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.
54768 Commits
32 Branches
301MB
Tree: 37f4aa133d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '37f4aa133d'
${ noResults }
FFmpeg/libswscale/utils.c

2045 lines
74KB
Raw Blame History 

  1. /*

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

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. #include "config.h"

  22. 

  23. #define _SVID_SOURCE // needed for MAP_ANONYMOUS

  24. #define _DARWIN_C_SOURCE // needed for MAP_ANON

  25. #include <inttypes.h>

  26. #include <math.h>

  27. #include <stdio.h>

  28. #include <string.h>

  29. #if HAVE_SYS_MMAN_H

  30. #include <sys/mman.h>

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

  32. #define MAP_ANONYMOUS MAP_ANON

  33. #endif

  34. #endif

  35. #if HAVE_VIRTUALALLOC

  36. #define WIN32_LEAN_AND_MEAN

  37. #include <windows.h>

  38. #endif

  39. 

  40. #include "libavutil/attributes.h"

  41. #include "libavutil/avassert.h"

  42. #include "libavutil/avutil.h"

  43. #include "libavutil/bswap.h"

  44. #include "libavutil/cpu.h"

  45. #include "libavutil/intreadwrite.h"

  46. #include "libavutil/mathematics.h"

  47. #include "libavutil/opt.h"

  48. #include "libavutil/pixdesc.h"

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

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

  51. #include "rgb2rgb.h"

  52. #include "swscale.h"

  53. #include "swscale_internal.h"

  54. 

  55. static void handle_formats(SwsContext *c);

  56. 

  57. unsigned swscale_version(void)

  58. {

  59.     av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);

  60.     return LIBSWSCALE_VERSION_INT;

  61. }

  62. 

  63. const char *swscale_configuration(void)

  64. {

  65.     return FFMPEG_CONFIGURATION;

  66. }

  67. 

  68. const char *swscale_license(void)

  69. {

  70. #define LICENSE_PREFIX "libswscale license: "

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

  72. }

  73. 

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

  75. 

  76. typedef struct FormatEntry {

  77.     uint8_t is_supported_in         :1;

  78.     uint8_t is_supported_out        :1;

  79.     uint8_t is_supported_endianness :1;

  80. } FormatEntry;

  81. 

  82. static const FormatEntry format_entries[AV_PIX_FMT_NB] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  209. };

  210. 

  211. int sws_isSupportedInput(enum AVPixelFormat pix_fmt)

  212. {

  213.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  214.            format_entries[pix_fmt].is_supported_in : 0;

  215. }

  216. 

  217. int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)

  218. {

  219.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  220.            format_entries[pix_fmt].is_supported_out : 0;

  221. }

  222. 

  223. int sws_isSupportedEndiannessConversion(enum AVPixelFormat pix_fmt)

  224. {

  225.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  226.            format_entries[pix_fmt].is_supported_endianness : 0;

  227. }

  228. 

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

  230. 

  231. #if FF_API_SWS_FORMAT_NAME

  232. const char *sws_format_name(enum AVPixelFormat format)

  233. {

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

  235.     if (desc)

  236.         return desc->name;

  237.     else

  238.         return "Unknown format";

  239. }

  240. #endif

  241. 

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

  243.                              double dist)

  244. {

  245.     if (dist <= 1.0)

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

  247.     else

  248.         return getSplineCoeff(0.0,

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

  250.                                c + 3.0 * d,

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

  252.                               dist - 1.0);

  253. }

  254. 

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

  256.                               int *outFilterSize, int xInc, int srcW,

  257.                               int dstW, int filterAlign, int one,

  258.                               int flags, int cpu_flags,

  259.                               SwsVector *srcFilter, SwsVector *dstFilter,

  260.                               double param[2])

  261. {

  262.     int i;

  263.     int filterSize;

  264.     int filter2Size;

  265.     int minFilterSize;

  266.     int64_t *filter    = NULL;

  267.     int64_t *filter2   = NULL;

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

  269.     int ret            = -1;

  270. 

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

  272. 

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

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

  275. 

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

  277.         int i;

  278.         filterSize = 1;

  279.         FF_ALLOCZ_OR_GOTO(NULL, filter,

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

  281. 

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

  283.             filter[i * filterSize] = fone;

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

  285.         }

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

  287.         int i;

  288.         int64_t xDstInSrc;

  289.         filterSize = 1;

  290.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  292. 

  293.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  296. 

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

  298.             filter[i]       = fone;

  299.             xDstInSrc      += xInc;

  300.         }

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

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

  303.         int i;

  304.         int64_t xDstInSrc;

  305.         filterSize = 2;

  306.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  308. 

  309.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  312.             int j;

  313. 

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

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

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

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

  318.                 if (coeff < 0)

  319.                     coeff = 0;

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

  321.                 xx++;

  322.             }

  323.             xDstInSrc += xInc;

  324.         }

  325.     } else {

  326.         int64_t xDstInSrc;

  327.         int sizeFactor;

  328. 

  329.         if (flags & SWS_BICUBIC)

  330.             sizeFactor = 4;

  331.         else if (flags & SWS_X)

  332.             sizeFactor = 8;

  333.         else if (flags & SWS_AREA)

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

  335.         else if (flags & SWS_GAUSS)

  336.             sizeFactor = 8;     // infinite ;)

  337.         else if (flags & SWS_LANCZOS)

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

  339.         else if (flags & SWS_SINC)

  340.             sizeFactor = 20;    // infinite ;)

  341.         else if (flags & SWS_SPLINE)

  342.             sizeFactor = 20;    // infinite ;)

  343.         else if (flags & SWS_BILINEAR)

  344.             sizeFactor = 2;

  345.         else {

  346.             av_assert0(0);

  347.         }

  348. 

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

  350.             filterSize = 1 + sizeFactor;    // upscale

  351.         else

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

  353. 

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

  355.         filterSize = FFMAX(filterSize, 1);

  356. 

  357.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  359. 

  360.         xDstInSrc = xInc - 0x10000;

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

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

  363.             int j;

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

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

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

  367.                 double floatd;

  368.                 int64_t coeff;

  369. 

  370.                 if (xInc > 1 << 16)

  371.                     d = d * dstW / srcW;

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

  373. 

  374.                 if (flags & SWS_BICUBIC) {

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

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

  377. 

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

  379.                         coeff = 0.0;

  380.                     } else {

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

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

  383. 

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

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

  386.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  387.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  388.                         else

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

  390.                                       (6 * B + 30 * C) * dd  +

  391.                                     (-12 * B - 48 * C) * d   +

  392.                                       (8 * B + 24 * C) * (1 << 30);

  393.                     }

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

  395.                 }

  396. #if 0

  397.                 else if (flags & SWS_X) {

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

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

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

  401.                 }

  402. #endif

  403.                 else if (flags & SWS_X) {

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

  405.                     double c;

  406. 

