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.
44245 Commits
32 Branches
301MB
Tree: ea8b730d8e
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'ea8b730d8e'
${ noResults }
FFmpeg/libswscale/utils.c

1809 lines
64KB
Raw Blame History 

  1. /*

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

  3.  *

  4.  * This file is part of Libav.

  5.  *

  6.  * Libav is free software; you can redistribute it and/or

  7.  * modify it under the terms of the GNU Lesser General Public

  8.  * License as published by the Free Software Foundation; either

  9.  * version 2.1 of the License, or (at your option) any later version.

  10.  *

  11.  * Libav is distributed in the hope that it will be useful,

  12.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  13.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  14.  * Lesser General Public License for more details.

  15.  *

  16.  * You should have received a copy of the GNU Lesser General Public

  17.  * License along with Libav; if not, write to the Free Software

  18.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  19.  */

  20. 

  21. #include "config.h"

  22. 

  23. #define _SVID_SOURCE // needed for MAP_ANONYMOUS

  24. #include <assert.h>

  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/avutil.h"

  42. #include "libavutil/bswap.h"

  43. #include "libavutil/cpu.h"

  44. #include "libavutil/intreadwrite.h"

  45. #include "libavutil/mathematics.h"

  46. #include "libavutil/opt.h"

  47. #include "libavutil/pixdesc.h"

  48. #include "libavutil/ppc/cpu.h"

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

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

  51. #include "rgb2rgb.h"

  52. #include "swscale.h"

  53. #include "swscale_internal.h"

  54. 

  55. unsigned swscale_version(void)

  56. {

  57.     return LIBSWSCALE_VERSION_INT;

  58. }

  59. 

  60. const char *swscale_configuration(void)

  61. {

  62.     return LIBAV_CONFIGURATION;

  63. }

  64. 

  65. const char *swscale_license(void)

  66. {

  67. #define LICENSE_PREFIX "libswscale license: "

  68.     return LICENSE_PREFIX LIBAV_LICENSE + sizeof(LICENSE_PREFIX) - 1;

  69. }

  70. 

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

  72. 

  73. typedef struct FormatEntry {

  74.     uint8_t is_supported_in         :1;

  75.     uint8_t is_supported_out        :1;

  76.     uint8_t is_supported_endianness :1;

  77. } FormatEntry;

  78. 

  79. static const FormatEntry format_entries[AV_PIX_FMT_NB] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  95.     [AV_PIX_FMT_YVYU422]     = { 1, 1 },

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  110.     [AV_PIX_FMT_GRAY12BE]    = { 1, 1 },

  111.     [AV_PIX_FMT_GRAY12LE]    = { 1, 1 },

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  159.     [AV_PIX_FMT_YA8]         = { 1, 0 },

  160.     [AV_PIX_FMT_YA16BE]      = { 1, 0 },

  161.     [AV_PIX_FMT_YA16LE]      = { 1, 0 },

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

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

  164.     [AV_PIX_FMT_BGRA64BE]    = { 0, 0, 1 },

  165.     [AV_PIX_FMT_BGRA64LE]    = { 0, 0, 1 },

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  194.     [AV_PIX_FMT_GBRAP12LE]   = { 1, 1 },

  195.     [AV_PIX_FMT_GBRAP12BE]   = { 1, 1 },

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

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

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

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

  200.     [AV_PIX_FMT_P010LE]      = { 1, 0 },

  201.     [AV_PIX_FMT_P010BE]      = { 1, 0 },

  202. };

  203. 

  204. int sws_isSupportedInput(enum AVPixelFormat pix_fmt)

  205. {

  206.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  207.            format_entries[pix_fmt].is_supported_in : 0;

  208. }

  209. 

  210. int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)

  211. {

  212.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  213.            format_entries[pix_fmt].is_supported_out : 0;

  214. }

  215. 

  216. int sws_isSupportedEndiannessConversion(enum AVPixelFormat pix_fmt)

  217. {

  218.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  219.            format_entries[pix_fmt].is_supported_endianness : 0;

  220. }

  221. 

  222. const char *sws_format_name(enum AVPixelFormat format)

  223. {

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

  225.     if (desc)

  226.         return desc->name;

  227.     else

  228.         return "Unknown format";

  229. }

  230. 

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

  232.                              double dist)

  233. {

  234.     if (dist <= 1.0)

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

  236.     else

  237.         return getSplineCoeff(0.0,

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

  239.                                c + 3.0 * d,

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

  241.                               dist - 1.0);

  242. }

  243. 

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

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

  246.                               int dstW, int filterAlign, int one,

  247.                               int flags, int cpu_flags,

  248.                               SwsVector *srcFilter, SwsVector *dstFilter,

  249.                               double param[2], int is_horizontal)

  250. {

  251.     int i;

  252.     int filterSize;

  253.     int filter2Size;

  254.     int minFilterSize;

  255.     int64_t *filter    = NULL;

  256.     int64_t *filter2   = NULL;

  257.     const int64_t fone = 1LL << 54;

  258.     int ret            = -1;

  259. 

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

  261. 

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

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

  264. 

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

  266.         int i;

  267.         filterSize = 1;

  268.         FF_ALLOCZ_OR_GOTO(NULL, filter,

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

  270. 

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

  272.             filter[i * filterSize] = fone;

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

  274.         }

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

  276.         int i;

  277.         int xDstInSrc;

  278.         filterSize = 1;

  279.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  281. 

  282.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  285. 

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

  287.             filter[i]       = fone;

  288.             xDstInSrc      += xInc;

  289.         }

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

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

  292.         int i;

  293.         int xDstInSrc;

  294.         filterSize = 2;

  295.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  297. 

  298.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  301.             int j;

  302. 

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

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

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

  306.                 int64_t coeff = fone - FFABS((xx << 16) - xDstInSrc) *

  307.                                 (fone >> 16);

  308.                 if (coeff < 0)

  309.                     coeff = 0;

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

  311.                 xx++;

  312.             }

  313.             xDstInSrc += xInc;

  314.         }

  315.     } else {

  316.         int64_t xDstInSrc;

  317.         int sizeFactor;

  318. 

  319.         if (flags & SWS_BICUBIC)

  320.             sizeFactor = 4;

  321.         else if (flags & SWS_X)

  322.             sizeFactor = 8;

  323.         else if (flags & SWS_AREA)

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

  325.         else if (flags & SWS_GAUSS)

  326.             sizeFactor = 8;     // infinite ;)

  327.         else if (flags & SWS_LANCZOS)

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

  329.         else if (flags & SWS_SINC)

  330.             sizeFactor = 20;    // infinite ;)

  331.         else if (flags & SWS_SPLINE)

  332.             sizeFactor = 20;    // infinite ;)

  333.         else if (flags & SWS_BILINEAR)

  334.             sizeFactor = 2;

  335.         else {

  336.             sizeFactor = 0;     // GCC warning killer

  337.             assert(0);

  338.         }

  339. 

