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.
35015 Commits
32 Branches
1.0GB
Tree: 90c02ae142
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '90c02ae142'
${ noResults }
FFmpeg/libswscale/utils.c

1538 lines
55KB
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_YUV422P9BE]  = { 1 , 1 },

  140.     [PIX_FMT_YUV422P9LE]  = { 1 , 1 },

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

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

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

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

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

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

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

  148. };

  149. 

  150. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  151. {

  152.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  153.         format_entries[pix_fmt].is_supported_in : 0;

  154. }

  155. 

  156. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  157. {

  158.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  159.         format_entries[pix_fmt].is_supported_out : 0;

  160. }

  161. 

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

  163. 

  164. #if FF_API_SWS_FORMAT_NAME

  165. const char *sws_format_name(enum PixelFormat format)

  166. {

  167.     return av_get_pix_fmt_name(format);

  168. }

  169. #endif

  170. 

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

  172. {

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

  174.     else           return getSplineCoeff(        0.0,

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

  176.                                                  c + 3.0*d,

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

  178.                                          dist-1.0);

  179. }

  180. 

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

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

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

  184. {

  185.     int i;

  186.     int filterSize;

  187.     int filter2Size;

  188.     int minFilterSize;

  189.     int64_t *filter=NULL;

  190.     int64_t *filter2=NULL;

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

  192.     int ret= -1;

  193. 

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

  195. 

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

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

  198. 

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

  200.         int i;

  201.         filterSize= 1;

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

  203. 

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

  205.             filter[i*filterSize]= fone;

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

  207.         }

  208. 

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

  210.         int i;

  211.         int xDstInSrc;

  212.         filterSize= 1;

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

  214. 

  215.         xDstInSrc= xInc/2 - 0x8000;

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

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

  218. 

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

  220.             filter[i]= fone;

  221.             xDstInSrc+= xInc;

  222.         }

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

  224.         int i;

  225.         int xDstInSrc;

  226.         filterSize= 2;

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

  228. 

  229.         xDstInSrc= xInc/2 - 0x8000;

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

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

  232.             int j;

  233. 

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

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

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

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

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

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

  240.                 xx++;

  241.             }

  242.             xDstInSrc+= xInc;

  243.         }

  244.     } else {

  245.         int xDstInSrc;

  246.         int sizeFactor;

  247. 

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

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

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

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

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

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

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

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

  256.         else {

  257.             sizeFactor= 0; //GCC warning killer

  258.             assert(0);

  259.         }

  260. 

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

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

  263. 

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

  265. 

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

  267. 

  268.         xDstInSrc= xInc - 0x10000;

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

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

  271.             int j;

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

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

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

  275.                 double floatd;

  276.                 int64_t coeff;

  277. 

  278.                 if (xInc > 1<<16)

  279.                     d= d*dstW/srcW;

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

  281. 

  282.                 if (flags & SWS_BICUBIC) {

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

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

  285. 

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

  287.                         coeff = 0.0;

  288.                     } else {

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

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

  291. 

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

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

  294.                         else

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

  296.                     }

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

  298.                 }

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

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

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

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

  303.                 }*/

  304.                 else if (flags & SWS_X) {

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

  306.                     double c;

  307. 

  308.                     if (floatd<1.0)

  309.                         c = cos(floatd*M_PI);

  310.                     else

  311.                         c=-1.0;

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

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

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

  315.                 } else if (flags & SWS_AREA) {

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

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

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

  319.                     else coeff=0.0;

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

  321.                 } else if (flags & SWS_GAUSS) {

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

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

  324.                 } else if (flags & SWS_SINC) {

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

  326.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  330.                 } else if (flags & SWS_BILINEAR) {

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

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

  333.                     coeff *= fone >> 30;

  334.                 } else if (flags & SWS_SPLINE) {

  335.                     double p=-2.196152422706632;

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

  337.                 } else {

  338.                     coeff= 0.0; //GCC warning killer

  339.                     assert(0);

  340.                 }

  341. 

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

  343.                 xx++;

  344.             }

  345.             xDstInSrc+= 2*xInc;

  346.         }

  347.     }

  348. 

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

  350.        av_free(filter);

  351.     */

  352.     assert(filterSize>0);

  353.     filter2Size= filterSize;

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

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

  356.     assert(filter2Size>0);

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

  358. 

