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.
40357 Commits
32 Branches
283MB
Tree: e96c3b81ca
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e96c3b81ca'
${ noResults }
FFmpeg/libswscale/utils.c

1783 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/ppc/cpu.h"

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

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

  51. #include "rgb2rgb.h"

  52. #include "swscale.h"

  53. #include "swscale_internal.h"

  54. 

  55. unsigned swscale_version(void)

  56. {

  57.     return LIBSWSCALE_VERSION_INT;

  58. }

  59. 

  60. const char *swscale_configuration(void)

  61. {

  62.     return LIBAV_CONFIGURATION;

  63. }

  64. 

  65. const char *swscale_license(void)

  66. {

  67. #define LICENSE_PREFIX "libswscale license: "

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

  69. }

  70. 

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

  72. 

  73. typedef struct FormatEntry {

  74.     uint8_t is_supported_in         :1;

  75.     uint8_t is_supported_out        :1;

  76.     uint8_t is_supported_endianness :1;

  77. } FormatEntry;

  78. 

  79. static const FormatEntry format_entries[AV_PIX_FMT_NB] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  95.     [AV_PIX_FMT_YVYU422]     = { 1, 1 },

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  137.     [AV_PIX_FMT_RGBA64BE]    = { 0, 0, 1 },

  138.     [AV_PIX_FMT_RGBA64LE]    = { 0, 0, 1 },

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  157.     [AV_PIX_FMT_YA8]         = { 1, 0 },

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

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

  160.     [AV_PIX_FMT_BGRA64BE]    = { 0, 0, 1 },

  161.     [AV_PIX_FMT_BGRA64LE]    = { 0, 0, 1 },

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  183. };

  184. 

  185. int sws_isSupportedInput(enum AVPixelFormat pix_fmt)

  186. {

  187.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  188.            format_entries[pix_fmt].is_supported_in : 0;

  189. }

  190. 

  191. int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)

  192. {

  193.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  194.            format_entries[pix_fmt].is_supported_out : 0;

  195. }

  196. 

  197. int sws_isSupportedEndiannessConversion(enum AVPixelFormat pix_fmt)

  198. {

  199.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  200.            format_entries[pix_fmt].is_supported_endianness : 0;

  201. }

  202. 

  203. const char *sws_format_name(enum AVPixelFormat format)

  204. {

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

  206.     if (desc)

  207.         return desc->name;

  208.     else

  209.         return "Unknown format";

  210. }

  211. 

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

  213.                              double dist)

  214. {

  215.     if (dist <= 1.0)

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

  217.     else

  218.         return getSplineCoeff(0.0,

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

  220.                                c + 3.0 * d,

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

  222.                               dist - 1.0);

  223. }

  224. 

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

  226.                               int *outFilterSize, int xInc, int srcW,

  227.                               int dstW, int filterAlign, int one,

  228.                               int flags, int cpu_flags,

  229.                               SwsVector *srcFilter, SwsVector *dstFilter,

  230.                               double param[2], int is_horizontal)

  231. {

  232.     int i;

  233.     int filterSize;

  234.     int filter2Size;

  235.     int minFilterSize;

  236.     int64_t *filter    = NULL;

  237.     int64_t *filter2   = NULL;

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

  239.     int ret            = -1;

  240. 

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

  242. 

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

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

  245. 

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

  247.         int i;

  248.         filterSize = 1;

  249.         FF_ALLOCZ_OR_GOTO(NULL, filter,

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

  251. 

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

  253.             filter[i * filterSize] = fone;

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

  255.         }

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

  257.         int i;

  258.         int xDstInSrc;

  259.         filterSize = 1;

  260.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  262. 

  263.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  266. 

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

  268.             filter[i]       = fone;

  269.             xDstInSrc      += xInc;

  270.         }

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

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

  273.         int i;

  274.         int xDstInSrc;

  275.         filterSize = 2;

  276.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  278. 

  279.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  282.             int j;

  283. 

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

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

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

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

  288.                                 (fone >> 16);

  289.                 if (coeff < 0)

  290.                     coeff = 0;

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

  292.                 xx++;

  293.             }

  294.             xDstInSrc += xInc;

  295.         }

  296.     } else {

  297.         int64_t xDstInSrc;

  298.         int sizeFactor;

  299. 

  300.         if (flags & SWS_BICUBIC)

  301.             sizeFactor = 4;

  302.         else if (flags & SWS_X)

  303.             sizeFactor = 8;

  304.         else if (flags & SWS_AREA)

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

  306.         else if (flags & SWS_GAUSS)

  307.             sizeFactor = 8;     // infinite ;)

  308.         else if (flags & SWS_LANCZOS)

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

  310.         else if (flags & SWS_SINC)

  311.             sizeFactor = 20;    // infinite ;)

  312.         else if (flags & SWS_SPLINE)

  313.             sizeFactor = 20;    // infinite ;)

  314.         else if (flags & SWS_BILINEAR)

  315.             sizeFactor = 2;

  316.         else {

  317.             sizeFactor = 0;     // GCC warning killer

  318.             assert(0);

  319.         }

  320. 

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

  322.             filterSize = 1 + sizeFactor;    // upscale

  323.         else

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

  325. 

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

  327.         filterSize = FFMAX(filterSize, 1);

  328. 

  329.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  331. 

  332.         xDstInSrc = xInc - 0x10000;

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

  334.             int xx = (xDstInSrc - ((int64_t)(filterSize - 2) << 16)) / (1 << 17);

  335.             int j;

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

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

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

  339.                 double floatd;

  340.                 int64_t coeff;

  341. 

  342.                 if (xInc > 1 << 16)

  343.                     d = d * dstW / srcW;

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

  345. 

  346.                 if (flags & SWS_BICUBIC) {

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

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

  349. 

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

  351.                         coeff = 0.0;

  352.                     } else {

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

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

  355. 

