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.
30822 Commits
32 Branches
552MB
Tree: 0c6cf3fa7c
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '0c6cf3fa7c'
${ noResults }
FFmpeg/libswscale/utils.c

1603 lines
58KB
Raw Blame History 

  1. /*

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

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

  6.  * FFmpeg 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.  * FFmpeg 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 FFmpeg; if not, write to the Free Software

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

  19.  */

  20. 

  21. #define _SVID_SOURCE //needed for MAP_ANONYMOUS

  22. #define _DARWIN_C_SOURCE // needed for MAP_ANON

  23. #include <inttypes.h>

  24. #include <string.h>

  25. #include <math.h>

  26. #include <stdio.h>

  27. #include "config.h"

  28. #include <assert.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. #include "swscale.h"

  40. #include "swscale_internal.h"

  41. #include "rgb2rgb.h"

  42. #include "libavutil/intreadwrite.h"

  43. #include "libavutil/x86_cpu.h"

  44. #include "libavutil/cpu.h"

  45. #include "libavutil/avutil.h"

  46. #include "libavutil/bswap.h"

  47. #include "libavutil/opt.h"

  48. #include "libavutil/pixdesc.h"

  49. 

  50. unsigned swscale_version(void)

  51. {

  52.     return LIBSWSCALE_VERSION_INT;

  53. }

  54. 

  55. const char *swscale_configuration(void)

  56. {

  57.     return FFMPEG_CONFIGURATION;

  58. }

  59. 

  60. const char *swscale_license(void)

  61. {

  62. #define LICENSE_PREFIX "libswscale license: "

  63.     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;

  64. }

  65. 

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

  67. 

  68. #define isSupportedIn(x)    (       \

  69.            (x)==PIX_FMT_YUV420P     \

  70.         || (x)==PIX_FMT_YUVA420P    \

  71.         || (x)==PIX_FMT_YUYV422     \

  72.         || (x)==PIX_FMT_UYVY422     \

  73.         || (x)==PIX_FMT_RGB48BE     \

  74.         || (x)==PIX_FMT_RGB48LE     \

  75.         || (x)==PIX_FMT_RGB32       \

  76.         || (x)==PIX_FMT_RGB32_1     \

  77.         || (x)==PIX_FMT_BGR48BE     \

  78.         || (x)==PIX_FMT_BGR48LE     \

  79.         || (x)==PIX_FMT_BGR24       \

  80.         || (x)==PIX_FMT_BGR565LE    \

  81.         || (x)==PIX_FMT_BGR565BE    \

  82.         || (x)==PIX_FMT_BGR555LE    \

  83.         || (x)==PIX_FMT_BGR555BE    \

  84.         || (x)==PIX_FMT_BGR32       \

  85.         || (x)==PIX_FMT_BGR32_1     \

  86.         || (x)==PIX_FMT_RGB24       \

  87.         || (x)==PIX_FMT_RGB565LE    \

  88.         || (x)==PIX_FMT_RGB565BE    \

  89.         || (x)==PIX_FMT_RGB555LE    \

  90.         || (x)==PIX_FMT_RGB555BE    \

  91.         || (x)==PIX_FMT_GRAY8       \

  92.         || (x)==PIX_FMT_GRAY8A      \

  93.         || (x)==PIX_FMT_YUV410P     \

  94.         || (x)==PIX_FMT_YUV440P     \

  95.         || (x)==PIX_FMT_NV12        \

  96.         || (x)==PIX_FMT_NV21        \

  97.         || (x)==PIX_FMT_GRAY16BE    \

  98.         || (x)==PIX_FMT_GRAY16LE    \

  99.         || (x)==PIX_FMT_YUV444P     \

  100.         || (x)==PIX_FMT_YUV422P     \

  101.         || (x)==PIX_FMT_YUV411P     \

  102.         || (x)==PIX_FMT_YUVJ420P    \

  103.         || (x)==PIX_FMT_YUVJ422P    \

  104.         || (x)==PIX_FMT_YUVJ440P    \

  105.         || (x)==PIX_FMT_YUVJ444P    \

  106.         || (x)==PIX_FMT_PAL8        \

  107.         || (x)==PIX_FMT_BGR8        \

  108.         || (x)==PIX_FMT_RGB8        \

  109.         || (x)==PIX_FMT_BGR4_BYTE   \

  110.         || (x)==PIX_FMT_RGB4_BYTE   \

  111.         || (x)==PIX_FMT_YUV440P     \

  112.         || (x)==PIX_FMT_MONOWHITE   \

  113.         || (x)==PIX_FMT_MONOBLACK   \

  114.         || (x)==PIX_FMT_YUV420P9LE    \

  115.         || (x)==PIX_FMT_YUV444P9LE    \

  116.         || (x)==PIX_FMT_YUV420P10LE   \

  117.         || (x)==PIX_FMT_YUV422P10LE   \

  118.         || (x)==PIX_FMT_YUV444P10LE   \

  119.         || (x)==PIX_FMT_YUV420P16LE   \

  120.         || (x)==PIX_FMT_YUV422P16LE   \

  121.         || (x)==PIX_FMT_YUV444P16LE   \

  122.         || (x)==PIX_FMT_YUV420P9BE    \

  123.         || (x)==PIX_FMT_YUV444P9BE    \

  124.         || (x)==PIX_FMT_YUV420P10BE   \

  125.         || (x)==PIX_FMT_YUV444P10BE   \

  126.         || (x)==PIX_FMT_YUV422P10BE   \

  127.         || (x)==PIX_FMT_YUV420P16BE   \

  128.         || (x)==PIX_FMT_YUV422P16BE   \

  129.         || (x)==PIX_FMT_YUV444P16BE   \

  130.         || (x)==PIX_FMT_YUV422P10     \

  131.     )

  132. 

  133. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  134. {

  135.     return isSupportedIn(pix_fmt);

  136. }

  137. 