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

  360.         int j, k;

  361. 

  362.         if(srcFilter) {

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

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

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

  366.             }

  367.         } else {

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

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

  370.         }

  371.         //FIXME dstFilter

  372. 

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

  374.     }

  375.     av_freep(&filter);

  376. 

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

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

  379.     minFilterSize= 0;

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

  381.         int min= filter2Size;

  382.         int j;

  383.         int64_t cutOff=0.0;

  384. 

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

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

  387.             int k;

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

  389. 

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

  391. 

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

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

  394. 

  395.             // move filter coefficients left

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

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

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

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

  400.         }

  401. 

  402.         cutOff=0;

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

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

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

  406. 

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

  408.             min--;

  409.         }

  410. 

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

  412.     }

  413. 

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

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

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

  417.         if (minFilterSize < 5)

  418.             filterAlign = 4;

  419. 

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

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

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

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

  424.         // decent-sized vector.

  425.         if (minFilterSize < 3)

  426.             filterAlign = 1;

  427.     }

  428. 

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

  430.         // special case for unscaled vertical filtering

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

  432.             filterAlign= 1;

  433.     }

  434. 

  435.     assert(minFilterSize > 0);

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

  437.     assert(filterSize > 0);

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

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

  440.         goto fail;

  441.     *outFilterSize= filterSize;

  442. 

  443.     if (flags&SWS_PRINT_INFO)

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

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

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

  447.         int j;

  448. 

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

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

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

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

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

  454.         }

  455.     }

  456. 

  457.     //FIXME try to align filterPos if possible

  458. 

  459.     //fix borders

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

  461.         int j;

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

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

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

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

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

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

  468.             }

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

  470.         }

  471. 

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

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

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

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

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

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

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

  479.             }

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

  481.         }

  482.     }

  483. 

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

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

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

  487. 

  488.     /* normalize & store in outFilter */

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

  490.         int j;

  491.         int64_t error=0;

  492.         int64_t sum=0;

  493. 

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

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

  496.         }

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

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

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

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

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

  502.             error= v - intV*sum;

  503.         }

  504.     }

  505. 

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

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

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

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

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

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

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

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

  514.     }

  515. 

  516.     ret=0;

  517. fail:

  518.     av_free(filter);

  519.     av_free(filter2);

  520.     return ret;

  521. }

  522. 

  523. #if HAVE_MMX2

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

  525. {

  526.     uint8_t *fragmentA;

  527.     x86_reg imm8OfPShufW1A;

  528.     x86_reg imm8OfPShufW2A;

  529.     x86_reg fragmentLengthA;

  530.     uint8_t *fragmentB;

  531.     x86_reg imm8OfPShufW1B;

  532.     x86_reg imm8OfPShufW2B;

  533.     x86_reg fragmentLengthB;

  534.     int fragmentPos;

  535. 

  536.     int xpos, i;

  537. 

  538.     // create an optimized horizontal scaling routine

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

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

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

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

  543.      * of fragmentA is used.

  544.      */

  545. 

  546.     //code fragment

  547. 

  548.     __asm__ volatile(

  549.         "jmp                         9f                 \n\t"

  550.     // Begin

  551.         "0:                                             \n\t"

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

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

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

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

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

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

  558.         "1:                                             \n\t"

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

  560.         "2:                                             \n\t"

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

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

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

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

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

  566. 

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

  568. 

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

  570.     // End

  571.         "9:                                             \n\t"

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

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

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

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

  576.         "dec                         %1                 \n\t"

  577.         "dec                         %2                 \n\t"

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

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

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

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

  582. 

  583. 

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

  585.         "=r" (fragmentLengthA)

  586.     );

  587. 

  588.     __asm__ volatile(

  589.         "jmp                         9f                 \n\t"

  590.     // Begin

  591.         "0:                                             \n\t"

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

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

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

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

  596.         "1:                                             \n\t"

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

  598.         "2:                                             \n\t"

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

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

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

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

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

  604. 

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

  606. 

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

  608.     // End

  609.         "9:                                             \n\t"

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

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

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

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

  614.         "dec                         %1                 \n\t"

  615.         "dec                         %2                 \n\t"

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

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

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

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

  620. 

  621. 

