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.
42239 Commits
32 Branches
283MB
Tree: fa36f33422
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'fa36f33422'
${ noResults }
FFmpeg/libswscale/utils.c

1768 lines
62KB
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/avassert.h"

  41. #include "libavutil/avutil.h"

  42. #include "libavutil/bswap.h"

  43. #include "libavutil/cpu.h"

  44. #include "libavutil/intreadwrite.h"

  45. #include "libavutil/mathematics.h"

  46. #include "libavutil/opt.h"

  47. #include "libavutil/pixdesc.h"

  48. #include "libavutil/x86_cpu.h"

  49. #include "rgb2rgb.h"

  50. #include "swscale.h"

  51. #include "swscale_internal.h"

  52. 

  53. unsigned swscale_version(void)

  54. {

  55.     av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);

  56.     return LIBSWSCALE_VERSION_INT;

  57. }

  58. 

  59. const char *swscale_configuration(void)

  60. {

  61.     return FFMPEG_CONFIGURATION;

  62. }

  63. 

  64. const char *swscale_license(void)

  65. {

  66. #define LICENSE_PREFIX "libswscale license: "

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

  68. }

  69. 

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

  71. 

  72. typedef struct FormatEntry {

  73.     int is_supported_in, is_supported_out;

  74. } FormatEntry;

  75. 

