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.
36271 Commits
32 Branches
283MB
Tree: d641ee94b5
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'd641ee94b5'
${ noResults }
FFmpeg/libswscale/utils.c

1745 lines
61KB
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.     int is_supported_in, is_supported_out;

  74. } FormatEntry;

  75. 

  76. static const FormatEntry format_entries[AV_PIX_FMT_NB] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  166.     [AV_PIX_FMT_GBRP]        = { 1, 0 },

  167.     [AV_PIX_FMT_GBRP9LE]     = { 1, 0 },

  168.     [AV_PIX_FMT_GBRP9BE]     = { 1, 0 },

  169.     [AV_PIX_FMT_GBRP10LE]    = { 1, 0 },

  170.     [AV_PIX_FMT_GBRP10BE]    = { 1, 0 },

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

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

  173. };

  174. 

  175. int sws_isSupportedInput(enum AVPixelFormat pix_fmt)

  176. {

  177.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  178.            format_entries[pix_fmt].is_supported_in : 0;

  179. }

  180. 

  181. int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)

  182. {

  183.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  184.            format_entries[pix_fmt].is_supported_out : 0;

  185. }

  186. 

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

  188. 

  189. const char *sws_format_name(enum AVPixelFormat format)

  190. {

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

  192.     if (desc)

  193.         return desc->name;

  194.     else

  195.         return "Unknown format";

  196. }

  197. 

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

  199.                              double dist)

  200. {

  201.     if (dist <= 1.0)

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

  203.     else

  204.         return getSplineCoeff(0.0,

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

  206.                                c + 3.0 * d,

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

  208.                               dist - 1.0);

  209. }

  210. 

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

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

  213.                       int filterAlign, int one, int flags, int cpu_flags,

  214.                       SwsVector *srcFilter, SwsVector *dstFilter,

  215.                       double param[2], int is_horizontal)

  216. {

  217.     int i;

  218.     int filterSize;

  219.     int filter2Size;

  220.     int minFilterSize;

  221.     int64_t *filter    = NULL;

  222.     int64_t *filter2   = NULL;

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

  224.     int ret            = -1;

  225. 

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

  227. 

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

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

  230. 

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

  232.         int i;

  233.         filterSize = 1;

  234.         FF_ALLOCZ_OR_GOTO(NULL, filter,

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

  236. 

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

  238.             filter[i * filterSize] = fone;

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

  240.         }

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

  242.         int i;

  243.         int xDstInSrc;

  244.         filterSize = 1;

  245.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  247. 

  248.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  251. 

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

  253.             filter[i]       = fone;

  254.             xDstInSrc      += xInc;

  255.         }

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

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

  258.         int i;

  259.         int xDstInSrc;

  260.         filterSize = 2;

  261.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  263. 

  264.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  267.             int j;

  268. 

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

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

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

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

  273.                                 (fone >> 16);

  274.                 if (coeff < 0)

  275.                     coeff = 0;

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

  277.                 xx++;

  278.             }

  279.             xDstInSrc += xInc;

  280.         }

  281.     } else {

  282.         int64_t xDstInSrc;

  283.         int sizeFactor;

  284. 

  285.         if (flags & SWS_BICUBIC)

  286.             sizeFactor = 4;

  287.         else if (flags & SWS_X)

  288.             sizeFactor = 8;

  289.         else if (flags & SWS_AREA)

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

  291.         else if (flags & SWS_GAUSS)

  292.             sizeFactor = 8;     // infinite ;)

  293.         else if (flags & SWS_LANCZOS)

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

  295.         else if (flags & SWS_SINC)

  296.             sizeFactor = 20;    // infinite ;)

  297.         else if (flags & SWS_SPLINE)

  298.             sizeFactor = 20;    // infinite ;)

  299.         else if (flags & SWS_BILINEAR)

  300.             sizeFactor = 2;

  301.         else {

  302.             sizeFactor = 0;     // GCC warning killer

  303.             assert(0);

  304.         }

  305. 

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

  307.             filterSize = 1 + sizeFactor;    // upscale

  308.         else

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

  310. 

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

  312.         filterSize = FFMAX(filterSize, 1);

  313. 

  314.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  316. 

  317.         xDstInSrc = xInc - 0x10000;

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

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

  320.             int j;

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

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

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

  324.                 double floatd;

  325.                 int64_t coeff;

  326. 

  327.                 if (xInc > 1 << 16)

  328.                     d = d * dstW / srcW;

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

  330. 

  331.                 if (flags & SWS_BICUBIC) {

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

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

  334. 

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

  336.                         coeff = 0.0;

  337.                     } else {

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

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

  340. 

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

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

  343.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  344.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  345.                         else

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

  347.                                       (6 * B + 30 * C) * dd  +

  348.                                     (-12 * B - 48 * C) * d   +

  349.                                       (8 * B + 24 * C) * (1 << 30);

  350.                     }

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

  352.                 }

  353. #if 0

  354.                 else if (flags & SWS_X) {

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

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

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

  358.                 }

  359. #endif

  360.                 else if (flags & SWS_X) {

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

  362.                     double c;

  363. 

