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.
33469 Commits
32 Branches
283MB
Tree: 4b4a6ebd68
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '4b4a6ebd68'
${ noResults }
FFmpeg/libswscale/utils.c

1524 lines
54KB
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. #include "libavutil/avassert.h"

  50. 

  51. unsigned swscale_version(void)

  52. {

  53.     return LIBSWSCALE_VERSION_INT;

  54. }

  55. 

  56. const char *swscale_configuration(void)

  57. {

  58.     return FFMPEG_CONFIGURATION;

  59. }

  60. 

  61. const char *swscale_license(void)

  62. {

  63. #define LICENSE_PREFIX "libswscale license: "

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

  65. }

  66. 

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

  68. 

  69. typedef struct FormatEntry {

  70.     int is_supported_in, is_supported_out;

  71. } FormatEntry;

  72. 

  73. const static FormatEntry format_entries[PIX_FMT_NB] = {

  74.     [PIX_FMT_YUV420P]     = { 1 , 1 },

  75.     [PIX_FMT_YUYV422]     = { 1 , 1 },

  76.     [PIX_FMT_RGB24]       = { 1 , 1 },

  77.     [PIX_FMT_BGR24]       = { 1 , 1 },

  78.     [PIX_FMT_YUV422P]     = { 1 , 1 },

  79.     [PIX_FMT_YUV444P]     = { 1 , 1 },

  80.     [PIX_FMT_YUV410P]     = { 1 , 1 },

  81.     [PIX_FMT_YUV411P]     = { 1 , 1 },

  82.     [PIX_FMT_GRAY8]       = { 1 , 1 },

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

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

  85.     [PIX_FMT_PAL8]        = { 1 , 0 },

  86.     [PIX_FMT_YUVJ420P]    = { 1 , 1 },

  87.     [PIX_FMT_YUVJ422P]    = { 1 , 1 },

  88.     [PIX_FMT_YUVJ444P]    = { 1 , 1 },

  89.     [PIX_FMT_UYVY422]     = { 1 , 1 },

  90.     [PIX_FMT_UYYVYY411]   = { 0 , 0 },

  91.     [PIX_FMT_BGR8]        = { 1 , 1 },

  92.     [PIX_FMT_BGR4]        = { 0 , 1 },

  93.     [PIX_FMT_BGR4_BYTE]   = { 1 , 1 },

  94.     [PIX_FMT_RGB8]        = { 1 , 1 },

  95.     [PIX_FMT_RGB4]        = { 0 , 1 },

  96.     [PIX_FMT_RGB4_BYTE]   = { 1 , 1 },

  97.     [PIX_FMT_NV12]        = { 1 , 1 },

  98.     [PIX_FMT_NV21]        = { 1 , 1 },

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

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

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

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

  103.     [PIX_FMT_GRAY16BE]    = { 1 , 1 },

  104.     [PIX_FMT_GRAY16LE]    = { 1 , 1 },

  105.     [PIX_FMT_YUV440P]     = { 1 , 1 },

  106.     [PIX_FMT_YUVJ440P]    = { 1 , 1 },

  107.     [PIX_FMT_YUVA420P]    = { 1 , 1 },

  108.     [PIX_FMT_RGB48BE]     = { 1 , 1 },

  109.     [PIX_FMT_RGB48LE]     = { 1 , 1 },

  110.     [PIX_FMT_RGBA64BE]    = { 0 , 0 },

  111.     [PIX_FMT_RGBA64LE]    = { 0 , 0 },

  112.     [PIX_FMT_RGB565BE]    = { 1 , 1 },

  113.     [PIX_FMT_RGB565LE]    = { 1 , 1 },

  114.     [PIX_FMT_RGB555BE]    = { 1 , 1 },

  115.     [PIX_FMT_RGB555LE]    = { 1 , 1 },

  116.     [PIX_FMT_BGR565BE]    = { 1 , 1 },

  117.     [PIX_FMT_BGR565LE]    = { 1 , 1 },

  118.     [PIX_FMT_BGR555BE]    = { 1 , 1 },

  119.     [PIX_FMT_BGR555LE]    = { 1 , 1 },

  120.     [PIX_FMT_YUV420P16LE] = { 1 , 1 },

  121.     [PIX_FMT_YUV420P16BE] = { 1 , 1 },

  122.     [PIX_FMT_YUV422P16LE] = { 1 , 1 },

  123.     [PIX_FMT_YUV422P16BE] = { 1 , 1 },

  124.     [PIX_FMT_YUV444P16LE] = { 1 , 1 },

  125.     [PIX_FMT_YUV444P16BE] = { 1 , 1 },

  126.     [PIX_FMT_RGB444LE]    = { 0 , 1 },

  127.     [PIX_FMT_RGB444BE]    = { 0 , 1 },

  128.     [PIX_FMT_BGR444LE]    = { 0 , 1 },

  129.     [PIX_FMT_BGR444BE]    = { 0 , 1 },

  130.     [PIX_FMT_Y400A]       = { 1 , 0 },

  131.     [PIX_FMT_BGR48BE]     = { 1 , 1 },

  132.     [PIX_FMT_BGR48LE]     = { 1 , 1 },

  133.     [PIX_FMT_BGRA64BE]    = { 0 , 0 },

  134.     [PIX_FMT_BGRA64LE]    = { 0 , 0 },

  135.     [PIX_FMT_YUV420P9BE]  = { 1 , 1 },

  136.     [PIX_FMT_YUV420P9LE]  = { 1 , 1 },

  137.     [PIX_FMT_YUV420P10BE] = { 1 , 1 },

  138.     [PIX_FMT_YUV420P10LE] = { 1 , 1 },

  139.     [PIX_FMT_YUV422P10BE] = { 1 , 1 },

  140.     [PIX_FMT_YUV422P10LE] = { 1 , 1 },

  141.     [PIX_FMT_YUV444P9BE]  = { 1 , 1 },

  142.     [PIX_FMT_YUV444P9LE]  = { 1 , 1 },

  143.     [PIX_FMT_YUV444P10BE] = { 1 , 1 },

  144.     [PIX_FMT_YUV444P10LE] = { 1 , 1 },

  145.     [PIX_FMT_GBR24P]      = { 1 , 0 },

  146. };

  147. 

  148. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  149. {

  150.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  151.         format_entries[pix_fmt].is_supported_in : 0;

  152. }

  153. 

  154. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  155. {

  156.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  157.         format_entries[pix_fmt].is_supported_out : 0;

  158. }

  159. 

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

  161. 

  162. #if FF_API_SWS_FORMAT_NAME

  163. const char *sws_format_name(enum PixelFormat format)

  164. {

  165.     return av_get_pix_fmt_name(format);

  166. }

  167. #endif

  168. 

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

  170. {

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

  172.     else           return getSplineCoeff(        0.0,

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

  174.                                                  c + 3.0*d,

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

  176.                                          dist-1.0);

  177. }

  178. 

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

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

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

  182. {

  183.     int i;

  184.     int filterSize;

  185.     int filter2Size;

  186.     int minFilterSize;

  187.     int64_t *filter=NULL;

  188.     int64_t *filter2=NULL;

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

  190.     int ret= -1;

  191. 

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