  407.                     if (floatd < 1.0)

  408.                         c = cos(floatd * M_PI);

  409.                     else

  410.                         c = -1.0;

  411.                     if (c < 0.0)

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

  413.                     else

  414.                         c = pow(c, A);

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

  416.                 } else if (flags & SWS_AREA) {

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

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

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

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

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

  422.                     else

  423.                         coeff = 0.0;

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

  425.                 } else if (flags & SWS_GAUSS) {

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

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

  428.                 } else if (flags & SWS_SINC) {

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

  430.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  434.                     if (floatd > p)

  435.                         coeff = 0;

  436.                 } else if (flags & SWS_BILINEAR) {

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

  438.                     if (coeff < 0)

  439.                         coeff = 0;

  440.                     coeff *= fone >> 30;

  441.                 } else if (flags & SWS_SPLINE) {

  442.                     double p = -2.196152422706632;

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

  444.                 } else {

  445.                     av_assert0(0);

  446.                 }

  447. 

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

  449.                 xx++;

  450.             }

  451.             xDstInSrc += 2 * xInc;

  452.         }

  453.     }

  454. 

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

  456.      * av_free(filter);

  457.      */

  458.     av_assert0(filterSize > 0);

  459.     filter2Size = filterSize;

  460.     if (srcFilter)

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

  462.     if (dstFilter)

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

  464.     av_assert0(filter2Size > 0);

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

  466. 

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

  468.         int j, k;

  469. 

  470.         if (srcFilter) {

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

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

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

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

  475.             }

  476.         } else {

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

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

  479.         }

  480.         // FIXME dstFilter

  481. 

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

  483.     }

  484.     av_freep(&filter);

  485. 

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

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

  488.     minFilterSize = 0;

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

  490.         int min = filter2Size;

  491.         int j;

  492.         int64_t cutOff = 0.0;

  493. 

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

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

  496.             int k;

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

  498. 

  499.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  500.                 break;

  501. 

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

  503.              * filter otherwise */

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

  505.                 break;

  506. 

  507.             // move filter coefficients left

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

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

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

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

  512.         }

  513. 

  514.         cutOff = 0;

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

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

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

  518. 

  519.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  520.                 break;

  521.             min--;

  522.         }

  523. 

  524.         if (min > minFilterSize)

  525.             minFilterSize = min;

  526.     }

  527. 

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

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

  530.         if (minFilterSize < 5)

  531.             filterAlign = 4;

  532. 

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

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

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

  536.         if (minFilterSize < 3)

  537.             filterAlign = 1;

  538.     }

  539. 

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

  541.         // special case for unscaled vertical filtering

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

  543.             filterAlign = 1;

  544.     }

  545. 

  546.     av_assert0(minFilterSize > 0);

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

  548.     av_assert0(filterSize > 0);

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

  550.     if (filterSize >= MAX_FILTER_SIZE * 16 /

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

  552.         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);

  553.         goto fail;

  554.     }

  555.     *outFilterSize = filterSize;

  556. 

  557.     if (flags & SWS_PRINT_INFO)

  558.         av_log(NULL, AV_LOG_VERBOSE,

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

  560.                filter2Size, filterSize);

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

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

  563.         int j;

  564. 

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

  566.             if (j >= filter2Size)

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

  568.             else

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

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

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

  572.         }

  573.     }

  574. 

  575.     // FIXME try to align filterPos if possible

  576. 

  577.     // fix borders

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

  579.         int j;

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

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

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

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

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

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

  586.             }

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

  588.         }

  589. 

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

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

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

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

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

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

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

  597.             }

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

  599.         }

  600.     }

  601. 

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

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

  604.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

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

  606. 

  607.     /* normalize & store in outFilter */

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

  609.         int j;

  610.         int64_t error = 0;

  611.         int64_t sum   = 0;

  612. 

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

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

  615.         }

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

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

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

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

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

  621.             error                                  = v - intV * sum;

  622.         }

  623.     }

  624. 

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

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

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

  628.                                                       * read over the end */

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

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

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

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

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

  634.     }

  635. 

  636.     ret = 0;

  637. 

  638. fail:

  639.     if(ret < 0)

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

  641.     av_free(filter);

  642.     av_free(filter2);

  643.     return ret;

  644. }

  645. 

  646. #if HAVE_MMXEXT_INLINE

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

  648.                                        int16_t *filter, int32_t *filterPos,

  649.                                        int numSplits)

  650. {

  651.     uint8_t *fragmentA;

  652.     x86_reg imm8OfPShufW1A;

  653.     x86_reg imm8OfPShufW2A;

  654.     x86_reg fragmentLengthA;

  655.     uint8_t *fragmentB;

  656.     x86_reg imm8OfPShufW1B;

  657.     x86_reg imm8OfPShufW2B;

  658.     x86_reg fragmentLengthB;

  659.     int fragmentPos;

  660. 

  661.     int xpos, i;

  662. 

  663.     // create an optimized horizontal scaling routine

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

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

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

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

  668.      */

  669. 

  670.     // code fragment

  671. 

  672.     __asm__ volatile (

  673.         "jmp                         9f                 \n\t"

  674.         // Begin

  675.         "0:                                             \n\t"

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

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

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

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

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

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

  682.         "1:                                             \n\t"

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

  684.         "2:                                             \n\t"

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

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

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

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

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

  690. 

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

  692. 

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

  694.         // End

  695.         "9:                                             \n\t"

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

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

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

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

  700.         "dec                         %1                 \n\t"

  701.         "dec                         %2                 \n\t"

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

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

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

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

  706. 

  707. 

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

  709.           "=r" (fragmentLengthA)

  710.         );

  711. 

  712.     __asm__ volatile (

  713.         "jmp                         9f                 \n\t"

  714.         // Begin

  715.         "0:                                             \n\t"

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

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

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

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

  720.         "1:                                             \n\t"

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

  722.         "2:                                             \n\t"

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

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

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

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

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

  728. 

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

  730. 

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

  732.         // End

  733.         "9:                                             \n\t"

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

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

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

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

  738.         "dec                         %1                 \n\t"

  739.         "dec                         %2                 \n\t"

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

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

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

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

  744. 

  745. 

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

  747.           "=r" (fragmentLengthB)

  748.         );

  749. 

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

  751.     fragmentPos = 0;

  752. 

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

  754.         int xx = xpos >> 16;

  755. 

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

  757.             int a                  = 0;

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

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

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

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

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

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

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

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

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

  767.             int shift              = 0;

  768. 

  769.             if (filterCode) {

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

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

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

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

  774.                 filterPos[i / 2] = xx;

  775. 

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

  777. 

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

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

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

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

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

  783.                                                                (c << 4) |

  784.                                                                (d << 6);

  785. 

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

  787.                     shift = maxShift;               // avoid overread

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

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

  790. 

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

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

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

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

  795.                 }

  796.             }

  797. 

  798.             fragmentPos += fragmentLength;

  799. 

  800.             if (filterCode)

  801.                 filterCode[fragmentPos] = RET;

  802.         }

  803.         xpos += xInc;

  804.     }

  805.     if (filterCode)

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

  807. 

  808.     return fragmentPos + 1;

  809. }

  810. #endif /* HAVE_MMXEXT_INLINE */

  811. 

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

  813. {

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

  815.     *h = desc->log2_chroma_w;

  816.     *v = desc->log2_chroma_h;

  817. }

  818. 