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

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

  358.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  359.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  360.                         else

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

  362.                                       (6 * B + 30 * C) * dd  +

  363.                                     (-12 * B - 48 * C) * d   +

  364.                                       (8 * B + 24 * C) * (1 << 30);

  365.                     }

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

  367.                 }

  368. #if 0

  369.                 else if (flags & SWS_X) {

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

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

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

  373.                 }

  374. #endif

  375.                 else if (flags & SWS_X) {

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

  377.                     double c;

  378. 

  379.                     if (floatd < 1.0)

  380.                         c = cos(floatd * M_PI);

  381.                     else

  382.                         c = -1.0;

  383.                     if (c < 0.0)

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

  385.                     else

  386.                         c = pow(c, A);

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

  388.                 } else if (flags & SWS_AREA) {

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

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

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

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

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

  394.                     else

  395.                         coeff = 0.0;

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

  397.                 } else if (flags & SWS_GAUSS) {

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

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

  400.                 } else if (flags & SWS_SINC) {

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

  402.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  406.                     if (floatd > p)

  407.                         coeff = 0;

  408.                 } else if (flags & SWS_BILINEAR) {

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

  410.                     if (coeff < 0)

  411.                         coeff = 0;

  412.                     coeff *= fone >> 30;

  413.                 } else if (flags & SWS_SPLINE) {

  414.                     double p = -2.196152422706632;

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

  416.                 } else {

  417.                     coeff = 0.0; // GCC warning killer

  418.                     assert(0);

  419.                 }

  420. 

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

  422.                 xx++;

  423.             }

  424.             xDstInSrc += 2 * xInc;

  425.         }

  426.     }

  427. 

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

  429.      * av_free(filter);

  430.      */

  431.     assert(filterSize > 0);

  432.     filter2Size = filterSize;

  433.     if (srcFilter)

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

  435.     if (dstFilter)

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

  437.     assert(filter2Size > 0);

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

  439. 

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

  441.         int j, k;

  442. 

  443.         if (srcFilter) {

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

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

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

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

  448.             }

  449.         } else {

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

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

  452.         }

  453.         // FIXME dstFilter

  454. 

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

  456.     }

  457.     av_freep(&filter);

  458. 

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

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

  461.     minFilterSize = 0;

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

  463.         int min = filter2Size;

  464.         int j;

  465.         int64_t cutOff = 0.0;

  466. 

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

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

  469.             int k;

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

  471. 

  472.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  473.                 break;

  474. 

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

  476.              * filter otherwise */

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

  478.                 break;

  479. 

  480.             // move filter coefficients left

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

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

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

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

  485.         }

  486. 

  487.         cutOff = 0;

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

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

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

  491. 

  492.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  493.                 break;

  494.             min--;

  495.         }

  496. 

  497.         if (min > minFilterSize)

  498.             minFilterSize = min;

  499.     }

  500. 

  501.     if (PPC_ALTIVEC(cpu_flags)) {

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

  503.         if (minFilterSize < 5)

  504.             filterAlign = 4;

  505. 

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

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

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

  509.         if (minFilterSize < 3)

  510.             filterAlign = 1;

  511.     }

  512. 

  513.     if (INLINE_MMX(cpu_flags)) {

  514.         // special case for unscaled vertical filtering

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

  516.             filterAlign = 1;

  517.     }

  518. 

  519.     assert(minFilterSize > 0);

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

  521.     assert(filterSize > 0);

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

  523.     if (filterSize >= MAX_FILTER_SIZE * 16 /

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

  525.         goto fail;

  526.     *outFilterSize = filterSize;

  527. 

  528.     if (flags & SWS_PRINT_INFO)

  529.         av_log(NULL, AV_LOG_VERBOSE,

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

  531.                filter2Size, filterSize);

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

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

  534.         int j;

  535. 

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

  537.             if (j >= filter2Size)

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

  539.             else

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

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

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

  543.         }

  544.     }

  545. 

  546.     // FIXME try to align filterPos if possible

  547. 

  548.     // fix borders

  549.     if (is_horizontal) {

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

  551.             int j;

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

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

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

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

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

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

  558.                 }

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

  560.             }

  561. 

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

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

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

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

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

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

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

  569.                 }

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

  571.             }

  572.         }

  573.     }

  574. 

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

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

  577.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

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

  579. 

  580.     /* normalize & store in outFilter */

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

  582.         int j;

  583.         int64_t error = 0;

  584.         int64_t sum   = 0;

  585. 

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

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

  588.         }

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

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

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

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

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

  594.             error                                  = v - intV * sum;

  595.         }

  596.     }

  597. 

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

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

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

  601.                                                       * read over the end */

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

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

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

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

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

  607.     }

  608. 

  609.     ret = 0;

  610. 

  611. fail:

  612.     av_free(filter);

  613.     av_free(filter2);

  614.     return ret;

  615. }

  616. 

  617. #if HAVE_MMXEXT_INLINE

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

  619.                                        int16_t *filter, int32_t *filterPos,

  620.                                        int numSplits)

  621. {

  622.     uint8_t *fragmentA;

  623.     x86_reg imm8OfPShufW1A;

  624.     x86_reg imm8OfPShufW2A;

  625.     x86_reg fragmentLengthA;

  626.     uint8_t *fragmentB;

  627.     x86_reg imm8OfPShufW1B;

  628.     x86_reg imm8OfPShufW2B;

  629.     x86_reg fragmentLengthB;

  630.     int fragmentPos;

  631. 

  632.     int xpos, i;

  633. 

  634.     // create an optimized horizontal scaling routine

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

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

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

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

  639.      */

  640. 

  641.     // code fragment

  642. 

  643.     __asm__ volatile (

  644.         "jmp                         9f                 \n\t"

  645.         // Begin

  646.         "0:                                             \n\t"

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

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

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

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

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

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

  653.         "1:                                             \n\t"

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

  655.         "2:                                             \n\t"

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

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

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

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

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

  661. 

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

  663. 

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

  665.         // End

  666.         "9:                                             \n\t"

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

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

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

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

  671.         "dec                         %1                 \n\t"

  672.         "dec                         %2                 \n\t"

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

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

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

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

  677. 