  193. 

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

  195.     FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+3)*sizeof(int16_t), fail);

  196. 

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

  198.         int i;

  199.         filterSize= 1;

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

  201. 

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

  203.             filter[i*filterSize]= fone;

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

  205.         }

  206. 

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

  208.         int i;

  209.         int xDstInSrc;

  210.         filterSize= 1;

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

  212. 

  213.         xDstInSrc= xInc/2 - 0x8000;

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

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

  216. 

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

  218.             filter[i]= fone;

  219.             xDstInSrc+= xInc;

  220.         }

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

  222.         int i;

  223.         int xDstInSrc;

  224.         filterSize= 2;

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

  226. 

  227.         xDstInSrc= xInc/2 - 0x8000;

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

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

  230.             int j;

  231. 

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

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

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

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

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

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

  238.                 xx++;

  239.             }

  240.             xDstInSrc+= xInc;

  241.         }

  242.     } else {

  243.         int xDstInSrc;

  244.         int sizeFactor;

  245. 

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

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

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

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

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

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

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

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

  254.         else {

  255.             sizeFactor= 0; //GCC warning killer

  256.             assert(0);

  257.         }

  258. 

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

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

  261. 

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

  263. 

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

  265. 

  266.         xDstInSrc= xInc - 0x10000;

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

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

  269.             int j;

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

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

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

  273.                 double floatd;

  274.                 int64_t coeff;

  275. 

  276.                 if (xInc > 1<<16)

  277.                     d= d*dstW/srcW;

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

  279. 

  280.                 if (flags & SWS_BICUBIC) {

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

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

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

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

  285. 

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

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

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

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

  290.                     else

  291.                         coeff=0.0;

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

  293.                 }

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

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

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

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

  298.                 }*/

  299.                 else if (flags & SWS_X) {

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

  301.                     double c;

  302. 

  303.                     if (floatd<1.0)

  304.                         c = cos(floatd*M_PI);

  305.                     else

  306.                         c=-1.0;

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

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

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

  310.                 } else if (flags & SWS_AREA) {

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

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

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

  314.                     else coeff=0.0;

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

  316.                 } else if (flags & SWS_GAUSS) {

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

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

  319.                 } else if (flags & SWS_SINC) {

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

  321.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  325.                 } else if (flags & SWS_BILINEAR) {

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

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

  328.                     coeff *= fone >> 30;

  329.                 } else if (flags & SWS_SPLINE) {

  330.                     double p=-2.196152422706632;

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

  332.                 } else {

  333.                     coeff= 0.0; //GCC warning killer

  334.                     assert(0);

  335.                 }

  336. 

