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.
37838 Commits
32 Branches
896MB
Tree: c7e921a54f
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c7e921a54f'
${ noResults }
FFmpeg/libswscale/utils.c

1773 lines
63KB
Raw Blame History 

  1. /*

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

  3.  *

  4.  * This file is part of Libav.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. #include "config.h"

  22. 

  23. #define _SVID_SOURCE // needed for MAP_ANONYMOUS

  24. #include <assert.h>

  25. #include <inttypes.h>

  26. #include <math.h>

  27. #include <stdio.h>

  28. #include <string.h>

  29. #if HAVE_SYS_MMAN_H

  30. #include <sys/mman.h>

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

  32. #define MAP_ANONYMOUS MAP_ANON

  33. #endif

  34. #endif

  35. #if HAVE_VIRTUALALLOC

  36. #define WIN32_LEAN_AND_MEAN

  37. #include <windows.h>

  38. #endif

  39. 

  40. #include "libavutil/attributes.h"

  41. #include "libavutil/avutil.h"

  42. #include "libavutil/bswap.h"

  43. #include "libavutil/cpu.h"

  44. #include "libavutil/intreadwrite.h"

  45. #include "libavutil/mathematics.h"

  46. #include "libavutil/opt.h"

  47. #include "libavutil/pixdesc.h"

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

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

  50. #include "rgb2rgb.h"

  51. #include "swscale.h"

  52. #include "swscale_internal.h"

  53. 

  54. unsigned swscale_version(void)

  55. {

  56.     return LIBSWSCALE_VERSION_INT;

  57. }

  58. 

  59. const char *swscale_configuration(void)

  60. {

  61.     return LIBAV_CONFIGURATION;

  62. }

  63. 

  64. const char *swscale_license(void)

  65. {

  66. #define LICENSE_PREFIX "libswscale license: "

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

  68. }

  69. 

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

  71. 

  72. typedef struct FormatEntry {

  73.     uint8_t is_supported_in         :1;

  74.     uint8_t is_supported_out        :1;

  75.     uint8_t is_supported_endianness :1;

  76. } FormatEntry;

  77. 

  78. static const FormatEntry format_entries[AV_PIX_FMT_NB] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  177. };

  178. 

  179. int sws_isSupportedInput(enum AVPixelFormat pix_fmt)

  180. {

  181.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  182.            format_entries[pix_fmt].is_supported_in : 0;

  183. }

  184. 

  185. int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)

  186. {

  187.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  188.            format_entries[pix_fmt].is_supported_out : 0;

  189. }

  190. 

  191. int sws_isSupportedEndiannessConversion(enum AVPixelFormat pix_fmt)

  192. {

  193.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  194.            format_entries[pix_fmt].is_supported_endianness : 0;

  195. }

  196. 

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

  198. 

  199. const char *sws_format_name(enum AVPixelFormat format)

  200. {

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

  202.     if (desc)

  203.         return desc->name;

  204.     else

  205.         return "Unknown format";

  206. }

  207. 

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

  209.                              double dist)

  210. {

  211.     if (dist <= 1.0)

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

  213.     else

  214.         return getSplineCoeff(0.0,

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

  216.                                c + 3.0 * d,

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

  218.                               dist - 1.0);

  219. }

  220. 

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

  222.                               int *outFilterSize, int xInc, int srcW,

  223.                               int dstW, int filterAlign, int one,

  224.                               int flags, int cpu_flags,

  225.                               SwsVector *srcFilter, SwsVector *dstFilter,

  226.                               double param[2], int is_horizontal)

  227. {

  228.     int i;

  229.     int filterSize;

  230.     int filter2Size;

  231.     int minFilterSize;

  232.     int64_t *filter    = NULL;

  233.     int64_t *filter2   = NULL;

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

  235.     int ret            = -1;

  236. 

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

  238. 

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

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

  241. 

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

  243.         int i;

  244.         filterSize = 1;

  245.         FF_ALLOCZ_OR_GOTO(NULL, filter,

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

  247. 

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

  249.             filter[i * filterSize] = fone;

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

  251.         }

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

  253.         int i;

  254.         int xDstInSrc;

  255.         filterSize = 1;

  256.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  258. 

  259.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  262. 

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

  264.             filter[i]       = fone;

  265.             xDstInSrc      += xInc;

  266.         }

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

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

  269.         int i;

  270.         int xDstInSrc;

  271.         filterSize = 2;

  272.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  274. 

  275.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  278.             int j;

  279. 

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

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

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

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

  284.                                 (fone >> 16);

  285.                 if (coeff < 0)

  286.                     coeff = 0;

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

  288.                 xx++;

  289.             }

  290.             xDstInSrc += xInc;

  291.         }

  292.     } else {

  293.         int64_t xDstInSrc;

  294.         int sizeFactor;

  295. 

  296.         if (flags & SWS_BICUBIC)

  297.             sizeFactor = 4;

  298.         else if (flags & SWS_X)

  299.             sizeFactor = 8;

  300.         else if (flags & SWS_AREA)

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

  302.         else if (flags & SWS_GAUSS)

  303.             sizeFactor = 8;     // infinite ;)

  304.         else if (flags & SWS_LANCZOS)

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

  306.         else if (flags & SWS_SINC)

  307.             sizeFactor = 20;    // infinite ;)

  308.         else if (flags & SWS_SPLINE)

  309.             sizeFactor = 20;    // infinite ;)

  310.         else if (flags & SWS_BILINEAR)

  311.             sizeFactor = 2;

  312.         else {

  313.             sizeFactor = 0;     // GCC warning killer

  314.             assert(0);

  315.         }

  316. 

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

  318.             filterSize = 1 + sizeFactor;    // upscale

  319.         else

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

  321. 

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

  323.         filterSize = FFMAX(filterSize, 1);

  324. 

  325.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  327. 