  678. 

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

  680.           "=r" (fragmentLengthA)

  681.         );

  682. 

  683.     __asm__ volatile (

  684.         "jmp                         9f                 \n\t"

  685.         // Begin

  686.         "0:                                             \n\t"

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

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

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

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

  691.         "1:                                             \n\t"

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

  693.         "2:                                             \n\t"

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

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

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

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

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

  699. 

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

  701. 

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

  703.         // End

  704.         "9:                                             \n\t"

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

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

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

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

  709.         "dec                         %1                 \n\t"

  710.         "dec                         %2                 \n\t"

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

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

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

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

  715. 

  716. 

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

  718.           "=r" (fragmentLengthB)

  719.         );

  720. 

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

  722.     fragmentPos = 0;

  723. 

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

  725.         int xx = xpos >> 16;

  726. 

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

  728.             int a                  = 0;

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

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

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

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

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

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

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

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

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

  738.             int shift              = 0;

  739. 

  740.             if (filterCode) {

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

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

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

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

  745.                 filterPos[i / 2] = xx;

  746. 

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

  748. 

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

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

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

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

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

  754.                                                                (c << 4) |

  755.                                                                (d << 6);

  756. 

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

  758.                     shift = maxShift;               // avoid overread

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

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

  761. 

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

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

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

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

  766.                 }

  767.             }

  768. 

  769.             fragmentPos += fragmentLength;

  770. 

  771.             if (filterCode)

  772.                 filterCode[fragmentPos] = RET;

  773.         }

  774.         xpos += xInc;

  775.     }

  776.     if (filterCode)

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

  778. 

  779.     return fragmentPos + 1;

  780. }

  781. #endif /* HAVE_MMXEXT_INLINE */

  782. 

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

  784. {

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

  786.     *h = desc->log2_chroma_w;

  787.     *v = desc->log2_chroma_h;

  788. }

  789. 

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

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

  792.                              int brightness, int contrast, int saturation)

  793. {

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

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

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

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

  798. 

  799.     c->brightness = brightness;

  800.     c->contrast   = contrast;

  801.     c->saturation = saturation;

  802.     c->srcRange   = srcRange;

  803.     c->dstRange   = dstRange;

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

  805.         return -1;

  806. 

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

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

  809. 

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

  811.                              contrast, saturation);

  812.     // FIXME factorize

  813. 

  814.     if (ARCH_PPC)

  815.         ff_yuv2rgb_init_tables_ppc(c, inv_table, brightness,

  816.                                    contrast, saturation);

  817.     return 0;

  818. }

  819. 

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

  821.                              int *srcRange, int **table, int *dstRange,

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

  823. {

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

  825.         return -1;

  826. 

  827.     *inv_table  = c->srcColorspaceTable;

  828.     *table      = c->dstColorspaceTable;

  829.     *srcRange   = c->srcRange;

  830.     *dstRange   = c->dstRange;

  831.     *brightness = c->brightness;

  832.     *contrast   = c->contrast;

  833.     *saturation = c->saturation;

  834. 

  835.     return 0;

  836. }

  837. 

  838. static int handle_jpeg(enum AVPixelFormat *format)

  839. {

  840.     switch (*format) {

  841.     case AV_PIX_FMT_YUVJ420P:

  842.         *format = AV_PIX_FMT_YUV420P;

  843.         return 1;

  844.     case AV_PIX_FMT_YUVJ422P:

  845.         *format = AV_PIX_FMT_YUV422P;

  846.         return 1;

  847.     case AV_PIX_FMT_YUVJ444P:

  848.         *format = AV_PIX_FMT_YUV444P;

  849.         return 1;

  850.     case AV_PIX_FMT_YUVJ440P:

  851.         *format = AV_PIX_FMT_YUV440P;

  852.         return 1;

  853.     default:

  854.         return 0;

  855.     }

  856. }

  857. 

  858. SwsContext *sws_alloc_context(void)

  859. {

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

  861. 

  862.     if (c) {

  863.         c->av_class = &sws_context_class;

  864.         av_opt_set_defaults(c);

  865.     }

  866. 

  867.     return c;

  868. }

  869. 

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

  871.                              SwsFilter *dstFilter)

  872. {

  873.     int i;

  874.     int usesVFilter, usesHFilter;

  875.     int unscaled;

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

  877.     int srcW              = c->srcW;

  878.     int srcH              = c->srcH;

  879.     int dstW              = c->dstW;

  880.     int dstH              = c->dstH;

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

  882.     int dst_stride_px     = dst_stride >> 1;

  883.     int flags, cpu_flags;

  884.     enum AVPixelFormat srcFormat = c->srcFormat;

  885.     enum AVPixelFormat dstFormat = c->dstFormat;

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

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

  888. 

  889.     cpu_flags = av_get_cpu_flags();

  890.     flags     = c->flags;

  891.     emms_c();

  892.     if (!rgb15to16)

  893.         sws_rgb2rgb_init();

  894. 

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

  896. 

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

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

  899.     if (!sws_isSupportedInput(srcFormat)) {

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

  901.                sws_format_name(srcFormat));

  902.         return AVERROR(EINVAL);

  903.     }

  904.     if (!sws_isSupportedOutput(dstFormat)) {

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

  906.                sws_format_name(dstFormat));

  907.         return AVERROR(EINVAL);

  908.     }

  909.     }

  910. 

  911.     i = flags & (SWS_POINT         |

  912.                  SWS_AREA          |

  913.                  SWS_BILINEAR      |

  914.                  SWS_FAST_BILINEAR |

  915.                  SWS_BICUBIC       |

  916.                  SWS_X             |

  917.                  SWS_GAUSS         |

  918.                  SWS_LANCZOS       |

  919.                  SWS_SINC          |

  920.                  SWS_SPLINE        |

  921.                  SWS_BICUBLIN);

  922. 