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

  338.                 xx++;

  339.             }

  340.             xDstInSrc+= 2*xInc;

  341.         }

  342.     }

  343. 

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

  345.        av_free(filter);

  346.     */

  347.     assert(filterSize>0);

  348.     filter2Size= filterSize;

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

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

  351.     assert(filter2Size>0);

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

  353. 

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

  355.         int j, k;

  356. 

  357.         if(srcFilter) {

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

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

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

  361.             }

  362.         } else {

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

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

  365.         }

  366.         //FIXME dstFilter

  367. 

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

  369.     }

  370.     av_freep(&filter);

  371. 

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

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

  374.     minFilterSize= 0;

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

  376.         int min= filter2Size;

  377.         int j;

  378.         int64_t cutOff=0.0;

  379. 

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

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

  382.             int k;

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

  384. 

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

  386. 

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

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

  389. 

  390.             // move filter coefficients left

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

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

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

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

  395.         }

  396. 

  397.         cutOff=0;

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

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

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

  401. 

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

  403.             min--;

  404.         }

  405. 

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

  407.     }

  408. 

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

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

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

  412.         if (minFilterSize < 5)

  413.             filterAlign = 4;

  414. 

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

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

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

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

  419.         // decent-sized vector.

  420.         if (minFilterSize < 3)

  421.             filterAlign = 1;

  422.     }

  423. 

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

  425.         // special case for unscaled vertical filtering

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

  427.             filterAlign= 1;

  428.     }

  429. 

  430.     assert(minFilterSize > 0);

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

  432.     assert(filterSize > 0);

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

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

  435.         goto fail;

  436.     *outFilterSize= filterSize;

  437. 

  438.     if (flags&SWS_PRINT_INFO)

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

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

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

  442.         int j;

  443. 

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

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

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

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

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

  449.         }

  450.     }

  451. 

  452.     //FIXME try to align filterPos if possible

  453. 

  454.     //fix borders

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

  456.         int j;

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

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

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

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

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

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

  463.             }

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

  465.         }

  466. 

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

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

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

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

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

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

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

  474.             }

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

  476.         }

  477.     }

  478. 

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

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

  481.     FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+3)*sizeof(int16_t), fail);

  482. 

  483.     /* normalize & store in outFilter */

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

  485.         int j;

  486.         int64_t error=0;

  487.         int64_t sum=0;

  488. 

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

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

  491.         }

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

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

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

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

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

  497.             error= v - intV*sum;

  498.         }

  499.     }

  500. 

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

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

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

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

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

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

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

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

  509.     }

  510. 

  511.     ret=0;

  512. fail:

  513.     av_free(filter);

  514.     av_free(filter2);

  515.     return ret;

  516. }

  517. 

  518. #if HAVE_MMX2

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

  520. {

  521.     uint8_t *fragmentA;

  522.     x86_reg imm8OfPShufW1A;

  523.     x86_reg imm8OfPShufW2A;

  524.     x86_reg fragmentLengthA;

  525.     uint8_t *fragmentB;

  526.     x86_reg imm8OfPShufW1B;

  527.     x86_reg imm8OfPShufW2B;

  528.     x86_reg fragmentLengthB;

  529.     int fragmentPos;

  530. 

  531.     int xpos, i;

  532. 

  533.     // create an optimized horizontal scaling routine

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

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

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

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

  538.      * of fragmentA is used.

  539.      */

  540. 

  541.     //code fragment

  542. 

  543.     __asm__ volatile(

  544.         "jmp                         9f                 \n\t"

  545.     // Begin

  546.         "0:                                             \n\t"

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

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

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

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

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

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

  553.         "1:                                             \n\t"

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

  555.         "2:                                             \n\t"

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

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

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

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

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

  561. 

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

  563. 

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

  565.     // End

  566.         "9:                                             \n\t"

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

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

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

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

  571.         "dec                         %1                 \n\t"

  572.         "dec                         %2                 \n\t"

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

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

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

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

  577. 

  578. 

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

  580.         "=r" (fragmentLengthA)

  581.     );

  582. 

  583.     __asm__ volatile(

  584.         "jmp                         9f                 \n\t"

  585.     // Begin

  586.         "0:                                             \n\t"

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

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

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

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

  591.         "1:                                             \n\t"

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

  593.         "2:                                             \n\t"

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

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

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

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

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

  599. 

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

  601. 

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

  603.     // End

  604.         "9:                                             \n\t"

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

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

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

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

  609.         "dec                         %1                 \n\t"

  610.         "dec                         %2                 \n\t"

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

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

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

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

  615. 

  616. 

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

  618.         "=r" (fragmentLengthB)

  619.     );

  620. 

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

  622.     fragmentPos=0;

  623. 

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

  625.         int xx=xpos>>16;

  626. 

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

  628.             int a=0;

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

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

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

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

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

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

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

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

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

  638.             int shift=0;

  639. 

  640.             if (filterCode) {

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

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

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

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

  645.                 filterPos[i/2]= xx;

  646. 