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

  820. {

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

  822.     int64_t vr =  table[0];

  823.     int64_t ub =  table[1];

  824.     int64_t ug = -table[2];

  825.     int64_t vg = -table[3];

  826.     int64_t ONE = 65536;

  827.     int64_t cy = ONE;

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

  829.     int i;

  830.     static const int8_t map[] = {

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

  832.     RY_IDX, -1    , GY_IDX, RY_IDX, RY_IDX, -1    , GY_IDX, RY_IDX,

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

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

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

  836.     RU_IDX, -1    , GU_IDX, RU_IDX, RU_IDX, -1    , GU_IDX, RU_IDX,

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

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

  839.     BV_IDX, GV_IDX, -1    , BV_IDX, BV_IDX, GV_IDX, -1    , BV_IDX,

  840.     RV_IDX, -1    , GV_IDX, RV_IDX, RV_IDX, -1    , GV_IDX, RV_IDX,

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

  842.     BV_IDX, -1    , GV_IDX, BV_IDX, BV_IDX, -1    , GV_IDX, BV_IDX,

  843.     RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,

  844.     BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,

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

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

  847.     RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,

  848.     BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,

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

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

  851.     RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,

  852.     BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,

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

  854.     -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, //23

  855.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //24

  856.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //25

  857.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //26

  858.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //27

  859.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //28

  860.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //29

  861.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //30

  862.     -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //31

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

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

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

  866.     };

  867. 

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

  869. 

  870.     if (!dstRange) {

  871.         cy = cy * 255 / 219;

  872.     } else {

  873.         vr = vr * 224 / 255;

  874.         ub = ub * 224 / 255;

  875.         ug = ug * 224 / 255;

  876.         vg = vg * 224 / 255;

  877.     }

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

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

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

  881. 

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

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

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

  885. 

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

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

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

  889. 

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

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

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

  893. 

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

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

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

  897. 

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

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

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

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

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

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

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

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

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

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

  908.     }

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

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

  911. }

  912. 

  913. static void fill_xyztables(struct SwsContext *c)

  914. {

  915.     int i;

  916.     double xyzgamma = XYZ_GAMMA;

  917.     double rgbgamma = 1.0 / RGB_GAMMA;

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

  919.         {13270, -6295, -2041},

  920.         {-3969,  7682,   170},

  921.         {  228,  -835,  4329} };

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

  923. 

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

  925.     c->xyzgamma = xyzgamma_tab;

  926.     c->rgbgamma = rgbgamma_tab;

  927. 

  928.     if (rgbgamma_tab[4095])

  929.         return;

  930. 

  931.     /* set gamma vectors */

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

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

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

  935.     }

  936. }

  937. 

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

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

  940.                              int brightness, int contrast, int saturation)

  941. {

  942.     const AVPixFmtDescriptor *desc_dst;

  943.     const AVPixFmtDescriptor *desc_src;

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

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

  946. 

  947.     handle_formats(c);

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

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

  950. 

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

  952.         dstRange = 0;

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

  954.         srcRange = 0;

  955. 

  956.     c->brightness = brightness;

  957.     c->contrast   = contrast;

  958.     c->saturation = saturation;

  959.     c->srcRange   = srcRange;

  960.     c->dstRange   = dstRange;

  961. 

  962.     fill_xyztables(c);

  963. 

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

  965.         return -1;

  966. 

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

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

  969. 

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

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

  972.                              contrast, saturation);

  973.     // FIXME factorize

  974. 

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

  976.         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,

  977.                                        contrast, saturation);

  978.     }

  979. 

  980.     fill_rgb2yuv_table(c, table, dstRange);

  981. 

  982.     return 0;

  983. }

  984. 

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

  986.                              int *srcRange, int **table, int *dstRange,

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

  988. {

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

  990.         return -1;

  991. 

  992.     *inv_table  = c->srcColorspaceTable;

  993.     *table      = c->dstColorspaceTable;

  994.     *srcRange   = c->srcRange;

  995.     *dstRange   = c->dstRange;

  996.     *brightness = c->brightness;

  997.     *contrast   = c->contrast;

  998.     *saturation = c->saturation;

  999. 

  1000.     return 0;

  1001. }

  1002. 

  1003. static int handle_jpeg(enum AVPixelFormat *format)

  1004. {

  1005.     switch (*format) {

  1006.     case AV_PIX_FMT_YUVJ420P:

  1007.         *format = AV_PIX_FMT_YUV420P;

  1008.         return 1;

  1009.     case AV_PIX_FMT_YUVJ411P:

  1010.         *format = AV_PIX_FMT_YUV411P;

  1011.         return 1;

  1012.     case AV_PIX_FMT_YUVJ422P:

  1013.         *format = AV_PIX_FMT_YUV422P;

  1014.         return 1;

  1015.     case AV_PIX_FMT_YUVJ444P:

  1016.         *format = AV_PIX_FMT_YUV444P;

  1017.         return 1;

  1018.     case AV_PIX_FMT_YUVJ440P:

  1019.         *format = AV_PIX_FMT_YUV440P;

  1020.         return 1;

  1021.     case AV_PIX_FMT_GRAY8:

  1022.         return 1;

  1023.     default:

  1024.         return 0;

  1025.     }

  1026. }

  1027. 

  1028. static int handle_0alpha(enum AVPixelFormat *format)

  1029. {

  1030.     switch (*format) {

  1031.     case AV_PIX_FMT_0BGR    : *format = AV_PIX_FMT_ABGR   ; return 1;

  1032.     case AV_PIX_FMT_BGR0    : *format = AV_PIX_FMT_BGRA   ; return 4;

  1033.     case AV_PIX_FMT_0RGB    : *format = AV_PIX_FMT_ARGB   ; return 1;

  1034.     case AV_PIX_FMT_RGB0    : *format = AV_PIX_FMT_RGBA   ; return 4;

  1035.     default:                                          return 0;

  1036.     }

  1037. }

  1038. 

  1039. static int handle_xyz(enum AVPixelFormat *format)

  1040. {

  1041.     switch (*format) {

  1042.     case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;

  1043.     case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;

  1044.     default:                                                return 0;

  1045.     }

  1046. }

  1047. 

  1048. static void handle_formats(SwsContext *c)

  1049. {

  1050.     c->src0Alpha |= handle_0alpha(&c->srcFormat);

  1051.     c->dst0Alpha |= handle_0alpha(&c->dstFormat);

  1052.     c->srcXYZ    |= handle_xyz(&c->srcFormat);

  1053.     c->dstXYZ    |= handle_xyz(&c->dstFormat);

  1054. }

  1055. 

  1056. SwsContext *sws_alloc_context(void)

  1057. {

  1058.     SwsContext *c = av_mallocz(sizeof(SwsContext));

  1059. 

  1060.     if (c) {

  1061.         c->av_class = &sws_context_class;

  1062.         av_opt_set_defaults(c);

  1063.     }

  1064. 

  1065.     return c;

  1066. }

  1067. 

