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.
37278 Commits
32 Branches
411MB
Tree: c1eb3e7fec
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c1eb3e7fec'
${ noResults }
FFmpeg/libswscale/utils.c

1771 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. };

  176. 

  177. int sws_isSupportedInput(enum AVPixelFormat pix_fmt)

  178. {

  179.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  180.            format_entries[pix_fmt].is_supported_in : 0;

  181. }

  182. 

  183. int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)

  184. {

  185.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  186.            format_entries[pix_fmt].is_supported_out : 0;

  187. }

  188. 

  189. int sws_isSupportedEndiannessConversion(enum AVPixelFormat pix_fmt)

  190. {

  191.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  192.            format_entries[pix_fmt].is_supported_endianness : 0;

  193. }

  194. 

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

  196. 

  197. const char *sws_format_name(enum AVPixelFormat format)

  198. {

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

  200.     if (desc)

  201.         return desc->name;

  202.     else

  203.         return "Unknown format";

  204. }

  205. 

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

  207.                              double dist)

  208. {

  209.     if (dist <= 1.0)

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

  211.     else

  212.         return getSplineCoeff(0.0,

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

  214.                                c + 3.0 * d,

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

  216.                               dist - 1.0);

  217. }

  218. 

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

  220.                               int *outFilterSize, int xInc, int srcW,

  221.                               int dstW, int filterAlign, int one,

  222.                               int flags, int cpu_flags,

  223.                               SwsVector *srcFilter, SwsVector *dstFilter,

  224.                               double param[2], int is_horizontal)

  225. {

  226.     int i;

  227.     int filterSize;

  228.     int filter2Size;

  229.     int minFilterSize;

  230.     int64_t *filter    = NULL;

  231.     int64_t *filter2   = NULL;

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

  233.     int ret            = -1;

  234. 

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

  236. 

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

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

  239. 

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

  241.         int i;

  242.         filterSize = 1;

  243.         FF_ALLOCZ_OR_GOTO(NULL, filter,

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

  245. 

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

  247.             filter[i * filterSize] = fone;

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

  249.         }

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

  251.         int i;

  252.         int xDstInSrc;

  253.         filterSize = 1;

  254.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  256. 

  257.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  260. 

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

  262.             filter[i]       = fone;

  263.             xDstInSrc      += xInc;

  264.         }

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

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

  267.         int i;

  268.         int xDstInSrc;

  269.         filterSize = 2;

  270.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  272. 

  273.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  276.             int j;

  277. 

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

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

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

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

  282.                                 (fone >> 16);

  283.                 if (coeff < 0)

  284.                     coeff = 0;

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

  286.                 xx++;

  287.             }

  288.             xDstInSrc += xInc;

  289.         }

  290.     } else {

  291.         int64_t xDstInSrc;

  292.         int sizeFactor;

  293. 

  294.         if (flags & SWS_BICUBIC)

  295.             sizeFactor = 4;

  296.         else if (flags & SWS_X)

  297.             sizeFactor = 8;

  298.         else if (flags & SWS_AREA)

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

  300.         else if (flags & SWS_GAUSS)

  301.             sizeFactor = 8;     // infinite ;)

  302.         else if (flags & SWS_LANCZOS)

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

  304.         else if (flags & SWS_SINC)

  305.             sizeFactor = 20;    // infinite ;)

  306.         else if (flags & SWS_SPLINE)

  307.             sizeFactor = 20;    // infinite ;)

  308.         else if (flags & SWS_BILINEAR)

  309.             sizeFactor = 2;

  310.         else {

  311.             sizeFactor = 0;     // GCC warning killer

  312.             assert(0);

  313.         }

  314. 

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

  316.             filterSize = 1 + sizeFactor;    // upscale

  317.         else

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

  319. 

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

  321.         filterSize = FFMAX(filterSize, 1);

  322. 

  323.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  325. 

  326.         xDstInSrc = xInc - 0x10000;

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

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

  329.             int j;

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

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

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

  333.                 double floatd;

  334.                 int64_t coeff;

  335. 

  336.                 if (xInc > 1 << 16)

  337.                     d = d * dstW / srcW;

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

  339. 

  340.                 if (flags & SWS_BICUBIC) {

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

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

  343. 

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

  345.                         coeff = 0.0;

  346.                     } else {

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

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

  349. 

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

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

  352.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  353.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  354.                         else

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

  356.                                       (6 * B + 30 * C) * dd  +

  357.                                     (-12 * B - 48 * C) * d   +

  358.                                       (8 * B + 24 * C) * (1 << 30);

  359.                     }

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

  361.                 }

  362. #if 0

  363.                 else if (flags & SWS_X) {

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

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

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

  367.                 }

  368. #endif

  369.                 else if (flags & SWS_X) {

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

  371.                     double c;

  372. 

  373.                     if (floatd < 1.0)

  374.                         c = cos(floatd * M_PI);

  375.                     else

  376.                         c = -1.0;

  377.                     if (c < 0.0)

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

  379.                     else

  380.                         c = pow(c, A);

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

  382.                 } else if (flags & SWS_AREA) {

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

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

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

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

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

  388.                     else

  389.                         coeff = 0.0;

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

  391.                 } else if (flags & SWS_GAUSS) {

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

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

  394.                 } else if (flags & SWS_SINC) {

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

  396.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  400.                     if (floatd > p)

  401.                         coeff = 0;

  402.                 } else if (flags & SWS_BILINEAR) {

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

  404.                     if (coeff < 0)

  405.                         coeff = 0;

  406.                     coeff *= fone >> 30;

  407.                 } else if (flags & SWS_SPLINE) {

  408.                     double p = -2.196152422706632;

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

  410.                 } else {

  411.                     coeff = 0.0; // GCC warning killer

  412.                     assert(0);

  413.                 }

  414. 

