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.
57240 Commits
32 Branches
283MB
Tree: 640c70ddc2
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '640c70ddc2'
${ noResults }
FFmpeg/libswscale/utils.c

2102 lines
76KB
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/ppc/cpu.h"

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

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

  52. #include "rgb2rgb.h"

  53. #include "swscale.h"

  54. #include "swscale_internal.h"

  55. 

  56. static void handle_formats(SwsContext *c);

  57. 

  58. unsigned swscale_version(void)

  59. {

  60.     av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);

  61.     return LIBSWSCALE_VERSION_INT;

  62. }

  63. 

  64. const char *swscale_configuration(void)

  65. {

  66.     return FFMPEG_CONFIGURATION;

  67. }

  68. 

  69. const char *swscale_license(void)

  70. {

  71. #define LICENSE_PREFIX "libswscale license: "

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

  73. }

  74. 

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

  76. 

  77. typedef struct FormatEntry {

  78.     uint8_t is_supported_in         :1;

  79.     uint8_t is_supported_out        :1;

  80.     uint8_t is_supported_endianness :1;

  81. } FormatEntry;

  82. 

  83. static const FormatEntry format_entries[AV_PIX_FMT_NB] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

  97.     [AV_PIX_FMT_YUVJ411P]    = { 1, 1 },

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  145.     [AV_PIX_FMT_RGBA64BE]    = { 1, 1 },

  146.     [AV_PIX_FMT_RGBA64LE]    = { 1, 1 },

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  205.     [AV_PIX_FMT_XYZ12BE]     = { 1, 1, 1 },

  206.     [AV_PIX_FMT_XYZ12LE]     = { 1, 1, 1 },

  207.     [AV_PIX_FMT_GBRAP]       = { 1, 1 },

  208.     [AV_PIX_FMT_GBRAP16LE]   = { 1, 0 },

  209.     [AV_PIX_FMT_GBRAP16BE]   = { 1, 0 },

  210. };

  211. 

  212. int sws_isSupportedInput(enum AVPixelFormat pix_fmt)

  213. {

  214.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  215.            format_entries[pix_fmt].is_supported_in : 0;

  216. }

  217. 

  218. int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)

  219. {

  220.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  221.            format_entries[pix_fmt].is_supported_out : 0;

  222. }

  223. 

  224. int sws_isSupportedEndiannessConversion(enum AVPixelFormat pix_fmt)

  225. {

  226.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  227.            format_entries[pix_fmt].is_supported_endianness : 0;

  228. }

  229. 

  230. #if FF_API_SWS_FORMAT_NAME

  231. const char *sws_format_name(enum AVPixelFormat format)

  232. {

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

  234.     if (desc)

  235.         return desc->name;

  236.     else

  237.         return "Unknown format";

  238. }

  239. #endif

  240. 

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

  242.                              double dist)

  243. {

  244.     if (dist <= 1.0)

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

  246.     else

  247.         return getSplineCoeff(0.0,

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

  249.                                c + 3.0 * d,

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

  251.                               dist - 1.0);

  252. }

  253. 

  254. static av_cold int get_local_pos(SwsContext *s, int chr_subsample, int pos, int dir)

  255. {

  256.     if (pos < 0) {

  257.         pos = (128 << chr_subsample) - 128;

  258.     }

  259.     pos += 128; // relative to ideal left edge

  260.     return pos >> chr_subsample;

  261. }

  262. 

  263. static av_cold int initFilter(int16_t **outFilter, int32_t **filterPos,

  264.                               int *outFilterSize, int xInc, int srcW,

  265.                               int dstW, int filterAlign, int one,

  266.                               int flags, int cpu_flags,

  267.                               SwsVector *srcFilter, SwsVector *dstFilter,

  268.                               double param[2], int srcPos, int dstPos)

  269. {

  270.     int i;

  271.     int filterSize;

  272.     int filter2Size;

  273.     int minFilterSize;

  274.     int64_t *filter    = NULL;

  275.     int64_t *filter2   = NULL;

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

  277.     int ret            = -1;

  278. 

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

  280. 

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

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

  283. 

  284.     if (FFABS(xInc - 0x10000) < 10 && srcPos == dstPos) { // unscaled

  285.         int i;

  286.         filterSize = 1;

  287.         FF_ALLOCZ_OR_GOTO(NULL, filter,

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

  289. 

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

  291.             filter[i * filterSize] = fone;

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

  293.         }

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

  295.         int i;

  296.         int64_t xDstInSrc;

  297.         filterSize = 1;

  298.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  300. 

  301.         xDstInSrc = ((dstPos*(int64_t)xInc)>>8) - ((srcPos*0x8000LL)>>7);

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

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

  304. 

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

  306.             filter[i]       = fone;

  307.             xDstInSrc      += xInc;

  308.         }

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

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

  311.         int i;

  312.         int64_t xDstInSrc;

  313.         filterSize = 2;

  314.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  316. 

  317.         xDstInSrc = ((dstPos*(int64_t)xInc)>>8) - ((srcPos*0x8000LL)>>7);

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

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

  320.             int j;

  321. 

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

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

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

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

  326.                 if (coeff < 0)

  327.                     coeff = 0;

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

  329.                 xx++;

  330.             }

  331.             xDstInSrc += xInc;

  332.         }

  333.     } else {

  334.         int64_t xDstInSrc;

  335.         int sizeFactor;

  336. 

  337.         if (flags & SWS_BICUBIC)

  338.             sizeFactor = 4;

  339.         else if (flags & SWS_X)

  340.             sizeFactor = 8;

  341.         else if (flags & SWS_AREA)

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

  343.         else if (flags & SWS_GAUSS)

  344.             sizeFactor = 8;     // infinite ;)

  345.         else if (flags & SWS_LANCZOS)

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

  347.         else if (flags & SWS_SINC)

  348.             sizeFactor = 20;    // infinite ;)

  349.         else if (flags & SWS_SPLINE)

  350.             sizeFactor = 20;    // infinite ;)

  351.         else if (flags & SWS_BILINEAR)

  352.             sizeFactor = 2;

  353.         else {

  354.             av_assert0(0);

  355.         }

  356. 

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

  358.             filterSize = 1 + sizeFactor;    // upscale

  359.         else

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

  361. 

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

  363.         filterSize = FFMAX(filterSize, 1);

  364. 

  365.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  367. 

  368.         xDstInSrc = ((dstPos*(int64_t)xInc)>>7) - ((srcPos*0x10000LL)>>7);

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

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

  371.             int j;

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

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

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

  375.                 double floatd;

  376.                 int64_t coeff;

  377. 

  378.                 if (xInc > 1 << 16)

  379.                     d = d * dstW / srcW;

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

  381. 

  382.                 if (flags & SWS_BICUBIC) {

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

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

  385. 

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

  387.                         coeff = 0.0;

  388.                     } else {

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

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

  391. 

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

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

  394.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  395.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  396.                         else

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

  398.                                       (6 * B + 30 * C) * dd  +

  399.                                     (-12 * B - 48 * C) * d   +

  400.                                       (8 * B + 24 * C) * (1 << 30);

  401.                     }

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

  403.                 }

  404. #if 0

  405.                 else if (flags & SWS_X) {

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

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

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

  409.                 }

  410. #endif

  411.                 else if (flags & SWS_X) {

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

  413.                     double c;

  414. 

  415.                     if (floatd < 1.0)

  416.                         c = cos(floatd * M_PI);

  417.                     else

  418.                         c = -1.0;

  419.                     if (c < 0.0)

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

  421.                     else

  422.                         c = pow(c, A);

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

  424.                 } else if (flags & SWS_AREA) {

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

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

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

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

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

  430.                     else

  431.                         coeff = 0.0;

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

  433.                 } else if (flags & SWS_GAUSS) {

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

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

  436.                 } else if (flags & SWS_SINC) {

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

  438.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  442.                     if (floatd > p)

  443.                         coeff = 0;

  444.                 } else if (flags & SWS_BILINEAR) {

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

  446.                     if (coeff < 0)

  447.                         coeff = 0;

  448.                     coeff *= fone >> 30;

  449.                 } else if (flags & SWS_SPLINE) {

  450.                     double p = -2.196152422706632;

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

  452.                 } else {

  453.                     av_assert0(0);

  454.                 }

  455. 

