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.
35271 Commits
32 Branches
380MB
Tree: 22c8cbc0da
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '22c8cbc0da'
${ noResults }
FFmpeg/libswscale/utils.c

1708 lines
60KB
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_RGB48BE]     = { 1, 1 },

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  153. };

  154. 

  155. int sws_isSupportedInput(enum AVPixelFormat pix_fmt)

  156. {

  157.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  158.            format_entries[pix_fmt].is_supported_in : 0;

  159. }

  160. 

  161. int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)

  162. {

  163.     return (unsigned)pix_fmt < AV_PIX_FMT_NB ?

  164.            format_entries[pix_fmt].is_supported_out : 0;

  165. }

  166. 

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

  168. 

  169. const char *sws_format_name(enum AVPixelFormat format)

  170. {

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

  172.     if (desc)

  173.         return desc->name;

  174.     else

  175.         return "Unknown format";

  176. }

  177. 

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

  179.                              double dist)

  180. {

  181.     if (dist <= 1.0)

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

  183.     else

  184.         return getSplineCoeff(0.0,

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

  186.                                c + 3.0 * d,

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

  188.                               dist - 1.0);

  189. }

  190. 

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

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

  193.                       int filterAlign, int one, int flags, int cpu_flags,

  194.                       SwsVector *srcFilter, SwsVector *dstFilter,

  195.                       double param[2], int is_horizontal)

  196. {

  197.     int i;

  198.     int filterSize;

  199.     int filter2Size;

  200.     int minFilterSize;

  201.     int64_t *filter    = NULL;

  202.     int64_t *filter2   = NULL;

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

  204.     int ret            = -1;

  205. 

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

  207. 

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

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

  210. 

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

  212.         int i;

  213.         filterSize = 1;

  214.         FF_ALLOCZ_OR_GOTO(NULL, filter,

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

  216. 

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

  218.             filter[i * filterSize] = fone;

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

  220.         }

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

  222.         int i;

  223.         int xDstInSrc;

  224.         filterSize = 1;

  225.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  227. 

  228.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  231. 

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

  233.             filter[i]       = fone;

  234.             xDstInSrc      += xInc;

  235.         }

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

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

  238.         int i;

  239.         int xDstInSrc;

  240.         filterSize = 2;

  241.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  243. 

  244.         xDstInSrc = xInc / 2 - 0x8000;

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

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

  247.             int j;

  248. 

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

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

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

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

  253.                                 (fone >> 16);

  254.                 if (coeff < 0)

  255.                     coeff = 0;

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

  257.                 xx++;

  258.             }

  259.             xDstInSrc += xInc;

  260.         }

  261.     } else {

  262.         int64_t xDstInSrc;

  263.         int sizeFactor;

  264. 

  265.         if (flags & SWS_BICUBIC)

  266.             sizeFactor = 4;

  267.         else if (flags & SWS_X)

  268.             sizeFactor = 8;

  269.         else if (flags & SWS_AREA)

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

  271.         else if (flags & SWS_GAUSS)

  272.             sizeFactor = 8;     // infinite ;)

  273.         else if (flags & SWS_LANCZOS)

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

  275.         else if (flags & SWS_SINC)

  276.             sizeFactor = 20;    // infinite ;)

  277.         else if (flags & SWS_SPLINE)

  278.             sizeFactor = 20;    // infinite ;)

  279.         else if (flags & SWS_BILINEAR)

  280.             sizeFactor = 2;

  281.         else {

  282.             sizeFactor = 0;     // GCC warning killer

  283.             assert(0);

  284.         }

  285. 

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

  287.             filterSize = 1 + sizeFactor;    // upscale

  288.         else

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

  290. 

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

  292.         filterSize = FFMAX(filterSize, 1);

  293. 

  294.         FF_ALLOC_OR_GOTO(NULL, filter,

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

  296. 

  297.         xDstInSrc = xInc - 0x10000;

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

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

  300.             int j;

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

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

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

  304.                 double floatd;

  305.                 int64_t coeff;

  306. 

  307.                 if (xInc > 1 << 16)

  308.                     d = d * dstW / srcW;

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

  310. 

  311.                 if (flags & SWS_BICUBIC) {

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

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

  314. 

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

  316.                         coeff = 0.0;

  317.                     } else {

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

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

  320. 

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

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

  323.                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +

  324.                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);

  325.                         else

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

  327.                                       (6 * B + 30 * C) * dd  +

  328.                                     (-12 * B - 48 * C) * d   +

  329.                                       (8 * B + 24 * C) * (1 << 30);

  330.                     }

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

  332.                 }

  333. #if 0

  334.                 else if (flags & SWS_X) {

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

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

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

  338.                 }

  339. #endif

  340.                 else if (flags & SWS_X) {

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

  342.                     double c;

  343. 

  344.                     if (floatd < 1.0)

  345.                         c = cos(floatd * M_PI);

  346.                     else

  347.                         c = -1.0;

  348.                     if (c < 0.0)

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

  350.                     else

  351.                         c = pow(c, A);

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

  353.                 } else if (flags & SWS_AREA) {

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

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

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

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

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

  359.                     else

  360.                         coeff = 0.0;

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

  362.                 } else if (flags & SWS_GAUSS) {

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

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

  365.                 } else if (flags & SWS_SINC) {

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

  367.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  371.                     if (floatd > p)

  372.                         coeff = 0;

  373.                 } else if (flags & SWS_BILINEAR) {

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

  375.                     if (coeff < 0)

  376.                         coeff = 0;

  377.                     coeff *= fone >> 30;

  378.                 } else if (flags & SWS_SPLINE) {

  379.                     double p = -2.196152422706632;

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

  381.                 } else {

  382.                     coeff = 0.0; // GCC warning killer

  383.                     assert(0);

  384.                 }

  385. 