  328.         xDstInSrc = xInc - 0x10000;

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

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

  331.             int j;

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

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

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

  335.                 double floatd;

  336.                 int64_t coeff;

  337. 

  338.                 if (xInc > 1 << 16)

  339.                     d = d * dstW / srcW;

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

  341. 

  342.                 if (flags & SWS_BICUBIC) {

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

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

  345. 

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

  347.                         coeff = 0.0;

  348.                     } else {

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

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

  351. 

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

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

  354.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  355.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  356.                         else

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

  358.                                       (6 * B + 30 * C) * dd  +

  359.                                     (-12 * B - 48 * C) * d   +

  360.                                       (8 * B + 24 * C) * (1 << 30);

  361.                     }

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

  363.                 }

  364. #if 0

  365.                 else if (flags & SWS_X) {

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

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

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

  369.                 }

  370. #endif

  371.                 else if (flags & SWS_X) {

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

  373.                     double c;

  374. 

  375.                     if (floatd < 1.0)

  376.                         c = cos(floatd * M_PI);

  377.                     else

  378.                         c = -1.0;

  379.                     if (c < 0.0)

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

  381.                     else

  382.                         c = pow(c, A);

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

  384.                 } else if (flags & SWS_AREA) {

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

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

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

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

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

  390.                     else

  391.                         coeff = 0.0;

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

  393.                 } else if (flags & SWS_GAUSS) {

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

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

  396.                 } else if (flags & SWS_SINC) {

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

  398.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  402.                     if (floatd > p)

  403.                         coeff = 0;

  404.                 } else if (flags & SWS_BILINEAR) {

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

  406.                     if (coeff < 0)

  407.                         coeff = 0;

  408.                     coeff *= fone >> 30;

  409.                 } else if (flags & SWS_SPLINE) {

  410.                     double p = -2.196152422706632;

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

  412.                 } else {

  413.                     coeff = 0.0; // GCC warning killer

  414.                     assert(0);

  415.                 }

  416. 

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

  418.                 xx++;

  419.             }

  420.             xDstInSrc += 2 * xInc;

  421.         }

  422.     }

  423. 

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

  425.      * av_free(filter);

  426.      */

  427.     assert(filterSize > 0);

  428.     filter2Size = filterSize;

  429.     if (srcFilter)

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

  431.     if (dstFilter)

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

  433.     assert(filter2Size > 0);

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

  435. 

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

  437.         int j, k;

  438. 

  439.         if (srcFilter) {

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

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

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

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

  444.             }

  445.         } else {

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

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

  448.         }

  449.         // FIXME dstFilter

  450. 

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

  452.     }

  453.     av_freep(&filter);

  454. 

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

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

  457.     minFilterSize = 0;

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

  459.         int min = filter2Size;

  460.         int j;

  461.         int64_t cutOff = 0.0;

  462. 

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

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

  465.             int k;

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

  467. 

  468.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  469.                 break;

  470. 

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

  472.              * filter otherwise */

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

  474.                 break;

  475. 

  476.             // move filter coefficients left

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

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

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

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

  481.         }

  482. 

  483.         cutOff = 0;

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

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

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

  487. 

  488.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  489.                 break;

  490.             min--;

  491.         }

  492. 

  493.         if (min > minFilterSize)

  494.             minFilterSize = min;

  495.     }

  496. 

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

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

  499.         if (minFilterSize < 5)

  500.             filterAlign = 4;

  501. 

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

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

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

  505.         if (minFilterSize < 3)

  506.             filterAlign = 1;

  507.     }

  508. 

  509.     if (INLINE_MMX(cpu_flags)) {

  510.         // special case for unscaled vertical filtering

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

  512.             filterAlign = 1;

  513.     }

  514. 

  515.     assert(minFilterSize > 0);

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

  517.     assert(filterSize > 0);

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

  519.     if (filterSize >= MAX_FILTER_SIZE * 16 /

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

  521.         goto fail;

  522.     *outFilterSize = filterSize;

  523. 

  524.     if (flags & SWS_PRINT_INFO)

  525.         av_log(NULL, AV_LOG_VERBOSE,

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

  527.                filter2Size, filterSize);

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

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

  530.         int j;

  531. 

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

  533.             if (j >= filter2Size)

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

  535.             else

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

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

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

  539.         }

  540.     }

  541. 

  542.     // FIXME try to align filterPos if possible

  543. 

  544.     // fix borders

  545.     if (is_horizontal) {

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

  547.             int j;

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

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

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

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

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

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

  554.                 }

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

  556.             }

  557. 

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

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

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

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

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

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

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

  565.                 }

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

  567.             }

  568.         }

  569.     }

  570. 

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

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

  573.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

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

  575. 

  576.     /* normalize & store in outFilter */

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

  578.         int j;

  579.         int64_t error = 0;

  580.         int64_t sum   = 0;

  581. 

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

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

  584.         }

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

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

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

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

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

  590.             error                                  = v - intV * sum;

  591.         }

  592.     }

  593. 

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

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

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

  597.                                                       * read over the end */

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

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

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

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

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

  603.     }

  604. 

  605.     ret = 0;

  606. 

  607. fail:

  608.     av_free(filter);

  609.     av_free(filter2);

  610.     return ret;

  611. }

  612. 

  613. #if HAVE_MMXEXT_INLINE

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

  615.                                        int16_t *filter, int32_t *filterPos,

  616.                                        int numSplits)

  617. {

  618.     uint8_t *fragmentA;

  619.     x86_reg imm8OfPShufW1A;

  620.     x86_reg imm8OfPShufW2A;

  621.     x86_reg fragmentLengthA;

  622.     uint8_t *fragmentB;

  623.     x86_reg imm8OfPShufW1B;

  624.     x86_reg imm8OfPShufW2B;

  625.     x86_reg fragmentLengthB;

  626.     int fragmentPos;

  627. 