  364.                     if (floatd < 1.0)

  365.                         c = cos(floatd * M_PI);

  366.                     else

  367.                         c = -1.0;

  368.                     if (c < 0.0)

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

  370.                     else

  371.                         c = pow(c, A);

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

  373.                 } else if (flags & SWS_AREA) {

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

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

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

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

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

  379.                     else

  380.                         coeff = 0.0;

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

  382.                 } else if (flags & SWS_GAUSS) {

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

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

  385.                 } else if (flags & SWS_SINC) {

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

  387.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  391.                     if (floatd > p)

  392.                         coeff = 0;

  393.                 } else if (flags & SWS_BILINEAR) {

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

  395.                     if (coeff < 0)

  396.                         coeff = 0;

  397.                     coeff *= fone >> 30;

  398.                 } else if (flags & SWS_SPLINE) {

  399.                     double p = -2.196152422706632;

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

  401.                 } else {

  402.                     coeff = 0.0; // GCC warning killer

  403.                     assert(0);

  404.                 }

  405. 

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

  407.                 xx++;

  408.             }

  409.             xDstInSrc += 2 * xInc;

  410.         }

  411.     }

  412. 

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

  414.      * av_free(filter);

  415.      */

  416.     assert(filterSize > 0);

  417.     filter2Size = filterSize;

  418.     if (srcFilter)

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

  420.     if (dstFilter)

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

  422.     assert(filter2Size > 0);

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

  424. 

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

  426.         int j, k;

  427. 

  428.         if (srcFilter) {

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

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

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

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

  433.             }

  434.         } else {

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

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

  437.         }

  438.         // FIXME dstFilter

  439. 

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

  441.     }

  442.     av_freep(&filter);

  443. 

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

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

  446.     minFilterSize = 0;

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

  448.         int min = filter2Size;

  449.         int j;

  450.         int64_t cutOff = 0.0;

  451. 

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

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

  454.             int k;

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

  456. 

  457.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  458.                 break;

  459. 

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

  461.              * filter otherwise */

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

  463.                 break;

  464. 

  465.             // move filter coefficients left

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

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

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

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

  470.         }

  471. 

  472.         cutOff = 0;

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

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

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

  476. 

  477.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  478.                 break;

  479.             min--;

  480.         }

  481. 

  482.         if (min > minFilterSize)

  483.             minFilterSize = min;

  484.     }

  485. 

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

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

  488.         if (minFilterSize < 5)

  489.             filterAlign = 4;

  490. 

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

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

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

  494.         if (minFilterSize < 3)

  495.             filterAlign = 1;

  496.     }

  497. 

  498.     if (INLINE_MMX(cpu_flags)) {

  499.         // special case for unscaled vertical filtering

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

  501.             filterAlign = 1;

  502.     }

  503. 

  504.     assert(minFilterSize > 0);

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

  506.     assert(filterSize > 0);

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

  508.     if (filterSize >= MAX_FILTER_SIZE * 16 /

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

  510.         goto fail;

  511.     *outFilterSize = filterSize;

  512. 

  513.     if (flags & SWS_PRINT_INFO)

  514.         av_log(NULL, AV_LOG_VERBOSE,

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

  516.                filter2Size, filterSize);

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

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

  519.         int j;

  520. 

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

  522.             if (j >= filter2Size)

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

  524.             else

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

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

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

  528.         }

  529.     }

  530. 

  531.     // FIXME try to align filterPos if possible

  532. 

  533.     // fix borders

  534.     if (is_horizontal) {

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

  536.             int j;

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

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

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

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

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

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

  543.                 }

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

  545.             }

  546. 

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

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

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

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

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

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

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

  554.                 }

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

  556.             }

  557.         }

  558.     }

  559. 

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

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

  562.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

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

  564. 

  565.     /* normalize & store in outFilter */

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

  567.         int j;

  568.         int64_t error = 0;

  569.         int64_t sum   = 0;

  570. 

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

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

  573.         }

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

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

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

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

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

  579.             error                                  = v - intV * sum;

  580.         }

  581.     }

  582. 

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

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

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

  586.                                                       * read over the end */

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

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

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

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

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

  592.     }

  593. 

  594.     ret = 0;

  595. 

  596. fail:

  597.     av_free(filter);

  598.     av_free(filter2);

  599.     return ret;

  600. }

  601. 

  602. #if HAVE_MMXEXT_INLINE

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

  604.                                int16_t *filter, int32_t *filterPos,

  605.                                int numSplits)

  606. {

  607.     uint8_t *fragmentA;

  608.     x86_reg imm8OfPShufW1A;

  609.     x86_reg imm8OfPShufW2A;

  610.     x86_reg fragmentLengthA;

  611.     uint8_t *fragmentB;

  612.     x86_reg imm8OfPShufW1B;

  613.     x86_reg imm8OfPShufW2B;

  614.     x86_reg fragmentLengthB;

  615.     int fragmentPos;

  616. 

  617.     int xpos, i;

  618. 

  619.     // create an optimized horizontal scaling routine

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

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

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

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

  624.      */

  625. 

  626.     // code fragment

  627. 

  628.     __asm__ volatile (

  629.         "jmp                         9f                 \n\t"

  630.         // Begin

  631.         "0:                                             \n\t"

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

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

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

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

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

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

  638.         "1:                                             \n\t"

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

  640.         "2:                                             \n\t"

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

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

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

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

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

  646. 

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

  648. 

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

  650.         // End

  651.         "9:                                             \n\t"

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

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

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

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

  656.         "dec                         %1                 \n\t"

  657.         "dec                         %2                 \n\t"

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

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

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

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

  662. 

  663. 

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

  665.           "=r" (fragmentLengthA)

  666.         );

  667. 

  668.     __asm__ volatile (

  669.         "jmp                         9f                 \n\t"

  670.         // Begin

  671.         "0:                                             \n\t"

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

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

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

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

  676.         "1:                                             \n\t"

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

  678.         "2:                                             \n\t"

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

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

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

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

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

  684. 