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

  648. 

  649.                 filterCode[fragmentPos + imm8OfPShufW1]=

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

  651.                 filterCode[fragmentPos + imm8OfPShufW2]=

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

  653. 

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

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

  656. 

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

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

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

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

  661.                 }

  662.             }

  663. 

  664.             fragmentPos+= fragmentLength;

  665. 

  666.             if (filterCode)

  667.                 filterCode[fragmentPos]= RET;

  668.         }

  669.         xpos+=xInc;

  670.     }

  671.     if (filterCode)

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

  673. 

  674.     return fragmentPos + 1;

  675. }

  676. #endif /* HAVE_MMX2 */

  677. 

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

  679. {

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

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

  682. }

  683. 

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

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

  686.                              int brightness, int contrast, int saturation)

  687. {

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

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

  690. 

  691.     c->brightness= brightness;

  692.     c->contrast  = contrast;

  693.     c->saturation= saturation;

  694.     c->srcRange  = srcRange;

  695.     c->dstRange  = dstRange;

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

  697. 

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

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

  700. 

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

  702.     //FIXME factorize

  703. 

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

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

  706.     return 0;

  707. }

  708. 

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

  710.                              int *srcRange, int **table, int *dstRange,

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

  712. {

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

  714. 

  715.     *inv_table = c->srcColorspaceTable;

  716.     *table     = c->dstColorspaceTable;

  717.     *srcRange  = c->srcRange;

  718.     *dstRange  = c->dstRange;

  719.     *brightness= c->brightness;

  720.     *contrast  = c->contrast;

  721.     *saturation= c->saturation;

  722. 

  723.     return 0;

  724. }

  725. 

  726. static int handle_jpeg(enum PixelFormat *format)

  727. {

  728.     switch (*format) {

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

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

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

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

  733.     default:                                          return 0;

  734.     }

  735. }

  736. 

  737. SwsContext *sws_alloc_context(void)

  738. {

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

  740. 

  741.     c->av_class = &sws_context_class;

  742.     av_opt_set_defaults(c);

  743. 

  744.     return c;

  745. }

  746. 

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

  748. {

  749.     int i, j;

  750.     int usesVFilter, usesHFilter;

  751.     int unscaled;

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

  753.     int srcW= c->srcW;

  754.     int srcH= c->srcH;

  755.     int dstW= c->dstW;

  756.     int dstH= c->dstH;

  757.     int dst_stride = FFALIGN(dstW * sizeof(int16_t)+66, 16);

  758.     int flags, cpu_flags;

  759.     enum PixelFormat srcFormat= c->srcFormat;

  760.     enum PixelFormat dstFormat= c->dstFormat;

  761. 

  762.     cpu_flags = av_get_cpu_flags();

  763.     flags     = c->flags;

  764.     emms_c();

  765.     if (!rgb15to16) sws_rgb2rgb_init();

  766. 

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

  768. 

  769.     if (!sws_isSupportedInput(srcFormat)) {

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

  771.         return AVERROR(EINVAL);

  772.     }

  773.     if (!sws_isSupportedOutput(dstFormat)) {

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

  775.         return AVERROR(EINVAL);

  776.     }

  777. 

  778.     i= flags & ( SWS_POINT

  779.                 |SWS_AREA

  780.                 |SWS_BILINEAR

  781.                 |SWS_FAST_BILINEAR

  782.                 |SWS_BICUBIC

  783.                 |SWS_X

  784.                 |SWS_GAUSS

  785.                 |SWS_LANCZOS

  786.                 |SWS_SINC

  787.                 |SWS_SPLINE

  788.                 |SWS_BICUBLIN);

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

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

  791.         return AVERROR(EINVAL);

  792.     }

  793.     /* sanity check */

  794.     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

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

  796.                srcW, srcH, dstW, dstH);

  797.         return AVERROR(EINVAL);

  798.     }

  799. 

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

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

  802. 

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

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

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

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

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

  808. 

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

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

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

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

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

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

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

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

  817. 

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

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

  820. 

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

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

  823. 

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

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

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

  827. 

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

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

  830.       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8

  831.       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4

  832.       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE

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

  834.         c->chrSrcHSubSample=1;

  835. 

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

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

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

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

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

  841. 

  842.     /* unscaled special cases */

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

  844.         ff_get_unscaled_swscale(c);

  845. 

  846.         if (c->swScale) {

  847.             if (flags&SWS_PRINT_INFO)

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

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

  850.             return 0;

  851.         }

  852.     }

  853. 

  854.     c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;

  855.     if (c->srcBpc < 8)

  856.         c->srcBpc = 8;

  857.     c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;

  858.     if (c->dstBpc < 8)

  859.         c->dstBpc = 8;

  860.     if (isAnyRGB(srcFormat) || srcFormat == PIX_FMT_PAL8)

  861.         c->srcBpc = 16;

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

  863.         dst_stride <<= 1;

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

  865.     if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->srcBpc == 8 && c->dstBpc <= 10) {

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

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

  868.             if (flags&SWS_PRINT_INFO)

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

  870.         }

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

  872.     }

  873.     else

  874.         c->canMMX2BeUsed=0;

  875. 

