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.
52724 Commits
32 Branches
944MB
Tree: e6d4e68755
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e6d4e68755'
${ noResults }
FFmpeg/libswscale/utils.c

2029 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.     [AV_PIX_FMT_GBRAP]       = { 0, 0 },

  202.     [AV_PIX_FMT_GBRAP16LE]   = { 0, 0 },

  203.     [AV_PIX_FMT_GBRAP16BE]   = { 0, 0 },

  204. };

  205. 

  206. int sws_isSupportedInput(enum AVPixelFormat pix_fmt)

  207. {

  208.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  209.            format_entries[pix_fmt].is_supported_in : 0;

  210. }

  211. 

  212. int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)

  213. {

  214.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  215.            format_entries[pix_fmt].is_supported_out : 0;

  216. }

  217. 

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

  219. 

  220. #if FF_API_SWS_FORMAT_NAME

  221. const char *sws_format_name(enum AVPixelFormat format)

  222. {

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

  224.     if (desc)

  225.         return desc->name;

  226.     else

  227.         return "Unknown format";

  228. }

  229. #endif

  230. 

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

  232.                              double dist)

  233. {

  234.     if (dist <= 1.0)

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

  236.     else

  237.         return getSplineCoeff(0.0,

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

  239.                                c + 3.0 * d,

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

  241.                               dist - 1.0);

  242. }

  243. 

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

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

  246.                       int filterAlign, int one, int flags, int cpu_flags,

  247.                       SwsVector *srcFilter, SwsVector *dstFilter,

  248.                       double param[2])

  249. {

  250.     int i;

  251.     int filterSize;

  252.     int filter2Size;

  253.     int minFilterSize;

  254.     int64_t *filter    = NULL;

  255.     int64_t *filter2   = NULL;

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

  257.     int ret            = -1;

  258. 

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

  260. 

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

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

  263. 

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

  265.         int i;

  266.         filterSize = 1;

  267.         FF_ALLOCZ_OR_GOTO(NULL, filter,

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

  269. 

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

  271.             filter[i * filterSize] = fone;

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

  273.         }

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

  275.         int i;

  276.         int64_t xDstInSrc;

  277.         filterSize = 1;

  278.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  280. 

  281.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  284. 

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

  286.             filter[i]       = fone;

  287.             xDstInSrc      += xInc;

  288.         }

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

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

  291.         int i;

  292.         int64_t xDstInSrc;

  293.         filterSize = 2;

  294.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  296. 

  297.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  300.             int j;

  301. 

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

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

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

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

  306.                 if (coeff < 0)

  307.                     coeff = 0;

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

  309.                 xx++;

  310.             }

  311.             xDstInSrc += xInc;

  312.         }

  313.     } else {

  314.         int64_t xDstInSrc;

  315.         int sizeFactor;

  316. 

  317.         if (flags & SWS_BICUBIC)

  318.             sizeFactor = 4;

  319.         else if (flags & SWS_X)

  320.             sizeFactor = 8;

  321.         else if (flags & SWS_AREA)

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

  323.         else if (flags & SWS_GAUSS)

  324.             sizeFactor = 8;     // infinite ;)

  325.         else if (flags & SWS_LANCZOS)

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

  327.         else if (flags & SWS_SINC)

  328.             sizeFactor = 20;    // infinite ;)

  329.         else if (flags & SWS_SPLINE)

  330.             sizeFactor = 20;    // infinite ;)

  331.         else if (flags & SWS_BILINEAR)

  332.             sizeFactor = 2;

  333.         else {

  334.             av_assert0(0);

  335.         }

  336. 

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

  338.             filterSize = 1 + sizeFactor;    // upscale

  339.         else

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

  341. 

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

  343.         filterSize = FFMAX(filterSize, 1);

  344. 

  345.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  347. 

  348.         xDstInSrc = xInc - 0x10000;

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

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

  351.             int j;

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

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

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

  355.                 double floatd;

  356.                 int64_t coeff;

  357. 

  358.                 if (xInc > 1 << 16)

  359.                     d = d * dstW / srcW;

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

  361. 

  362.                 if (flags & SWS_BICUBIC) {

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

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

  365. 

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

  367.                         coeff = 0.0;

  368.                     } else {

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

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

  371. 

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

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

  374.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  375.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  376.                         else

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

  378.                                       (6 * B + 30 * C) * dd  +

  379.                                     (-12 * B - 48 * C) * d   +

  380.                                       (8 * B + 24 * C) * (1 << 30);

  381.                     }

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

  383.                 }

  384. #if 0

  385.                 else if (flags & SWS_X) {

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

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

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

  389.                 }

  390. #endif

  391.                 else if (flags & SWS_X) {

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

  393.                     double c;

  394. 

  395.                     if (floatd < 1.0)

  396.                         c = cos(floatd * M_PI);

  397.                     else

  398.                         c = -1.0;

  399.                     if (c < 0.0)

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

  401.                     else

  402.                         c = pow(c, A);

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

  404.                 } else if (flags & SWS_AREA) {

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

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

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

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

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

  410.                     else

  411.                         coeff = 0.0;

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

  413.                 } else if (flags & SWS_GAUSS) {

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

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

  416.                 } else if (flags & SWS_SINC) {

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

  418.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  422.                     if (floatd > p)

  423.                         coeff = 0;

  424.                 } else if (flags & SWS_BILINEAR) {

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

  426.                     if (coeff < 0)

  427.                         coeff = 0;

  428.                     coeff *= fone >> 30;

  429.                 } else if (flags & SWS_SPLINE) {

  430.                     double p = -2.196152422706632;

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

  432.                 } else {

  433.                     av_assert0(0);

  434.                 }

  435. 

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

  437.                 xx++;

  438.             }

  439.             xDstInSrc += 2 * xInc;

  440.         }

  441.     }

  442. 

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

  444.      * av_free(filter);

  445.      */

  446.     av_assert0(filterSize > 0);

  447.     filter2Size = filterSize;

  448.     if (srcFilter)

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

  450.     if (dstFilter)

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

  452.     av_assert0(filter2Size > 0);

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

  454. 

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

  456.         int j, k;

  457. 

  458.         if (srcFilter) {

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

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

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

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

  463.             }

  464.         } else {

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

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

  467.         }

  468.         // FIXME dstFilter

  469. 

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

  471.     }

  472.     av_freep(&filter);

  473. 

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

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

  476.     minFilterSize = 0;

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

  478.         int min = filter2Size;

  479.         int j;

  480.         int64_t cutOff = 0.0;

  481. 

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

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

  484.             int k;

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

  486. 

  487.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  488.                 break;

  489. 

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

  491.              * filter otherwise */

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

  493.                 break;

  494. 

  495.             // move filter coefficients left

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

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

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

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

  500.         }

  501. 

  502.         cutOff = 0;

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

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

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

  506. 

  507.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  508.                 break;

  509.             min--;

  510.         }

  511. 

  512.         if (min > minFilterSize)

  513.             minFilterSize = min;

  514.     }

  515. 