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

  686. 

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

  688.         // End

  689.         "9:                                             \n\t"

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

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

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

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

  694.         "dec                         %1                 \n\t"

  695.         "dec                         %2                 \n\t"

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

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

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

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

  700. 

  701. 

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

  703.           "=r" (fragmentLengthB)

  704.         );

  705. 

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

  707.     fragmentPos = 0;

  708. 

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

  710.         int xx = xpos >> 16;

  711. 

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

  713.             int a                  = 0;

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

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

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

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

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

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

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

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

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

  723.             int shift              = 0;

  724. 

  725.             if (filterCode) {

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

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

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

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

  730.                 filterPos[i / 2] = xx;

  731. 

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

  733. 

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

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

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

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

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

  739.                                                                (c << 4) |

  740.                                                                (d << 6);

  741. 

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

  743.                     shift = maxShift;               // avoid overread

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

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

  746. 

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

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

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

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

  751.                 }

  752.             }

  753. 

  754.             fragmentPos += fragmentLength;

  755. 

  756.             if (filterCode)

  757.                 filterCode[fragmentPos] = RET;

  758.         }

  759.         xpos += xInc;

  760.     }

  761.     if (filterCode)

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

  763. 

  764.     return fragmentPos + 1;

  765. }

  766. #endif /* HAVE_MMXEXT_INLINE */

  767. 

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

  769. {

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

  771.     *h = desc->log2_chroma_w;

  772.     *v = desc->log2_chroma_h;

  773. }

  774. 

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

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

  777.                              int brightness, int contrast, int saturation)

  778. {

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

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

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

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

  783. 

  784.     c->brightness = brightness;

  785.     c->contrast   = contrast;

  786.     c->saturation = saturation;

  787.     c->srcRange   = srcRange;

  788.     c->dstRange   = dstRange;

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

  790.         return -1;

  791. 

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

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

  794. 

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

  796.                              contrast, saturation);

  797.     // FIXME factorize

  798. 

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

  800.         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,

  801.                                        contrast, saturation);

  802.     return 0;

  803. }

  804. 

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

  806.                              int *srcRange, int **table, int *dstRange,

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

  808. {

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

  810.         return -1;

  811. 

  812.     *inv_table  = c->srcColorspaceTable;

  813.     *table      = c->dstColorspaceTable;

  814.     *srcRange   = c->srcRange;

  815.     *dstRange   = c->dstRange;

  816.     *brightness = c->brightness;

  817.     *contrast   = c->contrast;

  818.     *saturation = c->saturation;

  819. 

  820.     return 0;

  821. }

  822. 

  823. static int handle_jpeg(enum AVPixelFormat *format)

  824. {

  825.     switch (*format) {

  826.     case AV_PIX_FMT_YUVJ420P:

  827.         *format = AV_PIX_FMT_YUV420P;

  828.         return 1;

  829.     case AV_PIX_FMT_YUVJ422P:

  830.         *format = AV_PIX_FMT_YUV422P;

  831.         return 1;

  832.     case AV_PIX_FMT_YUVJ444P:

  833.         *format = AV_PIX_FMT_YUV444P;

  834.         return 1;

  835.     case AV_PIX_FMT_YUVJ440P:

  836.         *format = AV_PIX_FMT_YUV440P;

  837.         return 1;

  838.     default:

  839.         return 0;

  840.     }

  841. }

  842. 

  843. SwsContext *sws_alloc_context(void)

  844. {

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

  846. 

  847.     if (c) {

  848.         c->av_class = &sws_context_class;

  849.         av_opt_set_defaults(c);

  850.     }

  851. 

  852.     return c;

  853. }

  854. 

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

  856.                              SwsFilter *dstFilter)

  857. {

  858.     int i;

  859.     int usesVFilter, usesHFilter;

  860.     int unscaled;

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

  862.     int srcW              = c->srcW;

  863.     int srcH              = c->srcH;

  864.     int dstW              = c->dstW;

  865.     int dstH              = c->dstH;

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

  867.     int dst_stride_px     = dst_stride >> 1;

  868.     int flags, cpu_flags;

  869.     enum AVPixelFormat srcFormat = c->srcFormat;

  870.     enum AVPixelFormat dstFormat = c->dstFormat;

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

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

  873. 

  874.     cpu_flags = av_get_cpu_flags();

  875.     flags     = c->flags;

  876.     emms_c();

  877.     if (!rgb15to16)

  878.         sws_rgb2rgb_init();

  879. 

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

  881. 

  882.     if (!sws_isSupportedInput(srcFormat)) {

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

  884.                sws_format_name(srcFormat));

  885.         return AVERROR(EINVAL);

  886.     }

  887.     if (!sws_isSupportedOutput(dstFormat)) {

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

  889.                sws_format_name(dstFormat));

  890.         return AVERROR(EINVAL);

  891.     }

  892. 

  893.     i = flags & (SWS_POINT         |

  894.                  SWS_AREA          |

  895.                  SWS_BILINEAR      |

  896.                  SWS_FAST_BILINEAR |

  897.                  SWS_BICUBIC       |

  898.                  SWS_X             |

  899.                  SWS_GAUSS         |

  900.                  SWS_LANCZOS       |

  901.                  SWS_SINC          |

  902.                  SWS_SPLINE        |

  903.                  SWS_BICUBLIN);

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

  905.         av_log(c, AV_LOG_ERROR,

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

  907.         return AVERROR(EINVAL);

  908.     }

  909.     /* sanity check */

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

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

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

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

  914.                srcW, srcH, dstW, dstH);

  915.         return AVERROR(EINVAL);

  916.     }

  917. 