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

  341.             filterSize = 1 + sizeFactor;    // upscale

  342.         else

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

  344. 

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

  346.         filterSize = FFMAX(filterSize, 1);

  347. 

  348.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  350. 

  351.         xDstInSrc = xInc - 0x10000;

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

  353.             int xx = (xDstInSrc - ((int64_t)(filterSize - 2) << 16)) / (1 << 17);

  354.             int j;

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

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

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

  358.                 double floatd;

  359.                 int64_t coeff;

  360. 

  361.                 if (xInc > 1 << 16)

  362.                     d = d * dstW / srcW;

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

  364. 

  365.                 if (flags & SWS_BICUBIC) {

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

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

  368. 

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

  370.                         coeff = 0.0;

  371.                     } else {

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

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

  374. 

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

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

  377.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  378.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  379.                         else

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

  381.                                       (6 * B + 30 * C) * dd  +

  382.                                     (-12 * B - 48 * C) * d   +

  383.                                       (8 * B + 24 * C) * (1 << 30);

  384.                     }

  385.                     coeff *= fone >> (30 + 24);

  386.                 }

  387.                 else if (flags & SWS_X) {

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

  389.                     double c;

  390. 

  391.                     if (floatd < 1.0)

  392.                         c = cos(floatd * M_PI);

  393.                     else

  394.                         c = -1.0;

  395.                     if (c < 0.0)

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

  397.                     else

  398.                         c = pow(c, A);

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

  400.                 } else if (flags & SWS_AREA) {

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

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

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

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

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

  406.                     else

  407.                         coeff = 0.0;

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

  409.                 } else if (flags & SWS_GAUSS) {

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

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

  412.                 } else if (flags & SWS_SINC) {

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

  414.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  418.                     if (floatd > p)

  419.                         coeff = 0;

  420.                 } else if (flags & SWS_BILINEAR) {

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

  422.                     if (coeff < 0)

  423.                         coeff = 0;

  424.                     coeff *= fone >> 30;

  425.                 } else if (flags & SWS_SPLINE) {

  426.                     double p = -2.196152422706632;

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

  428.                 } else {

  429.                     coeff = 0.0; // GCC warning killer

  430.                     assert(0);

  431.                 }

  432. 

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

  434.                 xx++;

  435.             }

  436.             xDstInSrc += 2 * xInc;

  437.         }

  438.     }

  439. 

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

  441.      * av_free(filter);

  442.      */

  443.     assert(filterSize > 0);

  444.     filter2Size = filterSize;

  445.     if (srcFilter)

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

  447.     if (dstFilter)

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

  449.     assert(filter2Size > 0);

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

  451. 

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

  453.         int j, k;

  454. 

  455.         if (srcFilter) {

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

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

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

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

  460.             }

  461.         } else {

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

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

  464.         }

  465.         // FIXME dstFilter

  466. 

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

  468.     }

  469.     av_freep(&filter);

  470. 

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

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

  473.     minFilterSize = 0;

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

  475.         int min = filter2Size;

  476.         int j;

  477.         int64_t cutOff = 0.0;

  478. 

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

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

  481.             int k;

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

  483. 

  484.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  485.                 break;

  486. 

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

  488.              * filter otherwise */

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

  490.                 break;

  491. 

  492.             // move filter coefficients left

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

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

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

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

  497.         }

  498. 

  499.         cutOff = 0;

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

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

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

  503. 

  504.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  505.                 break;

  506.             min--;

  507.         }

  508. 

  509.         if (min > minFilterSize)

  510.             minFilterSize = min;

  511.     }

  512. 

  513.     if (PPC_ALTIVEC(cpu_flags)) {

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

  515.         if (minFilterSize < 5)

  516.             filterAlign = 4;

  517. 

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

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

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

  521.         if (minFilterSize < 3)

  522.             filterAlign = 1;

  523.     }

  524. 

  525.     if (INLINE_MMX(cpu_flags)) {

  526.         // special case for unscaled vertical filtering

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

  528.             filterAlign = 1;

  529.     }

  530. 

  531.     assert(minFilterSize > 0);

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

  533.     assert(filterSize > 0);

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

  535.     if (filterSize >= MAX_FILTER_SIZE * 16 /

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

  537.         goto fail;

  538.     *outFilterSize = filterSize;

  539. 

  540.     if (flags & SWS_PRINT_INFO)

  541.         av_log(NULL, AV_LOG_VERBOSE,

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

  543.                filter2Size, filterSize);

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

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

  546.         int j;

  547. 

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

  549.             if (j >= filter2Size)

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

  551.             else

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

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

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

  555.         }

  556.     }

  557. 

  558.     // FIXME try to align filterPos if possible

  559. 

  560.     // fix borders

  561.     if (is_horizontal) {

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

  563.             int j;

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

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

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

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

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

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

  570.                 }

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

  572.             }

  573. 

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

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

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

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

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

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

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

  581.                 }

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

  583.             }

  584.         }

  585.     }

  586. 

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

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

  589.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

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

  591. 

  592.     /* normalize & store in outFilter */

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

  594.         int j;

  595.         int64_t error = 0;

  596.         int64_t sum   = 0;

  597. 

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

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

  600.         }

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

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

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

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

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

  606.             error                                  = v - intV * sum;

  607.         }

  608.     }

  609. 

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

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

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

  613.                                                       * read over the end */

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

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

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

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

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

  619.     }

  620. 

  621.     ret = 0;

  622. 

  623. fail:

  624.     av_free(filter);

  625.     av_free(filter2);

  626.     return ret;

  627. }

  628. 

  629. #if HAVE_MMXEXT_INLINE

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

  631.                                        int16_t *filter, int32_t *filterPos,

  632.                                        int numSplits)

  633. {

  634.     uint8_t *fragmentA;

  635.     x86_reg imm8OfPShufW1A;

  636.     x86_reg imm8OfPShufW2A;

  637.     x86_reg fragmentLengthA;

  638.     uint8_t *fragmentB;

  639.     x86_reg imm8OfPShufW1B;

  640.     x86_reg imm8OfPShufW2B;

  641.     x86_reg fragmentLengthB;

  642.     int fragmentPos;

  643. 

