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.
42039 Commits
32 Branches
301MB
Tree: a19e9f2d5c
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'a19e9f2d5c'
${ noResults }
FFmpeg/libswscale/utils.c

1752 lines
61KB
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_YUV422P9BE]  = { 1, 1 },

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  163. };

  164. 

  165. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  166. {

  167.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  168.            format_entries[pix_fmt].is_supported_in : 0;

  169. }

  170. 

  171. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  172. {

  173.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  174.            format_entries[pix_fmt].is_supported_out : 0;

  175. }

  176. 

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

  178. 

  179. #if FF_API_SWS_FORMAT_NAME

  180. const char *sws_format_name(enum PixelFormat format)

  181. {

  182.     return av_get_pix_fmt_name(format);

  183. }

  184. #endif

  185. 

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

  187.                              double dist)

  188. {

  189.     if (dist <= 1.0)

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

  191.     else

  192.         return getSplineCoeff(0.0,

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

  194.                                c + 3.0 * d,

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

  196.                               dist - 1.0);

  197. }

  198. 

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

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

  201.                       int filterAlign, int one, int flags, int cpu_flags,

  202.                       SwsVector *srcFilter, SwsVector *dstFilter,

  203.                       double param[2])

  204. {

  205.     int i;

  206.     int filterSize;

  207.     int filter2Size;

  208.     int minFilterSize;

  209.     int64_t *filter    = NULL;

  210.     int64_t *filter2   = NULL;

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

  212.     int ret            = -1;

  213. 

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

  215. 

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

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

  218. 

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

  220.         int i;

  221.         filterSize = 1;

  222.         FF_ALLOCZ_OR_GOTO(NULL, filter,

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

  224. 

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

  226.             filter[i * filterSize] = fone;

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

  228.         }

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

  230.         int i;

  231.         int64_t xDstInSrc;

  232.         filterSize = 1;

  233.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  235. 

  236.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  239. 

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

  241.             filter[i]       = fone;

  242.             xDstInSrc      += xInc;

  243.         }

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

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

  246.         int i;

  247.         int64_t xDstInSrc;

  248.         filterSize = 2;

  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.             int j;

  256. 

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

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

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

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

  261.                 if (coeff < 0)

  262.                     coeff = 0;

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

  264.                 xx++;

  265.             }

  266.             xDstInSrc += xInc;

  267.         }

  268.     } else {

  269.         int64_t xDstInSrc;

  270.         int sizeFactor;

  271. 

  272.         if (flags & SWS_BICUBIC)

  273.             sizeFactor = 4;

  274.         else if (flags & SWS_X)

  275.             sizeFactor = 8;

  276.         else if (flags & SWS_AREA)

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

  278.         else if (flags & SWS_GAUSS)

  279.             sizeFactor = 8;     // infinite ;)

  280.         else if (flags & SWS_LANCZOS)

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

  282.         else if (flags & SWS_SINC)

  283.             sizeFactor = 20;    // infinite ;)

  284.         else if (flags & SWS_SPLINE)

  285.             sizeFactor = 20;    // infinite ;)

  286.         else if (flags & SWS_BILINEAR)

  287.             sizeFactor = 2;

  288.         else {

  289.             av_assert0(0);

  290.         }

  291. 

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

  293.             filterSize = 1 + sizeFactor;    // upscale

  294.         else

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

  296. 

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

  298.         filterSize = FFMAX(filterSize, 1);

  299. 

  300.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  302. 

  303.         xDstInSrc = xInc - 0x10000;

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

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

  306.             int j;

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

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

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

  310.                 double floatd;

  311.                 int64_t coeff;

  312. 

  313.                 if (xInc > 1 << 16)

  314.                     d = d * dstW / srcW;

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

  316. 

  317.                 if (flags & SWS_BICUBIC) {

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

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

  320. 

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

  322.                         coeff = 0.0;

  323.                     } else {

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

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

  326. 

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

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

  329.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  330.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  331.                         else

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

  333.                                       (6 * B + 30 * C) * dd  +

  334.                                     (-12 * B - 48 * C) * d   +

  335.                                       (8 * B + 24 * C) * (1 << 30);

  336.                     }

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

  338.                 }

  339. #if 0

  340.                 else if (flags & SWS_X) {

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

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

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

  344.                 }

  345. #endif

  346.                 else if (flags & SWS_X) {

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

  348.                     double c;

  349. 

  350.                     if (floatd < 1.0)

  351.                         c = cos(floatd * M_PI);

  352.                     else

  353.                         c = -1.0;

  354.                     if (c < 0.0)

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

  356.                     else

  357.                         c = pow(c, A);

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

  359.                 } else if (flags & SWS_AREA) {

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

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

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

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

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

  365.                     else

  366.                         coeff = 0.0;

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

  368.                 } else if (flags & SWS_GAUSS) {

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

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

  371.                 } else if (flags & SWS_SINC) {

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

  373.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  377.                     if (floatd > p)

  378.                         coeff = 0;

  379.                 } else if (flags & SWS_BILINEAR) {

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

  381.                     if (coeff < 0)

  382.                         coeff = 0;

  383.                     coeff *= fone >> 30;

  384.                 } else if (flags & SWS_SPLINE) {

  385.                     double p = -2.196152422706632;

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

  387.                 } else {

  388.                     av_assert0(0);

  389.                 }

  390. 

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

  392.                 xx++;

  393.             }

  394.             xDstInSrc += 2 * xInc;

  395.         }

  396.     }

  397. 