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

  457.                 xx++;

  458.             }

  459.             xDstInSrc += 2 * xInc;

  460.         }

  461.     }

  462. 

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

  464.      * av_free(filter);

  465.      */

  466.     av_assert0(filterSize > 0);

  467.     filter2Size = filterSize;

  468.     if (srcFilter)

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

  470.     if (dstFilter)

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

  472.     av_assert0(filter2Size > 0);

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

  474. 

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

  476.         int j, k;

  477. 

  478.         if (srcFilter) {

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

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

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

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

  483.             }

  484.         } else {

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

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

  487.         }

  488.         // FIXME dstFilter

  489. 

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

  491.     }

  492.     av_freep(&filter);

  493. 

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

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

  496.     minFilterSize = 0;

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

  498.         int min = filter2Size;

  499.         int j;

  500.         int64_t cutOff = 0.0;

  501. 

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

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

  504.             int k;

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

  506. 

  507.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  508.                 break;

  509. 

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

  511.              * filter otherwise */

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

  513.                 break;

  514. 

  515.             // move filter coefficients left

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

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

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

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

  520.         }

  521. 

  522.         cutOff = 0;

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

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

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

  526. 

  527.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  528.                 break;

  529.             min--;

  530.         }

  531. 

  532.         if (min > minFilterSize)

  533.             minFilterSize = min;

  534.     }

  535. 

  536.     if (PPC_ALTIVEC(cpu_flags)) {

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

  538.         if (minFilterSize < 5)

  539.             filterAlign = 4;

  540. 

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

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

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

  544.         if (minFilterSize < 3)

  545.             filterAlign = 1;

  546.     }

  547. 

  548.     if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {

  549.         // special case for unscaled vertical filtering

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

  551.             filterAlign = 1;

  552.     }

  553. 

  554.     av_assert0(minFilterSize > 0);

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

  556.     av_assert0(filterSize > 0);

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

  558.     if (filterSize >= MAX_FILTER_SIZE * 16 /

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

  560.         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);

  561.         goto fail;

  562.     }

  563.     *outFilterSize = filterSize;

  564. 

  565.     if (flags & SWS_PRINT_INFO)

  566.         av_log(NULL, AV_LOG_VERBOSE,

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

  568.                filter2Size, filterSize);

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

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

  571.         int j;

  572. 

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

  574.             if (j >= filter2Size)

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

  576.             else

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

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

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

  580.         }

  581.     }

  582. 

  583.     // FIXME try to align filterPos if possible

  584. 

  585.     // fix borders

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

  587.         int j;

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

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

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

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

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

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

  594.             }

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

  596.         }

  597. 

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

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

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

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

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

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

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

  605.             }

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

  607.         }

  608.     }

  609. 

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

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

  612.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

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

  614. 

  615.     /* normalize & store in outFilter */

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

  617.         int j;

  618.         int64_t error = 0;

  619.         int64_t sum   = 0;

  620. 

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

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

  623.         }

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

  625.         if (!sum) {

  626.             av_log(NULL, AV_LOG_WARNING, "SwScaler: zero vector in scaling\n");

  627.             sum = 1;

  628.         }

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

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

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

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

  633.             error                                  = v - intV * sum;

  634.         }

  635.     }

  636. 

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

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

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

  640.                                                       * read over the end */

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

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

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

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

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

  646.     }

  647. 

  648.     ret = 0;

  649. 

  650. fail:

  651.     if(ret < 0)

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

  653.     av_free(filter);

  654.     av_free(filter2);

  655.     return ret;

  656. }

  657. 

  658. #if HAVE_MMXEXT_INLINE

  659. static av_cold int init_hscaler_mmxext(int dstW, int xInc, uint8_t *filterCode,

  660.                                        int16_t *filter, int32_t *filterPos,

  661.                                        int numSplits)

  662. {

  663.     uint8_t *fragmentA;

  664.     x86_reg imm8OfPShufW1A;

  665.     x86_reg imm8OfPShufW2A;

  666.     x86_reg fragmentLengthA;

  667.     uint8_t *fragmentB;

  668.     x86_reg imm8OfPShufW1B;

  669.     x86_reg imm8OfPShufW2B;

  670.     x86_reg fragmentLengthB;

  671.     int fragmentPos;

  672. 

  673.     int xpos, i;

  674. 

  675.     // create an optimized horizontal scaling routine

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

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

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

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

  680.      */

  681. 

  682.     // code fragment

  683. 

  684.     __asm__ volatile (

  685.         "jmp                         9f                 \n\t"

  686.         // Begin

  687.         "0:                                             \n\t"

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

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

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

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

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

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

  694.         "1:                                             \n\t"

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

  696.         "2:                                             \n\t"

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

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

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

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

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

  702. 

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

  704. 

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

  706.         // End

  707.         "9:                                             \n\t"

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

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

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

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

  712.         "dec                         %1                 \n\t"

  713.         "dec                         %2                 \n\t"

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

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

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

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

  718. 

  719. 

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

  721.           "=r" (fragmentLengthA)

  722.         );

  723. 

  724.     __asm__ volatile (

  725.         "jmp                         9f                 \n\t"

  726.         // Begin

  727.         "0:                                             \n\t"

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

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

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

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

  732.         "1:                                             \n\t"

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

  734.         "2:                                             \n\t"

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

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

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

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

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

  740. 

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

  742. 

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

  744.         // End

  745.         "9:                                             \n\t"

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

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

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

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

  750.         "dec                         %1                 \n\t"

  751.         "dec                         %2                 \n\t"

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

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

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

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

  756. 

  757. 

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

  759.           "=r" (fragmentLengthB)

  760.         );

  761. 

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

  763.     fragmentPos = 0;

  764. 

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

  766.         int xx = xpos >> 16;

  767. 

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

  769.             int a                  = 0;

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

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

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

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

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

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

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

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

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

  779.             int shift              = 0;

  780. 

  781.             if (filterCode) {

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

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

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

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

  786.                 filterPos[i / 2] = xx;

  787. 

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

  789. 

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

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

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

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

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

  795.                                                                (c << 4) |

  796.                                                                (d << 6);

  797. 

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

  799.                     shift = maxShift;               // avoid overread

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

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

  802. 

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

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

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

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

  807.                 }

  808.             }

  809. 

  810.             fragmentPos += fragmentLength;

  811. 

  812.             if (filterCode)

  813.                 filterCode[fragmentPos] = RET;

  814.         }

  815.         xpos += xInc;

  816.     }

  817.     if (filterCode)

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

  819. 

  820.     return fragmentPos + 1;

  821. }

  822. #endif /* HAVE_MMXEXT_INLINE */

  823. 

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

  825. {

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

  827.     *h = desc->log2_chroma_w;

  828.     *v = desc->log2_chroma_h;

  829. }

  830. 