  644.     int xpos, i;

  645. 

  646.     // create an optimized horizontal scaling routine

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

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

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

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

  651.      */

  652. 

  653.     // code fragment

  654. 

  655.     __asm__ volatile (

  656.         "jmp                         9f                 \n\t"

  657.         // Begin

  658.         "0:                                             \n\t"

  659.         "movq  (%%"FF_REG_d", %%"FF_REG_a"), %%mm3      \n\t"

  660.         "movd  (%%"FF_REG_c", %%"FF_REG_S"), %%mm0      \n\t"

  661.         "movd 1(%%"FF_REG_c", %%"FF_REG_S"), %%mm1      \n\t"

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

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

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

  665.         "1:                                             \n\t"

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

  667.         "2:                                             \n\t"

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

  669.         "movl 8(%%"FF_REG_b", %%"FF_REG_a"), %%esi      \n\t"

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

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

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

  673. 

  674.         "movq       %%mm0, (%%"FF_REG_D", %%"FF_REG_a") \n\t"

  675. 

  676.         "add                         $8, %%"FF_REG_a"   \n\t"

  677.         // End

  678.         "9:                                             \n\t"

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

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

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

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

  683.         "dec                         %1                 \n\t"

  684.         "dec                         %2                 \n\t"

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

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

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

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

  689. 

  690. 

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

  692.           "=r" (fragmentLengthA)

  693.         );

  694. 

  695.     __asm__ volatile (

  696.         "jmp                         9f                 \n\t"

  697.         // Begin

  698.         "0:                                             \n\t"

  699.         "movq (%%"FF_REG_d", %%"FF_REG_a"), %%mm3       \n\t"

  700.         "movd (%%"FF_REG_c", %%"FF_REG_S"), %%mm0       \n\t"

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

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

  703.         "1:                                             \n\t"

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

  705.         "2:                                             \n\t"

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

  707.         "movl 8(%%"FF_REG_b", %%"FF_REG_a"), %%esi      \n\t"

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

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

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

  711. 

  712.         "movq       %%mm0, (%%"FF_REG_D", %%"FF_REG_a") \n\t"

  713. 

  714.         "add                      $8, %%"FF_REG_a"      \n\t"

  715.         // End

  716.         "9:                                             \n\t"

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

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

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

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

  721.         "dec                         %1                 \n\t"

  722.         "dec                         %2                 \n\t"

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

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

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

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

  727. 

  728. 

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

  730.           "=r" (fragmentLengthB)

  731.         );

  732. 

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

  734.     fragmentPos = 0;

  735. 

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

  737.         int xx = xpos >> 16;

  738. 

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

  740.             int a                  = 0;

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

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

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

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

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

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

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

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

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

  750.             int shift              = 0;

  751. 

  752.             if (filterCode) {

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

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

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

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

  757.                 filterPos[i / 2] = xx;

  758. 

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

  760. 

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

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

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

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

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

  766.                                                                (c << 4) |

  767.                                                                (d << 6);

  768. 

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

  770.                     shift = maxShift;               // avoid overread

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

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

  773. 

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

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

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

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

  778.                 }

  779.             }

  780. 

  781.             fragmentPos += fragmentLength;

  782. 

  783.             if (filterCode)

  784.                 filterCode[fragmentPos] = RET;

  785.         }

  786.         xpos += xInc;

  787.     }

  788.     if (filterCode)

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

  790. 

  791.     return fragmentPos + 1;

  792. }

  793. #endif /* HAVE_MMXEXT_INLINE */

  794. 

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

  796. {

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

  798.     *h = desc->log2_chroma_w;

  799.     *v = desc->log2_chroma_h;

  800. }

  801. 

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

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

  804.                              int brightness, int contrast, int saturation)

  805. {

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

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

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

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

  810. 

  811.     c->brightness = brightness;

  812.     c->contrast   = contrast;

  813.     c->saturation = saturation;

  814.     c->srcRange   = srcRange;

  815.     c->dstRange   = dstRange;

  816.     if (isYUV(c->dstFormat) || isGray(c->dstFormat))

  817.         return -1;

  818. 

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

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

  821. 

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

  823.                              contrast, saturation);

  824.     // FIXME factorize

  825. 

  826.     if (ARCH_PPC)

  827.         ff_yuv2rgb_init_tables_ppc(c, inv_table, brightness,

  828.                                    contrast, saturation);

  829.     return 0;

  830. }

  831. 

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

  833.                              int *srcRange, int **table, int *dstRange,

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

  835. {

  836.     if (isYUV(c->dstFormat) || isGray(c->dstFormat))

  837.         return -1;

  838. 

  839.     *inv_table  = c->srcColorspaceTable;

  840.     *table      = c->dstColorspaceTable;

  841.     *srcRange   = c->srcRange;

  842.     *dstRange   = c->dstRange;

  843.     *brightness = c->brightness;

  844.     *contrast   = c->contrast;

  845.     *saturation = c->saturation;

  846. 

  847.     return 0;

  848. }

  849. 

  850. static int handle_jpeg(enum AVPixelFormat *format)

  851. {

  852.     switch (*format) {

  853.     case AV_PIX_FMT_YUVJ420P:

  854.         *format = AV_PIX_FMT_YUV420P;

  855.         return 1;

  856.     case AV_PIX_FMT_YUVJ422P:

  857.         *format = AV_PIX_FMT_YUV422P;

  858.         return 1;

  859.     case AV_PIX_FMT_YUVJ444P:

  860.         *format = AV_PIX_FMT_YUV444P;

  861.         return 1;

  862.     case AV_PIX_FMT_YUVJ440P:

  863.         *format = AV_PIX_FMT_YUV440P;

  864.         return 1;

  865.     default:

  866.         return 0;

  867.     }

  868. }

  869. 

  870. SwsContext *sws_alloc_context(void)

  871. {

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

  873. 

  874.     if (c) {

  875.         c->av_class = &ff_sws_context_class;

  876.         av_opt_set_defaults(c);

  877.     }

  878. 

  879.     return c;

  880. }

  881. 

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

  883.                              SwsFilter *dstFilter)

  884. {

  885.     int i;

  886.     int usesVFilter, usesHFilter;

  887.     int unscaled;

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

  889.     int srcW              = c->srcW;

  890.     int srcH              = c->srcH;

  891.     int dstW              = c->dstW;

  892.     int dstH              = c->dstH;

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

  894.     int dst_stride_px     = dst_stride >> 1;

  895.     int flags, cpu_flags;

  896.     enum AVPixelFormat srcFormat = c->srcFormat;

  897.     enum AVPixelFormat dstFormat = c->dstFormat;

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

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

  900. 