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

  399.      * av_free(filter);

  400.      */

  401.     av_assert0(filterSize > 0);

  402.     filter2Size = filterSize;

  403.     if (srcFilter)

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

  405.     if (dstFilter)

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

  407.     av_assert0(filter2Size > 0);

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

  409. 

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

  411.         int j, k;

  412. 

  413.         if (srcFilter) {

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

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

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

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

  418.             }

  419.         } else {

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

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

  422.         }

  423.         // FIXME dstFilter

  424. 

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

  426.     }

  427.     av_freep(&filter);

  428. 

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

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

  431.     minFilterSize = 0;

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

  433.         int min = filter2Size;

  434.         int j;

  435.         int64_t cutOff = 0.0;

  436. 

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

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

  439.             int k;

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

  441. 

  442.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  443.                 break;

  444. 

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

  446.              * filter otherwise */

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

  448.                 break;

  449. 

  450.             // move filter coefficients left

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

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

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

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

  455.         }

  456. 

  457.         cutOff = 0;

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

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

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

  461. 

  462.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  463.                 break;

  464.             min--;

  465.         }

  466. 

  467.         if (min > minFilterSize)

  468.             minFilterSize = min;

  469.     }

  470. 

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

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

  473.         if (minFilterSize < 5)

  474.             filterAlign = 4;

  475. 

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

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

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

  479.         if (minFilterSize < 3)

  480.             filterAlign = 1;

  481.     }

  482. 

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

  484.         // special case for unscaled vertical filtering

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

  486.             filterAlign = 1;

  487.     }

  488. 

  489.     av_assert0(minFilterSize > 0);

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

  491.     av_assert0(filterSize > 0);

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

  493.     if (filterSize >= MAX_FILTER_SIZE * 16 /

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

  495.         goto fail;

  496.     *outFilterSize = filterSize;

  497. 

  498.     if (flags & SWS_PRINT_INFO)

  499.         av_log(NULL, AV_LOG_VERBOSE,

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

  501.                filter2Size, filterSize);

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

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

  504.         int j;

  505. 

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

  507.             if (j >= filter2Size)

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

  509.             else

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

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

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

  513.         }

  514.     }

  515. 

  516.     // FIXME try to align filterPos if possible

  517. 

  518.     // fix borders

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

  520.         int j;

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

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

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

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

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

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

  527.             }

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

  529.         }

  530. 

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

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

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

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

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

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

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

  538.             }

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

  540.         }

  541.     }

  542. 

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

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

  545.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

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

  547. 

  548.     /* normalize & store in outFilter */

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

  550.         int j;

  551.         int64_t error = 0;

  552.         int64_t sum   = 0;

  553. 

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

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

  556.         }

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

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

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

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

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

  562.             error                                  = v - intV * sum;

  563.         }

  564.     }

  565. 

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

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

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

  569.                                                       * read over the end */

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

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

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

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

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

  575.     }

  576. 

  577.     ret = 0;

  578. 

  579. fail:

  580.     av_free(filter);

  581.     av_free(filter2);

  582.     return ret;

  583. }

  584. 

  585. #if HAVE_MMX2

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

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

  588. {

  589.     uint8_t *fragmentA;

  590.     x86_reg imm8OfPShufW1A;

  591.     x86_reg imm8OfPShufW2A;

  592.     x86_reg fragmentLengthA;

  593.     uint8_t *fragmentB;

  594.     x86_reg imm8OfPShufW1B;

  595.     x86_reg imm8OfPShufW2B;

  596.     x86_reg fragmentLengthB;

  597.     int fragmentPos;

  598. 

  599.     int xpos, i;

  600. 

  601.     // create an optimized horizontal scaling routine

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

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

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

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

  606.      */

  607. 

  608.     // code fragment

  609. 

  610.     __asm__ volatile (

  611.         "jmp                         9f                 \n\t"

  612.         // Begin

  613.         "0:                                             \n\t"

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

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

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

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

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

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

  620.         "1:                                             \n\t"

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

  622.         "2:                                             \n\t"

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

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

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

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

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

  628. 

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

  630. 

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

  632.         // End

  633.         "9:                                             \n\t"

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

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

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

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

  638.         "dec                         %1                 \n\t"

  639.         "dec                         %2                 \n\t"

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

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

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

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

  644. 

  645. 

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

  647.           "=r" (fragmentLengthA)

  648.         );

  649. 

  650.     __asm__ volatile (

  651.         "jmp                         9f                 \n\t"

  652.         // Begin

  653.         "0:                                             \n\t"

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

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

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

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

  658.         "1:                                             \n\t"

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

  660.         "2:                                             \n\t"

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

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

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

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

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

  666. 

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

  668. 

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

  670.         // End

  671.         "9:                                             \n\t"

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

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

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

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

  676.         "dec                         %1                 \n\t"

  677.         "dec                         %2                 \n\t"

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

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

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

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

  682. 

  683. 

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

  685.           "=r" (fragmentLengthB)

  686.         );

  687. 

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

  689.     fragmentPos = 0;

  690. 

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

  692.         int xx = xpos >> 16;

  693. 

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

  695.             int a                  = 0;

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

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

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

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

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

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

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

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

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

  705.             int shift              = 0;

  706. 

  707.             if (filterCode) {

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

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

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

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

  712.                 filterPos[i / 2] = xx;

  713. 

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

  715. 