  138. #define isSupportedOut(x)   (       \

  139.            (x)==PIX_FMT_YUV420P     \

  140.         || (x)==PIX_FMT_YUVA420P    \

  141.         || (x)==PIX_FMT_YUYV422     \

  142.         || (x)==PIX_FMT_UYVY422     \

  143.         || (x)==PIX_FMT_YUV444P     \

  144.         || (x)==PIX_FMT_YUV422P     \

  145.         || (x)==PIX_FMT_YUV411P     \

  146.         || (x)==PIX_FMT_YUVJ420P    \

  147.         || (x)==PIX_FMT_YUVJ422P    \

  148.         || (x)==PIX_FMT_YUVJ440P    \

  149.         || (x)==PIX_FMT_YUVJ444P    \

  150.         || isRGBinBytes(x)          \

  151.         || isBGRinBytes(x)          \

  152.         || (x)==PIX_FMT_RGB565      \

  153.         || (x)==PIX_FMT_RGB555      \

  154.         || (x)==PIX_FMT_RGB444      \

  155.         || (x)==PIX_FMT_BGR565      \

  156.         || (x)==PIX_FMT_BGR555      \

  157.         || (x)==PIX_FMT_BGR444      \

  158.         || (x)==PIX_FMT_RGB8        \

  159.         || (x)==PIX_FMT_BGR8        \

  160.         || (x)==PIX_FMT_RGB4_BYTE   \

  161.         || (x)==PIX_FMT_BGR4_BYTE   \

  162.         || (x)==PIX_FMT_RGB4        \

  163.         || (x)==PIX_FMT_BGR4        \

  164.         || (x)==PIX_FMT_MONOBLACK   \

  165.         || (x)==PIX_FMT_MONOWHITE   \

  166.         || (x)==PIX_FMT_NV12        \

  167.         || (x)==PIX_FMT_NV21        \

  168.         || (x)==PIX_FMT_GRAY16BE    \

  169.         || (x)==PIX_FMT_GRAY16LE    \

  170.         || (x)==PIX_FMT_GRAY8       \

  171.         || (x)==PIX_FMT_YUV410P     \

  172.         || (x)==PIX_FMT_YUV440P     \

  173.         || (x)==PIX_FMT_YUV422P10   \

  174.         || (x)==PIX_FMT_YUV420P9LE    \

  175.         || (x)==PIX_FMT_YUV420P10LE   \

  176.         || (x)==PIX_FMT_YUV420P16LE   \

  177.         || (x)==PIX_FMT_YUV422P16LE   \

  178.         || (x)==PIX_FMT_YUV444P16LE   \

  179.         || (x)==PIX_FMT_YUV420P9BE    \

  180.         || (x)==PIX_FMT_YUV420P10BE   \

  181.         || (x)==PIX_FMT_YUV420P16BE   \

  182.         || (x)==PIX_FMT_YUV422P16BE   \

  183.         || (x)==PIX_FMT_YUV444P16BE   \

  184.     )

  185. 

  186. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  187. {

  188.     return isSupportedOut(pix_fmt);

  189. }

  190. 

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

  192. 

  193. #if FF_API_SWS_FORMAT_NAME

  194. const char *sws_format_name(enum PixelFormat format)

  195. {

  196.     return av_get_pix_fmt_name(format);

  197. }

  198. #endif

  199. 

  200. static double getSplineCoeff(double a, double b, double c, double d, double dist)

  201. {

  202.     if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;

  203.     else           return getSplineCoeff(        0.0,

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

  205.                                                  c + 3.0*d,

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

  207.                                          dist-1.0);

  208. }

  209. 

  210. static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,

  211.                       int srcW, int dstW, int filterAlign, int one, int flags, int cpu_flags,

  212.                       SwsVector *srcFilter, SwsVector *dstFilter, double param[2])

  213. {

  214.     int i;

  215.     int filterSize;

  216.     int filter2Size;

  217.     int minFilterSize;

  218.     int64_t *filter=NULL;

  219.     int64_t *filter2=NULL;

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

  221.     int ret= -1;

  222. 

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

  224. 

  225.     // NOTE: the +1 is for the MMX scaler which reads over the end

  226.     FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);

  227. 

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

  229.         int i;

  230.         filterSize= 1;

  231.         FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);

  232. 

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

  234.             filter[i*filterSize]= fone;

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

  236.         }

  237. 

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

  239.         int i;

  240.         int xDstInSrc;

  241.         filterSize= 1;

  242.         FF_ALLOC_OR_GOTO(NULL, filter, 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. 

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

  249.             filter[i]= fone;

  250.             xDstInSrc+= xInc;

  251.         }

  252.     } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale

  253.         int i;

  254.         int xDstInSrc;

  255.         filterSize= 2;

  256.         FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);

  257. 

  258.         xDstInSrc= xInc/2 - 0x8000;

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

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

  261.             int j;

  262. 

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

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

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

  266.                 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);

  267.                 if (coeff<0) coeff=0;

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

  269.                 xx++;

  270.             }

  271.             xDstInSrc+= xInc;

  272.         }

  273.     } else {

  274.         int xDstInSrc;

  275.         int sizeFactor;

  276. 

  277.         if      (flags&SWS_BICUBIC)      sizeFactor=  4;

  278.         else if (flags&SWS_X)            sizeFactor=  8;

  279.         else if (flags&SWS_AREA)         sizeFactor=  1; //downscale only, for upscale it is bilinear

  280.         else if (flags&SWS_GAUSS)        sizeFactor=  8;   // infinite ;)

  281.         else if (flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;

  282.         else if (flags&SWS_SINC)         sizeFactor= 20; // infinite ;)

  283.         else if (flags&SWS_SPLINE)       sizeFactor= 20;  // infinite ;)

  284.         else if (flags&SWS_BILINEAR)     sizeFactor=  2;

  285.         else {

  286.             sizeFactor= 0; //GCC warning killer

  287.             assert(0);

  288.         }

  289. 

  290.         if (xInc <= 1<<16)      filterSize= 1 + sizeFactor; // upscale

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

  292. 

  293.         if (filterSize > srcW-2) filterSize=srcW-2;

  294. 

  295.         FF_ALLOC_OR_GOTO(NULL, filter, 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= ((int64_t)FFABS((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.                     int64_t dd = ( d*d)>>30;

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

  316. 