  628.     int xpos, i;

  629. 

  630.     // create an optimized horizontal scaling routine

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

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

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

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

  635.      */

  636. 

  637.     // code fragment

  638. 

  639.     __asm__ volatile (

  640.         "jmp                         9f                 \n\t"

  641.         // Begin

  642.         "0:                                             \n\t"

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

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

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

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

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

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

  649.         "1:                                             \n\t"

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

  651.         "2:                                             \n\t"

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

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

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

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

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

  657. 

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

  659. 

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

  661.         // End

  662.         "9:                                             \n\t"

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

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

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

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

  667.         "dec                         %1                 \n\t"

  668.         "dec                         %2                 \n\t"

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

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

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

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

  673. 

  674. 

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

  676.           "=r" (fragmentLengthA)

  677.         );

  678. 

  679.     __asm__ volatile (

  680.         "jmp                         9f                 \n\t"

  681.         // Begin

  682.         "0:                                             \n\t"

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

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

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

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

  687.         "1:                                             \n\t"

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

  689.         "2:                                             \n\t"

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

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

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

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

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

  695. 

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

  697. 

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

  699.         // End

  700.         "9:                                             \n\t"

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

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

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

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

  705.         "dec                         %1                 \n\t"

  706.         "dec                         %2                 \n\t"

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

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

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

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

  711. 

  712. 

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

  714.           "=r" (fragmentLengthB)

  715.         );

  716. 

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

  718.     fragmentPos = 0;

  719. 

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

  721.         int xx = xpos >> 16;

  722. 

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

  724.             int a                  = 0;

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

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

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

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

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

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

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

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

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

  734.             int shift              = 0;

  735. 

  736.             if (filterCode) {

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

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

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

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

  741.                 filterPos[i / 2] = xx;

  742. 

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

  744. 

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

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

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

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

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

  750.                                                                (c << 4) |

  751.                                                                (d << 6);

  752. 

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

  754.                     shift = maxShift;               // avoid overread

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

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

  757. 

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

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

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

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

  762.                 }

  763.             }

  764. 

  765.             fragmentPos += fragmentLength;

  766. 

  767.             if (filterCode)

  768.                 filterCode[fragmentPos] = RET;

  769.         }

  770.         xpos += xInc;

  771.     }

  772.     if (filterCode)

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

  774. 

  775.     return fragmentPos + 1;

  776. }

  777. #endif /* HAVE_MMXEXT_INLINE */

  778. 

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

  780. {

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

  782.     *h = desc->log2_chroma_w;

  783.     *v = desc->log2_chroma_h;

  784. }

  785. 

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

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

  788.                              int brightness, int contrast, int saturation)

  789. {

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

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

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

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

  794. 

  795.     c->brightness = brightness;

  796.     c->contrast   = contrast;

  797.     c->saturation = saturation;

  798.     c->srcRange   = srcRange;

  799.     c->dstRange   = dstRange;

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

  801.         return -1;

  802. 

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

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

  805. 

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

  807.                              contrast, saturation);

  808.     // FIXME factorize

  809. 

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

  811.         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,

  812.                                        contrast, saturation);

  813.     return 0;

  814. }

  815. 

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

  817.                              int *srcRange, int **table, int *dstRange,

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

  819. {

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

  821.         return -1;

  822. 

  823.     *inv_table  = c->srcColorspaceTable;

  824.     *table      = c->dstColorspaceTable;

  825.     *srcRange   = c->srcRange;

  826.     *dstRange   = c->dstRange;

  827.     *brightness = c->brightness;

  828.     *contrast   = c->contrast;

  829.     *saturation = c->saturation;

  830. 

  831.     return 0;

  832. }

  833. 

  834. static int handle_jpeg(enum AVPixelFormat *format)

  835. {

  836.     switch (*format) {

  837.     case AV_PIX_FMT_YUVJ420P:

  838.         *format = AV_PIX_FMT_YUV420P;

  839.         return 1;

  840.     case AV_PIX_FMT_YUVJ422P:

  841.         *format = AV_PIX_FMT_YUV422P;

  842.         return 1;

  843.     case AV_PIX_FMT_YUVJ444P:

  844.         *format = AV_PIX_FMT_YUV444P;

  845.         return 1;

  846.     case AV_PIX_FMT_YUVJ440P:

  847.         *format = AV_PIX_FMT_YUV440P;

  848.         return 1;

  849.     default:

  850.         return 0;

  851.     }

  852. }

  853. 

  854. SwsContext *sws_alloc_context(void)

  855. {

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

  857. 

  858.     if (c) {

  859.         c->av_class = &sws_context_class;

  860.         av_opt_set_defaults(c);

  861.     }

  862. 

  863.     return c;

  864. }

  865. 

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

  867.                              SwsFilter *dstFilter)

  868. {

  869.     int i;

  870.     int usesVFilter, usesHFilter;

  871.     int unscaled;

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

  873.     int srcW              = c->srcW;

  874.     int srcH              = c->srcH;

  875.     int dstW              = c->dstW;

  876.     int dstH              = c->dstH;

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

  878.     int dst_stride_px     = dst_stride >> 1;

  879.     int flags, cpu_flags;

  880.     enum AVPixelFormat srcFormat = c->srcFormat;

  881.     enum AVPixelFormat dstFormat = c->dstFormat;

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

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

  884. 

  885.     cpu_flags = av_get_cpu_flags();

  886.     flags     = c->flags;

  887.     emms_c();

  888.     if (!rgb15to16)

  889.         sws_rgb2rgb_init();

  890. 

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

  892. 

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

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

  895.     if (!sws_isSupportedInput(srcFormat)) {

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

  897.                sws_format_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.                sws_format_name(dstFormat));

  903.         return AVERROR(EINVAL);

  904.     }

  905.     }

  906. 

