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.
43714 Commits
32 Branches
283MB
Tree: 6c180b35c4
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '6c180b35c4'
${ noResults }
FFmpeg/libswscale/utils.c

1771 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/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 "rgb2rgb.h"

  51. #include "swscale.h"

  52. #include "swscale_internal.h"

  53. 

  54. unsigned swscale_version(void)

  55. {

  56.     av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);

  57.     return LIBSWSCALE_VERSION_INT;

  58. }

  59. 

  60. const char *swscale_configuration(void)

  61. {

  62.     return FFMPEG_CONFIGURATION;

  63. }

  64. 

  65. const char *swscale_license(void)

  66. {

  67. #define LICENSE_PREFIX "libswscale license: "

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

  69. }

  70. 

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

  72. 

  73. typedef struct FormatEntry {

  74.     int is_supported_in, is_supported_out;

  75. } FormatEntry;

  76. 

  77. static const FormatEntry format_entries[PIX_FMT_NB] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  180. };

  181. 

  182. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  183. {

  184.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  185.            format_entries[pix_fmt].is_supported_in : 0;

  186. }

  187. 

  188. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  189. {

  190.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  191.            format_entries[pix_fmt].is_supported_out : 0;

  192. }

  193. 

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

  195. 

  196. #if FF_API_SWS_FORMAT_NAME

  197. const char *sws_format_name(enum PixelFormat format)

  198. {

  199.     return av_get_pix_fmt_name(format);

  200. }

  201. #endif

  202. 

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

  204.                              double dist)

  205. {

  206.     if (dist <= 1.0)

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

  208.     else

  209.         return getSplineCoeff(0.0,

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

  211.                                c + 3.0 * d,

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

  213.                               dist - 1.0);

  214. }

  215. 

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

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

  218.                       int filterAlign, int one, int flags, int cpu_flags,

  219.                       SwsVector *srcFilter, SwsVector *dstFilter,

  220.                       double param[2])

  221. {

  222.     int i;

  223.     int filterSize;

  224.     int filter2Size;

  225.     int minFilterSize;

  226.     int64_t *filter    = NULL;

  227.     int64_t *filter2   = NULL;

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

  229.     int ret            = -1;

  230. 

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

  232. 

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

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

  235. 

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

  237.         int i;

  238.         filterSize = 1;

  239.         FF_ALLOCZ_OR_GOTO(NULL, filter,

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

  241. 

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

  243.             filter[i * filterSize] = fone;

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

  245.         }

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

  247.         int i;

  248.         int64_t xDstInSrc;

  249.         filterSize = 1;

  250.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  252. 

  253.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  256. 

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

  258.             filter[i]       = fone;

  259.             xDstInSrc      += xInc;

  260.         }

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

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

  263.         int i;

  264.         int64_t xDstInSrc;

  265.         filterSize = 2;

  266.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  268. 

  269.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  272.             int j;

  273. 

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

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

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

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

  278.                 if (coeff < 0)

  279.                     coeff = 0;

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

  281.                 xx++;

  282.             }

  283.             xDstInSrc += xInc;

  284.         }

  285.     } else {

  286.         int64_t xDstInSrc;

  287.         int sizeFactor;

  288. 

  289.         if (flags & SWS_BICUBIC)

  290.             sizeFactor = 4;

  291.         else if (flags & SWS_X)

  292.             sizeFactor = 8;

  293.         else if (flags & SWS_AREA)

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

  295.         else if (flags & SWS_GAUSS)

  296.             sizeFactor = 8;     // infinite ;)

  297.         else if (flags & SWS_LANCZOS)

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

  299.         else if (flags & SWS_SINC)

  300.             sizeFactor = 20;    // infinite ;)

  301.         else if (flags & SWS_SPLINE)

  302.             sizeFactor = 20;    // infinite ;)

  303.         else if (flags & SWS_BILINEAR)

  304.             sizeFactor = 2;

  305.         else {

  306.             av_assert0(0);

  307.         }

  308. 

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

  310.             filterSize = 1 + sizeFactor;    // upscale

  311.         else

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

  313. 

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

  315.         filterSize = FFMAX(filterSize, 1);

  316. 

  317.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  319. 

  320.         xDstInSrc = xInc - 0x10000;

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

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

  323.             int j;

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

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

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

  327.                 double floatd;

  328.                 int64_t coeff;

  329. 

  330.                 if (xInc > 1 << 16)

  331.                     d = d * dstW / srcW;

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

  333. 

  334.                 if (flags & SWS_BICUBIC) {

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

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

  337. 

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

  339.                         coeff = 0.0;

  340.                     } else {

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

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

  343. 

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

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

  346.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  347.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  348.                         else

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

  350.                                       (6 * B + 30 * C) * dd  +

  351.                                     (-12 * B - 48 * C) * d   +

  352.                                       (8 * B + 24 * C) * (1 << 30);

  353.                     }

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

  355.                 }

  356. #if 0

  357.                 else if (flags & SWS_X) {

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

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

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

  361.                 }

  362. #endif

  363.                 else if (flags & SWS_X) {

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

  365.                     double c;

  366. 

  367.                     if (floatd < 1.0)

  368.                         c = cos(floatd * M_PI);

  369.                     else

  370.                         c = -1.0;

  371.                     if (c < 0.0)

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

  373.                     else

  374.                         c = pow(c, A);

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

  376.                 } else if (flags & SWS_AREA) {

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

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

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

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

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

  382.                     else

  383.                         coeff = 0.0;

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

  385.                 } else if (flags & SWS_GAUSS) {

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

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

  388.                 } else if (flags & SWS_SINC) {

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

  390.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  394.                     if (floatd > p)

  395.                         coeff = 0;

  396.                 } else if (flags & SWS_BILINEAR) {

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

  398.                     if (coeff < 0)

  399.                         coeff = 0;

  400.                     coeff *= fone >> 30;

  401.                 } else if (flags & SWS_SPLINE) {

  402.                     double p = -2.196152422706632;

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

  404.                 } else {

  405.                     av_assert0(0);

  406.                 }

  407. 