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

  318.                         coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);

  319.                     else if (d < 1LL<<31)

  320.                         coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);

  321.                     else

  322.                         coeff=0.0;

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

  324.                 }

  325. /*                else if (flags & SWS_X) {

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

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

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

  329.                 }*/

  330.                 else if (flags & SWS_X) {

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

  332.                     double c;

  333. 

  334.                     if (floatd<1.0)

  335.                         c = cos(floatd*M_PI);

  336.                     else

  337.                         c=-1.0;

  338.                     if (c<0.0)      c= -pow(-c, A);

  339.                     else            c=  pow( c, A);

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

  341.                 } else if (flags & SWS_AREA) {

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

  343.                     if      (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));

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

  345.                     else coeff=0.0;

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

  347.                 } else if (flags & SWS_GAUSS) {

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

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

  350.                 } else if (flags & SWS_SINC) {

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

  352.                 } else if (flags & SWS_LANCZOS) {

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

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

  355.                     if (floatd>p) coeff=0;

  356.                 } else if (flags & SWS_BILINEAR) {

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

  358.                     if (coeff<0) coeff=0;

  359.                     coeff *= fone >> 30;

  360.                 } else if (flags & SWS_SPLINE) {

  361.                     double p=-2.196152422706632;

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

  363.                 } else {

  364.                     coeff= 0.0; //GCC warning killer

  365.                     assert(0);

  366.                 }

  367. 

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

  369.                 xx++;

  370.             }

  371.             xDstInSrc+= 2*xInc;

  372.         }

  373.     }

  374. 

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

  376.        av_free(filter);

  377.     */

  378.     assert(filterSize>0);

  379.     filter2Size= filterSize;

  380.     if (srcFilter) filter2Size+= srcFilter->length - 1;

  381.     if (dstFilter) filter2Size+= dstFilter->length - 1;

  382.     assert(filter2Size>0);

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

  384. 

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

  386.         int j, k;

  387. 

  388.         if(srcFilter) {

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

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

  391.                     filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];

  392.             }

  393.         } else {

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

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

  396.         }

  397.         //FIXME dstFilter

  398. 

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

  400.     }

  401.     av_freep(&filter);

  402. 

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

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

  405.     minFilterSize= 0;

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

  407.         int min= filter2Size;

  408.         int j;

  409.         int64_t cutOff=0.0;

  410. 

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

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

  413.             int k;

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

  415. 

  416.             if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;

  417. 

  418.             /* preserve monotonicity because the core can't handle the filter otherwise */

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

  420. 

  421.             // move filter coefficients left

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

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

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

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

  426.         }

  427. 

  428.         cutOff=0;

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

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

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

  432. 

  433.             if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;

  434.             min--;

  435.         }

  436. 

  437.         if (min>minFilterSize) minFilterSize= min;

  438.     }

  439. 

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

  441.         // we can handle the special case 4,

  442.         // so we don't want to go to the full 8

  443.         if (minFilterSize < 5)

  444.             filterAlign = 4;

  445. 

  446.         // We really don't want to waste our time

  447.         // doing useless computation, so fall back on

  448.         // the scalar C code for very small filters.

  449.         // Vectorizing is worth it only if you have a

  450.         // decent-sized vector.

  451.         if (minFilterSize < 3)

  452.             filterAlign = 1;

  453.     }

  454. 

  455.     if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {

  456.         // special case for unscaled vertical filtering

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

  458.             filterAlign= 1;

  459.     }

  460. 

  461.     assert(minFilterSize > 0);

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

  463.     assert(filterSize > 0);

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

  465.     if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)

  466.         goto fail;

  467.     *outFilterSize= filterSize;

  468. 

  469.     if (flags&SWS_PRINT_INFO)

  470.         av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);

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

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

  473.         int j;

  474. 

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

  476.             if (j>=filter2Size) filter[i*filterSize + j]= 0;

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

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

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

  480.         }

  481.     }

  482. 

  483.     //FIXME try to align filterPos if possible

  484. 

  485.     //fix borders

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

  487.         int j;

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

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

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

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

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

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

  494.             }

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

  496.         }

  497. 

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

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

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

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

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

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

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

  505.             }

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

  507.         }

  508.     }

  509. 

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

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

  512.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);

  513. 

  514.     /* normalize & store in outFilter */

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

  516.         int j;

  517.         int64_t error=0;

  518.         int64_t sum=0;

  519. 

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

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

  522.         }

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

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

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

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

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

  528.             error= v - intV*sum;

  529.         }

  530.     }

  531. 

  532.     (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end

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

  534.         int j= dstW*(*outFilterSize);

  535.         (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];

  536.     }

  537. 

  538.     ret=0;

  539. fail:

  540.     av_free(filter);

  541.     av_free(filter2);

  542.     return ret;

  543. }

  544. 

  545. #if HAVE_MMX2

  546. static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)

  547. {

  548.     uint8_t *fragmentA;

  549.     x86_reg imm8OfPShufW1A;

  550.     x86_reg imm8OfPShufW2A;

  551.     x86_reg fragmentLengthA;

  552.     uint8_t *fragmentB;

  553.     x86_reg imm8OfPShufW1B;

  554.     x86_reg imm8OfPShufW2B;

  555.     x86_reg fragmentLengthB;

  556.     int fragmentPos;

  557. 

  558.     int xpos, i;

  559. 

  560.     // create an optimized horizontal scaling routine

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

  562.      * tuned pshufw instructions. For every four output pixels, if four

  563.      * input pixels are enough for the fast bilinear scaling, then a chunk

  564.      * of fragmentB is used. If five input pixels are needed, then a chunk

  565.      * of fragmentA is used.

  566.      */

  567. 

  568.     //code fragment

  569. 

  570.     __asm__ volatile(

  571.         "jmp                         9f                 \n\t"

  572.     // Begin

  573.         "0:                                             \n\t"

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

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

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

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

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

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

  580.         "1:                                             \n\t"

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

  582.         "2:                                             \n\t"

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

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

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

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

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

  588. 

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

  590. 