  901.     cpu_flags = av_get_cpu_flags();

  902.     flags     = c->flags;

  903.     emms_c();

  904.     if (!rgb15to16)

  905.         ff_rgb2rgb_init();

  906. 

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

  908. 

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

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

  911.     if (!sws_isSupportedInput(srcFormat)) {

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

  913.                sws_format_name(srcFormat));

  914.         return AVERROR(EINVAL);

  915.     }

  916.     if (!sws_isSupportedOutput(dstFormat)) {

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

  918.                sws_format_name(dstFormat));

  919.         return AVERROR(EINVAL);

  920.     }

  921.     }

  922. 

  923.     i = flags & (SWS_POINT         |

  924.                  SWS_AREA          |

  925.                  SWS_BILINEAR      |

  926.                  SWS_FAST_BILINEAR |

  927.                  SWS_BICUBIC       |

  928.                  SWS_X             |

  929.                  SWS_GAUSS         |

  930.                  SWS_LANCZOS       |

  931.                  SWS_SINC          |

  932.                  SWS_SPLINE        |

  933.                  SWS_BICUBLIN);

  934. 

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

  936.     if (!i) {

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

  938.             flags |= SWS_GAUSS;

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

  940.             flags |= SWS_SINC;

  941.         else

  942.             flags |= SWS_LANCZOS;

  943.         c->flags = flags;

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

  945.         av_log(c, AV_LOG_ERROR,

  946.                "Exactly one scaler algorithm must be chosen\n");

  947.         return AVERROR(EINVAL);

  948.     }

  949.     /* sanity check */

  950.     if (srcW < 4 || srcH < 1 || dstW < 8 || dstH < 1) {

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

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

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

  954.                srcW, srcH, dstW, dstH);

  955.         return AVERROR(EINVAL);

  956.     }

  957. 

  958.     if (!dstFilter)

  959.         dstFilter = &dummyFilter;

  960.     if (!srcFilter)

  961.         srcFilter = &dummyFilter;

  962. 

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

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

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

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

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

  968. 

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

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

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

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

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

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

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

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

  977. 

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

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

  980. 

  981.     if (isPlanarRGB(dstFormat)) {

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

  983.             av_log(c, AV_LOG_DEBUG,

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

  985.                    av_get_pix_fmt_name(dstFormat));

  986.             flags   |= SWS_FULL_CHR_H_INT;

  987.             c->flags = flags;

  988.         }

  989.     }

  990. 

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

  992.      * chroma interpolation */

  993.     if (flags & SWS_FULL_CHR_H_INT &&

  994.         isAnyRGB(dstFormat)        &&

  995.         !isPlanarRGB(dstFormat)    &&

  996.         dstFormat != AV_PIX_FMT_RGBA  &&

  997.         dstFormat != AV_PIX_FMT_ARGB  &&

  998.         dstFormat != AV_PIX_FMT_BGRA  &&

  999.         dstFormat != AV_PIX_FMT_ABGR  &&

  1000.         dstFormat != AV_PIX_FMT_RGB24 &&

  1001.         dstFormat != AV_PIX_FMT_BGR24) {

  1002.         av_log(c, AV_LOG_ERROR,

  1003.                "full chroma interpolation for destination format '%s' not yet implemented\n",

  1004.                sws_format_name(dstFormat));

  1005.         flags   &= ~SWS_FULL_CHR_H_INT;

  1006.         c->flags = flags;

  1007.     }

  1008.     if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))

  1009.         c->chrDstHSubSample = 1;

  1010. 

  1011.     // drop some chroma lines if the user wants it

  1012.     c->vChrDrop          = (flags & SWS_SRC_V_CHR_DROP_MASK) >>

  1013.                            SWS_SRC_V_CHR_DROP_SHIFT;

  1014.     c->chrSrcVSubSample += c->vChrDrop;

  1015. 

  1016.     /* drop every other pixel for chroma calculation unless user

  1017.      * wants full chroma */

  1018.     if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP)   &&

  1019.         srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&

  1020.         srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&

  1021.         srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&

  1022.         srcFormat != AV_PIX_FMT_GBRP9BE   && srcFormat != AV_PIX_FMT_GBRP9LE  &&

  1023.         srcFormat != AV_PIX_FMT_GBRP10BE  && srcFormat != AV_PIX_FMT_GBRP10LE &&

  1024.         srcFormat != AV_PIX_FMT_GBRP12BE  && srcFormat != AV_PIX_FMT_GBRP12LE &&

  1025.         srcFormat != AV_PIX_FMT_GBRP16BE  && srcFormat != AV_PIX_FMT_GBRP16LE &&

  1026.         ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||

  1027.          (flags & SWS_FAST_BILINEAR)))

  1028.         c->chrSrcHSubSample = 1;

  1029. 

  1030.     // Note the AV_CEIL_RSHIFT is so that we always round toward +inf.

  1031.     c->chrSrcW = AV_CEIL_RSHIFT(srcW, c->chrSrcHSubSample);

  1032.     c->chrSrcH = AV_CEIL_RSHIFT(srcH, c->chrSrcVSubSample);

  1033.     c->chrDstW = AV_CEIL_RSHIFT(dstW, c->chrDstHSubSample);

  1034.     c->chrDstH = AV_CEIL_RSHIFT(dstH, c->chrDstVSubSample);

  1035. 

  1036.     /* unscaled special cases */

  1037.     if (unscaled && !usesHFilter && !usesVFilter &&

  1038.         (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {

  1039.         ff_get_unscaled_swscale(c);

  1040. 

  1041.         if (c->swscale) {

  1042.             if (flags & SWS_PRINT_INFO)

  1043.                 av_log(c, AV_LOG_INFO,

  1044.                        "using unscaled %s -> %s special converter\n",

  1045.                        sws_format_name(srcFormat), sws_format_name(dstFormat));

  1046.             return 0;

  1047.         }

  1048.     }

  1049. 