  76. static const FormatEntry format_entries[PIX_FMT_NB] = {

  77.     [PIX_FMT_YUV420P]     = { 1, 1 },

  78.     [PIX_FMT_YUYV422]     = { 1, 1 },

  79.     [PIX_FMT_RGB24]       = { 1, 1 },

  80.     [PIX_FMT_BGR24]       = { 1, 1 },

  81.     [PIX_FMT_YUV422P]     = { 1, 1 },

  82.     [PIX_FMT_YUV444P]     = { 1, 1 },

  83.     [PIX_FMT_YUV410P]     = { 1, 1 },

  84.     [PIX_FMT_YUV411P]     = { 1, 1 },

  85.     [PIX_FMT_GRAY8]       = { 1, 1 },

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

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

  88.     [PIX_FMT_PAL8]        = { 1, 0 },

  89.     [PIX_FMT_YUVJ420P]    = { 1, 1 },

  90.     [PIX_FMT_YUVJ422P]    = { 1, 1 },

  91.     [PIX_FMT_YUVJ444P]    = { 1, 1 },

  92.     [PIX_FMT_UYVY422]     = { 1, 1 },

  93.     [PIX_FMT_UYYVYY411]   = { 0, 0 },

  94.     [PIX_FMT_BGR8]        = { 1, 1 },

  95.     [PIX_FMT_BGR4]        = { 0, 1 },

  96.     [PIX_FMT_BGR4_BYTE]   = { 1, 1 },

  97.     [PIX_FMT_RGB8]        = { 1, 1 },

  98.     [PIX_FMT_RGB4]        = { 0, 1 },

  99.     [PIX_FMT_RGB4_BYTE]   = { 1, 1 },

  100.     [PIX_FMT_NV12]        = { 1, 1 },

  101.     [PIX_FMT_NV21]        = { 1, 1 },

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

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

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

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

  106.     [PIX_FMT_0RGB]        = { 1, 1 },

  107.     [PIX_FMT_RGB0]        = { 1, 1 },

  108.     [PIX_FMT_0BGR]        = { 1, 1 },

  109.     [PIX_FMT_BGR0]        = { 1, 1 },

  110.     [PIX_FMT_GRAY16BE]    = { 1, 1 },

  111.     [PIX_FMT_GRAY16LE]    = { 1, 1 },

  112.     [PIX_FMT_YUV440P]     = { 1, 1 },

  113.     [PIX_FMT_YUVJ440P]    = { 1, 1 },

  114.     [PIX_FMT_YUVA420P]    = { 1, 1 },

  115.     [PIX_FMT_YUVA422P]    = { 1, 1 },

  116.     [PIX_FMT_YUVA444P]    = { 1, 1 },

  117.     [PIX_FMT_RGB48BE]     = { 1, 1 },

  118.     [PIX_FMT_RGB48LE]     = { 1, 1 },

  119.     [PIX_FMT_RGBA64BE]    = { 1, 0 },

  120.     [PIX_FMT_RGBA64LE]    = { 1, 0 },

  121.     [PIX_FMT_RGB565BE]    = { 1, 1 },

  122.     [PIX_FMT_RGB565LE]    = { 1, 1 },

  123.     [PIX_FMT_RGB555BE]    = { 1, 1 },

  124.     [PIX_FMT_RGB555LE]    = { 1, 1 },

  125.     [PIX_FMT_BGR565BE]    = { 1, 1 },

  126.     [PIX_FMT_BGR565LE]    = { 1, 1 },

  127.     [PIX_FMT_BGR555BE]    = { 1, 1 },

  128.     [PIX_FMT_BGR555LE]    = { 1, 1 },

  129.     [PIX_FMT_YUV420P16LE] = { 1, 1 },

  130.     [PIX_FMT_YUV420P16BE] = { 1, 1 },

  131.     [PIX_FMT_YUV422P16LE] = { 1, 1 },

  132.     [PIX_FMT_YUV422P16BE] = { 1, 1 },

  133.     [PIX_FMT_YUV444P16LE] = { 1, 1 },

  134.     [PIX_FMT_YUV444P16BE] = { 1, 1 },

  135.     [PIX_FMT_RGB444LE]    = { 1, 1 },

  136.     [PIX_FMT_RGB444BE]    = { 1, 1 },

  137.     [PIX_FMT_BGR444LE]    = { 1, 1 },

  138.     [PIX_FMT_BGR444BE]    = { 1, 1 },

  139.     [PIX_FMT_Y400A]       = { 1, 0 },

  140.     [PIX_FMT_BGR48BE]     = { 1, 1 },

  141.     [PIX_FMT_BGR48LE]     = { 1, 1 },

  142.     [PIX_FMT_BGRA64BE]    = { 0, 0 },

  143.     [PIX_FMT_BGRA64LE]    = { 0, 0 },

  144.     [PIX_FMT_YUV420P9BE]  = { 1, 1 },

  145.     [PIX_FMT_YUV420P9LE]  = { 1, 1 },

  146.     [PIX_FMT_YUV420P10BE] = { 1, 1 },

  147.     [PIX_FMT_YUV420P10LE] = { 1, 1 },

  148.     [PIX_FMT_YUV420P12BE] = { 1, 1 },

  149.     [PIX_FMT_YUV420P12LE] = { 1, 1 },

  150.     [PIX_FMT_YUV420P14BE] = { 1, 1 },

  151.     [PIX_FMT_YUV420P14LE] = { 1, 1 },

  152.     [PIX_FMT_YUV422P9BE]  = { 1, 1 },

  153.     [PIX_FMT_YUV422P9LE]  = { 1, 1 },

  154.     [PIX_FMT_YUV422P10BE] = { 1, 1 },

  155.     [PIX_FMT_YUV422P10LE] = { 1, 1 },

  156.     [PIX_FMT_YUV422P12BE] = { 1, 1 },

  157.     [PIX_FMT_YUV422P12LE] = { 1, 1 },

  158.     [PIX_FMT_YUV422P14BE] = { 1, 1 },

  159.     [PIX_FMT_YUV422P14LE] = { 1, 1 },

  160.     [PIX_FMT_YUV444P9BE]  = { 1, 1 },

  161.     [PIX_FMT_YUV444P9LE]  = { 1, 1 },

  162.     [PIX_FMT_YUV444P10BE] = { 1, 1 },

  163.     [PIX_FMT_YUV444P10LE] = { 1, 1 },

  164.     [PIX_FMT_YUV444P12BE] = { 1, 1 },

  165.     [PIX_FMT_YUV444P12LE] = { 1, 1 },

  166.     [PIX_FMT_YUV444P14BE] = { 1, 1 },

  167.     [PIX_FMT_YUV444P14LE] = { 1, 1 },

  168.     [PIX_FMT_GBRP]        = { 1, 0 },

  169.     [PIX_FMT_GBRP9LE]     = { 1, 0 },

  170.     [PIX_FMT_GBRP9BE]     = { 1, 0 },

  171.     [PIX_FMT_GBRP10LE]    = { 1, 0 },

  172.     [PIX_FMT_GBRP10BE]    = { 1, 0 },

  173.     [PIX_FMT_GBRP12LE]    = { 1, 0 },

  174.     [PIX_FMT_GBRP12BE]    = { 1, 0 },

  175.     [PIX_FMT_GBRP14LE]    = { 1, 0 },

  176.     [PIX_FMT_GBRP14BE]    = { 1, 0 },

  177.     [PIX_FMT_GBRP16LE]    = { 1, 0 },

  178.     [PIX_FMT_GBRP16BE]    = { 1, 0 },

  179. };

  180. 

  181. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  182. {

  183.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  184.            format_entries[pix_fmt].is_supported_in : 0;

  185. }

  186. 

  187. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  188. {

  189.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  190.            format_entries[pix_fmt].is_supported_out : 0;

  191. }

  192. 

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

  194. 

  195. #if FF_API_SWS_FORMAT_NAME

  196. const char *sws_format_name(enum PixelFormat format)

  197. {

  198.     return av_get_pix_fmt_name(format);

  199. }

  200. #endif

  201. 

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

  203.                              double dist)

  204. {

  205.     if (dist <= 1.0)

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

  207.     else

  208.         return getSplineCoeff(0.0,

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

  210.                                c + 3.0 * d,

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

  212.                               dist - 1.0);

  213. }

  214. 

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

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

  217.                       int filterAlign, int one, int flags, int cpu_flags,

  218.                       SwsVector *srcFilter, SwsVector *dstFilter,

  219.                       double param[2])

  220. {

  221.     int i;

  222.     int filterSize;

  223.     int filter2Size;

  224.     int minFilterSize;

  225.     int64_t *filter    = NULL;

  226.     int64_t *filter2   = NULL;

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

  228.     int ret            = -1;

  229. 

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

  231. 

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

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

  234. 

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

  236.         int i;

  237.         filterSize = 1;

  238.         FF_ALLOCZ_OR_GOTO(NULL, filter,

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

  240. 

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

  242.             filter[i * filterSize] = fone;

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

  244.         }

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

  246.         int i;

  247.         int64_t xDstInSrc;

  248.         filterSize = 1;

  249.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  251. 

  252.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  255. 

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

  257.             filter[i]       = fone;

  258.             xDstInSrc      += xInc;

  259.         }

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

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

  262.         int i;

  263.         int64_t xDstInSrc;

  264.         filterSize = 2;

  265.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  267. 

  268.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  271.             int j;

  272. 

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

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

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

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

  277.                 if (coeff < 0)

  278.                     coeff = 0;

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

  280.                 xx++;

  281.             }

  282.             xDstInSrc += xInc;

  283.         }

  284.     } else {

  285.         int64_t xDstInSrc;

  286.         int sizeFactor;

  287. 

  288.         if (flags & SWS_BICUBIC)

  289.             sizeFactor = 4;

  290.         else if (flags & SWS_X)

  291.             sizeFactor = 8;

  292.         else if (flags & SWS_AREA)

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

  294.         else if (flags & SWS_GAUSS)

  295.             sizeFactor = 8;     // infinite ;)

  296.         else if (flags & SWS_LANCZOS)

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

  298.         else if (flags & SWS_SINC)

  299.             sizeFactor = 20;    // infinite ;)

  300.         else if (flags & SWS_SPLINE)

  301.             sizeFactor = 20;    // infinite ;)

  302.         else if (flags & SWS_BILINEAR)

  303.             sizeFactor = 2;

  304.         else {

  305.             av_assert0(0);

  306.         }

  307. 

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

  309.             filterSize = 1 + sizeFactor;    // upscale

  310.         else

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

  312. 

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

  314.         filterSize = FFMAX(filterSize, 1);

  315. 

  316.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  318. 

  319.         xDstInSrc = xInc - 0x10000;

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

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

  322.             int j;

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

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

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

  326.                 double floatd;

  327.                 int64_t coeff;

  328. 

  329.                 if (xInc > 1 << 16)

  330.                     d = d * dstW / srcW;

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

  332. 

  333.                 if (flags & SWS_BICUBIC) {

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

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

  336. 

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

  338.                         coeff = 0.0;

  339.                     } else {

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

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

  342. 

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

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

  345.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  346.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  347.                         else

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

  349.                                       (6 * B + 30 * C) * dd  +

  350.                                     (-12 * B - 48 * C) * d   +

  351.                                       (8 * B + 24 * C) * (1 << 30);

  352.                     }

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

  354.                 }

  355. #if 0

  356.                 else if (flags & SWS_X) {

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

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

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

  360.                 }

  361. #endif

  362.                 else if (flags & SWS_X) {

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

  364.                     double c;

  365. 