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

  592.     // End

  593.         "9:                                             \n\t"

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

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

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

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

  598.         "dec                         %1                 \n\t"

  599.         "dec                         %2                 \n\t"

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

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

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

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

  604. 

  605. 

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

  607.         "=r" (fragmentLengthA)

  608.     );

  609. 

  610.     __asm__ volatile(

  611.         "jmp                         9f                 \n\t"

  612.     // Begin

  613.         "0:                                             \n\t"

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

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

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

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

  618.         "1:                                             \n\t"

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

  620.         "2:                                             \n\t"

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

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

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

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

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

  626. 

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

  628. 

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

  630.     // End

  631.         "9:                                             \n\t"

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

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

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

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

  636.         "dec                         %1                 \n\t"

  637.         "dec                         %2                 \n\t"

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

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

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

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

  642. 

  643. 

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

  645.         "=r" (fragmentLengthB)

  646.     );

  647. 

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

  649.     fragmentPos=0;

  650. 

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

  652.         int xx=xpos>>16;

  653. 

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

  655.             int a=0;

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

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

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

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

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

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

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

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

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

  665.             int shift=0;

  666. 

  667.             if (filterCode) {

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

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

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

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

  672.                 filterPos[i/2]= xx;

  673. 

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

  675. 

  676.                 filterCode[fragmentPos + imm8OfPShufW1]=

  677.                     (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);

  678.                 filterCode[fragmentPos + imm8OfPShufW2]=

  679.                     a | (b<<2) | (c<<4) | (d<<6);

  680. 

  681.                 if (i+4-inc>=dstW) shift=maxShift; //avoid overread

  682.                 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align

  683. 

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

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

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

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

  688.                 }

  689.             }

  690. 

  691.             fragmentPos+= fragmentLength;

  692. 

  693.             if (filterCode)

  694.                 filterCode[fragmentPos]= RET;

  695.         }

  696.         xpos+=xInc;

  697.     }

  698.     if (filterCode)

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

  700. 

  701.     return fragmentPos + 1;

  702. }

  703. #endif /* HAVE_MMX2 */

  704. 

  705. static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)

  706. {

  707.     *h = av_pix_fmt_descriptors[format].log2_chroma_w;

  708.     *v = av_pix_fmt_descriptors[format].log2_chroma_h;

  709. }

  710. 

  711. int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)

  712. {

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

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

  715. 

  716.     c->brightness= brightness;

  717.     c->contrast  = contrast;

  718.     c->saturation= saturation;

  719.     c->srcRange  = srcRange;

  720.     c->dstRange  = dstRange;

  721.     if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;

  722. 

  723.     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);

  724.     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);

  725. 

  726.     ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);

  727.     //FIXME factorize

  728. 

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

  730.         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);

  731.     return 0;

  732. }

  733. 

  734. int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)

  735. {

  736.     if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;

  737. 

  738.     *inv_table = c->srcColorspaceTable;

  739.     *table     = c->dstColorspaceTable;

  740.     *srcRange  = c->srcRange;

  741.     *dstRange  = c->dstRange;

  742.     *brightness= c->brightness;

  743.     *contrast  = c->contrast;

  744.     *saturation= c->saturation;

  745. 

  746.     return 0;

  747. }

  748. 

  749. static int handle_jpeg(enum PixelFormat *format)

  750. {

  751.     switch (*format) {

  752.     case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;

  753.     case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;

  754.     case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;

  755.     case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;

  756.     default:                                          return 0;

  757.     }

  758. }

  759. 

  760. SwsContext *sws_alloc_context(void)

  761. {

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

  763. 

  764.     c->av_class = &sws_context_class;

  765.     av_opt_set_defaults(c);

  766. 

  767.     return c;

  768. }

  769. 

  770. int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)

  771. {

  772.     int i;

  773.     int usesVFilter, usesHFilter;

  774.     int unscaled;

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

  776.     int srcW= c->srcW;

  777.     int srcH= c->srcH;

  778.     int dstW= c->dstW;

  779.     int dstH= c->dstH;

  780.     int dst_stride = FFALIGN(dstW * sizeof(int16_t)+66, 16), dst_stride_px = dst_stride >> 1;

  781.     int flags, cpu_flags;

  782.     enum PixelFormat srcFormat= c->srcFormat;

  783.     enum PixelFormat dstFormat= c->dstFormat;

  784. 

  785.     cpu_flags = av_get_cpu_flags();

  786.     flags     = c->flags;

  787.     emms_c();

  788.     if (!rgb15to16) sws_rgb2rgb_init();

  789. 

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

  791. 

  792.     if (!isSupportedIn(srcFormat)) {

  793.         av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n", av_get_pix_fmt_name(srcFormat));

  794.         return AVERROR(EINVAL);

  795.     }

  796.     if (!isSupportedOut(dstFormat)) {

  797.         av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n", av_get_pix_fmt_name(dstFormat));

  798.         return AVERROR(EINVAL);

  799.     }

  800. 

  801.     i= flags & ( SWS_POINT

  802.                 |SWS_AREA

  803.                 |SWS_BILINEAR

  804.                 |SWS_FAST_BILINEAR

  805.                 |SWS_BICUBIC

  806.                 |SWS_X

  807.                 |SWS_GAUSS

  808.                 |SWS_LANCZOS

  809.                 |SWS_SINC

  810.                 |SWS_SPLINE

  811.                 |SWS_BICUBLIN);

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

  813.         av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen\n");

  814.         return AVERROR(EINVAL);

  815.     }

  816.     /* sanity check */

  817.     if (srcW<4 || srcH<1 || dstW<8 || dstH<1) { //FIXME check if these are enough and try to lowwer them after fixing the relevant parts of the code

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

  819.                srcW, srcH, dstW, dstH);

  820.         return AVERROR(EINVAL);

  821.     }

  822. 

  823.     if (!dstFilter) dstFilter= &dummyFilter;

  824.     if (!srcFilter) srcFilter= &dummyFilter;

  825. 

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

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

  828.     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);

  829.     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);

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

  831. 

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

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

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

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

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

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

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

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

  840. 

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

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

  843. 

  844.     // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation

  845.     if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;

  846. 