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

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

  878. 

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

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

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

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

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

  884.     // first and last pixel

  885.     if (flags&SWS_FAST_BILINEAR) {

  886.         if (c->canMMX2BeUsed) {

  887.             c->lumXInc+= 20;

  888.             c->chrXInc+= 20;

  889.         }

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

  891.         else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX && c->dstBpc <= 10) {

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

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

  894.         }

  895.     }

  896. 

  897.     /* precalculate horizontal scaler filter coefficients */

  898.     {

  899. #if HAVE_MMX2

  900. // can't downscale !!!

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

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

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

  904. 

  905. #ifdef MAP_ANONYMOUS

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

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

  908. #elif HAVE_VIRTUALALLOC

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

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

  911. #else

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

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

  914. #endif

  915. 

  916. #ifdef MAP_ANONYMOUS

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

  918. #else

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

  920. #endif

  921.                 return AVERROR(ENOMEM);

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

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

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

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

  926. 

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

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

  929. 

  930. #ifdef MAP_ANONYMOUS

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

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

  933. #endif

  934.         } else

  935. #endif /* HAVE_MMX2 */

  936.         {

  937.             const int filterAlign=

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

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

  940.                 1;

  941. 

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

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

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

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

  946.                 goto fail;

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

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

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

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

  951.                 goto fail;

  952.         }

  953.     } // initialize horizontal stuff

  954. 

  955.     /* precalculate vertical scaler filter coefficients */

  956.     {

  957.         const int filterAlign=

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

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

  960.             1;

  961. 

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

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

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

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

  966.             goto fail;

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

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

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

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

  971.             goto fail;

  972. 

  973. #if HAVE_ALTIVEC

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

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

  976. 

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

  978.             int j;

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

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

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

  982.         }

  983. 

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

  985.             int j;

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

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

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

  989.         }

  990. #endif

  991.     }

  992. 

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

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

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

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

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

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

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

  1000. 

  1001.         nextSlice>>= c->chrSrcVSubSample;

  1002.         nextSlice<<= c->chrSrcVSubSample;

  1003.         if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)

  1004.             c->vLumBufSize= nextSlice - c->vLumFilterPos[i];

  1005.         if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))

  1006.             c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];

  1007.     }

  1008. 

  1009.     // allocate pixbufs (we use dynamic allocation because otherwise we would need to

  1010.     // allocate several megabytes to handle all possible cases)

  1011.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1012.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1013.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1014.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1015.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1016.     //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)

  1017.     /* align at 16 bytes for AltiVec */

  1018.     for (i=0; i<c->vLumBufSize; i++) {

  1019.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1020.         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];

  1021.     }

  1022.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1023.     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);

  1024.     c->uv_offx2 = dst_stride + 16;

  1025.     for (i=0; i<c->vChrBufSize; i++) {

  1026.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);

  1027.         c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];

  1028.         c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1029.     }

  1030.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

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

  1032.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

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

  1034.         }

  1035. 

  1036.     //try to avoid drawing green stuff between the right end and the stride end

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

  1038.         if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){

  1039.             av_assert0(c->dstBpc > 10);

  1040.             for(j=0; j<dst_stride/2+1; j++)

  1041.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1042.         } else

  1043.             for(j=0; j<dst_stride+1; j++)

  1044.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1045. 

  1046.     assert(c->chrDstH <= dstH);

  1047. 

  1048.     if (flags&SWS_PRINT_INFO) {

  1049.         if      (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1050.         else if (flags&SWS_BILINEAR)      av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1051.         else if (flags&SWS_BICUBIC)       av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1052.         else if (flags&SWS_X)             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1053.         else if (flags&SWS_POINT)         av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1054.         else if (flags&SWS_AREA)          av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1055.         else if (flags&SWS_BICUBLIN)      av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1056.         else if (flags&SWS_GAUSS)         av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1057.         else if (flags&SWS_SINC)          av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1058.         else if (flags&SWS_LANCZOS)       av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1059.         else if (flags&SWS_SPLINE)        av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1060.         else                              av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1061. 

  1062.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1063.                av_get_pix_fmt_name(srcFormat),

  1064. #ifdef DITHER1XBPP

  1065.                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||

  1066.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1067.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",

  1068. #else

  1069.                "",

  1070. #endif

  1071.                av_get_pix_fmt_name(dstFormat));

  1072. 

  1073.         if      (HAVE_MMX2     && cpu_flags & AV_CPU_FLAG_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");

  1074.         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1075.         else if (HAVE_MMX      && cpu_flags & AV_CPU_FLAG_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");

  1076.         else if (HAVE_ALTIVEC  && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1077.         else                                   av_log(c, AV_LOG_INFO, "using C\n");

  1078. 