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

  832. {

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

  834.     int64_t vr =  table[0];

  835.     int64_t ub =  table[1];

  836.     int64_t ug = -table[2];

  837.     int64_t vg = -table[3];

  838.     int64_t ONE = 65536;

  839.     int64_t cy = ONE;

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

  841.     int i;

  842.     static const int8_t map[] = {

  843.     BY_IDX, GY_IDX, -1    , BY_IDX, BY_IDX, GY_IDX, -1    , BY_IDX,

  844.     RY_IDX, -1    , GY_IDX, RY_IDX, RY_IDX, -1    , GY_IDX, RY_IDX,

  845.     RY_IDX, GY_IDX, -1    , RY_IDX, RY_IDX, GY_IDX, -1    , RY_IDX,

  846.     BY_IDX, -1    , GY_IDX, BY_IDX, BY_IDX, -1    , GY_IDX, BY_IDX,

  847.     BU_IDX, GU_IDX, -1    , BU_IDX, BU_IDX, GU_IDX, -1    , BU_IDX,

  848.     RU_IDX, -1    , GU_IDX, RU_IDX, RU_IDX, -1    , GU_IDX, RU_IDX,

  849.     RU_IDX, GU_IDX, -1    , RU_IDX, RU_IDX, GU_IDX, -1    , RU_IDX,

  850.     BU_IDX, -1    , GU_IDX, BU_IDX, BU_IDX, -1    , GU_IDX, BU_IDX,

  851.     BV_IDX, GV_IDX, -1    , BV_IDX, BV_IDX, GV_IDX, -1    , BV_IDX,

  852.     RV_IDX, -1    , GV_IDX, RV_IDX, RV_IDX, -1    , GV_IDX, RV_IDX,

  853.     RV_IDX, GV_IDX, -1    , RV_IDX, RV_IDX, GV_IDX, -1    , RV_IDX,

  854.     BV_IDX, -1    , GV_IDX, BV_IDX, BV_IDX, -1    , GV_IDX, BV_IDX,

  855.     RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,

  856.     BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,

  857.     GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    ,

  858.     -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX,

  859.     RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,

  860.     BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,

  861.     GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    ,

  862.     -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX,

  863.     RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,

  864.     BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,

  865.     GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    ,

  866.     -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, //23

  867.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //24

  868.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //25

  869.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //26

  870.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //27

  871.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //28

  872.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //29

  873.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //30

  874.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //31

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

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

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

  878.     };

  879. 

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

  881. 

  882.     if (!dstRange) {

  883.         cy = cy * 255 / 219;

  884.     } else {

  885.         vr = vr * 224 / 255;

  886.         ub = ub * 224 / 255;

  887.         ug = ug * 224 / 255;

  888.         vg = vg * 224 / 255;

  889.     }

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

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

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

  893. 

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

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

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

  897. 

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

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

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

  901. 

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

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

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

  905. 

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

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

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

  909. 

  910.     if(/*!dstRange && */!memcmp(table, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], sizeof(ff_yuv2rgb_coeffs[SWS_CS_DEFAULT]))) {

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

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

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

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

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

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

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

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

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

  920.     }

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

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

  923. }

  924. 

  925. static void fill_xyztables(struct SwsContext *c)

  926. {

  927.     int i;

  928.     double xyzgamma = XYZ_GAMMA;

  929.     double rgbgamma = 1.0 / RGB_GAMMA;

  930.     double xyzgammainv = 1.0 / XYZ_GAMMA;

  931.     double rgbgammainv = RGB_GAMMA;

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

  933.         {13270, -6295, -2041},

  934.         {-3969,  7682,   170},

  935.         {  228,  -835,  4329} };

  936.     static const int16_t rgb2xyz_matrix[3][4] = {

  937.         {1689, 1464,  739},

  938.         { 871, 2929,  296},

  939.         {  79,  488, 3891} };

  940.     static int16_t xyzgamma_tab[4096], rgbgamma_tab[4096], xyzgammainv_tab[4096], rgbgammainv_tab[4096];

  941. 

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

  943.     memcpy(c->rgb2xyz_matrix, rgb2xyz_matrix, sizeof(c->rgb2xyz_matrix));

  944.     c->xyzgamma = xyzgamma_tab;

  945.     c->rgbgamma = rgbgamma_tab;

  946.     c->xyzgammainv = xyzgammainv_tab;

  947.     c->rgbgammainv = rgbgammainv_tab;

  948. 

  949.     if (rgbgamma_tab[4095])

  950.         return;

  951. 

  952.     /* set gamma vectors */

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

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

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

  956.         xyzgammainv_tab[i] = lrint(pow(i / 4095.0, xyzgammainv) * 4095.0);

  957.         rgbgammainv_tab[i] = lrint(pow(i / 4095.0, rgbgammainv) * 4095.0);

  958.     }

  959. }

  960. 

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

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

  963.                              int brightness, int contrast, int saturation)

  964. {

  965.     const AVPixFmtDescriptor *desc_dst;

  966.     const AVPixFmtDescriptor *desc_src;

  967.     memmove(c->srcColorspaceTable, inv_table, sizeof(int) * 4);

  968.     memmove(c->dstColorspaceTable, table, sizeof(int) * 4);

  969. 

  970.     handle_formats(c);

  971.     desc_dst = av_pix_fmt_desc_get(c->dstFormat);

  972.     desc_src = av_pix_fmt_desc_get(c->srcFormat);

  973. 

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

  975.         dstRange = 0;

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

  977.         srcRange = 0;

  978. 

  979.     c->brightness = brightness;

  980.     c->contrast   = contrast;

  981.     c->saturation = saturation;

  982.     c->srcRange   = srcRange;

  983.     c->dstRange   = dstRange;

  984. 

  985.     fill_xyztables(c);

  986. 

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

  988.         return -1;

  989. 

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

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

  992. 

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

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

  995.                                  contrast, saturation);

  996.         // FIXME factorize

  997. 

  998.         if (ARCH_PPC)

  999.             ff_yuv2rgb_init_tables_ppc(c, inv_table, brightness,

  1000.                                        contrast, saturation);

  1001.     }

  1002. 

  1003.     fill_rgb2yuv_table(c, table, dstRange);

  1004. 

  1005.     return 0;

  1006. }

  1007. 

  1008. int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,

  1009.                              int *srcRange, int **table, int *dstRange,

  1010.                              int *brightness, int *contrast, int *saturation)

  1011. {

  1012.     if (!c )

  1013.         return -1;

  1014. 

  1015.     *inv_table  = c->srcColorspaceTable;

  1016.     *table      = c->dstColorspaceTable;

  1017.     *srcRange   = c->srcRange;

  1018.     *dstRange   = c->dstRange;

  1019.     *brightness = c->brightness;

  1020.     *contrast   = c->contrast;

  1021.     *saturation = c->saturation;

  1022. 

  1023.     return 0;

  1024. }

  1025. 

  1026. static int handle_jpeg(enum AVPixelFormat *format)

  1027. {

  1028.     switch (*format) {

  1029.     case AV_PIX_FMT_YUVJ420P:

  1030.         *format = AV_PIX_FMT_YUV420P;

  1031.         return 1;

  1032.     case AV_PIX_FMT_YUVJ411P:

  1033.         *format = AV_PIX_FMT_YUV411P;

  1034.         return 1;

  1035.     case AV_PIX_FMT_YUVJ422P:

  1036.         *format = AV_PIX_FMT_YUV422P;

  1037.         return 1;

  1038.     case AV_PIX_FMT_YUVJ444P:

  1039.         *format = AV_PIX_FMT_YUV444P;

  1040.         return 1;

  1041.     case AV_PIX_FMT_YUVJ440P:

  1042.         *format = AV_PIX_FMT_YUV440P;

  1043.         return 1;

  1044.     case AV_PIX_FMT_GRAY8:

  1045.         return 1;

  1046.     default:

  1047.         return 0;

  1048.     }

  1049. }

  1050. 

  1051. static int handle_0alpha(enum AVPixelFormat *format)

  1052. {

  1053.     switch (*format) {

  1054.     case AV_PIX_FMT_0BGR    : *format = AV_PIX_FMT_ABGR   ; return 1;

  1055.     case AV_PIX_FMT_BGR0    : *format = AV_PIX_FMT_BGRA   ; return 4;

  1056.     case AV_PIX_FMT_0RGB    : *format = AV_PIX_FMT_ARGB   ; return 1;

  1057.     case AV_PIX_FMT_RGB0    : *format = AV_PIX_FMT_RGBA   ; return 4;

  1058.     default:                                          return 0;

  1059.     }

  1060. }

  1061. 

  1062. static int handle_xyz(enum AVPixelFormat *format)

  1063. {

  1064.     switch (*format) {

  1065.     case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;

  1066.     case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;

  1067.     default:                                                return 0;

  1068.     }

  1069. }

  1070. 