  366.                     if (floatd < 1.0)

  367.                         c = cos(floatd * M_PI);

  368.                     else

  369.                         c = -1.0;

  370.                     if (c < 0.0)

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

  372.                     else

  373.                         c = pow(c, A);

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

  375.                 } else if (flags & SWS_AREA) {

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

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

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

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

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

  381.                     else

  382.                         coeff = 0.0;

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

  384.                 } else if (flags & SWS_GAUSS) {

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

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

  387.                 } else if (flags & SWS_SINC) {

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

  389.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  393.                     if (floatd > p)

  394.                         coeff = 0;

  395.                 } else if (flags & SWS_BILINEAR) {

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

  397.                     if (coeff < 0)

  398.                         coeff = 0;

  399.                     coeff *= fone >> 30;

  400.                 } else if (flags & SWS_SPLINE) {

  401.                     double p = -2.196152422706632;

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

  403.                 } else {

  404.                     av_assert0(0);

  405.                 }

  406. 

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

  408.                 xx++;

  409.             }

  410.             xDstInSrc += 2 * xInc;

  411.         }

  412.     }

  413. 

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

  415.      * av_free(filter);

  416.      */

  417.     av_assert0(filterSize > 0);

  418.     filter2Size = filterSize;

  419.     if (srcFilter)

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

  421.     if (dstFilter)

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

  423.     av_assert0(filter2Size > 0);

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

  425. 

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

  427.         int j, k;

  428. 

  429.         if (srcFilter) {

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

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

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

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

  434.             }

  435.         } else {

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

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

  438.         }

  439.         // FIXME dstFilter

  440. 

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

  442.     }

  443.     av_freep(&filter);

  444. 

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

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

  447.     minFilterSize = 0;

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

  449.         int min = filter2Size;

  450.         int j;

  451.         int64_t cutOff = 0.0;

  452. 

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

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

  455.             int k;

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

  457. 

  458.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  459.                 break;

  460. 

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

  462.              * filter otherwise */

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

  464.                 break;

  465. 

  466.             // move filter coefficients left

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

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

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

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

  471.         }

  472. 

  473.         cutOff = 0;

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

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

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

  477. 

  478.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  479.                 break;

  480.             min--;

  481.         }

  482. 

  483.         if (min > minFilterSize)

  484.             minFilterSize = min;

  485.     }

  486. 

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

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

  489.         if (minFilterSize < 5)

  490.             filterAlign = 4;

  491. 

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

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

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

  495.         if (minFilterSize < 3)

  496.             filterAlign = 1;

  497.     }

  498. 

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

  500.         // special case for unscaled vertical filtering

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

  502.             filterAlign = 1;

  503.     }

  504. 

  505.     av_assert0(minFilterSize > 0);

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

  507.     av_assert0(filterSize > 0);

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

  509.     if (filterSize >= MAX_FILTER_SIZE * 16 /

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

  511.         goto fail;

  512.     *outFilterSize = filterSize;

  513. 

  514.     if (flags & SWS_PRINT_INFO)

  515.         av_log(NULL, AV_LOG_VERBOSE,

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

  517.                filter2Size, filterSize);

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

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

  520.         int j;

  521. 

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

  523.             if (j >= filter2Size)

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

  525.             else

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

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

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

  529.         }

  530.     }

  531. 

  532.     // FIXME try to align filterPos if possible

  533. 

  534.     // fix borders

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

  536.         int j;

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

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

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

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

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

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

  543.             }

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

  545.         }

  546. 

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

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

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

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

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

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

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

  554.             }

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

  556.         }

  557.     }

  558. 

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

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

  561.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

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

  563. 

  564.     /* normalize & store in outFilter */

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

  566.         int j;

  567.         int64_t error = 0;

  568.         int64_t sum   = 0;

  569. 

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

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

  572.         }

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

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

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

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

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

  578.             error                                  = v - intV * sum;

  579.         }

  580.     }

  581. 

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

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

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

  585.                                                       * read over the end */

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

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

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

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

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

  591.     }

  592. 

  593.     ret = 0;

  594. 

  595. fail:

  596.     av_free(filter);

  597.     av_free(filter2);

  598.     return ret;

  599. }

  600. 

  601. #if HAVE_MMX2

  602. static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode,

  603.                            int16_t *filter, int32_t *filterPos, int numSplits)

  604. {

  605.     uint8_t *fragmentA;

  606.     x86_reg imm8OfPShufW1A;

  607.     x86_reg imm8OfPShufW2A;

  608.     x86_reg fragmentLengthA;

  609.     uint8_t *fragmentB;

  610.     x86_reg imm8OfPShufW1B;

  611.     x86_reg imm8OfPShufW2B;

  612.     x86_reg fragmentLengthB;

  613.     int fragmentPos;

  614. 

  615.     int xpos, i;

  616. 

  617.     // create an optimized horizontal scaling routine

  618.     /* This scaler is made of runtime-generated MMX2 code using specially tuned

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

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

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

  622.      */

  623. 

  624.     // code fragment

  625. 

  626.     __asm__ volatile (

  627.         "jmp                         9f                 \n\t"

  628.         // Begin

  629.         "0:                                             \n\t"

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

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

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

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

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

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

  636.         "1:                                             \n\t"

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

  638.         "2:                                             \n\t"

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

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

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

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

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

  644. 

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

  646. 

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

  648.         // End

  649.         "9:                                             \n\t"

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

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

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

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

  654.         "dec                         %1                 \n\t"

  655.         "dec                         %2                 \n\t"

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

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

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

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

  660. 

  661. 

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

  663.           "=r" (fragmentLengthA)

  664.         );

  665. 

  666.     __asm__ volatile (

  667.         "jmp                         9f                 \n\t"

  668.         // Begin

  669.         "0:                                             \n\t"

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

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

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

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

  674.         "1:                                             \n\t"

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

  676.         "2:                                             \n\t"

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

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

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

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

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

  682. 

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

  684. 

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

  686.         // End

  687.         "9:                                             \n\t"

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

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

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

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

  692.         "dec                         %1                 \n\t"

  693.         "dec                         %2                 \n\t"

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

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

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

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

  698. 

  699. 

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

  701.           "=r" (fragmentLengthB)

  702.         );

  703. 

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

  705.     fragmentPos = 0;

  706. 

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

  708.         int xx = xpos >> 16;

  709. 

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

  711.             int a                  = 0;

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

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

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

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

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

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

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

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

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

  721.             int shift              = 0;

  722. 

  723.             if (filterCode) {

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

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

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

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

  728.                 filterPos[i / 2] = xx;

  729. 

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

  731. 

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

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

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

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

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

  737.                                                                (c << 4) |

  738.                                                                (d << 6);

  739. 

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

  741.                     shift = maxShift;               // avoid overread

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

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

  744. 

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

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

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

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

  749.                 }

  750.             }

  751. 

  752.             fragmentPos += fragmentLength;

  753. 

  754.             if (filterCode)

  755.                 filterCode[fragmentPos] = RET;

  756.         }

  757.         xpos += xInc;

  758.     }

  759.     if (filterCode)

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

  761. 