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

  623.         "=r" (fragmentLengthB)

  624.     );

  625. 

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

  627.     fragmentPos=0;

  628. 

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

  630.         int xx=xpos>>16;

  631. 

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

  633.             int a=0;

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

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

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

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

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

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

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

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

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

  643.             int shift=0;

  644. 

  645.             if (filterCode) {

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

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

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

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

  650.                 filterPos[i/2]= xx;

  651. 

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

  653. 

  654.                 filterCode[fragmentPos + imm8OfPShufW1]=

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

  656.                 filterCode[fragmentPos + imm8OfPShufW2]=

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

  658. 

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

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

  661. 

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

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

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

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

  666.                 }

  667.             }

  668. 

  669.             fragmentPos+= fragmentLength;

  670. 

  671.             if (filterCode)

  672.                 filterCode[fragmentPos]= RET;

  673.         }

  674.         xpos+=xInc;

  675.     }

  676.     if (filterCode)

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

  678. 

  679.     return fragmentPos + 1;

  680. }

  681. #endif /* HAVE_MMX2 */

  682. 

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

  684. {

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

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

  687. }

  688. 

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

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

  691.                              int brightness, int contrast, int saturation)

  692. {

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

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

  695. 

  696.     c->brightness= brightness;

  697.     c->contrast  = contrast;

  698.     c->saturation= saturation;

  699.     c->srcRange  = srcRange;

  700.     c->dstRange  = dstRange;

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

  702. 

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

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

  705. 

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

  707.     //FIXME factorize

  708. 

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

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

  711.     return 0;

  712. }

  713. 

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

  715.                              int *srcRange, int **table, int *dstRange,

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

  717. {

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

  719. 

  720.     *inv_table = c->srcColorspaceTable;

  721.     *table     = c->dstColorspaceTable;

  722.     *srcRange  = c->srcRange;

  723.     *dstRange  = c->dstRange;

  724.     *brightness= c->brightness;

  725.     *contrast  = c->contrast;

  726.     *saturation= c->saturation;

  727. 

  728.     return 0;

  729. }

  730. 

  731. static int handle_jpeg(enum PixelFormat *format)

  732. {

  733.     switch (*format) {

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

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

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

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

  738.     default:                                          return 0;

  739.     }

  740. }

  741. 

  742. SwsContext *sws_alloc_context(void)

  743. {

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

  745. 

  746.     c->av_class = &sws_context_class;

  747.     av_opt_set_defaults(c);

  748. 

  749.     return c;

  750. }

  751. 

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

  753. {

  754.     int i, j;

  755.     int usesVFilter, usesHFilter;

  756.     int unscaled;

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

  758.     int srcW= c->srcW;

  759.     int srcH= c->srcH;

  760.     int dstW= c->dstW;

  761.     int dstH= c->dstH;

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

  763.     int flags, cpu_flags;

  764.     enum PixelFormat srcFormat= c->srcFormat;

  765.     enum PixelFormat dstFormat= c->dstFormat;

  766. 

  767.     cpu_flags = av_get_cpu_flags();

  768.     flags     = c->flags;

  769.     emms_c();

  770.     if (!rgb15to16) sws_rgb2rgb_init();

  771. 

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

  773. 

  774.     handle_jpeg(&srcFormat);

  775.     handle_jpeg(&dstFormat);

  776. 

  777.     if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){

  778.         av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");

  779.         c->srcFormat= srcFormat;

  780.         c->dstFormat= dstFormat;

  781.     }

  782. 

  783.     if (!sws_isSupportedInput(srcFormat)) {

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

  785.         return AVERROR(EINVAL);

  786.     }

  787.     if (!sws_isSupportedOutput(dstFormat)) {

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

  789.         return AVERROR(EINVAL);

  790.     }

  791. 

  792.     i= flags & ( SWS_POINT

  793.                 |SWS_AREA

  794.                 |SWS_BILINEAR

  795.                 |SWS_FAST_BILINEAR

  796.                 |SWS_BICUBIC

  797.                 |SWS_X

  798.                 |SWS_GAUSS

  799.                 |SWS_LANCZOS

  800.                 |SWS_SINC

  801.                 |SWS_SPLINE

  802.                 |SWS_BICUBLIN);

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

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

  805.         return AVERROR(EINVAL);

  806.     }

  807.     /* sanity check */

  808.     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

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

  810.                srcW, srcH, dstW, dstH);

  811.         return AVERROR(EINVAL);

  812.     }

  813. 