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

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

  518.         if (minFilterSize < 5)

  519.             filterAlign = 4;

  520. 

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

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

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

  524.         if (minFilterSize < 3)

  525.             filterAlign = 1;

  526.     }

  527. 

  528.     if (INLINE_MMX(cpu_flags)) {

  529.         // special case for unscaled vertical filtering

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

  531.             filterAlign = 1;

  532.     }

  533. 

  534.     av_assert0(minFilterSize > 0);

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

  536.     av_assert0(filterSize > 0);

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

  538.     if (filterSize >= MAX_FILTER_SIZE * 16 /

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

  540.         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);

  541.         goto fail;

  542.     }

  543.     *outFilterSize = filterSize;

  544. 

  545.     if (flags & SWS_PRINT_INFO)

  546.         av_log(NULL, AV_LOG_VERBOSE,

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

  548.                filter2Size, filterSize);

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

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

  551.         int j;

  552. 

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

  554.             if (j >= filter2Size)

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

  556.             else

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

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

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

  560.         }

  561.     }

  562. 

  563.     // FIXME try to align filterPos if possible

  564. 

  565.     // fix borders

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

  567.         int j;

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

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

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

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

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

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

  574.             }

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

  576.         }

  577. 

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

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

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

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

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

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

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

  585.             }

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

  587.         }

  588.     }

  589. 

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

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

  592.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

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

  594. 

  595.     /* normalize & store in outFilter */

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

  597.         int j;

  598.         int64_t error = 0;

  599.         int64_t sum   = 0;

  600. 

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

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

  603.         }

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

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

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

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

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

  609.             error                                  = v - intV * sum;

  610.         }

  611.     }

  612. 

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

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

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

  616.                                                       * read over the end */

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

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

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

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

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

  622.     }

  623. 

  624.     ret = 0;

  625. 

  626. fail:

  627.     if(ret < 0)

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

  629.     av_free(filter);

  630.     av_free(filter2);

  631.     return ret;

  632. }

  633. 

  634. #if HAVE_MMXEXT_INLINE

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

  636.                                int16_t *filter, int32_t *filterPos,

  637.                                int numSplits)

  638. {

  639.     uint8_t *fragmentA;

  640.     x86_reg imm8OfPShufW1A;

  641.     x86_reg imm8OfPShufW2A;

  642.     x86_reg fragmentLengthA;

  643.     uint8_t *fragmentB;

  644.     x86_reg imm8OfPShufW1B;

  645.     x86_reg imm8OfPShufW2B;

  646.     x86_reg fragmentLengthB;

  647.     int fragmentPos;

  648. 

  649.     int xpos, i;

  650. 

  651.     // create an optimized horizontal scaling routine

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

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

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

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

  656.      */

  657. 

  658.     // code fragment

  659. 

  660.     __asm__ volatile (

  661.         "jmp                         9f                 \n\t"

  662.         // Begin

  663.         "0:                                             \n\t"

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

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

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

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

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

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

  670.         "1:                                             \n\t"

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

  672.         "2:                                             \n\t"

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

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

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

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

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

  678. 

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

  680. 

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

  682.         // End

  683.         "9:                                             \n\t"

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

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

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

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

  688.         "dec                         %1                 \n\t"

  689.         "dec                         %2                 \n\t"

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

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

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

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

  694. 

  695. 

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

  697.           "=r" (fragmentLengthA)

  698.         );

  699. 

  700.     __asm__ volatile (

  701.         "jmp                         9f                 \n\t"

  702.         // Begin

  703.         "0:                                             \n\t"

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

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

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

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

  708.         "1:                                             \n\t"

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

  710.         "2:                                             \n\t"

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

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

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

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

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

  716. 

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

  718. 

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

  720.         // End

  721.         "9:                                             \n\t"

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

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

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

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

  726.         "dec                         %1                 \n\t"

  727.         "dec                         %2                 \n\t"

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

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

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

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

  732. 

  733. 

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

  735.           "=r" (fragmentLengthB)

  736.         );

  737. 

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

  739.     fragmentPos = 0;

  740. 

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

  742.         int xx = xpos >> 16;

  743. 

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

  745.             int a                  = 0;

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

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

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

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

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

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

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

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

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

  755.             int shift              = 0;

  756. 

  757.             if (filterCode) {

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

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

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

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

  762.                 filterPos[i / 2] = xx;

  763. 

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

  765. 

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

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

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

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

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

  771.                                                                (c << 4) |

  772.                                                                (d << 6);

  773. 

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

  775.                     shift = maxShift;               // avoid overread

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

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

  778. 

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

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

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

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

  783.                 }

  784.             }

  785. 

  786.             fragmentPos += fragmentLength;

  787. 

  788.             if (filterCode)

  789.                 filterCode[fragmentPos] = RET;

  790.         }

  791.         xpos += xInc;

  792.     }

  793.     if (filterCode)

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

  795. 

  796.     return fragmentPos + 1;

  797. }

  798. #endif /* HAVE_MMXEXT_INLINE */

  799. 

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

  801. {

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

  803.     *h = desc->log2_chroma_w;

  804.     *v = desc->log2_chroma_h;

  805. }

  806. 

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

  808. {

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

  810.     int64_t vr =  table[0];

  811.     int64_t ub =  table[1];

  812.     int64_t ug = -table[2];

  813.     int64_t vg = -table[3];

  814.     int64_t ONE = 65536;

  815.     int64_t cy = ONE;

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

  817.     int i;

  818.     static const int8_t map[] = {

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

  820.     RY_IDX, -1    , GY_IDX, RY_IDX, RY_IDX, -1    , GY_IDX, RY_IDX,

  821.     RY_IDX, GY_IDX, -1    , RY_IDX, RY_IDX, GY_IDX, -1    , RY_IDX,

  822.     BY_IDX, -1    , GY_IDX, BY_IDX, BY_IDX, -1    , GY_IDX, BY_IDX,

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

  824.     RU_IDX, -1    , GU_IDX, RU_IDX, RU_IDX, -1    , GU_IDX, RU_IDX,

  825.     RU_IDX, GU_IDX, -1    , RU_IDX, RU_IDX, GU_IDX, -1    , RU_IDX,

  826.     BU_IDX, -1    , GU_IDX, BU_IDX, BU_IDX, -1    , GU_IDX, BU_IDX,

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

  828.     RV_IDX, -1    , GV_IDX, RV_IDX, RV_IDX, -1    , GV_IDX, RV_IDX,

  829.     RV_IDX, GV_IDX, -1    , RV_IDX, RV_IDX, GV_IDX, -1    , RV_IDX,

  830.     BV_IDX, -1    , GV_IDX, BV_IDX, BV_IDX, -1    , GV_IDX, BV_IDX,

  831.     RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,

  832.     BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,

  833.     GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    ,

  834.     -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX,

  835.     RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,

  836.     BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,

  837.     GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    ,

  838.     -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX,

  839.     RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,

  840.     BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,

  841.     GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    ,

  842.     -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, //23

  843.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //24

  844.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //25

  845.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //26

  846.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //27

  847.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //28

  848.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //29

  849.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //30

  850.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //31

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

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

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

  854.     };

  855. 