  1050.     c->srcBpc = desc_src->comp[0].depth;

  1051.     if (c->srcBpc < 8)

  1052.         c->srcBpc = 8;

  1053.     c->dstBpc = desc_dst->comp[0].depth;

  1054.     if (c->dstBpc < 8)

  1055.         c->dstBpc = 8;

  1056.     if (c->dstBpc == 16)

  1057.         dst_stride <<= 1;

  1058.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,

  1059.                      (FFALIGN(srcW, 16) * 2 * FFALIGN(c->srcBpc, 8) >> 3) + 16,

  1060.                      fail);

  1061.     if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 12) {

  1062.         c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&

  1063.                               (srcW & 15) == 0) ? 1 : 0;

  1064.         if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0

  1065.             && (flags & SWS_FAST_BILINEAR)) {

  1066.             if (flags & SWS_PRINT_INFO)

  1067.                 av_log(c, AV_LOG_INFO,

  1068.                        "output width is not a multiple of 32 -> no MMXEXT scaler\n");

  1069.         }

  1070.         if (usesHFilter)

  1071.             c->canMMXEXTBeUsed = 0;

  1072.     } else

  1073.         c->canMMXEXTBeUsed = 0;

  1074. 

  1075.     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;

  1076.     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

  1077. 

  1078.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  1079.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  1080.      * correct scaling.

  1081.      * n-2 is the last chrominance sample available.

  1082.      * This is not perfect, but no one should notice the difference, the more

  1083.      * correct variant would be like the vertical one, but that would require

  1084.      * some special code for the first and last pixel */

  1085.     if (flags & SWS_FAST_BILINEAR) {

  1086.         if (c->canMMXEXTBeUsed) {

  1087.             c->lumXInc += 20;

  1088.             c->chrXInc += 20;

  1089.         }

  1090.         // we don't use the x86 asm scaler if MMX is available

  1091.         else if (INLINE_MMX(cpu_flags)) {

  1092.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1093.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1094.         }

  1095.     }

  1096. 

  1097. #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)

  1098. 

  1099.     /* precalculate horizontal scaler filter coefficients */

  1100.     {

  1101. #if HAVE_MMXEXT_INLINE

  1102. // can't downscale !!!

  1103.         if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {

  1104.             c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,

  1105.                                                              NULL, NULL, 8);

  1106.             c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,

  1107.                                                              NULL, NULL, NULL, 4);

  1108. 

  1109. #if USE_MMAP

  1110.             c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,

  1111.                                           PROT_READ | PROT_WRITE,

  1112.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1113.                                           -1, 0);

  1114.             c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,

  1115.                                           PROT_READ | PROT_WRITE,

  1116.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1117.                                           -1, 0);

  1118. #elif HAVE_VIRTUALALLOC

  1119.             c->lumMmxextFilterCode = VirtualAlloc(NULL,

  1120.                                                   c->lumMmxextFilterCodeSize,

  1121.                                                   MEM_COMMIT,

  1122.                                                   PAGE_EXECUTE_READWRITE);

  1123.             c->chrMmxextFilterCode = VirtualAlloc(NULL,

  1124.                                                   c->chrMmxextFilterCodeSize,

  1125.                                                   MEM_COMMIT,

  1126.                                                   PAGE_EXECUTE_READWRITE);

  1127. #else

  1128.             c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);

  1129.             c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);

  1130. #endif

  1131. 

  1132.             if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)

  1133.                 return AVERROR(ENOMEM);

  1134.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1135.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1136.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1137.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1138. 

  1139.             init_hscaler_mmxext(dstW, c->lumXInc, c->lumMmxextFilterCode,

  1140.                                 c->hLumFilter, c->hLumFilterPos, 8);

  1141.             init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,

  1142.                                 c->hChrFilter, c->hChrFilterPos, 4);

  1143. 

  1144. #if USE_MMAP

  1145.             mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1146.             mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1147. #endif

  1148.         } else

  1149. #endif /* HAVE_MMXEXT_INLINE */

  1150.         {

  1151.             const int filterAlign = X86_MMX(cpu_flags)     ? 4 :

  1152.                                     PPC_ALTIVEC(cpu_flags) ? 8 : 1;

  1153. 

  1154.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1155.                            &c->hLumFilterSize, c->lumXInc,

  1156.                            srcW, dstW, filterAlign, 1 << 14,

  1157.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1158.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1159.                            c->param, 1) < 0)

  1160.                 goto fail;

  1161.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1162.                            &c->hChrFilterSize, c->chrXInc,

  1163.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1164.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1165.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1166.                            c->param, 1) < 0)

  1167.                 goto fail;

  1168.         }

  1169.     } // initialize horizontal stuff

  1170. 

  1171.     /* precalculate vertical scaler filter coefficients */

  1172.     {

  1173.         const int filterAlign = X86_MMX(cpu_flags)     ? 2 :

  1174.                                 PPC_ALTIVEC(cpu_flags) ? 8 : 1;

  1175. 

  1176.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1177.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

  1178.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1179.                        cpu_flags, srcFilter->lumV, dstFilter->lumV,

  1180.                        c->param, 0) < 0)

  1181.             goto fail;

  1182.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1183.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1184.                        filterAlign, (1 << 12),

  1185.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1186.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1187.                        c->param, 0) < 0)

  1188.             goto fail;

  1189. 

  1190. #if HAVE_ALTIVEC

  1191.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1192.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1193. 

  1194.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1195.             int j;

  1196.             short *p = (short *)&c->vYCoeffsBank[i];

  1197.             for (j = 0; j < 8; j++)

  1198.                 p[j] = c->vLumFilter[i];

  1199.         }

  1200. 

  1201.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1202.             int j;

  1203.             short *p = (short *)&c->vCCoeffsBank[i];

  1204.             for (j = 0; j < 8; j++)

  1205.                 p[j] = c->vChrFilter[i];

  1206.         }

  1207. #endif

  1208.     }

  1209. 

  1210.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1211.     c->vLumBufSize = c->vLumFilterSize;

  1212.     c->vChrBufSize = c->vChrFilterSize;

  1213.     for (i = 0; i < dstH; i++) {

  1214.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1215.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1216.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1217.                                << c->chrSrcVSubSample));

  1218. 

  1219.         nextSlice >>= c->chrSrcVSubSample;

  1220.         nextSlice <<= c->chrSrcVSubSample;

  1221.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1222.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1223.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1224.             (nextSlice >> c->chrSrcVSubSample))

  1225.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1226.                              c->vChrFilterPos[chrI];

  1227.     }

  1228. 