  1068. av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,

  1069.                              SwsFilter *dstFilter)

  1070. {

  1071.     int i, j;

  1072.     int usesVFilter, usesHFilter;

  1073.     int unscaled;

  1074.     SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };

  1075.     int srcW              = c->srcW;

  1076.     int srcH              = c->srcH;

  1077.     int dstW              = c->dstW;

  1078.     int dstH              = c->dstH;

  1079.     int dst_stride        = FFALIGN(dstW * sizeof(int16_t) + 66, 16);

  1080.     int flags, cpu_flags;

  1081.     enum AVPixelFormat srcFormat = c->srcFormat;

  1082.     enum AVPixelFormat dstFormat = c->dstFormat;

  1083.     const AVPixFmtDescriptor *desc_src;

  1084.     const AVPixFmtDescriptor *desc_dst;

  1085. 

  1086.     cpu_flags = av_get_cpu_flags();

  1087.     flags     = c->flags;

  1088.     emms_c();

  1089.     if (!rgb15to16)

  1090.         sws_rgb2rgb_init();

  1091. 

  1092.     unscaled = (srcW == dstW && srcH == dstH);

  1093. 

  1094.     c->srcRange |= handle_jpeg(&c->srcFormat);

  1095.     c->dstRange |= handle_jpeg(&c->dstFormat);

  1096. 

  1097.     if (!c->contrast && !c->saturation && !c->dstFormatBpp)

  1098.         sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1099.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  1100.                                  c->dstRange, 0, 1 << 16, 1 << 16);

  1101. 

  1102.     if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat)

  1103.         av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");

  1104.     handle_formats(c);

  1105.     srcFormat = c->srcFormat;

  1106.     dstFormat = c->dstFormat;

  1107.     desc_src = av_pix_fmt_desc_get(srcFormat);

  1108.     desc_dst = av_pix_fmt_desc_get(dstFormat);

  1109. 

  1110.     if (!(unscaled && sws_isSupportedEndiannessConversion(srcFormat) &&

  1111.           av_pix_fmt_swap_endianness(srcFormat) == dstFormat)) {

  1112.     if (!sws_isSupportedInput(srcFormat)) {

  1113.         av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",

  1114.                av_get_pix_fmt_name(srcFormat));

  1115.         return AVERROR(EINVAL);

  1116.     }

  1117.     if (!sws_isSupportedOutput(dstFormat)) {

  1118.         av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",

  1119.                av_get_pix_fmt_name(dstFormat));

  1120.         return AVERROR(EINVAL);

  1121.     }

  1122.     }

  1123. 

  1124.     i = flags & (SWS_POINT         |

  1125.                  SWS_AREA          |

  1126.                  SWS_BILINEAR      |

  1127.                  SWS_FAST_BILINEAR |

  1128.                  SWS_BICUBIC       |

  1129.                  SWS_X             |

  1130.                  SWS_GAUSS         |

  1131.                  SWS_LANCZOS       |

  1132.                  SWS_SINC          |

  1133.                  SWS_SPLINE        |

  1134.                  SWS_BICUBLIN);

  1135.     if (!i || (i & (i - 1))) {

  1136.         av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);

  1137.         return AVERROR(EINVAL);

  1138.     }

  1139.     /* sanity check */

  1140.     if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {

  1141.         /* FIXME check if these are enough and try to lower them after

  1142.          * fixing the relevant parts of the code */

  1143.         av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",

  1144.                srcW, srcH, dstW, dstH);

  1145.         return AVERROR(EINVAL);

  1146.     }

  1147. 

  1148.     if (!dstFilter)

  1149.         dstFilter = &dummyFilter;

  1150.     if (!srcFilter)

  1151.         srcFilter = &dummyFilter;

  1152. 

  1153.     c->lumXInc      = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;

  1154.     c->lumYInc      = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;

  1155.     c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);

  1156.     c->srcFormatBpp = av_get_bits_per_pixel(desc_src);

  1157.     c->vRounder     = 4 * 0x0001000100010001ULL;

  1158. 

  1159.     usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||

  1160.                   (srcFilter->chrV && srcFilter->chrV->length > 1) ||

  1161.                   (dstFilter->lumV && dstFilter->lumV->length > 1) ||

  1162.                   (dstFilter->chrV && dstFilter->chrV->length > 1);

  1163.     usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||

  1164.                   (srcFilter->chrH && srcFilter->chrH->length > 1) ||

  1165.                   (dstFilter->lumH && dstFilter->lumH->length > 1) ||

  1166.                   (dstFilter->chrH && dstFilter->chrH->length > 1);

  1167. 

  1168.     getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);

  1169.     getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);

  1170. 

  1171.     if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {

  1172.         if (dstW&1) {

  1173.             av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");

  1174.             flags |= SWS_FULL_CHR_H_INT;

  1175.             c->flags = flags;

  1176.         }

  1177.     }

  1178. 

  1179.     if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||

  1180.        dstFormat == AV_PIX_FMT_RGB4_BYTE ||

  1181.        dstFormat == AV_PIX_FMT_BGR8 ||

  1182.        dstFormat == AV_PIX_FMT_RGB8) {

  1183.         if (flags & SWS_ERROR_DIFFUSION && !(flags & SWS_FULL_CHR_H_INT)) {

  1184.             av_log(c, AV_LOG_DEBUG,

  1185.                 "Error diffusion dither is only supported in full chroma interpolation for destination format '%s'\n",

  1186.                 av_get_pix_fmt_name(dstFormat));

  1187.             flags   |= SWS_FULL_CHR_H_INT;

  1188.             c->flags = flags;

  1189.         }

  1190.         if (!(flags & SWS_ERROR_DIFFUSION) && (flags & SWS_FULL_CHR_H_INT)) {

  1191.             av_log(c, AV_LOG_DEBUG,

  1192.                 "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",

  1193.                 av_get_pix_fmt_name(dstFormat));

  1194.             flags   |= SWS_ERROR_DIFFUSION;

  1195.             c->flags = flags;

  1196.         }

  1197.     }

  1198.     if (isPlanarRGB(dstFormat)) {

  1199.         if (!(flags & SWS_FULL_CHR_H_INT)) {

  1200.             av_log(c, AV_LOG_DEBUG,

  1201.                    "%s output is not supported with half chroma resolution, switching to full\n",

  1202.                    av_get_pix_fmt_name(dstFormat));

  1203.             flags   |= SWS_FULL_CHR_H_INT;

  1204.             c->flags = flags;

  1205.         }

  1206.     }

  1207. 

  1208.     /* reuse chroma for 2 pixels RGB/BGR unless user wants full

  1209.      * chroma interpolation */

  1210.     if (flags & SWS_FULL_CHR_H_INT &&

  1211.         isAnyRGB(dstFormat)        &&

  1212.         !isPlanarRGB(dstFormat)    &&

  1213.         dstFormat != AV_PIX_FMT_RGBA  &&

  1214.         dstFormat != AV_PIX_FMT_ARGB  &&

  1215.         dstFormat != AV_PIX_FMT_BGRA  &&

  1216.         dstFormat != AV_PIX_FMT_ABGR  &&

  1217.         dstFormat != AV_PIX_FMT_RGB24 &&

  1218.         dstFormat != AV_PIX_FMT_BGR24 &&

  1219.         dstFormat != AV_PIX_FMT_BGR4_BYTE &&

  1220.         dstFormat != AV_PIX_FMT_RGB4_BYTE &&

  1221.         dstFormat != AV_PIX_FMT_BGR8 &&

  1222.         dstFormat != AV_PIX_FMT_RGB8

  1223.     ) {

  1224.         av_log(c, AV_LOG_WARNING,

  1225.                "full chroma interpolation for destination format '%s' not yet implemented\n",

  1226.                av_get_pix_fmt_name(dstFormat));

  1227.         flags   &= ~SWS_FULL_CHR_H_INT;

  1228.         c->flags = flags;

  1229.     }

  1230.     if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))

  1231.         c->chrDstHSubSample = 1;

  1232. 