  762.     return fragmentPos + 1;

  763. }

  764. #endif /* HAVE_MMX2 */

  765. 

  766. static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)

  767. {

  768.     *h = av_pix_fmt_descriptors[format].log2_chroma_w;

  769.     *v = av_pix_fmt_descriptors[format].log2_chroma_h;

  770. }

  771. 

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

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

  774.                              int brightness, int contrast, int saturation)

  775. {

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

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

  778. 

  779.     c->brightness = brightness;

  780.     c->contrast   = contrast;

  781.     c->saturation = saturation;

  782.     c->srcRange   = srcRange;

  783.     c->dstRange   = dstRange;

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

  785.         return -1;

  786. 

  787.     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);

  788.     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);

  789. 

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

  791.                              contrast, saturation);

  792.     // FIXME factorize

  793. 

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

  795.         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,

  796.                                        contrast, saturation);

  797.     return 0;

  798. }

  799. 

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

  801.                              int *srcRange, int **table, int *dstRange,

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

  803. {

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

  805.         return -1;

  806. 

  807.     *inv_table  = c->srcColorspaceTable;

  808.     *table      = c->dstColorspaceTable;

  809.     *srcRange   = c->srcRange;

  810.     *dstRange   = c->dstRange;

  811.     *brightness = c->brightness;

  812.     *contrast   = c->contrast;

  813.     *saturation = c->saturation;

  814. 

  815.     return 0;

  816. }

  817. 

  818. static int handle_jpeg(enum PixelFormat *format)

  819. {

  820.     switch (*format) {

  821.     case PIX_FMT_YUVJ420P:

  822.         *format = PIX_FMT_YUV420P;

  823.         return 1;

  824.     case PIX_FMT_YUVJ422P:

  825.         *format = PIX_FMT_YUV422P;

  826.         return 1;

  827.     case PIX_FMT_YUVJ444P:

  828.         *format = PIX_FMT_YUV444P;

  829.         return 1;

  830.     case PIX_FMT_YUVJ440P:

  831.         *format = PIX_FMT_YUV440P;

  832.         return 1;

  833.     default:

  834.         return 0;

  835.     }

  836. }

  837. 

  838. static int handle_0alpha(enum PixelFormat *format)

  839. {

  840.     switch (*format) {

  841.     case PIX_FMT_0BGR    : *format = PIX_FMT_ABGR   ; return 1;

  842.     case PIX_FMT_BGR0    : *format = PIX_FMT_BGRA   ; return 4;

  843.     case PIX_FMT_0RGB    : *format = PIX_FMT_ARGB   ; return 1;

  844.     case PIX_FMT_RGB0    : *format = PIX_FMT_RGBA   ; return 4;

  845.     default:                                          return 0;

  846.     }

  847. }

  848. 

  849. SwsContext *sws_alloc_context(void)

  850. {

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

  852. 

  853.     c->av_class = &sws_context_class;

  854.     av_opt_set_defaults(c);

  855. 

  856.     return c;

  857. }

  858. 

  859. int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)

  860. {

  861.     int i, j;

  862.     int usesVFilter, usesHFilter;

  863.     int unscaled;

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

  865.     int srcW              = c->srcW;

  866.     int srcH              = c->srcH;

  867.     int dstW              = c->dstW;

  868.     int dstH              = c->dstH;

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

  870.     int flags, cpu_flags;

  871.     enum PixelFormat srcFormat = c->srcFormat;

  872.     enum PixelFormat dstFormat = c->dstFormat;

  873. 

  874.     cpu_flags = av_get_cpu_flags();

  875.     flags     = c->flags;

  876.     emms_c();

  877.     if (!rgb15to16)

  878.         sws_rgb2rgb_init();

  879. 

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

  881. 

  882.     handle_jpeg(&srcFormat);

  883.     handle_jpeg(&dstFormat);

  884.     handle_0alpha(&srcFormat);

  885.     handle_0alpha(&dstFormat);

  886. 

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

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

  889.         c->srcFormat= srcFormat;

  890.         c->dstFormat= dstFormat;

  891.     }

  892. 

  893.     if (!sws_isSupportedInput(srcFormat)) {

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

  895.                av_get_pix_fmt_name(srcFormat));

  896.         return AVERROR(EINVAL);

  897.     }

  898.     if (!sws_isSupportedOutput(dstFormat)) {

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

  900.                av_get_pix_fmt_name(dstFormat));

  901.         return AVERROR(EINVAL);

  902.     }

  903. 

  904.     i = flags & (SWS_POINT         |

  905.                  SWS_AREA          |

  906.                  SWS_BILINEAR      |

  907.                  SWS_FAST_BILINEAR |

  908.                  SWS_BICUBIC       |

  909.                  SWS_X             |

  910.                  SWS_GAUSS         |

  911.                  SWS_LANCZOS       |

  912.                  SWS_SINC          |

  913.                  SWS_SPLINE        |

  914.                  SWS_BICUBLIN);

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

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

  917.         return AVERROR(EINVAL);

  918.     }

  919.     /* sanity check */

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

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

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

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

  924.                srcW, srcH, dstW, dstH);

  925.         return AVERROR(EINVAL);

  926.     }

  927. 

  928.     if (!dstFilter)

  929.         dstFilter = &dummyFilter;

  930.     if (!srcFilter)

  931.         srcFilter = &dummyFilter;

  932. 

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

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

  935.     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);

  936.     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);

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

  938. 

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

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

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

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

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

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

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

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

  947. 

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

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

  950. 

  951. 

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

  953.         if (dstW&1) {

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

  955.             flags |= SWS_FULL_CHR_H_INT;

  956.             c->flags = flags;

  957.         }

  958.     }

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

  960.      * chroma interpolation */

  961.     if (flags & SWS_FULL_CHR_H_INT &&

  962.         isAnyRGB(dstFormat)        &&

  963.         dstFormat != PIX_FMT_RGBA  &&

  964.         dstFormat != PIX_FMT_ARGB  &&

  965.         dstFormat != PIX_FMT_BGRA  &&

  966.         dstFormat != PIX_FMT_ABGR  &&

  967.         dstFormat != PIX_FMT_RGB24 &&

  968.         dstFormat != PIX_FMT_BGR24) {

  969.         av_log(c, AV_LOG_WARNING,

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

  971.                av_get_pix_fmt_name(dstFormat));

  972.         flags   &= ~SWS_FULL_CHR_H_INT;

  973.         c->flags = flags;

  974.     }

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

  976.         c->chrDstHSubSample = 1;

  977. 

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

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

  980.                            SWS_SRC_V_CHR_DROP_SHIFT;

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

  982. 