  1071. static void handle_formats(SwsContext *c)

  1072. {

  1073.     c->src0Alpha |= handle_0alpha(&c->srcFormat);

  1074.     c->dst0Alpha |= handle_0alpha(&c->dstFormat);

  1075.     c->srcXYZ    |= handle_xyz(&c->srcFormat);

  1076.     c->dstXYZ    |= handle_xyz(&c->dstFormat);

  1077. }

  1078. 

  1079. SwsContext *sws_alloc_context(void)

  1080. {

  1081.     SwsContext *c = av_mallocz(sizeof(SwsContext));

  1082. 

  1083.     if (c) {

  1084.         c->av_class = &sws_context_class;

  1085.         av_opt_set_defaults(c);

  1086.     }

  1087. 

  1088.     return c;

  1089. }

  1090. 

  1091. av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,

  1092.                              SwsFilter *dstFilter)

  1093. {

  1094.     int i, j;

  1095.     int usesVFilter, usesHFilter;

  1096.     int unscaled;

  1097.     SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };

  1098.     int srcW              = c->srcW;

  1099.     int srcH              = c->srcH;

  1100.     int dstW              = c->dstW;

  1101.     int dstH              = c->dstH;

  1102.     int dst_stride        = FFALIGN(dstW * sizeof(int16_t) + 66, 16);

  1103.     int flags, cpu_flags;

  1104.     enum AVPixelFormat srcFormat = c->srcFormat;

  1105.     enum AVPixelFormat dstFormat = c->dstFormat;

  1106.     const AVPixFmtDescriptor *desc_src;

  1107.     const AVPixFmtDescriptor *desc_dst;

  1108. 

  1109.     cpu_flags = av_get_cpu_flags();

  1110.     flags     = c->flags;

  1111.     emms_c();

  1112.     if (!rgb15to16)

  1113.         sws_rgb2rgb_init();

  1114. 

  1115.     unscaled = (srcW == dstW && srcH == dstH);

  1116. 

  1117.     c->srcRange |= handle_jpeg(&c->srcFormat);

  1118.     c->dstRange |= handle_jpeg(&c->dstFormat);

  1119. 

  1120.     if (!c->contrast && !c->saturation && !c->dstFormatBpp)

  1121.         sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1122.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  1123.                                  c->dstRange, 0, 1 << 16, 1 << 16);

  1124. 

  1125.     if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat)

  1126.         av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");

  1127.     handle_formats(c);

  1128.     srcFormat = c->srcFormat;

  1129.     dstFormat = c->dstFormat;

  1130.     desc_src = av_pix_fmt_desc_get(srcFormat);

  1131.     desc_dst = av_pix_fmt_desc_get(dstFormat);

  1132. 

  1133.     if (!(unscaled && sws_isSupportedEndiannessConversion(srcFormat) &&

  1134.           av_pix_fmt_swap_endianness(srcFormat) == dstFormat)) {

  1135.     if (!sws_isSupportedInput(srcFormat)) {

  1136.         av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",

  1137.                av_get_pix_fmt_name(srcFormat));

  1138.         return AVERROR(EINVAL);

  1139.     }

  1140.     if (!sws_isSupportedOutput(dstFormat)) {

  1141.         av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",

  1142.                av_get_pix_fmt_name(dstFormat));

  1143.         return AVERROR(EINVAL);

  1144.     }

  1145.     }

  1146. 

  1147.     i = flags & (SWS_POINT         |

  1148.                  SWS_AREA          |

  1149.                  SWS_BILINEAR      |

  1150.                  SWS_FAST_BILINEAR |

  1151.                  SWS_BICUBIC       |

  1152.                  SWS_X             |

  1153.                  SWS_GAUSS         |

  1154.                  SWS_LANCZOS       |

  1155.                  SWS_SINC          |

  1156.                  SWS_SPLINE        |

  1157.                  SWS_BICUBLIN);

  1158. 

  1159.     /* provide a default scaler if not set by caller */

  1160.     if (!i) {

  1161.         if (dstW < srcW && dstH < srcH)

  1162.             flags |= SWS_BICUBIC;

  1163.         else if (dstW > srcW && dstH > srcH)

  1164.             flags |= SWS_BICUBIC;

  1165.         else

  1166.             flags |= SWS_BICUBIC;

  1167.         c->flags = flags;

  1168.     } else if (i & (i - 1)) {

  1169.         av_log(c, AV_LOG_ERROR,

  1170.                "Exactly one scaler algorithm must be chosen, got %X\n", i);

  1171.         return AVERROR(EINVAL);

  1172.     }

  1173.     /* sanity check */

  1174.     if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {

  1175.         /* FIXME check if these are enough and try to lower them after

  1176.          * fixing the relevant parts of the code */

  1177.         av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",

  1178.                srcW, srcH, dstW, dstH);

  1179.         return AVERROR(EINVAL);

  1180.     }

  1181. 

  1182.     if (!dstFilter)

  1183.         dstFilter = &dummyFilter;

  1184.     if (!srcFilter)

  1185.         srcFilter = &dummyFilter;

  1186. 

  1187.     c->lumXInc      = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;

  1188.     c->lumYInc      = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;

  1189.     c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);

  1190.     c->srcFormatBpp = av_get_bits_per_pixel(desc_src);

  1191.     c->vRounder     = 4 * 0x0001000100010001ULL;

  1192. 

  1193.     usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||

  1194.                   (srcFilter->chrV && srcFilter->chrV->length > 1) ||

  1195.                   (dstFilter->lumV && dstFilter->lumV->length > 1) ||

  1196.                   (dstFilter->chrV && dstFilter->chrV->length > 1);

  1197.     usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||

  1198.                   (srcFilter->chrH && srcFilter->chrH->length > 1) ||

  1199.                   (dstFilter->lumH && dstFilter->lumH->length > 1) ||

  1200.                   (dstFilter->chrH && dstFilter->chrH->length > 1);

  1201. 

  1202.     getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);

  1203.     getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);

  1204. 

  1205.     if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {

  1206.         if (dstW&1) {

  1207.             av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");

  1208.             flags |= SWS_FULL_CHR_H_INT;

  1209.             c->flags = flags;

  1210.         }

  1211. 

  1212.         if (   c->chrSrcHSubSample == 0

  1213.             && c->chrSrcVSubSample == 0

  1214.             && c->dither != SWS_DITHER_BAYER //SWS_FULL_CHR_H_INT is currently not supported with SWS_DITHER_BAYER

  1215.             && !(c->flags & SWS_FAST_BILINEAR)

  1216.         ) {

  1217.             av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to input having non subsampled chroma\n");

  1218.             flags |= SWS_FULL_CHR_H_INT;

  1219.             c->flags = flags;

  1220.         }

  1221.     }

  1222. 

  1223.     if (c->dither == SWS_DITHER_AUTO) {

  1224.         if (flags & SWS_ERROR_DIFFUSION)

  1225.             c->dither = SWS_DITHER_ED;

  1226.     }

  1227. 

  1228.     if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||

  1229.        dstFormat == AV_PIX_FMT_RGB4_BYTE ||

  1230.        dstFormat == AV_PIX_FMT_BGR8 ||

  1231.        dstFormat == AV_PIX_FMT_RGB8) {

  1232.         if (c->dither == SWS_DITHER_AUTO)

  1233.             c->dither = (flags & SWS_FULL_CHR_H_INT) ? SWS_DITHER_ED : SWS_DITHER_BAYER;

  1234.         if (!(flags & SWS_FULL_CHR_H_INT)) {

  1235.             if (c->dither == SWS_DITHER_ED) {

  1236.                 av_log(c, AV_LOG_DEBUG,

  1237.                     "Desired dithering only supported in full chroma interpolation for destination format '%s'\n",

  1238.                     av_get_pix_fmt_name(dstFormat));

  1239.                 flags   |= SWS_FULL_CHR_H_INT;

  1240.                 c->flags = flags;

  1241.             }

  1242.         }

  1243.         if (flags & SWS_FULL_CHR_H_INT) {

  1244.             if (c->dither == SWS_DITHER_BAYER) {

  1245.                 av_log(c, AV_LOG_DEBUG,

  1246.                     "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",

  1247.                     av_get_pix_fmt_name(dstFormat));

  1248.                 c->dither = SWS_DITHER_ED;

  1249.             }

  1250.         }

  1251.     }

  1252.     if (isPlanarRGB(dstFormat)) {

  1253.         if (!(flags & SWS_FULL_CHR_H_INT)) {

  1254.             av_log(c, AV_LOG_DEBUG,

  1255.                    "%s output is not supported with half chroma resolution, switching to full\n",

  1256.                    av_get_pix_fmt_name(dstFormat));

  1257.             flags   |= SWS_FULL_CHR_H_INT;

  1258.             c->flags = flags;

  1259.         }

  1260.     }

  1261. 