  907.     i = flags & (SWS_POINT         |

  908.                  SWS_AREA          |

  909.                  SWS_BILINEAR      |

  910.                  SWS_FAST_BILINEAR |

  911.                  SWS_BICUBIC       |

  912.                  SWS_X             |

  913.                  SWS_GAUSS         |

  914.                  SWS_LANCZOS       |

  915.                  SWS_SINC          |

  916.                  SWS_SPLINE        |

  917.                  SWS_BICUBLIN);

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

  919.         av_log(c, AV_LOG_ERROR,

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

  921.         return AVERROR(EINVAL);

  922.     }

  923.     /* sanity check */

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

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

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

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

  928.                srcW, srcH, dstW, dstH);

  929.         return AVERROR(EINVAL);

  930.     }

  931. 

  932.     if (!dstFilter)

  933.         dstFilter = &dummyFilter;

  934.     if (!srcFilter)

  935.         srcFilter = &dummyFilter;

  936. 

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

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

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

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

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

  942. 

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

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

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

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

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

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

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

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

  951. 

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

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

  954. 

  955.     if (isPlanarRGB(dstFormat)) {

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

  957.             av_log(c, AV_LOG_DEBUG,

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

  959.                    av_get_pix_fmt_name(dstFormat));

  960.             flags   |= SWS_FULL_CHR_H_INT;

  961.             c->flags = flags;

  962.         }

  963.     }

  964. 

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

  966.      * chroma interpolation */

  967.     if (flags & SWS_FULL_CHR_H_INT &&

  968.         isAnyRGB(dstFormat)        &&

  969.         !isPlanarRGB(dstFormat)    &&

  970.         dstFormat != AV_PIX_FMT_RGBA  &&

  971.         dstFormat != AV_PIX_FMT_ARGB  &&

  972.         dstFormat != AV_PIX_FMT_BGRA  &&

  973.         dstFormat != AV_PIX_FMT_ABGR  &&

  974.         dstFormat != AV_PIX_FMT_RGB24 &&

  975.         dstFormat != AV_PIX_FMT_BGR24) {

  976.         av_log(c, AV_LOG_ERROR,

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

  978.                sws_format_name(dstFormat));

  979.         flags   &= ~SWS_FULL_CHR_H_INT;

  980.         c->flags = flags;

  981.     }

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

  983.         c->chrDstHSubSample = 1;

  984. 

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

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

  987.                            SWS_SRC_V_CHR_DROP_SHIFT;

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

  989. 

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

  991.      * wants full chroma */

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

  993.         srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&

  994.         srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&

  995.         srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&

  996.         srcFormat != AV_PIX_FMT_GBRP9BE   && srcFormat != AV_PIX_FMT_GBRP9LE  &&

  997.         srcFormat != AV_PIX_FMT_GBRP10BE  && srcFormat != AV_PIX_FMT_GBRP10LE &&

  998.         srcFormat != AV_PIX_FMT_GBRP16BE  && srcFormat != AV_PIX_FMT_GBRP16LE &&

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

  1000.          (flags & SWS_FAST_BILINEAR)))

  1001.         c->chrSrcHSubSample = 1;

  1002. 

  1003.     // Note the -((-x)>>y) is so that we always round toward +inf.

  1004.     c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);

  1005.     c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);

  1006.     c->chrDstW = -((-dstW) >> c->chrDstHSubSample);

  1007.     c->chrDstH = -((-dstH) >> c->chrDstVSubSample);

  1008. 

  1009.     /* unscaled special cases */

  1010.     if (unscaled && !usesHFilter && !usesVFilter &&

  1011.         (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {

  1012.         ff_get_unscaled_swscale(c);

  1013. 

  1014.         if (c->swScale) {

  1015.             if (flags & SWS_PRINT_INFO)

  1016.                 av_log(c, AV_LOG_INFO,

  1017.                        "using unscaled %s -> %s special converter\n",

  1018.                        sws_format_name(srcFormat), sws_format_name(dstFormat));

  1019.             return 0;

  1020.         }

  1021.     }

  1022. 

  1023.     c->srcBpc = 1 + desc_src->comp[0].depth_minus1;

  1024.     if (c->srcBpc < 8)

  1025.         c->srcBpc = 8;

  1026.     c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;

  1027.     if (c->dstBpc < 8)

  1028.         c->dstBpc = 8;

  1029.     if (c->dstBpc == 16)

  1030.         dst_stride <<= 1;

  1031.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,

  1032.                      (FFALIGN(srcW, 16) * 2 * FFALIGN(c->srcBpc, 8) >> 3) + 16,

  1033.                      fail);

  1034.     if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 10) {

  1035.         c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&

  1036.                               (srcW & 15) == 0) ? 1 : 0;

  1037.         if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0

  1038.             && (flags & SWS_FAST_BILINEAR)) {

  1039.             if (flags & SWS_PRINT_INFO)

  1040.                 av_log(c, AV_LOG_INFO,

  1041.                        "output width is not a multiple of 32 -> no MMXEXT scaler\n");

  1042.         }

  1043.         if (usesHFilter)

  1044.             c->canMMXEXTBeUsed = 0;

  1045.     } else

  1046.         c->canMMXEXTBeUsed = 0;

  1047. 

  1048.     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;

  1049.     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

  1050. 

  1051.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  1052.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  1053.      * correct scaling.

  1054.      * n-2 is the last chrominance sample available.

  1055.      * This is not perfect, but no one should notice the difference, the more

  1056.      * correct variant would be like the vertical one, but that would require

  1057.      * some special code for the first and last pixel */

  1058.     if (flags & SWS_FAST_BILINEAR) {

  1059.         if (c->canMMXEXTBeUsed) {

  1060.             c->lumXInc += 20;

  1061.             c->chrXInc += 20;

  1062.         }

  1063.         // we don't use the x86 asm scaler if MMX is available

  1064.         else if (INLINE_MMX(cpu_flags)) {

  1065.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1066.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1067.         }

  1068.     }

  1069. 