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

  409.                 xx++;

  410.             }

  411.             xDstInSrc += 2 * xInc;

  412.         }

  413.     }

  414. 

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

  416.      * av_free(filter);

  417.      */

  418.     av_assert0(filterSize > 0);

  419.     filter2Size = filterSize;

  420.     if (srcFilter)

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

  422.     if (dstFilter)

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

  424.     av_assert0(filter2Size > 0);

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

  426. 

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

  428.         int j, k;

  429. 

  430.         if (srcFilter) {

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

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

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

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

  435.             }

  436.         } else {

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

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

  439.         }

  440.         // FIXME dstFilter

  441. 

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

  443.     }

  444.     av_freep(&filter);

  445. 

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

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

  448.     minFilterSize = 0;

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

  450.         int min = filter2Size;

  451.         int j;

  452.         int64_t cutOff = 0.0;

  453. 

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

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

  456.             int k;

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

  458. 

  459.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  460.                 break;

  461. 

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

  463.              * filter otherwise */

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

  465.                 break;

  466. 

  467.             // move filter coefficients left

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

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

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

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

  472.         }

  473. 

  474.         cutOff = 0;

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

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

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

  478. 

  479.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  480.                 break;

  481.             min--;

  482.         }

  483. 

  484.         if (min > minFilterSize)

  485.             minFilterSize = min;

  486.     }

  487. 

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

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

  490.         if (minFilterSize < 5)

  491.             filterAlign = 4;

  492. 

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

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

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

  496.         if (minFilterSize < 3)

  497.             filterAlign = 1;

  498.     }

  499. 

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

  501.         // special case for unscaled vertical filtering

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

  503.             filterAlign = 1;

  504.     }

  505. 

  506.     av_assert0(minFilterSize > 0);

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

  508.     av_assert0(filterSize > 0);

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

  510.     if (filterSize >= MAX_FILTER_SIZE * 16 /

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

  512.         goto fail;

  513.     *outFilterSize = filterSize;

  514. 

  515.     if (flags & SWS_PRINT_INFO)

  516.         av_log(NULL, AV_LOG_VERBOSE,

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

  518.                filter2Size, filterSize);

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

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

  521.         int j;

  522. 

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

  524.             if (j >= filter2Size)

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

  526.             else

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

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

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

  530.         }

  531.     }

  532. 

  533.     // FIXME try to align filterPos if possible

  534. 

  535.     // fix borders

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

  537.         int j;

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

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

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

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

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

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

  544.             }

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

  546.         }

  547. 

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

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

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

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

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

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

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

  555.             }

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

  557.         }

  558.     }

  559. 

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

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

  562.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

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

  564. 

  565.     /* normalize & store in outFilter */

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

  567.         int j;

  568.         int64_t error = 0;

  569.         int64_t sum   = 0;

  570. 

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

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

  573.         }

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

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

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

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

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

  579.             error                                  = v - intV * sum;

  580.         }

  581.     }

  582. 

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

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

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

  586.                                                       * read over the end */

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

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

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

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

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

  592.     }

  593. 

  594.     ret = 0;

  595. 

  596. fail:

  597.     av_free(filter);

  598.     av_free(filter2);

  599.     return ret;

  600. }

  601. 

  602. #if HAVE_MMXEXT && HAVE_INLINE_ASM

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

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

  605. {

  606.     uint8_t *fragmentA;

  607.     x86_reg imm8OfPShufW1A;

  608.     x86_reg imm8OfPShufW2A;

  609.     x86_reg fragmentLengthA;

  610.     uint8_t *fragmentB;

  611.     x86_reg imm8OfPShufW1B;

  612.     x86_reg imm8OfPShufW2B;

  613.     x86_reg fragmentLengthB;

  614.     int fragmentPos;

  615. 

  616.     int xpos, i;

  617. 

  618.     // create an optimized horizontal scaling routine

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

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

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

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

  623.      */

  624. 

  625.     // code fragment

  626. 

  627.     __asm__ volatile (

  628.         "jmp                         9f                 \n\t"

  629.         // Begin

  630.         "0:                                             \n\t"

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

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

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

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

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

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

  637.         "1:                                             \n\t"

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

  639.         "2:                                             \n\t"

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

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

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

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

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

  645. 

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

  647. 

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

  649.         // End

  650.         "9:                                             \n\t"

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

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

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

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

  655.         "dec                         %1                 \n\t"

  656.         "dec                         %2                 \n\t"

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

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

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

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

  661. 

  662. 

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

  664.           "=r" (fragmentLengthA)

  665.         );

  666. 

  667.     __asm__ volatile (

  668.         "jmp                         9f                 \n\t"

  669.         // Begin

  670.         "0:                                             \n\t"

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

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

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

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

  675.         "1:                                             \n\t"

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

  677.         "2:                                             \n\t"

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

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

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

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

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

  683. 

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

  685. 

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

  687.         // End

  688.         "9:                                             \n\t"

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

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

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

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

  693.         "dec                         %1                 \n\t"

  694.         "dec                         %2                 \n\t"

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

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

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

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

  699. 

  700. 

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

  702.           "=r" (fragmentLengthB)

  703.         );

  704. 

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

  706.     fragmentPos = 0;

  707. 