  1262.     /* reuse chroma for 2 pixels RGB/BGR unless user wants full

  1263.      * chroma interpolation */

  1264.     if (flags & SWS_FULL_CHR_H_INT &&

  1265.         isAnyRGB(dstFormat)        &&

  1266.         !isPlanarRGB(dstFormat)    &&

  1267.         dstFormat != AV_PIX_FMT_RGBA  &&

  1268.         dstFormat != AV_PIX_FMT_ARGB  &&

  1269.         dstFormat != AV_PIX_FMT_BGRA  &&

  1270.         dstFormat != AV_PIX_FMT_ABGR  &&

  1271.         dstFormat != AV_PIX_FMT_RGB24 &&

  1272.         dstFormat != AV_PIX_FMT_BGR24 &&

  1273.         dstFormat != AV_PIX_FMT_BGR4_BYTE &&

  1274.         dstFormat != AV_PIX_FMT_RGB4_BYTE &&

  1275.         dstFormat != AV_PIX_FMT_BGR8 &&

  1276.         dstFormat != AV_PIX_FMT_RGB8

  1277.     ) {

  1278.         av_log(c, AV_LOG_WARNING,

  1279.                "full chroma interpolation for destination format '%s' not yet implemented\n",

  1280.                av_get_pix_fmt_name(dstFormat));

  1281.         flags   &= ~SWS_FULL_CHR_H_INT;

  1282.         c->flags = flags;

  1283.     }

  1284.     if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))

  1285.         c->chrDstHSubSample = 1;

  1286. 

  1287.     // drop some chroma lines if the user wants it

  1288.     c->vChrDrop          = (flags & SWS_SRC_V_CHR_DROP_MASK) >>

  1289.                            SWS_SRC_V_CHR_DROP_SHIFT;

  1290.     c->chrSrcVSubSample += c->vChrDrop;

  1291. 

  1292.     /* drop every other pixel for chroma calculation unless user

  1293.      * wants full chroma */

  1294.     if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP)   &&

  1295.         srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&

  1296.         srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&

  1297.         srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&

  1298.         srcFormat != AV_PIX_FMT_GBRP9BE   && srcFormat != AV_PIX_FMT_GBRP9LE  &&

  1299.         srcFormat != AV_PIX_FMT_GBRP10BE  && srcFormat != AV_PIX_FMT_GBRP10LE &&

  1300.         srcFormat != AV_PIX_FMT_GBRP12BE  && srcFormat != AV_PIX_FMT_GBRP12LE &&

  1301.         srcFormat != AV_PIX_FMT_GBRP14BE  && srcFormat != AV_PIX_FMT_GBRP14LE &&

  1302.         srcFormat != AV_PIX_FMT_GBRP16BE  && srcFormat != AV_PIX_FMT_GBRP16LE &&

  1303.         ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||

  1304.          (flags & SWS_FAST_BILINEAR)))

  1305.         c->chrSrcHSubSample = 1;

  1306. 

  1307.     // Note the FF_CEIL_RSHIFT is so that we always round toward +inf.

  1308.     c->chrSrcW = FF_CEIL_RSHIFT(srcW, c->chrSrcHSubSample);

  1309.     c->chrSrcH = FF_CEIL_RSHIFT(srcH, c->chrSrcVSubSample);

  1310.     c->chrDstW = FF_CEIL_RSHIFT(dstW, c->chrDstHSubSample);

  1311.     c->chrDstH = FF_CEIL_RSHIFT(dstH, c->chrDstVSubSample);

  1312. 

  1313.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);

  1314. 

  1315.     /* unscaled special cases */

  1316.     if (unscaled && !usesHFilter && !usesVFilter &&

  1317.         (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {

  1318.         ff_get_unscaled_swscale(c);

  1319. 

  1320.         if (c->swscale) {

  1321.             if (flags & SWS_PRINT_INFO)

  1322.                 av_log(c, AV_LOG_INFO,

  1323.                        "using unscaled %s -> %s special converter\n",

  1324.                        av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));

  1325.             return 0;

  1326.         }

  1327.     }

  1328. 

  1329.     c->srcBpc = 1 + desc_src->comp[0].depth_minus1;

  1330.     if (c->srcBpc < 8)

  1331.         c->srcBpc = 8;

  1332.     c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;

  1333.     if (c->dstBpc < 8)

  1334.         c->dstBpc = 8;

  1335.     if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)

  1336.         c->srcBpc = 16;

  1337.     if (c->dstBpc == 16)

  1338.         dst_stride <<= 1;

  1339. 

  1340.     if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {

  1341.         c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&

  1342.                               (srcW & 15) == 0) ? 1 : 0;

  1343.         if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0

  1344. 

  1345.             && (flags & SWS_FAST_BILINEAR)) {

  1346.             if (flags & SWS_PRINT_INFO)

  1347.                 av_log(c, AV_LOG_INFO,

  1348.                        "output width is not a multiple of 32 -> no MMXEXT scaler\n");

  1349.         }

  1350.         if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))

  1351.             c->canMMXEXTBeUsed = 0;

  1352.     } else

  1353.         c->canMMXEXTBeUsed = 0;

  1354. 

  1355.     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;

  1356.     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

  1357. 

  1358.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  1359.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  1360.      * correct scaling.

  1361.      * n-2 is the last chrominance sample available.

  1362.      * This is not perfect, but no one should notice the difference, the more

  1363.      * correct variant would be like the vertical one, but that would require

  1364.      * some special code for the first and last pixel */

  1365.     if (flags & SWS_FAST_BILINEAR) {

  1366.         if (c->canMMXEXTBeUsed) {

  1367.             c->lumXInc += 20;

  1368.             c->chrXInc += 20;

  1369.         }

  1370.         // we don't use the x86 asm scaler if MMX is available

  1371.         else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {

  1372.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1373.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1374.         }

  1375.     }

  1376. 

  1377. #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)

  1378. 

  1379.     /* precalculate horizontal scaler filter coefficients */

  1380.     {

  1381. #if HAVE_MMXEXT_INLINE

  1382. // can't downscale !!!

  1383.         if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {

  1384.             c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,

  1385.                                                              NULL, NULL, 8);

  1386.             c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,

  1387.                                                              NULL, NULL, NULL, 4);

  1388. 

  1389. #if USE_MMAP

  1390.             c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,

  1391.                                           PROT_READ | PROT_WRITE,

  1392.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1393.                                           -1, 0);

  1394.             c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,

  1395.                                           PROT_READ | PROT_WRITE,

  1396.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1397.                                           -1, 0);

  1398. #elif HAVE_VIRTUALALLOC

  1399.             c->lumMmxextFilterCode = VirtualAlloc(NULL,

  1400.                                                   c->lumMmxextFilterCodeSize,

  1401.                                                   MEM_COMMIT,

  1402.                                                   PAGE_EXECUTE_READWRITE);

  1403.             c->chrMmxextFilterCode = VirtualAlloc(NULL,

  1404.                                                   c->chrMmxextFilterCodeSize,

  1405.                                                   MEM_COMMIT,

  1406.                                                   PAGE_EXECUTE_READWRITE);

  1407. #else

  1408.             c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);

  1409.             c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);

  1410. #endif

  1411. 