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

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

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

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

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

  721.                                                                (c << 4) |

  722.                                                                (d << 6);

  723. 

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

  725.                     shift = maxShift;               // avoid overread

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

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

  728. 

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

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

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

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

  733.                 }

  734.             }

  735. 

  736.             fragmentPos += fragmentLength;

  737. 

  738.             if (filterCode)

  739.                 filterCode[fragmentPos] = RET;

  740.         }

  741.         xpos += xInc;

  742.     }

  743.     if (filterCode)

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

  745. 

  746.     return fragmentPos + 1;

  747. }

  748. #endif /* HAVE_MMX2 */

  749. 

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

  751. {

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

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

  754. }

  755. 

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

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

  758.                              int brightness, int contrast, int saturation)

  759. {

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

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

  762. 

  763.     c->brightness = brightness;

  764.     c->contrast   = contrast;

  765.     c->saturation = saturation;

  766.     c->srcRange   = srcRange;

  767.     c->dstRange   = dstRange;

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

  769.         return -1;

  770. 

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

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

  773. 

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

  775.                              contrast, saturation);

  776.     // FIXME factorize

  777. 

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

  779.         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,

  780.                                        contrast, saturation);

  781.     return 0;

  782. }

  783. 

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

  785.                              int *srcRange, int **table, int *dstRange,

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

  787. {

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

  789.         return -1;

  790. 

  791.     *inv_table  = c->srcColorspaceTable;

  792.     *table      = c->dstColorspaceTable;

  793.     *srcRange   = c->srcRange;

  794.     *dstRange   = c->dstRange;

  795.     *brightness = c->brightness;

  796.     *contrast   = c->contrast;

  797.     *saturation = c->saturation;

  798. 

  799.     return 0;

  800. }

  801. 

  802. static int handle_jpeg(enum PixelFormat *format)

  803. {

  804.     switch (*format) {

  805.     case PIX_FMT_YUVJ420P:

  806.         *format = PIX_FMT_YUV420P;

  807.         return 1;

  808.     case PIX_FMT_YUVJ422P:

  809.         *format = PIX_FMT_YUV422P;

  810.         return 1;

  811.     case PIX_FMT_YUVJ444P:

  812.         *format = PIX_FMT_YUV444P;

  813.         return 1;

  814.     case PIX_FMT_YUVJ440P:

  815.         *format = PIX_FMT_YUV440P;

  816.         return 1;

  817.     default:

  818.         return 0;

  819.     }

  820. }

  821. 

  822. static int handle_0alpha(enum PixelFormat *format)

  823. {

  824.     switch (*format) {

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

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

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

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

  829.     default:                                          return 0;

  830.     }

  831. }

  832. 

  833. SwsContext *sws_alloc_context(void)

  834. {

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

  836. 

  837.     c->av_class = &sws_context_class;

  838.     av_opt_set_defaults(c);

  839. 

  840.     return c;

  841. }

  842. 

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

  844. {

  845.     int i, j;

  846.     int usesVFilter, usesHFilter;

  847.     int unscaled;

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

  849.     int srcW              = c->srcW;

  850.     int srcH              = c->srcH;

  851.     int dstW              = c->dstW;

  852.     int dstH              = c->dstH;

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

  854.     int flags, cpu_flags;

  855.     enum PixelFormat srcFormat = c->srcFormat;

  856.     enum PixelFormat dstFormat = c->dstFormat;

  857. 

  858.     cpu_flags = av_get_cpu_flags();

  859.     flags     = c->flags;

  860.     emms_c();

  861.     if (!rgb15to16)

  862.         sws_rgb2rgb_init();

  863. 

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

  865. 

  866.     handle_jpeg(&srcFormat);

  867.     handle_jpeg(&dstFormat);

  868.     handle_0alpha(&srcFormat);

  869.     handle_0alpha(&dstFormat);

  870. 

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

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

  873.         c->srcFormat= srcFormat;

  874.         c->dstFormat= dstFormat;

  875.     }

  876. 

  877.     if (!sws_isSupportedInput(srcFormat)) {

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

  879.                av_get_pix_fmt_name(srcFormat));

  880.         return AVERROR(EINVAL);

  881.     }

  882.     if (!sws_isSupportedOutput(dstFormat)) {

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

  884.                av_get_pix_fmt_name(dstFormat));

  885.         return AVERROR(EINVAL);

  886.     }

  887. 

  888.     i = flags & (SWS_POINT         |

  889.                  SWS_AREA          |

  890.                  SWS_BILINEAR      |

  891.                  SWS_FAST_BILINEAR |

  892.                  SWS_BICUBIC       |

  893.                  SWS_X             |

  894.                  SWS_GAUSS         |

  895.                  SWS_LANCZOS       |

  896.                  SWS_SINC          |

  897.                  SWS_SPLINE        |

  898.                  SWS_BICUBLIN);

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

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

  901.         return AVERROR(EINVAL);

  902.     }

  903.     /* sanity check */

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

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

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

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

  908.                srcW, srcH, dstW, dstH);

  909.         return AVERROR(EINVAL);

  910.     }

  911. 

  912.     if (!dstFilter)

  913.         dstFilter = &dummyFilter;

  914.     if (!srcFilter)

  915.         srcFilter = &dummyFilter;

  916. 

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

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

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

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

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

  922. 

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

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

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

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

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

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

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

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

  931. 

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

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

  934. 

  935. 