  918.     if (!dstFilter)

  919.         dstFilter = &dummyFilter;

  920.     if (!srcFilter)

  921.         srcFilter = &dummyFilter;

  922. 

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

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

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

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

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

  928. 

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

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

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

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

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

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

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

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

  937. 

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

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

  940. 

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

  942.      * chroma interpolation */

  943.     if (flags & SWS_FULL_CHR_H_INT &&

  944.         isAnyRGB(dstFormat)        &&

  945.         dstFormat != AV_PIX_FMT_RGBA  &&

  946.         dstFormat != AV_PIX_FMT_ARGB  &&

  947.         dstFormat != AV_PIX_FMT_BGRA  &&

  948.         dstFormat != AV_PIX_FMT_ABGR  &&

  949.         dstFormat != AV_PIX_FMT_RGB24 &&

  950.         dstFormat != AV_PIX_FMT_BGR24) {

  951.         av_log(c, AV_LOG_ERROR,

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

  953.                sws_format_name(dstFormat));

  954.         flags   &= ~SWS_FULL_CHR_H_INT;

  955.         c->flags = flags;

  956.     }

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

  958.         c->chrDstHSubSample = 1;

  959. 

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

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

  962.                            SWS_SRC_V_CHR_DROP_SHIFT;

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

  964. 

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

  966.      * wants full chroma */

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

  968.         srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&

  969.         srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&

  970.         srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&

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

  972.          (flags & SWS_FAST_BILINEAR)))

  973.         c->chrSrcHSubSample = 1;

  974. 

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

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

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

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

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

  980. 

  981.     /* unscaled special cases */

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

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

  984.         ff_get_unscaled_swscale(c);

  985. 

  986.         if (c->swScale) {

  987.             if (flags & SWS_PRINT_INFO)

  988.                 av_log(c, AV_LOG_INFO,

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

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

  991.             return 0;

  992.         }

  993.     }

  994. 

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

  996.     if (c->srcBpc < 8)

  997.         c->srcBpc = 8;

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

  999.     if (c->dstBpc < 8)

  1000.         c->dstBpc = 8;

  1001.     if (c->dstBpc == 16)

  1002.         dst_stride <<= 1;

  1003.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,

  1004.                      (FFALIGN(srcW, 16) * 2 * FFALIGN(c->srcBpc, 8) >> 3) + 16,

  1005.                      fail);

  1006.     if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 10) {

  1007.         c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&

  1008.                               (srcW & 15) == 0) ? 1 : 0;

  1009.         if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0

  1010.             && (flags & SWS_FAST_BILINEAR)) {

  1011.             if (flags & SWS_PRINT_INFO)

  1012.                 av_log(c, AV_LOG_INFO,

  1013.                        "output width is not a multiple of 32 -> no MMXEXT scaler\n");

  1014.         }

  1015.         if (usesHFilter)

  1016.             c->canMMXEXTBeUsed = 0;

  1017.     } else

  1018.         c->canMMXEXTBeUsed = 0;

  1019. 

  1020.     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;

  1021.     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

  1022. 

  1023.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  1024.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  1025.      * correct scaling.

  1026.      * n-2 is the last chrominance sample available.

  1027.      * This is not perfect, but no one should notice the difference, the more

  1028.      * correct variant would be like the vertical one, but that would require

  1029.      * some special code for the first and last pixel */

  1030.     if (flags & SWS_FAST_BILINEAR) {

  1031.         if (c->canMMXEXTBeUsed) {

  1032.             c->lumXInc += 20;

  1033.             c->chrXInc += 20;

  1034.         }

  1035.         // we don't use the x86 asm scaler if MMX is available

  1036.         else if (INLINE_MMX(cpu_flags)) {

  1037.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1038.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1039.         }

  1040.     }

  1041. 

  1042. #define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)

  1043. 

  1044.     /* precalculate horizontal scaler filter coefficients */

  1045.     {

  1046. #if HAVE_MMXEXT_INLINE

  1047. // can't downscale !!!

  1048.         if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {

  1049.             c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,

  1050.                                                              NULL, NULL, 8);

  1051.             c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,

  1052.                                                              NULL, NULL, NULL, 4);

  1053. 

  1054. #if USE_MMAP

  1055.             c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,

  1056.                                           PROT_READ | PROT_WRITE,

  1057.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1058.                                           -1, 0);

  1059.             c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,

  1060.                                           PROT_READ | PROT_WRITE,

  1061.                                           MAP_PRIVATE | MAP_ANONYMOUS,

  1062.                                           -1, 0);

  1063. #elif HAVE_VIRTUALALLOC

  1064.             c->lumMmxextFilterCode = VirtualAlloc(NULL,

  1065.                                                   c->lumMmxextFilterCodeSize,

  1066.                                                   MEM_COMMIT,

  1067.                                                   PAGE_EXECUTE_READWRITE);

  1068.             c->chrMmxextFilterCode = VirtualAlloc(NULL,

  1069.                                                   c->chrMmxextFilterCodeSize,

  1070.                                                   MEM_COMMIT,

  1071.                                                   PAGE_EXECUTE_READWRITE);

  1072. #else

  1073.             c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);

  1074.             c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);

  1075. #endif

  1076. 

  1077.             if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)

  1078.                 return AVERROR(ENOMEM);

  1079.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1080.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1081.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1082.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1083. 