  1070. #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)

  1071. 

  1072.     /* precalculate horizontal scaler filter coefficients */

  1073.     {

  1074. #if HAVE_MMXEXT_INLINE

  1075. // can't downscale !!!

  1076.         if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {

  1077.             c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,

  1078.                                                              NULL, NULL, 8);

  1079.             c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,

  1080.                                                              NULL, NULL, NULL, 4);

  1081. 

  1082. #if USE_MMAP

  1083.             c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,

  1084.                                           PROT_READ | PROT_WRITE,

  1085.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1086.                                           -1, 0);

  1087.             c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,

  1088.                                           PROT_READ | PROT_WRITE,

  1089.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1090.                                           -1, 0);

  1091. #elif HAVE_VIRTUALALLOC

  1092.             c->lumMmxextFilterCode = VirtualAlloc(NULL,

  1093.                                                   c->lumMmxextFilterCodeSize,

  1094.                                                   MEM_COMMIT,

  1095.                                                   PAGE_EXECUTE_READWRITE);

  1096.             c->chrMmxextFilterCode = VirtualAlloc(NULL,

  1097.                                                   c->chrMmxextFilterCodeSize,

  1098.                                                   MEM_COMMIT,

  1099.                                                   PAGE_EXECUTE_READWRITE);

  1100. #else

  1101.             c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);

  1102.             c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);

  1103. #endif

  1104. 

  1105.             if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)

  1106.                 return AVERROR(ENOMEM);

  1107.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1108.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1109.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1110.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1111. 

  1112.             init_hscaler_mmxext(dstW, c->lumXInc, c->lumMmxextFilterCode,

  1113.                                 c->hLumFilter, c->hLumFilterPos, 8);

  1114.             init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,

  1115.                                 c->hChrFilter, c->hChrFilterPos, 4);

  1116. 

  1117. #if USE_MMAP

  1118.             mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1119.             mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1120. #endif

  1121.         } else

  1122. #endif /* HAVE_MMXEXT_INLINE */

  1123.         {

  1124.             const int filterAlign =

  1125.                 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :

  1126.                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1127.                 1;

  1128. 

  1129.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1130.                            &c->hLumFilterSize, c->lumXInc,

  1131.                            srcW, dstW, filterAlign, 1 << 14,

  1132.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1133.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1134.                            c->param, 1) < 0)

  1135.                 goto fail;

  1136.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1137.                            &c->hChrFilterSize, c->chrXInc,

  1138.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1139.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1140.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1141.                            c->param, 1) < 0)

  1142.                 goto fail;

  1143.         }

  1144.     } // initialize horizontal stuff

  1145. 

  1146.     /* precalculate vertical scaler filter coefficients */

  1147.     {

  1148.         const int filterAlign =

  1149.             (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :

  1150.             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1151.             1;

  1152. 

  1153.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1154.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

  1155.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1156.                        cpu_flags, srcFilter->lumV, dstFilter->lumV,

  1157.                        c->param, 0) < 0)

  1158.             goto fail;

  1159.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1160.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1161.                        filterAlign, (1 << 12),

  1162.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1163.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1164.                        c->param, 0) < 0)

  1165.             goto fail;

  1166. 

  1167. #if HAVE_ALTIVEC

  1168.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1169.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1170. 

  1171.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1172.             int j;

  1173.             short *p = (short *)&c->vYCoeffsBank[i];

  1174.             for (j = 0; j < 8; j++)

  1175.                 p[j] = c->vLumFilter[i];

  1176.         }

  1177. 

  1178.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1179.             int j;

  1180.             short *p = (short *)&c->vCCoeffsBank[i];

  1181.             for (j = 0; j < 8; j++)

  1182.                 p[j] = c->vChrFilter[i];

  1183.         }

  1184. #endif

  1185.     }

  1186. 

  1187.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1188.     c->vLumBufSize = c->vLumFilterSize;

  1189.     c->vChrBufSize = c->vChrFilterSize;

  1190.     for (i = 0; i < dstH; i++) {

  1191.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1192.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1193.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1194.                                << c->chrSrcVSubSample));

  1195. 

  1196.         nextSlice >>= c->chrSrcVSubSample;

  1197.         nextSlice <<= c->chrSrcVSubSample;

  1198.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1199.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1200.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1201.             (nextSlice >> c->chrSrcVSubSample))

  1202.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1203.                              c->vChrFilterPos[chrI];

  1204.     }

  1205. 

  1206.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1207.      * need to allocate several megabytes to handle all possible cases) */

  1208.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1209.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1210.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1211.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1212.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1213.     /* Note we need at least one pixel more at the end because of the MMX code

  1214.      * (just in case someone wants to replace the 4000/8000). */

  1215.     /* align at 16 bytes for AltiVec */

  1216.     for (i = 0; i < c->vLumBufSize; i++) {

  1217.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1218.                           dst_stride + 16, fail);

  1219.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1220.     }

  1221.     // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)

  1222.     c->uv_off_px   = dst_stride_px + 64 / (c->dstBpc & ~7);

  1223.     c->uv_off_byte = dst_stride + 16;

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

  1225.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1226.                          dst_stride * 2 + 32, fail);

  1227.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1228.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1229.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1230.     }

  1231.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1232.         for (i = 0; i < c->vLumBufSize; i++) {

  1233.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],

  1234.                               dst_stride + 16, fail);

  1235.             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];

  1236.         }

  1237. 

  1238.     // try to avoid drawing green stuff between the right end and the stride end

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

  1240.         memset(c->chrUPixBuf[i], 64, dst_stride * 2 + 1);

  1241. 

  1242.     assert(c->chrDstH <= dstH);

  1243. 