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

  416.                 xx++;

  417.             }

  418.             xDstInSrc += 2 * xInc;

  419.         }

  420.     }

  421. 

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

  423.      * av_free(filter);

  424.      */

  425.     assert(filterSize > 0);

  426.     filter2Size = filterSize;

  427.     if (srcFilter)

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

  429.     if (dstFilter)

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

  431.     assert(filter2Size > 0);

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

  433. 

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

  435.         int j, k;

  436. 

  437.         if (srcFilter) {

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

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

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

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

  442.             }

  443.         } else {

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

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

  446.         }

  447.         // FIXME dstFilter

  448. 

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

  450.     }

  451.     av_freep(&filter);

  452. 

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

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

  455.     minFilterSize = 0;

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

  457.         int min = filter2Size;

  458.         int j;

  459.         int64_t cutOff = 0.0;

  460. 

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

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

  463.             int k;

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

  465. 

  466.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  467.                 break;

  468. 

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

  470.              * filter otherwise */

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

  472.                 break;

  473. 

  474.             // move filter coefficients left

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

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

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

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

  479.         }

  480. 

  481.         cutOff = 0;

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

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

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

  485. 

  486.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  487.                 break;

  488.             min--;

  489.         }

  490. 

  491.         if (min > minFilterSize)

  492.             minFilterSize = min;

  493.     }

  494. 

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

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

  497.         if (minFilterSize < 5)

  498.             filterAlign = 4;

  499. 

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

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

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

  503.         if (minFilterSize < 3)

  504.             filterAlign = 1;

  505.     }

  506. 

  507.     if (INLINE_MMX(cpu_flags)) {

  508.         // special case for unscaled vertical filtering

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

  510.             filterAlign = 1;

  511.     }

  512. 

  513.     assert(minFilterSize > 0);

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

  515.     assert(filterSize > 0);

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

  517.     if (filterSize >= MAX_FILTER_SIZE * 16 /

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

  519.         goto fail;

  520.     *outFilterSize = filterSize;

  521. 

  522.     if (flags & SWS_PRINT_INFO)

  523.         av_log(NULL, AV_LOG_VERBOSE,

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

  525.                filter2Size, filterSize);

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

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

  528.         int j;

  529. 

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

  531.             if (j >= filter2Size)

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

  533.             else

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

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

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

  537.         }

  538.     }

  539. 

  540.     // FIXME try to align filterPos if possible

  541. 

  542.     // fix borders

  543.     if (is_horizontal) {

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

  545.             int j;

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

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

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

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

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

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

  552.                 }

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

  554.             }

  555. 

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

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

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

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

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

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

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

  563.                 }

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

  565.             }

  566.         }

  567.     }

  568. 

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

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

  571.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

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

  573. 

  574.     /* normalize & store in outFilter */

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

  576.         int j;

  577.         int64_t error = 0;

  578.         int64_t sum   = 0;

  579. 

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

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

  582.         }

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

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

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

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

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

  588.             error                                  = v - intV * sum;

  589.         }

  590.     }

  591. 

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

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

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

  595.                                                       * read over the end */

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

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

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

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

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

  601.     }

  602. 

  603.     ret = 0;

  604. 

  605. fail:

  606.     av_free(filter);

  607.     av_free(filter2);

  608.     return ret;

  609. }

  610. 

  611. #if HAVE_MMXEXT_INLINE

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

  613.                                        int16_t *filter, int32_t *filterPos,

  614.                                        int numSplits)

  615. {

  616.     uint8_t *fragmentA;

  617.     x86_reg imm8OfPShufW1A;

  618.     x86_reg imm8OfPShufW2A;

  619.     x86_reg fragmentLengthA;

  620.     uint8_t *fragmentB;

  621.     x86_reg imm8OfPShufW1B;

  622.     x86_reg imm8OfPShufW2B;

  623.     x86_reg fragmentLengthB;

  624.     int fragmentPos;

  625. 

  626.     int xpos, i;

  627. 

  628.     // create an optimized horizontal scaling routine

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

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

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

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

  633.      */

  634. 

  635.     // code fragment

  636. 

  637.     __asm__ volatile (

  638.         "jmp                         9f                 \n\t"

  639.         // Begin

  640.         "0:                                             \n\t"

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

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

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

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

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

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

  647.         "1:                                             \n\t"

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

  649.         "2:                                             \n\t"

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

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

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

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

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

  655. 

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

  657. 

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

  659.         // End

  660.         "9:                                             \n\t"

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

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

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

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

  665.         "dec                         %1                 \n\t"

  666.         "dec                         %2                 \n\t"

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

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

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

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

  671. 

  672. 

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

  674.           "=r" (fragmentLengthA)

  675.         );

  676. 

  677.     __asm__ volatile (

  678.         "jmp                         9f                 \n\t"

  679.         // Begin

  680.         "0:                                             \n\t"

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

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

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

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

  685.         "1:                                             \n\t"

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

  687.         "2:                                             \n\t"

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

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

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

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

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

  693. 

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

  695. 

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

  697.         // End

  698.         "9:                                             \n\t"

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

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

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

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

  703.         "dec                         %1                 \n\t"

  704.         "dec                         %2                 \n\t"

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

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

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

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

  709. 

  710. 

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

  712.           "=r" (fragmentLengthB)

  713.         );

  714. 

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

  716.     fragmentPos = 0;

  717. 