  1084.             init_hscaler_mmxext(dstW, c->lumXInc, c->lumMmxextFilterCode,

  1085.                                 c->hLumFilter, c->hLumFilterPos, 8);

  1086.             init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,

  1087.                                 c->hChrFilter, c->hChrFilterPos, 4);

  1088. 

  1089. #if USE_MMAP

  1090.             mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1091.             mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ);

  1092. #endif

  1093.         } else

  1094. #endif /* HAVE_MMXEXT_INLINE */

  1095.         {

  1096.             const int filterAlign =

  1097.                 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :

  1098.                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1099.                 1;

  1100. 

  1101.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1102.                            &c->hLumFilterSize, c->lumXInc,

  1103.                            srcW, dstW, filterAlign, 1 << 14,

  1104.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1105.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1106.                            c->param, 1) < 0)

  1107.                 goto fail;

  1108.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1109.                            &c->hChrFilterSize, c->chrXInc,

  1110.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1111.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1112.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1113.                            c->param, 1) < 0)

  1114.                 goto fail;

  1115.         }

  1116.     } // initialize horizontal stuff

  1117. 

  1118.     /* precalculate vertical scaler filter coefficients */

  1119.     {

  1120.         const int filterAlign =

  1121.             (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :

  1122.             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1123.             1;

  1124. 

  1125.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1126.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

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

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

  1129.                        c->param, 0) < 0)

  1130.             goto fail;

  1131.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1132.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1133.                        filterAlign, (1 << 12),

  1134.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1135.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1136.                        c->param, 0) < 0)

  1137.             goto fail;

  1138. 

  1139. #if HAVE_ALTIVEC

  1140.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1141.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1142. 

  1143.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1144.             int j;

  1145.             short *p = (short *)&c->vYCoeffsBank[i];

  1146.             for (j = 0; j < 8; j++)

  1147.                 p[j] = c->vLumFilter[i];

  1148.         }

  1149. 

  1150.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1151.             int j;

  1152.             short *p = (short *)&c->vCCoeffsBank[i];

  1153.             for (j = 0; j < 8; j++)

  1154.                 p[j] = c->vChrFilter[i];

  1155.         }

  1156. #endif

  1157.     }

  1158. 

  1159.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1160.     c->vLumBufSize = c->vLumFilterSize;

  1161.     c->vChrBufSize = c->vChrFilterSize;

  1162.     for (i = 0; i < dstH; i++) {

  1163.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1164.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1165.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1166.                                << c->chrSrcVSubSample));

  1167. 

  1168.         nextSlice >>= c->chrSrcVSubSample;

  1169.         nextSlice <<= c->chrSrcVSubSample;

  1170.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1171.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1172.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1173.             (nextSlice >> c->chrSrcVSubSample))

  1174.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1175.                              c->vChrFilterPos[chrI];

  1176.     }

  1177. 

  1178.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1179.      * need to allocate several megabytes to handle all possible cases) */

  1180.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1181.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1182.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1183.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1184.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1185.     /* Note we need at least one pixel more at the end because of the MMX code

  1186.      * (just in case someone wants to replace the 4000/8000). */

  1187.     /* align at 16 bytes for AltiVec */

  1188.     for (i = 0; i < c->vLumBufSize; i++) {

  1189.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1190.                           dst_stride + 16, fail);

  1191.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1192.     }

  1193.     // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)

  1194.     c->uv_off_px   = dst_stride_px + 64 / (c->dstBpc & ~7);

  1195.     c->uv_off_byte = dst_stride + 16;

  1196.     for (i = 0; i < c->vChrBufSize; i++) {

  1197.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1198.                          dst_stride * 2 + 32, fail);

  1199.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1200.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1201.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1202.     }

  1203.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

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

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

  1206.                               dst_stride + 16, fail);

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

  1208.         }

  1209. 

  1210.     // try to avoid drawing green stuff between the right end and the stride end

  1211.     for (i = 0; i < c->vChrBufSize; i++)

  1212.         memset(c->chrUPixBuf[i], 64, dst_stride * 2 + 1);

  1213. 

  1214.     assert(c->chrDstH <= dstH);

  1215. 

  1216.     if (flags & SWS_PRINT_INFO) {

  1217.         if (flags & SWS_FAST_BILINEAR)

  1218.             av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1219.         else if (flags & SWS_BILINEAR)

  1220.             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1221.         else if (flags & SWS_BICUBIC)

  1222.             av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1223.         else if (flags & SWS_X)

  1224.             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1225.         else if (flags & SWS_POINT)

  1226.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1227.         else if (flags & SWS_AREA)

  1228.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1229.         else if (flags & SWS_BICUBLIN)

  1230.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1231.         else if (flags & SWS_GAUSS)

  1232.             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1233.         else if (flags & SWS_SINC)

  1234.             av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1235.         else if (flags & SWS_LANCZOS)

  1236.             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1237.         else if (flags & SWS_SPLINE)

  1238.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1239.         else

  1240.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1241. 

  1242.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1243.                sws_format_name(srcFormat),

  1244. #ifdef DITHER1XBPP

  1245.                dstFormat == AV_PIX_FMT_BGR555   || dstFormat == AV_PIX_FMT_BGR565   ||

  1246.                dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||

  1247.                dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?

  1248.                                                              "dithered " : "",

  1249. #else

  1250.                "",

  1251. #endif

  1252.                sws_format_name(dstFormat));

  1253. 