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

  848.     c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;

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

  850. 

  851.     // drop every other pixel for chroma calculation unless user wants full chroma

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

  853.       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8

  854.       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4

  855.       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE

  856.       && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))

  857.         c->chrSrcHSubSample=1;

  858. 

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

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

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

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

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

  864. 

  865.     /* unscaled special cases */

  866.     if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {

  867.         ff_get_unscaled_swscale(c);

  868. 

  869.         if (c->swScale) {

  870.             if (flags&SWS_PRINT_INFO)

  871.                 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",

  872.                        av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));

  873.             return 0;

  874.         }

  875.     }

  876. 

  877.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);

  878.     if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) {

  879.         c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;

  880.         if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {

  881.             if (flags&SWS_PRINT_INFO)

  882.                 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");

  883.         }

  884.         if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat)) c->canMMX2BeUsed=0;

  885.     }

  886.     else

  887.         c->canMMX2BeUsed=0;

  888. 

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

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

  891. 

  892.     // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst

  893.     // but only for the FAST_BILINEAR mode otherwise do correct scaling

  894.     // n-2 is the last chrominance sample available

  895.     // this is not perfect, but no one should notice the difference, the more correct variant

  896.     // would be like the vertical one, but that would require some special code for the

  897.     // first and last pixel

  898.     if (flags&SWS_FAST_BILINEAR) {

  899.         if (c->canMMX2BeUsed) {

  900.             c->lumXInc+= 20;

  901.             c->chrXInc+= 20;

  902.         }

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

  904.         else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {

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

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

  907.         }

  908.     }

  909. 

  910.     /* precalculate horizontal scaler filter coefficients */

  911.     {

  912. #if HAVE_MMX2

  913. // can't downscale !!!

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

  915.             c->lumMmx2FilterCodeSize = initMMX2HScaler(      dstW, c->lumXInc, NULL, NULL, NULL, 8);

  916.             c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);

  917. 

  918. #ifdef MAP_ANONYMOUS

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

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

  921. #elif HAVE_VIRTUALALLOC

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

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

  924. #else

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

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

  927. #endif

  928. 

  929. #ifdef MAP_ANONYMOUS

  930.             if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)

  931. #else

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

  933. #endif

  934.                 return AVERROR(ENOMEM);

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

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

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

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

  939. 

  940.             initMMX2HScaler(      dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);

  941.             initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);

  942. 

  943. #ifdef MAP_ANONYMOUS

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

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

  946. #endif

  947.         } else

  948. #endif /* HAVE_MMX2 */

  949.         {

  950.             const int filterAlign=

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

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

  953.                 1;

  954. 

  955.             if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,

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

  957.                            (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags, cpu_flags,

  958.                            srcFilter->lumH, dstFilter->lumH, c->param) < 0)

  959.                 goto fail;

  960.             if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,

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

  962.                            (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,

  963.                            srcFilter->chrH, dstFilter->chrH, c->param) < 0)

  964.                 goto fail;

  965.         }

  966.     } // initialize horizontal stuff

  967. 

  968.     /* precalculate vertical scaler filter coefficients */

  969.     {

  970.         const int filterAlign=

  971.             (HAVE_MMX     && cpu_flags & AV_CPU_FLAG_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :

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

  973.             1;

  974. 

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

  976.                        srcH      ,        dstH, filterAlign, (1<<12),

  977.                        (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags, cpu_flags,

  978.                        srcFilter->lumV, dstFilter->lumV, c->param) < 0)

  979.             goto fail;

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

  981.                        c->chrSrcH, c->chrDstH, filterAlign, (1<<12),

  982.                        (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,

  983.                        srcFilter->chrV, dstFilter->chrV, c->param) < 0)

  984.             goto fail;

  985. 

  986. #if HAVE_ALTIVEC

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

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

  989. 

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

  991.             int j;

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

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

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

  995.         }

  996. 

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

  998.             int j;

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

  1000.             for (j=0;j<8;j++)

  1001.                 p[j] = c->vChrFilter[i];

  1002.         }

  1003. #endif

  1004.     }

  1005. 

  1006.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1007.     c->vLumBufSize= c->vLumFilterSize;

  1008.     c->vChrBufSize= c->vChrFilterSize;

  1009.     for (i=0; i<dstH; i++) {

  1010.         int chrI= (int64_t)i*c->chrDstH / dstH;

  1011.         int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,

  1012.                            ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));

  1013. 

  1014.         nextSlice>>= c->chrSrcVSubSample;

  1015.         nextSlice<<= c->chrSrcVSubSample;

  1016.         if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)

  1017.             c->vLumBufSize= nextSlice - c->vLumFilterPos[i];

  1018.         if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))

  1019.             c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];

  1020.     }

  1021. 

  1022.     // allocate pixbufs (we use dynamic allocation because otherwise we would need to

  1023.     // allocate several megabytes to handle all possible cases)

  1024.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1025.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1026.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1027.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1028.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1029.     //Note we need at least one pixel more at the end because of the MMX code (just in case someone wanna replace the 4000/8000)

  1030.     /* align at 16 bytes for AltiVec */

  1031.     for (i=0; i<c->vLumBufSize; i++) {

  1032.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+1, fail);

  1033.         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];

  1034.     }

  1035.     c->uv_off = dst_stride_px;

  1036.     c->uv_offx2 = dst_stride;

  1037.     for (i=0; i<c->vChrBufSize; i++) {

  1038.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+1, fail);

  1039.         c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];

  1040.         c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + dst_stride_px;

  1041.     }

  1042.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1043.         for (i=0; i<c->vLumBufSize; i++) {

  1044.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+1, fail);

  1045.             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];

  1046.         }

  1047. 

  1048.     //try to avoid drawing green stuff between the right end and the stride end

  1049.     for (i=0; i<c->vChrBufSize; i++)

  1050.         memset(c->chrUPixBuf[i], 64, dst_stride*2+1);

  1051. 

  1052.     assert(c->chrDstH <= dstH);

  1053. 