  1244.     if (flags & SWS_PRINT_INFO) {

  1245.         if (flags & SWS_FAST_BILINEAR)

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

  1247.         else if (flags & SWS_BILINEAR)

  1248.             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1249.         else if (flags & SWS_BICUBIC)

  1250.             av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1251.         else if (flags & SWS_X)

  1252.             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1253.         else if (flags & SWS_POINT)

  1254.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1255.         else if (flags & SWS_AREA)

  1256.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1257.         else if (flags & SWS_BICUBLIN)

  1258.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1259.         else if (flags & SWS_GAUSS)

  1260.             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1261.         else if (flags & SWS_SINC)

  1262.             av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1263.         else if (flags & SWS_LANCZOS)

  1264.             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1265.         else if (flags & SWS_SPLINE)

  1266.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1267.         else

  1268.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1269. 

  1270.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1271.                sws_format_name(srcFormat),

  1272. #ifdef DITHER1XBPP

  1273.                dstFormat == AV_PIX_FMT_BGR555   || dstFormat == AV_PIX_FMT_BGR565   ||

  1274.                dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||

  1275.                dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?

  1276.                                                              "dithered " : "",

  1277. #else

  1278.                "",

  1279. #endif

  1280.                sws_format_name(dstFormat));

  1281. 

  1282.         if (INLINE_MMXEXT(cpu_flags))

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

  1284.         else if (INLINE_AMD3DNOW(cpu_flags))

  1285.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1286.         else if (INLINE_MMX(cpu_flags))

  1287.             av_log(c, AV_LOG_INFO, "using MMX\n");

  1288.         else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)

  1289.             av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1290.         else

  1291.             av_log(c, AV_LOG_INFO, "using C\n");

  1292. 

  1293.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1294.         av_log(c, AV_LOG_DEBUG,

  1295.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1296.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1297.         av_log(c, AV_LOG_DEBUG,

  1298.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1299.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1300.                c->chrXInc, c->chrYInc);

  1301.     }

  1302. 

  1303.     c->swScale = ff_getSwsFunc(c);

  1304.     return 0;

  1305. fail: // FIXME replace things by appropriate error codes

  1306.     return -1;

  1307. }

  1308. 

  1309. #if FF_API_SWS_GETCONTEXT

  1310. SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,

  1311.                            int dstW, int dstH, enum AVPixelFormat dstFormat,

  1312.                            int flags, SwsFilter *srcFilter,

  1313.                            SwsFilter *dstFilter, const double *param)

  1314. {

  1315.     SwsContext *c;

  1316. 

  1317.     if (!(c = sws_alloc_context()))

  1318.         return NULL;

  1319. 

  1320.     c->flags     = flags;

  1321.     c->srcW      = srcW;

  1322.     c->srcH      = srcH;

  1323.     c->dstW      = dstW;

  1324.     c->dstH      = dstH;

  1325.     c->srcRange  = handle_jpeg(&srcFormat);

  1326.     c->dstRange  = handle_jpeg(&dstFormat);

  1327.     c->srcFormat = srcFormat;

  1328.     c->dstFormat = dstFormat;

  1329. 

  1330.     if (param) {

  1331.         c->param[0] = param[0];

  1332.         c->param[1] = param[1];

  1333.     }

  1334.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1335.                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1336.                              c->dstRange, 0, 1 << 16, 1 << 16);

  1337. 

  1338.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1339.         sws_freeContext(c);

  1340.         return NULL;

  1341.     }

  1342. 

  1343.     return c;

  1344. }

  1345. #endif

  1346. 

  1347. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1348.                                 float lumaSharpen, float chromaSharpen,

  1349.                                 float chromaHShift, float chromaVShift,

  1350.                                 int verbose)

  1351. {

  1352.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1353.     if (!filter)

  1354.         return NULL;

  1355. 

  1356.     if (lumaGBlur != 0.0) {

  1357.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1358.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1359.     } else {

  1360.         filter->lumH = sws_getIdentityVec();

  1361.         filter->lumV = sws_getIdentityVec();

  1362.     }

  1363. 

  1364.     if (chromaGBlur != 0.0) {

  1365.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1366.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1367.     } else {

  1368.         filter->chrH = sws_getIdentityVec();

  1369.         filter->chrV = sws_getIdentityVec();

  1370.     }

  1371. 

  1372.     if (chromaSharpen != 0.0) {

  1373.         SwsVector *id = sws_getIdentityVec();

  1374.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1375.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1376.         sws_addVec(filter->chrH, id);

  1377.         sws_addVec(filter->chrV, id);

  1378.         sws_freeVec(id);

  1379.     }

  1380. 

  1381.     if (lumaSharpen != 0.0) {

  1382.         SwsVector *id = sws_getIdentityVec();

  1383.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1384.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1385.         sws_addVec(filter->lumH, id);

  1386.         sws_addVec(filter->lumV, id);

  1387.         sws_freeVec(id);

  1388.     }

  1389. 

  1390.     if (chromaHShift != 0.0)

  1391.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1392. 

  1393.     if (chromaVShift != 0.0)

  1394.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1395. 

  1396.     sws_normalizeVec(filter->chrH, 1.0);

  1397.     sws_normalizeVec(filter->chrV, 1.0);

  1398.     sws_normalizeVec(filter->lumH, 1.0);

  1399.     sws_normalizeVec(filter->lumV, 1.0);

  1400. 

  1401.     if (verbose)

  1402.         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1403.     if (verbose)

  1404.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1405. 

  1406.     return filter;

  1407. }

  1408. 

  1409. SwsVector *sws_allocVec(int length)

  1410. {

  1411.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1412.     if (!vec)

  1413.         return NULL;

  1414.     vec->length = length;

  1415.     vec->coeff  = av_malloc(sizeof(double) * length);

  1416.     if (!vec->coeff)

  1417.         av_freep(&vec);

  1418.     return vec;

  1419. }

  1420. 

