -1    , GV_IDX, BV_IDX, BV_IDX, -1    , GV_IDX, BV_IDX,

  828.     RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,

  829.     BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,

  830.     GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    ,

  831.     -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX,

  832.     RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,

  833.     BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,

  834.     GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    ,

  835.     -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX,

  836.     RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,

  837.     BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,

  838.     GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    ,

  839.     -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, //23

  840.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //24

  841.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //25

  842.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //26

  843.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //27

  844.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //28

  845.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //29

  846.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //30

  847.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //31

  848.     BY_IDX, GY_IDX, RY_IDX, -1    , -1    , -1    , -1    , -1    , //32

  849.     BU_IDX, GU_IDX, RU_IDX, -1    , -1    , -1    , -1    , -1    , //33

  850.     BV_IDX, GV_IDX, RV_IDX, -1    , -1    , -1    , -1    , -1    , //34

  851.     };

  852. 

  853.     dstRange = 0; //FIXME range = 1 is handled elsewhere

  854. 

  855.     if (!dstRange) {

  856.         cy = cy * 255 / 219;

  857.     } else {

  858.         vr = vr * 224 / 255;

  859.         ub = ub * 224 / 255;

  860.         ug = ug * 224 / 255;

  861.         vg = vg * 224 / 255;

  862.     }

  863.     W = ROUNDED_DIV(ONE*ONE*ug, ub);

  864.     V = ROUNDED_DIV(ONE*ONE*vg, vr);

  865.     Z = ONE*ONE-W-V;

  866. 

  867.     Cy = ROUNDED_DIV(cy*Z, ONE);

  868.     Cu = ROUNDED_DIV(ub*Z, ONE);

  869.     Cv = ROUNDED_DIV(vr*Z, ONE);

  870. 

  871.     c->input_rgb2yuv_table[RY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V        , Cy);

  872.     c->input_rgb2yuv_table[GY_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE  , Cy);

  873.     c->input_rgb2yuv_table[BY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W        , Cy);

  874. 

  875.     c->input_rgb2yuv_table[RU_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V        , Cu);

  876.     c->input_rgb2yuv_table[GU_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE  , Cu);

  877.     c->input_rgb2yuv_table[BU_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(Z+W)    , Cu);

  878. 

  879.     c->input_rgb2yuv_table[RV_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(V+Z)    , Cv);

  880.     c->input_rgb2yuv_table[GV_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE  , Cv);

  881.     c->input_rgb2yuv_table[BV_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W        , Cv);

  882. 

  883.     if(/*!dstRange && */table == ff_yuv2rgb_coeffs[SWS_CS_DEFAULT]) {

  884.         c->input_rgb2yuv_table[BY_IDX] =  ((int)(0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  885.         c->input_rgb2yuv_table[BV_IDX] = (-(int)(0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  886.         c->input_rgb2yuv_table[BU_IDX] =  ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  887.         c->input_rgb2yuv_table[GY_IDX] =  ((int)(0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  888.         c->input_rgb2yuv_table[GV_IDX] = (-(int)(0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  889.         c->input_rgb2yuv_table[GU_IDX] = (-(int)(0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  890.         c->input_rgb2yuv_table[RY_IDX] =  ((int)(0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  891.         c->input_rgb2yuv_table[RV_IDX] =  ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  892.         c->input_rgb2yuv_table[RU_IDX] = (-(int)(0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));

  893.     }

  894.     for(i=0; i<FF_ARRAY_ELEMS(map); i++)

  895.         AV_WL16(p + 16*4 + 2*i, map[i] >= 0 ? c->input_rgb2yuv_table[map[i]] : 0);

  896. }

  897. 

  898. static void fill_xyztables(struct SwsContext *c)

  899. {

  900.     int i;

  901.     double xyzgamma = XYZ_GAMMA;

  902.     double rgbgamma = 1.0 / RGB_GAMMA;

  903.     static const int16_t xyz2rgb_matrix[3][4] = {

  904.         {13270, -6295, -2041},

  905.         {-3969,  7682,   170},

  906.         {  228,  -835,  4329} };

  907. 

  908.     /* set gamma vectors */

  909.     for (i = 0; i < 4096; i++) {

  910.         c->xyzgamma[i] = lrint(pow(i / 4095.0, xyzgamma) * 4095.0);

  911.         c->rgbgamma[i] = lrint(pow(i / 4095.0, rgbgamma) * 4095.0);

  912.     }

  913.     memcpy(c->xyz2rgb_matrix, xyz2rgb_matrix, sizeof(c->xyz2rgb_matrix));

  914. }

  915. 

  916. int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],

  917.                              int srcRange, const int table[4], int dstRange,

  918.                              int brightness, int contrast, int saturation)

  919. {

  920.     const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(c->dstFormat);

  921.     const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(c->srcFormat);

  922.     memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);

  923.     memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);

  924. 

  925.     if(!isYUV(c->dstFormat) && !isGray(c->dstFormat))

  926.         dstRange = 0;

  927.     if(!isYUV(c->srcFormat) && !isGray(c->srcFormat))

  928.         srcRange = 0;

  929. 

  930.     c->brightness = brightness;

  931.     c->contrast   = contrast;

  932.     c->saturation = saturation;

  933.     c->srcRange   = srcRange;

  934.     c->dstRange   = dstRange;

  935. 

  936.     fill_xyztables(c);

  937. 

  938.     if ((isYUV(c->dstFormat) || isGray(c->dstFormat)) && (isYUV(c->srcFormat) || isGray(c->srcFormat)))

  939.         return -1;

  940. 

  941.     c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);

  942.     c->srcFormatBpp = av_get_bits_per_pixel(desc_src);

  943. 