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

  857. 

  858.     if (!dstRange) {

  859.         cy = cy * 255 / 219;

  860.     } else {

  861.         vr = vr * 224 / 255;

  862.         ub = ub * 224 / 255;

  863.         ug = ug * 224 / 255;

  864.         vg = vg * 224 / 255;

  865.     }

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

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

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

  869. 

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

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

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

  873. 

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

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

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

  877. 

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

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

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

  881. 

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

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

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

  885. 

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

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

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

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

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

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

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

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

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

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

  896.     }

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

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

  899. }

  900. 

  901. static void fill_xyztables(struct SwsContext *c)

  902. {

  903.     int i;

  904.     double xyzgamma = XYZ_GAMMA;

  905.     double rgbgamma = 1.0 / RGB_GAMMA;

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

  907.         {13270, -6295, -2041},

  908.         {-3969,  7682,   170},

  909.         {  228,  -835,  4329} };

  910.     static int16_t xyzgamma_tab[4096], rgbgamma_tab[4096];

  911. 

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

  913.     c->xyzgamma = xyzgamma_tab;

  914.     c->rgbgamma = rgbgamma_tab;

  915. 

  916.     if (rgbgamma_tab[4095])

  917.         return;

  918. 

  919.     /* set gamma vectors */

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

  921.         xyzgamma_tab[i] = lrint(pow(i / 4095.0, xyzgamma) * 4095.0);

  922.         rgbgamma_tab[i] = lrint(pow(i / 4095.0, rgbgamma) * 4095.0);

  923.     }

  924. }

  925. 

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

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

  928.                              int brightness, int contrast, int saturation)

  929. {

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

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

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

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

  934. 

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

  936.         dstRange = 0;

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

  938.         srcRange = 0;

  939. 

  940.     c->brightness = brightness;

  941.     c->contrast   = contrast;

  942.     c->saturation = saturation;

  943.     c->srcRange   = srcRange;

  944.     c->dstRange   = dstRange;

  945. 

  946.     fill_xyztables(c);

  947. 

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

  949.         return -1;

  950. 

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

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

  953. 

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

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

  956.                              contrast, saturation);

  957.     // FIXME factorize

  958. 

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

  960.         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,

  961.                                        contrast, saturation);

  962.     }

  963. 

  964.     fill_rgb2yuv_table(c, table, dstRange);

  965. 

  966.     return 0;

  967. }

  968. 

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

  970.                              int *srcRange, int **table, int *dstRange,

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

  972. {

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

  974.         return -1;

  975. 

  976.     *inv_table  = c->srcColorspaceTable;

  977.     *table      = c->dstColorspaceTable;

  978.     *srcRange   = c->srcRange;

  979.     *dstRange   = c->dstRange;

  980.     *brightness = c->brightness;

  981.     *contrast   = c->contrast;

  982.     *saturation = c->saturation;

  983. 

  984.     return 0;

  985. }

  986. 

  987. static int handle_jpeg(enum AVPixelFormat *format)

  988. {

  989.     switch (*format) {

  990.     case AV_PIX_FMT_YUVJ420P:

  991.         *format = AV_PIX_FMT_YUV420P;

  992.         return 1;

  993.     case AV_PIX_FMT_YUVJ422P:

  994.         *format = AV_PIX_FMT_YUV422P;

  995.         return 1;

  996.     case AV_PIX_FMT_YUVJ444P:

  997.         *format = AV_PIX_FMT_YUV444P;

  998.         return 1;

  999.     case AV_PIX_FMT_YUVJ440P:

  1000.         *format = AV_PIX_FMT_YUV440P;

  1001.         return 1;

  1002.     default:

  1003.         return 0;

  1004.     }

  1005. }

  1006. 

  1007. static int handle_0alpha(enum AVPixelFormat *format)

  1008. {

  1009.     switch (*format) {

  1010.     case AV_PIX_FMT_0BGR    : *format = AV_PIX_FMT_ABGR   ; return 1;

  1011.     case AV_PIX_FMT_BGR0    : *format = AV_PIX_FMT_BGRA   ; return 4;

  1012.     case AV_PIX_FMT_0RGB    : *format = AV_PIX_FMT_ARGB   ; return 1;

  1013.     case AV_PIX_FMT_RGB0    : *format = AV_PIX_FMT_RGBA   ; return 4;

  1014.     default:                                          return 0;

  1015.     }

  1016. }

  1017. 

  1018. static int handle_xyz(enum AVPixelFormat *format)

  1019. {

  1020.     switch (*format) {

  1021.     case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;

  1022.     case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;

  1023.     default:                                                return 0;

  1024.     }

  1025. }

  1026. 

  1027. SwsContext *sws_alloc_context(void)

  1028. {

  1029.     SwsContext *c = av_mallocz(sizeof(SwsContext));

  1030. 

  1031.     if (c) {

  1032.         c->av_class = &sws_context_class;

  1033.         av_opt_set_defaults(c);

  1034.     }

  1035. 

  1036.     return c;

  1037. }

  1038. 

  1039. av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,

  1040.                              SwsFilter *dstFilter)

  1041. {

  1042.     int i, j;

  1043.     int usesVFilter, usesHFilter;

  1044.     int unscaled;

  1045.     SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };

  1046.     int srcW              = c->srcW;

  1047.     int srcH              = c->srcH;

  1048.     int dstW              = c->dstW;

  1049.     int dstH              = c->dstH;

  1050.     int dst_stride        = FFALIGN(dstW * sizeof(int16_t) + 66, 16);

  1051.     int flags, cpu_flags;

  1052.     enum AVPixelFormat srcFormat = c->srcFormat;

  1053.     enum AVPixelFormat dstFormat = c->dstFormat;

  1054.     const AVPixFmtDescriptor *desc_src = av_pix_fmt_desc_get(srcFormat);

  1055.     const AVPixFmtDescriptor *desc_dst = av_pix_fmt_desc_get(dstFormat);

  1056. 

  1057.     cpu_flags = av_get_cpu_flags();

  1058.     flags     = c->flags;

  1059.     emms_c();

  1060.     if (!rgb15to16)

  1061.         sws_rgb2rgb_init();

  1062. 

  1063.     unscaled = (srcW == dstW && srcH == dstH);

  1064. 

  1065.     handle_jpeg(&srcFormat);

  1066.     handle_jpeg(&dstFormat);

  1067.     handle_0alpha(&srcFormat);

  1068.     handle_0alpha(&dstFormat);

  1069.     handle_xyz(&srcFormat);

  1070.     handle_xyz(&dstFormat);

  1071. 

  1072.     if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){

  1073.         av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");

  1074.         c->srcFormat= srcFormat;

  1075.         c->dstFormat= dstFormat;

  1076.     }

  1077. 