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

  984.      * wants full chroma */

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

  986.         srcFormat != PIX_FMT_RGB8 && srcFormat != PIX_FMT_BGR8 &&

  987.         srcFormat != PIX_FMT_RGB4 && srcFormat != PIX_FMT_BGR4 &&

  988.         srcFormat != PIX_FMT_RGB4_BYTE && srcFormat != PIX_FMT_BGR4_BYTE &&

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

  990.          (flags & SWS_FAST_BILINEAR)))

  991.         c->chrSrcHSubSample = 1;

  992. 

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

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

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

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

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

  998. 

  999.     /* unscaled special cases */

  1000.     if (unscaled && !usesHFilter && !usesVFilter &&

  1001.         (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {

  1002.         ff_get_unscaled_swscale(c);

  1003. 

  1004.         if (c->swScale) {

  1005.             if (flags & SWS_PRINT_INFO)

  1006.                 av_log(c, AV_LOG_INFO,

  1007.                        "using unscaled %s -> %s special converter\n",

  1008.                        av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));

  1009.             return 0;

  1010.         }

  1011.     }

  1012. 

  1013.     c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;

  1014.     if (c->srcBpc < 8)

  1015.         c->srcBpc = 8;

  1016.     c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;

  1017.     if (c->dstBpc < 8)

  1018.         c->dstBpc = 8;

  1019.     if (isAnyRGB(srcFormat) || srcFormat == PIX_FMT_PAL8)

  1020.         c->srcBpc = 16;

  1021.     if (c->dstBpc == 16)

  1022.         dst_stride <<= 1;

  1023.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);

  1024.     if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 &&

  1025.         c->srcBpc == 8 && c->dstBpc <= 14) {

  1026.         c->canMMX2BeUsed = (dstW >= srcW && (dstW & 31) == 0 &&

  1027.                             (srcW & 15) == 0) ? 1 : 0;

  1028.         if (!c->canMMX2BeUsed && dstW >= srcW && (srcW & 15) == 0

  1029.             && (flags & SWS_FAST_BILINEAR)) {

  1030.             if (flags & SWS_PRINT_INFO)

  1031.                 av_log(c, AV_LOG_INFO,

  1032.                        "output width is not a multiple of 32 -> no MMX2 scaler\n");

  1033.         }

  1034.         if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))

  1035.             c->canMMX2BeUsed=0;

  1036.     } else

  1037.         c->canMMX2BeUsed = 0;

  1038. 

  1039.     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;

  1040.     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

  1041. 

  1042.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  1043.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  1044.      * correct scaling.

  1045.      * n-2 is the last chrominance sample available.

  1046.      * This is not perfect, but no one should notice the difference, the more

  1047.      * correct variant would be like the vertical one, but that would require

  1048.      * some special code for the first and last pixel */

  1049.     if (flags & SWS_FAST_BILINEAR) {

  1050.         if (c->canMMX2BeUsed) {

  1051.             c->lumXInc += 20;

  1052.             c->chrXInc += 20;

  1053.         }

  1054.         // we don't use the x86 asm scaler if MMX is available

  1055.         else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX && c->dstBpc <= 14) {

  1056.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1057.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1058.         }

  1059.     }

  1060. 

  1061.     /* precalculate horizontal scaler filter coefficients */

  1062.     {

  1063. #if HAVE_MMX2

  1064. // can't downscale !!!

  1065.         if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {

  1066.             c->lumMmx2FilterCodeSize = initMMX2HScaler(dstW, c->lumXInc, NULL,

  1067.                                                        NULL, NULL, 8);

  1068.             c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc,

  1069.                                                        NULL, NULL, NULL, 4);

  1070. 

  1071. #ifdef MAP_ANONYMOUS

  1072.             c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);

  1073.             c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);

  1074. #elif HAVE_VIRTUALALLOC

  1075.             c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);

  1076.             c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);

  1077. #else

  1078.             c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);

  1079.             c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);

  1080. #endif

  1081. 

  1082. #ifdef MAP_ANONYMOUS

  1083.             if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)

  1084. #else

  1085.             if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)

  1086. #endif

  1087.             {

  1088.                 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");

  1089.                 return AVERROR(ENOMEM);

  1090.             }

  1091. 

  1092.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1093.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1094.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1095.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1096. 

  1097.             initMMX2HScaler(      dstW, c->lumXInc, c->lumMmx2FilterCode,

  1098.                             c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);

  1099.             initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode,

  1100.                             c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);

  1101. 

  1102. #ifdef MAP_ANONYMOUS

  1103.             mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);

  1104.             mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);

  1105. #endif

  1106.         } else

  1107. #endif /* HAVE_MMX2 */

  1108.         {

  1109.             const int filterAlign =

  1110.                 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :

  1111.                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1112.                 1;

  1113. 

  1114.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1115.                            &c->hLumFilterSize, c->lumXInc,

  1116.                            srcW, dstW, filterAlign, 1 << 14,

  1117.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1118.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1119.                            c->param) < 0)

  1120.                 goto fail;

  1121.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1122.                            &c->hChrFilterSize, c->chrXInc,

  1123.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1124.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1125.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1126.                            c->param) < 0)

  1127.                 goto fail;

  1128.         }

  1129.     } // initialize horizontal stuff

  1130. 

  1131.     /* precalculate vertical scaler filter coefficients */

  1132.     {

  1133.         const int filterAlign =

  1134.             (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :

  1135.             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1136.             1;

  1137. 

  1138.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1139.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

  1140.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1141.                        cpu_flags, srcFilter->lumV, dstFilter->lumV,

  1142.                        c->param) < 0)

  1143.             goto fail;

  1144.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1145.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1146.                        filterAlign, (1 << 12),

  1147.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1148.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1149.                        c->param) < 0)

  1150.             goto fail;

  1151. 

  1152. #if HAVE_ALTIVEC

  1153.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1154.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1155. 

  1156.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1157.             int j;

  1158.             short *p = (short *)&c->vYCoeffsBank[i];

  1159.             for (j = 0; j < 8; j++)

  1160.                 p[j] = c->vLumFilter[i];

  1161.         }

  1162. 

  1163.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1164.             int j;

  1165.             short *p = (short *)&c->vCCoeffsBank[i];

  1166.             for (j = 0; j < 8; j++)

  1167.                 p[j] = c->vChrFilter[i];

  1168.         }

  1169. #endif

  1170.     }

  1171. 

  1172.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1173.     c->vLumBufSize = c->vLumFilterSize;

  1174.     c->vChrBufSize = c->vChrFilterSize;

  1175.     for (i = 0; i < dstH; i++) {

  1176.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1177.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1178.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1179.                                << c->chrSrcVSubSample));

  1180. 