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

  387.                 xx++;

  388.             }

  389.             xDstInSrc += 2 * xInc;

  390.         }

  391.     }

  392. 

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

  394.      * av_free(filter);

  395.      */

  396.     assert(filterSize > 0);

  397.     filter2Size = filterSize;

  398.     if (srcFilter)

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

  400.     if (dstFilter)

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

  402.     assert(filter2Size > 0);

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

  404. 

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

  406.         int j, k;

  407. 

  408.         if (srcFilter) {

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

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

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

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

  413.             }

  414.         } else {

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

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

  417.         }

  418.         // FIXME dstFilter

  419. 

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

  421.     }

  422.     av_freep(&filter);

  423. 

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

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

  426.     minFilterSize = 0;

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

  428.         int min = filter2Size;

  429.         int j;

  430.         int64_t cutOff = 0.0;

  431. 

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

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

  434.             int k;

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

  436. 

  437.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  438.                 break;

  439. 

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

  441.              * filter otherwise */

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

  443.                 break;

  444. 

  445.             // move filter coefficients left

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

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

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

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

  450.         }

  451. 

  452.         cutOff = 0;

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

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

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

  456. 

  457.             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)

  458.                 break;

  459.             min--;

  460.         }

  461. 

  462.         if (min > minFilterSize)

  463.             minFilterSize = min;

  464.     }

  465. 

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

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

  468.         if (minFilterSize < 5)

  469.             filterAlign = 4;

  470. 

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

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

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

  474.         if (minFilterSize < 3)

  475.             filterAlign = 1;

  476.     }

  477. 

  478.     if (INLINE_MMX(cpu_flags)) {

  479.         // special case for unscaled vertical filtering

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

  481.             filterAlign = 1;

  482.     }

  483. 

  484.     assert(minFilterSize > 0);

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

  486.     assert(filterSize > 0);

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

  488.     if (filterSize >= MAX_FILTER_SIZE * 16 /

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

  490.         goto fail;

  491.     *outFilterSize = filterSize;

  492. 

  493.     if (flags & SWS_PRINT_INFO)

  494.         av_log(NULL, AV_LOG_VERBOSE,

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

  496.                filter2Size, filterSize);

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

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

  499.         int j;

  500. 

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

  502.             if (j >= filter2Size)

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

  504.             else

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

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

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

  508.         }

  509.     }

  510. 

  511.     // FIXME try to align filterPos if possible

  512. 

  513.     // fix borders

  514.     if (is_horizontal) {

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

  516.             int j;

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

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

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

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

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

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

  523.                 }

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

  525.             }

  526. 

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

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

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

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

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

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

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

  534.                 }

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

  536.             }

  537.         }

  538.     }

  539. 

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

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

  542.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,

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

  544. 

  545.     /* normalize & store in outFilter */

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

  547.         int j;

  548.         int64_t error = 0;

  549.         int64_t sum   = 0;

  550. 

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

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

  553.         }

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

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

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

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

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

  559.             error                                  = v - intV * sum;

  560.         }

  561.     }

  562. 

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

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

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

  566.                                                       * read over the end */

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

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

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

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

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

  572.     }

  573. 

  574.     ret = 0;

  575. 

  576. fail:

  577.     av_free(filter);

  578.     av_free(filter2);

  579.     return ret;

  580. }

  581. 

  582. #if HAVE_MMXEXT_INLINE

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

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

  585. {

  586.     uint8_t *fragmentA;

  587.     x86_reg imm8OfPShufW1A;

  588.     x86_reg imm8OfPShufW2A;

  589.     x86_reg fragmentLengthA;

  590.     uint8_t *fragmentB;

  591.     x86_reg imm8OfPShufW1B;

  592.     x86_reg imm8OfPShufW2B;

  593.     x86_reg fragmentLengthB;

  594.     int fragmentPos;

  595. 

  596.     int xpos, i;

  597. 

  598.     // create an optimized horizontal scaling routine

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

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

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

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

  603.      */

  604. 

  605.     // code fragment

  606. 

  607.     __asm__ volatile (

  608.         "jmp                         9f                 \n\t"

  609.         // Begin

  610.         "0:                                             \n\t"

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

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

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

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

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

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

  617.         "1:                                             \n\t"

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

  619.         "2:                                             \n\t"

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

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

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

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

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

  625. 

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

  627. 

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

  629.         // End

  630.         "9:                                             \n\t"

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

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

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

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

  635.         "dec                         %1                 \n\t"

  636.         "dec                         %2                 \n\t"

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

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

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

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

  641. 

  642. 

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

  644.           "=r" (fragmentLengthA)

  645.         );

  646. 

  647.     __asm__ volatile (

  648.         "jmp                         9f                 \n\t"

  649.         // Begin

  650.         "0:                                             \n\t"

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

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

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

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

  655.         "1:                                             \n\t"

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

  657.         "2:                                             \n\t"

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

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

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

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

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

  663. 

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

  665. 

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

  667.         // End

  668.         "9:                                             \n\t"

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

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

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

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

  673.         "dec                         %1                 \n\t"

  674.         "dec                         %2                 \n\t"

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

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

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

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

  679. 

  680. 