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

  709.         int xx = xpos >> 16;

  710. 

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

  712.             int a                  = 0;

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

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

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

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

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

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

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

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

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

  722.             int shift              = 0;

  723. 

  724.             if (filterCode) {

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

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

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

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

  729.                 filterPos[i / 2] = xx;

  730. 

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

  732. 

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

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

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

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

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

  738.                                                                (c << 4) |

  739.                                                                (d << 6);

  740. 

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

  742.                     shift = maxShift;               // avoid overread

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

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

  745. 

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

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

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

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

  750.                 }

  751.             }

  752. 

  753.             fragmentPos += fragmentLength;

  754. 

  755.             if (filterCode)

  756.                 filterCode[fragmentPos] = RET;

  757.         }

  758.         xpos += xInc;

  759.     }

  760.     if (filterCode)

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

  762. 

  763.     return fragmentPos + 1;

  764. }

  765. #endif /* HAVE_MMXEXT && HAVE_INLINE_ASM */

  766. 

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

  768. {

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

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

  771. }

  772. 

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

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

  775.                              int brightness, int contrast, int saturation)

  776. {

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

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

  779. 

  780.     c->brightness = brightness;

  781.     c->contrast   = contrast;

  782.     c->saturation = saturation;

  783.     c->srcRange   = srcRange;

  784.     c->dstRange   = dstRange;

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

  786.         return -1;

  787. 

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

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

  790. 

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

  792.                              contrast, saturation);

  793.     // FIXME factorize

  794. 

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

  796.         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,

  797.                                        contrast, saturation);

  798.     return 0;

  799. }

  800. 

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

  802.                              int *srcRange, int **table, int *dstRange,

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

  804. {

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

  806.         return -1;

  807. 

  808.     *inv_table  = c->srcColorspaceTable;

  809.     *table      = c->dstColorspaceTable;

  810.     *srcRange   = c->srcRange;

  811.     *dstRange   = c->dstRange;

  812.     *brightness = c->brightness;

  813.     *contrast   = c->contrast;

  814.     *saturation = c->saturation;

  815. 

  816.     return 0;

  817. }

  818. 

  819. static int handle_jpeg(enum PixelFormat *format)

  820. {

  821.     switch (*format) {

  822.     case PIX_FMT_YUVJ420P:

  823.         *format = PIX_FMT_YUV420P;

  824.         return 1;

  825.     case PIX_FMT_YUVJ422P:

  826.         *format = PIX_FMT_YUV422P;

  827.         return 1;

  828.     case PIX_FMT_YUVJ444P:

  829.         *format = PIX_FMT_YUV444P;

  830.         return 1;

  831.     case PIX_FMT_YUVJ440P:

  832.         *format = PIX_FMT_YUV440P;

  833.         return 1;

  834.     default:

  835.         return 0;

  836.     }

  837. }

  838. 

  839. static int handle_0alpha(enum PixelFormat *format)

  840. {

  841.     switch (*format) {

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

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

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

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

  846.     default:                                          return 0;

  847.     }

  848. }

  849. 

  850. SwsContext *sws_alloc_context(void)

  851. {

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

  853. 

  854.     c->av_class = &sws_context_class;

  855.     av_opt_set_defaults(c);

  856. 

  857.     return c;

  858. }

  859. 

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

  861.                              SwsFilter *dstFilter)

  862. {

  863.     int i, j;

  864.     int usesVFilter, usesHFilter;

  865.     int unscaled;

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

  867.     int srcW              = c->srcW;

  868.     int srcH              = c->srcH;

  869.     int dstW              = c->dstW;

  870.     int dstH              = c->dstH;

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

  872.     int flags, cpu_flags;

  873.     enum PixelFormat srcFormat = c->srcFormat;

  874.     enum PixelFormat dstFormat = c->dstFormat;

  875. 

  876.     cpu_flags = av_get_cpu_flags();

  877.     flags     = c->flags;

  878.     emms_c();

  879.     if (!rgb15to16)

  880.         sws_rgb2rgb_init();

  881. 

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

  883. 

  884.     handle_jpeg(&srcFormat);

  885.     handle_jpeg(&dstFormat);

  886.     handle_0alpha(&srcFormat);

  887.     handle_0alpha(&dstFormat);

  888. 

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

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

  891.         c->srcFormat= srcFormat;

  892.         c->dstFormat= dstFormat;

  893.     }

  894. 

  895.     if (!sws_isSupportedInput(srcFormat)) {

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

  897.                av_get_pix_fmt_name(srcFormat));

  898.         return AVERROR(EINVAL);

  899.     }

  900.     if (!sws_isSupportedOutput(dstFormat)) {

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

  902.                av_get_pix_fmt_name(dstFormat));

  903.         return AVERROR(EINVAL);

  904.     }

  905. 

  906.     i = flags & (SWS_POINT         |

  907.                  SWS_AREA          |

  908.                  SWS_BILINEAR      |

  909.                  SWS_FAST_BILINEAR |

  910.                  SWS_BICUBIC       |

  911.                  SWS_X             |

  912.                  SWS_GAUSS         |

  913.                  SWS_LANCZOS       |

  914.                  SWS_SINC          |

  915.                  SWS_SPLINE        |

  916.                  SWS_BICUBLIN);

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

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

  919.         return AVERROR(EINVAL);

  920.     }

  921.     /* sanity check */

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

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

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

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

  926.                srcW, srcH, dstW, dstH);

  927.         return AVERROR(EINVAL);

  928.     }

  929. 

  930.     if (!dstFilter)

  931.         dstFilter = &dummyFilter;

  932.     if (!srcFilter)

  933.         srcFilter = &dummyFilter;

  934. 