  1181.         nextSlice >>= c->chrSrcVSubSample;

  1182.         nextSlice <<= c->chrSrcVSubSample;

  1183.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1184.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1185.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1186.             (nextSlice >> c->chrSrcVSubSample))

  1187.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1188.                              c->vChrFilterPos[chrI];

  1189.     }

  1190. 

  1191.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1192.      * need to allocate several megabytes to handle all possible cases) */

  1193.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1194.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1195.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1196.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1197.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1198.     /* Note we need at least one pixel more at the end because of the MMX code

  1199.      * (just in case someone wants to replace the 4000/8000). */

  1200.     /* align at 16 bytes for AltiVec */

  1201.     for (i = 0; i < c->vLumBufSize; i++) {

  1202.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1203.                           dst_stride + 16, fail);

  1204.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1205.     }

  1206.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1207.     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);

  1208.     c->uv_offx2 = dst_stride + 16;

  1209.     for (i = 0; i < c->vChrBufSize; i++) {

  1210.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1211.                          dst_stride * 2 + 32, fail);

  1212.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1213.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1214.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1215.     }

  1216.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1217.         for (i = 0; i < c->vLumBufSize; i++) {

  1218.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],

  1219.                               dst_stride + 16, fail);

  1220.             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];

  1221.         }

  1222. 

  1223.     // try to avoid drawing green stuff between the right end and the stride end

  1224.     for (i = 0; i < c->vChrBufSize; i++)

  1225.         if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){

  1226.             av_assert0(c->dstBpc > 14);

  1227.             for(j=0; j<dst_stride/2+1; j++)

  1228.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1229.         } else

  1230.             for(j=0; j<dst_stride+1; j++)

  1231.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1232. 

  1233.     av_assert0(c->chrDstH <= dstH);

  1234. 

  1235.     if (flags & SWS_PRINT_INFO) {

  1236.         if (flags & SWS_FAST_BILINEAR)

  1237.             av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1238.         else if (flags & SWS_BILINEAR)

  1239.             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1240.         else if (flags & SWS_BICUBIC)

  1241.             av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1242.         else if (flags & SWS_X)

  1243.             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1244.         else if (flags & SWS_POINT)

  1245.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1246.         else if (flags & SWS_AREA)

  1247.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1248.         else if (flags & SWS_BICUBLIN)

  1249.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1250.         else if (flags & SWS_GAUSS)

  1251.             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1252.         else if (flags & SWS_SINC)

  1253.             av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1254.         else if (flags & SWS_LANCZOS)

  1255.             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1256.         else if (flags & SWS_SPLINE)

  1257.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1258.         else

  1259.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1260. 

  1261.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1262.                av_get_pix_fmt_name(srcFormat),

  1263. #ifdef DITHER1XBPP

  1264.                dstFormat == PIX_FMT_BGR555   || dstFormat == PIX_FMT_BGR565   ||

  1265.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1266.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ?

  1267.                                                              "dithered " : "",

  1268. #else

  1269.                "",

  1270. #endif

  1271.                av_get_pix_fmt_name(dstFormat));

  1272. 

  1273.         if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2)

  1274.             av_log(c, AV_LOG_INFO, "using MMX2\n");

  1275.         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)

  1276.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1277.         else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX)

  1278.             av_log(c, AV_LOG_INFO, "using MMX\n");

  1279.         else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)

  1280.             av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1281.         else

  1282.             av_log(c, AV_LOG_INFO, "using C\n");

  1283. 

  1284.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1285.         av_log(c, AV_LOG_DEBUG,

  1286.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1287.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1288.         av_log(c, AV_LOG_DEBUG,

  1289.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1290.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1291.                c->chrXInc, c->chrYInc);

  1292.     }

  1293. 

  1294.     c->swScale = ff_getSwsFunc(c);

  1295.     return 0;

  1296. fail: // FIXME replace things by appropriate error codes

  1297.     return -1;

  1298. }

  1299. 

  1300. #if FF_API_SWS_GETCONTEXT

  1301. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1302.                            int dstW, int dstH, enum PixelFormat dstFormat,

  1303.                            int flags, SwsFilter *srcFilter,

  1304.                            SwsFilter *dstFilter, const double *param)

  1305. {

  1306.     SwsContext *c;

  1307. 

  1308.     if (!(c = sws_alloc_context()))

  1309.         return NULL;

  1310. 

  1311.     c->flags     = flags;

  1312.     c->srcW      = srcW;

  1313.     c->srcH      = srcH;

  1314.     c->dstW      = dstW;

  1315.     c->dstH      = dstH;

  1316.     c->srcRange  = handle_jpeg(&srcFormat);

  1317.     c->dstRange  = handle_jpeg(&dstFormat);

  1318.     c->src0Alpha = handle_0alpha(&srcFormat);

  1319.     c->dst0Alpha = handle_0alpha(&dstFormat);

  1320.     c->srcFormat = srcFormat;

  1321.     c->dstFormat = dstFormat;

  1322. 

  1323.     if (param) {

  1324.         c->param[0] = param[0];

  1325.         c->param[1] = param[1];

  1326.     }

  1327.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1328.                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1329.                              c->dstRange, 0, 1 << 16, 1 << 16);

  1330. 

  1331.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1332.         sws_freeContext(c);

  1333.         return NULL;

  1334.     }

  1335. 

  1336.     return c;

  1337. }

  1338. #endif

  1339. 

  1340. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1341.                                 float lumaSharpen, float chromaSharpen,

  1342.                                 float chromaHShift, float chromaVShift,

  1343.                                 int verbose)

  1344. {

  1345.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1346.     if (!filter)

  1347.         return NULL;

  1348. 

  1349.     if (lumaGBlur != 0.0) {

  1350.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1351.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1352.     } else {

  1353.         filter->lumH = sws_getIdentityVec();

  1354.         filter->lumV = sws_getIdentityVec();

  1355.     }

  1356. 

  1357.     if (chromaGBlur != 0.0) {

  1358.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1359.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1360.     } else {

  1361.         filter->chrH = sws_getIdentityVec();

  1362.         filter->chrV = sws_getIdentityVec();

  1363.     }

  1364. 

  1365.     if (chromaSharpen != 0.0) {

  1366.         SwsVector *id = sws_getIdentityVec();

  1367.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1368.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1369.         sws_addVec(filter->chrH, id);

  1370.         sws_addVec(filter->chrV, id);

  1371.         sws_freeVec(id);

  1372.     }

  1373. 

  1374.     if (lumaSharpen != 0.0) {

  1375.         SwsVector *id = sws_getIdentityVec();

  1376.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1377.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1378.         sws_addVec(filter->lumH, id);

  1379.         sws_addVec(filter->lumV, id);

  1380.         sws_freeVec(id);

  1381.     }

  1382. 