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

  682.           "=r" (fragmentLengthB)

  683.         );

  684. 

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

  686.     fragmentPos = 0;

  687. 

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

  689.         int xx = xpos >> 16;

  690. 

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

  692.             int a                  = 0;

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

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

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

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

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

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

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

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

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

  702.             int shift              = 0;

  703. 

  704.             if (filterCode) {

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

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

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

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

  709.                 filterPos[i / 2] = xx;

  710. 

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

  712. 

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

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

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

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

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

  718.                                                                (c << 4) |

  719.                                                                (d << 6);

  720. 

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

  722.                     shift = maxShift;               // avoid overread

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

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

  725. 

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

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

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

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

  730.                 }

  731.             }

  732. 

  733.             fragmentPos += fragmentLength;

  734. 

  735.             if (filterCode)

  736.                 filterCode[fragmentPos] = RET;

  737.         }

  738.         xpos += xInc;

  739.     }

  740.     if (filterCode)

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

  742. 

  743.     return fragmentPos + 1;

  744. }

  745. #endif /* HAVE_MMXEXT_INLINE */

  746. 

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

  748. {

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

  750.     *h = desc->log2_chroma_w;

  751.     *v = desc->log2_chroma_h;

  752. }

  753. 

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

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

  756.                              int brightness, int contrast, int saturation)

  757. {

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

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

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

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

  762. 

  763.     c->brightness = brightness;

  764.     c->contrast   = contrast;

  765.     c->saturation = saturation;

  766.     c->srcRange   = srcRange;

  767.     c->dstRange   = dstRange;

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

  769.         return -1;

  770. 

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

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

  773. 

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

  775.                              contrast, saturation);

  776.     // FIXME factorize

  777. 

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

  779.         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,

  780.                                        contrast, saturation);

  781.     return 0;

  782. }

  783. 

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

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

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

  787. {

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

  789.         return -1;

  790. 

  791.     *inv_table  = c->srcColorspaceTable;

  792.     *table      = c->dstColorspaceTable;

  793.     *srcRange   = c->srcRange;

  794.     *dstRange   = c->dstRange;

  795.     *brightness = c->brightness;

  796.     *contrast   = c->contrast;

  797.     *saturation = c->saturation;

  798. 

  799.     return 0;

  800. }

  801. 

  802. static int handle_jpeg(enum AVPixelFormat *format)

  803. {

  804.     switch (*format) {

  805.     case AV_PIX_FMT_YUVJ420P:

  806.         *format = AV_PIX_FMT_YUV420P;

  807.         return 1;

  808.     case AV_PIX_FMT_YUVJ422P:

  809.         *format = AV_PIX_FMT_YUV422P;

  810.         return 1;

  811.     case AV_PIX_FMT_YUVJ444P:

  812.         *format = AV_PIX_FMT_YUV444P;

  813.         return 1;

  814.     case AV_PIX_FMT_YUVJ440P:

  815.         *format = AV_PIX_FMT_YUV440P;

  816.         return 1;

  817.     default:

  818.         return 0;

  819.     }

  820. }

  821. 

  822. SwsContext *sws_alloc_context(void)

  823. {

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

  825. 

  826.     c->av_class = &sws_context_class;

  827.     av_opt_set_defaults(c);

  828. 

  829.     return c;

  830. }

  831. 

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

  833.                              SwsFilter *dstFilter)

  834. {

  835.     int i;

  836.     int usesVFilter, usesHFilter;

  837.     int unscaled;

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

  839.     int srcW              = c->srcW;

  840.     int srcH              = c->srcH;

  841.     int dstW              = c->dstW;

  842.     int dstH              = c->dstH;

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

  844.     int dst_stride_px     = dst_stride >> 1;

  845.     int flags, cpu_flags;

  846.     enum AVPixelFormat srcFormat = c->srcFormat;

  847.     enum AVPixelFormat dstFormat = c->dstFormat;

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

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

  850. 

  851.     cpu_flags = av_get_cpu_flags();

  852.     flags     = c->flags;

  853.     emms_c();

  854.     if (!rgb15to16)

  855.         sws_rgb2rgb_init();

  856. 

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

  858. 

  859.     if (!sws_isSupportedInput(srcFormat)) {

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

  861.                sws_format_name(srcFormat));

  862.         return AVERROR(EINVAL);

  863.     }

  864.     if (!sws_isSupportedOutput(dstFormat)) {

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

  866.                sws_format_name(dstFormat));

  867.         return AVERROR(EINVAL);

  868.     }

  869. 

  870.     i = flags & (SWS_POINT         |

  871.                  SWS_AREA          |

  872.                  SWS_BILINEAR      |

  873.                  SWS_FAST_BILINEAR |

  874.                  SWS_BICUBIC       |

  875.                  SWS_X             |

  876.                  SWS_GAUSS         |

  877.                  SWS_LANCZOS       |

  878.                  SWS_SINC          |

  879.                  SWS_SPLINE        |

  880.                  SWS_BICUBLIN);

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

  882.         av_log(c, AV_LOG_ERROR,

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

  884.         return AVERROR(EINVAL);

  885.     }

  886.     /* sanity check */

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

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

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

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

  891.                srcW, srcH, dstW, dstH);

  892.         return AVERROR(EINVAL);

  893.     }

  894. 

  895.     if (!dstFilter)

  896.         dstFilter = &dummyFilter;

  897.     if (!srcFilter)

  898.         srcFilter = &dummyFilter;

  899. 

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

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

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

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

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

  905. 

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

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

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

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

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

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

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

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

  914. 