  923.     /* provide a default scaler if not set by caller */

  924.     if (!i) {

  925.         if (dstW < srcW && dstH < srcH)

  926.             flags |= SWS_GAUSS;

  927.         else if (dstW > srcW && dstH > srcH)

  928.             flags |= SWS_SINC;

  929.         else

  930.             flags |= SWS_LANCZOS;

  931.         c->flags = flags;

  932.     } else if (i & (i - 1)) {

  933.         av_log(c, AV_LOG_ERROR,

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

  935.         return AVERROR(EINVAL);

  936.     }

  937.     /* sanity check */

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

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

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

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

  942.                srcW, srcH, dstW, dstH);

  943.         return AVERROR(EINVAL);

  944.     }

  945. 

  946.     if (!dstFilter)

  947.         dstFilter = &dummyFilter;

  948.     if (!srcFilter)

  949.         srcFilter = &dummyFilter;

  950. 

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

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

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

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

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

  956. 

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

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

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

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

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

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

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

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

  965. 

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

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

  968. 

  969.     if (isPlanarRGB(dstFormat)) {

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

  971.             av_log(c, AV_LOG_DEBUG,

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

  973.                    av_get_pix_fmt_name(dstFormat));

  974.             flags   |= SWS_FULL_CHR_H_INT;

  975.             c->flags = flags;

  976.         }

  977.     }

  978. 

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

  980.      * chroma interpolation */

  981.     if (flags & SWS_FULL_CHR_H_INT &&

  982.         isAnyRGB(dstFormat)        &&

  983.         !isPlanarRGB(dstFormat)    &&

  984.         dstFormat != AV_PIX_FMT_RGBA  &&

  985.         dstFormat != AV_PIX_FMT_ARGB  &&

  986.         dstFormat != AV_PIX_FMT_BGRA  &&

  987.         dstFormat != AV_PIX_FMT_ABGR  &&

  988.         dstFormat != AV_PIX_FMT_RGB24 &&

  989.         dstFormat != AV_PIX_FMT_BGR24) {

  990.         av_log(c, AV_LOG_ERROR,

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

  992.                sws_format_name(dstFormat));

  993.         flags   &= ~SWS_FULL_CHR_H_INT;

  994.         c->flags = flags;

  995.     }

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

  997.         c->chrDstHSubSample = 1;

  998. 

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

  1000.     c->vChrDrop          = (flags & SWS_SRC_V_CHR_DROP_MASK) >>

  1001.                            SWS_SRC_V_CHR_DROP_SHIFT;

  1002.     c->chrSrcVSubSample += c->vChrDrop;

  1003. 

  1004.     /* drop every other pixel for chroma calculation unless user

  1005.      * wants full chroma */

  1006.     if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP)   &&

  1007.         srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&

  1008.         srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&

  1009.         srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&

  1010.         srcFormat != AV_PIX_FMT_GBRP9BE   && srcFormat != AV_PIX_FMT_GBRP9LE  &&

  1011.         srcFormat != AV_PIX_FMT_GBRP10BE  && srcFormat != AV_PIX_FMT_GBRP10LE &&

  1012.         srcFormat != AV_PIX_FMT_GBRP16BE  && srcFormat != AV_PIX_FMT_GBRP16LE &&

  1013.         ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||

  1014.          (flags & SWS_FAST_BILINEAR)))

  1015.         c->chrSrcHSubSample = 1;

  1016. 

  1017.     // Note the -((-x)>>y) is so that we always round toward +inf.

  1018.     c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);

  1019.     c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);

  1020.     c->chrDstW = -((-dstW) >> c->chrDstHSubSample);

  1021.     c->chrDstH = -((-dstH) >> c->chrDstVSubSample);

  1022. 

  1023.     /* unscaled special cases */

  1024.     if (unscaled && !usesHFilter && !usesVFilter &&

  1025.         (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {

  1026.         ff_get_unscaled_swscale(c);

  1027. 

  1028.         if (c->swscale) {

  1029.             if (flags & SWS_PRINT_INFO)

  1030.                 av_log(c, AV_LOG_INFO,

  1031.                        "using unscaled %s -> %s special converter\n",

  1032.                        sws_format_name(srcFormat), sws_format_name(dstFormat));

  1033.             return 0;

  1034.         }

  1035.     }

  1036. 

  1037.     c->srcBpc = 1 + desc_src->comp[0].depth_minus1;

  1038.     if (c->srcBpc < 8)

  1039.         c->srcBpc = 8;

  1040.     c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;

  1041.     if (c->dstBpc < 8)

  1042.         c->dstBpc = 8;

  1043.     if (c->dstBpc == 16)

  1044.         dst_stride <<= 1;

  1045.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,

  1046.                      (FFALIGN(srcW, 16) * 2 * FFALIGN(c->srcBpc, 8) >> 3) + 16,

  1047.                      fail);

  1048.     if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 10) {

  1049.         c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&

  1050.                               (srcW & 15) == 0) ? 1 : 0;

  1051.         if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0

  1052.             && (flags & SWS_FAST_BILINEAR)) {

  1053.             if (flags & SWS_PRINT_INFO)

  1054.                 av_log(c, AV_LOG_INFO,

  1055.                        "output width is not a multiple of 32 -> no MMXEXT scaler\n");

  1056.         }

  1057.         if (usesHFilter)

  1058.             c->canMMXEXTBeUsed = 0;

  1059.     } else

  1060.         c->canMMXEXTBeUsed = 0;

  1061. 

  1062.     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;

  1063.     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

  1064. 

  1065.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  1066.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  1067.      * correct scaling.

  1068.      * n-2 is the last chrominance sample available.

  1069.      * This is not perfect, but no one should notice the difference, the more

  1070.      * correct variant would be like the vertical one, but that would require

  1071.      * some special code for the first and last pixel */

  1072.     if (flags & SWS_FAST_BILINEAR) {

  1073.         if (c->canMMXEXTBeUsed) {

  1074.             c->lumXInc += 20;

  1075.             c->chrXInc += 20;

  1076.         }

  1077.         // we don't use the x86 asm scaler if MMX is available

  1078.         else if (INLINE_MMX(cpu_flags)) {

  1079.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1080.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1081.         }

  1082.     }

  1083. 