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

  937.         if (dstW&1) {

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

  939.             flags |= SWS_FULL_CHR_H_INT;

  940.             c->flags = flags;

  941.         }

  942.     }

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

  944.      * chroma interpolation */

  945.     if (flags & SWS_FULL_CHR_H_INT &&

  946.         isAnyRGB(dstFormat)        &&

  947.         dstFormat != PIX_FMT_RGBA  &&

  948.         dstFormat != PIX_FMT_ARGB  &&

  949.         dstFormat != PIX_FMT_BGRA  &&

  950.         dstFormat != PIX_FMT_ABGR  &&

  951.         dstFormat != PIX_FMT_RGB24 &&

  952.         dstFormat != PIX_FMT_BGR24) {

  953.         av_log(c, AV_LOG_WARNING,

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

  955.                av_get_pix_fmt_name(dstFormat));

  956.         flags   &= ~SWS_FULL_CHR_H_INT;

  957.         c->flags = flags;

  958.     }

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

  960.         c->chrDstHSubSample = 1;

  961. 

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

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

  964.                            SWS_SRC_V_CHR_DROP_SHIFT;

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

  966. 

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

  968.      * wants full chroma */

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

  970.         srcFormat != PIX_FMT_RGB8 && srcFormat != PIX_FMT_BGR8 &&

  971.         srcFormat != PIX_FMT_RGB4 && srcFormat != PIX_FMT_BGR4 &&

  972.         srcFormat != PIX_FMT_RGB4_BYTE && srcFormat != PIX_FMT_BGR4_BYTE &&

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

  974.          (flags & SWS_FAST_BILINEAR)))

  975.         c->chrSrcHSubSample = 1;

  976. 

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

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

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

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

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

  982. 

  983.     /* unscaled special cases */

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

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

  986.         ff_get_unscaled_swscale(c);

  987. 

  988.         if (c->swScale) {

  989.             if (flags & SWS_PRINT_INFO)

  990.                 av_log(c, AV_LOG_INFO,

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

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

  993.             return 0;

  994.         }

  995.     }

  996. 

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

  998.     if (c->srcBpc < 8)

  999.         c->srcBpc = 8;

  1000.     c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;

  1001.     if (c->dstBpc < 8)

  1002.         c->dstBpc = 8;

  1003.     if (isAnyRGB(srcFormat) || srcFormat == PIX_FMT_PAL8)

  1004.         c->srcBpc = 16;

  1005.     if (c->dstBpc == 16)

  1006.         dst_stride <<= 1;

  1007.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);

  1008.     if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 &&

  1009.         c->srcBpc == 8 && c->dstBpc <= 10) {

  1010.         c->canMMX2BeUsed = (dstW >= srcW && (dstW & 31) == 0 &&

  1011.                             (srcW & 15) == 0) ? 1 : 0;

  1012.         if (!c->canMMX2BeUsed && dstW >= srcW && (srcW & 15) == 0

  1013.             && (flags & SWS_FAST_BILINEAR)) {

  1014.             if (flags & SWS_PRINT_INFO)

  1015.                 av_log(c, AV_LOG_INFO,

  1016.                        "output width is not a multiple of 32 -> no MMX2 scaler\n");

  1017.         }

  1018.         if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))

  1019.             c->canMMX2BeUsed=0;

  1020.     } else

  1021.         c->canMMX2BeUsed = 0;

  1022. 

  1023.     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;

  1024.     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

  1025. 

  1026.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  1027.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  1028.      * correct scaling.

  1029.      * n-2 is the last chrominance sample available.

  1030.      * This is not perfect, but no one should notice the difference, the more

  1031.      * correct variant would be like the vertical one, but that would require

  1032.      * some special code for the first and last pixel */

  1033.     if (flags & SWS_FAST_BILINEAR) {

  1034.         if (c->canMMX2BeUsed) {

  1035.             c->lumXInc += 20;

  1036.             c->chrXInc += 20;

  1037.         }

  1038.         // we don't use the x86 asm scaler if MMX is available

  1039.         else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX && c->dstBpc <= 10) {

  1040.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1041.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1042.         }

  1043.     }

  1044. 

  1045.     /* precalculate horizontal scaler filter coefficients */

  1046.     {

  1047. #if HAVE_MMX2

  1048. // can't downscale !!!

  1049.         if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {

  1050.             c->lumMmx2FilterCodeSize = initMMX2HScaler(dstW, c->lumXInc, NULL,

  1051.                                                        NULL, NULL, 8);

  1052.             c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc,

  1053.                                                        NULL, NULL, NULL, 4);

  1054. 

  1055. #ifdef MAP_ANONYMOUS

  1056.             c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);

  1057.             c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);

  1058. #elif HAVE_VIRTUALALLOC

  1059.             c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);

  1060.             c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);

  1061. #else

  1062.             c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);

  1063.             c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);

  1064. #endif

  1065. 

  1066. #ifdef MAP_ANONYMOUS

  1067.             if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)

  1068. #else

  1069.             if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)

  1070. #endif

  1071.             {

  1072.                 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");

  1073.                 return AVERROR(ENOMEM);

  1074.             }

  1075. 

  1076.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1077.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1078.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1079.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1080. 

  1081.             initMMX2HScaler(      dstW, c->lumXInc, c->lumMmx2FilterCode,

  1082.                             c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);

  1083.             initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode,

  1084.                             c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);

  1085. 