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

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

  917. 

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

  919.      * chroma interpolation */

  920.     if (flags & SWS_FULL_CHR_H_INT &&

  921.         isAnyRGB(dstFormat)        &&

  922.         dstFormat != AV_PIX_FMT_RGBA  &&

  923.         dstFormat != AV_PIX_FMT_ARGB  &&

  924.         dstFormat != AV_PIX_FMT_BGRA  &&

  925.         dstFormat != AV_PIX_FMT_ABGR  &&

  926.         dstFormat != AV_PIX_FMT_RGB24 &&

  927.         dstFormat != AV_PIX_FMT_BGR24) {

  928.         av_log(c, AV_LOG_ERROR,

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

  930.                sws_format_name(dstFormat));

  931.         flags   &= ~SWS_FULL_CHR_H_INT;

  932.         c->flags = flags;

  933.     }

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

  935.         c->chrDstHSubSample = 1;

  936. 

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

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

  939.                            SWS_SRC_V_CHR_DROP_SHIFT;

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

  941. 

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

  943.      * wants full chroma */

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

  945.         srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&

  946.         srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&

  947.         srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&

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

  949.          (flags & SWS_FAST_BILINEAR)))

  950.         c->chrSrcHSubSample = 1;

  951. 

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

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

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

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

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

  957. 

  958.     /* unscaled special cases */

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

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

  961.         ff_get_unscaled_swscale(c);

  962. 

  963.         if (c->swScale) {

  964.             if (flags & SWS_PRINT_INFO)

  965.                 av_log(c, AV_LOG_INFO,

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

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

  968.             return 0;

  969.         }

  970.     }

  971. 

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

  973.     if (c->srcBpc < 8)

  974.         c->srcBpc = 8;

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

  976.     if (c->dstBpc < 8)

  977.         c->dstBpc = 8;

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

  979.         dst_stride <<= 1;

  980.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,

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

  982.                      fail);

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

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

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

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

  987.             && (flags & SWS_FAST_BILINEAR)) {

  988.             if (flags & SWS_PRINT_INFO)

  989.                 av_log(c, AV_LOG_INFO,

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

  991.         }

  992.         if (usesHFilter)

  993.             c->canMMX2BeUsed = 0;

  994.     } else

  995.         c->canMMX2BeUsed = 0;

  996. 

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

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

  999. 

  1000.     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src

  1001.      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do

  1002.      * correct scaling.

  1003.      * n-2 is the last chrominance sample available.

  1004.      * This is not perfect, but no one should notice the difference, the more

  1005.      * correct variant would be like the vertical one, but that would require

  1006.      * some special code for the first and last pixel */

  1007.     if (flags & SWS_FAST_BILINEAR) {

  1008.         if (c->canMMX2BeUsed) {

  1009.             c->lumXInc += 20;

  1010.             c->chrXInc += 20;

  1011.         }

  1012.         // we don't use the x86 asm scaler if MMX is available

  1013.         else if (INLINE_MMX(cpu_flags)) {

  1014.             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;

  1015.             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;

  1016.         }

  1017.     }

  1018. 

  1019.     /* precalculate horizontal scaler filter coefficients */

  1020.     {

  1021. #if HAVE_MMXEXT_INLINE

  1022. // can't downscale !!!

  1023.         if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {

  1024.             c->lumMmx2FilterCodeSize = initMMX2HScaler(dstW, c->lumXInc, NULL,

  1025.                                                        NULL, NULL, 8);

  1026.             c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc,

  1027.                                                        NULL, NULL, NULL, 4);

  1028. 

  1029. #ifdef MAP_ANONYMOUS

  1030.             c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);

  1031.             c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);

  1032. #elif HAVE_VIRTUALALLOC

  1033.             c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);

  1034.             c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);

  1035. #else

  1036.             c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);

  1037.             c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);

  1038. #endif

  1039. 

  1040.             if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)

  1041.                 return AVERROR(ENOMEM);

  1042.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);

  1043.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);

  1044.             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);

  1045.             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);

  1046. 

  1047.             initMMX2HScaler(dstW, c->lumXInc, c->lumMmx2FilterCode,

  1048.                             c->hLumFilter, c->hLumFilterPos, 8);

  1049.             initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode,

  1050.                             c->hChrFilter, c->hChrFilterPos, 4);

  1051. 

  1052. #ifdef MAP_ANONYMOUS

  1053.             mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);

  1054.             mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);

  1055. #endif

  1056.         } else

  1057. #endif /* HAVE_MMXEXT_INLINE */

  1058.         {

  1059.             const int filterAlign =

  1060.                 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :

  1061.                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1062.                 1;

  1063. 

  1064.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,

  1065.                            &c->hLumFilterSize, c->lumXInc,

  1066.                            srcW, dstW, filterAlign, 1 << 14,

  1067.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1068.                            cpu_flags, srcFilter->lumH, dstFilter->lumH,

  1069.                            c->param, 1) < 0)

  1070.                 goto fail;

  1071.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,

  1072.                            &c->hChrFilterSize, c->chrXInc,

  1073.                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,

  1074.                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1075.                            cpu_flags, srcFilter->chrH, dstFilter->chrH,

  1076.                            c->param, 1) < 0)

  1077.                 goto fail;

  1078.         }

  1079.     } // initialize horizontal stuff

  1080. 

  1081.     /* precalculate vertical scaler filter coefficients */

  1082.     {

  1083.         const int filterAlign =

  1084.             (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :

  1085.             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :

  1086.             1;

  1087. 