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

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

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

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

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

  940. 

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

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

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

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

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

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

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

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

  949. 

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

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

  952. 

  953. 

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

  955.         if (dstW&1) {

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

  957.             flags |= SWS_FULL_CHR_H_INT;

  958.             c->flags = flags;

  959.         }

  960.     }

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

  962.      * chroma interpolation */

  963.     if (flags & SWS_FULL_CHR_H_INT &&

  964.         isAnyRGB(dstFormat)        &&

  965.         dstFormat != PIX_FMT_RGBA  &&

  966.         dstFormat != PIX_FMT_ARGB  &&

  967.         dstFormat != PIX_FMT_BGRA  &&

  968.         dstFormat != PIX_FMT_ABGR  &&

  969.         dstFormat != PIX_FMT_RGB24 &&

  970.         dstFormat != PIX_FMT_BGR24) {

  971.         av_log(c, AV_LOG_WARNING,

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

  973.                av_get_pix_fmt_name(dstFormat));

  974.         flags   &= ~SWS_FULL_CHR_H_INT;

  975.         c->flags = flags;

  976.     }

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

  978.         c->chrDstHSubSample = 1;

  979. 

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

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

  982.                            SWS_SRC_V_CHR_DROP_SHIFT;

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

  984. 

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

  986.      * wants full chroma */

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

  988.         srcFormat != PIX_FMT_RGB8 && srcFormat != PIX_FMT_BGR8 &&

  989.         srcFormat != PIX_FMT_RGB4 && srcFormat != PIX_FMT_BGR4 &&

  990.         srcFormat != PIX_FMT_RGB4_BYTE && srcFormat != PIX_FMT_BGR4_BYTE &&

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

  992.          (flags & SWS_FAST_BILINEAR)))

  993.         c->chrSrcHSubSample = 1;

  994. 

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

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

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

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

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

  1000. 

  1001.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);

  1002. 

  1003.     /* unscaled special cases */

  1004.     if (unscaled && !usesHFilter && !usesVFilter &&

  1005.         (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {

  1006.         ff_get_unscaled_swscale(c);

  1007. 

  1008.         if (c->swScale) {

  1009.             if (flags & SWS_PRINT_INFO)

  1010.                 av_log(c, AV_LOG_INFO,

  1011.                        "using unscaled %s -> %s special converter\n",

  1012.                        av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));

  1013.             return 0;

  1014.         }

  1015.     }

  1016. 

  1017.     c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;

  1018.     if (c->srcBpc < 8)

  1019.         c->srcBpc = 8;

  1020.     c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;

  1021.     if (c->dstBpc < 8)

  1022.         c->dstBpc = 8;

  1023.     if (isAnyRGB(srcFormat) || srcFormat == PIX_FMT_PAL8)

  1024.         c->srcBpc = 16;

  1025.     if (c->dstBpc == 16)

  1026.         dst_stride <<= 1;

  1027.     if (HAVE_MMXEXT && HAVE_INLINE_ASM && cpu_flags & AV_CPU_FLAG_MMXEXT &&

  1028.         c->srcBpc == 8 && c->dstBpc <= 14) {

  1029.         c->canMMX2BeUsed = (dstW >= srcW && (dstW & 31) == 0 &&

  1030.                             (srcW & 15) == 0) ? 1 : 0;

  1031.         if (!c->canMMX2BeUsed && dstW >= srcW && (srcW & 15) == 0

  1032.             && (flags & SWS_FAST_BILINEAR)) {

  1033.             if (flags & SWS_PRINT_INFO)

  1034.                 av_log(c, AV_LOG_INFO,

  1035.                        "output width is not a multiple of 32 -> no MMX2 scaler\n");

  1036.         }

  1037.         if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))

  1038.             c->canMMX2BeUsed=0;

  1039.     } else

  1040.         c->canMMX2BeUsed = 0;

  1041. 

  1042.     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;

  1043.     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

  1044. 

  1045.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  1046.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  1047.      * correct scaling.

  1048.      * n-2 is the last chrominance sample available.

  1049.      * This is not perfect, but no one should notice the difference, the more

  1050.      * correct variant would be like the vertical one, but that would require

  1051.      * some special code for the first and last pixel */

  1052.     if (flags & SWS_FAST_BILINEAR) {

  1053.         if (c->canMMX2BeUsed) {

  1054.             c->lumXInc += 20;

  1055.             c->chrXInc += 20;

  1056.         }

  1057.         // we don't use the x86 asm scaler if MMX is available

  1058.         else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX && c->dstBpc <= 14) {

  1059.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1060.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1061.         }

  1062.     }

  1063. 

  1064.     /* precalculate horizontal scaler filter coefficients */

  1065.     {

  1066. #if HAVE_MMXEXT && HAVE_INLINE_ASM

  1067. // can't downscale !!!

  1068.         if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {

  1069.             c->lumMmx2FilterCodeSize = initMMX2HScaler(dstW, c->lumXInc, NULL,

  1070.                                                        NULL, NULL, 8);

  1071.             c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc,

  1072.                                                        NULL, NULL, NULL, 4);

  1073. 

  1074. #ifdef MAP_ANONYMOUS

  1075.             c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);

  1076.             c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);

  1077. #elif HAVE_VIRTUALALLOC

  1078.             c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);

  1079.             c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);

  1080. #else

  1081.             c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);

  1082.             c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);

  1083. #endif

  1084. 

  1085. #ifdef MAP_ANONYMOUS

  1086.             if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)

  1087. #else

  1088.             if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)

  1089. #endif

  1090.             {

  1091.                 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");

  1092.                 return AVERROR(ENOMEM);

  1093.             }

  1094. 