  944.     if (!isYUV(c->dstFormat) && !isGray(c->dstFormat)) {

  945.     ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,

  946.                              contrast, saturation);

  947.     // FIXME factorize

  948. 

  949.     if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)

  950.         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,

  951.                                        contrast, saturation);

  952.     }

  953. 

  954.     fill_rgb2yuv_table(c, table, dstRange);

  955. 

  956.     return 0;

  957. }

  958. 

  959. int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,

  960.                              int *srcRange, int **table, int *dstRange,

  961.                              int *brightness, int *contrast, int *saturation)

  962. {

  963.     if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat))

  964.         return -1;

  965. 

  966.     *inv_table  = c->srcColorspaceTable;

  967.     *table      = c->dstColorspaceTable;

  968.     *srcRange   = c->srcRange;

  969.     *dstRange   = c->dstRange;

  970.     *brightness = c->brightness;

  971.     *contrast   = c->contrast;

  972.     *saturation = c->saturation;

  973. 

  974.     return 0;

  975. }

  976. 

  977. static int handle_jpeg(enum AVPixelFormat *format)

  978. {

  979.     switch (*format) {

  980.     case AV_PIX_FMT_YUVJ420P:

  981.         *format = AV_PIX_FMT_YUV420P;

  982.         return 1;

  983.     case AV_PIX_FMT_YUVJ422P:

  984.         *format = AV_PIX_FMT_YUV422P;

  985.         return 1;

  986.     case AV_PIX_FMT_YUVJ444P:

  987.         *format = AV_PIX_FMT_YUV444P;

  988.         return 1;

  989.     case AV_PIX_FMT_YUVJ440P:

  990.         *format = AV_PIX_FMT_YUV440P;

  991.         return 1;

  992.     default:

  993.         return 0;

  994.     }

  995. }

  996. 

  997. static int handle_0alpha(enum AVPixelFormat *format)

  998. {

  999.     switch (*format) {

  1000.     case AV_PIX_FMT_0BGR    : *format = AV_PIX_FMT_ABGR   ; return 1;

  1001.     case AV_PIX_FMT_BGR0    : *format = AV_PIX_FMT_BGRA   ; return 4;

  1002.     case AV_PIX_FMT_0RGB    : *format = AV_PIX_FMT_ARGB   ; return 1;

  1003.     case AV_PIX_FMT_RGB0    : *format = AV_PIX_FMT_RGBA   ; return 4;

  1004.     default:                                          return 0;

  1005.     }

  1006. }

  1007. 

  1008. static int handle_xyz(enum AVPixelFormat *format)

  1009. {

  1010.     switch (*format) {

  1011.     case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;

  1012.     case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;

  1013.     default:                                                return 0;

  1014.     }

  1015. }

  1016. 

  1017. SwsContext *sws_alloc_context(void)

  1018. {

  1019.     SwsContext *c = av_mallocz(sizeof(SwsContext));

  1020. 

  1021.     if (c) {

  1022.         c->av_class = &sws_context_class;

  1023.         av_opt_set_defaults(c);

  1024.     }

  1025. 

  1026.     return c;

  1027. }

  1028. 

  1029. av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,

  1030.                              SwsFilter *dstFilter)

  1031. {

  1032.     int i, j;

  1033.     int usesVFilter, usesHFilter;

  1034.     int unscaled;

  1035.     SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };

  1036.     int srcW              = c->srcW;

  1037.     int srcH              = c->srcH;

  1038.     int dstW              = c->dstW;

  1039.     int dstH              = c->dstH;

  1040.     int dst_stride        = FFALIGN(dstW * sizeof(int16_t) + 66, 16);

  1041.     int flags, cpu_flags;

  1042.     enum AVPixelFormat srcFormat = c->srcFormat;

  1043.     enum AVPixelFormat dstFormat = c->dstFormat;

  1044.     const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(srcFormat);

  1045.     const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(dstFormat);

  1046. 

  1047.     cpu_flags = av_get_cpu_flags();

  1048.     flags     = c->flags;

  1049.     emms_c();

  1050.     if (!rgb15to16)

  1051.         sws_rgb2rgb_init();

  1052. 

  1053.     unscaled = (srcW == dstW && srcH == dstH);

  1054. 

  1055.     handle_jpeg(&srcFormat);

  1056.     handle_jpeg(&dstFormat);

  1057.     handle_0alpha(&srcFormat);

  1058.     handle_0alpha(&dstFormat);

  1059.     handle_xyz(&srcFormat);

  1060.     handle_xyz(&dstFormat);

  1061. 

  1062.     if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){

  1063.         av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");

  1064.         c->srcFormat= srcFormat;

  1065.         c->dstFormat= dstFormat;

  1066.     }

  1067. 

  1068.     if (!sws_isSupportedInput(srcFormat)) {

  1069.         av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",

  1070.                av_get_pix_fmt_name(srcFormat));

  1071.         return AVERROR(EINVAL);

  1072.     }

  1073.     if (!sws_isSupportedOutput(dstFormat)) {

  1074.         av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",

  1075.                av_get_pix_fmt_name(dstFormat));

  1076.         return AVERROR(EINVAL);

  1077.     }

  1078. 

  1079.     i = flags & (SWS_POINT         |

  1080.                  SWS_AREA          |

  1081.                  SWS_BILINEAR      |

  1082.                  SWS_FAST_BILINEAR |

  1083.                  SWS_BICUBIC       |

  1084.                  SWS_X             |

  1085.                  SWS_GAUSS         |

  1086.                  SWS_LANCZOS       |

  1087.                  SWS_SINC          |

  1088.                  SWS_SPLINE        |

  1089.                  SWS_BICUBLIN);

  1090.     if (!i || (i & (i - 1))) {

  1091.         av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);

  1092.         return AVERROR(EINVAL);

  1093.     }

  1094.     /* sanity check */

  1095.     if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {

  1096.         /* FIXME check if these are enough and try to lower them after

  1097.          * fixing the relevant parts of the code */

  1098.         av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",

  1099.                srcW, srcH, dstW, dstH);

  1100.         return AVERROR(EINVAL);

  1101.     }

  1102. 