  1088.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,

  1089.                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),

  1090.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,

  1091.                        cpu_flags, srcFilter->lumV, dstFilter->lumV,

  1092.                        c->param, 0) < 0)

  1093.             goto fail;

  1094.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,

  1095.                        c->chrYInc, c->chrSrcH, c->chrDstH,

  1096.                        filterAlign, (1 << 12),

  1097.                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,

  1098.                        cpu_flags, srcFilter->chrV, dstFilter->chrV,

  1099.                        c->param, 0) < 0)

  1100.             goto fail;

  1101. 

  1102. #if HAVE_ALTIVEC

  1103.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);

  1104.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);

  1105. 

  1106.         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {

  1107.             int j;

  1108.             short *p = (short *)&c->vYCoeffsBank[i];

  1109.             for (j = 0; j < 8; j++)

  1110.                 p[j] = c->vLumFilter[i];

  1111.         }

  1112. 

  1113.         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {

  1114.             int j;

  1115.             short *p = (short *)&c->vCCoeffsBank[i];

  1116.             for (j = 0; j < 8; j++)

  1117.                 p[j] = c->vChrFilter[i];

  1118.         }

  1119. #endif

  1120.     }

  1121. 

  1122.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1123.     c->vLumBufSize = c->vLumFilterSize;

  1124.     c->vChrBufSize = c->vChrFilterSize;

  1125.     for (i = 0; i < dstH; i++) {

  1126.         int chrI      = (int64_t)i * c->chrDstH / dstH;

  1127.         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,

  1128.                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)

  1129.                                << c->chrSrcVSubSample));

  1130. 

  1131.         nextSlice >>= c->chrSrcVSubSample;

  1132.         nextSlice <<= c->chrSrcVSubSample;

  1133.         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)

  1134.             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];

  1135.         if (c->vChrFilterPos[chrI] + c->vChrBufSize <

  1136.             (nextSlice >> c->chrSrcVSubSample))

  1137.             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -

  1138.                              c->vChrFilterPos[chrI];

  1139.     }

  1140. 

  1141.     /* Allocate pixbufs (we use dynamic allocation because otherwise we would

  1142.      * need to allocate several megabytes to handle all possible cases) */

  1143.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1144.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1145.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);

  1146.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1147.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);

  1148.     /* Note we need at least one pixel more at the end because of the MMX code

  1149.      * (just in case someone wants to replace the 4000/8000). */

  1150.     /* align at 16 bytes for AltiVec */

  1151.     for (i = 0; i < c->vLumBufSize; i++) {

  1152.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],

  1153.                           dst_stride + 16, fail);

  1154.         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];

  1155.     }

  1156.     // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)

  1157.     c->uv_off_px   = dst_stride_px + 64 / (c->dstBpc & ~7);

  1158.     c->uv_off_byte = dst_stride + 16;

  1159.     for (i = 0; i < c->vChrBufSize; i++) {

  1160.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],

  1161.                          dst_stride * 2 + 32, fail);

  1162.         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];

  1163.         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]

  1164.                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1165.     }

  1166.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1167.         for (i = 0; i < c->vLumBufSize; i++) {

  1168.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],

  1169.                               dst_stride + 16, fail);

  1170.             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];

  1171.         }

  1172. 

  1173.     // try to avoid drawing green stuff between the right end and the stride end

  1174.     for (i = 0; i < c->vChrBufSize; i++)

  1175.         memset(c->chrUPixBuf[i], 64, dst_stride * 2 + 1);

  1176. 

  1177.     assert(c->chrDstH <= dstH);

  1178. 

  1179.     if (flags & SWS_PRINT_INFO) {

  1180.         if (flags & SWS_FAST_BILINEAR)

  1181.             av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1182.         else if (flags & SWS_BILINEAR)

  1183.             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1184.         else if (flags & SWS_BICUBIC)

  1185.             av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1186.         else if (flags & SWS_X)

  1187.             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1188.         else if (flags & SWS_POINT)

  1189.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1190.         else if (flags & SWS_AREA)

  1191.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1192.         else if (flags & SWS_BICUBLIN)

  1193.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1194.         else if (flags & SWS_GAUSS)

  1195.             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1196.         else if (flags & SWS_SINC)

  1197.             av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1198.         else if (flags & SWS_LANCZOS)

  1199.             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1200.         else if (flags & SWS_SPLINE)

  1201.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1202.         else

  1203.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1204. 

  1205.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1206.                sws_format_name(srcFormat),

  1207. #ifdef DITHER1XBPP

  1208.                dstFormat == AV_PIX_FMT_BGR555   || dstFormat == AV_PIX_FMT_BGR565   ||

  1209.                dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||

  1210.                dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?

  1211.                                                              "dithered " : "",

  1212. #else

  1213.                "",

  1214. #endif

  1215.                sws_format_name(dstFormat));

  1216. 

  1217.         if (INLINE_MMXEXT(cpu_flags))

  1218.             av_log(c, AV_LOG_INFO, "using MMX2\n");

  1219.         else if (INLINE_AMD3DNOW(cpu_flags))

  1220.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1221.         else if (INLINE_MMX(cpu_flags))

  1222.             av_log(c, AV_LOG_INFO, "using MMX\n");

  1223.         else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)

  1224.             av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1225.         else

  1226.             av_log(c, AV_LOG_INFO, "using C\n");

  1227. 