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

  719.         int xx = xpos >> 16;

  720. 

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

  722.             int a                  = 0;

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

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

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

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

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

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

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

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

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

  732.             int shift              = 0;

  733. 

  734.             if (filterCode) {

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

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

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

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

  739.                 filterPos[i / 2] = xx;

  740. 

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

  742. 

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

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

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

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

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

  748.                                                                (c << 4) |

  749.                                                                (d << 6);

  750. 

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

  752.                     shift = maxShift;               // avoid overread

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

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

  755. 

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

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

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

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

  760.                 }

  761.             }

  762. 

  763.             fragmentPos += fragmentLength;

  764. 

  765.             if (filterCode)

  766.                 filterCode[fragmentPos] = RET;

  767.         }

  768.         xpos += xInc;

  769.     }

  770.     if (filterCode)

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

  772. 

  773.     return fragmentPos + 1;

  774. }

  775. #endif /* HAVE_MMXEXT_INLINE */

  776. 

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

  778. {

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

  780.     *h = desc->log2_chroma_w;

  781.     *v = desc->log2_chroma_h;

  782. }

  783. 

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

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

  786.                              int brightness, int contrast, int saturation)

  787. {

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

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

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

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

  792. 

  793.     c->brightness = brightness;

  794.     c->contrast   = contrast;

  795.     c->saturation = saturation;

  796.     c->srcRange   = srcRange;

  797.     c->dstRange   = dstRange;

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

  799.         return -1;

  800. 

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

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

  803. 

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

  805.                              contrast, saturation);

  806.     // FIXME factorize

  807. 

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

  809.         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,

  810.                                        contrast, saturation);

  811.     return 0;

  812. }

  813. 

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

  815.                              int *srcRange, int **table, int *dstRange,

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

  817. {

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

  819.         return -1;

  820. 

  821.     *inv_table  = c->srcColorspaceTable;

  822.     *table      = c->dstColorspaceTable;

  823.     *srcRange   = c->srcRange;

  824.     *dstRange   = c->dstRange;

  825.     *brightness = c->brightness;

  826.     *contrast   = c->contrast;

  827.     *saturation = c->saturation;

  828. 

  829.     return 0;

  830. }

  831. 

  832. static int handle_jpeg(enum AVPixelFormat *format)

  833. {

  834.     switch (*format) {

  835.     case AV_PIX_FMT_YUVJ420P:

  836.         *format = AV_PIX_FMT_YUV420P;

  837.         return 1;

  838.     case AV_PIX_FMT_YUVJ422P:

  839.         *format = AV_PIX_FMT_YUV422P;

  840.         return 1;

  841.     case AV_PIX_FMT_YUVJ444P:

  842.         *format = AV_PIX_FMT_YUV444P;

  843.         return 1;

  844.     case AV_PIX_FMT_YUVJ440P:

  845.         *format = AV_PIX_FMT_YUV440P;

  846.         return 1;

  847.     default:

  848.         return 0;

  849.     }

  850. }

  851. 

  852. SwsContext *sws_alloc_context(void)

  853. {

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

  855. 

  856.     if (c) {

  857.         c->av_class = &sws_context_class;

  858.         av_opt_set_defaults(c);

  859.     }

  860. 

  861.     return c;

  862. }

  863. 

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

  865.                              SwsFilter *dstFilter)

  866. {

  867.     int i;

  868.     int usesVFilter, usesHFilter;

  869.     int unscaled;

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

  871.     int srcW              = c->srcW;

  872.     int srcH              = c->srcH;

  873.     int dstW              = c->dstW;

  874.     int dstH              = c->dstH;

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

  876.     int dst_stride_px     = dst_stride >> 1;

  877.     int flags, cpu_flags;

  878.     enum AVPixelFormat srcFormat = c->srcFormat;

  879.     enum AVPixelFormat dstFormat = c->dstFormat;

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

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

  882. 

  883.     cpu_flags = av_get_cpu_flags();

  884.     flags     = c->flags;

  885.     emms_c();

  886.     if (!rgb15to16)

  887.         sws_rgb2rgb_init();

  888. 

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

  890. 

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

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

  893.     if (!sws_isSupportedInput(srcFormat)) {

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

  895.                sws_format_name(srcFormat));

  896.         return AVERROR(EINVAL);

  897.     }

  898.     if (!sws_isSupportedOutput(dstFormat)) {

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

  900.                sws_format_name(dstFormat));

  901.         return AVERROR(EINVAL);

  902.     }

  903.     }

  904. 

  905.     i = flags & (SWS_POINT         |

  906.                  SWS_AREA          |

  907.                  SWS_BILINEAR      |

  908.                  SWS_FAST_BILINEAR |

  909.                  SWS_BICUBIC       |

  910.                  SWS_X             |

  911.                  SWS_GAUSS         |

  912.                  SWS_LANCZOS       |

  913.                  SWS_SINC          |

  914.                  SWS_SPLINE        |

  915.                  SWS_BICUBLIN);

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

  917.         av_log(c, AV_LOG_ERROR,

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

  919.         return AVERROR(EINVAL);

  920.     }

  921.     /* sanity check */

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

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

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

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

  926.                srcW, srcH, dstW, dstH);

  927.         return AVERROR(EINVAL);

  928.     }

  929. 

  930.     if (!dstFilter)

  931.         dstFilter = &dummyFilter;

  932.     if (!srcFilter)

  933.         srcFilter = &dummyFilter;

  934. 

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

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

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

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

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

  940. 

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

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

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

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

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

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

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

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

  949. 

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

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

  952. 