  1086. #ifdef MAP_ANONYMOUS

  1087.             mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);

  1088.             mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);

  1089. #endif

  1090.         } else

  1091. #endif /* HAVE_MMX2 */

  1092.         {

  1093.             const int filterAlign =

  1094.                 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :

  1095.                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1096.                 1;

  1097. 

  1098.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1099.                            &c->hLumFilterSize, c->lumXInc,

  1100.                            srcW, dstW, filterAlign, 1 << 14,

  1101.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1102.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1103.                            c->param) < 0)

  1104.                 goto fail;

  1105.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1106.                            &c->hChrFilterSize, c->chrXInc,

  1107.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1108.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1109.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1110.                            c->param) < 0)

  1111.                 goto fail;

  1112.         }

  1113.     } // initialize horizontal stuff

  1114. 

  1115.     /* precalculate vertical scaler filter coefficients */

  1116.     {

  1117.         const int filterAlign =

  1118.             (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :

  1119.             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1120.             1;

  1121. 

  1122.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1123.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

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

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

  1126.                        c->param) < 0)

  1127.             goto fail;

  1128.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1129.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1130.                        filterAlign, (1 << 12),

  1131.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1132.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1133.                        c->param) < 0)

  1134.             goto fail;

  1135. 

  1136. #if HAVE_ALTIVEC

  1137.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1138.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1139. 

  1140.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1141.             int j;

  1142.             short *p = (short *)&c->vYCoeffsBank[i];

  1143.             for (j = 0; j < 8; j++)

  1144.                 p[j] = c->vLumFilter[i];

  1145.         }

  1146. 

  1147.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1148.             int j;

  1149.             short *p = (short *)&c->vCCoeffsBank[i];

  1150.             for (j = 0; j < 8; j++)

  1151.                 p[j] = c->vChrFilter[i];

  1152.         }

  1153. #endif

  1154.     }

  1155. 

  1156.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1157.     c->vLumBufSize = c->vLumFilterSize;

  1158.     c->vChrBufSize = c->vChrFilterSize;

  1159.     for (i = 0; i < dstH; i++) {

  1160.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1161.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1162.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1163.                                << c->chrSrcVSubSample));

  1164. 

  1165.         nextSlice >>= c->chrSrcVSubSample;

  1166.         nextSlice <<= c->chrSrcVSubSample;

  1167.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1168.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1169.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1170.             (nextSlice >> c->chrSrcVSubSample))

  1171.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1172.                              c->vChrFilterPos[chrI];

  1173.     }

  1174. 

  1175.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1176.      * need to allocate several megabytes to handle all possible cases) */

  1177.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1178.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1179.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1180.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1181.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1182.     /* Note we need at least one pixel more at the end because of the MMX code

  1183.      * (just in case someone wants to replace the 4000/8000). */

  1184.     /* align at 16 bytes for AltiVec */

  1185.     for (i = 0; i < c->vLumBufSize; i++) {

  1186.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1187.                           dst_stride + 16, fail);

  1188.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1189.     }

  1190.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1191.     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);

  1192.     c->uv_offx2 = dst_stride + 16;

  1193.     for (i = 0; i < c->vChrBufSize; i++) {

  1194.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1195.                          dst_stride * 2 + 32, fail);

  1196.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1197.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1198.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1199.     }

  1200.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

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

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

  1203.                               dst_stride + 16, fail);

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

  1205.         }

  1206. 

  1207.     // try to avoid drawing green stuff between the right end and the stride end

  1208.     for (i = 0; i < c->vChrBufSize; i++)

  1209.         if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){

  1210.             av_assert0(c->dstBpc > 10);

  1211.             for(j=0; j<dst_stride/2+1; j++)

  1212.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1213.         } else

  1214.             for(j=0; j<dst_stride+1; j++)

  1215.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1216. 

  1217.     av_assert0(c->chrDstH <= dstH);

  1218. 

  1219.     if (flags & SWS_PRINT_INFO) {

  1220.         if (flags & SWS_FAST_BILINEAR)

  1221.             av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1222.         else if (flags & SWS_BILINEAR)

  1223.             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1224.         else if (flags & SWS_BICUBIC)

  1225.             av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1226.         else if (flags & SWS_X)

  1227.             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1228.         else if (flags & SWS_POINT)

  1229.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1230.         else if (flags & SWS_AREA)

  1231.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1232.         else if (flags & SWS_BICUBLIN)

  1233.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1234.         else if (flags & SWS_GAUSS)

  1235.             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1236.         else if (flags & SWS_SINC)

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

  1238.         else if (flags & SWS_LANCZOS)

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

  1240.         else if (flags & SWS_SPLINE)

  1241.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1242.         else

  1243.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1244. 

  1245.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1246.                av_get_pix_fmt_name(srcFormat),

  1247. #ifdef DITHER1XBPP

  1248.                dstFormat == PIX_FMT_BGR555   || dstFormat == PIX_FMT_BGR565   ||

  1249.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1250.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ?

  1251.                                                              "dithered " : "",

  1252. #else

  1253.                "",

  1254. #endif

  1255.                av_get_pix_fmt_name(dstFormat));

  1256. 

  1257.         if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2)

  1258.             av_log(c, AV_LOG_INFO, "using MMX2\n");

  1259.         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)

  1260.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1261.         else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX)

  1262.             av_log(c, AV_LOG_INFO, "using MMX\n");

  1263.         else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)

  1264.             av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1265.         else

  1266.             av_log(c, AV_LOG_INFO, "using C\n");

  1267. 