  1412. #ifdef MAP_ANONYMOUS

  1413.             if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)

  1414. #else

  1415.             if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)

  1416. #endif

  1417.             {

  1418.                 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");

  1419.                 return AVERROR(ENOMEM);

  1420.             }

  1421. 

  1422.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1423.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1424.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1425.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1426. 

  1427.             init_hscaler_mmxext(      dstW, c->lumXInc, c->lumMmxextFilterCode,

  1428.                                 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);

  1429.             init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,

  1430.                                 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);

  1431. 

  1432. #if USE_MMAP

  1433.             mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1434.             mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1435. #endif

  1436.         } else

  1437. #endif /* HAVE_MMXEXT_INLINE */

  1438.         {

  1439.             const int filterAlign = X86_MMX(cpu_flags)     ? 4 :

  1440.                                     PPC_ALTIVEC(cpu_flags) ? 8 : 1;

  1441. 

  1442.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1443.                            &c->hLumFilterSize, c->lumXInc,

  1444.                            srcW, dstW, filterAlign, 1 << 14,

  1445.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1446.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1447.                            c->param,

  1448.                            get_local_pos(c, 0, 0, 0),

  1449.                            get_local_pos(c, 0, 0, 0)) < 0)

  1450.                 goto fail;

  1451.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1452.                            &c->hChrFilterSize, c->chrXInc,

  1453.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1454.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1455.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1456.                            c->param,

  1457.                            get_local_pos(c, c->chrSrcHSubSample, c->src_h_chr_pos, 0),

  1458.                            get_local_pos(c, c->chrDstHSubSample, c->dst_h_chr_pos, 0)) < 0)

  1459.                 goto fail;

  1460.         }

  1461.     } // initialize horizontal stuff

  1462. 

  1463.     /* precalculate vertical scaler filter coefficients */

  1464.     {

  1465.         const int filterAlign = X86_MMX(cpu_flags)     ? 2 :

  1466.                                 PPC_ALTIVEC(cpu_flags) ? 8 : 1;

  1467. 

  1468.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1469.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

  1470.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1471.                        cpu_flags, srcFilter->lumV, dstFilter->lumV,

  1472.                        c->param,

  1473.                        get_local_pos(c, 0, 0, 1),

  1474.                        get_local_pos(c, 0, 0, 1)) < 0)

  1475.             goto fail;

  1476.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1477.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1478.                        filterAlign, (1 << 12),

  1479.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1480.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1481.                        c->param,

  1482.                        get_local_pos(c, c->chrSrcVSubSample, c->src_v_chr_pos, 1),

  1483.                        get_local_pos(c, c->chrDstVSubSample, c->dst_v_chr_pos, 1)) < 0)

  1484. 

  1485.             goto fail;

  1486. 

  1487. #if HAVE_ALTIVEC

  1488.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1489.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1490. 

  1491.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1492.             int j;

  1493.             short *p = (short *)&c->vYCoeffsBank[i];

  1494.             for (j = 0; j < 8; j++)

  1495.                 p[j] = c->vLumFilter[i];

  1496.         }

  1497. 

  1498.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1499.             int j;

  1500.             short *p = (short *)&c->vCCoeffsBank[i];

  1501.             for (j = 0; j < 8; j++)

  1502.                 p[j] = c->vChrFilter[i];

  1503.         }

  1504. #endif

  1505.     }

  1506. 

  1507.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1508.     c->vLumBufSize = c->vLumFilterSize;

  1509.     c->vChrBufSize = c->vChrFilterSize;

  1510.     for (i = 0; i < dstH; i++) {

  1511.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1512.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1513.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1514.                                << c->chrSrcVSubSample));

  1515. 

  1516.         nextSlice >>= c->chrSrcVSubSample;

  1517.         nextSlice <<= c->chrSrcVSubSample;

  1518.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1519.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1520.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1521.             (nextSlice >> c->chrSrcVSubSample))

  1522.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1523.                              c->vChrFilterPos[chrI];

  1524.     }

  1525. 

  1526.     for (i = 0; i < 4; i++)

  1527.         FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);

  1528. 

  1529.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1530.      * need to allocate several megabytes to handle all possible cases) */

  1531.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1532.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1533.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1534.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1535.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1536.     /* Note we need at least one pixel more at the end because of the MMX code

  1537.      * (just in case someone wants to replace the 4000/8000). */

  1538.     /* align at 16 bytes for AltiVec */

  1539.     for (i = 0; i < c->vLumBufSize; i++) {

  1540.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1541.                           dst_stride + 16, fail);

  1542.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1543.     }

  1544.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1545.     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);

  1546.     c->uv_offx2 = dst_stride + 16;

  1547.     for (i = 0; i < c->vChrBufSize; i++) {

  1548.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1549.                          dst_stride * 2 + 32, fail);

  1550.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1551.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1552.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1553.     }

  1554.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1555.         for (i = 0; i < c->vLumBufSize; i++) {

  1556.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],

  1557.                               dst_stride + 16, fail);

  1558.             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];

  1559.         }

  1560. 

  1561.     // try to avoid drawing green stuff between the right end and the stride end

  1562.     for (i = 0; i < c->vChrBufSize; i++)

  1563.         if(desc_dst->comp[0].depth_minus1 == 15){

  1564.             av_assert0(c->dstBpc > 14);

  1565.             for(j=0; j<dst_stride/2+1; j++)

  1566.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1567.         } else

  1568.             for(j=0; j<dst_stride+1; j++)

  1569.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1570. 

  1571.     av_assert0(c->chrDstH <= dstH);

  1572. 

  1573.     if (flags & SWS_PRINT_INFO) {

  1574.         const char *scaler, *cpucaps;

  1575.         if (flags & SWS_FAST_BILINEAR)

  1576.             scaler = "FAST_BILINEAR scaler";

  1577.         else if (flags & SWS_BILINEAR)

  1578.             scaler = "BILINEAR scaler";

  1579.         else if (flags & SWS_BICUBIC)

  1580.             scaler = "BICUBIC scaler";

  1581.         else if (flags & SWS_X)

  1582.             scaler = "Experimental scaler";

  1583.         else if (flags & SWS_POINT)

  1584.             scaler = "Nearest Neighbor / POINT scaler";

  1585.         else if (flags & SWS_AREA)

  1586.             scaler = "Area Averaging scaler";

  1587.         else if (flags & SWS_BICUBLIN)

  1588.             scaler = "luma BICUBIC / chroma BILINEAR scaler";

  1589.         else if (flags & SWS_GAUSS)

  1590.             scaler = "Gaussian scaler";

  1591.         else if (flags & SWS_SINC)

  1592.             scaler = "Sinc scaler";

  1593.         else if (flags & SWS_LANCZOS)

  1594.             scaler = "Lanczos scaler";

  1595.         else if (flags & SWS_SPLINE)

  1596.             scaler = "Bicubic spline scaler";

  1597.         else

  1598.             scaler = "ehh flags invalid?!";

  1599. 

  1600.         av_log(c, AV_LOG_INFO, "%s, from %s to %s%s ",

  1601.                scaler,

  1602.                av_get_pix_fmt_name(srcFormat),

  1603. #ifdef DITHER1XBPP

  1604.                dstFormat == AV_PIX_FMT_BGR555   || dstFormat == AV_PIX_FMT_BGR565   ||

  1605.                dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||

  1606.                dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?

  1607.                                                              "dithered " : "",

  1608. #else

  1609.                "",

  1610. #endif

  1611.                av_get_pix_fmt_name(dstFormat));

  1612. 

  1613.         if (INLINE_MMXEXT(cpu_flags))

  1614.             cpucaps = "MMXEXT";

  1615.         else if (INLINE_AMD3DNOW(cpu_flags))

  1616.             cpucaps = "3DNOW";

  1617.         else if (INLINE_MMX(cpu_flags))

  1618.             cpucaps = "MMX";

  1619.         else if (PPC_ALTIVEC(cpu_flags))

  1620.             cpucaps = "AltiVec";

  1621.         else

  1622.             cpucaps = "C";

  1623. 

  1624.         av_log(c, AV_LOG_INFO, "using %s\n", cpucaps);

  1625. 