  953.     if (isPlanarRGB(dstFormat)) {

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

  955.             av_log(c, AV_LOG_DEBUG,

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

  957.                    av_get_pix_fmt_name(dstFormat));

  958.             flags   |= SWS_FULL_CHR_H_INT;

  959.             c->flags = flags;

  960.         }

  961.     }

  962. 

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

  964.      * chroma interpolation */

  965.     if (flags & SWS_FULL_CHR_H_INT &&

  966.         isAnyRGB(dstFormat)        &&

  967.         !isPlanarRGB(dstFormat)    &&

  968.         dstFormat != AV_PIX_FMT_RGBA  &&

  969.         dstFormat != AV_PIX_FMT_ARGB  &&

  970.         dstFormat != AV_PIX_FMT_BGRA  &&

  971.         dstFormat != AV_PIX_FMT_ABGR  &&

  972.         dstFormat != AV_PIX_FMT_RGB24 &&

  973.         dstFormat != AV_PIX_FMT_BGR24) {

  974.         av_log(c, AV_LOG_ERROR,

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

  976.                sws_format_name(dstFormat));

  977.         flags   &= ~SWS_FULL_CHR_H_INT;

  978.         c->flags = flags;

  979.     }

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

  981.         c->chrDstHSubSample = 1;

  982. 

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

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

  985.                            SWS_SRC_V_CHR_DROP_SHIFT;

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

  987. 

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

  989.      * wants full chroma */

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

  991.         srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&

  992.         srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&

  993.         srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&

  994.         srcFormat != AV_PIX_FMT_GBRP9BE   && srcFormat != AV_PIX_FMT_GBRP9LE  &&

  995.         srcFormat != AV_PIX_FMT_GBRP10BE  && srcFormat != AV_PIX_FMT_GBRP10LE &&

  996.         srcFormat != AV_PIX_FMT_GBRP16BE  && srcFormat != AV_PIX_FMT_GBRP16LE &&

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

  998.          (flags & SWS_FAST_BILINEAR)))

  999.         c->chrSrcHSubSample = 1;

  1000. 

  1001.     // Note the -((-x)>>y) is so that we always round toward +inf.

  1002.     c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);

  1003.     c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);

  1004.     c->chrDstW = -((-dstW) >> c->chrDstHSubSample);

  1005.     c->chrDstH = -((-dstH) >> c->chrDstVSubSample);

  1006. 

  1007.     /* unscaled special cases */

  1008.     if (unscaled && !usesHFilter && !usesVFilter &&

  1009.         (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {

  1010.         ff_get_unscaled_swscale(c);

  1011. 

  1012.         if (c->swScale) {

  1013.             if (flags & SWS_PRINT_INFO)

  1014.                 av_log(c, AV_LOG_INFO,

  1015.                        "using unscaled %s -> %s special converter\n",

  1016.                        sws_format_name(srcFormat), sws_format_name(dstFormat));

  1017.             return 0;

  1018.         }

  1019.     }

  1020. 

  1021.     c->srcBpc = 1 + desc_src->comp[0].depth_minus1;

  1022.     if (c->srcBpc < 8)

  1023.         c->srcBpc = 8;

  1024.     c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;

  1025.     if (c->dstBpc < 8)

  1026.         c->dstBpc = 8;

  1027.     if (c->dstBpc == 16)

  1028.         dst_stride <<= 1;

  1029.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,

  1030.                      (FFALIGN(srcW, 16) * 2 * FFALIGN(c->srcBpc, 8) >> 3) + 16,

  1031.                      fail);

  1032.     if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 10) {

  1033.         c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&

  1034.                               (srcW & 15) == 0) ? 1 : 0;

  1035.         if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0

  1036.             && (flags & SWS_FAST_BILINEAR)) {

  1037.             if (flags & SWS_PRINT_INFO)

  1038.                 av_log(c, AV_LOG_INFO,

  1039.                        "output width is not a multiple of 32 -> no MMXEXT scaler\n");

  1040.         }

  1041.         if (usesHFilter)

  1042.             c->canMMXEXTBeUsed = 0;

  1043.     } else

  1044.         c->canMMXEXTBeUsed = 0;

  1045. 

  1046.     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;

  1047.     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

  1048. 

  1049.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  1050.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  1051.      * correct scaling.

  1052.      * n-2 is the last chrominance sample available.

  1053.      * This is not perfect, but no one should notice the difference, the more

  1054.      * correct variant would be like the vertical one, but that would require

  1055.      * some special code for the first and last pixel */

  1056.     if (flags & SWS_FAST_BILINEAR) {

  1057.         if (c->canMMXEXTBeUsed) {

  1058.             c->lumXInc += 20;

  1059.             c->chrXInc += 20;

  1060.         }

  1061.         // we don't use the x86 asm scaler if MMX is available

  1062.         else if (INLINE_MMX(cpu_flags)) {

  1063.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1064.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1065.         }

  1066.     }

  1067. 

  1068. #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)

  1069. 

  1070.     /* precalculate horizontal scaler filter coefficients */

  1071.     {

  1072. #if HAVE_MMXEXT_INLINE

  1073. // can't downscale !!!

  1074.         if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {

  1075.             c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,

  1076.                                                              NULL, NULL, 8);

  1077.             c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,

  1078.                                                              NULL, NULL, NULL, 4);

  1079. 