  1228.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1229.         av_log(c, AV_LOG_DEBUG,

  1230.                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1231.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1232.         av_log(c, AV_LOG_DEBUG,

  1233.                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1234.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,

  1235.                c->chrXInc, c->chrYInc);

  1236.     }

  1237. 

  1238.     c->swScale = ff_getSwsFunc(c);

  1239.     return 0;

  1240. fail: // FIXME replace things by appropriate error codes

  1241.     return -1;

  1242. }

  1243. 

  1244. #if FF_API_SWS_GETCONTEXT

  1245. SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,

  1246.                            int dstW, int dstH, enum AVPixelFormat dstFormat,

  1247.                            int flags, SwsFilter *srcFilter,

  1248.                            SwsFilter *dstFilter, const double *param)

  1249. {

  1250.     SwsContext *c;

  1251. 

  1252.     if (!(c = sws_alloc_context()))

  1253.         return NULL;

  1254. 

  1255.     c->flags     = flags;

  1256.     c->srcW      = srcW;

  1257.     c->srcH      = srcH;

  1258.     c->dstW      = dstW;

  1259.     c->dstH      = dstH;

  1260.     c->srcRange  = handle_jpeg(&srcFormat);

  1261.     c->dstRange  = handle_jpeg(&dstFormat);

  1262.     c->srcFormat = srcFormat;

  1263.     c->dstFormat = dstFormat;

  1264. 

  1265.     if (param) {

  1266.         c->param[0] = param[0];

  1267.         c->param[1] = param[1];

  1268.     }

  1269.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,

  1270.                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1271.                              c->dstRange, 0, 1 << 16, 1 << 16);

  1272. 

  1273.     if (sws_init_context(c, srcFilter, dstFilter) < 0) {

  1274.         sws_freeContext(c);

  1275.         return NULL;

  1276.     }

  1277. 

  1278.     return c;

  1279. }

  1280. #endif

  1281. 

  1282. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1283.                                 float lumaSharpen, float chromaSharpen,

  1284.                                 float chromaHShift, float chromaVShift,

  1285.                                 int verbose)

  1286. {

  1287.     SwsFilter *filter = av_malloc(sizeof(SwsFilter));

  1288.     if (!filter)

  1289.         return NULL;

  1290. 

  1291.     if (lumaGBlur != 0.0) {

  1292.         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);

  1293.         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);

  1294.     } else {

  1295.         filter->lumH = sws_getIdentityVec();

  1296.         filter->lumV = sws_getIdentityVec();

  1297.     }

  1298. 

  1299.     if (chromaGBlur != 0.0) {

  1300.         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);

  1301.         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);

  1302.     } else {

  1303.         filter->chrH = sws_getIdentityVec();

  1304.         filter->chrV = sws_getIdentityVec();

  1305.     }

  1306. 

  1307.     if (chromaSharpen != 0.0) {

  1308.         SwsVector *id = sws_getIdentityVec();

  1309.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1310.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1311.         sws_addVec(filter->chrH, id);

  1312.         sws_addVec(filter->chrV, id);

  1313.         sws_freeVec(id);

  1314.     }

  1315. 

  1316.     if (lumaSharpen != 0.0) {

  1317.         SwsVector *id = sws_getIdentityVec();

  1318.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1319.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1320.         sws_addVec(filter->lumH, id);

  1321.         sws_addVec(filter->lumV, id);

  1322.         sws_freeVec(id);

  1323.     }

  1324. 

  1325.     if (chromaHShift != 0.0)

  1326.         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));

  1327. 

  1328.     if (chromaVShift != 0.0)

  1329.         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));

  1330. 

  1331.     sws_normalizeVec(filter->chrH, 1.0);

  1332.     sws_normalizeVec(filter->chrV, 1.0);

  1333.     sws_normalizeVec(filter->lumH, 1.0);

  1334.     sws_normalizeVec(filter->lumV, 1.0);

  1335. 

  1336.     if (verbose)

  1337.         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1338.     if (verbose)

  1339.         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1340. 

  1341.     return filter;

  1342. }

  1343. 

  1344. SwsVector *sws_allocVec(int length)

  1345. {

  1346.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1347.     if (!vec)

  1348.         return NULL;

  1349.     vec->length = length;

  1350.     vec->coeff  = av_malloc(sizeof(double) * length);

  1351.     if (!vec->coeff)

  1352.         av_freep(&vec);

  1353.     return vec;

  1354. }

  1355. 

  1356. SwsVector *sws_getGaussianVec(double variance, double quality)

  1357. {

  1358.     const int length = (int)(variance * quality + 0.5) | 1;

  1359.     int i;

  1360.     double middle  = (length - 1) * 0.5;

  1361.     SwsVector *vec = sws_allocVec(length);

  1362. 

  1363.     if (!vec)

  1364.         return NULL;

  1365. 

  1366.     for (i = 0; i < length; i++) {

  1367.         double dist = i - middle;

  1368.         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /

  1369.                         sqrt(2 * variance * M_PI);

  1370.     }

  1371. 

  1372.     sws_normalizeVec(vec, 1.0);

  1373. 

  1374.     return vec;

  1375. }

  1376. 

  1377. SwsVector *sws_getConstVec(double c, int length)

  1378. {

  1379.     int i;

  1380.     SwsVector *vec = sws_allocVec(length);

  1381. 

  1382.     if (!vec)

  1383.         return NULL;

  1384. 

  1385.     for (i = 0; i < length; i++)

  1386.         vec->coeff[i] = c;

  1387. 

  1388.     return vec;

  1389. }

  1390. 