  1078.     if (!sws_isSupportedInput(srcFormat)) {

  1079.         av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",

  1080.                av_get_pix_fmt_name(srcFormat));

  1081.         return AVERROR(EINVAL);

  1082.     }

  1083.     if (!sws_isSupportedOutput(dstFormat)) {

  1084.         av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",

  1085.                av_get_pix_fmt_name(dstFormat));

  1086.         return AVERROR(EINVAL);

  1087.     }

  1088. 

  1089.     i = flags & (SWS_POINT         |

  1090.                  SWS_AREA          |

  1091.                  SWS_BILINEAR      |

  1092.                  SWS_FAST_BILINEAR |

  1093.                  SWS_BICUBIC       |

  1094.                  SWS_X             |

  1095.                  SWS_GAUSS         |

  1096.                  SWS_LANCZOS       |

  1097.                  SWS_SINC          |

  1098.                  SWS_SPLINE        |

  1099.                  SWS_BICUBLIN);

  1100.     if (!i || (i & (i - 1))) {

  1101.         av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);

  1102.         return AVERROR(EINVAL);

  1103.     }

  1104.     /* sanity check */

  1105.     if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {

  1106.         /* FIXME check if these are enough and try to lower them after

  1107.          * fixing the relevant parts of the code */

  1108.         av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",

  1109.                srcW, srcH, dstW, dstH);

  1110.         return AVERROR(EINVAL);

  1111.     }

  1112. 

  1113.     if (!dstFilter)

  1114.         dstFilter = &dummyFilter;

  1115.     if (!srcFilter)

  1116.         srcFilter = &dummyFilter;

  1117. 

  1118.     c->lumXInc      = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;

  1119.     c->lumYInc      = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;

  1120.     c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);

  1121.     c->srcFormatBpp = av_get_bits_per_pixel(desc_src);

  1122.     c->vRounder     = 4 * 0x0001000100010001ULL;

  1123. 

  1124.     usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||

  1125.                   (srcFilter->chrV && srcFilter->chrV->length > 1) ||

  1126.                   (dstFilter->lumV && dstFilter->lumV->length > 1) ||

  1127.                   (dstFilter->chrV && dstFilter->chrV->length > 1);

  1128.     usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||

  1129.                   (srcFilter->chrH && srcFilter->chrH->length > 1) ||

  1130.                   (dstFilter->lumH && dstFilter->lumH->length > 1) ||

  1131.                   (dstFilter->chrH && dstFilter->chrH->length > 1);

  1132. 

  1133.     getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);

  1134.     getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);

  1135. 

  1136.     if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {

  1137.         if (dstW&1) {

  1138.             av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");

  1139.             flags |= SWS_FULL_CHR_H_INT;

  1140.             c->flags = flags;

  1141.         }

  1142.     }

  1143. 

  1144.     if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||

  1145.        dstFormat == AV_PIX_FMT_RGB4_BYTE ||

  1146.        dstFormat == AV_PIX_FMT_BGR8 ||

  1147.        dstFormat == AV_PIX_FMT_RGB8) {

  1148.         if (flags & SWS_ERROR_DIFFUSION && !(flags & SWS_FULL_CHR_H_INT)) {

  1149.             av_log(c, AV_LOG_DEBUG,

  1150.                 "Error diffusion dither is only supported in full chroma interpolation for destination format '%s'\n",

  1151.                 av_get_pix_fmt_name(dstFormat));

  1152.             flags   |= SWS_FULL_CHR_H_INT;

  1153.             c->flags = flags;

  1154.         }

  1155.         if (!(flags & SWS_ERROR_DIFFUSION) && (flags & SWS_FULL_CHR_H_INT)) {

  1156.             av_log(c, AV_LOG_DEBUG,

  1157.                 "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",

  1158.                 av_get_pix_fmt_name(dstFormat));

  1159.             flags   |= SWS_ERROR_DIFFUSION;

  1160.             c->flags = flags;

  1161.         }

  1162.     }

  1163.     if (isPlanarRGB(dstFormat)) {

  1164.         if (!(flags & SWS_FULL_CHR_H_INT)) {

  1165.             av_log(c, AV_LOG_DEBUG,

  1166.                    "%s output is not supported with half chroma resolution, switching to full\n",

  1167.                    av_get_pix_fmt_name(dstFormat));

  1168.             flags   |= SWS_FULL_CHR_H_INT;

  1169.             c->flags = flags;

  1170.         }

  1171.     }

  1172. 

  1173.     /* reuse chroma for 2 pixels RGB/BGR unless user wants full

  1174.      * chroma interpolation */

  1175.     if (flags & SWS_FULL_CHR_H_INT &&

  1176.         isAnyRGB(dstFormat)        &&

  1177.         !isPlanarRGB(dstFormat)    &&

  1178.         dstFormat != AV_PIX_FMT_RGBA  &&

  1179.         dstFormat != AV_PIX_FMT_ARGB  &&

  1180.         dstFormat != AV_PIX_FMT_BGRA  &&

  1181.         dstFormat != AV_PIX_FMT_ABGR  &&

  1182.         dstFormat != AV_PIX_FMT_RGB24 &&

  1183.         dstFormat != AV_PIX_FMT_BGR24 &&

  1184.         dstFormat != AV_PIX_FMT_BGR4_BYTE &&

  1185.         dstFormat != AV_PIX_FMT_RGB4_BYTE &&

  1186.         dstFormat != AV_PIX_FMT_BGR8 &&

  1187.         dstFormat != AV_PIX_FMT_RGB8

  1188.     ) {

  1189.         av_log(c, AV_LOG_WARNING,

  1190.                "full chroma interpolation for destination format '%s' not yet implemented\n",

  1191.                av_get_pix_fmt_name(dstFormat));

  1192.         flags   &= ~SWS_FULL_CHR_H_INT;

  1193.         c->flags = flags;

  1194.     }

  1195.     if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))

  1196.         c->chrDstHSubSample = 1;

  1197. 

  1198.     // drop some chroma lines if the user wants it

  1199.     c->vChrDrop          = (flags & SWS_SRC_V_CHR_DROP_MASK) >>

  1200.                            SWS_SRC_V_CHR_DROP_SHIFT;

  1201.     c->chrSrcVSubSample += c->vChrDrop;

  1202. 

  1203.     /* drop every other pixel for chroma calculation unless user

  1204.      * wants full chroma */

  1205.     if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP)   &&

  1206.         srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&

  1207.         srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&

  1208.         srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&

  1209.         srcFormat != AV_PIX_FMT_GBRP9BE   && srcFormat != AV_PIX_FMT_GBRP9LE  &&

  1210.         srcFormat != AV_PIX_FMT_GBRP10BE  && srcFormat != AV_PIX_FMT_GBRP10LE &&

  1211.         srcFormat != AV_PIX_FMT_GBRP12BE  && srcFormat != AV_PIX_FMT_GBRP12LE &&

  1212.         srcFormat != AV_PIX_FMT_GBRP14BE  && srcFormat != AV_PIX_FMT_GBRP14LE &&

  1213.         srcFormat != AV_PIX_FMT_GBRP16BE  && srcFormat != AV_PIX_FMT_GBRP16LE &&

  1214.         ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||

  1215.          (flags & SWS_FAST_BILINEAR)))

  1216.         c->chrSrcHSubSample = 1;

  1217. 