  1080. #if USE_MMAP

  1081.             c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,

  1082.                                           PROT_READ | PROT_WRITE,

  1083.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1084.                                           -1, 0);

  1085.             c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,

  1086.                                           PROT_READ | PROT_WRITE,

  1087.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1088.                                           -1, 0);

  1089. #elif HAVE_VIRTUALALLOC

  1090.             c->lumMmxextFilterCode = VirtualAlloc(NULL,

  1091.                                                   c->lumMmxextFilterCodeSize,

  1092.                                                   MEM_COMMIT,

  1093.                                                   PAGE_EXECUTE_READWRITE);

  1094.             c->chrMmxextFilterCode = VirtualAlloc(NULL,

  1095.                                                   c->chrMmxextFilterCodeSize,

  1096.                                                   MEM_COMMIT,

  1097.                                                   PAGE_EXECUTE_READWRITE);

  1098. #else

  1099.             c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);

  1100.             c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);

  1101. #endif

  1102. 

  1103.             if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)

  1104.                 return AVERROR(ENOMEM);

  1105.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1106.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1107.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1108.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1109. 

  1110.             init_hscaler_mmxext(dstW, c->lumXInc, c->lumMmxextFilterCode,

  1111.                                 c->hLumFilter, c->hLumFilterPos, 8);

  1112.             init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,

  1113.                                 c->hChrFilter, c->hChrFilterPos, 4);

  1114. 

  1115. #if USE_MMAP

  1116.             mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1117.             mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1118. #endif

  1119.         } else

  1120. #endif /* HAVE_MMXEXT_INLINE */

  1121.         {

  1122.             const int filterAlign =

  1123.                 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :

  1124.                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1125.                 1;

  1126. 

  1127.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1128.                            &c->hLumFilterSize, c->lumXInc,

  1129.                            srcW, dstW, filterAlign, 1 << 14,

  1130.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1131.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1132.                            c->param, 1) < 0)

  1133.                 goto fail;

  1134.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1135.                            &c->hChrFilterSize, c->chrXInc,

  1136.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1137.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1138.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1139.                            c->param, 1) < 0)

  1140.                 goto fail;

  1141.         }

  1142.     } // initialize horizontal stuff

  1143. 

  1144.     /* precalculate vertical scaler filter coefficients */

  1145.     {

  1146.         const int filterAlign =

  1147.             (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :

  1148.             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1149.             1;

  1150. 

  1151.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1152.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

  1153.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1154.                        cpu_flags, srcFilter->lumV, dstFilter->lumV,

  1155.                        c->param, 0) < 0)

  1156.             goto fail;

  1157.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1158.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1159.                        filterAlign, (1 << 12),

  1160.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1161.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1162.                        c->param, 0) < 0)

  1163.             goto fail;

  1164. 

  1165. #if HAVE_ALTIVEC

  1166.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1167.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1168. 

  1169.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1170.             int j;

  1171.             short *p = (short *)&c->vYCoeffsBank[i];

  1172.             for (j = 0; j < 8; j++)

  1173.                 p[j] = c->vLumFilter[i];

  1174.         }

  1175. 

  1176.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1177.             int j;

  1178.             short *p = (short *)&c->vCCoeffsBank[i];

  1179.             for (j = 0; j < 8; j++)

  1180.                 p[j] = c->vChrFilter[i];

  1181.         }

  1182. #endif

  1183.     }

  1184. 

  1185.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1186.     c->vLumBufSize = c->vLumFilterSize;

  1187.     c->vChrBufSize = c->vChrFilterSize;

  1188.     for (i = 0; i < dstH; i++) {

  1189.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1190.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1191.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1192.                                << c->chrSrcVSubSample));

  1193. 

  1194.         nextSlice >>= c->chrSrcVSubSample;

  1195.         nextSlice <<= c->chrSrcVSubSample;

  1196.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1197.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1198.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1199.             (nextSlice >> c->chrSrcVSubSample))

  1200.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1201.                              c->vChrFilterPos[chrI];

  1202.     }

  1203. 

  1204.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1205.      * need to allocate several megabytes to handle all possible cases) */

  1206.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1207.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1208.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1209.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1210.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1211.     /* Note we need at least one pixel more at the end because of the MMX code

  1212.      * (just in case someone wants to replace the 4000/8000). */

  1213.     /* align at 16 bytes for AltiVec */

  1214.     for (i = 0; i < c->vLumBufSize; i++) {

  1215.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1216.                           dst_stride + 16, fail);

  1217.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1218.     }

  1219.     // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)

  1220.     c->uv_off_px   = dst_stride_px + 64 / (c->dstBpc & ~7);

  1221.     c->uv_off_byte = dst_stride + 16;

  1222.     for (i = 0; i < c->vChrBufSize; i++) {

  1223.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1224.                          dst_stride * 2 + 32, fail);

  1225.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1226.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1227.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1228.     }

  1229.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1230.         for (i = 0; i < c->vLumBufSize; i++) {

  1231.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],

  1232.                               dst_stride + 16, fail);

  1233.             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];

  1234.         }

  1235. 

  1236.     // try to avoid drawing green stuff between the right end and the stride end

  1237.     for (i = 0; i < c->vChrBufSize; i++)

  1238.         memset(c->chrUPixBuf[i], 64, dst_stride * 2 + 1);

  1239. 

  1240.     assert(c->chrDstH <= dstH);

  1241. 

  1242.     if (flags & SWS_PRINT_INFO) {

  1243.         if (flags & SWS_FAST_BILINEAR)

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

  1245.         else if (flags & SWS_BILINEAR)

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

  1247.         else if (flags & SWS_BICUBIC)

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

  1249.         else if (flags & SWS_X)

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

  1251.         else if (flags & SWS_POINT)

  1252.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1253.         else if (flags & SWS_AREA)

  1254.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1255.         else if (flags & SWS_BICUBLIN)

  1256.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1257.         else if (flags & SWS_GAUSS)

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

  1259.         else if (flags & SWS_SINC)

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

  1261.         else if (flags & SWS_LANCZOS)

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

  1263.         else if (flags & SWS_SPLINE)

  1264.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1265.         else

  1266.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1267. 