  1254.         if (INLINE_MMXEXT(cpu_flags))

  1255.             av_log(c, AV_LOG_INFO, "using MMXEXT\n");

  1256.         else if (INLINE_AMD3DNOW(cpu_flags))

  1257.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1258.         else if (INLINE_MMX(cpu_flags))

  1259.             av_log(c, AV_LOG_INFO, "using MMX\n");

  1260.         else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)

  1261.             av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1262.         else

  1263.             av_log(c, AV_LOG_INFO, "using C\n");

  1264. 

  1265.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1266.         av_log(c, AV_LOG_DEBUG,

  1267.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1268.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1269.         av_log(c, AV_LOG_DEBUG,

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

  1271.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1272.                c->chrXInc, c->chrYInc);

  1273.     }

  1274. 

  1275.     c->swScale = ff_getSwsFunc(c);

  1276.     return 0;

  1277. fail: // FIXME replace things by appropriate error codes

  1278.     return -1;

  1279. }

  1280. 

  1281. #if FF_API_SWS_GETCONTEXT

  1282. SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,

  1283.                            int dstW, int dstH, enum AVPixelFormat dstFormat,

  1284.                            int flags, SwsFilter *srcFilter,

  1285.                            SwsFilter *dstFilter, const double *param)

  1286. {

  1287.     SwsContext *c;

  1288. 

  1289.     if (!(c = sws_alloc_context()))

  1290.         return NULL;

  1291. 

  1292.     c->flags     = flags;

  1293.     c->srcW      = srcW;

  1294.     c->srcH      = srcH;

  1295.     c->dstW      = dstW;

  1296.     c->dstH      = dstH;

  1297.     c->srcRange  = handle_jpeg(&srcFormat);

  1298.     c->dstRange  = handle_jpeg(&dstFormat);

  1299.     c->srcFormat = srcFormat;

  1300.     c->dstFormat = dstFormat;

  1301. 

  1302.     if (param) {

  1303.         c->param[0] = param[0];

  1304.         c->param[1] = param[1];

  1305.     }

  1306.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1307.                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1308.                              c->dstRange, 0, 1 << 16, 1 << 16);

  1309. 

  1310.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1311.         sws_freeContext(c);

  1312.         return NULL;

  1313.     }

  1314. 

  1315.     return c;

  1316. }

  1317. #endif

  1318. 

  1319. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1320.                                 float lumaSharpen, float chromaSharpen,

  1321.                                 float chromaHShift, float chromaVShift,

  1322.                                 int verbose)

  1323. {

  1324.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1325.     if (!filter)

  1326.         return NULL;

  1327. 

  1328.     if (lumaGBlur != 0.0) {

  1329.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1330.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1331.     } else {

  1332.         filter->lumH = sws_getIdentityVec();

  1333.         filter->lumV = sws_getIdentityVec();

  1334.     }

  1335. 

  1336.     if (chromaGBlur != 0.0) {

  1337.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1338.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1339.     } else {

  1340.         filter->chrH = sws_getIdentityVec();

  1341.         filter->chrV = sws_getIdentityVec();

  1342.     }

  1343. 

  1344.     if (chromaSharpen != 0.0) {

  1345.         SwsVector *id = sws_getIdentityVec();

  1346.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1347.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1348.         sws_addVec(filter->chrH, id);

  1349.         sws_addVec(filter->chrV, id);

  1350.         sws_freeVec(id);

  1351.     }

  1352. 

  1353.     if (lumaSharpen != 0.0) {

  1354.         SwsVector *id = sws_getIdentityVec();

  1355.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1356.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1357.         sws_addVec(filter->lumH, id);

  1358.         sws_addVec(filter->lumV, id);

  1359.         sws_freeVec(id);

  1360.     }

  1361. 

  1362.     if (chromaHShift != 0.0)

  1363.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1364. 

  1365.     if (chromaVShift != 0.0)

  1366.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1367. 

  1368.     sws_normalizeVec(filter->chrH, 1.0);

  1369.     sws_normalizeVec(filter->chrV, 1.0);

  1370.     sws_normalizeVec(filter->lumH, 1.0);

  1371.     sws_normalizeVec(filter->lumV, 1.0);

  1372. 

  1373.     if (verbose)

  1374.         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1375.     if (verbose)

  1376.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1377. 

  1378.     return filter;

  1379. }

  1380. 

  1381. SwsVector *sws_allocVec(int length)

  1382. {

  1383.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1384.     if (!vec)

  1385.         return NULL;

  1386.     vec->length = length;

  1387.     vec->coeff  = av_malloc(sizeof(double) * length);

  1388.     if (!vec->coeff)

  1389.         av_freep(&vec);

  1390.     return vec;

  1391. }

  1392. 

  1393. SwsVector *sws_getGaussianVec(double variance, double quality)

  1394. {

  1395.     const int length = (int)(variance * quality + 0.5) | 1;

  1396.     int i;

  1397.     double middle  = (length - 1) * 0.5;

  1398.     SwsVector *vec = sws_allocVec(length);

  1399. 

  1400.     if (!vec)

  1401.         return NULL;

  1402. 

  1403.     for (i = 0; i < length; i++) {

  1404.         double dist = i - middle;

  1405.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1406.                         sqrt(2 * variance * M_PI);

  1407.     }

  1408. 

  1409.     sws_normalizeVec(vec, 1.0);

  1410. 

  1411.     return vec;

  1412. }

  1413. 

  1414. SwsVector *sws_getConstVec(double c, int length)

  1415. {

  1416.     int i;

  1417.     SwsVector *vec = sws_allocVec(length);

  1418. 

  1419.     if (!vec)

  1420.         return NULL;

  1421. 