  1103.     if (!dstFilter)

  1104.         dstFilter = &dummyFilter;

  1105.     if (!srcFilter)

  1106.         srcFilter = &dummyFilter;

  1107. 

  1108.     c->lumXInc      = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;

  1109.     c->lumYInc      = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;

  1110.     c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);

  1111.     c->srcFormatBpp = av_get_bits_per_pixel(desc_src);

  1112.     c->vRounder     = 4 * 0x0001000100010001ULL;

  1113. 

  1114.     usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||

  1115.                   (srcFilter->chrV && srcFilter->chrV->length > 1) ||

  1116.                   (dstFilter->lumV && dstFilter->lumV->length > 1) ||

  1117.                   (dstFilter->chrV && dstFilter->chrV->length > 1);

  1118.     usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||

  1119.                   (srcFilter->chrH && srcFilter->chrH->length > 1) ||

  1120.                   (dstFilter->lumH && dstFilter->lumH->length > 1) ||

  1121.                   (dstFilter->chrH && dstFilter->chrH->length > 1);

  1122. 

  1123.     getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);

  1124.     getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);

  1125. 

  1126.     if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {

  1127.         if (dstW&1) {

  1128.             av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");

  1129.             flags |= SWS_FULL_CHR_H_INT;

  1130.             c->flags = flags;

  1131.         }

  1132.     }

  1133. 

  1134.     if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||

  1135.        dstFormat == AV_PIX_FMT_RGB4_BYTE ||

  1136.        dstFormat == AV_PIX_FMT_BGR8 ||

  1137.        dstFormat == AV_PIX_FMT_RGB8) {

  1138.         if (flags & SWS_ERROR_DIFFUSION && !(flags & SWS_FULL_CHR_H_INT)) {

  1139.             av_log(c, AV_LOG_DEBUG,

  1140.                 "Error diffusion dither is only supported in full chroma interpolation for destination format '%s'\n",

  1141.                 av_get_pix_fmt_name(dstFormat));

  1142.             flags   |= SWS_FULL_CHR_H_INT;

  1143.             c->flags = flags;

  1144.         }

  1145.         if (!(flags & SWS_ERROR_DIFFUSION) && (flags & SWS_FULL_CHR_H_INT)) {

  1146.             av_log(c, AV_LOG_DEBUG,

  1147.                 "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",

  1148.                 av_get_pix_fmt_name(dstFormat));

  1149.             flags   |= SWS_ERROR_DIFFUSION;

  1150.             c->flags = flags;

  1151.         }

  1152.     }

  1153.     if (isPlanarRGB(dstFormat)) {

  1154.         if (!(flags & SWS_FULL_CHR_H_INT)) {

  1155.             av_log(c, AV_LOG_DEBUG,

  1156.                    "%s output is not supported with half chroma resolution, switching to full\n",

  1157.                    av_get_pix_fmt_name(dstFormat));

  1158.             flags   |= SWS_FULL_CHR_H_INT;

  1159.             c->flags = flags;

  1160.         }

  1161.     }

  1162. 

  1163.     /* reuse chroma for 2 pixels RGB/BGR unless user wants full

  1164.      * chroma interpolation */

  1165.     if (flags & SWS_FULL_CHR_H_INT &&

  1166.         isAnyRGB(dstFormat)        &&

  1167.         !isPlanarRGB(dstFormat)    &&

  1168.         dstFormat != AV_PIX_FMT_RGBA  &&

  1169.         dstFormat != AV_PIX_FMT_ARGB  &&

  1170.         dstFormat != AV_PIX_FMT_BGRA  &&

  1171.         dstFormat != AV_PIX_FMT_ABGR  &&

  1172.         dstFormat != AV_PIX_FMT_RGB24 &&

  1173.         dstFormat != AV_PIX_FMT_BGR24 &&

  1174.         dstFormat != AV_PIX_FMT_BGR4_BYTE &&

  1175.         dstFormat != AV_PIX_FMT_RGB4_BYTE &&

  1176.         dstFormat != AV_PIX_FMT_BGR8 &&

  1177.         dstFormat != AV_PIX_FMT_RGB8

  1178.     ) {

  1179.         av_log(c, AV_LOG_WARNING,

  1180.                "full chroma interpolation for destination format '%s' not yet implemented\n",

  1181.                av_get_pix_fmt_name(dstFormat));

  1182.         flags   &= ~SWS_FULL_CHR_H_INT;

  1183.         c->flags = flags;

  1184.     }

  1185.     if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))

  1186.         c->chrDstHSubSample = 1;

  1187. 

  1188.     // drop some chroma lines if the user wants it

  1189.     c->vChrDrop          = (flags & SWS_SRC_V_CHR_DROP_MASK) >>

  1190.                            SWS_SRC_V_CHR_DROP_SHIFT;

  1191.     c->chrSrcVSubSample += c->vChrDrop;

  1192. 