  1095.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1096.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1097.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1098.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1099. 

  1100.             initMMX2HScaler(      dstW, c->lumXInc, c->lumMmx2FilterCode,

  1101.                             c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);

  1102.             initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode,

  1103.                             c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);

  1104. 

  1105. #ifdef MAP_ANONYMOUS

  1106.             mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);

  1107.             mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);

  1108. #endif

  1109.         } else

  1110. #endif /* HAVE_MMXEXT && HAVE_INLINE_ASM */

  1111.         {

  1112.             const int filterAlign =

  1113.                 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :

  1114.                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1115.                 1;

  1116. 

  1117.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1118.                            &c->hLumFilterSize, c->lumXInc,

  1119.                            srcW, dstW, filterAlign, 1 << 14,

  1120.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1121.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1122.                            c->param) < 0)

  1123.                 goto fail;

  1124.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1125.                            &c->hChrFilterSize, c->chrXInc,

  1126.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1127.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1128.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1129.                            c->param) < 0)

  1130.                 goto fail;

  1131.         }

  1132.     } // initialize horizontal stuff

  1133. 

  1134.     /* precalculate vertical scaler filter coefficients */

  1135.     {

  1136.         const int filterAlign =

  1137.             (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :

  1138.             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1139.             1;

  1140. 

  1141.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1142.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

  1143.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1144.                        cpu_flags, srcFilter->lumV, dstFilter->lumV,

  1145.                        c->param) < 0)

  1146.             goto fail;

  1147.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1148.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1149.                        filterAlign, (1 << 12),

  1150.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1151.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1152.                        c->param) < 0)

  1153.             goto fail;

  1154. 

  1155. #if HAVE_ALTIVEC

  1156.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1157.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1158. 

  1159.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1160.             int j;

  1161.             short *p = (short *)&c->vYCoeffsBank[i];

  1162.             for (j = 0; j < 8; j++)

  1163.                 p[j] = c->vLumFilter[i];

  1164.         }

  1165. 

  1166.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1167.             int j;

  1168.             short *p = (short *)&c->vCCoeffsBank[i];

  1169.             for (j = 0; j < 8; j++)

  1170.                 p[j] = c->vChrFilter[i];

  1171.         }

  1172. #endif

  1173.     }

  1174. 

  1175.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1176.     c->vLumBufSize = c->vLumFilterSize;

  1177.     c->vChrBufSize = c->vChrFilterSize;

  1178.     for (i = 0; i < dstH; i++) {

  1179.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1180.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1181.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1182.                                << c->chrSrcVSubSample));

  1183. 

  1184.         nextSlice >>= c->chrSrcVSubSample;

  1185.         nextSlice <<= c->chrSrcVSubSample;

  1186.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1187.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1188.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1189.             (nextSlice >> c->chrSrcVSubSample))

  1190.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1191.                              c->vChrFilterPos[chrI];

  1192.     }

  1193. 

  1194.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1195.      * need to allocate several megabytes to handle all possible cases) */

  1196.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1197.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1198.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1199.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1200.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1201.     /* Note we need at least one pixel more at the end because of the MMX code

  1202.      * (just in case someone wants to replace the 4000/8000). */

  1203.     /* align at 16 bytes for AltiVec */

  1204.     for (i = 0; i < c->vLumBufSize; i++) {

  1205.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1206.                           dst_stride + 16, fail);

  1207.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1208.     }

  1209.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1210.     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);

  1211.     c->uv_offx2 = dst_stride + 16;

  1212.     for (i = 0; i < c->vChrBufSize; i++) {

  1213.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1214.                          dst_stride * 2 + 32, fail);

  1215.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1216.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1217.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1218.     }

  1219.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1220.         for (i = 0; i < c->vLumBufSize; i++) {

  1221.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],

  1222.                               dst_stride + 16, fail);

  1223.             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];

  1224.         }

  1225. 

  1226.     // try to avoid drawing green stuff between the right end and the stride end

  1227.     for (i = 0; i < c->vChrBufSize; i++)

  1228.         if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){

  1229.             av_assert0(c->dstBpc > 14);

  1230.             for(j=0; j<dst_stride/2+1; j++)

  1231.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1232.         } else

  1233.             for(j=0; j<dst_stride+1; j++)

  1234.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1235. 

  1236.     av_assert0(c->chrDstH <= dstH);

  1237. 

  1238.     if (flags & SWS_PRINT_INFO) {

  1239.         if (flags & SWS_FAST_BILINEAR)

  1240.             av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1241.         else if (flags & SWS_BILINEAR)

  1242.             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1243.         else if (flags & SWS_BICUBIC)

  1244.             av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1245.         else if (flags & SWS_X)

  1246.             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1247.         else if (flags & SWS_POINT)

  1248.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1249.         else if (flags & SWS_AREA)

  1250.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1251.         else if (flags & SWS_BICUBLIN)

  1252.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1253.         else if (flags & SWS_GAUSS)

  1254.             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1255.         else if (flags & SWS_SINC)

  1256.             av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1257.         else if (flags & SWS_LANCZOS)

  1258.             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1259.         else if (flags & SWS_SPLINE)

  1260.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1261.         else

  1262.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1263. 

  1264.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1265.                av_get_pix_fmt_name(srcFormat),

  1266. #ifdef DITHER1XBPP

  1267.                dstFormat == PIX_FMT_BGR555   || dstFormat == PIX_FMT_BGR565   ||

  1268.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1269.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ?

  1270.                                                              "dithered " : "",

  1271. #else

  1272.                "",

  1273. #endif

  1274.                av_get_pix_fmt_name(dstFormat));

  1275. 