  1422.     for (i = 0; i < length; i++)

  1423.         vec->coeff[i] = c;

  1424. 

  1425.     return vec;

  1426. }

  1427. 

  1428. SwsVector *sws_getIdentityVec(void)

  1429. {

  1430.     return sws_getConstVec(1.0, 1);

  1431. }

  1432. 

  1433. static double sws_dcVec(SwsVector *a)

  1434. {

  1435.     int i;

  1436.     double sum = 0;

  1437. 

  1438.     for (i = 0; i < a->length; i++)

  1439.         sum += a->coeff[i];

  1440. 

  1441.     return sum;

  1442. }

  1443. 

  1444. void sws_scaleVec(SwsVector *a, double scalar)

  1445. {

  1446.     int i;

  1447. 

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

  1449.         a->coeff[i] *= scalar;

  1450. }

  1451. 

  1452. void sws_normalizeVec(SwsVector *a, double height)

  1453. {

  1454.     sws_scaleVec(a, height / sws_dcVec(a));

  1455. }

  1456. 

  1457. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1458. {

  1459.     int length = a->length + b->length - 1;

  1460.     int i, j;

  1461.     SwsVector *vec = sws_getConstVec(0.0, length);

  1462. 

  1463.     if (!vec)

  1464.         return NULL;

  1465. 

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

  1467.         for (j = 0; j < b->length; j++) {

  1468.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1469.         }

  1470.     }

  1471. 

  1472.     return vec;

  1473. }

  1474. 

  1475. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1476. {

  1477.     int length = FFMAX(a->length, b->length);

  1478.     int i;

  1479.     SwsVector *vec = sws_getConstVec(0.0, length);

  1480. 

  1481.     if (!vec)

  1482.         return NULL;

  1483. 

  1484.     for (i = 0; i < a->length; i++)

  1485.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

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

  1487.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

  1488. 

  1489.     return vec;

  1490. }

  1491. 

  1492. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1493. {

  1494.     int length = FFMAX(a->length, b->length);

  1495.     int i;

  1496.     SwsVector *vec = sws_getConstVec(0.0, length);

  1497. 

  1498.     if (!vec)

  1499.         return NULL;

  1500. 

  1501.     for (i = 0; i < a->length; i++)

  1502.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1503.     for (i = 0; i < b->length; i++)

  1504.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

  1505. 

  1506.     return vec;

  1507. }

  1508. 

  1509. /* shift left / or right if "shift" is negative */

  1510. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1511. {

  1512.     int length = a->length + FFABS(shift) * 2;

  1513.     int i;

  1514.     SwsVector *vec = sws_getConstVec(0.0, length);

  1515. 

  1516.     if (!vec)

  1517.         return NULL;

  1518. 

  1519.     for (i = 0; i < a->length; i++) {

  1520.         vec->coeff[i + (length    - 1) / 2 -

  1521.                        (a->length - 1) / 2 - shift] = a->coeff[i];

  1522.     }

  1523. 

  1524.     return vec;

  1525. }

  1526. 

  1527. void sws_shiftVec(SwsVector *a, int shift)

  1528. {

  1529.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1530.     av_free(a->coeff);

  1531.     a->coeff  = shifted->coeff;

  1532.     a->length = shifted->length;

  1533.     av_free(shifted);

  1534. }

  1535. 

  1536. void sws_addVec(SwsVector *a, SwsVector *b)

  1537. {

  1538.     SwsVector *sum = sws_sumVec(a, b);

  1539.     av_free(a->coeff);

  1540.     a->coeff  = sum->coeff;

  1541.     a->length = sum->length;

  1542.     av_free(sum);

  1543. }

  1544. 

  1545. void sws_subVec(SwsVector *a, SwsVector *b)

  1546. {

  1547.     SwsVector *diff = sws_diffVec(a, b);

  1548.     av_free(a->coeff);

  1549.     a->coeff  = diff->coeff;

  1550.     a->length = diff->length;

  1551.     av_free(diff);

  1552. }

  1553. 

  1554. void sws_convVec(SwsVector *a, SwsVector *b)

  1555. {

  1556.     SwsVector *conv = sws_getConvVec(a, b);

  1557.     av_free(a->coeff);

  1558.     a->coeff  = conv->coeff;

  1559.     a->length = conv->length;

  1560.     av_free(conv);

  1561. }

  1562. 

  1563. SwsVector *sws_cloneVec(SwsVector *a)

  1564. {

  1565.     int i;

  1566.     SwsVector *vec = sws_allocVec(a->length);

  1567. 

  1568.     if (!vec)

  1569.         return NULL;

  1570. 

  1571.     for (i = 0; i < a->length; i++)

  1572.         vec->coeff[i] = a->coeff[i];

  1573. 

  1574.     return vec;

  1575. }

  1576. 

  1577. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1578. {

  1579.     int i;

  1580.     double max = 0;

  1581.     double min = 0;

  1582.     double range;

  1583. 

  1584.     for (i = 0; i < a->length; i++)

  1585.         if (a->coeff[i] > max)

  1586.             max = a->coeff[i];

  1587. 

  1588.     for (i = 0; i < a->length; i++)

  1589.         if (a->coeff[i] < min)

  1590.             min = a->coeff[i];

  1591. 

  1592.     range = max - min;

  1593. 

  1594.     for (i = 0; i < a->length; i++) {

  1595.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1596.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1597.         for (; x > 0; x--)

  1598.             av_log(log_ctx, log_level, " ");

  1599.         av_log(log_ctx, log_level, "|\n");

  1600.     }

  1601. }

  1602. 