  1084. #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)

  1085. 

  1086.     /* precalculate horizontal scaler filter coefficients */

  1087.     {

  1088. #if HAVE_MMXEXT_INLINE

  1089. // can't downscale !!!

  1090.         if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {

  1091.             c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,

  1092.                                                              NULL, NULL, 8);

  1093.             c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,

  1094.                                                              NULL, NULL, NULL, 4);

  1095. 

  1096. #if USE_MMAP

  1097.             c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,

  1098.                                           PROT_READ | PROT_WRITE,

  1099.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1100.                                           -1, 0);

  1101.             c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,

  1102.                                           PROT_READ | PROT_WRITE,

  1103.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1104.                                           -1, 0);

  1105. #elif HAVE_VIRTUALALLOC

  1106.             c->lumMmxextFilterCode = VirtualAlloc(NULL,

  1107.                                                   c->lumMmxextFilterCodeSize,

  1108.                                                   MEM_COMMIT,

  1109.                                                   PAGE_EXECUTE_READWRITE);

  1110.             c->chrMmxextFilterCode = VirtualAlloc(NULL,

  1111.                                                   c->chrMmxextFilterCodeSize,

  1112.                                                   MEM_COMMIT,

  1113.                                                   PAGE_EXECUTE_READWRITE);

  1114. #else

  1115.             c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);

  1116.             c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);

  1117. #endif

  1118. 

  1119.             if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)

  1120.                 return AVERROR(ENOMEM);

  1121.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1122.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1123.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1124.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1125. 

  1126.             init_hscaler_mmxext(dstW, c->lumXInc, c->lumMmxextFilterCode,

  1127.                                 c->hLumFilter, c->hLumFilterPos, 8);

  1128.             init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,

  1129.                                 c->hChrFilter, c->hChrFilterPos, 4);

  1130. 

  1131. #if USE_MMAP

  1132.             mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1133.             mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1134. #endif

  1135.         } else

  1136. #endif /* HAVE_MMXEXT_INLINE */

  1137.         {

  1138.             const int filterAlign = X86_MMX(cpu_flags)     ? 4 :

  1139.                                     PPC_ALTIVEC(cpu_flags) ? 8 : 1;

  1140. 

  1141.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1142.                            &c->hLumFilterSize, c->lumXInc,

  1143.                            srcW, dstW, filterAlign, 1 << 14,

  1144.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1145.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1146.                            c->param, 1) < 0)

  1147.                 goto fail;

  1148.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1149.                            &c->hChrFilterSize, c->chrXInc,

  1150.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1151.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1152.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1153.                            c->param, 1) < 0)

  1154.                 goto fail;

  1155.         }

  1156.     } // initialize horizontal stuff

  1157. 

  1158.     /* precalculate vertical scaler filter coefficients */

  1159.     {

  1160.         const int filterAlign = X86_MMX(cpu_flags)     ? 2 :

  1161.                                 PPC_ALTIVEC(cpu_flags) ? 8 : 1;

  1162. 

  1163.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1164.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

  1165.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1166.                        cpu_flags, srcFilter->lumV, dstFilter->lumV,

  1167.                        c->param, 0) < 0)

  1168.             goto fail;

  1169.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1170.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1171.                        filterAlign, (1 << 12),

  1172.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1173.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1174.                        c->param, 0) < 0)

  1175.             goto fail;

  1176. 

  1177. #if HAVE_ALTIVEC

  1178.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1179.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1180. 

  1181.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1182.             int j;

  1183.             short *p = (short *)&c->vYCoeffsBank[i];

  1184.             for (j = 0; j < 8; j++)

  1185.                 p[j] = c->vLumFilter[i];

  1186.         }

  1187. 

  1188.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1189.             int j;

  1190.             short *p = (short *)&c->vCCoeffsBank[i];

  1191.             for (j = 0; j < 8; j++)

  1192.                 p[j] = c->vChrFilter[i];

  1193.         }

  1194. #endif

  1195.     }

  1196. 

  1197.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1198.     c->vLumBufSize = c->vLumFilterSize;

  1199.     c->vChrBufSize = c->vChrFilterSize;

  1200.     for (i = 0; i < dstH; i++) {

  1201.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1202.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1203.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1204.                                << c->chrSrcVSubSample));

  1205. 

  1206.         nextSlice >>= c->chrSrcVSubSample;

  1207.         nextSlice <<= c->chrSrcVSubSample;

  1208.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1209.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1210.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1211.             (nextSlice >> c->chrSrcVSubSample))

  1212.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1213.                              c->vChrFilterPos[chrI];

  1214.     }

  1215. 

  1216.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1217.      * need to allocate several megabytes to handle all possible cases) */

  1218.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1219.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1220.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1221.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1222.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1223.     /* Note we need at least one pixel more at the end because of the MMX code

  1224.      * (just in case someone wants to replace the 4000/8000). */

  1225.     /* align at 16 bytes for AltiVec */

  1226.     for (i = 0; i < c->vLumBufSize; i++) {

  1227.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1228.                           dst_stride + 16, fail);

  1229.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1230.     }

  1231.     // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)

  1232.     c->uv_off_px   = dst_stride_px + 64 / (c->dstBpc & ~7);

  1233.     c->uv_off_byte = dst_stride + 16;

  1234.     for (i = 0; i < c->vChrBufSize; i++) {

  1235.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1236.                          dst_stride * 2 + 32, fail);

  1237.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1238.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1239.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1240.     }

  1241.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1242.         for (i = 0; i < c->vLumBufSize; i++) {

  1243.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],

  1244.                               dst_stride + 16, fail);

  1245.             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];

  1246.         }

  1247. 

  1248.     // try to avoid drawing green stuff between the right end and the stride end

  1249.     for (i = 0; i < c->vChrBufSize; i++)

  1250.         memset(c->chrUPixBuf[i], 64, dst_stride * 2 + 1);

  1251. 