  1054.     if (flags&SWS_PRINT_INFO) {

  1055.         if      (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1056.         else if (flags&SWS_BILINEAR)      av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1057.         else if (flags&SWS_BICUBIC)       av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1058.         else if (flags&SWS_X)             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1059.         else if (flags&SWS_POINT)         av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1060.         else if (flags&SWS_AREA)          av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1061.         else if (flags&SWS_BICUBLIN)      av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1062.         else if (flags&SWS_GAUSS)         av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1063.         else if (flags&SWS_SINC)          av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1064.         else if (flags&SWS_LANCZOS)       av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1065.         else if (flags&SWS_SPLINE)        av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1066.         else                              av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1067. 

  1068.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1069.                av_get_pix_fmt_name(srcFormat),

  1070. #ifdef DITHER1XBPP

  1071.                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||

  1072.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1073.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",

  1074. #else

  1075.                "",

  1076. #endif

  1077.                av_get_pix_fmt_name(dstFormat));

  1078. 

  1079.         if      (HAVE_MMX2     && cpu_flags & AV_CPU_FLAG_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");

  1080.         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1081.         else if (HAVE_MMX      && cpu_flags & AV_CPU_FLAG_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");

  1082.         else if (HAVE_ALTIVEC  && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1083.         else                                   av_log(c, AV_LOG_INFO, "using C\n");

  1084. 

  1085.         if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {

  1086.             if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))

  1087.                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");

  1088.             else {

  1089.                 if (c->hLumFilterSize==4)

  1090.                     av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");

  1091.                 else if (c->hLumFilterSize==8)

  1092.                     av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");

  1093.                 else

  1094.                     av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");

  1095. 

  1096.                 if (c->hChrFilterSize==4)

  1097.                     av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");

  1098.                 else if (c->hChrFilterSize==8)

  1099.                     av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");

  1100.                 else

  1101.                     av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");

  1102.             }

  1103.         } else {

  1104. #if HAVE_MMX

  1105.             av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");

  1106. #else

  1107.             if (flags & SWS_FAST_BILINEAR)

  1108.                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");

  1109.             else

  1110.                 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");

  1111. #endif

  1112.         }

  1113.         if (isPlanarYUV(dstFormat)) {

  1114.             if (c->vLumFilterSize==1)

  1115.                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n",

  1116.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1117.             else

  1118.                 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n",

  1119.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1120.         } else {

  1121.             if (c->vLumFilterSize==1 && c->vChrFilterSize==2)

  1122.                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"

  1123.                        "      2-tap scaler for vertical chrominance scaling (BGR)\n",

  1124.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1125.             else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)

  1126.                 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n",

  1127.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1128.             else

  1129.                 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n",

  1130.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1131.         }

  1132. 

  1133.         if (dstFormat==PIX_FMT_BGR24)

  1134.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",

  1135.                    (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) ? "MMX2" :

  1136.                    ((HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"));

  1137.         else if (dstFormat==PIX_FMT_RGB32)

  1138.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n",

  1139.                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1140.         else if (dstFormat==PIX_FMT_BGR565)

  1141.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n",

  1142.                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1143.         else if (dstFormat==PIX_FMT_BGR555)

  1144.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n",

  1145.                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1146.         else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1147.                  dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)

  1148.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n",

  1149.                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1150. 

  1151.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1152.         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1153.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1154.         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1155.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1156.     }

  1157. 

  1158.     c->swScale= ff_getSwsFunc(c);

  1159.     return 0;

  1160. fail: //FIXME replace things by appropriate error codes

  1161.     return -1;

  1162. }

  1163. 

  1164. #if FF_API_SWS_GETCONTEXT

  1165. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1166.                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1167.                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1168. {

  1169.     SwsContext *c;

  1170. 

  1171.     if(!(c=sws_alloc_context()))

  1172.         return NULL;

  1173. 

  1174.     c->flags= flags;

  1175.     c->srcW= srcW;

  1176.     c->srcH= srcH;

  1177.     c->dstW= dstW;

  1178.     c->dstH= dstH;

  1179.     c->srcRange = handle_jpeg(&srcFormat);

  1180.     c->dstRange = handle_jpeg(&dstFormat);

  1181.     c->srcFormat= srcFormat;

  1182.     c->dstFormat= dstFormat;

  1183. 

  1184.     if (param) {

  1185.         c->param[0] = param[0];

  1186.         c->param[1] = param[1];

  1187.     }

  1188.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, c->dstRange, 0, 1<<16, 1<<16);

  1189. 

  1190.     if(sws_init_context(c, srcFilter, dstFilter) < 0){

  1191.         sws_freeContext(c);

  1192.         return NULL;

  1193.     }

  1194. 

  1195.     return c;

  1196. }

  1197. #endif

  1198. 

  1199. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1200.                                 float lumaSharpen, float chromaSharpen,

  1201.                                 float chromaHShift, float chromaVShift,

  1202.                                 int verbose)

  1203. {

  1204.     SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  1205.     if (!filter)

  1206.         return NULL;

  1207. 

  1208.     if (lumaGBlur!=0.0) {

  1209.         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  1210.         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  1211.     } else {

  1212.         filter->lumH= sws_getIdentityVec();

  1213.         filter->lumV= sws_getIdentityVec();

  1214.     }

  1215. 

  1216.     if (chromaGBlur!=0.0) {

  1217.         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  1218.         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  1219.     } else {

  1220.         filter->chrH= sws_getIdentityVec();

  1221.         filter->chrV= sws_getIdentityVec();

  1222.     }

  1223. 

  1224.     if (chromaSharpen!=0.0) {

  1225.         SwsVector *id= sws_getIdentityVec();

  1226.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1227.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1228.         sws_addVec(filter->chrH, id);

  1229.         sws_addVec(filter->chrV, id);

  1230.         sws_freeVec(id);

  1231.     }

  1232. 

  1233.     if (lumaSharpen!=0.0) {

  1234.         SwsVector *id= sws_getIdentityVec();

  1235.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1236.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1237.         sws_addVec(filter->lumH, id);

  1238.         sws_addVec(filter->lumV, id);

  1239.         sws_freeVec(id);

  1240.     }

  1241. 