  1603. void sws_freeVec(SwsVector *a)

  1604. {

  1605.     if (!a)

  1606.         return;

  1607.     av_freep(&a->coeff);

  1608.     a->length = 0;

  1609.     av_free(a);

  1610. }

  1611. 

  1612. void sws_freeFilter(SwsFilter *filter)

  1613. {

  1614.     if (!filter)

  1615.         return;

  1616. 

  1617.     if (filter->lumH)

  1618.         sws_freeVec(filter->lumH);

  1619.     if (filter->lumV)

  1620.         sws_freeVec(filter->lumV);

  1621.     if (filter->chrH)

  1622.         sws_freeVec(filter->chrH);

  1623.     if (filter->chrV)

  1624.         sws_freeVec(filter->chrV);

  1625.     av_free(filter);

  1626. }

  1627. 

  1628. void sws_freeContext(SwsContext *c)

  1629. {

  1630.     int i;

  1631.     if (!c)

  1632.         return;

  1633. 

  1634.     if (c->lumPixBuf) {

  1635.         for (i = 0; i < c->vLumBufSize; i++)

  1636.             av_freep(&c->lumPixBuf[i]);

  1637.         av_freep(&c->lumPixBuf);

  1638.     }

  1639. 

  1640.     if (c->chrUPixBuf) {

  1641.         for (i = 0; i < c->vChrBufSize; i++)

  1642.             av_freep(&c->chrUPixBuf[i]);

  1643.         av_freep(&c->chrUPixBuf);

  1644.         av_freep(&c->chrVPixBuf);

  1645.     }

  1646. 

  1647.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1648.         for (i = 0; i < c->vLumBufSize; i++)

  1649.             av_freep(&c->alpPixBuf[i]);

  1650.         av_freep(&c->alpPixBuf);

  1651.     }

  1652. 

  1653.     av_freep(&c->vLumFilter);

  1654.     av_freep(&c->vChrFilter);

  1655.     av_freep(&c->hLumFilter);

  1656.     av_freep(&c->hChrFilter);

  1657. #if HAVE_ALTIVEC

  1658.     av_freep(&c->vYCoeffsBank);

  1659.     av_freep(&c->vCCoeffsBank);

  1660. #endif

  1661. 

  1662.     av_freep(&c->vLumFilterPos);

  1663.     av_freep(&c->vChrFilterPos);

  1664.     av_freep(&c->hLumFilterPos);

  1665.     av_freep(&c->hChrFilterPos);

  1666. 

  1667. #if HAVE_MMX_INLINE

  1668. #if USE_MMAP

  1669.     if (c->lumMmxextFilterCode)

  1670.         munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);

  1671.     if (c->chrMmxextFilterCode)

  1672.         munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);

  1673. #elif HAVE_VIRTUALALLOC

  1674.     if (c->lumMmxextFilterCode)

  1675.         VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);

  1676.     if (c->chrMmxextFilterCode)

  1677.         VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);

  1678. #else

  1679.     av_free(c->lumMmxextFilterCode);

  1680.     av_free(c->chrMmxextFilterCode);

  1681. #endif

  1682.     c->lumMmxextFilterCode = NULL;

  1683.     c->chrMmxextFilterCode = NULL;

  1684. #endif /* HAVE_MMX_INLINE */

  1685. 

  1686.     av_freep(&c->yuvTable);

  1687.     av_free(c->formatConvBuffer);

  1688. 

  1689.     av_free(c);

  1690. }

  1691. 

  1692. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  1693.                                         int srcH, enum AVPixelFormat srcFormat,

  1694.                                         int dstW, int dstH,

  1695.                                         enum AVPixelFormat dstFormat, int flags,

  1696.                                         SwsFilter *srcFilter,

  1697.                                         SwsFilter *dstFilter,

  1698.                                         const double *param)

  1699. {

  1700.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  1701.                                              SWS_PARAM_DEFAULT };

  1702. 

  1703.     if (!param)

  1704.         param = default_param;

  1705. 

  1706.     if (context &&

  1707.         (context->srcW      != srcW      ||

  1708.          context->srcH      != srcH      ||

  1709.          context->srcFormat != srcFormat ||

  1710.          context->dstW      != dstW      ||

  1711.          context->dstH      != dstH      ||

  1712.          context->dstFormat != dstFormat ||

  1713.          context->flags     != flags     ||

  1714.          context->param[0]  != param[0]  ||

  1715.          context->param[1]  != param[1])) {

  1716.         sws_freeContext(context);

  1717.         context = NULL;

  1718.     }

  1719. 

  1720.     if (!context) {

  1721.         if (!(context = sws_alloc_context()))

  1722.             return NULL;

  1723.         context->srcW      = srcW;

  1724.         context->srcH      = srcH;

  1725.         context->srcRange  = handle_jpeg(&srcFormat);

  1726.         context->srcFormat = srcFormat;

  1727.         context->dstW      = dstW;

  1728.         context->dstH      = dstH;

  1729.         context->dstRange  = handle_jpeg(&dstFormat);

  1730.         context->dstFormat = dstFormat;

  1731.         context->flags     = flags;

  1732.         context->param[0]  = param[0];

  1733.         context->param[1]  = param[1];

  1734.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  1735.                                  context->srcRange,

  1736.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1737.                                  context->dstRange, 0, 1 << 16, 1 << 16);

  1738.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1739.             sws_freeContext(context);

  1740.             return NULL;

  1741.         }

  1742.     }

  1743.     return context;

  1744. }