  1229.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1230.      * need to allocate several megabytes to handle all possible cases) */

  1231.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1232.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1233.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1234.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1235.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1236.     /* Note we need at least one pixel more at the end because of the MMX code

  1237.      * (just in case someone wants to replace the 4000/8000). */

  1238.     /* align at 16 bytes for AltiVec */

  1239.     for (i = 0; i < c->vLumBufSize; i++) {

  1240.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1241.                           dst_stride + 16, fail);

  1242.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1243.     }

  1244.     // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)

  1245.     c->uv_off_px   = dst_stride_px + 64 / (c->dstBpc & ~7);

  1246.     c->uv_off_byte = dst_stride + 16;

  1247.     for (i = 0; i < c->vChrBufSize; i++) {

  1248.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1249.                          dst_stride * 2 + 32, fail);

  1250.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1251.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1252.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1253.     }

  1254.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1255.         for (i = 0; i < c->vLumBufSize; i++) {

  1256.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],

  1257.                               dst_stride + 16, fail);

  1258.             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];

  1259.         }

  1260. 

  1261.     // try to avoid drawing green stuff between the right end and the stride end

  1262.     for (i = 0; i < c->vChrBufSize; i++)

  1263.         memset(c->chrUPixBuf[i], 64, dst_stride * 2 + 1);

  1264. 

  1265.     assert(c->chrDstH <= dstH);

  1266. 

  1267.     if (flags & SWS_PRINT_INFO) {

  1268.         if (flags & SWS_FAST_BILINEAR)

  1269.             av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1270.         else if (flags & SWS_BILINEAR)

  1271.             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1272.         else if (flags & SWS_BICUBIC)

  1273.             av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1274.         else if (flags & SWS_X)

  1275.             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1276.         else if (flags & SWS_POINT)

  1277.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1278.         else if (flags & SWS_AREA)

  1279.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1280.         else if (flags & SWS_BICUBLIN)

  1281.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1282.         else if (flags & SWS_GAUSS)

  1283.             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1284.         else if (flags & SWS_SINC)

  1285.             av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1286.         else if (flags & SWS_LANCZOS)

  1287.             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1288.         else if (flags & SWS_SPLINE)

  1289.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1290.         else

  1291.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1292. 

  1293.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1294.                sws_format_name(srcFormat),

  1295. #ifdef DITHER1XBPP

  1296.                dstFormat == AV_PIX_FMT_BGR555   || dstFormat == AV_PIX_FMT_BGR565   ||

  1297.                dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||

  1298.                dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?

  1299.                                                              "dithered " : "",

  1300. #else

  1301.                "",

  1302. #endif

  1303.                sws_format_name(dstFormat));

  1304. 

  1305.         if (INLINE_MMXEXT(cpu_flags))

  1306.             av_log(c, AV_LOG_INFO, "using MMXEXT\n");

  1307.         else if (INLINE_AMD3DNOW(cpu_flags))

  1308.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1309.         else if (INLINE_MMX(cpu_flags))

  1310.             av_log(c, AV_LOG_INFO, "using MMX\n");

  1311.         else if (PPC_ALTIVEC(cpu_flags))

  1312.             av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1313.         else

  1314.             av_log(c, AV_LOG_INFO, "using C\n");

  1315. 

  1316.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1317.         av_log(c, AV_LOG_DEBUG,

  1318.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1319.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1320.         av_log(c, AV_LOG_DEBUG,

  1321.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1322.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1323.                c->chrXInc, c->chrYInc);

  1324.     }

  1325. 

  1326.     c->swscale = ff_getSwsFunc(c);

  1327.     return 0;

  1328. fail: // FIXME replace things by appropriate error codes

  1329.     return -1;

  1330. }

  1331. 

  1332. SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,

  1333.                            int dstW, int dstH, enum AVPixelFormat dstFormat,

  1334.                            int flags, SwsFilter *srcFilter,

  1335.                            SwsFilter *dstFilter, const double *param)

  1336. {

  1337.     SwsContext *c;

  1338. 

  1339.     if (!(c = sws_alloc_context()))

  1340.         return NULL;

  1341. 

  1342.     c->flags     = flags;

  1343.     c->srcW      = srcW;

  1344.     c->srcH      = srcH;

  1345.     c->dstW      = dstW;

  1346.     c->dstH      = dstH;

  1347.     c->srcRange  = handle_jpeg(&srcFormat);

  1348.     c->dstRange  = handle_jpeg(&dstFormat);

  1349.     c->srcFormat = srcFormat;

  1350.     c->dstFormat = dstFormat;

  1351. 

  1352.     if (param) {

  1353.         c->param[0] = param[0];

  1354.         c->param[1] = param[1];

  1355.     }

  1356.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1357.                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1358.                              c->dstRange, 0, 1 << 16, 1 << 16);

  1359. 

  1360.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1361.         sws_freeContext(c);

  1362.         return NULL;

  1363.     }

  1364. 

  1365.     return c;

  1366. }

  1367. 

  1368. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1369.                                 float lumaSharpen, float chromaSharpen,

  1370.                                 float chromaHShift, float chromaVShift,

  1371.                                 int verbose)

  1372. {

  1373.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1374.     if (!filter)

  1375.         return NULL;

  1376. 

  1377.     if (lumaGBlur != 0.0) {

  1378.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1379.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1380.     } else {

  1381.         filter->lumH = sws_getIdentityVec();

  1382.         filter->lumV = sws_getIdentityVec();

  1383.     }

  1384. 

  1385.     if (chromaGBlur != 0.0) {

  1386.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1387.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1388.     } else {

  1389.         filter->chrH = sws_getIdentityVec();

  1390.         filter->chrV = sws_getIdentityVec();

  1391.     }

  1392. 

  1393.     if (!filter->lumH || !filter->lumV || !filter->chrH || !filter->chrV)

  1394.         goto fail;

  1395. 

  1396.     if (chromaSharpen != 0.0) {

  1397.         SwsVector *id = sws_getIdentityVec();

  1398.         if (!id)

  1399.             goto fail;

  1400.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1401.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1402.         sws_addVec(filter->chrH, id);

  1403.         sws_addVec(filter->chrV, id);

  1404.         sws_freeVec(id);

  1405.     }

  1406. 

  1407.     if (lumaSharpen != 0.0) {

  1408.         SwsVector *id = sws_getIdentityVec();

  1409.         if (!id)

  1410.             goto fail;

  1411.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1412.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1413.         sws_addVec(filter->lumH, id);

  1414.         sws_addVec(filter->lumV, id);

  1415.         sws_freeVec(id);

  1416.     }

  1417. 

  1418.     if (chromaHShift != 0.0)

  1419.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1420. 

  1421.     if (chromaVShift != 0.0)

  1422.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1423. 

  1424.     sws_normalizeVec(filter->chrH, 1.0);

  1425.     sws_normalizeVec(filter->chrV, 1.0);

  1426.     sws_normalizeVec(filter->lumH, 1.0);

  1427.     sws_normalizeVec(filter->lumV, 1.0);

  1428. 