  1268.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1269.                sws_format_name(srcFormat),

  1270. #ifdef DITHER1XBPP

  1271.                dstFormat == AV_PIX_FMT_BGR555   || dstFormat == AV_PIX_FMT_BGR565   ||

  1272.                dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||

  1273.                dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?

  1274.                                                              "dithered " : "",

  1275. #else

  1276.                "",

  1277. #endif

  1278.                sws_format_name(dstFormat));

  1279. 

  1280.         if (INLINE_MMXEXT(cpu_flags))

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

  1282.         else if (INLINE_AMD3DNOW(cpu_flags))

  1283.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1284.         else if (INLINE_MMX(cpu_flags))

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

  1286.         else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)

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

  1288.         else

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

  1290. 

  1291.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1292.         av_log(c, AV_LOG_DEBUG,

  1293.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1294.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1295.         av_log(c, AV_LOG_DEBUG,

  1296.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1297.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1298.                c->chrXInc, c->chrYInc);

  1299.     }

  1300. 

  1301.     c->swScale = ff_getSwsFunc(c);

  1302.     return 0;

  1303. fail: // FIXME replace things by appropriate error codes

  1304.     return -1;

  1305. }

  1306. 

  1307. #if FF_API_SWS_GETCONTEXT

  1308. SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,

  1309.                            int dstW, int dstH, enum AVPixelFormat dstFormat,

  1310.                            int flags, SwsFilter *srcFilter,

  1311.                            SwsFilter *dstFilter, const double *param)

  1312. {

  1313.     SwsContext *c;

  1314. 

  1315.     if (!(c = sws_alloc_context()))

  1316.         return NULL;

  1317. 

  1318.     c->flags     = flags;

  1319.     c->srcW      = srcW;

  1320.     c->srcH      = srcH;

  1321.     c->dstW      = dstW;

  1322.     c->dstH      = dstH;

  1323.     c->srcRange  = handle_jpeg(&srcFormat);

  1324.     c->dstRange  = handle_jpeg(&dstFormat);

  1325.     c->srcFormat = srcFormat;

  1326.     c->dstFormat = dstFormat;

  1327. 

  1328.     if (param) {

  1329.         c->param[0] = param[0];

  1330.         c->param[1] = param[1];

  1331.     }

  1332.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1333.                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1334.                              c->dstRange, 0, 1 << 16, 1 << 16);

  1335. 

  1336.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1337.         sws_freeContext(c);

  1338.         return NULL;

  1339.     }

  1340. 

  1341.     return c;

  1342. }

  1343. #endif

  1344. 

  1345. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1346.                                 float lumaSharpen, float chromaSharpen,

  1347.                                 float chromaHShift, float chromaVShift,

  1348.                                 int verbose)

  1349. {

  1350.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1351.     if (!filter)

  1352.         return NULL;

  1353. 

  1354.     if (lumaGBlur != 0.0) {

  1355.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1356.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1357.     } else {

  1358.         filter->lumH = sws_getIdentityVec();

  1359.         filter->lumV = sws_getIdentityVec();

  1360.     }

  1361. 

  1362.     if (chromaGBlur != 0.0) {

  1363.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1364.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1365.     } else {

  1366.         filter->chrH = sws_getIdentityVec();

  1367.         filter->chrV = sws_getIdentityVec();

  1368.     }

  1369. 

  1370.     if (chromaSharpen != 0.0) {

  1371.         SwsVector *id = sws_getIdentityVec();

  1372.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1373.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1374.         sws_addVec(filter->chrH, id);

  1375.         sws_addVec(filter->chrV, id);

  1376.         sws_freeVec(id);

  1377.     }

  1378. 

  1379.     if (lumaSharpen != 0.0) {

  1380.         SwsVector *id = sws_getIdentityVec();

  1381.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1382.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1383.         sws_addVec(filter->lumH, id);

  1384.         sws_addVec(filter->lumV, id);

  1385.         sws_freeVec(id);

  1386.     }

  1387. 

  1388.     if (chromaHShift != 0.0)

  1389.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1390. 

  1391.     if (chromaVShift != 0.0)

  1392.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1393. 

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

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

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

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

  1398. 

  1399.     if (verbose)

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

  1401.     if (verbose)

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

  1403. 

  1404.     return filter;

  1405. }

  1406. 

  1407. SwsVector *sws_allocVec(int length)

  1408. {

  1409.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1410.     if (!vec)

  1411.         return NULL;

  1412.     vec->length = length;

  1413.     vec->coeff  = av_malloc(sizeof(double) * length);

  1414.     if (!vec->coeff)

  1415.         av_freep(&vec);

  1416.     return vec;

  1417. }

  1418. 

  1419. SwsVector *sws_getGaussianVec(double variance, double quality)

  1420. {

  1421.     const int length = (int)(variance * quality + 0.5) | 1;

  1422.     int i;

  1423.     double middle  = (length - 1) * 0.5;

  1424.     SwsVector *vec = sws_allocVec(length);

  1425. 

  1426.     if (!vec)

  1427.         return NULL;

  1428. 

  1429.     for (i = 0; i < length; i++) {

  1430.         double dist = i - middle;

  1431.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1432.                         sqrt(2 * variance * M_PI);

  1433.     }

  1434. 