  1233.     // drop some chroma lines if the user wants it

  1234.     c->vChrDrop          = (flags & SWS_SRC_V_CHR_DROP_MASK) >>

  1235.                            SWS_SRC_V_CHR_DROP_SHIFT;

  1236.     c->chrSrcVSubSample += c->vChrDrop;

  1237. 

  1238.     /* drop every other pixel for chroma calculation unless user

  1239.      * wants full chroma */

  1240.     if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP)   &&

  1241.         srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&

  1242.         srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&

  1243.         srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&

  1244.         srcFormat != AV_PIX_FMT_GBRP9BE   && srcFormat != AV_PIX_FMT_GBRP9LE  &&

  1245.         srcFormat != AV_PIX_FMT_GBRP10BE  && srcFormat != AV_PIX_FMT_GBRP10LE &&

  1246.         srcFormat != AV_PIX_FMT_GBRP12BE  && srcFormat != AV_PIX_FMT_GBRP12LE &&

  1247.         srcFormat != AV_PIX_FMT_GBRP14BE  && srcFormat != AV_PIX_FMT_GBRP14LE &&

  1248.         srcFormat != AV_PIX_FMT_GBRP16BE  && srcFormat != AV_PIX_FMT_GBRP16LE &&

  1249.         ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||

  1250.          (flags & SWS_FAST_BILINEAR)))

  1251.         c->chrSrcHSubSample = 1;

  1252. 

  1253.     // Note the FF_CEIL_RSHIFT is so that we always round toward +inf.

  1254.     c->chrSrcW = FF_CEIL_RSHIFT(srcW, c->chrSrcHSubSample);

  1255.     c->chrSrcH = FF_CEIL_RSHIFT(srcH, c->chrSrcVSubSample);

  1256.     c->chrDstW = FF_CEIL_RSHIFT(dstW, c->chrDstHSubSample);

  1257.     c->chrDstH = FF_CEIL_RSHIFT(dstH, c->chrDstVSubSample);

  1258. 

  1259.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);

  1260. 

  1261.     /* unscaled special cases */

  1262.     if (unscaled && !usesHFilter && !usesVFilter &&

  1263.         (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {

  1264.         ff_get_unscaled_swscale(c);

  1265. 

  1266.         if (c->swScale) {

  1267.             if (flags & SWS_PRINT_INFO)

  1268.                 av_log(c, AV_LOG_INFO,

  1269.                        "using unscaled %s -> %s special converter\n",

  1270.                        av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));

  1271.             return 0;

  1272.         }

  1273.     }

  1274. 

  1275.     c->srcBpc = 1 + desc_src->comp[0].depth_minus1;

  1276.     if (c->srcBpc < 8)

  1277.         c->srcBpc = 8;

  1278.     c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;

  1279.     if (c->dstBpc < 8)

  1280.         c->dstBpc = 8;

  1281.     if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)

  1282.         c->srcBpc = 16;

  1283.     if (c->dstBpc == 16)

  1284.         dst_stride <<= 1;

  1285. 

  1286.     if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {

  1287.         c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&

  1288.                               (srcW & 15) == 0) ? 1 : 0;

  1289.         if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0

  1290. 

  1291.             && (flags & SWS_FAST_BILINEAR)) {

  1292.             if (flags & SWS_PRINT_INFO)

  1293.                 av_log(c, AV_LOG_INFO,

  1294.                        "output width is not a multiple of 32 -> no MMXEXT scaler\n");

  1295.         }

  1296.         if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))

  1297.             c->canMMXEXTBeUsed = 0;

  1298.     } else

  1299.         c->canMMXEXTBeUsed = 0;

  1300. 

  1301.     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;

  1302.     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

  1303. 

  1304.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  1305.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  1306.      * correct scaling.

  1307.      * n-2 is the last chrominance sample available.

  1308.      * This is not perfect, but no one should notice the difference, the more

  1309.      * correct variant would be like the vertical one, but that would require

  1310.      * some special code for the first and last pixel */

  1311.     if (flags & SWS_FAST_BILINEAR) {

  1312.         if (c->canMMXEXTBeUsed) {

  1313.             c->lumXInc += 20;

  1314.             c->chrXInc += 20;

  1315.         }

  1316.         // we don't use the x86 asm scaler if MMX is available

  1317.         else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {

  1318.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1319.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1320.         }

  1321.     }

  1322. 

  1323. #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)

  1324. 

  1325.     /* precalculate horizontal scaler filter coefficients */

  1326.     {

  1327. #if HAVE_MMXEXT_INLINE

  1328. // can't downscale !!!

  1329.         if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {

  1330.             c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,

  1331.                                                              NULL, NULL, 8);

  1332.             c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,

  1333.                                                              NULL, NULL, NULL, 4);

  1334. 

  1335. #if USE_MMAP

  1336.             c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,

  1337.                                           PROT_READ | PROT_WRITE,

  1338.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1339.                                           -1, 0);

  1340.             c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,

  1341.                                           PROT_READ | PROT_WRITE,

  1342.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1343.                                           -1, 0);

  1344. #elif HAVE_VIRTUALALLOC

  1345.             c->lumMmxextFilterCode = VirtualAlloc(NULL,

  1346.                                                   c->lumMmxextFilterCodeSize,

  1347.                                                   MEM_COMMIT,

  1348.                                                   PAGE_EXECUTE_READWRITE);

  1349.             c->chrMmxextFilterCode = VirtualAlloc(NULL,

  1350.                                                   c->chrMmxextFilterCodeSize,

  1351.                                                   MEM_COMMIT,

  1352.                                                   PAGE_EXECUTE_READWRITE);

  1353. #else

  1354.             c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);

  1355.             c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);

  1356. #endif

  1357. 

  1358. #ifdef MAP_ANONYMOUS

  1359.             if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)

  1360. #else

  1361.             if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)

  1362. #endif

  1363.             {

  1364.                 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");

  1365.                 return AVERROR(ENOMEM);

  1366.             }

  1367. 

  1368.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1369.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1370.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1371.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1372. 

  1373.             init_hscaler_mmxext(      dstW, c->lumXInc, c->lumMmxextFilterCode,

  1374.                                 c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);

  1375.             init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,

  1376.                                 c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);

  1377. 

  1378. #if USE_MMAP

  1379.             mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1380.             mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1381. #endif

  1382.         } else

  1383. #endif /* HAVE_MMXEXT_INLINE */

  1384.         {

  1385.             const int filterAlign =

  1386.                 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :

  1387.                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1388.                 1;

  1389. 