  1626.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1627.         av_log(c, AV_LOG_DEBUG,

  1628.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1629.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1630.         av_log(c, AV_LOG_DEBUG,

  1631.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1632.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1633.                c->chrXInc, c->chrYInc);

  1634.     }

  1635. 

  1636.     c->swscale = ff_getSwsFunc(c);

  1637.     return 0;

  1638. fail: // FIXME replace things by appropriate error codes

  1639.     return -1;

  1640. }

  1641. 

  1642. #if FF_API_SWS_GETCONTEXT

  1643. SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,

  1644.                            int dstW, int dstH, enum AVPixelFormat dstFormat,

  1645.                            int flags, SwsFilter *srcFilter,

  1646.                            SwsFilter *dstFilter, const double *param)

  1647. {

  1648.     SwsContext *c;

  1649. 

  1650.     if (!(c = sws_alloc_context()))

  1651.         return NULL;

  1652. 

  1653.     c->flags     = flags;

  1654.     c->srcW      = srcW;

  1655.     c->srcH      = srcH;

  1656.     c->dstW      = dstW;

  1657.     c->dstH      = dstH;

  1658.     c->srcFormat = srcFormat;

  1659.     c->dstFormat = dstFormat;

  1660. 

  1661.     if (param) {

  1662.         c->param[0] = param[0];

  1663.         c->param[1] = param[1];

  1664.     }

  1665. 

  1666.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1667.         sws_freeContext(c);

  1668.         return NULL;

  1669.     }

  1670. 

  1671.     return c;

  1672. }

  1673. #endif

  1674. 

  1675. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1676.                                 float lumaSharpen, float chromaSharpen,

  1677.                                 float chromaHShift, float chromaVShift,

  1678.                                 int verbose)

  1679. {

  1680.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1681.     if (!filter)

  1682.         return NULL;

  1683. 

  1684.     if (lumaGBlur != 0.0) {

  1685.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1686.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1687.     } else {

  1688.         filter->lumH = sws_getIdentityVec();

  1689.         filter->lumV = sws_getIdentityVec();

  1690.     }

  1691. 

  1692.     if (chromaGBlur != 0.0) {

  1693.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1694.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1695.     } else {

  1696.         filter->chrH = sws_getIdentityVec();

  1697.         filter->chrV = sws_getIdentityVec();

  1698.     }

  1699. 

  1700.     if (chromaSharpen != 0.0) {

  1701.         SwsVector *id = sws_getIdentityVec();

  1702.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1703.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1704.         sws_addVec(filter->chrH, id);

  1705.         sws_addVec(filter->chrV, id);

  1706.         sws_freeVec(id);

  1707.     }

  1708. 

  1709.     if (lumaSharpen != 0.0) {

  1710.         SwsVector *id = sws_getIdentityVec();

  1711.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1712.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1713.         sws_addVec(filter->lumH, id);

  1714.         sws_addVec(filter->lumV, id);

  1715.         sws_freeVec(id);

  1716.     }

  1717. 

  1718.     if (chromaHShift != 0.0)

  1719.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1720. 

  1721.     if (chromaVShift != 0.0)

  1722.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1723. 

  1724.     sws_normalizeVec(filter->chrH, 1.0);

  1725.     sws_normalizeVec(filter->chrV, 1.0);

  1726.     sws_normalizeVec(filter->lumH, 1.0);

  1727.     sws_normalizeVec(filter->lumV, 1.0);

  1728. 

  1729.     if (verbose)

  1730.         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1731.     if (verbose)

  1732.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1733. 

  1734.     return filter;

  1735. }

  1736. 

  1737. SwsVector *sws_allocVec(int length)

  1738. {

  1739.     SwsVector *vec;

  1740. 

  1741.     if(length <= 0 || length > INT_MAX/ sizeof(double))

  1742.         return NULL;

  1743. 

  1744.     vec = av_malloc(sizeof(SwsVector));

  1745.     if (!vec)

  1746.         return NULL;

  1747.     vec->length = length;

  1748.     vec->coeff  = av_malloc(sizeof(double) * length);

  1749.     if (!vec->coeff)

  1750.         av_freep(&vec);

  1751.     return vec;

  1752. }

  1753. 

  1754. SwsVector *sws_getGaussianVec(double variance, double quality)

  1755. {

  1756.     const int length = (int)(variance * quality + 0.5) | 1;

  1757.     int i;

  1758.     double middle  = (length - 1) * 0.5;

  1759.     SwsVector *vec;

  1760. 

  1761.     if(variance < 0 || quality < 0)

  1762.         return NULL;

  1763. 

  1764.     vec = sws_allocVec(length);

  1765. 

  1766.     if (!vec)

  1767.         return NULL;

  1768. 

  1769.     for (i = 0; i < length; i++) {

  1770.         double dist = i - middle;

  1771.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1772.                         sqrt(2 * variance * M_PI);

  1773.     }

  1774. 

  1775.     sws_normalizeVec(vec, 1.0);

  1776. 

  1777.     return vec;

  1778. }

  1779. 

  1780. SwsVector *sws_getConstVec(double c, int length)

  1781. {

  1782.     int i;

  1783.     SwsVector *vec = sws_allocVec(length);

  1784. 

  1785.     if (!vec)

  1786.         return NULL;

  1787. 

  1788.     for (i = 0; i < length; i++)

  1789.         vec->coeff[i] = c;

  1790. 

  1791.     return vec;

  1792. }

  1793. 

  1794. SwsVector *sws_getIdentityVec(void)

  1795. {

  1796.     return sws_getConstVec(1.0, 1);

  1797. }

  1798. 

  1799. static double sws_dcVec(SwsVector *a)

  1800. {

  1801.     int i;

  1802.     double sum = 0;

  1803. 

  1804.     for (i = 0; i < a->length; i++)

  1805.         sum += a->coeff[i];

  1806. 

  1807.     return sum;

  1808. }

  1809. 

  1810. void sws_scaleVec(SwsVector *a, double scalar)

  1811. {

  1812.     int i;

  1813. 

  1814.     for (i = 0; i < a->length; i++)

  1815.         a->coeff[i] *= scalar;

  1816. }

  1817. 

  1818. void sws_normalizeVec(SwsVector *a, double height)

  1819. {

  1820.     sws_scaleVec(a, height / sws_dcVec(a));

  1821. }

  1822. 

  1823. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1824. {

  1825.     int length = a->length + b->length - 1;

  1826.     int i, j;

  1827.     SwsVector *vec = sws_getConstVec(0.0, length);

  1828. 

  1829.     if (!vec)

  1830.         return NULL;

  1831. 

  1832.     for (i = 0; i < a->length; i++) {

  1833.         for (j = 0; j < b->length; j++) {

  1834.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1835.         }

  1836.     }

  1837. 

  1838.     return vec;

  1839. }

  1840. 

  1841. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1842. {

  1843.     int length = FFMAX(a->length, b->length);

  1844.     int i;

  1845.     SwsVector *vec = sws_getConstVec(0.0, length);

  1846. 

  1847.     if (!vec)

  1848.         return NULL;

  1849. 

  1850.     for (i = 0; i < a->length; i++)

  1851.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1852.     for (i = 0; i < b->length; i++)

  1853.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

  1854. 

  1855.     return vec;

  1856. }

  1857. 

  1858. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1859. {

  1860.     int length = FFMAX(a->length, b->length);

  1861.     int i;

  1862.     SwsVector *vec = sws_getConstVec(0.0, length);

  1863. 

  1864.     if (!vec)

  1865.         return NULL;

  1866. 

  1867.     for (i = 0; i < a->length; i++)

  1868.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1869.     for (i = 0; i < b->length; i++)

  1870.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

  1871. 

  1872.     return vec;

  1873. }

  1874. 

  1875. /* shift left / or right if "shift" is negative */

  1876. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1877. {

  1878.     int length = a->length + FFABS(shift) * 2;

  1879.     int i;

  1880.     SwsVector *vec = sws_getConstVec(0.0, length);

  1881. 