  1435.     sws_normalizeVec(vec, 1.0);

  1436. 

  1437.     return vec;

  1438. }

  1439. 

  1440. SwsVector *sws_getConstVec(double c, int length)

  1441. {

  1442.     int i;

  1443.     SwsVector *vec = sws_allocVec(length);

  1444. 

  1445.     if (!vec)

  1446.         return NULL;

  1447. 

  1448.     for (i = 0; i < length; i++)

  1449.         vec->coeff[i] = c;

  1450. 

  1451.     return vec;

  1452. }

  1453. 

  1454. SwsVector *sws_getIdentityVec(void)

  1455. {

  1456.     return sws_getConstVec(1.0, 1);

  1457. }

  1458. 

  1459. static double sws_dcVec(SwsVector *a)

  1460. {

  1461.     int i;

  1462.     double sum = 0;

  1463. 

  1464.     for (i = 0; i < a->length; i++)

  1465.         sum += a->coeff[i];

  1466. 

  1467.     return sum;

  1468. }

  1469. 

  1470. void sws_scaleVec(SwsVector *a, double scalar)

  1471. {

  1472.     int i;

  1473. 

  1474.     for (i = 0; i < a->length; i++)

  1475.         a->coeff[i] *= scalar;

  1476. }

  1477. 

  1478. void sws_normalizeVec(SwsVector *a, double height)

  1479. {

  1480.     sws_scaleVec(a, height / sws_dcVec(a));

  1481. }

  1482. 

  1483. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1484. {

  1485.     int length = a->length + b->length - 1;

  1486.     int i, j;

  1487.     SwsVector *vec = sws_getConstVec(0.0, length);

  1488. 

  1489.     if (!vec)

  1490.         return NULL;

  1491. 

  1492.     for (i = 0; i < a->length; i++) {

  1493.         for (j = 0; j < b->length; j++) {

  1494.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1495.         }

  1496.     }

  1497. 

  1498.     return vec;

  1499. }

  1500. 

  1501. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1502. {

  1503.     int length = FFMAX(a->length, b->length);

  1504.     int i;

  1505.     SwsVector *vec = sws_getConstVec(0.0, length);

  1506. 

  1507.     if (!vec)

  1508.         return NULL;

  1509. 

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

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

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

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

  1514. 

  1515.     return vec;

  1516. }

  1517. 

  1518. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1519. {

  1520.     int length = FFMAX(a->length, b->length);

  1521.     int i;

  1522.     SwsVector *vec = sws_getConstVec(0.0, length);

  1523. 

  1524.     if (!vec)

  1525.         return NULL;

  1526. 

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

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

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

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

  1531. 

  1532.     return vec;

  1533. }

  1534. 

  1535. /* shift left / or right if "shift" is negative */

  1536. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1537. {

  1538.     int length = a->length + FFABS(shift) * 2;

  1539.     int i;

  1540.     SwsVector *vec = sws_getConstVec(0.0, length);

  1541. 

  1542.     if (!vec)

  1543.         return NULL;

  1544. 

  1545.     for (i = 0; i < a->length; i++) {

  1546.         vec->coeff[i + (length    - 1) / 2 -

  1547.                        (a->length - 1) / 2 - shift] = a->coeff[i];

  1548.     }

  1549. 

  1550.     return vec;

  1551. }

  1552. 

  1553. void sws_shiftVec(SwsVector *a, int shift)

  1554. {

  1555.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1556.     av_free(a->coeff);

  1557.     a->coeff  = shifted->coeff;

  1558.     a->length = shifted->length;

  1559.     av_free(shifted);

  1560. }

  1561. 

  1562. void sws_addVec(SwsVector *a, SwsVector *b)

  1563. {

  1564.     SwsVector *sum = sws_sumVec(a, b);

  1565.     av_free(a->coeff);

  1566.     a->coeff  = sum->coeff;

  1567.     a->length = sum->length;

  1568.     av_free(sum);

  1569. }

  1570. 

  1571. void sws_subVec(SwsVector *a, SwsVector *b)

  1572. {

  1573.     SwsVector *diff = sws_diffVec(a, b);

  1574.     av_free(a->coeff);

  1575.     a->coeff  = diff->coeff;

  1576.     a->length = diff->length;

  1577.     av_free(diff);

  1578. }

  1579. 

  1580. void sws_convVec(SwsVector *a, SwsVector *b)

  1581. {

  1582.     SwsVector *conv = sws_getConvVec(a, b);

  1583.     av_free(a->coeff);

  1584.     a->coeff  = conv->coeff;

  1585.     a->length = conv->length;

  1586.     av_free(conv);

  1587. }

  1588. 

  1589. SwsVector *sws_cloneVec(SwsVector *a)

  1590. {

  1591.     int i;

  1592.     SwsVector *vec = sws_allocVec(a->length);

  1593. 

  1594.     if (!vec)

  1595.         return NULL;

  1596. 

  1597.     for (i = 0; i < a->length; i++)

  1598.         vec->coeff[i] = a->coeff[i];

  1599. 

  1600.     return vec;

  1601. }

  1602. 

  1603. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1604. {

  1605.     int i;

  1606.     double max = 0;

  1607.     double min = 0;

  1608.     double range;

  1609. 

  1610.     for (i = 0; i < a->length; i++)

  1611.         if (a->coeff[i] > max)

  1612.             max = a->coeff[i];

  1613. 

  1614.     for (i = 0; i < a->length; i++)

  1615.         if (a->coeff[i] < min)

  1616.             min = a->coeff[i];

  1617. 

  1618.     range = max - min;

  1619. 