  1079.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1080.         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1081.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1082.         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1083.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1084.     }

  1085. 

  1086.     c->swScale= ff_getSwsFunc(c);

  1087.     return 0;

  1088. fail: //FIXME replace things by appropriate error codes

  1089.     return -1;

  1090. }

  1091. 

  1092. #if FF_API_SWS_GETCONTEXT

  1093. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1094.                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1095.                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1096. {

  1097.     SwsContext *c;

  1098. 

  1099.     if(!(c=sws_alloc_context()))

  1100.         return NULL;

  1101. 

  1102.     c->flags= flags;

  1103.     c->srcW= srcW;

  1104.     c->srcH= srcH;

  1105.     c->dstW= dstW;

  1106.     c->dstH= dstH;

  1107.     c->srcRange = handle_jpeg(&srcFormat);

  1108.     c->dstRange = handle_jpeg(&dstFormat);

  1109.     c->srcFormat= srcFormat;

  1110.     c->dstFormat= dstFormat;

  1111. 

  1112.     if (param) {

  1113.         c->param[0] = param[0];

  1114.         c->param[1] = param[1];

  1115.     }

  1116.     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);

  1117. 

  1118.     if(sws_init_context(c, srcFilter, dstFilter) < 0){

  1119.         sws_freeContext(c);

  1120.         return NULL;

  1121.     }

  1122. 

  1123.     return c;

  1124. }

  1125. #endif

  1126. 

  1127. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1128.                                 float lumaSharpen, float chromaSharpen,

  1129.                                 float chromaHShift, float chromaVShift,

  1130.                                 int verbose)

  1131. {

  1132.     SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  1133.     if (!filter)

  1134.         return NULL;

  1135. 

  1136.     if (lumaGBlur!=0.0) {

  1137.         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  1138.         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  1139.     } else {

  1140.         filter->lumH= sws_getIdentityVec();

  1141.         filter->lumV= sws_getIdentityVec();

  1142.     }

  1143. 

  1144.     if (chromaGBlur!=0.0) {

  1145.         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  1146.         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  1147.     } else {

  1148.         filter->chrH= sws_getIdentityVec();

  1149.         filter->chrV= sws_getIdentityVec();

  1150.     }

  1151. 

  1152.     if (chromaSharpen!=0.0) {

  1153.         SwsVector *id= sws_getIdentityVec();

  1154.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1155.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1156.         sws_addVec(filter->chrH, id);

  1157.         sws_addVec(filter->chrV, id);

  1158.         sws_freeVec(id);

  1159.     }

  1160. 

  1161.     if (lumaSharpen!=0.0) {

  1162.         SwsVector *id= sws_getIdentityVec();

  1163.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1164.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1165.         sws_addVec(filter->lumH, id);

  1166.         sws_addVec(filter->lumV, id);

  1167.         sws_freeVec(id);

  1168.     }

  1169. 

  1170.     if (chromaHShift != 0.0)

  1171.         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  1172. 

  1173.     if (chromaVShift != 0.0)

  1174.         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  1175. 

  1176.     sws_normalizeVec(filter->chrH, 1.0);

  1177.     sws_normalizeVec(filter->chrV, 1.0);

  1178.     sws_normalizeVec(filter->lumH, 1.0);

  1179.     sws_normalizeVec(filter->lumV, 1.0);

  1180. 

  1181.     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1182.     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1183. 

  1184.     return filter;

  1185. }

  1186. 

  1187. SwsVector *sws_allocVec(int length)

  1188. {

  1189.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1190.     if (!vec)

  1191.         return NULL;

  1192.     vec->length = length;

  1193.     vec->coeff  = av_malloc(sizeof(double) * length);

  1194.     if (!vec->coeff)

  1195.         av_freep(&vec);

  1196.     return vec;

  1197. }

  1198. 

  1199. SwsVector *sws_getGaussianVec(double variance, double quality)

  1200. {

  1201.     const int length= (int)(variance*quality + 0.5) | 1;

  1202.     int i;

  1203.     double middle= (length-1)*0.5;

  1204.     SwsVector *vec= sws_allocVec(length);

  1205. 

  1206.     if (!vec)

  1207.         return NULL;

  1208. 

  1209.     for (i=0; i<length; i++) {

  1210.         double dist= i-middle;

  1211.         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);

  1212.     }

  1213. 

  1214.     sws_normalizeVec(vec, 1.0);

  1215. 

  1216.     return vec;

  1217. }

  1218. 

  1219. SwsVector *sws_getConstVec(double c, int length)

  1220. {

  1221.     int i;

  1222.     SwsVector *vec= sws_allocVec(length);

  1223. 

  1224.     if (!vec)

  1225.         return NULL;

  1226. 

  1227.     for (i=0; i<length; i++)

  1228.         vec->coeff[i]= c;

  1229. 

  1230.     return vec;

  1231. }

  1232. 