  1383.     if (chromaHShift != 0.0)

  1384.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1385. 

  1386.     if (chromaVShift != 0.0)

  1387.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1388. 

  1389.     sws_normalizeVec(filter->chrH, 1.0);

  1390.     sws_normalizeVec(filter->chrV, 1.0);

  1391.     sws_normalizeVec(filter->lumH, 1.0);

  1392.     sws_normalizeVec(filter->lumV, 1.0);

  1393. 

  1394.     if (verbose)

  1395.         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1396.     if (verbose)

  1397.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1398. 

  1399.     return filter;

  1400. }

  1401. 

  1402. SwsVector *sws_allocVec(int length)

  1403. {

  1404.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1405.     if (!vec)

  1406.         return NULL;

  1407.     vec->length = length;

  1408.     vec->coeff  = av_malloc(sizeof(double) * length);

  1409.     if (!vec->coeff)

  1410.         av_freep(&vec);

  1411.     return vec;

  1412. }

  1413. 

  1414. SwsVector *sws_getGaussianVec(double variance, double quality)

  1415. {

  1416.     const int length = (int)(variance * quality + 0.5) | 1;

  1417.     int i;

  1418.     double middle  = (length - 1) * 0.5;

  1419.     SwsVector *vec = sws_allocVec(length);

  1420. 

  1421.     if (!vec)

  1422.         return NULL;

  1423. 

  1424.     for (i = 0; i < length; i++) {

  1425.         double dist = i - middle;

  1426.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1427.                         sqrt(2 * variance * M_PI);

  1428.     }

  1429. 

  1430.     sws_normalizeVec(vec, 1.0);

  1431. 

  1432.     return vec;

  1433. }

  1434. 

  1435. SwsVector *sws_getConstVec(double c, int length)

  1436. {

  1437.     int i;

  1438.     SwsVector *vec = sws_allocVec(length);

  1439. 

  1440.     if (!vec)

  1441.         return NULL;

  1442. 

  1443.     for (i = 0; i < length; i++)

  1444.         vec->coeff[i] = c;

  1445. 

  1446.     return vec;

  1447. }

  1448. 

  1449. SwsVector *sws_getIdentityVec(void)

  1450. {

  1451.     return sws_getConstVec(1.0, 1);

  1452. }

  1453. 

  1454. static double sws_dcVec(SwsVector *a)

  1455. {

  1456.     int i;

  1457.     double sum = 0;

  1458. 

  1459.     for (i = 0; i < a->length; i++)

  1460.         sum += a->coeff[i];

  1461. 

  1462.     return sum;

  1463. }

  1464. 

  1465. void sws_scaleVec(SwsVector *a, double scalar)

  1466. {

  1467.     int i;

  1468. 

  1469.     for (i = 0; i < a->length; i++)

  1470.         a->coeff[i] *= scalar;

  1471. }

  1472. 

  1473. void sws_normalizeVec(SwsVector *a, double height)

  1474. {

  1475.     sws_scaleVec(a, height / sws_dcVec(a));

  1476. }

  1477. 

  1478. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1479. {

  1480.     int length = a->length + b->length - 1;

  1481.     int i, j;

  1482.     SwsVector *vec = sws_getConstVec(0.0, length);

  1483. 

  1484.     if (!vec)

  1485.         return NULL;

  1486. 

  1487.     for (i = 0; i < a->length; i++) {

  1488.         for (j = 0; j < b->length; j++) {

  1489.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1490.         }

  1491.     }

  1492. 

  1493.     return vec;

  1494. }

  1495. 

  1496. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1497. {

  1498.     int length = FFMAX(a->length, b->length);

  1499.     int i;

  1500.     SwsVector *vec = sws_getConstVec(0.0, length);

  1501. 

  1502.     if (!vec)

  1503.         return NULL;

  1504. 

  1505.     for (i = 0; i < a->length; i++)

  1506.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1507.     for (i = 0; i < b->length; i++)

  1508.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

  1509. 

  1510.     return vec;

  1511. }

  1512. 

  1513. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1514. {

  1515.     int length = FFMAX(a->length, b->length);

  1516.     int i;

  1517.     SwsVector *vec = sws_getConstVec(0.0, length);

  1518. 

  1519.     if (!vec)

  1520.         return NULL;

  1521. 

  1522.     for (i = 0; i < a->length; i++)

  1523.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1524.     for (i = 0; i < b->length; i++)

  1525.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

  1526. 

  1527.     return vec;

  1528. }

  1529. 

  1530. /* shift left / or right if "shift" is negative */

  1531. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1532. {

  1533.     int length = a->length + FFABS(shift) * 2;

  1534.     int i;

  1535.     SwsVector *vec = sws_getConstVec(0.0, length);

  1536. 

  1537.     if (!vec)

  1538.         return NULL;

  1539. 

  1540.     for (i = 0; i < a->length; i++) {

  1541.         vec->coeff[i + (length    - 1) / 2 -

  1542.                        (a->length - 1) / 2 - shift] = a->coeff[i];

  1543.     }

  1544. 

  1545.     return vec;

  1546. }

  1547. 

  1548. void sws_shiftVec(SwsVector *a, int shift)

  1549. {

  1550.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1551.     av_free(a->coeff);

  1552.     a->coeff  = shifted->coeff;

  1553.     a->length = shifted->length;

  1554.     av_free(shifted);

  1555. }

  1556. 

  1557. void sws_addVec(SwsVector *a, SwsVector *b)

  1558. {

  1559.     SwsVector *sum = sws_sumVec(a, b);

  1560.     av_free(a->coeff);

  1561.     a->coeff  = sum->coeff;

  1562.     a->length = sum->length;

  1563.     av_free(sum);

  1564. }

  1565. 

  1566. void sws_subVec(SwsVector *a, SwsVector *b)

  1567. {

  1568.     SwsVector *diff = sws_diffVec(a, b);

  1569.     av_free(a->coeff);

  1570.     a->coeff  = diff->coeff;

  1571.     a->length = diff->length;

  1572.     av_free(diff);

  1573. }

  1574. 

  1575. void sws_convVec(SwsVector *a, SwsVector *b)

  1576. {

  1577.     SwsVector *conv = sws_getConvVec(a, b);

  1578.     av_free(a->coeff);

  1579.     a->coeff  = conv->coeff;

  1580.     a->length = conv->length;

  1581.     av_free(conv);

  1582. }

  1583. 

  1584. SwsVector *sws_cloneVec(SwsVector *a)

  1585. {

  1586.     int i;

  1587.     SwsVector *vec = sws_allocVec(a->length);

  1588. 