  1276.         if (HAVE_MMXEXT && cpu_flags & AV_CPU_FLAG_MMXEXT)

  1277.             av_log(c, AV_LOG_INFO, "using MMX2\n");

  1278.         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)

  1279.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1280.         else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX)

  1281.             av_log(c, AV_LOG_INFO, "using MMX\n");

  1282.         else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)

  1283.             av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1284.         else

  1285.             av_log(c, AV_LOG_INFO, "using C\n");

  1286. 

  1287.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1288.         av_log(c, AV_LOG_DEBUG,

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

  1290.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1291.         av_log(c, AV_LOG_DEBUG,

  1292.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1293.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1294.                c->chrXInc, c->chrYInc);

  1295.     }

  1296. 

  1297.     c->swScale = ff_getSwsFunc(c);

  1298.     return 0;

  1299. fail: // FIXME replace things by appropriate error codes

  1300.     return -1;

  1301. }

  1302. 

  1303. #if FF_API_SWS_GETCONTEXT

  1304. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1305.                            int dstW, int dstH, enum PixelFormat dstFormat,

  1306.                            int flags, SwsFilter *srcFilter,

  1307.                            SwsFilter *dstFilter, const double *param)

  1308. {

  1309.     SwsContext *c;

  1310. 

  1311.     if (!(c = sws_alloc_context()))

  1312.         return NULL;

  1313. 

  1314.     c->flags     = flags;

  1315.     c->srcW      = srcW;

  1316.     c->srcH      = srcH;

  1317.     c->dstW      = dstW;

  1318.     c->dstH      = dstH;

  1319.     c->srcRange  = handle_jpeg(&srcFormat);

  1320.     c->dstRange  = handle_jpeg(&dstFormat);

  1321.     c->src0Alpha = handle_0alpha(&srcFormat);

  1322.     c->dst0Alpha = handle_0alpha(&dstFormat);

  1323.     c->srcFormat = srcFormat;

  1324.     c->dstFormat = dstFormat;

  1325. 

  1326.     if (param) {

  1327.         c->param[0] = param[0];

  1328.         c->param[1] = param[1];

  1329.     }

  1330.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1331.                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1332.                              c->dstRange, 0, 1 << 16, 1 << 16);

  1333. 

  1334.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1335.         sws_freeContext(c);

  1336.         return NULL;

  1337.     }

  1338. 

  1339.     return c;

  1340. }

  1341. #endif

  1342. 

  1343. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1344.                                 float lumaSharpen, float chromaSharpen,

  1345.                                 float chromaHShift, float chromaVShift,

  1346.                                 int verbose)

  1347. {

  1348.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1349.     if (!filter)

  1350.         return NULL;

  1351. 

  1352.     if (lumaGBlur != 0.0) {

  1353.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1354.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1355.     } else {

  1356.         filter->lumH = sws_getIdentityVec();

  1357.         filter->lumV = sws_getIdentityVec();

  1358.     }

  1359. 

  1360.     if (chromaGBlur != 0.0) {

  1361.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1362.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1363.     } else {

  1364.         filter->chrH = sws_getIdentityVec();

  1365.         filter->chrV = sws_getIdentityVec();

  1366.     }

  1367. 

  1368.     if (chromaSharpen != 0.0) {

  1369.         SwsVector *id = sws_getIdentityVec();

  1370.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1371.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1372.         sws_addVec(filter->chrH, id);

  1373.         sws_addVec(filter->chrV, id);

  1374.         sws_freeVec(id);

  1375.     }

  1376. 

  1377.     if (lumaSharpen != 0.0) {

  1378.         SwsVector *id = sws_getIdentityVec();

  1379.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1380.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1381.         sws_addVec(filter->lumH, id);

  1382.         sws_addVec(filter->lumV, id);

  1383.         sws_freeVec(id);

  1384.     }

  1385. 

  1386.     if (chromaHShift != 0.0)

  1387.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1388. 

  1389.     if (chromaVShift != 0.0)

  1390.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1391. 

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

  1393.     sws_normalizeVec(filter->chrV, 1.0);

  1394.     sws_normalizeVec(filter->lumH, 1.0);

  1395.     sws_normalizeVec(filter->lumV, 1.0);

  1396. 

  1397.     if (verbose)

  1398.         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1399.     if (verbose)

  1400.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1401. 

  1402.     return filter;

  1403. }

  1404. 

  1405. SwsVector *sws_allocVec(int length)

  1406. {

  1407.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1408.     if (!vec)

  1409.         return NULL;

  1410.     vec->length = length;

  1411.     vec->coeff  = av_malloc(sizeof(double) * length);

  1412.     if (!vec->coeff)

  1413.         av_freep(&vec);

  1414.     return vec;

  1415. }

  1416. 

  1417. SwsVector *sws_getGaussianVec(double variance, double quality)

  1418. {

  1419.     const int length = (int)(variance * quality + 0.5) | 1;

  1420.     int i;

  1421.     double middle  = (length - 1) * 0.5;

  1422.     SwsVector *vec = sws_allocVec(length);

  1423. 

  1424.     if (!vec)

  1425.         return NULL;

  1426. 

  1427.     for (i = 0; i < length; i++) {

  1428.         double dist = i - middle;

  1429.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1430.                         sqrt(2 * variance * M_PI);

  1431.     }

  1432. 

  1433.     sws_normalizeVec(vec, 1.0);

  1434. 

  1435.     return vec;

  1436. }

  1437. 

  1438. SwsVector *sws_getConstVec(double c, int length)

  1439. {

  1440.     int i;

  1441.     SwsVector *vec = sws_allocVec(length);

  1442. 

  1443.     if (!vec)

  1444.         return NULL;

  1445. 