  1421. SwsVector *sws_getGaussianVec(double variance, double quality)

  1422. {

  1423.     const int length = (int)(variance * quality + 0.5) | 1;

  1424.     int i;

  1425.     double middle  = (length - 1) * 0.5;

  1426.     SwsVector *vec = sws_allocVec(length);

  1427. 

  1428.     if (!vec)

  1429.         return NULL;

  1430. 

  1431.     for (i = 0; i < length; i++) {

  1432.         double dist = i - middle;

  1433.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1434.                         sqrt(2 * variance * M_PI);

  1435.     }

  1436. 

  1437.     sws_normalizeVec(vec, 1.0);

  1438. 

  1439.     return vec;

  1440. }

  1441. 

  1442. SwsVector *sws_getConstVec(double c, int length)

  1443. {

  1444.     int i;

  1445.     SwsVector *vec = sws_allocVec(length);

  1446. 

  1447.     if (!vec)

  1448.         return NULL;

  1449. 

  1450.     for (i = 0; i < length; i++)

  1451.         vec->coeff[i] = c;

  1452. 

  1453.     return vec;

  1454. }

  1455. 

  1456. SwsVector *sws_getIdentityVec(void)

  1457. {

  1458.     return sws_getConstVec(1.0, 1);

  1459. }

  1460. 

  1461. static double sws_dcVec(SwsVector *a)

  1462. {

  1463.     int i;

  1464.     double sum = 0;

  1465. 

  1466.     for (i = 0; i < a->length; i++)

  1467.         sum += a->coeff[i];

  1468. 

  1469.     return sum;

  1470. }

  1471. 

  1472. void sws_scaleVec(SwsVector *a, double scalar)

  1473. {

  1474.     int i;

  1475. 

  1476.     for (i = 0; i < a->length; i++)

  1477.         a->coeff[i] *= scalar;

  1478. }

  1479. 

  1480. void sws_normalizeVec(SwsVector *a, double height)

  1481. {

  1482.     sws_scaleVec(a, height / sws_dcVec(a));

  1483. }

  1484. 

  1485. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1486. {

  1487.     int length = a->length + b->length - 1;

  1488.     int i, j;

  1489.     SwsVector *vec = sws_getConstVec(0.0, length);

  1490. 

  1491.     if (!vec)

  1492.         return NULL;

  1493. 

  1494.     for (i = 0; i < a->length; i++) {

  1495.         for (j = 0; j < b->length; j++) {

  1496.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1497.         }

  1498.     }

  1499. 

  1500.     return vec;

  1501. }

  1502. 

  1503. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1504. {

  1505.     int length = FFMAX(a->length, b->length);

  1506.     int i;

  1507.     SwsVector *vec = sws_getConstVec(0.0, length);

  1508. 

  1509.     if (!vec)

  1510.         return NULL;

  1511. 

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

  1513.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1514.     for (i = 0; i < b->length; i++)

  1515.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

  1516. 

  1517.     return vec;

  1518. }

  1519. 

  1520. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1521. {

  1522.     int length = FFMAX(a->length, b->length);

  1523.     int i;

  1524.     SwsVector *vec = sws_getConstVec(0.0, length);

  1525. 

  1526.     if (!vec)

  1527.         return NULL;

  1528. 

  1529.     for (i = 0; i < a->length; i++)

  1530.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

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

  1532.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

  1533. 

  1534.     return vec;

  1535. }

  1536. 

  1537. /* shift left / or right if "shift" is negative */

  1538. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1539. {

  1540.     int length = a->length + FFABS(shift) * 2;

  1541.     int i;

  1542.     SwsVector *vec = sws_getConstVec(0.0, length);

  1543. 

  1544.     if (!vec)

  1545.         return NULL;

  1546. 

  1547.     for (i = 0; i < a->length; i++) {

  1548.         vec->coeff[i + (length    - 1) / 2 -

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

  1550.     }

  1551. 

  1552.     return vec;

  1553. }

  1554. 

  1555. void sws_shiftVec(SwsVector *a, int shift)

  1556. {

  1557.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1558.     av_free(a->coeff);

  1559.     a->coeff  = shifted->coeff;

  1560.     a->length = shifted->length;

  1561.     av_free(shifted);

  1562. }

  1563. 

  1564. void sws_addVec(SwsVector *a, SwsVector *b)

  1565. {

  1566.     SwsVector *sum = sws_sumVec(a, b);

  1567.     av_free(a->coeff);

  1568.     a->coeff  = sum->coeff;

  1569.     a->length = sum->length;

  1570.     av_free(sum);

  1571. }

  1572. 

  1573. void sws_subVec(SwsVector *a, SwsVector *b)

  1574. {

  1575.     SwsVector *diff = sws_diffVec(a, b);

  1576.     av_free(a->coeff);

  1577.     a->coeff  = diff->coeff;

  1578.     a->length = diff->length;

  1579.     av_free(diff);

  1580. }

  1581. 

  1582. void sws_convVec(SwsVector *a, SwsVector *b)

  1583. {

  1584.     SwsVector *conv = sws_getConvVec(a, b);

  1585.     av_free(a->coeff);

  1586.     a->coeff  = conv->coeff;

  1587.     a->length = conv->length;

  1588.     av_free(conv);

  1589. }

  1590. 

  1591. SwsVector *sws_cloneVec(SwsVector *a)

  1592. {

  1593.     int i;

  1594.     SwsVector *vec = sws_allocVec(a->length);

  1595. 

  1596.     if (!vec)

  1597.         return NULL;

  1598. 

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

  1600.         vec->coeff[i] = a->coeff[i];

  1601. 

  1602.     return vec;

  1603. }

  1604. 

  1605. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1606. {

  1607.     int i;

  1608.     double max = 0;

  1609.     double min = 0;

  1610.     double range;

  1611. 

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

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

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

  1615. 