  1268.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1269.         av_log(c, AV_LOG_DEBUG,

  1270.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1271.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1272.         av_log(c, AV_LOG_DEBUG,

  1273.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1274.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1275.                c->chrXInc, c->chrYInc);

  1276.     }

  1277. 

  1278.     c->swScale = ff_getSwsFunc(c);

  1279.     return 0;

  1280. fail: // FIXME replace things by appropriate error codes

  1281.     return -1;

  1282. }

  1283. 

  1284. #if FF_API_SWS_GETCONTEXT

  1285. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1286.                            int dstW, int dstH, enum PixelFormat dstFormat,

  1287.                            int flags, SwsFilter *srcFilter,

  1288.                            SwsFilter *dstFilter, const double *param)

  1289. {

  1290.     SwsContext *c;

  1291. 

  1292.     if (!(c = sws_alloc_context()))

  1293.         return NULL;

  1294. 

  1295.     c->flags     = flags;

  1296.     c->srcW      = srcW;

  1297.     c->srcH      = srcH;

  1298.     c->dstW      = dstW;

  1299.     c->dstH      = dstH;

  1300.     c->srcRange  = handle_jpeg(&srcFormat);

  1301.     c->dstRange  = handle_jpeg(&dstFormat);

  1302.     c->src0Alpha = handle_0alpha(&srcFormat);

  1303.     c->dst0Alpha = handle_0alpha(&dstFormat);

  1304.     c->srcFormat = srcFormat;

  1305.     c->dstFormat = dstFormat;

  1306. 

  1307.     if (param) {

  1308.         c->param[0] = param[0];

  1309.         c->param[1] = param[1];

  1310.     }

  1311.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1312.                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1313.                              c->dstRange, 0, 1 << 16, 1 << 16);

  1314. 

  1315.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1316.         sws_freeContext(c);

  1317.         return NULL;

  1318.     }

  1319. 

  1320.     return c;

  1321. }

  1322. #endif

  1323. 

  1324. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1325.                                 float lumaSharpen, float chromaSharpen,

  1326.                                 float chromaHShift, float chromaVShift,

  1327.                                 int verbose)

  1328. {

  1329.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1330.     if (!filter)

  1331.         return NULL;

  1332. 

  1333.     if (lumaGBlur != 0.0) {

  1334.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1335.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1336.     } else {

  1337.         filter->lumH = sws_getIdentityVec();

  1338.         filter->lumV = sws_getIdentityVec();

  1339.     }

  1340. 

  1341.     if (chromaGBlur != 0.0) {

  1342.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1343.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1344.     } else {

  1345.         filter->chrH = sws_getIdentityVec();

  1346.         filter->chrV = sws_getIdentityVec();

  1347.     }

  1348. 

  1349.     if (chromaSharpen != 0.0) {

  1350.         SwsVector *id = sws_getIdentityVec();

  1351.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1352.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1353.         sws_addVec(filter->chrH, id);

  1354.         sws_addVec(filter->chrV, id);

  1355.         sws_freeVec(id);

  1356.     }

  1357. 

  1358.     if (lumaSharpen != 0.0) {

  1359.         SwsVector *id = sws_getIdentityVec();

  1360.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1361.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1362.         sws_addVec(filter->lumH, id);

  1363.         sws_addVec(filter->lumV, id);

  1364.         sws_freeVec(id);

  1365.     }

  1366. 

  1367.     if (chromaHShift != 0.0)

  1368.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1369. 

  1370.     if (chromaVShift != 0.0)

  1371.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1372. 

  1373.     sws_normalizeVec(filter->chrH, 1.0);

  1374.     sws_normalizeVec(filter->chrV, 1.0);

  1375.     sws_normalizeVec(filter->lumH, 1.0);

  1376.     sws_normalizeVec(filter->lumV, 1.0);

  1377. 

  1378.     if (verbose)

  1379.         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1380.     if (verbose)

  1381.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1382. 

  1383.     return filter;

  1384. }

  1385. 

  1386. SwsVector *sws_allocVec(int length)

  1387. {

  1388.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1389.     if (!vec)

  1390.         return NULL;

  1391.     vec->length = length;

  1392.     vec->coeff  = av_malloc(sizeof(double) * length);

  1393.     if (!vec->coeff)

  1394.         av_freep(&vec);

  1395.     return vec;

  1396. }

  1397. 

  1398. SwsVector *sws_getGaussianVec(double variance, double quality)

  1399. {

  1400.     const int length = (int)(variance * quality + 0.5) | 1;

  1401.     int i;

  1402.     double middle  = (length - 1) * 0.5;

  1403.     SwsVector *vec = sws_allocVec(length);

  1404. 

  1405.     if (!vec)

  1406.         return NULL;

  1407. 

  1408.     for (i = 0; i < length; i++) {

  1409.         double dist = i - middle;

  1410.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1411.                         sqrt(2 * variance * M_PI);

  1412.     }

  1413. 

  1414.     sws_normalizeVec(vec, 1.0);

  1415. 

  1416.     return vec;

  1417. }

  1418. 

  1419. SwsVector *sws_getConstVec(double c, int length)

  1420. {

  1421.     int i;

  1422.     SwsVector *vec = sws_allocVec(length);

  1423. 

  1424.     if (!vec)

  1425.         return NULL;

  1426. 

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

  1428.         vec->coeff[i] = c;

  1429. 

  1430.     return vec;

  1431. }

  1432. 