  1233. SwsVector *sws_getIdentityVec(void)

  1234. {

  1235.     return sws_getConstVec(1.0, 1);

  1236. }

  1237. 

  1238. static double sws_dcVec(SwsVector *a)

  1239. {

  1240.     int i;

  1241.     double sum=0;

  1242. 

  1243.     for (i=0; i<a->length; i++)

  1244.         sum+= a->coeff[i];

  1245. 

  1246.     return sum;

  1247. }

  1248. 

  1249. void sws_scaleVec(SwsVector *a, double scalar)

  1250. {

  1251.     int i;

  1252. 

  1253.     for (i=0; i<a->length; i++)

  1254.         a->coeff[i]*= scalar;

  1255. }

  1256. 

  1257. void sws_normalizeVec(SwsVector *a, double height)

  1258. {

  1259.     sws_scaleVec(a, height/sws_dcVec(a));

  1260. }

  1261. 

  1262. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1263. {

  1264.     int length= a->length + b->length - 1;

  1265.     int i, j;

  1266.     SwsVector *vec= sws_getConstVec(0.0, length);

  1267. 

  1268.     if (!vec)

  1269.         return NULL;

  1270. 

  1271.     for (i=0; i<a->length; i++) {

  1272.         for (j=0; j<b->length; j++) {

  1273.             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1274.         }

  1275.     }

  1276. 

  1277.     return vec;

  1278. }

  1279. 

  1280. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1281. {

  1282.     int length= FFMAX(a->length, b->length);

  1283.     int i;

  1284.     SwsVector *vec= sws_getConstVec(0.0, length);

  1285. 

  1286.     if (!vec)

  1287.         return NULL;

  1288. 

  1289.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1290.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];

  1291. 

  1292.     return vec;

  1293. }

  1294. 

  1295. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1296. {

  1297.     int length= FFMAX(a->length, b->length);

  1298.     int i;

  1299.     SwsVector *vec= sws_getConstVec(0.0, length);

  1300. 

  1301.     if (!vec)

  1302.         return NULL;

  1303. 

  1304.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1305.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1306. 

  1307.     return vec;

  1308. }

  1309. 

  1310. /* shift left / or right if "shift" is negative */

  1311. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1312. {

  1313.     int length= a->length + FFABS(shift)*2;

  1314.     int i;

  1315.     SwsVector *vec= sws_getConstVec(0.0, length);

  1316. 

  1317.     if (!vec)

  1318.         return NULL;

  1319. 

  1320.     for (i=0; i<a->length; i++) {

  1321.         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1322.     }

  1323. 

  1324.     return vec;

  1325. }

  1326. 

  1327. void sws_shiftVec(SwsVector *a, int shift)

  1328. {

  1329.     SwsVector *shifted= sws_getShiftedVec(a, shift);

  1330.     av_free(a->coeff);

  1331.     a->coeff= shifted->coeff;

  1332.     a->length= shifted->length;

  1333.     av_free(shifted);

  1334. }

  1335. 

  1336. void sws_addVec(SwsVector *a, SwsVector *b)

  1337. {

  1338.     SwsVector *sum= sws_sumVec(a, b);

  1339.     av_free(a->coeff);

  1340.     a->coeff= sum->coeff;

  1341.     a->length= sum->length;

  1342.     av_free(sum);

  1343. }

  1344. 

  1345. void sws_subVec(SwsVector *a, SwsVector *b)

  1346. {

  1347.     SwsVector *diff= sws_diffVec(a, b);

  1348.     av_free(a->coeff);

  1349.     a->coeff= diff->coeff;

  1350.     a->length= diff->length;

  1351.     av_free(diff);

  1352. }

  1353. 

  1354. void sws_convVec(SwsVector *a, SwsVector *b)

  1355. {

  1356.     SwsVector *conv= sws_getConvVec(a, b);

  1357.     av_free(a->coeff);

  1358.     a->coeff= conv->coeff;

  1359.     a->length= conv->length;

  1360.     av_free(conv);

  1361. }

  1362. 

  1363. SwsVector *sws_cloneVec(SwsVector *a)

  1364. {

  1365.     int i;

  1366.     SwsVector *vec= sws_allocVec(a->length);

  1367. 

  1368.     if (!vec)

  1369.         return NULL;

  1370. 

  1371.     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];

  1372. 

  1373.     return vec;

  1374. }

  1375. 

  1376. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1377. {

  1378.     int i;

  1379.     double max=0;

  1380.     double min=0;

  1381.     double range;

  1382. 

  1383.     for (i=0; i<a->length; i++)

  1384.         if (a->coeff[i]>max) max= a->coeff[i];

  1385. 

  1386.     for (i=0; i<a->length; i++)

  1387.         if (a->coeff[i]<min) min= a->coeff[i];

  1388. 

  1389.     range= max - min;

  1390. 