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

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

  816. 

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

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

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

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

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

  822. 

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

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

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

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

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

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

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

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

  831. 

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

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

  834. 

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

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

  837. 

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

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

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

  841. 

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

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

  844.       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8

  845.       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4

  846.       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE

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

  848.         c->chrSrcHSubSample=1;

  849. 

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

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

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

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

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

  855. 

  856.     /* unscaled special cases */

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

  858.         ff_get_unscaled_swscale(c);

  859. 

  860.         if (c->swScale) {

  861.             if (flags&SWS_PRINT_INFO)

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

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

  864.             return 0;

  865.         }

  866.     }

  867. 

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

  869.     if (c->srcBpc < 8)

  870.         c->srcBpc = 8;

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

  872.     if (c->dstBpc < 8)

  873.         c->dstBpc = 8;

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

  875.         c->srcBpc = 16;

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

  877.         dst_stride <<= 1;

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

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

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

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

  882.             if (flags&SWS_PRINT_INFO)

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

  884.         }

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

  886.     }

  887.     else

  888.         c->canMMX2BeUsed=0;

  889. 

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

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

  892. 

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

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

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

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

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

  898.     // first and last pixel

  899.     if (flags&SWS_FAST_BILINEAR) {

  900.         if (c->canMMX2BeUsed) {

  901.             c->lumXInc+= 20;

  902.             c->chrXInc+= 20;

  903.         }

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

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

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

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

  908.         }

  909.     }

  910. 

  911.     /* precalculate horizontal scaler filter coefficients */

  912.     {

  913. #if HAVE_MMX2

  914. // can't downscale !!!

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

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

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

  918. 

  919. #ifdef MAP_ANONYMOUS

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

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

  922. #elif HAVE_VIRTUALALLOC

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

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

  925. #else

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

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

  928. #endif

  929. 

  930. #ifdef MAP_ANONYMOUS

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

  932. #else

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

  934. #endif

  935.                 return AVERROR(ENOMEM);

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

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

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

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

  940. 

  941.             initMMX2HScaler(      dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);

  942.             initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);

  943. 

  944. #ifdef MAP_ANONYMOUS

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

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

  947. #endif

  948.         } else

  949. #endif /* HAVE_MMX2 */

  950.         {

  951.             const int filterAlign=

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

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

  954.                 1;

  955. 

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

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

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

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

  960.                 goto fail;

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

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

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

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

  965.                 goto fail;

  966.         }

  967.     } // initialize horizontal stuff

  968. 

  969.     /* precalculate vertical scaler filter coefficients */

  970.     {

  971.         const int filterAlign=

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

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

  974.             1;

  975. 

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

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

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

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

  980.             goto fail;

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

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

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

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

  985.             goto fail;

  986. 

  987. #if HAVE_ALTIVEC

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

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

  990. 

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

  992.             int j;

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

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

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

  996.         }

  997. 

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

  999.             int j;

  1000.             short *p = (short *)&c->vCCoeffsBank[i];

  1001.             for (j=0;j<8;j++)

  1002.                 p[j] = c->vChrFilter[i];

  1003.         }

  1004. #endif

  1005.     }

  1006. 

  1007.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1008.     c->vLumBufSize= c->vLumFilterSize;

  1009.     c->vChrBufSize= c->vChrFilterSize;

  1010.     for (i=0; i<dstH; i++) {

  1011.         int chrI= (int64_t)i*c->chrDstH / dstH;

  1012.         int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,

  1013.                            ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));

  1014. 

  1015.         nextSlice>>= c->chrSrcVSubSample;

  1016.         nextSlice<<= c->chrSrcVSubSample;

  1017.         if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)

  1018.             c->vLumBufSize= nextSlice - c->vLumFilterPos[i];

  1019.         if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))

  1020.             c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];

  1021.     }

  1022. 