  1433. SwsVector *sws_getIdentityVec(void)

  1434. {

  1435.     return sws_getConstVec(1.0, 1);

  1436. }

  1437. 

  1438. static double sws_dcVec(SwsVector *a)

  1439. {

  1440.     int i;

  1441.     double sum = 0;

  1442. 

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

  1444.         sum += a->coeff[i];

  1445. 

  1446.     return sum;

  1447. }

  1448. 

  1449. void sws_scaleVec(SwsVector *a, double scalar)

  1450. {

  1451.     int i;

  1452. 

  1453.     for (i = 0; i < a->length; i++)

  1454.         a->coeff[i] *= scalar;

  1455. }

  1456. 

  1457. void sws_normalizeVec(SwsVector *a, double height)

  1458. {

  1459.     sws_scaleVec(a, height / sws_dcVec(a));

  1460. }

  1461. 

  1462. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1463. {

  1464.     int length = a->length + b->length - 1;

  1465.     int i, j;

  1466.     SwsVector *vec = sws_getConstVec(0.0, length);

  1467. 

  1468.     if (!vec)

  1469.         return NULL;

  1470. 

  1471.     for (i = 0; i < a->length; i++) {

  1472.         for (j = 0; j < b->length; j++) {

  1473.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1474.         }

  1475.     }

  1476. 

  1477.     return vec;

  1478. }

  1479. 

  1480. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1481. {

  1482.     int length = FFMAX(a->length, b->length);

  1483.     int i;

  1484.     SwsVector *vec = sws_getConstVec(0.0, length);

  1485. 

  1486.     if (!vec)

  1487.         return NULL;

  1488. 

  1489.     for (i = 0; i < a->length; i++)

  1490.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

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

  1492.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

  1493. 

  1494.     return vec;

  1495. }

  1496. 

  1497. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1498. {

  1499.     int length = FFMAX(a->length, b->length);

  1500.     int i;

  1501.     SwsVector *vec = sws_getConstVec(0.0, length);

  1502. 

  1503.     if (!vec)

  1504.         return NULL;

  1505. 

  1506.     for (i = 0; i < a->length; i++)

  1507.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

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

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

  1510. 

  1511.     return vec;

  1512. }

  1513. 

  1514. /* shift left / or right if "shift" is negative */

  1515. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1516. {

  1517.     int length = a->length + FFABS(shift) * 2;

  1518.     int i;

  1519.     SwsVector *vec = sws_getConstVec(0.0, length);

  1520. 

  1521.     if (!vec)

  1522.         return NULL;

  1523. 

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

  1525.         vec->coeff[i + (length    - 1) / 2 -

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

  1527.     }

  1528. 

  1529.     return vec;

  1530. }

  1531. 

  1532. void sws_shiftVec(SwsVector *a, int shift)

  1533. {

  1534.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1535.     av_free(a->coeff);

  1536.     a->coeff  = shifted->coeff;

  1537.     a->length = shifted->length;

  1538.     av_free(shifted);

  1539. }

  1540. 

  1541. void sws_addVec(SwsVector *a, SwsVector *b)

  1542. {

  1543.     SwsVector *sum = sws_sumVec(a, b);

  1544.     av_free(a->coeff);

  1545.     a->coeff  = sum->coeff;

  1546.     a->length = sum->length;

  1547.     av_free(sum);

  1548. }

  1549. 

  1550. void sws_subVec(SwsVector *a, SwsVector *b)

  1551. {

  1552.     SwsVector *diff = sws_diffVec(a, b);

  1553.     av_free(a->coeff);

  1554.     a->coeff  = diff->coeff;

  1555.     a->length = diff->length;

  1556.     av_free(diff);

  1557. }

  1558. 

  1559. void sws_convVec(SwsVector *a, SwsVector *b)

  1560. {

  1561.     SwsVector *conv = sws_getConvVec(a, b);

  1562.     av_free(a->coeff);

  1563.     a->coeff  = conv->coeff;

  1564.     a->length = conv->length;

  1565.     av_free(conv);

  1566. }

  1567. 

  1568. SwsVector *sws_cloneVec(SwsVector *a)

  1569. {

  1570.     int i;

  1571.     SwsVector *vec = sws_allocVec(a->length);

  1572. 

  1573.     if (!vec)

  1574.         return NULL;

  1575. 

  1576.     for (i = 0; i < a->length; i++)

  1577.         vec->coeff[i] = a->coeff[i];

  1578. 

  1579.     return vec;

  1580. }

  1581. 

  1582. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1583. {

  1584.     int i;

  1585.     double max = 0;

  1586.     double min = 0;

  1587.     double range;

  1588. 

  1589.     for (i = 0; i < a->length; i++)

  1590.         if (a->coeff[i] > max)

  1591.             max = a->coeff[i];

  1592. 

  1593.     for (i = 0; i < a->length; i++)

  1594.         if (a->coeff[i] < min)

  1595.             min = a->coeff[i];

  1596. 

  1597.     range = max - min;

  1598. 

  1599.     for (i = 0; i < a->length; i++) {

  1600.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1601.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1602.         for (; x > 0; x--)

  1603.             av_log(log_ctx, log_level, " ");

  1604.         av_log(log_ctx, log_level, "|\n");

  1605.     }

  1606. }

  1607. 

  1608. void sws_freeVec(SwsVector *a)

  1609. {

  1610.     if (!a)

  1611.         return;

  1612.     av_freep(&a->coeff);

  1613.     a->length = 0;

  1614.     av_free(a);

  1615. }

  1616. 