  1589.     if (!vec)

  1590.         return NULL;

  1591. 

  1592.     for (i = 0; i < a->length; i++)

  1593.         vec->coeff[i] = a->coeff[i];

  1594. 

  1595.     return vec;

  1596. }

  1597. 

  1598. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1599. {

  1600.     int i;

  1601.     double max = 0;

  1602.     double min = 0;

  1603.     double range;

  1604. 

  1605.     for (i = 0; i < a->length; i++)

  1606.         if (a->coeff[i] > max)

  1607.             max = a->coeff[i];

  1608. 

  1609.     for (i = 0; i < a->length; i++)

  1610.         if (a->coeff[i] < min)

  1611.             min = a->coeff[i];

  1612. 

  1613.     range = max - min;

  1614. 

  1615.     for (i = 0; i < a->length; i++) {

  1616.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1617.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1618.         for (; x > 0; x--)

  1619.             av_log(log_ctx, log_level, " ");

  1620.         av_log(log_ctx, log_level, "|\n");

  1621.     }

  1622. }

  1623. 

  1624. void sws_freeVec(SwsVector *a)

  1625. {

  1626.     if (!a)

  1627.         return;

  1628.     av_freep(&a->coeff);

  1629.     a->length = 0;

  1630.     av_free(a);

  1631. }

  1632. 

  1633. void sws_freeFilter(SwsFilter *filter)

  1634. {

  1635.     if (!filter)

  1636.         return;

  1637. 

  1638.     if (filter->lumH)

  1639.         sws_freeVec(filter->lumH);

  1640.     if (filter->lumV)

  1641.         sws_freeVec(filter->lumV);

  1642.     if (filter->chrH)

  1643.         sws_freeVec(filter->chrH);

  1644.     if (filter->chrV)

  1645.         sws_freeVec(filter->chrV);

  1646.     av_free(filter);

  1647. }

  1648. 

  1649. void sws_freeContext(SwsContext *c)

  1650. {

  1651.     int i;

  1652.     if (!c)

  1653.         return;

  1654. 

  1655.     if (c->lumPixBuf) {

  1656.         for (i = 0; i < c->vLumBufSize; i++)

  1657.             av_freep(&c->lumPixBuf[i]);

  1658.         av_freep(&c->lumPixBuf);

  1659.     }

  1660. 

  1661.     if (c->chrUPixBuf) {

  1662.         for (i = 0; i < c->vChrBufSize; i++)

  1663.             av_freep(&c->chrUPixBuf[i]);

  1664.         av_freep(&c->chrUPixBuf);

  1665.         av_freep(&c->chrVPixBuf);

  1666.     }

  1667. 

  1668.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1669.         for (i = 0; i < c->vLumBufSize; i++)

  1670.             av_freep(&c->alpPixBuf[i]);

  1671.         av_freep(&c->alpPixBuf);

  1672.     }

  1673. 

  1674.     av_freep(&c->vLumFilter);

  1675.     av_freep(&c->vChrFilter);

  1676.     av_freep(&c->hLumFilter);

  1677.     av_freep(&c->hChrFilter);

  1678. #if HAVE_ALTIVEC

  1679.     av_freep(&c->vYCoeffsBank);

  1680.     av_freep(&c->vCCoeffsBank);

  1681. #endif

  1682. 

  1683.     av_freep(&c->vLumFilterPos);

  1684.     av_freep(&c->vChrFilterPos);

  1685.     av_freep(&c->hLumFilterPos);

  1686.     av_freep(&c->hChrFilterPos);

  1687. 

  1688. #if HAVE_MMX

  1689. #ifdef MAP_ANONYMOUS

  1690.     if (c->lumMmx2FilterCode)

  1691.         munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1692.     if (c->chrMmx2FilterCode)

  1693.         munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1694. #elif HAVE_VIRTUALALLOC

  1695.     if (c->lumMmx2FilterCode)

  1696.         VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1697.     if (c->chrMmx2FilterCode)

  1698.         VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1699. #else

  1700.     av_free(c->lumMmx2FilterCode);

  1701.     av_free(c->chrMmx2FilterCode);

  1702. #endif

  1703.     c->lumMmx2FilterCode = NULL;

  1704.     c->chrMmx2FilterCode = NULL;

  1705. #endif /* HAVE_MMX */

  1706. 

  1707.     av_freep(&c->yuvTable);

  1708.     av_freep(&c->formatConvBuffer);

  1709. 

  1710.     av_free(c);

  1711. }

  1712. 

  1713. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  1714.                                         int srcH, enum PixelFormat srcFormat,

  1715.                                         int dstW, int dstH,

  1716.                                         enum PixelFormat dstFormat, int flags,

  1717.                                         SwsFilter *srcFilter,

  1718.                                         SwsFilter *dstFilter,

  1719.                                         const double *param)

  1720. {

  1721.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  1722.                                              SWS_PARAM_DEFAULT };

  1723. 

  1724.     if (!param)

  1725.         param = default_param;

  1726. 

  1727.     if (context &&

  1728.         (context->srcW      != srcW      ||

  1729.          context->srcH      != srcH      ||

  1730.          context->srcFormat != srcFormat ||

  1731.          context->dstW      != dstW      ||

  1732.          context->dstH      != dstH      ||

  1733.          context->dstFormat != dstFormat ||

  1734.          context->flags     != flags     ||

  1735.          context->param[0]  != param[0]  ||

  1736.          context->param[1]  != param[1])) {

  1737.         sws_freeContext(context);

  1738.         context = NULL;

  1739.     }

  1740. 

  1741.     if (!context) {

  1742.         if (!(context = sws_alloc_context()))

  1743.             return NULL;

  1744.         context->srcW      = srcW;

  1745.         context->srcH      = srcH;

  1746.         context->srcRange  = handle_jpeg(&srcFormat);

  1747.         context->src0Alpha = handle_0alpha(&srcFormat);

  1748.         context->srcFormat = srcFormat;

  1749.         context->dstW      = dstW;

  1750.         context->dstH      = dstH;

  1751.         context->dstRange  = handle_jpeg(&dstFormat);

  1752.         context->dst0Alpha = handle_0alpha(&dstFormat);

  1753.         context->dstFormat = dstFormat;

  1754.         context->flags     = flags;

  1755.         context->param[0]  = param[0];

  1756.         context->param[1]  = param[1];

  1757.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  1758.                                  context->srcRange,

  1759.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1760.                                  context->dstRange, 0, 1 << 16, 1 << 16);

  1761.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1762.             sws_freeContext(context);

  1763.             return NULL;

  1764.         }

  1765.     }

  1766.     return context;

  1767. }