  1023.     // allocate pixbufs (we use dynamic allocation because otherwise we would need to

  1024.     // allocate several megabytes to handle all possible cases)

  1025.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

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

  1027.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1028.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1029.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1030.     //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)

  1031.     /* align at 16 bytes for AltiVec */

  1032.     for (i=0; i<c->vLumBufSize; i++) {

  1033.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1034.         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];

  1035.     }

  1036.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1037.     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);

  1038.     c->uv_offx2 = dst_stride + 16;

  1039.     for (i=0; i<c->vChrBufSize; i++) {

  1040.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);

  1041.         c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];

  1042.         c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1043.     }

  1044.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1045.         for (i=0; i<c->vLumBufSize; i++) {

  1046.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1047.             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];

  1048.         }

  1049. 

  1050.     //try to avoid drawing green stuff between the right end and the stride end

  1051.     for (i=0; i<c->vChrBufSize; i++)

  1052.         if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){

  1053.             av_assert0(c->dstBpc > 10);

  1054.             for(j=0; j<dst_stride/2+1; j++)

  1055.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1056.         } else

  1057.             for(j=0; j<dst_stride+1; j++)

  1058.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1059. 

  1060.     assert(c->chrDstH <= dstH);

  1061. 

  1062.     if (flags&SWS_PRINT_INFO) {

  1063.         if      (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1064.         else if (flags&SWS_BILINEAR)      av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1065.         else if (flags&SWS_BICUBIC)       av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1066.         else if (flags&SWS_X)             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1067.         else if (flags&SWS_POINT)         av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1068.         else if (flags&SWS_AREA)          av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1069.         else if (flags&SWS_BICUBLIN)      av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1070.         else if (flags&SWS_GAUSS)         av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1071.         else if (flags&SWS_SINC)          av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1072.         else if (flags&SWS_LANCZOS)       av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1073.         else if (flags&SWS_SPLINE)        av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1074.         else                              av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1075. 

  1076.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1077.                av_get_pix_fmt_name(srcFormat),

  1078. #ifdef DITHER1XBPP

  1079.                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||

  1080.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1081.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",

  1082. #else

  1083.                "",

  1084. #endif

  1085.                av_get_pix_fmt_name(dstFormat));

  1086. 

  1087.         if      (HAVE_MMX2     && cpu_flags & AV_CPU_FLAG_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");

  1088.         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1089.         else if (HAVE_MMX      && cpu_flags & AV_CPU_FLAG_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");

  1090.         else if (HAVE_ALTIVEC  && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1091.         else                                   av_log(c, AV_LOG_INFO, "using C\n");

  1092. 

  1093.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1094.         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1095.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1096.         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1097.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1098.     }

  1099. 

  1100.     c->swScale= ff_getSwsFunc(c);

  1101.     return 0;

  1102. fail: //FIXME replace things by appropriate error codes

  1103.     return -1;

  1104. }

  1105. 

  1106. #if FF_API_SWS_GETCONTEXT

  1107. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1108.                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1109.                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1110. {

  1111.     SwsContext *c;

  1112. 

  1113.     if(!(c=sws_alloc_context()))

  1114.         return NULL;

  1115. 

  1116.     c->flags= flags;

  1117.     c->srcW= srcW;

  1118.     c->srcH= srcH;

  1119.     c->dstW= dstW;

  1120.     c->dstH= dstH;

  1121.     c->srcRange = handle_jpeg(&srcFormat);

  1122.     c->dstRange = handle_jpeg(&dstFormat);

  1123.     c->srcFormat= srcFormat;

  1124.     c->dstFormat= dstFormat;

  1125. 

  1126.     if (param) {

  1127.         c->param[0] = param[0];

  1128.         c->param[1] = param[1];

  1129.     }

  1130.     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);

  1131. 

  1132.     if(sws_init_context(c, srcFilter, dstFilter) < 0){

  1133.         sws_freeContext(c);

  1134.         return NULL;

  1135.     }

  1136. 

  1137.     return c;

  1138. }

  1139. #endif

  1140. 

  1141. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1142.                                 float lumaSharpen, float chromaSharpen,

  1143.                                 float chromaHShift, float chromaVShift,

  1144.                                 int verbose)

  1145. {

  1146.     SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  1147.     if (!filter)

  1148.         return NULL;

  1149. 

  1150.     if (lumaGBlur!=0.0) {

  1151.         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  1152.         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  1153.     } else {

  1154.         filter->lumH= sws_getIdentityVec();

  1155.         filter->lumV= sws_getIdentityVec();

  1156.     }

  1157. 