  1252.     assert(c->chrDstH <= dstH);

  1253. 

  1254.     if (flags & SWS_PRINT_INFO) {

  1255.         if (flags & SWS_FAST_BILINEAR)

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

  1257.         else if (flags & SWS_BILINEAR)

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

  1259.         else if (flags & SWS_BICUBIC)

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

  1261.         else if (flags & SWS_X)

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

  1263.         else if (flags & SWS_POINT)

  1264.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1265.         else if (flags & SWS_AREA)

  1266.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1267.         else if (flags & SWS_BICUBLIN)

  1268.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1269.         else if (flags & SWS_GAUSS)

  1270.             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1271.         else if (flags & SWS_SINC)

  1272.             av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1273.         else if (flags & SWS_LANCZOS)

  1274.             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1275.         else if (flags & SWS_SPLINE)

  1276.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1277.         else

  1278.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1279. 

  1280.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1281.                sws_format_name(srcFormat),

  1282. #ifdef DITHER1XBPP

  1283.                dstFormat == AV_PIX_FMT_BGR555   || dstFormat == AV_PIX_FMT_BGR565   ||

  1284.                dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||

  1285.                dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?

  1286.                                                              "dithered " : "",

  1287. #else

  1288.                "",

  1289. #endif

  1290.                sws_format_name(dstFormat));

  1291. 

  1292.         if (INLINE_MMXEXT(cpu_flags))

  1293.             av_log(c, AV_LOG_INFO, "using MMXEXT\n");

  1294.         else if (INLINE_AMD3DNOW(cpu_flags))

  1295.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1296.         else if (INLINE_MMX(cpu_flags))

  1297.             av_log(c, AV_LOG_INFO, "using MMX\n");

  1298.         else if (PPC_ALTIVEC(cpu_flags))

  1299.             av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1300.         else

  1301.             av_log(c, AV_LOG_INFO, "using C\n");

  1302. 

  1303.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1304.         av_log(c, AV_LOG_DEBUG,

  1305.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1306.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1307.         av_log(c, AV_LOG_DEBUG,

  1308.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1309.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1310.                c->chrXInc, c->chrYInc);

  1311.     }

  1312. 

  1313.     c->swscale = ff_getSwsFunc(c);

  1314.     return 0;

  1315. fail: // FIXME replace things by appropriate error codes

  1316.     return -1;

  1317. }

  1318. 

  1319. #if FF_API_SWS_GETCONTEXT

  1320. SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,

  1321.                            int dstW, int dstH, enum AVPixelFormat dstFormat,

  1322.                            int flags, SwsFilter *srcFilter,

  1323.                            SwsFilter *dstFilter, const double *param)

  1324. {

  1325.     SwsContext *c;

  1326. 

  1327.     if (!(c = sws_alloc_context()))

  1328.         return NULL;

  1329. 

  1330.     c->flags     = flags;

  1331.     c->srcW      = srcW;

  1332.     c->srcH      = srcH;

  1333.     c->dstW      = dstW;

  1334.     c->dstH      = dstH;

  1335.     c->srcRange  = handle_jpeg(&srcFormat);

  1336.     c->dstRange  = handle_jpeg(&dstFormat);

  1337.     c->srcFormat = srcFormat;

  1338.     c->dstFormat = dstFormat;

  1339. 

  1340.     if (param) {

  1341.         c->param[0] = param[0];

  1342.         c->param[1] = param[1];

  1343.     }

  1344.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1345.                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1346.                              c->dstRange, 0, 1 << 16, 1 << 16);

  1347. 

  1348.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1349.         sws_freeContext(c);

  1350.         return NULL;

  1351.     }

  1352. 

  1353.     return c;

  1354. }

  1355. #endif

  1356. 

  1357. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1358.                                 float lumaSharpen, float chromaSharpen,

  1359.                                 float chromaHShift, float chromaVShift,

  1360.                                 int verbose)

  1361. {

  1362.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1363.     if (!filter)

  1364.         return NULL;

  1365. 

  1366.     if (lumaGBlur != 0.0) {

  1367.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1368.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1369.     } else {

  1370.         filter->lumH = sws_getIdentityVec();

  1371.         filter->lumV = sws_getIdentityVec();

  1372.     }

  1373. 

  1374.     if (chromaGBlur != 0.0) {

  1375.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1376.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1377.     } else {

  1378.         filter->chrH = sws_getIdentityVec();

  1379.         filter->chrV = sws_getIdentityVec();

  1380.     }

  1381. 

  1382.     if (chromaSharpen != 0.0) {

  1383.         SwsVector *id = sws_getIdentityVec();

  1384.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1385.         sws_scaleVec(filter->chrV, -chromaSharpen);

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

  1387.         sws_addVec(filter->chrV, id);

  1388.         sws_freeVec(id);

  1389.     }

  1390. 

  1391.     if (lumaSharpen != 0.0) {

  1392.         SwsVector *id = sws_getIdentityVec();

  1393.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1394.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1395.         sws_addVec(filter->lumH, id);

  1396.         sws_addVec(filter->lumV, id);

  1397.         sws_freeVec(id);

  1398.     }

  1399. 

  1400.     if (chromaHShift != 0.0)

  1401.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1402. 

  1403.     if (chromaVShift != 0.0)

  1404.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1405. 

  1406.     sws_normalizeVec(filter->chrH, 1.0);

  1407.     sws_normalizeVec(filter->chrV, 1.0);

  1408.     sws_normalizeVec(filter->lumH, 1.0);

  1409.     sws_normalizeVec(filter->lumV, 1.0);

  1410. 

  1411.     if (verbose)

  1412.         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1413.     if (verbose)

  1414.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1415. 

  1416.     return filter;

  1417. }

  1418. 