  1242.     if (chromaHShift != 0.0)

  1243.         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  1244. 

  1245.     if (chromaVShift != 0.0)

  1246.         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  1247. 

  1248.     sws_normalizeVec(filter->chrH, 1.0);

  1249.     sws_normalizeVec(filter->chrV, 1.0);

  1250.     sws_normalizeVec(filter->lumH, 1.0);

  1251.     sws_normalizeVec(filter->lumV, 1.0);

  1252. 

  1253.     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1254.     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1255. 

  1256.     return filter;

  1257. }

  1258. 

  1259. SwsVector *sws_allocVec(int length)

  1260. {

  1261.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1262.     if (!vec)

  1263.         return NULL;

  1264.     vec->length = length;

  1265.     vec->coeff  = av_malloc(sizeof(double) * length);

  1266.     if (!vec->coeff)

  1267.         av_freep(&vec);

  1268.     return vec;

  1269. }

  1270. 

  1271. SwsVector *sws_getGaussianVec(double variance, double quality)

  1272. {

  1273.     const int length= (int)(variance*quality + 0.5) | 1;

  1274.     int i;

  1275.     double middle= (length-1)*0.5;

  1276.     SwsVector *vec= sws_allocVec(length);

  1277. 

  1278.     if (!vec)

  1279.         return NULL;

  1280. 

  1281.     for (i=0; i<length; i++) {

  1282.         double dist= i-middle;

  1283.         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);

  1284.     }

  1285. 

  1286.     sws_normalizeVec(vec, 1.0);

  1287. 

  1288.     return vec;

  1289. }

  1290. 

  1291. SwsVector *sws_getConstVec(double c, int length)

  1292. {

  1293.     int i;

  1294.     SwsVector *vec= sws_allocVec(length);

  1295. 

  1296.     if (!vec)

  1297.         return NULL;

  1298. 

  1299.     for (i=0; i<length; i++)

  1300.         vec->coeff[i]= c;

  1301. 

  1302.     return vec;

  1303. }

  1304. 

  1305. SwsVector *sws_getIdentityVec(void)

  1306. {

  1307.     return sws_getConstVec(1.0, 1);

  1308. }

  1309. 

  1310. static double sws_dcVec(SwsVector *a)

  1311. {

  1312.     int i;

  1313.     double sum=0;

  1314. 

  1315.     for (i=0; i<a->length; i++)

  1316.         sum+= a->coeff[i];

  1317. 

  1318.     return sum;

  1319. }

  1320. 

  1321. void sws_scaleVec(SwsVector *a, double scalar)

  1322. {

  1323.     int i;

  1324. 

  1325.     for (i=0; i<a->length; i++)

  1326.         a->coeff[i]*= scalar;

  1327. }

  1328. 

  1329. void sws_normalizeVec(SwsVector *a, double height)

  1330. {

  1331.     sws_scaleVec(a, height/sws_dcVec(a));

  1332. }

  1333. 

  1334. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1335. {

  1336.     int length= a->length + b->length - 1;

  1337.     int i, j;

  1338.     SwsVector *vec= sws_getConstVec(0.0, length);

  1339. 

  1340.     if (!vec)

  1341.         return NULL;

  1342. 

  1343.     for (i=0; i<a->length; i++) {

  1344.         for (j=0; j<b->length; j++) {

  1345.             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1346.         }

  1347.     }

  1348. 

  1349.     return vec;

  1350. }

  1351. 

  1352. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1353. {

  1354.     int length= FFMAX(a->length, b->length);

  1355.     int i;

  1356.     SwsVector *vec= sws_getConstVec(0.0, length);

  1357. 

  1358.     if (!vec)

  1359.         return NULL;

  1360. 

  1361.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1362.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];

  1363. 

  1364.     return vec;

  1365. }

  1366. 

  1367. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1368. {

  1369.     int length= FFMAX(a->length, b->length);

  1370.     int i;

  1371.     SwsVector *vec= sws_getConstVec(0.0, length);

  1372. 

  1373.     if (!vec)

  1374.         return NULL;

  1375. 

  1376.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1377.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1378. 

  1379.     return vec;

  1380. }

  1381. 

  1382. /* shift left / or right if "shift" is negative */

  1383. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1384. {

  1385.     int length= a->length + FFABS(shift)*2;

  1386.     int i;

  1387.     SwsVector *vec= sws_getConstVec(0.0, length);

  1388. 

  1389.     if (!vec)

  1390.         return NULL;

  1391. 

  1392.     for (i=0; i<a->length; i++) {

  1393.         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1394.     }

  1395. 

  1396.     return vec;

  1397. }

  1398. 

  1399. void sws_shiftVec(SwsVector *a, int shift)

  1400. {

  1401.     SwsVector *shifted= sws_getShiftedVec(a, shift);

  1402.     av_free(a->coeff);

  1403.     a->coeff= shifted->coeff;

  1404.     a->length= shifted->length;

  1405.     av_free(shifted);

  1406. }

  1407. 

  1408. void sws_addVec(SwsVector *a, SwsVector *b)

  1409. {

  1410.     SwsVector *sum= sws_sumVec(a, b);

  1411.     av_free(a->coeff);

  1412.     a->coeff= sum->coeff;

  1413.     a->length= sum->length;

  1414.     av_free(sum);

  1415. }

  1416. 

  1417. void sws_subVec(SwsVector *a, SwsVector *b)

  1418. {

  1419.     SwsVector *diff= sws_diffVec(a, b);

  1420.     av_free(a->coeff);

  1421.     a->coeff= diff->coeff;

  1422.     a->length= diff->length;

  1423.     av_free(diff);

  1424. }

  1425. 

  1426. void sws_convVec(SwsVector *a, SwsVector *b)

  1427. {

  1428.     SwsVector *conv= sws_getConvVec(a, b);

  1429.     av_free(a->coeff);

  1430.     a->coeff= conv->coeff;

  1431.     a->length= conv->length;

  1432.     av_free(conv);

  1433. }

  1434. 

  1435. SwsVector *sws_cloneVec(SwsVector *a)

  1436. {

  1437.     int i;

  1438.     SwsVector *vec= sws_allocVec(a->length);

  1439. 