  1193.     /* drop every other pixel for chroma calculation unless user

  1194.      * wants full chroma */

  1195.     if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP)   &&

  1196.         srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&

  1197.         srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&

  1198.         srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&

  1199.         srcFormat != AV_PIX_FMT_GBRP9BE   && srcFormat != AV_PIX_FMT_GBRP9LE  &&

  1200.         srcFormat != AV_PIX_FMT_GBRP10BE  && srcFormat != AV_PIX_FMT_GBRP10LE &&

  1201.         srcFormat != AV_PIX_FMT_GBRP12BE  && srcFormat != AV_PIX_FMT_GBRP12LE &&

  1202.         srcFormat != AV_PIX_FMT_GBRP14BE  && srcFormat != AV_PIX_FMT_GBRP14LE &&

  1203.         srcFormat != AV_PIX_FMT_GBRP16BE  && srcFormat != AV_PIX_FMT_GBRP16LE &&

  1204.         ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||

  1205.          (flags & SWS_FAST_BILINEAR)))

  1206.         c->chrSrcHSubSample = 1;

  1207. 

  1208.     // Note the -((-x)>>y) is so that we always round toward +inf.

  1209.     c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);

  1210.     c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);

  1211.     c->chrDstW = -((-dstW) >> c->chrDstHSubSample);

  1212.     c->chrDstH = -((-dstH) >> c->chrDstVSubSample);

  1213. 

  1214.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);

  1215. 

  1216.     /* unscaled special cases */

  1217.     if (unscaled && !usesHFilter && !usesVFilter &&

  1218.         (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {

  1219.         ff_get_unscaled_swscale(c);

  1220. 

  1221.         if (c->swScale) {

  1222.             if (flags & SWS_PRINT_INFO)

  1223.                 av_log(c, AV_LOG_INFO,

  1224.                        "using unscaled %s -> %s special converter\n",

  1225.                        av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));

  1226.             return 0;

  1227.         }

  1228.     }

  1229. 

  1230.     c->srcBpc = 1 + desc_src->comp[0].depth_minus1;

  1231.     if (c->srcBpc < 8)

  1232.         c->srcBpc = 8;

  1233.     c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;

  1234.     if (c->dstBpc < 8)

  1235.         c->dstBpc = 8;

  1236.     if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)

  1237.         c->srcBpc = 16;

  1238.     if (c->dstBpc == 16)

  1239.         dst_stride <<= 1;

  1240. 

  1241.     if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {

  1242.         c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&

  1243.                               (srcW & 15) == 0) ? 1 : 0;

  1244.         if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0

  1245. 

  1246.             && (flags & SWS_FAST_BILINEAR)) {

  1247.             if (flags & SWS_PRINT_INFO)

  1248.                 av_log(c, AV_LOG_INFO,

  1249.                        "output width is not a multiple of 32 -> no MMXEXT scaler\n");

  1250.         }

  1251.         if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))

  1252.             c->canMMXEXTBeUsed = 0;

  1253.     } else

  1254.         c->canMMXEXTBeUsed = 0;

  1255. 

  1256.     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;

  1257.     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

  1258. 

  1259.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  1260.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  1261.      * correct scaling.

  1262.      * n-2 is the last chrominance sample available.

  1263.      * This is not perfect, but no one should notice the difference, the more

  1264.      * correct variant would be like the vertical one, but that would require

  1265.      * some special code for the first and last pixel */

  1266.     if (flags & SWS_FAST_BILINEAR) {

  1267.         if (c->canMMXEXTBeUsed) {

  1268.             c->lumXInc += 20;

  1269.             c->chrXInc += 20;

  1270.         }

  1271.         // we don't use the x86 asm scaler if MMX is available

  1272.         else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {

  1273.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1274.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1275.         }

  1276.     }

  1277. 

  1278. #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)

  1279. 

  1280.     /* precalculate horizontal scaler filter coefficients */

  1281.     {

  1282. #if HAVE_MMXEXT_INLINE

  1283. // can't downscale !!!

  1284.         if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {

  1285.             c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,

  1286.                                                              NULL, NULL, 8);

  1287.             c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,

  1288.                                                              NULL, NULL, NULL, 4);

  1289. 

  1290. #if USE_MMAP

  1291.             c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,

  1292.                                           PROT_READ | PROT_WRITE,

  1293.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1294.                                           -1, 0);

  1295.             c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,

  1296.                                           PROT_READ | PROT_WRITE,

  1297.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1298.                                           -1, 0);

  1299. #elif HAVE_VIRTUALALLOC

  1300.             c->lumMmxextFilterCode = VirtualAlloc(NULL,

  1301.                                                   c->lumMmxextFilterCodeSize,

  1302.                                                   MEM_COMMIT,

  1303.                                                   PAGE_EXECUTE_READWRITE);

  1304.             c->chrMmxextFilterCode = VirtualAlloc(NULL,

  1305.                                                   c->chrMmxextFilterCodeSize,

  1306.                                                   MEM_COMMIT,

  1307.                                                   PAGE_EXECUTE_READWRITE);

  1308. #else

  1309.             c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);

  1310.             c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);

  1311. #endif

  1312. 

  1313. #ifdef MAP_ANONYMOUS

  1314.             if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)

  1315. #else

  1316.             if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)

  1317. #endif

  1318.             {

  1319.                 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");

  1320.                 return AVERROR(ENOMEM);

  1321.             }

  1322. 

  1323.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1324.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1325.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1326.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1327. 

  1328.             init_hscaler_mmxext(      dstW, c->lumXInc, c->lumMmxextFilterCode,

  1329.                                 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);

  1330.             init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,

  1331.                                 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);

  1332. 

  1333. #if USE_MMAP

  1334.             mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1335.             mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1336. #endif

  1337.         } else

  1338. #endif /* HAVE_MMXEXT_INLINE */

  1339.         {

  1340.             const int filterAlign =

  1341.                 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :

  1342.                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1343.                 1;

  1344. 

  1345.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1346.                            &c->hLumFilterSize, c->lumXInc,

  1347.                            srcW, dstW, filterAlign, 1 << 14,

  1348.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1349.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1350.                            c->param) < 0)

  1351.                 goto fail;

  1352.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1353.                            &c->hChrFilterSize, c->chrXInc,

  1354.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1355.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1356.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1357.                            c->param) < 0)

  1358.                 goto fail;

  1359.         }

  1360.     } // initialize horizontal stuff

  1361. 