  1419. SwsVector *sws_allocVec(int length)

  1420. {

  1421.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1422.     if (!vec)

  1423.         return NULL;

  1424.     vec->length = length;

  1425.     vec->coeff  = av_malloc(sizeof(double) * length);

  1426.     if (!vec->coeff)

  1427.         av_freep(&vec);

  1428.     return vec;

  1429. }

  1430. 

  1431. SwsVector *sws_getGaussianVec(double variance, double quality)

  1432. {

  1433.     const int length = (int)(variance * quality + 0.5) | 1;

  1434.     int i;

  1435.     double middle  = (length - 1) * 0.5;

  1436.     SwsVector *vec = sws_allocVec(length);

  1437. 

  1438.     if (!vec)

  1439.         return NULL;

  1440. 

  1441.     for (i = 0; i < length; i++) {

  1442.         double dist = i - middle;

  1443.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1444.                         sqrt(2 * variance * M_PI);

  1445.     }

  1446. 

  1447.     sws_normalizeVec(vec, 1.0);

  1448. 

  1449.     return vec;

  1450. }

  1451. 

  1452. SwsVector *sws_getConstVec(double c, int length)

  1453. {

  1454.     int i;

  1455.     SwsVector *vec = sws_allocVec(length);

  1456. 

  1457.     if (!vec)

  1458.         return NULL;

  1459. 

  1460.     for (i = 0; i < length; i++)

  1461.         vec->coeff[i] = c;

  1462. 

  1463.     return vec;

  1464. }

  1465. 

  1466. SwsVector *sws_getIdentityVec(void)

  1467. {

  1468.     return sws_getConstVec(1.0, 1);

  1469. }

  1470. 

  1471. static double sws_dcVec(SwsVector *a)

  1472. {

  1473.     int i;

  1474.     double sum = 0;

  1475. 

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

  1477.         sum += a->coeff[i];

  1478. 

  1479.     return sum;

  1480. }

  1481. 

  1482. void sws_scaleVec(SwsVector *a, double scalar)

  1483. {

  1484.     int i;

  1485. 

  1486.     for (i = 0; i < a->length; i++)

  1487.         a->coeff[i] *= scalar;

  1488. }

  1489. 

  1490. void sws_normalizeVec(SwsVector *a, double height)

  1491. {

  1492.     sws_scaleVec(a, height / sws_dcVec(a));

  1493. }

  1494. 

  1495. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1496. {

  1497.     int length = a->length + b->length - 1;

  1498.     int i, j;

  1499.     SwsVector *vec = sws_getConstVec(0.0, length);

  1500. 

  1501.     if (!vec)

  1502.         return NULL;

  1503. 

  1504.     for (i = 0; i < a->length; i++) {

  1505.         for (j = 0; j < b->length; j++) {

  1506.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1507.         }

  1508.     }

  1509. 

  1510.     return vec;

  1511. }

  1512. 

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

  1514. {

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

  1516.     int i;

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

  1518. 

  1519.     if (!vec)

  1520.         return NULL;

  1521. 

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

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

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

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

  1526. 

  1527.     return vec;

  1528. }

  1529. 

  1530. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1531. {

  1532.     int length = FFMAX(a->length, b->length);

  1533.     int i;

  1534.     SwsVector *vec = sws_getConstVec(0.0, length);

  1535. 

  1536.     if (!vec)

  1537.         return NULL;

  1538. 

  1539.     for (i = 0; i < a->length; i++)

  1540.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1541.     for (i = 0; i < b->length; i++)

  1542.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

  1543. 

  1544.     return vec;

  1545. }

  1546. 

  1547. /* shift left / or right if "shift" is negative */

  1548. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1549. {

  1550.     int length = a->length + FFABS(shift) * 2;

  1551.     int i;

  1552.     SwsVector *vec = sws_getConstVec(0.0, length);

  1553. 

  1554.     if (!vec)

  1555.         return NULL;

  1556. 

  1557.     for (i = 0; i < a->length; i++) {

  1558.         vec->coeff[i + (length    - 1) / 2 -

  1559.                        (a->length - 1) / 2 - shift] = a->coeff[i];

  1560.     }

  1561. 

  1562.     return vec;

  1563. }

  1564. 

  1565. void sws_shiftVec(SwsVector *a, int shift)

  1566. {

  1567.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1568.     av_free(a->coeff);

  1569.     a->coeff  = shifted->coeff;

  1570.     a->length = shifted->length;

  1571.     av_free(shifted);

  1572. }

  1573. 

  1574. void sws_addVec(SwsVector *a, SwsVector *b)

  1575. {

  1576.     SwsVector *sum = sws_sumVec(a, b);

  1577.     av_free(a->coeff);

  1578.     a->coeff  = sum->coeff;

  1579.     a->length = sum->length;

  1580.     av_free(sum);

  1581. }

  1582. 

  1583. void sws_subVec(SwsVector *a, SwsVector *b)

  1584. {

  1585.     SwsVector *diff = sws_diffVec(a, b);

  1586.     av_free(a->coeff);

  1587.     a->coeff  = diff->coeff;

  1588.     a->length = diff->length;

  1589.     av_free(diff);

  1590. }

  1591. 

  1592. void sws_convVec(SwsVector *a, SwsVector *b)

  1593. {

  1594.     SwsVector *conv = sws_getConvVec(a, b);

  1595.     av_free(a->coeff);

  1596.     a->coeff  = conv->coeff;

  1597.     a->length = conv->length;

  1598.     av_free(conv);

  1599. }

  1600. 

  1601. SwsVector *sws_cloneVec(SwsVector *a)

  1602. {

  1603.     int i;

  1604.     SwsVector *vec = sws_allocVec(a->length);

  1605. 

  1606.     if (!vec)

  1607.         return NULL;

  1608. 

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

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

  1611. 

  1612.     return vec;

  1613. }

  1614. 

  1615. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1616. {

  1617.     int i;

  1618.     double max = 0;

  1619.     double min = 0;

  1620.     double range;

  1621. 

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

  1623.         if (a->coeff[i] > max)

  1624.             max = a->coeff[i];

  1625. 