  1391. SwsVector *sws_getIdentityVec(void)

  1392. {

  1393.     return sws_getConstVec(1.0, 1);

  1394. }

  1395. 

  1396. static double sws_dcVec(SwsVector *a)

  1397. {

  1398.     int i;

  1399.     double sum = 0;

  1400. 

  1401.     for (i = 0; i < a->length; i++)

  1402.         sum += a->coeff[i];

  1403. 

  1404.     return sum;

  1405. }

  1406. 

  1407. void sws_scaleVec(SwsVector *a, double scalar)

  1408. {

  1409.     int i;

  1410. 

  1411.     for (i = 0; i < a->length; i++)

  1412.         a->coeff[i] *= scalar;

  1413. }

  1414. 

  1415. void sws_normalizeVec(SwsVector *a, double height)

  1416. {

  1417.     sws_scaleVec(a, height / sws_dcVec(a));

  1418. }

  1419. 

  1420. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1421. {

  1422.     int length = a->length + b->length - 1;

  1423.     int i, j;

  1424.     SwsVector *vec = sws_getConstVec(0.0, length);

  1425. 

  1426.     if (!vec)

  1427.         return NULL;

  1428. 

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

  1430.         for (j = 0; j < b->length; j++) {

  1431.             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];

  1432.         }

  1433.     }

  1434. 

  1435.     return vec;

  1436. }

  1437. 

  1438. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1439. {

  1440.     int length = FFMAX(a->length, b->length);

  1441.     int i;

  1442.     SwsVector *vec = sws_getConstVec(0.0, length);

  1443. 

  1444.     if (!vec)

  1445.         return NULL;

  1446. 

  1447.     for (i = 0; i < a->length; i++)

  1448.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

  1449.     for (i = 0; i < b->length; i++)

  1450.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];

  1451. 

  1452.     return vec;

  1453. }

  1454. 

  1455. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1456. {

  1457.     int length = FFMAX(a->length, b->length);

  1458.     int i;

  1459.     SwsVector *vec = sws_getConstVec(0.0, length);

  1460. 

  1461.     if (!vec)

  1462.         return NULL;

  1463. 

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

  1465.         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];

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

  1467.         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];

  1468. 

  1469.     return vec;

  1470. }

  1471. 

  1472. /* shift left / or right if "shift" is negative */

  1473. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1474. {

  1475.     int length = a->length + FFABS(shift) * 2;

  1476.     int i;

  1477.     SwsVector *vec = sws_getConstVec(0.0, length);

  1478. 

  1479.     if (!vec)

  1480.         return NULL;

  1481. 

  1482.     for (i = 0; i < a->length; i++) {

  1483.         vec->coeff[i + (length    - 1) / 2 -

  1484.                        (a->length - 1) / 2 - shift] = a->coeff[i];

  1485.     }

  1486. 

  1487.     return vec;

  1488. }

  1489. 

  1490. void sws_shiftVec(SwsVector *a, int shift)

  1491. {

  1492.     SwsVector *shifted = sws_getShiftedVec(a, shift);

  1493.     av_free(a->coeff);

  1494.     a->coeff  = shifted->coeff;

  1495.     a->length = shifted->length;

  1496.     av_free(shifted);

  1497. }

  1498. 

  1499. void sws_addVec(SwsVector *a, SwsVector *b)

  1500. {

  1501.     SwsVector *sum = sws_sumVec(a, b);

  1502.     av_free(a->coeff);

  1503.     a->coeff  = sum->coeff;

  1504.     a->length = sum->length;

  1505.     av_free(sum);

  1506. }

  1507. 

  1508. void sws_subVec(SwsVector *a, SwsVector *b)

  1509. {

  1510.     SwsVector *diff = sws_diffVec(a, b);

  1511.     av_free(a->coeff);

  1512.     a->coeff  = diff->coeff;

  1513.     a->length = diff->length;

  1514.     av_free(diff);

  1515. }

  1516. 

  1517. void sws_convVec(SwsVector *a, SwsVector *b)

  1518. {

  1519.     SwsVector *conv = sws_getConvVec(a, b);

  1520.     av_free(a->coeff);

  1521.     a->coeff  = conv->coeff;

  1522.     a->length = conv->length;

  1523.     av_free(conv);

  1524. }

  1525. 

  1526. SwsVector *sws_cloneVec(SwsVector *a)

  1527. {

  1528.     int i;

  1529.     SwsVector *vec = sws_allocVec(a->length);

  1530. 

  1531.     if (!vec)

  1532.         return NULL;

  1533. 

  1534.     for (i = 0; i < a->length; i++)

  1535.         vec->coeff[i] = a->coeff[i];

  1536. 

  1537.     return vec;

  1538. }

  1539. 

  1540. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1541. {

  1542.     int i;

  1543.     double max = 0;

  1544.     double min = 0;

  1545.     double range;

  1546. 

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

  1548.         if (a->coeff[i] > max)

  1549.             max = a->coeff[i];

  1550. 

  1551.     for (i = 0; i < a->length; i++)

  1552.         if (a->coeff[i] < min)

  1553.             min = a->coeff[i];

  1554. 

  1555.     range = max - min;

  1556. 

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

  1558.         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);

  1559.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1560.         for (; x > 0; x--)

  1561.             av_log(log_ctx, log_level, " ");

  1562.         av_log(log_ctx, log_level, "|\n");

  1563.     }

  1564. }

  1565. 