  1446.     for (i = 0; i < length; i++)

  1447.         vec->coeff[i] = c;

  1448. 

  1449.     return vec;

  1450. }

  1451. 

  1452. SwsVector *sws_getIdentityVec(void)

  1453. {

  1454.     return sws_getConstVec(1.0, 1);

  1455. }

  1456. 

  1457. static double sws_dcVec(SwsVector *a)

  1458. {

  1459.     int i;

  1460.     double sum = 0;

  1461. 

  1462.     for (i = 0; i < a->length; i++)

  1463.         sum += a->coeff[i];

  1464. 

  1465.     return sum;

  1466. }

  1467. 

  1468. void sws_scaleVec(SwsVector *a, double scalar)

  1469. {

  1470.     int i;

  1471. 

  1472.     for (i = 0; i < a->length; i++)

  1473.         a->coeff[i] *= scalar;

  1474. }

  1475. 

  1476. void sws_normalizeVec(SwsVector *a, double height)

  1477. {

  1478.     sws_scaleVec(a, height / sws_dcVec(a));

  1479. }

  1480. 

  1481. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1482. {

  1483.     int length = a->length + b->length - 1;

  1484.     int i, j;

  1485.     SwsVector *vec = sws_getConstVec(0.0, length);

  1486. 

  1487.     if (!vec)

  1488.         return NULL;

  1489. 

  1490.     for (i = 0; i < a->length; i++) {

  1491.         for (j = 0; j < b->length; j++) {

  1492.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1493.         }

  1494.     }

  1495. 

  1496.     return vec;

  1497. }

  1498. 

  1499. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1500. {

  1501.     int length = FFMAX(a->length, b->length);

  1502.     int i;

  1503.     SwsVector *vec = sws_getConstVec(0.0, length);

  1504. 

  1505.     if (!vec)

  1506.         return NULL;

  1507. 

  1508.     for (i = 0; i < a->length; i++)

  1509.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1510.     for (i = 0; i < b->length; i++)

  1511.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

  1512. 

  1513.     return vec;

  1514. }

  1515. 

  1516. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1517. {

  1518.     int length = FFMAX(a->length, b->length);

  1519.     int i;

  1520.     SwsVector *vec = sws_getConstVec(0.0, length);

  1521. 

  1522.     if (!vec)

  1523.         return NULL;

  1524. 

  1525.     for (i = 0; i < a->length; i++)

  1526.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1527.     for (i = 0; i < b->length; i++)

  1528.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

  1529. 

  1530.     return vec;

  1531. }

  1532. 

  1533. /* shift left / or right if "shift" is negative */

  1534. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1535. {

  1536.     int length = a->length + FFABS(shift) * 2;

  1537.     int i;

  1538.     SwsVector *vec = sws_getConstVec(0.0, length);

  1539. 

  1540.     if (!vec)

  1541.         return NULL;

  1542. 

  1543.     for (i = 0; i < a->length; i++) {

  1544.         vec->coeff[i + (length    - 1) / 2 -

  1545.                        (a->length - 1) / 2 - shift] = a->coeff[i];

  1546.     }

  1547. 

  1548.     return vec;

  1549. }

  1550. 

  1551. void sws_shiftVec(SwsVector *a, int shift)

  1552. {

  1553.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1554.     av_free(a->coeff);

  1555.     a->coeff  = shifted->coeff;

  1556.     a->length = shifted->length;

  1557.     av_free(shifted);

  1558. }

  1559. 

  1560. void sws_addVec(SwsVector *a, SwsVector *b)

  1561. {

  1562.     SwsVector *sum = sws_sumVec(a, b);

  1563.     av_free(a->coeff);

  1564.     a->coeff  = sum->coeff;

  1565.     a->length = sum->length;

  1566.     av_free(sum);

  1567. }

  1568. 

  1569. void sws_subVec(SwsVector *a, SwsVector *b)

  1570. {

  1571.     SwsVector *diff = sws_diffVec(a, b);

  1572.     av_free(a->coeff);

  1573.     a->coeff  = diff->coeff;

  1574.     a->length = diff->length;

  1575.     av_free(diff);

  1576. }

  1577. 

  1578. void sws_convVec(SwsVector *a, SwsVector *b)

  1579. {

  1580.     SwsVector *conv = sws_getConvVec(a, b);

  1581.     av_free(a->coeff);

  1582.     a->coeff  = conv->coeff;

  1583.     a->length = conv->length;

  1584.     av_free(conv);

  1585. }

  1586. 

  1587. SwsVector *sws_cloneVec(SwsVector *a)

  1588. {

  1589.     int i;

  1590.     SwsVector *vec = sws_allocVec(a->length);

  1591. 

  1592.     if (!vec)

  1593.         return NULL;

  1594. 

  1595.     for (i = 0; i < a->length; i++)

  1596.         vec->coeff[i] = a->coeff[i];

  1597. 

  1598.     return vec;

  1599. }

  1600. 

  1601. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1602. {

  1603.     int i;

  1604.     double max = 0;

  1605.     double min = 0;

  1606.     double range;

  1607. 

  1608.     for (i = 0; i < a->length; i++)

  1609.         if (a->coeff[i] > max)

  1610.             max = a->coeff[i];

  1611. 

  1612.     for (i = 0; i < a->length; i++)

  1613.         if (a->coeff[i] < min)

  1614.             min = a->coeff[i];

  1615. 

  1616.     range = max - min;

  1617. 

  1618.     for (i = 0; i < a->length; i++) {

  1619.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1620.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1621.         for (; x > 0; x--)

  1622.             av_log(log_ctx, log_level, " ");

  1623.         av_log(log_ctx, log_level, "|\n");

  1624.     }

  1625. }

  1626. 