  1429.     if (verbose)

  1430.         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1431.     if (verbose)

  1432.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1433. 

  1434.     return filter;

  1435. 

  1436. fail:

  1437.     sws_freeVec(filter->lumH);

  1438.     sws_freeVec(filter->lumV);

  1439.     sws_freeVec(filter->chrH);

  1440.     sws_freeVec(filter->chrV);

  1441.     av_freep(&filter);

  1442.     return NULL;

  1443. }

  1444. 

  1445. SwsVector *sws_allocVec(int length)

  1446. {

  1447.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1448.     if (!vec)

  1449.         return NULL;

  1450.     vec->length = length;

  1451.     vec->coeff  = av_malloc(sizeof(double) * length);

  1452.     if (!vec->coeff)

  1453.         av_freep(&vec);

  1454.     return vec;

  1455. }

  1456. 

  1457. SwsVector *sws_getGaussianVec(double variance, double quality)

  1458. {

  1459.     const int length = (int)(variance * quality + 0.5) | 1;

  1460.     int i;

  1461.     double middle  = (length - 1) * 0.5;

  1462.     SwsVector *vec = sws_allocVec(length);

  1463. 

  1464.     if (!vec)

  1465.         return NULL;

  1466. 

  1467.     for (i = 0; i < length; i++) {

  1468.         double dist = i - middle;

  1469.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1470.                         sqrt(2 * variance * M_PI);

  1471.     }

  1472. 

  1473.     sws_normalizeVec(vec, 1.0);

  1474. 

  1475.     return vec;

  1476. }

  1477. 

  1478. SwsVector *sws_getConstVec(double c, int length)

  1479. {

  1480.     int i;

  1481.     SwsVector *vec = sws_allocVec(length);

  1482. 

  1483.     if (!vec)

  1484.         return NULL;

  1485. 

  1486.     for (i = 0; i < length; i++)

  1487.         vec->coeff[i] = c;

  1488. 

  1489.     return vec;

  1490. }

  1491. 

  1492. SwsVector *sws_getIdentityVec(void)

  1493. {

  1494.     return sws_getConstVec(1.0, 1);

  1495. }

  1496. 

  1497. static double sws_dcVec(SwsVector *a)

  1498. {

  1499.     int i;

  1500.     double sum = 0;

  1501. 

  1502.     for (i = 0; i < a->length; i++)

  1503.         sum += a->coeff[i];

  1504. 

  1505.     return sum;

  1506. }

  1507. 

  1508. void sws_scaleVec(SwsVector *a, double scalar)

  1509. {

  1510.     int i;

  1511. 

  1512.     for (i = 0; i < a->length; i++)

  1513.         a->coeff[i] *= scalar;

  1514. }

  1515. 

  1516. void sws_normalizeVec(SwsVector *a, double height)

  1517. {

  1518.     sws_scaleVec(a, height / sws_dcVec(a));

  1519. }

  1520. 

  1521. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1522. {

  1523.     int length = a->length + b->length - 1;

  1524.     int i, j;

  1525.     SwsVector *vec = sws_getConstVec(0.0, length);

  1526. 

  1527.     if (!vec)

  1528.         return NULL;

  1529. 

  1530.     for (i = 0; i < a->length; i++) {

  1531.         for (j = 0; j < b->length; j++) {

  1532.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1533.         }

  1534.     }

  1535. 

  1536.     return vec;

  1537. }

  1538. 

  1539. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1540. {

  1541.     int length = FFMAX(a->length, b->length);

  1542.     int i;

  1543.     SwsVector *vec = sws_getConstVec(0.0, length);

  1544. 

  1545.     if (!vec)

  1546.         return NULL;

  1547. 

  1548.     for (i = 0; i < a->length; i++)

  1549.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1550.     for (i = 0; i < b->length; i++)

  1551.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

  1552. 

  1553.     return vec;

  1554. }

  1555. 

  1556. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1557. {

  1558.     int length = FFMAX(a->length, b->length);

  1559.     int i;

  1560.     SwsVector *vec = sws_getConstVec(0.0, length);

  1561. 

  1562.     if (!vec)

  1563.         return NULL;

  1564. 

  1565.     for (i = 0; i < a->length; i++)

  1566.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1567.     for (i = 0; i < b->length; i++)

  1568.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

  1569. 

  1570.     return vec;

  1571. }

  1572. 

  1573. /* shift left / or right if "shift" is negative */

  1574. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1575. {

  1576.     int length = a->length + FFABS(shift) * 2;

  1577.     int i;

  1578.     SwsVector *vec = sws_getConstVec(0.0, length);

  1579. 

  1580.     if (!vec)

  1581.         return NULL;

  1582. 

  1583.     for (i = 0; i < a->length; i++) {

  1584.         vec->coeff[i + (length    - 1) / 2 -

  1585.                        (a->length - 1) / 2 - shift] = a->coeff[i];

  1586.     }

  1587. 

  1588.     return vec;

  1589. }

  1590. 

  1591. void sws_shiftVec(SwsVector *a, int shift)

  1592. {

  1593.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1594.     av_free(a->coeff);

  1595.     a->coeff  = shifted->coeff;

  1596.     a->length = shifted->length;

  1597.     av_free(shifted);

  1598. }

  1599. 

  1600. void sws_addVec(SwsVector *a, SwsVector *b)

  1601. {

  1602.     SwsVector *sum = sws_sumVec(a, b);

  1603.     av_free(a->coeff);

  1604.     a->coeff  = sum->coeff;

  1605.     a->length = sum->length;

  1606.     av_free(sum);

  1607. }

  1608. 

  1609. void sws_subVec(SwsVector *a, SwsVector *b)

  1610. {

  1611.     SwsVector *diff = sws_diffVec(a, b);

  1612.     av_free(a->coeff);

  1613.     a->coeff  = diff->coeff;

  1614.     a->length = diff->length;

  1615.     av_free(diff);

  1616. }

  1617. 

  1618. void sws_convVec(SwsVector *a, SwsVector *b)

  1619. {

  1620.     SwsVector *conv = sws_getConvVec(a, b);

  1621.     av_free(a->coeff);

  1622.     a->coeff  = conv->coeff;

  1623.     a->length = conv->length;

  1624.     av_free(conv);

  1625. }

  1626. 

  1627. SwsVector *sws_cloneVec(SwsVector *a)

  1628. {

  1629.     int i;

  1630.     SwsVector *vec = sws_allocVec(a->length);

  1631. 