  1566. void sws_freeVec(SwsVector *a)

  1567. {

  1568.     if (!a)

  1569.         return;

  1570.     av_freep(&a->coeff);

  1571.     a->length = 0;

  1572.     av_free(a);

  1573. }

  1574. 

  1575. void sws_freeFilter(SwsFilter *filter)

  1576. {

  1577.     if (!filter)

  1578.         return;

  1579. 

  1580.     if (filter->lumH)

  1581.         sws_freeVec(filter->lumH);

  1582.     if (filter->lumV)

  1583.         sws_freeVec(filter->lumV);

  1584.     if (filter->chrH)

  1585.         sws_freeVec(filter->chrH);

  1586.     if (filter->chrV)

  1587.         sws_freeVec(filter->chrV);

  1588.     av_free(filter);

  1589. }

  1590. 

  1591. void sws_freeContext(SwsContext *c)

  1592. {

  1593.     int i;

  1594.     if (!c)

  1595.         return;

  1596. 

  1597.     if (c->lumPixBuf) {

  1598.         for (i = 0; i < c->vLumBufSize; i++)

  1599.             av_freep(&c->lumPixBuf[i]);

  1600.         av_freep(&c->lumPixBuf);

  1601.     }

  1602. 

  1603.     if (c->chrUPixBuf) {

  1604.         for (i = 0; i < c->vChrBufSize; i++)

  1605.             av_freep(&c->chrUPixBuf[i]);

  1606.         av_freep(&c->chrUPixBuf);

  1607.         av_freep(&c->chrVPixBuf);

  1608.     }

  1609. 

  1610.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1611.         for (i = 0; i < c->vLumBufSize; i++)

  1612.             av_freep(&c->alpPixBuf[i]);

  1613.         av_freep(&c->alpPixBuf);

  1614.     }

  1615. 

  1616.     av_freep(&c->vLumFilter);

  1617.     av_freep(&c->vChrFilter);

  1618.     av_freep(&c->hLumFilter);

  1619.     av_freep(&c->hChrFilter);

  1620. #if HAVE_ALTIVEC

  1621.     av_freep(&c->vYCoeffsBank);

  1622.     av_freep(&c->vCCoeffsBank);

  1623. #endif

  1624. 

  1625.     av_freep(&c->vLumFilterPos);

  1626.     av_freep(&c->vChrFilterPos);

  1627.     av_freep(&c->hLumFilterPos);

  1628.     av_freep(&c->hChrFilterPos);

  1629. 

  1630. #if HAVE_MMX_INLINE

  1631. #ifdef MAP_ANONYMOUS

  1632.     if (c->lumMmx2FilterCode)

  1633.         munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1634.     if (c->chrMmx2FilterCode)

  1635.         munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1636. #elif HAVE_VIRTUALALLOC

  1637.     if (c->lumMmx2FilterCode)

  1638.         VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1639.     if (c->chrMmx2FilterCode)

  1640.         VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1641. #else

  1642.     av_free(c->lumMmx2FilterCode);

  1643.     av_free(c->chrMmx2FilterCode);

  1644. #endif

  1645.     c->lumMmx2FilterCode = NULL;

  1646.     c->chrMmx2FilterCode = NULL;

  1647. #endif /* HAVE_MMX_INLINE */

  1648. 

  1649.     av_freep(&c->yuvTable);

  1650.     av_free(c->formatConvBuffer);

  1651. 

  1652.     av_free(c);

  1653. }

  1654. 

  1655. struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,

  1656.                                         int srcH, enum AVPixelFormat srcFormat,

  1657.                                         int dstW, int dstH,

  1658.                                         enum AVPixelFormat dstFormat, int flags,

  1659.                                         SwsFilter *srcFilter,

  1660.                                         SwsFilter *dstFilter,

  1661.                                         const double *param)

  1662. {

  1663.     static const double default_param[2] = { SWS_PARAM_DEFAULT,

  1664.                                              SWS_PARAM_DEFAULT };

  1665. 

  1666.     if (!param)

  1667.         param = default_param;

  1668. 

  1669.     if (context &&

  1670.         (context->srcW      != srcW      ||

  1671.          context->srcH      != srcH      ||

  1672.          context->srcFormat != srcFormat ||

  1673.          context->dstW      != dstW      ||

  1674.          context->dstH      != dstH      ||

  1675.          context->dstFormat != dstFormat ||

  1676.          context->flags     != flags     ||

  1677.          context->param[0]  != param[0]  ||

  1678.          context->param[1]  != param[1])) {

  1679.         sws_freeContext(context);

  1680.         context = NULL;

  1681.     }

  1682. 

  1683.     if (!context) {

  1684.         if (!(context = sws_alloc_context()))

  1685.             return NULL;

  1686.         context->srcW      = srcW;

  1687.         context->srcH      = srcH;

  1688.         context->srcRange  = handle_jpeg(&srcFormat);

  1689.         context->srcFormat = srcFormat;

  1690.         context->dstW      = dstW;

  1691.         context->dstH      = dstH;

  1692.         context->dstRange  = handle_jpeg(&dstFormat);

  1693.         context->dstFormat = dstFormat;

  1694.         context->flags     = flags;

  1695.         context->param[0]  = param[0];

  1696.         context->param[1]  = param[1];

  1697.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],

  1698.                                  context->srcRange,

  1699.                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,

  1700.                                  context->dstRange, 0, 1 << 16, 1 << 16);

  1701.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1702.             sws_freeContext(context);

  1703.             return NULL;

  1704.         }

  1705.     }

  1706.     return context;

  1707. }