  1620.     for (i = 0; i < a->length; i++) {

  1621.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1622.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1623.         for (; x > 0; x--)

  1624.             av_log(log_ctx, log_level, " ");

  1625.         av_log(log_ctx, log_level, "|\n");

  1626.     }

  1627. }

  1628. 

  1629. void sws_freeVec(SwsVector *a)

  1630. {

  1631.     if (!a)

  1632.         return;

  1633.     av_freep(&a->coeff);

  1634.     a->length = 0;

  1635.     av_free(a);

  1636. }

  1637. 

  1638. void sws_freeFilter(SwsFilter *filter)

  1639. {

  1640.     if (!filter)

  1641.         return;

  1642. 

  1643.     if (filter->lumH)

  1644.         sws_freeVec(filter->lumH);

  1645.     if (filter->lumV)

  1646.         sws_freeVec(filter->lumV);

  1647.     if (filter->chrH)

  1648.         sws_freeVec(filter->chrH);

  1649.     if (filter->chrV)

  1650.         sws_freeVec(filter->chrV);

  1651.     av_free(filter);

  1652. }

  1653. 

  1654. void sws_freeContext(SwsContext *c)

  1655. {

  1656.     int i;

  1657.     if (!c)

  1658.         return;

  1659. 

  1660.     if (c->lumPixBuf) {

  1661.         for (i = 0; i < c->vLumBufSize; i++)

  1662.             av_freep(&c->lumPixBuf[i]);

  1663.         av_freep(&c->lumPixBuf);

  1664.     }

  1665. 

  1666.     if (c->chrUPixBuf) {

  1667.         for (i = 0; i < c->vChrBufSize; i++)

  1668.             av_freep(&c->chrUPixBuf[i]);

  1669.         av_freep(&c->chrUPixBuf);

  1670.         av_freep(&c->chrVPixBuf);

  1671.     }

  1672. 

  1673.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1674.         for (i = 0; i < c->vLumBufSize; i++)

  1675.             av_freep(&c->alpPixBuf[i]);

  1676.         av_freep(&c->alpPixBuf);

  1677.     }

  1678. 

  1679.     av_freep(&c->vLumFilter);

  1680.     av_freep(&c->vChrFilter);

  1681.     av_freep(&c->hLumFilter);

  1682.     av_freep(&c->hChrFilter);

  1683. #if HAVE_ALTIVEC

  1684.     av_freep(&c->vYCoeffsBank);

  1685.     av_freep(&c->vCCoeffsBank);

  1686. #endif

  1687. 

  1688.     av_freep(&c->vLumFilterPos);

  1689.     av_freep(&c->vChrFilterPos);

  1690.     av_freep(&c->hLumFilterPos);

  1691.     av_freep(&c->hChrFilterPos);

  1692. 

  1693. #if HAVE_MMX_INLINE

  1694. #if USE_MMAP

  1695.     if (c->lumMmxextFilterCode)

  1696.         munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);

  1697.     if (c->chrMmxextFilterCode)

  1698.         munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);

  1699. #elif HAVE_VIRTUALALLOC

  1700.     if (c->lumMmxextFilterCode)

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

  1702.     if (c->chrMmxextFilterCode)

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

  1704. #else

  1705.     av_free(c->lumMmxextFilterCode);

  1706.     av_free(c->chrMmxextFilterCode);

  1707. #endif

  1708.     c->lumMmxextFilterCode = NULL;

  1709.     c->chrMmxextFilterCode = NULL;

  1710. #endif /* HAVE_MMX_INLINE */

  1711. 

  1712.     av_freep(&c->yuvTable);

  1713.     av_free(c->formatConvBuffer);

  1714. 

  1715.     av_free(c);

  1716. }

  1717. 

  1718. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  1719.                                         int srcH, enum AVPixelFormat srcFormat,

  1720.                                         int dstW, int dstH,

  1721.                                         enum AVPixelFormat dstFormat, int flags,

  1722.                                         SwsFilter *srcFilter,

  1723.                                         SwsFilter *dstFilter,

  1724.                                         const double *param)

  1725. {

  1726.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  1727.                                              SWS_PARAM_DEFAULT };

  1728. 

  1729.     if (!param)

  1730.         param = default_param;

  1731. 

  1732.     if (context &&

  1733.         (context->srcW      != srcW      ||

  1734.          context->srcH      != srcH      ||

  1735.          context->srcFormat != srcFormat ||

  1736.          context->dstW      != dstW      ||

  1737.          context->dstH      != dstH      ||

  1738.          context->dstFormat != dstFormat ||

  1739.          context->flags     != flags     ||

  1740.          context->param[0]  != param[0]  ||

  1741.          context->param[1]  != param[1])) {

  1742.         sws_freeContext(context);

  1743.         context = NULL;

  1744.     }

  1745. 

  1746.     if (!context) {

  1747.         if (!(context = sws_alloc_context()))

  1748.             return NULL;

  1749.         context->srcW      = srcW;

  1750.         context->srcH      = srcH;

  1751.         context->srcRange  = handle_jpeg(&srcFormat);

  1752.         context->srcFormat = srcFormat;

  1753.         context->dstW      = dstW;

  1754.         context->dstH      = dstH;

  1755.         context->dstRange  = handle_jpeg(&dstFormat);

  1756.         context->dstFormat = dstFormat;

  1757.         context->flags     = flags;

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

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

  1760.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  1761.                                  context->srcRange,

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

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

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

  1765.             sws_freeContext(context);

  1766.             return NULL;

  1767.         }

  1768.     }

  1769.     return context;

  1770. }