  1627. void sws_freeVec(SwsVector *a)

  1628. {

  1629.     if (!a)

  1630.         return;

  1631.     av_freep(&a->coeff);

  1632.     a->length = 0;

  1633.     av_free(a);

  1634. }

  1635. 

  1636. void sws_freeFilter(SwsFilter *filter)

  1637. {

  1638.     if (!filter)

  1639.         return;

  1640. 

  1641.     if (filter->lumH)

  1642.         sws_freeVec(filter->lumH);

  1643.     if (filter->lumV)

  1644.         sws_freeVec(filter->lumV);

  1645.     if (filter->chrH)

  1646.         sws_freeVec(filter->chrH);

  1647.     if (filter->chrV)

  1648.         sws_freeVec(filter->chrV);

  1649.     av_free(filter);

  1650. }

  1651. 

  1652. void sws_freeContext(SwsContext *c)

  1653. {

  1654.     int i;

  1655.     if (!c)

  1656.         return;

  1657. 

  1658.     if (c->lumPixBuf) {

  1659.         for (i = 0; i < c->vLumBufSize; i++)

  1660.             av_freep(&c->lumPixBuf[i]);

  1661.         av_freep(&c->lumPixBuf);

  1662.     }

  1663. 

  1664.     if (c->chrUPixBuf) {

  1665.         for (i = 0; i < c->vChrBufSize; i++)

  1666.             av_freep(&c->chrUPixBuf[i]);

  1667.         av_freep(&c->chrUPixBuf);

  1668.         av_freep(&c->chrVPixBuf);

  1669.     }

  1670. 

  1671.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1672.         for (i = 0; i < c->vLumBufSize; i++)

  1673.             av_freep(&c->alpPixBuf[i]);

  1674.         av_freep(&c->alpPixBuf);

  1675.     }

  1676. 

  1677.     av_freep(&c->vLumFilter);

  1678.     av_freep(&c->vChrFilter);

  1679.     av_freep(&c->hLumFilter);

  1680.     av_freep(&c->hChrFilter);

  1681. #if HAVE_ALTIVEC

  1682.     av_freep(&c->vYCoeffsBank);

  1683.     av_freep(&c->vCCoeffsBank);

  1684. #endif

  1685. 

  1686.     av_freep(&c->vLumFilterPos);

  1687.     av_freep(&c->vChrFilterPos);

  1688.     av_freep(&c->hLumFilterPos);

  1689.     av_freep(&c->hChrFilterPos);

  1690. 

  1691. #if HAVE_MMX

  1692. #ifdef MAP_ANONYMOUS

  1693.     if (c->lumMmx2FilterCode)

  1694.         munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1695.     if (c->chrMmx2FilterCode)

  1696.         munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1697. #elif HAVE_VIRTUALALLOC

  1698.     if (c->lumMmx2FilterCode)

  1699.         VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1700.     if (c->chrMmx2FilterCode)

  1701.         VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1702. #else

  1703.     av_free(c->lumMmx2FilterCode);

  1704.     av_free(c->chrMmx2FilterCode);

  1705. #endif

  1706.     c->lumMmx2FilterCode = NULL;

  1707.     c->chrMmx2FilterCode = NULL;

  1708. #endif /* HAVE_MMX */

  1709. 

  1710.     av_freep(&c->yuvTable);

  1711.     av_freep(&c->formatConvBuffer);

  1712. 

  1713.     av_free(c);

  1714. }

  1715. 

  1716. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  1717.                                         int srcH, enum PixelFormat srcFormat,

  1718.                                         int dstW, int dstH,

  1719.                                         enum PixelFormat dstFormat, int flags,

  1720.                                         SwsFilter *srcFilter,

  1721.                                         SwsFilter *dstFilter,

  1722.                                         const double *param)

  1723. {

  1724.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  1725.                                              SWS_PARAM_DEFAULT };

  1726. 

  1727.     if (!param)

  1728.         param = default_param;

  1729. 

  1730.     if (context &&

  1731.         (context->srcW      != srcW      ||

  1732.          context->srcH      != srcH      ||

  1733.          context->srcFormat != srcFormat ||

  1734.          context->dstW      != dstW      ||

  1735.          context->dstH      != dstH      ||

  1736.          context->dstFormat != dstFormat ||

  1737.          context->flags     != flags     ||

  1738.          context->param[0]  != param[0]  ||

  1739.          context->param[1]  != param[1])) {

  1740.         sws_freeContext(context);

  1741.         context = NULL;

  1742.     }

  1743. 

  1744.     if (!context) {

  1745.         if (!(context = sws_alloc_context()))

  1746.             return NULL;

  1747.         context->srcW      = srcW;

  1748.         context->srcH      = srcH;

  1749.         context->srcRange  = handle_jpeg(&srcFormat);

  1750.         context->src0Alpha = handle_0alpha(&srcFormat);

  1751.         context->srcFormat = srcFormat;

  1752.         context->dstW      = dstW;

  1753.         context->dstH      = dstH;

  1754.         context->dstRange  = handle_jpeg(&dstFormat);

  1755.         context->dst0Alpha = handle_0alpha(&dstFormat);

  1756.         context->dstFormat = dstFormat;

  1757.         context->flags     = flags;

  1758.         context->param[0]  = param[0];

  1759.         context->param[1]  = param[1];

  1760.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  1761.                                  context->srcRange,

  1762.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1763.                                  context->dstRange, 0, 1 << 16, 1 << 16);

  1764.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1765.             sws_freeContext(context);

  1766.             return NULL;

  1767.         }

  1768.     }

  1769.     return context;

  1770. }