  1626.     for (i = 0; i < a->length; i++)

  1627.         if (a->coeff[i] < min)

  1628.             min = a->coeff[i];

  1629. 

  1630.     range = max - min;

  1631. 

  1632.     for (i = 0; i < a->length; i++) {

  1633.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1634.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1635.         for (; x > 0; x--)

  1636.             av_log(log_ctx, log_level, " ");

  1637.         av_log(log_ctx, log_level, "|\n");

  1638.     }

  1639. }

  1640. 

  1641. void sws_freeVec(SwsVector *a)

  1642. {

  1643.     if (!a)

  1644.         return;

  1645.     av_freep(&a->coeff);

  1646.     a->length = 0;

  1647.     av_free(a);

  1648. }

  1649. 

  1650. void sws_freeFilter(SwsFilter *filter)

  1651. {

  1652.     if (!filter)

  1653.         return;

  1654. 

  1655.     if (filter->lumH)

  1656.         sws_freeVec(filter->lumH);

  1657.     if (filter->lumV)

  1658.         sws_freeVec(filter->lumV);

  1659.     if (filter->chrH)

  1660.         sws_freeVec(filter->chrH);

  1661.     if (filter->chrV)

  1662.         sws_freeVec(filter->chrV);

  1663.     av_free(filter);

  1664. }

  1665. 

  1666. void sws_freeContext(SwsContext *c)

  1667. {

  1668.     int i;

  1669.     if (!c)

  1670.         return;

  1671. 

  1672.     if (c->lumPixBuf) {

  1673.         for (i = 0; i < c->vLumBufSize; i++)

  1674.             av_freep(&c->lumPixBuf[i]);

  1675.         av_freep(&c->lumPixBuf);

  1676.     }

  1677. 

  1678.     if (c->chrUPixBuf) {

  1679.         for (i = 0; i < c->vChrBufSize; i++)

  1680.             av_freep(&c->chrUPixBuf[i]);

  1681.         av_freep(&c->chrUPixBuf);

  1682.         av_freep(&c->chrVPixBuf);

  1683.     }

  1684. 

  1685.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1686.         for (i = 0; i < c->vLumBufSize; i++)

  1687.             av_freep(&c->alpPixBuf[i]);

  1688.         av_freep(&c->alpPixBuf);

  1689.     }

  1690. 

  1691.     av_freep(&c->vLumFilter);

  1692.     av_freep(&c->vChrFilter);

  1693.     av_freep(&c->hLumFilter);

  1694.     av_freep(&c->hChrFilter);

  1695. #if HAVE_ALTIVEC

  1696.     av_freep(&c->vYCoeffsBank);

  1697.     av_freep(&c->vCCoeffsBank);

  1698. #endif

  1699. 

  1700.     av_freep(&c->vLumFilterPos);

  1701.     av_freep(&c->vChrFilterPos);

  1702.     av_freep(&c->hLumFilterPos);

  1703.     av_freep(&c->hChrFilterPos);

  1704. 

  1705. #if HAVE_MMX_INLINE

  1706. #if USE_MMAP

  1707.     if (c->lumMmxextFilterCode)

  1708.         munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);

  1709.     if (c->chrMmxextFilterCode)

  1710.         munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);

  1711. #elif HAVE_VIRTUALALLOC

  1712.     if (c->lumMmxextFilterCode)

  1713.         VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);

  1714.     if (c->chrMmxextFilterCode)

  1715.         VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);

  1716. #else

  1717.     av_free(c->lumMmxextFilterCode);

  1718.     av_free(c->chrMmxextFilterCode);

  1719. #endif

  1720.     c->lumMmxextFilterCode = NULL;

  1721.     c->chrMmxextFilterCode = NULL;

  1722. #endif /* HAVE_MMX_INLINE */

  1723. 

  1724.     av_freep(&c->yuvTable);

  1725.     av_free(c->formatConvBuffer);

  1726. 

  1727.     av_free(c);

  1728. }

  1729. 

  1730. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  1731.                                         int srcH, enum AVPixelFormat srcFormat,

  1732.                                         int dstW, int dstH,

  1733.                                         enum AVPixelFormat dstFormat, int flags,

  1734.                                         SwsFilter *srcFilter,

  1735.                                         SwsFilter *dstFilter,

  1736.                                         const double *param)

  1737. {

  1738.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  1739.                                              SWS_PARAM_DEFAULT };

  1740. 

  1741.     if (!param)

  1742.         param = default_param;

  1743. 

  1744.     if (context &&

  1745.         (context->srcW      != srcW      ||

  1746.          context->srcH      != srcH      ||

  1747.          context->srcFormat != srcFormat ||

  1748.          context->dstW      != dstW      ||

  1749.          context->dstH      != dstH      ||

  1750.          context->dstFormat != dstFormat ||

  1751.          context->flags     != flags     ||

  1752.          context->param[0]  != param[0]  ||

  1753.          context->param[1]  != param[1])) {

  1754.         sws_freeContext(context);

  1755.         context = NULL;

  1756.     }

  1757. 

  1758.     if (!context) {

  1759.         if (!(context = sws_alloc_context()))

  1760.             return NULL;

  1761.         context->srcW      = srcW;

  1762.         context->srcH      = srcH;

  1763.         context->srcRange  = handle_jpeg(&srcFormat);

  1764.         context->srcFormat = srcFormat;

  1765.         context->dstW      = dstW;

  1766.         context->dstH      = dstH;

  1767.         context->dstRange  = handle_jpeg(&dstFormat);

  1768.         context->dstFormat = dstFormat;

  1769.         context->flags     = flags;

  1770.         context->param[0]  = param[0];

  1771.         context->param[1]  = param[1];

  1772.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  1773.                                  context->srcRange,

  1774.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1775.                                  context->dstRange, 0, 1 << 16, 1 << 16);

  1776.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1777.             sws_freeContext(context);

  1778.             return NULL;

  1779.         }

  1780.     }

  1781.     return context;

  1782. }