  1218.     // Note the -((-x)>>y) is so that we always round toward +inf.

  1219.     c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);

  1220.     c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);

  1221.     c->chrDstW = -((-dstW) >> c->chrDstHSubSample);

  1222.     c->chrDstH = -((-dstH) >> c->chrDstVSubSample);

  1223. 

  1224.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);

  1225. 

  1226.     /* unscaled special cases */

  1227.     if (unscaled && !usesHFilter && !usesVFilter &&

  1228.         (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {

  1229.         ff_get_unscaled_swscale(c);

  1230. 

  1231.         if (c->swScale) {

  1232.             if (flags & SWS_PRINT_INFO)

  1233.                 av_log(c, AV_LOG_INFO,

  1234.                        "using unscaled %s -> %s special converter\n",

  1235.                        av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));

  1236.             return 0;

  1237.         }

  1238.     }

  1239. 

  1240.     c->srcBpc = 1 + desc_src->comp[0].depth_minus1;

  1241.     if (c->srcBpc < 8)

  1242.         c->srcBpc = 8;

  1243.     c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;

  1244.     if (c->dstBpc < 8)

  1245.         c->dstBpc = 8;

  1246.     if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)

  1247.         c->srcBpc = 16;

  1248.     if (c->dstBpc == 16)

  1249.         dst_stride <<= 1;

  1250. 

  1251.     if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {

  1252.         c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&

  1253.                               (srcW & 15) == 0) ? 1 : 0;

  1254.         if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0

  1255. 

  1256.             && (flags & SWS_FAST_BILINEAR)) {

  1257.             if (flags & SWS_PRINT_INFO)

  1258.                 av_log(c, AV_LOG_INFO,

  1259.                        "output width is not a multiple of 32 -> no MMXEXT scaler\n");

  1260.         }

  1261.         if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))

  1262.             c->canMMXEXTBeUsed = 0;

  1263.     } else

  1264.         c->canMMXEXTBeUsed = 0;

  1265. 

  1266.     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;

  1267.     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

  1268. 

  1269.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  1270.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  1271.      * correct scaling.

  1272.      * n-2 is the last chrominance sample available.

  1273.      * This is not perfect, but no one should notice the difference, the more

  1274.      * correct variant would be like the vertical one, but that would require

  1275.      * some special code for the first and last pixel */

  1276.     if (flags & SWS_FAST_BILINEAR) {

  1277.         if (c->canMMXEXTBeUsed) {

  1278.             c->lumXInc += 20;

  1279.             c->chrXInc += 20;

  1280.         }

  1281.         // we don't use the x86 asm scaler if MMX is available

  1282.         else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {

  1283.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1284.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1285.         }

  1286.     }

  1287. 

  1288. #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)

  1289. 

  1290.     /* precalculate horizontal scaler filter coefficients */

  1291.     {

  1292. #if HAVE_MMXEXT_INLINE

  1293. // can't downscale !!!

  1294.         if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {

  1295.             c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,

  1296.                                                              NULL, NULL, 8);

  1297.             c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,

  1298.                                                              NULL, NULL, NULL, 4);

  1299. 

  1300. #if USE_MMAP

  1301.             c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,

  1302.                                           PROT_READ | PROT_WRITE,

  1303.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1304.                                           -1, 0);

  1305.             c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,

  1306.                                           PROT_READ | PROT_WRITE,

  1307.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1308.                                           -1, 0);

  1309. #elif HAVE_VIRTUALALLOC

  1310.             c->lumMmxextFilterCode = VirtualAlloc(NULL,

  1311.                                                   c->lumMmxextFilterCodeSize,

  1312.                                                   MEM_COMMIT,

  1313.                                                   PAGE_EXECUTE_READWRITE);

  1314.             c->chrMmxextFilterCode = VirtualAlloc(NULL,

  1315.                                                   c->chrMmxextFilterCodeSize,

  1316.                                                   MEM_COMMIT,

  1317.                                                   PAGE_EXECUTE_READWRITE);

  1318. #else

  1319.             c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);

  1320.             c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);

  1321. #endif

  1322. 

  1323. #ifdef MAP_ANONYMOUS

  1324.             if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)

  1325. #else

  1326.             if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)

  1327. #endif

  1328.             {

  1329.                 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");

  1330.                 return AVERROR(ENOMEM);

  1331.             }

  1332. 

  1333.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1334.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1335.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1336.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1337. 

  1338.             init_hscaler_mmxext(      dstW, c->lumXInc, c->lumMmxextFilterCode,

  1339.                                 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);

  1340.             init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,

  1341.                                 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);

  1342. 

  1343. #if USE_MMAP

  1344.             mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1345.             mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1346. #endif

  1347.         } else

  1348. #endif /* HAVE_MMXEXT_INLINE */

  1349.         {

  1350.             const int filterAlign =

  1351.                 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :

  1352.                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1353.                 1;

  1354. 

  1355.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1356.                            &c->hLumFilterSize, c->lumXInc,

  1357.                            srcW, dstW, filterAlign, 1 << 14,

  1358.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1359.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1360.                            c->param) < 0)

  1361.                 goto fail;

  1362.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1363.                            &c->hChrFilterSize, c->chrXInc,

  1364.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1365.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1366.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1367.                            c->param) < 0)

  1368.                 goto fail;

  1369.         }

  1370.     } // initialize horizontal stuff

  1371. 

  1372.     /* precalculate vertical scaler filter coefficients */

  1373.     {

  1374.         const int filterAlign =

  1375.             (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :

  1376.             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1377.             1;

  1378. 

  1379.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1380.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

  1381.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1382.                        cpu_flags, srcFilter->lumV, dstFilter->lumV,

  1383.                        c->param) < 0)

  1384.             goto fail;

  1385.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1386.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1387.                        filterAlign, (1 << 12),

  1388.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1389.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1390.                        c->param) < 0)

  1391.             goto fail;

  1392. 

  1393. #if HAVE_ALTIVEC

  1394.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1395.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1396. 

  1397.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1398.             int j;

  1399.             short *p = (short *)&c->vYCoeffsBank[i];

  1400.             for (j = 0; j < 8; j++)

  1401.                 p[j] = c->vLumFilter[i];

  1402.         }

  1403. 

  1404.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1405.             int j;

  1406.             short *p = (short *)&c->vCCoeffsBank[i];

  1407.             for (j = 0; j < 8; j++)

  1408.                 p[j] = c->vChrFilter[i];

  1409.         }

  1410. #endif

  1411.     }

  1412. 

  1413.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1414.     c->vLumBufSize = c->vLumFilterSize;

  1415.     c->vChrBufSize = c->vChrFilterSize;

  1416.     for (i = 0; i < dstH; i++) {

  1417.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1418.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1419.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1420.                                << c->chrSrcVSubSample));

  1421. 