  1632.     if (!vec)

  1633.         return NULL;

  1634. 

  1635.     for (i = 0; i < a->length; i++)

  1636.         vec->coeff[i] = a->coeff[i];

  1637. 

  1638.     return vec;

  1639. }

  1640. 

  1641. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1642. {

  1643.     int i;

  1644.     double max = 0;

  1645.     double min = 0;

  1646.     double range;

  1647. 

  1648.     for (i = 0; i < a->length; i++)

  1649.         if (a->coeff[i] > max)

  1650.             max = a->coeff[i];

  1651. 

  1652.     for (i = 0; i < a->length; i++)

  1653.         if (a->coeff[i] < min)

  1654.             min = a->coeff[i];

  1655. 

  1656.     range = max - min;

  1657. 

  1658.     for (i = 0; i < a->length; i++) {

  1659.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1660.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1661.         for (; x > 0; x--)

  1662.             av_log(log_ctx, log_level, " ");

  1663.         av_log(log_ctx, log_level, "|\n");

  1664.     }

  1665. }

  1666. 

  1667. void sws_freeVec(SwsVector *a)

  1668. {

  1669.     if (!a)

  1670.         return;

  1671.     av_freep(&a->coeff);

  1672.     a->length = 0;

  1673.     av_free(a);

  1674. }

  1675. 

  1676. void sws_freeFilter(SwsFilter *filter)

  1677. {

  1678.     if (!filter)

  1679.         return;

  1680. 

  1681.     if (filter->lumH)

  1682.         sws_freeVec(filter->lumH);

  1683.     if (filter->lumV)

  1684.         sws_freeVec(filter->lumV);

  1685.     if (filter->chrH)

  1686.         sws_freeVec(filter->chrH);

  1687.     if (filter->chrV)

  1688.         sws_freeVec(filter->chrV);

  1689.     av_free(filter);

  1690. }

  1691. 

  1692. void sws_freeContext(SwsContext *c)

  1693. {

  1694.     int i;

  1695.     if (!c)

  1696.         return;

  1697. 

  1698.     if (c->lumPixBuf) {

  1699.         for (i = 0; i < c->vLumBufSize; i++)

  1700.             av_freep(&c->lumPixBuf[i]);

  1701.         av_freep(&c->lumPixBuf);

  1702.     }

  1703. 

  1704.     if (c->chrUPixBuf) {

  1705.         for (i = 0; i < c->vChrBufSize; i++)

  1706.             av_freep(&c->chrUPixBuf[i]);

  1707.         av_freep(&c->chrUPixBuf);

  1708.         av_freep(&c->chrVPixBuf);

  1709.     }

  1710. 

  1711.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1712.         for (i = 0; i < c->vLumBufSize; i++)

  1713.             av_freep(&c->alpPixBuf[i]);

  1714.         av_freep(&c->alpPixBuf);

  1715.     }

  1716. 

  1717.     av_freep(&c->vLumFilter);

  1718.     av_freep(&c->vChrFilter);

  1719.     av_freep(&c->hLumFilter);

  1720.     av_freep(&c->hChrFilter);

  1721. #if HAVE_ALTIVEC

  1722.     av_freep(&c->vYCoeffsBank);

  1723.     av_freep(&c->vCCoeffsBank);

  1724. #endif

  1725. 

  1726.     av_freep(&c->vLumFilterPos);

  1727.     av_freep(&c->vChrFilterPos);

  1728.     av_freep(&c->hLumFilterPos);

  1729.     av_freep(&c->hChrFilterPos);

  1730. 

  1731. #if HAVE_MMX_INLINE

  1732. #if USE_MMAP

  1733.     if (c->lumMmxextFilterCode)

  1734.         munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);

  1735.     if (c->chrMmxextFilterCode)

  1736.         munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);

  1737. #elif HAVE_VIRTUALALLOC

  1738.     if (c->lumMmxextFilterCode)

  1739.         VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);

  1740.     if (c->chrMmxextFilterCode)

  1741.         VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);

  1742. #else

  1743.     av_free(c->lumMmxextFilterCode);

  1744.     av_free(c->chrMmxextFilterCode);

  1745. #endif

  1746.     c->lumMmxextFilterCode = NULL;

  1747.     c->chrMmxextFilterCode = NULL;

  1748. #endif /* HAVE_MMX_INLINE */

  1749. 

  1750.     av_freep(&c->yuvTable);

  1751.     av_free(c->formatConvBuffer);

  1752. 

  1753.     av_free(c);

  1754. }

  1755. 

  1756. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  1757.                                         int srcH, enum AVPixelFormat srcFormat,

  1758.                                         int dstW, int dstH,

  1759.                                         enum AVPixelFormat dstFormat, int flags,

  1760.                                         SwsFilter *srcFilter,

  1761.                                         SwsFilter *dstFilter,

  1762.                                         const double *param)

  1763. {

  1764.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  1765.                                              SWS_PARAM_DEFAULT };

  1766. 

  1767.     if (!param)

  1768.         param = default_param;

  1769. 

  1770.     if (context &&

  1771.         (context->srcW      != srcW      ||

  1772.          context->srcH      != srcH      ||

  1773.          context->srcFormat != srcFormat ||

  1774.          context->dstW      != dstW      ||

  1775.          context->dstH      != dstH      ||

  1776.          context->dstFormat != dstFormat ||

  1777.          context->flags     != flags     ||

  1778.          context->param[0]  != param[0]  ||

  1779.          context->param[1]  != param[1])) {

  1780.         sws_freeContext(context);

  1781.         context = NULL;

  1782.     }

  1783. 

  1784.     if (!context) {

  1785.         if (!(context = sws_alloc_context()))

  1786.             return NULL;

  1787.         context->srcW      = srcW;

  1788.         context->srcH      = srcH;

  1789.         context->srcRange  = handle_jpeg(&srcFormat);

  1790.         context->srcFormat = srcFormat;

  1791.         context->dstW      = dstW;

  1792.         context->dstH      = dstH;

  1793.         context->dstRange  = handle_jpeg(&dstFormat);

  1794.         context->dstFormat = dstFormat;

  1795.         context->flags     = flags;

  1796.         context->param[0]  = param[0];

  1797.         context->param[1]  = param[1];

  1798.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  1799.                                  context->srcRange,

  1800.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1801.                                  context->dstRange, 0, 1 << 16, 1 << 16);

  1802.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1803.             sws_freeContext(context);

  1804.             return NULL;

  1805.         }

  1806.     }

  1807.     return context;

  1808. }