  1390.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1391.                            &c->hLumFilterSize, c->lumXInc,

  1392.                            srcW, dstW, filterAlign, 1 << 14,

  1393.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1394.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1395.                            c->param) < 0)

  1396.                 goto fail;

  1397.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1398.                            &c->hChrFilterSize, c->chrXInc,

  1399.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1400.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1401.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1402.                            c->param) < 0)

  1403.                 goto fail;

  1404.         }

  1405.     } // initialize horizontal stuff

  1406. 

  1407.     /* precalculate vertical scaler filter coefficients */

  1408.     {

  1409.         const int filterAlign =

  1410.             (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :

  1411.             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1412.             1;

  1413. 

  1414.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1415.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

  1416.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1417.                        cpu_flags, srcFilter->lumV, dstFilter->lumV,

  1418.                        c->param) < 0)

  1419.             goto fail;

  1420.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1421.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1422.                        filterAlign, (1 << 12),

  1423.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1424.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1425.                        c->param) < 0)

  1426.             goto fail;

  1427. 

  1428. #if HAVE_ALTIVEC

  1429.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1430.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1431. 

  1432.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1433.             int j;

  1434.             short *p = (short *)&c->vYCoeffsBank[i];

  1435.             for (j = 0; j < 8; j++)

  1436.                 p[j] = c->vLumFilter[i];

  1437.         }

  1438. 

  1439.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1440.             int j;

  1441.             short *p = (short *)&c->vCCoeffsBank[i];

  1442.             for (j = 0; j < 8; j++)

  1443.                 p[j] = c->vChrFilter[i];

  1444.         }

  1445. #endif

  1446.     }

  1447. 

  1448.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1449.     c->vLumBufSize = c->vLumFilterSize;

  1450.     c->vChrBufSize = c->vChrFilterSize;

  1451.     for (i = 0; i < dstH; i++) {

  1452.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1453.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1454.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1455.                                << c->chrSrcVSubSample));

  1456. 

  1457.         nextSlice >>= c->chrSrcVSubSample;

  1458.         nextSlice <<= c->chrSrcVSubSample;

  1459.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1460.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1461.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1462.             (nextSlice >> c->chrSrcVSubSample))

  1463.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1464.                              c->vChrFilterPos[chrI];

  1465.     }

  1466. 

  1467.     for (i = 0; i < 4; i++)

  1468.         FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);

  1469. 

  1470.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1471.      * need to allocate several megabytes to handle all possible cases) */

  1472.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1473.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1474.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1475.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1476.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1477.     /* Note we need at least one pixel more at the end because of the MMX code

  1478.      * (just in case someone wants to replace the 4000/8000). */

  1479.     /* align at 16 bytes for AltiVec */

  1480.     for (i = 0; i < c->vLumBufSize; i++) {

  1481.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1482.                           dst_stride + 16, fail);

  1483.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1484.     }

  1485.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1486.     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);

  1487.     c->uv_offx2 = dst_stride + 16;

  1488.     for (i = 0; i < c->vChrBufSize; i++) {

  1489.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1490.                          dst_stride * 2 + 32, fail);

  1491.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1492.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1493.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1494.     }

  1495.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1496.         for (i = 0; i < c->vLumBufSize; i++) {

  1497.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],

  1498.                               dst_stride + 16, fail);

  1499.             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];

  1500.         }

  1501. 

  1502.     // try to avoid drawing green stuff between the right end and the stride end

  1503.     for (i = 0; i < c->vChrBufSize; i++)

  1504.         if(desc_dst->comp[0].depth_minus1 == 15){

  1505.             av_assert0(c->dstBpc > 14);

  1506.             for(j=0; j<dst_stride/2+1; j++)

  1507.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1508.         } else

  1509.             for(j=0; j<dst_stride+1; j++)

  1510.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1511. 

  1512.     av_assert0(c->chrDstH <= dstH);

  1513. 

  1514.     if (flags & SWS_PRINT_INFO) {

  1515.         if (flags & SWS_FAST_BILINEAR)

  1516.             av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1517.         else if (flags & SWS_BILINEAR)

  1518.             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1519.         else if (flags & SWS_BICUBIC)

  1520.             av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1521.         else if (flags & SWS_X)

  1522.             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1523.         else if (flags & SWS_POINT)

  1524.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1525.         else if (flags & SWS_AREA)

  1526.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1527.         else if (flags & SWS_BICUBLIN)

  1528.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1529.         else if (flags & SWS_GAUSS)

  1530.             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1531.         else if (flags & SWS_SINC)

  1532.             av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1533.         else if (flags & SWS_LANCZOS)

  1534.             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1535.         else if (flags & SWS_SPLINE)

  1536.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1537.         else

  1538.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1539. 

  1540.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1541.                av_get_pix_fmt_name(srcFormat),

  1542. #ifdef DITHER1XBPP

  1543.                dstFormat == AV_PIX_FMT_BGR555   || dstFormat == AV_PIX_FMT_BGR565   ||

  1544.                dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||

  1545.                dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?

  1546.                                                              "dithered " : "",

  1547. #else

  1548.                "",

  1549. #endif

  1550.                av_get_pix_fmt_name(dstFormat));

  1551. 

  1552.         if (INLINE_MMXEXT(cpu_flags))

  1553.             av_log(c, AV_LOG_INFO, "using MMXEXT\n");

  1554.         else if (INLINE_AMD3DNOW(cpu_flags))

  1555.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1556.         else if (INLINE_MMX(cpu_flags))

  1557.             av_log(c, AV_LOG_INFO, "using MMX\n");

  1558.         else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)

  1559.             av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1560.         else

  1561.             av_log(c, AV_LOG_INFO, "using C\n");

  1562. 

  1563.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1564.         av_log(c, AV_LOG_DEBUG,

  1565.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1566.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1567.         av_log(c, AV_LOG_DEBUG,

  1568.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1569.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1570.                c->chrXInc, c->chrYInc);

  1571.     }

  1572. 

  1573.     c->swScale = ff_getSwsFunc(c);

  1574.     return 0;

  1575. fail: // FIXME replace things by appropriate error codes

  1576.     return -1;

  1577. }

  1578. 

  1579. #if FF_API_SWS_GETCONTEXT

  1580. SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,

  1581.                            int dstW, int dstH, enum AVPixelFormat dstFormat,

  1582.                            int flags, SwsFilter *srcFilter,

  1583.                            SwsFilter *dstFilter, const double *param)

  1584. {

  1585.     SwsContext *c;

  1586. 

  1587.     if (!(c = sws_alloc_context()))

  1588.         return NULL;

  1589. 

  1590.     c->flags     = flags;

  1591.     c->srcW      = srcW;

  1592.     c->srcH      = srcH;

  1593.     c->dstW      = dstW;

  1594.     c->dstH      = dstH;

  1595.     c->srcFormat = srcFormat;

  1596.     c->dstFormat = dstFormat;

  1597. 

  1598.     if (param) {

  1599.         c->param[0] = param[0];

  1600.         c->param[1] = param[1];

  1601.     }

  1602. 

  1603.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1604.         sws_freeContext(c);

  1605.         return NULL;

  1606.     }

  1607. 

  1608.     return c;

  1609. }

  1610. #endif

  1611. 