  1440.     if (!vec)

  1441.         return NULL;

  1442. 

  1443.     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];

  1444. 

  1445.     return vec;

  1446. }

  1447. 

  1448. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1449. {

  1450.     int i;

  1451.     double max=0;

  1452.     double min=0;

  1453.     double range;

  1454. 

  1455.     for (i=0; i<a->length; i++)

  1456.         if (a->coeff[i]>max) max= a->coeff[i];

  1457. 

  1458.     for (i=0; i<a->length; i++)

  1459.         if (a->coeff[i]<min) min= a->coeff[i];

  1460. 

  1461.     range= max - min;

  1462. 

  1463.     for (i=0; i<a->length; i++) {

  1464.         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1465.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1466.         for (;x>0; x--) av_log(log_ctx, log_level, " ");

  1467.         av_log(log_ctx, log_level, "|\n");

  1468.     }

  1469. }

  1470. 

  1471. #if LIBSWSCALE_VERSION_MAJOR < 1

  1472. void sws_printVec(SwsVector *a)

  1473. {

  1474.     sws_printVec2(a, NULL, AV_LOG_DEBUG);

  1475. }

  1476. #endif

  1477. 

  1478. void sws_freeVec(SwsVector *a)

  1479. {

  1480.     if (!a) return;

  1481.     av_freep(&a->coeff);

  1482.     a->length=0;

  1483.     av_free(a);

  1484. }

  1485. 

  1486. void sws_freeFilter(SwsFilter *filter)

  1487. {

  1488.     if (!filter) return;

  1489. 

  1490.     if (filter->lumH) sws_freeVec(filter->lumH);

  1491.     if (filter->lumV) sws_freeVec(filter->lumV);

  1492.     if (filter->chrH) sws_freeVec(filter->chrH);

  1493.     if (filter->chrV) sws_freeVec(filter->chrV);

  1494.     av_free(filter);

  1495. }

  1496. 

  1497. void sws_freeContext(SwsContext *c)

  1498. {

  1499.     int i;

  1500.     if (!c) return;

  1501. 

  1502.     if (c->lumPixBuf) {

  1503.         for (i=0; i<c->vLumBufSize; i++)

  1504.             av_freep(&c->lumPixBuf[i]);

  1505.         av_freep(&c->lumPixBuf);

  1506.     }

  1507. 

  1508.     if (c->chrUPixBuf) {

  1509.         for (i=0; i<c->vChrBufSize; i++)

  1510.             av_freep(&c->chrUPixBuf[i]);

  1511.         av_freep(&c->chrUPixBuf);

  1512.         av_freep(&c->chrVPixBuf);

  1513.     }

  1514. 

  1515.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1516.         for (i=0; i<c->vLumBufSize; i++)

  1517.             av_freep(&c->alpPixBuf[i]);

  1518.         av_freep(&c->alpPixBuf);

  1519.     }

  1520. 

  1521.     av_freep(&c->vLumFilter);

  1522.     av_freep(&c->vChrFilter);

  1523.     av_freep(&c->hLumFilter);

  1524.     av_freep(&c->hChrFilter);

  1525. #if HAVE_ALTIVEC

  1526.     av_freep(&c->vYCoeffsBank);

  1527.     av_freep(&c->vCCoeffsBank);

  1528. #endif

  1529. 

  1530.     av_freep(&c->vLumFilterPos);

  1531.     av_freep(&c->vChrFilterPos);

  1532.     av_freep(&c->hLumFilterPos);

  1533.     av_freep(&c->hChrFilterPos);

  1534. 

  1535. #if HAVE_MMX

  1536. #ifdef MAP_ANONYMOUS

  1537.     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1538.     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1539. #elif HAVE_VIRTUALALLOC

  1540.     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1541.     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1542. #else

  1543.     av_free(c->lumMmx2FilterCode);

  1544.     av_free(c->chrMmx2FilterCode);

  1545. #endif

  1546.     c->lumMmx2FilterCode=NULL;

  1547.     c->chrMmx2FilterCode=NULL;

  1548. #endif /* HAVE_MMX */

  1549. 

  1550.     av_freep(&c->yuvTable);

  1551.     av_freep(&c->formatConvBuffer);

  1552. 

  1553.     av_free(c);

  1554. }

  1555. 

  1556. struct SwsContext *sws_getCachedContext(struct SwsContext *context,

  1557.                                         int srcW, int srcH, enum PixelFormat srcFormat,

  1558.                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1559.                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1560. {

  1561.     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};

  1562. 

  1563.     if (!param)

  1564.         param = default_param;

  1565. 

  1566.     if (context &&

  1567.         (context->srcW      != srcW      ||

  1568.          context->srcH      != srcH      ||

  1569.          context->srcFormat != srcFormat ||

  1570.          context->dstW      != dstW      ||

  1571.          context->dstH      != dstH      ||

  1572.          context->dstFormat != dstFormat ||

  1573.          context->flags     != flags     ||

  1574.          context->param[0]  != param[0]  ||

  1575.          context->param[1]  != param[1])) {

  1576.         sws_freeContext(context);

  1577.         context = NULL;

  1578.     }

  1579. 

  1580.     if (!context) {

  1581.         if (!(context = sws_alloc_context()))

  1582.             return NULL;

  1583.         context->srcW      = srcW;

  1584.         context->srcH      = srcH;

  1585.         context->srcRange  = handle_jpeg(&srcFormat);

  1586.         context->srcFormat = srcFormat;

  1587.         context->dstW      = dstW;

  1588.         context->dstH      = dstH;

  1589.         context->dstRange  = handle_jpeg(&dstFormat);

  1590.         context->dstFormat = dstFormat;

  1591.         context->flags     = flags;

  1592.         context->param[0]  = param[0];

  1593.         context->param[1]  = param[1];

  1594.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], context->srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, context->dstRange, 0, 1<<16, 1<<16);

  1595.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1596.             sws_freeContext(context);

  1597.             return NULL;

  1598.         }

  1599.     }

  1600.     return context;

  1601. }