  1616.     for (i = 0; i < a->length; i++)

  1617.         if (a->coeff[i] < min)

  1618.             min = a->coeff[i];

  1619. 

  1620.     range = max - min;

  1621. 

  1622.     for (i = 0; i < a->length; i++) {

  1623.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1624.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1625.         for (; x > 0; x--)

  1626.             av_log(log_ctx, log_level, " ");

  1627.         av_log(log_ctx, log_level, "|\n");

  1628.     }

  1629. }

  1630. 

  1631. void sws_freeVec(SwsVector *a)

  1632. {

  1633.     if (!a)

  1634.         return;

  1635.     av_freep(&a->coeff);

  1636.     a->length = 0;

  1637.     av_free(a);

  1638. }

  1639. 

  1640. void sws_freeFilter(SwsFilter *filter)

  1641. {

  1642.     if (!filter)

  1643.         return;

  1644. 

  1645.     if (filter->lumH)

  1646.         sws_freeVec(filter->lumH);

  1647.     if (filter->lumV)

  1648.         sws_freeVec(filter->lumV);

  1649.     if (filter->chrH)

  1650.         sws_freeVec(filter->chrH);

  1651.     if (filter->chrV)

  1652.         sws_freeVec(filter->chrV);

  1653.     av_free(filter);

  1654. }

  1655. 

  1656. void sws_freeContext(SwsContext *c)

  1657. {

  1658.     int i;

  1659.     if (!c)

  1660.         return;

  1661. 

  1662.     if (c->lumPixBuf) {

  1663.         for (i = 0; i < c->vLumBufSize; i++)

  1664.             av_freep(&c->lumPixBuf[i]);

  1665.         av_freep(&c->lumPixBuf);

  1666.     }

  1667. 

  1668.     if (c->chrUPixBuf) {

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

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

  1671.         av_freep(&c->chrUPixBuf);

  1672.         av_freep(&c->chrVPixBuf);

  1673.     }

  1674. 

  1675.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1676.         for (i = 0; i < c->vLumBufSize; i++)

  1677.             av_freep(&c->alpPixBuf[i]);

  1678.         av_freep(&c->alpPixBuf);

  1679.     }

  1680. 

  1681.     av_freep(&c->vLumFilter);

  1682.     av_freep(&c->vChrFilter);

  1683.     av_freep(&c->hLumFilter);

  1684.     av_freep(&c->hChrFilter);

  1685. #if HAVE_ALTIVEC

  1686.     av_freep(&c->vYCoeffsBank);

  1687.     av_freep(&c->vCCoeffsBank);

  1688. #endif

  1689. 

  1690.     av_freep(&c->vLumFilterPos);

  1691.     av_freep(&c->vChrFilterPos);

  1692.     av_freep(&c->hLumFilterPos);

  1693.     av_freep(&c->hChrFilterPos);

  1694. 

  1695. #if HAVE_MMX_INLINE

  1696. #if USE_MMAP

  1697.     if (c->lumMmxextFilterCode)

  1698.         munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);

  1699.     if (c->chrMmxextFilterCode)

  1700.         munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);

  1701. #elif HAVE_VIRTUALALLOC

  1702.     if (c->lumMmxextFilterCode)

  1703.         VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);

  1704.     if (c->chrMmxextFilterCode)

  1705.         VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);

  1706. #else

  1707.     av_free(c->lumMmxextFilterCode);

  1708.     av_free(c->chrMmxextFilterCode);

  1709. #endif

  1710.     c->lumMmxextFilterCode = NULL;

  1711.     c->chrMmxextFilterCode = NULL;

  1712. #endif /* HAVE_MMX_INLINE */

  1713. 

  1714.     av_freep(&c->yuvTable);

  1715.     av_free(c->formatConvBuffer);

  1716. 

  1717.     av_free(c);

  1718. }

  1719. 

  1720. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  1721.                                         int srcH, enum AVPixelFormat srcFormat,

  1722.                                         int dstW, int dstH,

  1723.                                         enum AVPixelFormat dstFormat, int flags,

  1724.                                         SwsFilter *srcFilter,

  1725.                                         SwsFilter *dstFilter,

  1726.                                         const double *param)

  1727. {

  1728.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  1729.                                              SWS_PARAM_DEFAULT };

  1730. 

  1731.     if (!param)

  1732.         param = default_param;

  1733. 

  1734.     if (context &&

  1735.         (context->srcW      != srcW      ||

  1736.          context->srcH      != srcH      ||

  1737.          context->srcFormat != srcFormat ||

  1738.          context->dstW      != dstW      ||

  1739.          context->dstH      != dstH      ||

  1740.          context->dstFormat != dstFormat ||

  1741.          context->flags     != flags     ||

  1742.          context->param[0]  != param[0]  ||

  1743.          context->param[1]  != param[1])) {

  1744.         sws_freeContext(context);

  1745.         context = NULL;

  1746.     }

  1747. 

  1748.     if (!context) {

  1749.         if (!(context = sws_alloc_context()))

  1750.             return NULL;

  1751.         context->srcW      = srcW;

  1752.         context->srcH      = srcH;

  1753.         context->srcRange  = handle_jpeg(&srcFormat);

  1754.         context->srcFormat = srcFormat;

  1755.         context->dstW      = dstW;

  1756.         context->dstH      = dstH;

  1757.         context->dstRange  = handle_jpeg(&dstFormat);

  1758.         context->dstFormat = dstFormat;

  1759.         context->flags     = flags;

  1760.         context->param[0]  = param[0];

  1761.         context->param[1]  = param[1];

  1762.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  1763.                                  context->srcRange,

  1764.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1765.                                  context->dstRange, 0, 1 << 16, 1 << 16);

  1766.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1767.             sws_freeContext(context);

  1768.             return NULL;

  1769.         }

  1770.     }

  1771.     return context;

  1772. }