  1612. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1613.                                 float lumaSharpen, float chromaSharpen,

  1614.                                 float chromaHShift, float chromaVShift,

  1615.                                 int verbose)

  1616. {

  1617.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1618.     if (!filter)

  1619.         return NULL;

  1620. 

  1621.     if (lumaGBlur != 0.0) {

  1622.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1623.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1624.     } else {

  1625.         filter->lumH = sws_getIdentityVec();

  1626.         filter->lumV = sws_getIdentityVec();

  1627.     }

  1628. 

  1629.     if (chromaGBlur != 0.0) {

  1630.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1631.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1632.     } else {

  1633.         filter->chrH = sws_getIdentityVec();

  1634.         filter->chrV = sws_getIdentityVec();

  1635.     }

  1636. 

  1637.     if (chromaSharpen != 0.0) {

  1638.         SwsVector *id = sws_getIdentityVec();

  1639.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1640.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1641.         sws_addVec(filter->chrH, id);

  1642.         sws_addVec(filter->chrV, id);

  1643.         sws_freeVec(id);

  1644.     }

  1645. 

  1646.     if (lumaSharpen != 0.0) {

  1647.         SwsVector *id = sws_getIdentityVec();

  1648.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1649.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1650.         sws_addVec(filter->lumH, id);

  1651.         sws_addVec(filter->lumV, id);

  1652.         sws_freeVec(id);

  1653.     }

  1654. 

  1655.     if (chromaHShift != 0.0)

  1656.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1657. 

  1658.     if (chromaVShift != 0.0)

  1659.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1660. 

  1661.     sws_normalizeVec(filter->chrH, 1.0);

  1662.     sws_normalizeVec(filter->chrV, 1.0);

  1663.     sws_normalizeVec(filter->lumH, 1.0);

  1664.     sws_normalizeVec(filter->lumV, 1.0);

  1665. 

  1666.     if (verbose)

  1667.         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1668.     if (verbose)

  1669.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1670. 

  1671.     return filter;

  1672. }

  1673. 

  1674. SwsVector *sws_allocVec(int length)

  1675. {

  1676.     SwsVector *vec;

  1677. 

  1678.     if(length <= 0 || length > INT_MAX/ sizeof(double))

  1679.         return NULL;

  1680. 

  1681.     vec = av_malloc(sizeof(SwsVector));

  1682.     if (!vec)

  1683.         return NULL;

  1684.     vec->length = length;

  1685.     vec->coeff  = av_malloc(sizeof(double) * length);

  1686.     if (!vec->coeff)

  1687.         av_freep(&vec);

  1688.     return vec;

  1689. }

  1690. 

  1691. SwsVector *sws_getGaussianVec(double variance, double quality)

  1692. {

  1693.     const int length = (int)(variance * quality + 0.5) | 1;

  1694.     int i;

  1695.     double middle  = (length - 1) * 0.5;

  1696.     SwsVector *vec;

  1697. 

  1698.     if(variance < 0 || quality < 0)

  1699.         return NULL;

  1700. 

  1701.     vec = sws_allocVec(length);

  1702. 

  1703.     if (!vec)

  1704.         return NULL;

  1705. 

  1706.     for (i = 0; i < length; i++) {

  1707.         double dist = i - middle;

  1708.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1709.                         sqrt(2 * variance * M_PI);

  1710.     }

  1711. 

  1712.     sws_normalizeVec(vec, 1.0);

  1713. 

  1714.     return vec;

  1715. }

  1716. 

  1717. SwsVector *sws_getConstVec(double c, int length)

  1718. {

  1719.     int i;

  1720.     SwsVector *vec = sws_allocVec(length);

  1721. 

  1722.     if (!vec)

  1723.         return NULL;

  1724. 

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

  1726.         vec->coeff[i] = c;

  1727. 

  1728.     return vec;

  1729. }

  1730. 

  1731. SwsVector *sws_getIdentityVec(void)

  1732. {

  1733.     return sws_getConstVec(1.0, 1);

  1734. }

  1735. 

  1736. static double sws_dcVec(SwsVector *a)

  1737. {

  1738.     int i;

  1739.     double sum = 0;

  1740. 

  1741.     for (i = 0; i < a->length; i++)

  1742.         sum += a->coeff[i];

  1743. 

  1744.     return sum;

  1745. }

  1746. 

  1747. void sws_scaleVec(SwsVector *a, double scalar)

  1748. {

  1749.     int i;

  1750. 

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

  1752.         a->coeff[i] *= scalar;

  1753. }

  1754. 

  1755. void sws_normalizeVec(SwsVector *a, double height)

  1756. {

  1757.     sws_scaleVec(a, height / sws_dcVec(a));

  1758. }

  1759. 

  1760. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1761. {

  1762.     int length = a->length + b->length - 1;

  1763.     int i, j;

  1764.     SwsVector *vec = sws_getConstVec(0.0, length);

  1765. 

  1766.     if (!vec)

  1767.         return NULL;

  1768. 

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

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

  1771.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1772.         }

  1773.     }

  1774. 

  1775.     return vec;

  1776. }

  1777. 

  1778. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1779. {

  1780.     int length = FFMAX(a->length, b->length);

  1781.     int i;

  1782.     SwsVector *vec = sws_getConstVec(0.0, length);

  1783. 

  1784.     if (!vec)

  1785.         return NULL;

  1786. 

  1787.     for (i = 0; i < a->length; i++)

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

  1789.     for (i = 0; i < b->length; i++)

  1790.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

  1791. 

  1792.     return vec;

  1793. }

  1794. 

  1795. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1796. {

  1797.     int length = FFMAX(a->length, b->length);

  1798.     int i;

  1799.     SwsVector *vec = sws_getConstVec(0.0, length);

  1800. 

  1801.     if (!vec)

  1802.         return NULL;

  1803. 

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

  1805.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1806.     for (i = 0; i < b->length; i++)

  1807.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

  1808. 

  1809.     return vec;

  1810. }

  1811. 

  1812. /* shift left / or right if "shift" is negative */

  1813. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1814. {

  1815.     int length = a->length + FFABS(shift) * 2;

  1816.     int i;

  1817.     SwsVector *vec = sws_getConstVec(0.0, length);

  1818. 

  1819.     if (!vec)

  1820.         return NULL;

  1821. 

  1822.     for (i = 0; i < a->length; i++) {

  1823.         vec->coeff[i + (length    - 1) / 2 -

  1824.                        (a->length - 1) / 2 - shift] = a->coeff[i];

  1825.     }

  1826. 

  1827.     return vec;

  1828. }

  1829. 

  1830. void sws_shiftVec(SwsVector *a, int shift)

  1831. {

  1832.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1833.     av_free(a->coeff);

  1834.     a->coeff  = shifted->coeff;

  1835.     a->length = shifted->length;

  1836.     av_free(shifted);

  1837. }

  1838. 

  1839. void sws_addVec(SwsVector *a, SwsVector *b)

  1840. {

  1841.     SwsVector *sum = sws_sumVec(a, b);

  1842.     av_free(a->coeff);

  1843.     a->coeff  = sum->coeff;

  1844.     a->length = sum->length;

  1845.     av_free(sum);

  1846. }

  1847. 