  1617. void sws_freeFilter(SwsFilter *filter)

  1618. {

  1619.     if (!filter)

  1620.         return;

  1621. 

  1622.     if (filter->lumH)

  1623.         sws_freeVec(filter->lumH);

  1624.     if (filter->lumV)

  1625.         sws_freeVec(filter->lumV);

  1626.     if (filter->chrH)

  1627.         sws_freeVec(filter->chrH);

  1628.     if (filter->chrV)

  1629.         sws_freeVec(filter->chrV);

  1630.     av_free(filter);

  1631. }

  1632. 

  1633. void sws_freeContext(SwsContext *c)

  1634. {

  1635.     int i;

  1636.     if (!c)

  1637.         return;

  1638. 

  1639.     if (c->lumPixBuf) {

  1640.         for (i = 0; i < c->vLumBufSize; i++)

  1641.             av_freep(&c->lumPixBuf[i]);

  1642.         av_freep(&c->lumPixBuf);

  1643.     }

  1644. 

  1645.     if (c->chrUPixBuf) {

  1646.         for (i = 0; i < c->vChrBufSize; i++)

  1647.             av_freep(&c->chrUPixBuf[i]);

  1648.         av_freep(&c->chrUPixBuf);

  1649.         av_freep(&c->chrVPixBuf);

  1650.     }

  1651. 

  1652.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1653.         for (i = 0; i < c->vLumBufSize; i++)

  1654.             av_freep(&c->alpPixBuf[i]);

  1655.         av_freep(&c->alpPixBuf);

  1656.     }

  1657. 

  1658.     av_freep(&c->vLumFilter);

  1659.     av_freep(&c->vChrFilter);

  1660.     av_freep(&c->hLumFilter);

  1661.     av_freep(&c->hChrFilter);

  1662. #if HAVE_ALTIVEC

  1663.     av_freep(&c->vYCoeffsBank);

  1664.     av_freep(&c->vCCoeffsBank);

  1665. #endif

  1666. 

  1667.     av_freep(&c->vLumFilterPos);

  1668.     av_freep(&c->vChrFilterPos);

  1669.     av_freep(&c->hLumFilterPos);

  1670.     av_freep(&c->hChrFilterPos);

  1671. 

  1672. #if HAVE_MMX

  1673. #ifdef MAP_ANONYMOUS

  1674.     if (c->lumMmx2FilterCode)

  1675.         munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1676.     if (c->chrMmx2FilterCode)

  1677.         munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1678. #elif HAVE_VIRTUALALLOC

  1679.     if (c->lumMmx2FilterCode)

  1680.         VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1681.     if (c->chrMmx2FilterCode)

  1682.         VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1683. #else

  1684.     av_free(c->lumMmx2FilterCode);

  1685.     av_free(c->chrMmx2FilterCode);

  1686. #endif

  1687.     c->lumMmx2FilterCode = NULL;

  1688.     c->chrMmx2FilterCode = NULL;

  1689. #endif /* HAVE_MMX */

  1690. 

  1691.     av_freep(&c->yuvTable);

  1692.     av_freep(&c->formatConvBuffer);

  1693. 

  1694.     av_free(c);

  1695. }

  1696. 

  1697. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  1698.                                         int srcH, enum PixelFormat srcFormat,

  1699.                                         int dstW, int dstH,

  1700.                                         enum PixelFormat dstFormat, int flags,

  1701.                                         SwsFilter *srcFilter,

  1702.                                         SwsFilter *dstFilter,

  1703.                                         const double *param)

  1704. {

  1705.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  1706.                                              SWS_PARAM_DEFAULT };

  1707. 

  1708.     if (!param)

  1709.         param = default_param;

  1710. 

  1711.     if (context &&

  1712.         (context->srcW      != srcW      ||

  1713.          context->srcH      != srcH      ||

  1714.          context->srcFormat != srcFormat ||

  1715.          context->dstW      != dstW      ||

  1716.          context->dstH      != dstH      ||

  1717.          context->dstFormat != dstFormat ||

  1718.          context->flags     != flags     ||

  1719.          context->param[0]  != param[0]  ||

  1720.          context->param[1]  != param[1])) {

  1721.         sws_freeContext(context);

  1722.         context = NULL;

  1723.     }

  1724. 

  1725.     if (!context) {

  1726.         if (!(context = sws_alloc_context()))

  1727.             return NULL;

  1728.         context->srcW      = srcW;

  1729.         context->srcH      = srcH;

  1730.         context->srcRange  = handle_jpeg(&srcFormat);

  1731.         context->src0Alpha = handle_0alpha(&srcFormat);

  1732.         context->srcFormat = srcFormat;

  1733.         context->dstW      = dstW;

  1734.         context->dstH      = dstH;

  1735.         context->dstRange  = handle_jpeg(&dstFormat);

  1736.         context->dst0Alpha = handle_0alpha(&dstFormat);

  1737.         context->dstFormat = dstFormat;

  1738.         context->flags     = flags;

  1739.         context->param[0]  = param[0];

  1740.         context->param[1]  = param[1];

  1741.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  1742.                                  context->srcRange,

  1743.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1744.                                  context->dstRange, 0, 1 << 16, 1 << 16);

  1745.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1746.             sws_freeContext(context);

  1747.             return NULL;

  1748.         }

  1749.     }

  1750.     return context;

  1751. }