  1422.         nextSlice >>= c->chrSrcVSubSample;

  1423.         nextSlice <<= c->chrSrcVSubSample;

  1424.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1425.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1426.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1427.             (nextSlice >> c->chrSrcVSubSample))

  1428.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1429.                              c->vChrFilterPos[chrI];

  1430.     }

  1431. 

  1432.     for (i = 0; i < 4; i++)

  1433.         FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);

  1434. 

  1435.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1436.      * need to allocate several megabytes to handle all possible cases) */

  1437.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1438.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1439.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1440.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1441.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1442.     /* Note we need at least one pixel more at the end because of the MMX code

  1443.      * (just in case someone wants to replace the 4000/8000). */

  1444.     /* align at 16 bytes for AltiVec */

  1445.     for (i = 0; i < c->vLumBufSize; i++) {

  1446.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1447.                           dst_stride + 16, fail);

  1448.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1449.     }

  1450.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1451.     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);

  1452.     c->uv_offx2 = dst_stride + 16;

  1453.     for (i = 0; i < c->vChrBufSize; i++) {

  1454.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1455.                          dst_stride * 2 + 32, fail);

  1456.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1457.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1458.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1459.     }

  1460.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1461.         for (i = 0; i < c->vLumBufSize; i++) {

  1462.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],

  1463.                               dst_stride + 16, fail);

  1464.             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];

  1465.         }

  1466. 

  1467.     // try to avoid drawing green stuff between the right end and the stride end

  1468.     for (i = 0; i < c->vChrBufSize; i++)

  1469.         if(desc_dst->comp[0].depth_minus1 == 15){

  1470.             av_assert0(c->dstBpc > 14);

  1471.             for(j=0; j<dst_stride/2+1; j++)

  1472.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1473.         } else

  1474.             for(j=0; j<dst_stride+1; j++)

  1475.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1476. 

  1477.     av_assert0(c->chrDstH <= dstH);

  1478. 

  1479.     if (flags & SWS_PRINT_INFO) {

  1480.         if (flags & SWS_FAST_BILINEAR)

  1481.             av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1482.         else if (flags & SWS_BILINEAR)

  1483.             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1484.         else if (flags & SWS_BICUBIC)

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

  1486.         else if (flags & SWS_X)

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

  1488.         else if (flags & SWS_POINT)

  1489.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1490.         else if (flags & SWS_AREA)

  1491.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1492.         else if (flags & SWS_BICUBLIN)

  1493.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1494.         else if (flags & SWS_GAUSS)

  1495.             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1496.         else if (flags & SWS_SINC)

  1497.             av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1498.         else if (flags & SWS_LANCZOS)

  1499.             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1500.         else if (flags & SWS_SPLINE)

  1501.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1502.         else

  1503.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1504. 

  1505.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1506.                av_get_pix_fmt_name(srcFormat),

  1507. #ifdef DITHER1XBPP

  1508.                dstFormat == AV_PIX_FMT_BGR555   || dstFormat == AV_PIX_FMT_BGR565   ||

  1509.                dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||

  1510.                dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?

  1511.                                                              "dithered " : "",

  1512. #else

  1513.                "",

  1514. #endif

  1515.                av_get_pix_fmt_name(dstFormat));

  1516. 

  1517.         if (INLINE_MMXEXT(cpu_flags))

  1518.             av_log(c, AV_LOG_INFO, "using MMXEXT\n");

  1519.         else if (INLINE_AMD3DNOW(cpu_flags))

  1520.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1521.         else if (INLINE_MMX(cpu_flags))

  1522.             av_log(c, AV_LOG_INFO, "using MMX\n");

  1523.         else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)

  1524.             av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1525.         else

  1526.             av_log(c, AV_LOG_INFO, "using C\n");

  1527. 

  1528.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1529.         av_log(c, AV_LOG_DEBUG,

  1530.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1531.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1532.         av_log(c, AV_LOG_DEBUG,

  1533.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1534.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1535.                c->chrXInc, c->chrYInc);

  1536.     }

  1537. 

  1538.     c->swScale = ff_getSwsFunc(c);

  1539.     return 0;

  1540. fail: // FIXME replace things by appropriate error codes

  1541.     return -1;

  1542. }

  1543. 

  1544. #if FF_API_SWS_GETCONTEXT

  1545. SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,

  1546.                            int dstW, int dstH, enum AVPixelFormat dstFormat,

  1547.                            int flags, SwsFilter *srcFilter,

  1548.                            SwsFilter *dstFilter, const double *param)

  1549. {

  1550.     SwsContext *c;

  1551. 

  1552.     if (!(c = sws_alloc_context()))

  1553.         return NULL;

  1554. 

  1555.     c->flags     = flags;

  1556.     c->srcW      = srcW;

  1557.     c->srcH      = srcH;

  1558.     c->dstW      = dstW;

  1559.     c->dstH      = dstH;

  1560.     c->srcRange  = handle_jpeg(&srcFormat);

  1561.     c->dstRange  = handle_jpeg(&dstFormat);

  1562.     c->src0Alpha = handle_0alpha(&srcFormat);

  1563.     c->dst0Alpha = handle_0alpha(&dstFormat);

  1564.     c->srcXYZ    = handle_xyz(&srcFormat);

  1565.     c->dstXYZ    = handle_xyz(&dstFormat);

  1566.     c->srcFormat = srcFormat;

  1567.     c->dstFormat = dstFormat;

  1568. 

  1569.     if (param) {

  1570.         c->param[0] = param[0];

  1571.         c->param[1] = param[1];

  1572.     }

  1573.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1574.                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1575.                              c->dstRange, 0, 1 << 16, 1 << 16);

  1576. 

  1577.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1578.         sws_freeContext(c);

  1579.         return NULL;

  1580.     }

  1581. 

  1582.     return c;

  1583. }

  1584. #endif

  1585. 

  1586. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1587.                                 float lumaSharpen, float chromaSharpen,

  1588.                                 float chromaHShift, float chromaVShift,

  1589.                                 int verbose)

  1590. {

  1591.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1592.     if (!filter)

  1593.         return NULL;

  1594. 

  1595.     if (lumaGBlur != 0.0) {

  1596.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1597.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1598.     } else {

  1599.         filter->lumH = sws_getIdentityVec();

  1600.         filter->lumV = sws_getIdentityVec();

  1601.     }

  1602. 

  1603.     if (chromaGBlur != 0.0) {

  1604.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1605.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1606.     } else {

  1607.         filter->chrH = sws_getIdentityVec();

  1608.         filter->chrV = sws_getIdentityVec();

  1609.     }

  1610. 

  1611.     if (chromaSharpen != 0.0) {

  1612.         SwsVector *id = sws_getIdentityVec();

  1613.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1614.         sws_scaleVec(filter->chrV, -chromaSharpen);

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

  1616.         sws_addVec(filter->chrV, id);

  1617.         sws_freeVec(id);

  1618.     }

  1619. 