  1362.     /* precalculate vertical scaler filter coefficients */

  1363.     {

  1364.         const int filterAlign =

  1365.             (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :

  1366.             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1367.             1;

  1368. 

  1369.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1370.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

  1371.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1372.                        cpu_flags, srcFilter->lumV, dstFilter->lumV,

  1373.                        c->param) < 0)

  1374.             goto fail;

  1375.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1376.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1377.                        filterAlign, (1 << 12),

  1378.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1379.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1380.                        c->param) < 0)

  1381.             goto fail;

  1382. 

  1383. #if HAVE_ALTIVEC

  1384.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1385.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1386. 

  1387.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1388.             int j;

  1389.             short *p = (short *)&c->vYCoeffsBank[i];

  1390.             for (j = 0; j < 8; j++)

  1391.                 p[j] = c->vLumFilter[i];

  1392.         }

  1393. 

  1394.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1395.             int j;

  1396.             short *p = (short *)&c->vCCoeffsBank[i];

  1397.             for (j = 0; j < 8; j++)

  1398.                 p[j] = c->vChrFilter[i];

  1399.         }

  1400. #endif

  1401.     }

  1402. 

  1403.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1404.     c->vLumBufSize = c->vLumFilterSize;

  1405.     c->vChrBufSize = c->vChrFilterSize;

  1406.     for (i = 0; i < dstH; i++) {

  1407.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1408.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1409.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1410.                                << c->chrSrcVSubSample));

  1411. 

  1412.         nextSlice >>= c->chrSrcVSubSample;

  1413.         nextSlice <<= c->chrSrcVSubSample;

  1414.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1415.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1416.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1417.             (nextSlice >> c->chrSrcVSubSample))

  1418.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1419.                              c->vChrFilterPos[chrI];

  1420.     }

  1421. 

  1422.     for (i = 0; i < 4; i++)

  1423.         FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);

  1424. 

  1425.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1426.      * need to allocate several megabytes to handle all possible cases) */

  1427.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1428.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1429.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1430.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1431.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1432.     /* Note we need at least one pixel more at the end because of the MMX code

  1433.      * (just in case someone wants to replace the 4000/8000). */

  1434.     /* align at 16 bytes for AltiVec */

  1435.     for (i = 0; i < c->vLumBufSize; i++) {

  1436.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1437.                           dst_stride + 16, fail);

  1438.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1439.     }

  1440.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1441.     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);

  1442.     c->uv_offx2 = dst_stride + 16;

  1443.     for (i = 0; i < c->vChrBufSize; i++) {

  1444.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1445.                          dst_stride * 2 + 32, fail);

  1446.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1447.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1448.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1449.     }

  1450.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1451.         for (i = 0; i < c->vLumBufSize; i++) {

  1452.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],

  1453.                               dst_stride + 16, fail);

  1454.             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];

  1455.         }

  1456. 

  1457.     // try to avoid drawing green stuff between the right end and the stride end

  1458.     for (i = 0; i < c->vChrBufSize; i++)

  1459.         if(desc_dst->comp[0].depth_minus1 == 15){

  1460.             av_assert0(c->dstBpc > 14);

  1461.             for(j=0; j<dst_stride/2+1; j++)

  1462.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1463.         } else

  1464.             for(j=0; j<dst_stride+1; j++)

  1465.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1466. 

  1467.     av_assert0(c->chrDstH <= dstH);

  1468. 

  1469.     if (flags & SWS_PRINT_INFO) {

  1470.         if (flags & SWS_FAST_BILINEAR)

  1471.             av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1472.         else if (flags & SWS_BILINEAR)

  1473.             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1474.         else if (flags & SWS_BICUBIC)

  1475.             av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1476.         else if (flags & SWS_X)

  1477.             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1478.         else if (flags & SWS_POINT)

  1479.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1480.         else if (flags & SWS_AREA)

  1481.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1482.         else if (flags & SWS_BICUBLIN)

  1483.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1484.         else if (flags & SWS_GAUSS)

  1485.             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1486.         else if (flags & SWS_SINC)

  1487.             av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1488.         else if (flags & SWS_LANCZOS)

  1489.             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1490.         else if (flags & SWS_SPLINE)

  1491.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1492.         else

  1493.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1494. 

  1495.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1496.                av_get_pix_fmt_name(srcFormat),

  1497. #ifdef DITHER1XBPP

  1498.                dstFormat == AV_PIX_FMT_BGR555   || dstFormat == AV_PIX_FMT_BGR565   ||

  1499.                dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||

  1500.                dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?

  1501.                                                              "dithered " : "",

  1502. #else

  1503.                "",

  1504. #endif

  1505.                av_get_pix_fmt_name(dstFormat));

  1506. 

  1507.         if (INLINE_MMXEXT(cpu_flags))

  1508.             av_log(c, AV_LOG_INFO, "using MMXEXT\n");

  1509.         else if (INLINE_AMD3DNOW(cpu_flags))

  1510.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1511.         else if (INLINE_MMX(cpu_flags))

  1512.             av_log(c, AV_LOG_INFO, "using MMX\n");

  1513.         else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)

  1514.             av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1515.         else

  1516.             av_log(c, AV_LOG_INFO, "using C\n");

  1517. 

  1518.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1519.         av_log(c, AV_LOG_DEBUG,

  1520.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1521.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1522.         av_log(c, AV_LOG_DEBUG,

  1523.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1524.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1525.                c->chrXInc, c->chrYInc);

  1526.     }

  1527. 