  1848. void sws_subVec(SwsVector *a, SwsVector *b)

  1849. {

  1850.     SwsVector *diff = sws_diffVec(a, b);

  1851.     av_free(a->coeff);

  1852.     a->coeff  = diff->coeff;

  1853.     a->length = diff->length;

  1854.     av_free(diff);

  1855. }

  1856. 

  1857. void sws_convVec(SwsVector *a, SwsVector *b)

  1858. {

  1859.     SwsVector *conv = sws_getConvVec(a, b);

  1860.     av_free(a->coeff);

  1861.     a->coeff  = conv->coeff;

  1862.     a->length = conv->length;

  1863.     av_free(conv);

  1864. }

  1865. 

  1866. SwsVector *sws_cloneVec(SwsVector *a)

  1867. {

  1868.     int i;

  1869.     SwsVector *vec = sws_allocVec(a->length);

  1870. 

  1871.     if (!vec)

  1872.         return NULL;

  1873. 

  1874.     for (i = 0; i < a->length; i++)

  1875.         vec->coeff[i] = a->coeff[i];

  1876. 

  1877.     return vec;

  1878. }

  1879. 

  1880. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1881. {

  1882.     int i;

  1883.     double max = 0;

  1884.     double min = 0;

  1885.     double range;

  1886. 

  1887.     for (i = 0; i < a->length; i++)

  1888.         if (a->coeff[i] > max)

  1889.             max = a->coeff[i];

  1890. 

  1891.     for (i = 0; i < a->length; i++)

  1892.         if (a->coeff[i] < min)

  1893.             min = a->coeff[i];

  1894. 

  1895.     range = max - min;

  1896. 

  1897.     for (i = 0; i < a->length; i++) {

  1898.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1899.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1900.         for (; x > 0; x--)

  1901.             av_log(log_ctx, log_level, " ");

  1902.         av_log(log_ctx, log_level, "|\n");

  1903.     }

  1904. }

  1905. 

  1906. void sws_freeVec(SwsVector *a)

  1907. {

  1908.     if (!a)

  1909.         return;

  1910.     av_freep(&a->coeff);

  1911.     a->length = 0;

  1912.     av_free(a);

  1913. }

  1914. 

  1915. void sws_freeFilter(SwsFilter *filter)

  1916. {

  1917.     if (!filter)

  1918.         return;

  1919. 

  1920.     if (filter->lumH)

  1921.         sws_freeVec(filter->lumH);

  1922.     if (filter->lumV)

  1923.         sws_freeVec(filter->lumV);

  1924.     if (filter->chrH)

  1925.         sws_freeVec(filter->chrH);

  1926.     if (filter->chrV)

  1927.         sws_freeVec(filter->chrV);

  1928.     av_free(filter);

  1929. }

  1930. 

  1931. void sws_freeContext(SwsContext *c)

  1932. {

  1933.     int i;

  1934.     if (!c)

  1935.         return;

  1936. 

  1937.     if (c->lumPixBuf) {

  1938.         for (i = 0; i < c->vLumBufSize; i++)

  1939.             av_freep(&c->lumPixBuf[i]);

  1940.         av_freep(&c->lumPixBuf);

  1941.     }

  1942. 

  1943.     if (c->chrUPixBuf) {

  1944.         for (i = 0; i < c->vChrBufSize; i++)

  1945.             av_freep(&c->chrUPixBuf[i]);

  1946.         av_freep(&c->chrUPixBuf);

  1947.         av_freep(&c->chrVPixBuf);

  1948.     }

  1949. 

  1950.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1951.         for (i = 0; i < c->vLumBufSize; i++)

  1952.             av_freep(&c->alpPixBuf[i]);

  1953.         av_freep(&c->alpPixBuf);

  1954.     }

  1955. 

  1956.     for (i = 0; i < 4; i++)

  1957.         av_freep(&c->dither_error[i]);

  1958. 

  1959.     av_freep(&c->vLumFilter);

  1960.     av_freep(&c->vChrFilter);

  1961.     av_freep(&c->hLumFilter);

  1962.     av_freep(&c->hChrFilter);

  1963. #if HAVE_ALTIVEC

  1964.     av_freep(&c->vYCoeffsBank);

  1965.     av_freep(&c->vCCoeffsBank);

  1966. #endif

  1967. 

  1968.     av_freep(&c->vLumFilterPos);

  1969.     av_freep(&c->vChrFilterPos);

  1970.     av_freep(&c->hLumFilterPos);

  1971.     av_freep(&c->hChrFilterPos);

  1972. 

  1973. #if HAVE_MMX_INLINE

  1974. #if USE_MMAP

  1975.     if (c->lumMmxextFilterCode)

  1976.         munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);

  1977.     if (c->chrMmxextFilterCode)

  1978.         munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);

  1979. #elif HAVE_VIRTUALALLOC

  1980.     if (c->lumMmxextFilterCode)

  1981.         VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);

  1982.     if (c->chrMmxextFilterCode)

  1983.         VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);

  1984. #else

  1985.     av_free(c->lumMmxextFilterCode);

  1986.     av_free(c->chrMmxextFilterCode);

  1987. #endif

  1988.     c->lumMmxextFilterCode = NULL;

  1989.     c->chrMmxextFilterCode = NULL;

  1990. #endif /* HAVE_MMX_INLINE */

  1991. 

  1992.     av_freep(&c->yuvTable);

  1993.     av_freep(&c->formatConvBuffer);

  1994. 

  1995.     av_free(c);

  1996. }

  1997. 

  1998. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  1999.                                         int srcH, enum AVPixelFormat srcFormat,

  2000.                                         int dstW, int dstH,

  2001.                                         enum AVPixelFormat dstFormat, int flags,

  2002.                                         SwsFilter *srcFilter,

  2003.                                         SwsFilter *dstFilter,

  2004.                                         const double *param)

  2005. {

  2006.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  2007.                                              SWS_PARAM_DEFAULT };

  2008. 

  2009.     if (!param)

  2010.         param = default_param;

  2011. 

  2012.     if (context &&

  2013.         (context->srcW      != srcW      ||

  2014.          context->srcH      != srcH      ||

  2015.          context->srcFormat != srcFormat ||

  2016.          context->dstW      != dstW      ||

  2017.          context->dstH      != dstH      ||

  2018.          context->dstFormat != dstFormat ||

  2019.          context->flags     != flags     ||

  2020.          context->param[0]  != param[0]  ||

  2021.          context->param[1]  != param[1])) {

  2022.         sws_freeContext(context);

  2023.         context = NULL;

  2024.     }

  2025. 

  2026.     if (!context) {

  2027.         if (!(context = sws_alloc_context()))

  2028.             return NULL;

  2029.         context->srcW      = srcW;

  2030.         context->srcH      = srcH;

  2031.         context->srcFormat = srcFormat;

  2032.         context->dstW      = dstW;

  2033.         context->dstH      = dstH;

  2034.         context->dstFormat = dstFormat;

  2035.         context->flags     = flags;

  2036.         context->param[0]  = param[0];

  2037.         context->param[1]  = param[1];

  2038.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  2039.             sws_freeContext(context);

  2040.             return NULL;

  2041.         }

  2042.     }

  2043.     return context;

  2044. }