  1620.     if (lumaSharpen != 0.0) {

  1621.         SwsVector *id = sws_getIdentityVec();

  1622.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1623.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1624.         sws_addVec(filter->lumH, id);

  1625.         sws_addVec(filter->lumV, id);

  1626.         sws_freeVec(id);

  1627.     }

  1628. 

  1629.     if (chromaHShift != 0.0)

  1630.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1631. 

  1632.     if (chromaVShift != 0.0)

  1633.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1634. 

  1635.     sws_normalizeVec(filter->chrH, 1.0);

  1636.     sws_normalizeVec(filter->chrV, 1.0);

  1637.     sws_normalizeVec(filter->lumH, 1.0);

  1638.     sws_normalizeVec(filter->lumV, 1.0);

  1639. 

  1640.     if (verbose)

  1641.         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1642.     if (verbose)

  1643.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1644. 

  1645.     return filter;

  1646. }

  1647. 

  1648. SwsVector *sws_allocVec(int length)

  1649. {

  1650.     SwsVector *vec;

  1651. 

  1652.     if(length <= 0 || length > INT_MAX/ sizeof(double))

  1653.         return NULL;

  1654. 

  1655.     vec = av_malloc(sizeof(SwsVector));

  1656.     if (!vec)

  1657.         return NULL;

  1658.     vec->length = length;

  1659.     vec->coeff  = av_malloc(sizeof(double) * length);

  1660.     if (!vec->coeff)

  1661.         av_freep(&vec);

  1662.     return vec;

  1663. }

  1664. 

  1665. SwsVector *sws_getGaussianVec(double variance, double quality)

  1666. {

  1667.     const int length = (int)(variance * quality + 0.5) | 1;

  1668.     int i;

  1669.     double middle  = (length - 1) * 0.5;

  1670.     SwsVector *vec;

  1671. 

  1672.     if(variance < 0 || quality < 0)

  1673.         return NULL;

  1674. 

  1675.     vec = sws_allocVec(length);

  1676. 

  1677.     if (!vec)

  1678.         return NULL;

  1679. 

  1680.     for (i = 0; i < length; i++) {

  1681.         double dist = i - middle;

  1682.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1683.                         sqrt(2 * variance * M_PI);

  1684.     }

  1685. 

  1686.     sws_normalizeVec(vec, 1.0);

  1687. 

  1688.     return vec;

  1689. }

  1690. 

  1691. SwsVector *sws_getConstVec(double c, int length)

  1692. {

  1693.     int i;

  1694.     SwsVector *vec = sws_allocVec(length);

  1695. 

  1696.     if (!vec)

  1697.         return NULL;

  1698. 

  1699.     for (i = 0; i < length; i++)

  1700.         vec->coeff[i] = c;

  1701. 

  1702.     return vec;

  1703. }

  1704. 

  1705. SwsVector *sws_getIdentityVec(void)

  1706. {

  1707.     return sws_getConstVec(1.0, 1);

  1708. }

  1709. 

  1710. static double sws_dcVec(SwsVector *a)

  1711. {

  1712.     int i;

  1713.     double sum = 0;

  1714. 

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

  1716.         sum += a->coeff[i];

  1717. 

  1718.     return sum;

  1719. }

  1720. 

  1721. void sws_scaleVec(SwsVector *a, double scalar)

  1722. {

  1723.     int i;

  1724. 

  1725.     for (i = 0; i < a->length; i++)

  1726.         a->coeff[i] *= scalar;

  1727. }

  1728. 

  1729. void sws_normalizeVec(SwsVector *a, double height)

  1730. {

  1731.     sws_scaleVec(a, height / sws_dcVec(a));

  1732. }

  1733. 

  1734. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1735. {

  1736.     int length = a->length + b->length - 1;

  1737.     int i, j;

  1738.     SwsVector *vec = sws_getConstVec(0.0, length);

  1739. 

  1740.     if (!vec)

  1741.         return NULL;

  1742. 

  1743.     for (i = 0; i < a->length; i++) {

  1744.         for (j = 0; j < b->length; j++) {

  1745.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1746.         }

  1747.     }

  1748. 

  1749.     return vec;

  1750. }

  1751. 

  1752. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1753. {

  1754.     int length = FFMAX(a->length, b->length);

  1755.     int i;

  1756.     SwsVector *vec = sws_getConstVec(0.0, length);

  1757. 

  1758.     if (!vec)

  1759.         return NULL;

  1760. 

  1761.     for (i = 0; i < a->length; i++)

  1762.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1763.     for (i = 0; i < b->length; i++)

  1764.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

  1765. 

  1766.     return vec;

  1767. }

  1768. 

  1769. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1770. {

  1771.     int length = FFMAX(a->length, b->length);

  1772.     int i;

  1773.     SwsVector *vec = sws_getConstVec(0.0, length);

  1774. 

  1775.     if (!vec)

  1776.         return NULL;

  1777. 

  1778.     for (i = 0; i < a->length; i++)

  1779.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1780.     for (i = 0; i < b->length; i++)

  1781.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

  1782. 

  1783.     return vec;

  1784. }

  1785. 

  1786. /* shift left / or right if "shift" is negative */

  1787. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1788. {

  1789.     int length = a->length + FFABS(shift) * 2;

  1790.     int i;

  1791.     SwsVector *vec = sws_getConstVec(0.0, length);

  1792. 

  1793.     if (!vec)

  1794.         return NULL;

  1795. 

  1796.     for (i = 0; i < a->length; i++) {

  1797.         vec->coeff[i + (length    - 1) / 2 -

  1798.                        (a->length - 1) / 2 - shift] = a->coeff[i];

  1799.     }

  1800. 

  1801.     return vec;

  1802. }

  1803. 

  1804. void sws_shiftVec(SwsVector *a, int shift)

  1805. {

  1806.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1807.     av_free(a->coeff);

  1808.     a->coeff  = shifted->coeff;

  1809.     a->length = shifted->length;

  1810.     av_free(shifted);

  1811. }

  1812. 

  1813. void sws_addVec(SwsVector *a, SwsVector *b)

  1814. {

  1815.     SwsVector *sum = sws_sumVec(a, b);

  1816.     av_free(a->coeff);

  1817.     a->coeff  = sum->coeff;

  1818.     a->length = sum->length;

  1819.     av_free(sum);

  1820. }

  1821. 

  1822. void sws_subVec(SwsVector *a, SwsVector *b)

  1823. {

  1824.     SwsVector *diff = sws_diffVec(a, b);

  1825.     av_free(a->coeff);

  1826.     a->coeff  = diff->coeff;

  1827.     a->length = diff->length;

  1828.     av_free(diff);

  1829. }

  1830. 

  1831. void sws_convVec(SwsVector *a, SwsVector *b)

  1832. {

  1833.     SwsVector *conv = sws_getConvVec(a, b);

  1834.     av_free(a->coeff);

  1835.     a->coeff  = conv->coeff;

  1836.     a->length = conv->length;

  1837.     av_free(conv);

  1838. }

  1839. 