  1528.     c->swScale = ff_getSwsFunc(c);

  1529.     return 0;

  1530. fail: // FIXME replace things by appropriate error codes

  1531.     return -1;

  1532. }

  1533. 

  1534. #if FF_API_SWS_GETCONTEXT

  1535. SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,

  1536.                            int dstW, int dstH, enum AVPixelFormat dstFormat,

  1537.                            int flags, SwsFilter *srcFilter,

  1538.                            SwsFilter *dstFilter, const double *param)

  1539. {

  1540.     SwsContext *c;

  1541. 

  1542.     if (!(c = sws_alloc_context()))

  1543.         return NULL;

  1544. 

  1545.     c->flags     = flags;

  1546.     c->srcW      = srcW;

  1547.     c->srcH      = srcH;

  1548.     c->dstW      = dstW;

  1549.     c->dstH      = dstH;

  1550.     c->srcRange  = handle_jpeg(&srcFormat);

  1551.     c->dstRange  = handle_jpeg(&dstFormat);

  1552.     c->src0Alpha = handle_0alpha(&srcFormat);

  1553.     c->dst0Alpha = handle_0alpha(&dstFormat);

  1554.     c->srcXYZ    = handle_xyz(&srcFormat);

  1555.     c->dstXYZ    = handle_xyz(&dstFormat);

  1556.     c->srcFormat = srcFormat;

  1557.     c->dstFormat = dstFormat;

  1558. 

  1559.     if (param) {

  1560.         c->param[0] = param[0];

  1561.         c->param[1] = param[1];

  1562.     }

  1563.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1564.                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1565.                              c->dstRange, 0, 1 << 16, 1 << 16);

  1566. 

  1567.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1568.         sws_freeContext(c);

  1569.         return NULL;

  1570.     }

  1571. 

  1572.     return c;

  1573. }

  1574. #endif

  1575. 

  1576. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1577.                                 float lumaSharpen, float chromaSharpen,

  1578.                                 float chromaHShift, float chromaVShift,

  1579.                                 int verbose)

  1580. {

  1581.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1582.     if (!filter)

  1583.         return NULL;

  1584. 

  1585.     if (lumaGBlur != 0.0) {

  1586.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1587.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1588.     } else {

  1589.         filter->lumH = sws_getIdentityVec();

  1590.         filter->lumV = sws_getIdentityVec();

  1591.     }

  1592. 

  1593.     if (chromaGBlur != 0.0) {

  1594.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1595.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1596.     } else {

  1597.         filter->chrH = sws_getIdentityVec();

  1598.         filter->chrV = sws_getIdentityVec();

  1599.     }

  1600. 

  1601.     if (chromaSharpen != 0.0) {

  1602.         SwsVector *id = sws_getIdentityVec();

  1603.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1604.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1605.         sws_addVec(filter->chrH, id);

  1606.         sws_addVec(filter->chrV, id);

  1607.         sws_freeVec(id);

  1608.     }

  1609. 

  1610.     if (lumaSharpen != 0.0) {

  1611.         SwsVector *id = sws_getIdentityVec();

  1612.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1613.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1614.         sws_addVec(filter->lumH, id);

  1615.         sws_addVec(filter->lumV, id);

  1616.         sws_freeVec(id);

  1617.     }

  1618. 

  1619.     if (chromaHShift != 0.0)

  1620.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1621. 

  1622.     if (chromaVShift != 0.0)

  1623.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1624. 

  1625.     sws_normalizeVec(filter->chrH, 1.0);

  1626.     sws_normalizeVec(filter->chrV, 1.0);

  1627.     sws_normalizeVec(filter->lumH, 1.0);

  1628.     sws_normalizeVec(filter->lumV, 1.0);

  1629. 

  1630.     if (verbose)

  1631.         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1632.     if (verbose)

  1633.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1634. 

  1635.     return filter;

  1636. }

  1637. 

  1638. SwsVector *sws_allocVec(int length)

  1639. {

  1640.     SwsVector *vec;

  1641. 

  1642.     if(length <= 0 || length > INT_MAX/ sizeof(double))

  1643.         return NULL;

  1644. 

  1645.     vec = av_malloc(sizeof(SwsVector));

  1646.     if (!vec)

  1647.         return NULL;

  1648.     vec->length = length;

  1649.     vec->coeff  = av_malloc(sizeof(double) * length);

  1650.     if (!vec->coeff)

  1651.         av_freep(&vec);

  1652.     return vec;

  1653. }

  1654. 

  1655. SwsVector *sws_getGaussianVec(double variance, double quality)

  1656. {

  1657.     const int length = (int)(variance * quality + 0.5) | 1;

  1658.     int i;

  1659.     double middle  = (length - 1) * 0.5;

  1660.     SwsVector *vec;

  1661. 

  1662.     if(variance < 0 || quality < 0)

  1663.         return NULL;

  1664. 

  1665.     vec = sws_allocVec(length);

  1666. 

  1667.     if (!vec)

  1668.         return NULL;

  1669. 

  1670.     for (i = 0; i < length; i++) {

  1671.         double dist = i - middle;

  1672.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1673.                         sqrt(2 * variance * M_PI);

  1674.     }

  1675. 

  1676.     sws_normalizeVec(vec, 1.0);

  1677. 

  1678.     return vec;

  1679. }

  1680. 

  1681. SwsVector *sws_getConstVec(double c, int length)

  1682. {

  1683.     int i;

  1684.     SwsVector *vec = sws_allocVec(length);

  1685. 

  1686.     if (!vec)

  1687.         return NULL;

  1688. 

  1689.     for (i = 0; i < length; i++)

  1690.         vec->coeff[i] = c;

  1691. 

  1692.     return vec;

  1693. }

  1694. 