  1158.     if (chromaGBlur!=0.0) {

  1159.         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  1160.         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  1161.     } else {

  1162.         filter->chrH= sws_getIdentityVec();

  1163.         filter->chrV= sws_getIdentityVec();

  1164.     }

  1165. 

  1166.     if (chromaSharpen!=0.0) {

  1167.         SwsVector *id= sws_getIdentityVec();

  1168.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1169.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1170.         sws_addVec(filter->chrH, id);

  1171.         sws_addVec(filter->chrV, id);

  1172.         sws_freeVec(id);

  1173.     }

  1174. 

  1175.     if (lumaSharpen!=0.0) {

  1176.         SwsVector *id= sws_getIdentityVec();

  1177.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1178.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1179.         sws_addVec(filter->lumH, id);

  1180.         sws_addVec(filter->lumV, id);

  1181.         sws_freeVec(id);

  1182.     }

  1183. 

  1184.     if (chromaHShift != 0.0)

  1185.         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  1186. 

  1187.     if (chromaVShift != 0.0)

  1188.         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  1189. 

  1190.     sws_normalizeVec(filter->chrH, 1.0);

  1191.     sws_normalizeVec(filter->chrV, 1.0);

  1192.     sws_normalizeVec(filter->lumH, 1.0);

  1193.     sws_normalizeVec(filter->lumV, 1.0);

  1194. 

  1195.     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1196.     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1197. 

  1198.     return filter;

  1199. }

  1200. 

  1201. SwsVector *sws_allocVec(int length)

  1202. {

  1203.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1204.     if (!vec)

  1205.         return NULL;

  1206.     vec->length = length;

  1207.     vec->coeff  = av_malloc(sizeof(double) * length);

  1208.     if (!vec->coeff)

  1209.         av_freep(&vec);

  1210.     return vec;

  1211. }

  1212. 

  1213. SwsVector *sws_getGaussianVec(double variance, double quality)

  1214. {

  1215.     const int length= (int)(variance*quality + 0.5) | 1;

  1216.     int i;

  1217.     double middle= (length-1)*0.5;

  1218.     SwsVector *vec= sws_allocVec(length);

  1219. 

  1220.     if (!vec)

  1221.         return NULL;

  1222. 

  1223.     for (i=0; i<length; i++) {

  1224.         double dist= i-middle;

  1225.         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);

  1226.     }

  1227. 

  1228.     sws_normalizeVec(vec, 1.0);

  1229. 

  1230.     return vec;

  1231. }

  1232. 

  1233. SwsVector *sws_getConstVec(double c, int length)

  1234. {

  1235.     int i;

  1236.     SwsVector *vec= sws_allocVec(length);

  1237. 

  1238.     if (!vec)

  1239.         return NULL;

  1240. 

  1241.     for (i=0; i<length; i++)

  1242.         vec->coeff[i]= c;

  1243. 

  1244.     return vec;

  1245. }

  1246. 

  1247. SwsVector *sws_getIdentityVec(void)

  1248. {

  1249.     return sws_getConstVec(1.0, 1);

  1250. }

  1251. 

  1252. static double sws_dcVec(SwsVector *a)

  1253. {

  1254.     int i;

  1255.     double sum=0;

  1256. 

  1257.     for (i=0; i<a->length; i++)

  1258.         sum+= a->coeff[i];

  1259. 

  1260.     return sum;

  1261. }

  1262. 

  1263. void sws_scaleVec(SwsVector *a, double scalar)

  1264. {

  1265.     int i;

  1266. 

  1267.     for (i=0; i<a->length; i++)

  1268.         a->coeff[i]*= scalar;

  1269. }

  1270. 

  1271. void sws_normalizeVec(SwsVector *a, double height)

  1272. {

  1273.     sws_scaleVec(a, height/sws_dcVec(a));

  1274. }

  1275. 

  1276. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1277. {

  1278.     int length= a->length + b->length - 1;

  1279.     int i, j;

  1280.     SwsVector *vec= sws_getConstVec(0.0, length);

  1281. 

  1282.     if (!vec)

  1283.         return NULL;

  1284. 

  1285.     for (i=0; i<a->length; i++) {

  1286.         for (j=0; j<b->length; j++) {

  1287.             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1288.         }

  1289.     }

  1290. 