  1391.     for (i=0; i<a->length; i++) {

  1392.         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1393.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1394.         for (;x>0; x--) av_log(log_ctx, log_level, " ");

  1395.         av_log(log_ctx, log_level, "|\n");

  1396.     }

  1397. }

  1398. 

  1399. void sws_freeVec(SwsVector *a)

  1400. {

  1401.     if (!a) return;

  1402.     av_freep(&a->coeff);

  1403.     a->length=0;

  1404.     av_free(a);

  1405. }

  1406. 

  1407. void sws_freeFilter(SwsFilter *filter)

  1408. {

  1409.     if (!filter) return;

  1410. 

  1411.     if (filter->lumH) sws_freeVec(filter->lumH);

  1412.     if (filter->lumV) sws_freeVec(filter->lumV);

  1413.     if (filter->chrH) sws_freeVec(filter->chrH);

  1414.     if (filter->chrV) sws_freeVec(filter->chrV);

  1415.     av_free(filter);

  1416. }

  1417. 

  1418. void sws_freeContext(SwsContext *c)

  1419. {

  1420.     int i;

  1421.     if (!c) return;

  1422. 

  1423.     if (c->lumPixBuf) {

  1424.         for (i=0; i<c->vLumBufSize; i++)

  1425.             av_freep(&c->lumPixBuf[i]);

  1426.         av_freep(&c->lumPixBuf);

  1427.     }

  1428. 

  1429.     if (c->chrUPixBuf) {

  1430.         for (i=0; i<c->vChrBufSize; i++)

  1431.             av_freep(&c->chrUPixBuf[i]);

  1432.         av_freep(&c->chrUPixBuf);

  1433.         av_freep(&c->chrVPixBuf);

  1434.     }

  1435. 

  1436.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1437.         for (i=0; i<c->vLumBufSize; i++)

  1438.             av_freep(&c->alpPixBuf[i]);

  1439.         av_freep(&c->alpPixBuf);

  1440.     }

  1441. 

  1442.     av_freep(&c->vLumFilter);

  1443.     av_freep(&c->vChrFilter);

  1444.     av_freep(&c->hLumFilter);

  1445.     av_freep(&c->hChrFilter);

  1446. #if HAVE_ALTIVEC

  1447.     av_freep(&c->vYCoeffsBank);

  1448.     av_freep(&c->vCCoeffsBank);

  1449. #endif

  1450. 

  1451.     av_freep(&c->vLumFilterPos);

  1452.     av_freep(&c->vChrFilterPos);

  1453.     av_freep(&c->hLumFilterPos);

  1454.     av_freep(&c->hChrFilterPos);

  1455. 

  1456. #if HAVE_MMX

  1457. #ifdef MAP_ANONYMOUS

  1458.     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1459.     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1460. #elif HAVE_VIRTUALALLOC

  1461.     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1462.     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1463. #else

  1464.     av_free(c->lumMmx2FilterCode);

  1465.     av_free(c->chrMmx2FilterCode);

  1466. #endif

  1467.     c->lumMmx2FilterCode=NULL;

  1468.     c->chrMmx2FilterCode=NULL;

  1469. #endif /* HAVE_MMX */

  1470. 

  1471.     av_freep(&c->yuvTable);

  1472.     av_freep(&c->formatConvBuffer);

  1473. 

  1474.     av_free(c);

  1475. }

  1476. 

  1477. struct SwsContext *sws_getCachedContext(struct SwsContext *context,

  1478.                                         int srcW, int srcH, enum PixelFormat srcFormat,

  1479.                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1480.                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1481. {

  1482.     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};

  1483. 

  1484.     if (!param)

  1485.         param = default_param;

  1486. 

  1487.     if (context &&

  1488.         (context->srcW      != srcW      ||

  1489.          context->srcH      != srcH      ||

  1490.          context->srcFormat != srcFormat ||

  1491.          context->dstW      != dstW      ||

  1492.          context->dstH      != dstH      ||

  1493.          context->dstFormat != dstFormat ||

  1494.          context->flags     != flags     ||

  1495.          context->param[0]  != param[0]  ||

  1496.          context->param[1]  != param[1])) {

  1497.         sws_freeContext(context);

  1498.         context = NULL;

  1499.     }

  1500. 

  1501.     if (!context) {

  1502.         if (!(context = sws_alloc_context()))

  1503.             return NULL;

  1504.         context->srcW      = srcW;

  1505.         context->srcH      = srcH;

  1506.         context->srcRange  = handle_jpeg(&srcFormat);

  1507.         context->srcFormat = srcFormat;

  1508.         context->dstW      = dstW;

  1509.         context->dstH      = dstH;

  1510.         context->dstRange  = handle_jpeg(&dstFormat);

  1511.         context->dstFormat = dstFormat;

  1512.         context->flags     = flags;

  1513.         context->param[0]  = param[0];

  1514.         context->param[1]  = param[1];

  1515.         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);

  1516.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1517.             sws_freeContext(context);

  1518.             return NULL;

  1519.         }

  1520.     }

  1521.     return context;

  1522. }