  1695. SwsVector *sws_getIdentityVec(void)

  1696. {

  1697.     return sws_getConstVec(1.0, 1);

  1698. }

  1699. 

  1700. static double sws_dcVec(SwsVector *a)

  1701. {

  1702.     int i;

  1703.     double sum = 0;

  1704. 

  1705.     for (i = 0; i < a->length; i++)

  1706.         sum += a->coeff[i];

  1707. 

  1708.     return sum;

  1709. }

  1710. 

  1711. void sws_scaleVec(SwsVector *a, double scalar)

  1712. {

  1713.     int i;

  1714. 

  1715.     for (i = 0; i < a->length; i++)

  1716.         a->coeff[i] *= scalar;

  1717. }

  1718. 

  1719. void sws_normalizeVec(SwsVector *a, double height)

  1720. {

  1721.     sws_scaleVec(a, height / sws_dcVec(a));

  1722. }

  1723. 

  1724. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1725. {

  1726.     int length = a->length + b->length - 1;

  1727.     int i, j;

  1728.     SwsVector *vec = sws_getConstVec(0.0, length);

  1729. 

  1730.     if (!vec)

  1731.         return NULL;

  1732. 

  1733.     for (i = 0; i < a->length; i++) {

  1734.         for (j = 0; j < b->length; j++) {

  1735.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1736.         }

  1737.     }

  1738. 

  1739.     return vec;

  1740. }

  1741. 

  1742. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1743. {

  1744.     int length = FFMAX(a->length, b->length);

  1745.     int i;

  1746.     SwsVector *vec = sws_getConstVec(0.0, length);

  1747. 

  1748.     if (!vec)

  1749.         return NULL;

  1750. 

  1751.     for (i = 0; i < a->length; i++)

  1752.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1753.     for (i = 0; i < b->length; i++)

  1754.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

  1755. 

  1756.     return vec;

  1757. }

  1758. 

  1759. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1760. {

  1761.     int length = FFMAX(a->length, b->length);

  1762.     int i;

  1763.     SwsVector *vec = sws_getConstVec(0.0, length);

  1764. 

  1765.     if (!vec)

  1766.         return NULL;

  1767. 

  1768.     for (i = 0; i < a->length; i++)

  1769.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1770.     for (i = 0; i < b->length; i++)

  1771.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

  1772. 

  1773.     return vec;

  1774. }

  1775. 

  1776. /* shift left / or right if "shift" is negative */

  1777. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1778. {

  1779.     int length = a->length + FFABS(shift) * 2;

  1780.     int i;

  1781.     SwsVector *vec = sws_getConstVec(0.0, length);

  1782. 

  1783.     if (!vec)

  1784.         return NULL;

  1785. 

  1786.     for (i = 0; i < a->length; i++) {

  1787.         vec->coeff[i + (length    - 1) / 2 -

  1788.                        (a->length - 1) / 2 - shift] = a->coeff[i];

  1789.     }

  1790. 

  1791.     return vec;

  1792. }

  1793. 

  1794. void sws_shiftVec(SwsVector *a, int shift)

  1795. {

  1796.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1797.     av_free(a->coeff);

  1798.     a->coeff  = shifted->coeff;

  1799.     a->length = shifted->length;

  1800.     av_free(shifted);

  1801. }

  1802. 

  1803. void sws_addVec(SwsVector *a, SwsVector *b)

  1804. {

  1805.     SwsVector *sum = sws_sumVec(a, b);

  1806.     av_free(a->coeff);

  1807.     a->coeff  = sum->coeff;

  1808.     a->length = sum->length;

  1809.     av_free(sum);

  1810. }

  1811. 

  1812. void sws_subVec(SwsVector *a, SwsVector *b)

  1813. {

  1814.     SwsVector *diff = sws_diffVec(a, b);

  1815.     av_free(a->coeff);

  1816.     a->coeff  = diff->coeff;

  1817.     a->length = diff->length;

  1818.     av_free(diff);

  1819. }

  1820. 

  1821. void sws_convVec(SwsVector *a, SwsVector *b)

  1822. {

  1823.     SwsVector *conv = sws_getConvVec(a, b);

  1824.     av_free(a->coeff);

  1825.     a->coeff  = conv->coeff;

  1826.     a->length = conv->length;

  1827.     av_free(conv);

  1828. }

  1829. 

  1830. SwsVector *sws_cloneVec(SwsVector *a)

  1831. {

  1832.     int i;

  1833.     SwsVector *vec = sws_allocVec(a->length);

  1834. 

  1835.     if (!vec)

  1836.         return NULL;

  1837. 

  1838.     for (i = 0; i < a->length; i++)

  1839.         vec->coeff[i] = a->coeff[i];

  1840. 

  1841.     return vec;

  1842. }

  1843. 

  1844. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1845. {

  1846.     int i;

  1847.     double max = 0;

  1848.     double min = 0;

  1849.     double range;

  1850. 

  1851.     for (i = 0; i < a->length; i++)

  1852.         if (a->coeff[i] > max)

  1853.             max = a->coeff[i];

  1854. 

  1855.     for (i = 0; i < a->length; i++)

  1856.         if (a->coeff[i] < min)

  1857.             min = a->coeff[i];

  1858. 

  1859.     range = max - min;

  1860. 

  1861.     for (i = 0; i < a->length; i++) {

  1862.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1863.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1864.         for (; x > 0; x--)

  1865.             av_log(log_ctx, log_level, " ");

  1866.         av_log(log_ctx, log_level, "|\n");

  1867.     }

  1868. }

  1869. 

  1870. void sws_freeVec(SwsVector *a)

  1871. {

  1872.     if (!a)

  1873.         return;

  1874.     av_freep(&a->coeff);

  1875.     a->length = 0;

  1876.     av_free(a);

  1877. }

  1878. 