  1291.     return vec;

  1292. }

  1293. 

  1294. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1295. {

  1296.     int length= FFMAX(a->length, b->length);

  1297.     int i;

  1298.     SwsVector *vec= sws_getConstVec(0.0, length);

  1299. 

  1300.     if (!vec)

  1301.         return NULL;

  1302. 

  1303.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

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

  1305. 

  1306.     return vec;

  1307. }

  1308. 

  1309. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1310. {

  1311.     int length= FFMAX(a->length, b->length);

  1312.     int i;

  1313.     SwsVector *vec= sws_getConstVec(0.0, length);

  1314. 

  1315.     if (!vec)

  1316.         return NULL;

  1317. 

  1318.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1319.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1320. 

  1321.     return vec;

  1322. }

  1323. 

  1324. /* shift left / or right if "shift" is negative */

  1325. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1326. {

  1327.     int length= a->length + FFABS(shift)*2;

  1328.     int i;

  1329.     SwsVector *vec= sws_getConstVec(0.0, length);

  1330. 

  1331.     if (!vec)

  1332.         return NULL;

  1333. 

  1334.     for (i=0; i<a->length; i++) {

  1335.         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1336.     }

  1337. 

  1338.     return vec;

  1339. }

  1340. 

  1341. void sws_shiftVec(SwsVector *a, int shift)

  1342. {

  1343.     SwsVector *shifted= sws_getShiftedVec(a, shift);

  1344.     av_free(a->coeff);

  1345.     a->coeff= shifted->coeff;

  1346.     a->length= shifted->length;

  1347.     av_free(shifted);

  1348. }

  1349. 

  1350. void sws_addVec(SwsVector *a, SwsVector *b)

  1351. {

  1352.     SwsVector *sum= sws_sumVec(a, b);

  1353.     av_free(a->coeff);

  1354.     a->coeff= sum->coeff;

  1355.     a->length= sum->length;

  1356.     av_free(sum);

  1357. }

  1358. 

  1359. void sws_subVec(SwsVector *a, SwsVector *b)

  1360. {

  1361.     SwsVector *diff= sws_diffVec(a, b);

  1362.     av_free(a->coeff);

  1363.     a->coeff= diff->coeff;

  1364.     a->length= diff->length;

  1365.     av_free(diff);

  1366. }

  1367. 

  1368. void sws_convVec(SwsVector *a, SwsVector *b)

  1369. {

  1370.     SwsVector *conv= sws_getConvVec(a, b);

  1371.     av_free(a->coeff);

  1372.     a->coeff= conv->coeff;

  1373.     a->length= conv->length;

  1374.     av_free(conv);

  1375. }

  1376. 

  1377. SwsVector *sws_cloneVec(SwsVector *a)

  1378. {

  1379.     int i;

  1380.     SwsVector *vec= sws_allocVec(a->length);

  1381. 

  1382.     if (!vec)

  1383.         return NULL;

  1384. 

  1385.     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];

  1386. 

  1387.     return vec;

  1388. }

  1389. 

  1390. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1391. {

  1392.     int i;

  1393.     double max=0;

  1394.     double min=0;

  1395.     double range;

  1396. 

  1397.     for (i=0; i<a->length; i++)

  1398.         if (a->coeff[i]>max) max= a->coeff[i];

  1399. 

  1400.     for (i=0; i<a->length; i++)

  1401.         if (a->coeff[i]<min) min= a->coeff[i];

  1402. 

  1403.     range= max - min;

  1404. 

  1405.     for (i=0; i<a->length; i++) {

  1406.         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1407.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1408.         for (;x>0; x--) av_log(log_ctx, log_level, " ");

  1409.         av_log(log_ctx, log_level, "|\n");

  1410.     }

  1411. }

  1412. 

  1413. void sws_freeVec(SwsVector *a)

  1414. {

  1415.     if (!a) return;

  1416.     av_freep(&a->coeff);

  1417.     a->length=0;

  1418.     av_free(a);

  1419. }

  1420. 

  1421. void sws_freeFilter(SwsFilter *filter)

  1422. {

  1423.     if (!filter) return;

  1424. 