  1840. SwsVector *sws_cloneVec(SwsVector *a)

  1841. {

  1842.     int i;

  1843.     SwsVector *vec = sws_allocVec(a->length);

  1844. 

  1845.     if (!vec)

  1846.         return NULL;

  1847. 

  1848.     for (i = 0; i < a->length; i++)

  1849.         vec->coeff[i] = a->coeff[i];

  1850. 

  1851.     return vec;

  1852. }

  1853. 

  1854. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1855. {

  1856.     int i;

  1857.     double max = 0;

  1858.     double min = 0;

  1859.     double range;

  1860. 

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

  1862.         if (a->coeff[i] > max)

  1863.             max = a->coeff[i];

  1864. 

  1865.     for (i = 0; i < a->length; i++)

  1866.         if (a->coeff[i] < min)

  1867.             min = a->coeff[i];

  1868. 

  1869.     range = max - min;

  1870. 

  1871.     for (i = 0; i < a->length; i++) {

  1872.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1873.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1874.         for (; x > 0; x--)

  1875.             av_log(log_ctx, log_level, " ");

  1876.         av_log(log_ctx, log_level, "|\n");

  1877.     }

  1878. }

  1879. 

  1880. void sws_freeVec(SwsVector *a)

  1881. {

  1882.     if (!a)

  1883.         return;

  1884.     av_freep(&a->coeff);

  1885.     a->length = 0;

  1886.     av_free(a);

  1887. }

  1888. 

  1889. void sws_freeFilter(SwsFilter *filter)

  1890. {

  1891.     if (!filter)

  1892.         return;

  1893. 

  1894.     if (filter->lumH)

  1895.         sws_freeVec(filter->lumH);

  1896.     if (filter->lumV)

  1897.         sws_freeVec(filter->lumV);

  1898.     if (filter->chrH)

  1899.         sws_freeVec(filter->chrH);

  1900.     if (filter->chrV)

  1901.         sws_freeVec(filter->chrV);

  1902.     av_free(filter);

  1903. }

  1904. 

  1905. void sws_freeContext(SwsContext *c)

  1906. {

  1907.     int i;

  1908.     if (!c)

  1909.         return;

  1910. 

  1911.     if (c->lumPixBuf) {

  1912.         for (i = 0; i < c->vLumBufSize; i++)

  1913.             av_freep(&c->lumPixBuf[i]);

  1914.         av_freep(&c->lumPixBuf);

  1915.     }

  1916. 

  1917.     if (c->chrUPixBuf) {

  1918.         for (i = 0; i < c->vChrBufSize; i++)

  1919.             av_freep(&c->chrUPixBuf[i]);

  1920.         av_freep(&c->chrUPixBuf);

  1921.         av_freep(&c->chrVPixBuf);

  1922.     }

  1923. 

  1924.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1925.         for (i = 0; i < c->vLumBufSize; i++)

  1926.             av_freep(&c->alpPixBuf[i]);

  1927.         av_freep(&c->alpPixBuf);

  1928.     }

  1929. 

  1930.     for (i = 0; i < 4; i++)

  1931.         av_freep(&c->dither_error[i]);

  1932. 

  1933.     av_freep(&c->vLumFilter);

  1934.     av_freep(&c->vChrFilter);

  1935.     av_freep(&c->hLumFilter);

  1936.     av_freep(&c->hChrFilter);

  1937. #if HAVE_ALTIVEC

  1938.     av_freep(&c->vYCoeffsBank);

  1939.     av_freep(&c->vCCoeffsBank);

  1940. #endif

  1941. 

  1942.     av_freep(&c->vLumFilterPos);

  1943.     av_freep(&c->vChrFilterPos);

  1944.     av_freep(&c->hLumFilterPos);

  1945.     av_freep(&c->hChrFilterPos);

  1946. 

  1947. #if HAVE_MMX_INLINE

  1948. #if USE_MMAP

  1949.     if (c->lumMmxextFilterCode)

  1950.         munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);

  1951.     if (c->chrMmxextFilterCode)

  1952.         munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);

  1953. #elif HAVE_VIRTUALALLOC

  1954.     if (c->lumMmxextFilterCode)

  1955.         VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);

  1956.     if (c->chrMmxextFilterCode)

  1957.         VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);

  1958. #else

  1959.     av_free(c->lumMmxextFilterCode);

  1960.     av_free(c->chrMmxextFilterCode);

  1961. #endif

  1962.     c->lumMmxextFilterCode = NULL;

  1963.     c->chrMmxextFilterCode = NULL;

  1964. #endif /* HAVE_MMX_INLINE */

  1965. 

  1966.     av_freep(&c->yuvTable);

  1967.     av_freep(&c->formatConvBuffer);

  1968. 

  1969.     av_free(c);

  1970. }

  1971. 

  1972. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  1973.                                         int srcH, enum AVPixelFormat srcFormat,

  1974.                                         int dstW, int dstH,

  1975.                                         enum AVPixelFormat dstFormat, int flags,

  1976.                                         SwsFilter *srcFilter,

  1977.                                         SwsFilter *dstFilter,

  1978.                                         const double *param)

  1979. {

  1980.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  1981.                                              SWS_PARAM_DEFAULT };

  1982. 

  1983.     if (!param)

  1984.         param = default_param;

  1985. 

  1986.     if (context &&

  1987.         (context->srcW      != srcW      ||

  1988.          context->srcH      != srcH      ||

  1989.          context->srcFormat != srcFormat ||

  1990.          context->dstW      != dstW      ||

  1991.          context->dstH      != dstH      ||

  1992.          context->dstFormat != dstFormat ||

  1993.          context->flags     != flags     ||

  1994.          context->param[0]  != param[0]  ||

  1995.          context->param[1]  != param[1])) {

  1996.         sws_freeContext(context);

  1997.         context = NULL;

  1998.     }

  1999. 

  2000.     if (!context) {

  2001.         if (!(context = sws_alloc_context()))

  2002.             return NULL;

  2003.         context->srcW      = srcW;

  2004.         context->srcH      = srcH;

  2005.         context->srcRange  = handle_jpeg(&srcFormat);

  2006.         context->src0Alpha = handle_0alpha(&srcFormat);

  2007.         context->srcXYZ    = handle_xyz(&srcFormat);

  2008.         context->srcFormat = srcFormat;

  2009.         context->dstW      = dstW;

  2010.         context->dstH      = dstH;

  2011.         context->dstRange  = handle_jpeg(&dstFormat);

  2012.         context->dst0Alpha = handle_0alpha(&dstFormat);

  2013.         context->dstXYZ    = handle_xyz(&dstFormat);

  2014.         context->dstFormat = dstFormat;

  2015.         context->flags     = flags;

  2016.         context->param[0]  = param[0];

  2017.         context->param[1]  = param[1];

  2018.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  2019.                                  context->srcRange,

  2020.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  2021.                                  context->dstRange, 0, 1 << 16, 1 << 16);

  2022.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  2023.             sws_freeContext(context);

  2024.             return NULL;

  2025.         }

  2026.     }

  2027.     return context;

  2028. }