  1879. void sws_freeFilter(SwsFilter *filter)

  1880. {

  1881.     if (!filter)

  1882.         return;

  1883. 

  1884.     if (filter->lumH)

  1885.         sws_freeVec(filter->lumH);

  1886.     if (filter->lumV)

  1887.         sws_freeVec(filter->lumV);

  1888.     if (filter->chrH)

  1889.         sws_freeVec(filter->chrH);

  1890.     if (filter->chrV)

  1891.         sws_freeVec(filter->chrV);

  1892.     av_free(filter);

  1893. }

  1894. 

  1895. void sws_freeContext(SwsContext *c)

  1896. {

  1897.     int i;

  1898.     if (!c)

  1899.         return;

  1900. 

  1901.     if (c->lumPixBuf) {

  1902.         for (i = 0; i < c->vLumBufSize; i++)

  1903.             av_freep(&c->lumPixBuf[i]);

  1904.         av_freep(&c->lumPixBuf);

  1905.     }

  1906. 

  1907.     if (c->chrUPixBuf) {

  1908.         for (i = 0; i < c->vChrBufSize; i++)

  1909.             av_freep(&c->chrUPixBuf[i]);

  1910.         av_freep(&c->chrUPixBuf);

  1911.         av_freep(&c->chrVPixBuf);

  1912.     }

  1913. 

  1914.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1915.         for (i = 0; i < c->vLumBufSize; i++)

  1916.             av_freep(&c->alpPixBuf[i]);

  1917.         av_freep(&c->alpPixBuf);

  1918.     }

  1919. 

  1920.     for (i = 0; i < 4; i++)

  1921.         av_freep(&c->dither_error[i]);

  1922. 

  1923.     av_freep(&c->vLumFilter);

  1924.     av_freep(&c->vChrFilter);

  1925.     av_freep(&c->hLumFilter);

  1926.     av_freep(&c->hChrFilter);

  1927. #if HAVE_ALTIVEC

  1928.     av_freep(&c->vYCoeffsBank);

  1929.     av_freep(&c->vCCoeffsBank);

  1930. #endif

  1931. 

  1932.     av_freep(&c->vLumFilterPos);

  1933.     av_freep(&c->vChrFilterPos);

  1934.     av_freep(&c->hLumFilterPos);

  1935.     av_freep(&c->hChrFilterPos);

  1936. 

  1937. #if HAVE_MMX_INLINE

  1938. #if USE_MMAP

  1939.     if (c->lumMmxextFilterCode)

  1940.         munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);

  1941.     if (c->chrMmxextFilterCode)

  1942.         munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);

  1943. #elif HAVE_VIRTUALALLOC

  1944.     if (c->lumMmxextFilterCode)

  1945.         VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);

  1946.     if (c->chrMmxextFilterCode)

  1947.         VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);

  1948. #else

  1949.     av_free(c->lumMmxextFilterCode);

  1950.     av_free(c->chrMmxextFilterCode);

  1951. #endif

  1952.     c->lumMmxextFilterCode = NULL;

  1953.     c->chrMmxextFilterCode = NULL;

  1954. #endif /* HAVE_MMX_INLINE */

  1955. 

  1956.     av_freep(&c->yuvTable);

  1957.     av_freep(&c->formatConvBuffer);

  1958. 

  1959.     av_free(c);

  1960. }

  1961. 

  1962. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  1963.                                         int srcH, enum AVPixelFormat srcFormat,

  1964.                                         int dstW, int dstH,

  1965.                                         enum AVPixelFormat dstFormat, int flags,

  1966.                                         SwsFilter *srcFilter,

  1967.                                         SwsFilter *dstFilter,

  1968.                                         const double *param)

  1969. {

  1970.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  1971.                                              SWS_PARAM_DEFAULT };

  1972. 

  1973.     if (!param)

  1974.         param = default_param;

  1975. 

  1976.     if (context &&

  1977.         (context->srcW      != srcW      ||

  1978.          context->srcH      != srcH      ||

  1979.          context->srcFormat != srcFormat ||

  1980.          context->dstW      != dstW      ||

  1981.          context->dstH      != dstH      ||

  1982.          context->dstFormat != dstFormat ||

  1983.          context->flags     != flags     ||

  1984.          context->param[0]  != param[0]  ||

  1985.          context->param[1]  != param[1])) {

  1986.         sws_freeContext(context);

  1987.         context = NULL;

  1988.     }

  1989. 

  1990.     if (!context) {

  1991.         if (!(context = sws_alloc_context()))

  1992.             return NULL;

  1993.         context->srcW      = srcW;

  1994.         context->srcH      = srcH;

  1995.         context->srcRange  = handle_jpeg(&srcFormat);

  1996.         context->src0Alpha = handle_0alpha(&srcFormat);

  1997.         context->srcXYZ    = handle_xyz(&srcFormat);

  1998.         context->srcFormat = srcFormat;

  1999.         context->dstW      = dstW;

  2000.         context->dstH      = dstH;

  2001.         context->dstRange  = handle_jpeg(&dstFormat);

  2002.         context->dst0Alpha = handle_0alpha(&dstFormat);

  2003.         context->dstXYZ    = handle_xyz(&dstFormat);

  2004.         context->dstFormat = dstFormat;

  2005.         context->flags     = flags;

  2006.         context->param[0]  = param[0];

  2007.         context->param[1]  = param[1];

  2008.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  2009.                                  context->srcRange,

  2010.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  2011.                                  context->dstRange, 0, 1 << 16, 1 << 16);

  2012.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  2013.             sws_freeContext(context);

  2014.             return NULL;

  2015.         }

  2016.     }

  2017.     return context;

  2018. }