  1425.     if (filter->lumH) sws_freeVec(filter->lumH);

  1426.     if (filter->lumV) sws_freeVec(filter->lumV);

  1427.     if (filter->chrH) sws_freeVec(filter->chrH);

  1428.     if (filter->chrV) sws_freeVec(filter->chrV);

  1429.     av_free(filter);

  1430. }

  1431. 

  1432. void sws_freeContext(SwsContext *c)

  1433. {

  1434.     int i;

  1435.     if (!c) return;

  1436. 

  1437.     if (c->lumPixBuf) {

  1438.         for (i=0; i<c->vLumBufSize; i++)

  1439.             av_freep(&c->lumPixBuf[i]);

  1440.         av_freep(&c->lumPixBuf);

  1441.     }

  1442. 

  1443.     if (c->chrUPixBuf) {

  1444.         for (i=0; i<c->vChrBufSize; i++)

  1445.             av_freep(&c->chrUPixBuf[i]);

  1446.         av_freep(&c->chrUPixBuf);

  1447.         av_freep(&c->chrVPixBuf);

  1448.     }

  1449. 

  1450.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1451.         for (i=0; i<c->vLumBufSize; i++)

  1452.             av_freep(&c->alpPixBuf[i]);

  1453.         av_freep(&c->alpPixBuf);

  1454.     }

  1455. 

  1456.     av_freep(&c->vLumFilter);

  1457.     av_freep(&c->vChrFilter);

  1458.     av_freep(&c->hLumFilter);

  1459.     av_freep(&c->hChrFilter);

  1460. #if HAVE_ALTIVEC

  1461.     av_freep(&c->vYCoeffsBank);

  1462.     av_freep(&c->vCCoeffsBank);

  1463. #endif

  1464. 

  1465.     av_freep(&c->vLumFilterPos);

  1466.     av_freep(&c->vChrFilterPos);

  1467.     av_freep(&c->hLumFilterPos);

  1468.     av_freep(&c->hChrFilterPos);

  1469. 

  1470. #if HAVE_MMX

  1471. #ifdef MAP_ANONYMOUS

  1472.     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1473.     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1474. #elif HAVE_VIRTUALALLOC

  1475.     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1476.     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1477. #else

  1478.     av_free(c->lumMmx2FilterCode);

  1479.     av_free(c->chrMmx2FilterCode);

  1480. #endif

  1481.     c->lumMmx2FilterCode=NULL;

  1482.     c->chrMmx2FilterCode=NULL;

  1483. #endif /* HAVE_MMX */

  1484. 

  1485.     av_freep(&c->yuvTable);

  1486.     av_freep(&c->formatConvBuffer);

  1487. 

  1488.     av_free(c);

  1489. }

  1490. 

  1491. struct SwsContext *sws_getCachedContext(struct SwsContext *context,

  1492.                                         int srcW, int srcH, enum PixelFormat srcFormat,

  1493.                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1494.                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1495. {

  1496.     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};

  1497. 

  1498.     if (!param)

  1499.         param = default_param;

  1500. 

  1501.     if (context &&

  1502.         (context->srcW      != srcW      ||

  1503.          context->srcH      != srcH      ||

  1504.          context->srcFormat != srcFormat ||

  1505.          context->dstW      != dstW      ||

  1506.          context->dstH      != dstH      ||

  1507.          context->dstFormat != dstFormat ||

  1508.          context->flags     != flags     ||

  1509.          context->param[0]  != param[0]  ||

  1510.          context->param[1]  != param[1])) {

  1511.         sws_freeContext(context);

  1512.         context = NULL;

  1513.     }

  1514. 

  1515.     if (!context) {

  1516.         if (!(context = sws_alloc_context()))

  1517.             return NULL;

  1518.         context->srcW      = srcW;

  1519.         context->srcH      = srcH;

  1520.         context->srcRange  = handle_jpeg(&srcFormat);

  1521.         context->srcFormat = srcFormat;

  1522.         context->dstW      = dstW;

  1523.         context->dstH      = dstH;

  1524.         context->dstRange  = handle_jpeg(&dstFormat);

  1525.         context->dstFormat = dstFormat;

  1526.         context->flags     = flags;

  1527.         context->param[0]  = param[0];

  1528.         context->param[1]  = param[1];

  1529.         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);

  1530.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1531.             sws_freeContext(context);

  1532.             return NULL;

  1533.         }

  1534.     }

  1535.     return context;

  1536. }