  1882.     if (!vec)

  1883.         return NULL;

  1884. 

  1885.     for (i = 0; i < a->length; i++) {

  1886.         vec->coeff[i + (length    - 1) / 2 -

  1887.                        (a->length - 1) / 2 - shift] = a->coeff[i];

  1888.     }

  1889. 

  1890.     return vec;

  1891. }

  1892. 

  1893. void sws_shiftVec(SwsVector *a, int shift)

  1894. {

  1895.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1896.     av_free(a->coeff);

  1897.     a->coeff  = shifted->coeff;

  1898.     a->length = shifted->length;

  1899.     av_free(shifted);

  1900. }

  1901. 

  1902. void sws_addVec(SwsVector *a, SwsVector *b)

  1903. {

  1904.     SwsVector *sum = sws_sumVec(a, b);

  1905.     av_free(a->coeff);

  1906.     a->coeff  = sum->coeff;

  1907.     a->length = sum->length;

  1908.     av_free(sum);

  1909. }

  1910. 

  1911. void sws_subVec(SwsVector *a, SwsVector *b)

  1912. {

  1913.     SwsVector *diff = sws_diffVec(a, b);

  1914.     av_free(a->coeff);

  1915.     a->coeff  = diff->coeff;

  1916.     a->length = diff->length;

  1917.     av_free(diff);

  1918. }

  1919. 

  1920. void sws_convVec(SwsVector *a, SwsVector *b)

  1921. {

  1922.     SwsVector *conv = sws_getConvVec(a, b);

  1923.     av_free(a->coeff);

  1924.     a->coeff  = conv->coeff;

  1925.     a->length = conv->length;

  1926.     av_free(conv);

  1927. }

  1928. 

  1929. SwsVector *sws_cloneVec(SwsVector *a)

  1930. {

  1931.     SwsVector *vec = sws_allocVec(a->length);

  1932. 

  1933.     if (!vec)

  1934.         return NULL;

  1935. 

  1936.     memcpy(vec->coeff, a->coeff, a->length * sizeof(*a->coeff));

  1937. 

  1938.     return vec;

  1939. }

  1940. 

  1941. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1942. {

  1943.     int i;

  1944.     double max = 0;

  1945.     double min = 0;

  1946.     double range;

  1947. 

  1948.     for (i = 0; i < a->length; i++)

  1949.         if (a->coeff[i] > max)

  1950.             max = a->coeff[i];

  1951. 

  1952.     for (i = 0; i < a->length; i++)

  1953.         if (a->coeff[i] < min)

  1954.             min = a->coeff[i];

  1955. 

  1956.     range = max - min;

  1957. 

  1958.     for (i = 0; i < a->length; i++) {

  1959.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1960.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1961.         for (; x > 0; x--)

  1962.             av_log(log_ctx, log_level, " ");

  1963.         av_log(log_ctx, log_level, "|\n");

  1964.     }

  1965. }

  1966. 

  1967. void sws_freeVec(SwsVector *a)

  1968. {

  1969.     if (!a)

  1970.         return;

  1971.     av_freep(&a->coeff);

  1972.     a->length = 0;

  1973.     av_free(a);

  1974. }

  1975. 

  1976. void sws_freeFilter(SwsFilter *filter)

  1977. {

  1978.     if (!filter)

  1979.         return;

  1980. 

  1981.     sws_freeVec(filter->lumH);

  1982.     sws_freeVec(filter->lumV);

  1983.     sws_freeVec(filter->chrH);

  1984.     sws_freeVec(filter->chrV);

  1985.     av_free(filter);

  1986. }

  1987. 

  1988. void sws_freeContext(SwsContext *c)

  1989. {

  1990.     int i;

  1991.     if (!c)

  1992.         return;

  1993. 

  1994.     if (c->lumPixBuf) {

  1995.         for (i = 0; i < c->vLumBufSize; i++)

  1996.             av_freep(&c->lumPixBuf[i]);

  1997.         av_freep(&c->lumPixBuf);

  1998.     }

  1999. 

  2000.     if (c->chrUPixBuf) {

  2001.         for (i = 0; i < c->vChrBufSize; i++)

  2002.             av_freep(&c->chrUPixBuf[i]);

  2003.         av_freep(&c->chrUPixBuf);

  2004.         av_freep(&c->chrVPixBuf);

  2005.     }

  2006. 

  2007.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  2008.         for (i = 0; i < c->vLumBufSize; i++)

  2009.             av_freep(&c->alpPixBuf[i]);

  2010.         av_freep(&c->alpPixBuf);

  2011.     }

  2012. 

  2013.     for (i = 0; i < 4; i++)

  2014.         av_freep(&c->dither_error[i]);

  2015. 

  2016.     av_freep(&c->vLumFilter);

  2017.     av_freep(&c->vChrFilter);

  2018.     av_freep(&c->hLumFilter);

  2019.     av_freep(&c->hChrFilter);

  2020. #if HAVE_ALTIVEC

  2021.     av_freep(&c->vYCoeffsBank);

  2022.     av_freep(&c->vCCoeffsBank);

  2023. #endif

  2024. 

  2025.     av_freep(&c->vLumFilterPos);

  2026.     av_freep(&c->vChrFilterPos);

  2027.     av_freep(&c->hLumFilterPos);

  2028.     av_freep(&c->hChrFilterPos);

  2029. 

  2030. #if HAVE_MMX_INLINE

  2031. #if USE_MMAP

  2032.     if (c->lumMmxextFilterCode)

  2033.         munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);

  2034.     if (c->chrMmxextFilterCode)

  2035.         munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);

  2036. #elif HAVE_VIRTUALALLOC

  2037.     if (c->lumMmxextFilterCode)

  2038.         VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);

  2039.     if (c->chrMmxextFilterCode)

  2040.         VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);

  2041. #else

  2042.     av_free(c->lumMmxextFilterCode);

  2043.     av_free(c->chrMmxextFilterCode);

  2044. #endif

  2045.     c->lumMmxextFilterCode = NULL;

  2046.     c->chrMmxextFilterCode = NULL;

  2047. #endif /* HAVE_MMX_INLINE */

  2048. 

  2049.     av_freep(&c->yuvTable);

  2050.     av_freep(&c->formatConvBuffer);

  2051. 

  2052.     av_free(c);

  2053. }

  2054. 

  2055. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  2056.                                         int srcH, enum AVPixelFormat srcFormat,

  2057.                                         int dstW, int dstH,

  2058.                                         enum AVPixelFormat dstFormat, int flags,

  2059.                                         SwsFilter *srcFilter,

  2060.                                         SwsFilter *dstFilter,

  2061.                                         const double *param)

  2062. {

  2063.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  2064.                                              SWS_PARAM_DEFAULT };

  2065. 

  2066.     if (!param)

  2067.         param = default_param;

  2068. 

  2069.     if (context &&

  2070.         (context->srcW      != srcW      ||

  2071.          context->srcH      != srcH      ||

  2072.          context->srcFormat != srcFormat ||

  2073.          context->dstW      != dstW      ||

  2074.          context->dstH      != dstH      ||

  2075.          context->dstFormat != dstFormat ||

  2076.          context->flags     != flags     ||

  2077.          context->param[0]  != param[0]  ||

  2078.          context->param[1]  != param[1])) {

  2079.         sws_freeContext(context);

  2080.         context = NULL;

  2081.     }

  2082. 

  2083.     if (!context) {

  2084.         if (!(context = sws_alloc_context()))

  2085.             return NULL;

  2086.         context->srcW      = srcW;

  2087.         context->srcH      = srcH;

  2088.         context->srcFormat = srcFormat;

  2089.         context->dstW      = dstW;

  2090.         context->dstH      = dstH;

  2091.         context->dstFormat = dstFormat;

  2092.         context->flags     = flags;

  2093.         context->param[0]  = param[0];

  2094.         context->param[1]  = param[1];

  2095.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  2096.             sws_freeContext(context);

  2097.             return NULL;

  2098.         }

  2099.     }

  2100.     return context;

  2101. }