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.
31116 Commits
32 Branches
301MB
Tree: 5089ce1b5a
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '5089ce1b5a'
${ noResults }
FFmpeg/libswscale/utils.c

1534 lines
55KB
Raw Blame History 

  1. /*

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

  3.  *

  4.  * This file is part of Libav.

  5.  *

  6.  * Libav is free software; you can redistribute it and/or

  7.  * modify it under the terms of the GNU Lesser General Public

  8.  * License as published by the Free Software Foundation; either

  9.  * version 2.1 of the License, or (at your option) any later version.

  10.  *

  11.  * Libav is distributed in the hope that it will be useful,

  12.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  13.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  14.  * Lesser General Public License for more details.

  15.  *

  16.  * You should have received a copy of the GNU Lesser General Public

  17.  * License along with Libav; if not, write to the Free Software

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

  19.  */

  20. 

  21. #define _SVID_SOURCE //needed for MAP_ANONYMOUS

  22. #include <inttypes.h>

  23. #include <string.h>

  24. #include <math.h>

  25. #include <stdio.h>

  26. #include "config.h"

  27. #include <assert.h>

  28. #if HAVE_SYS_MMAN_H

  29. #include <sys/mman.h>

  30. #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)

  31. #define MAP_ANONYMOUS MAP_ANON

  32. #endif

  33. #endif

  34. #if HAVE_VIRTUALALLOC

  35. #define WIN32_LEAN_AND_MEAN

  36. #include <windows.h>

  37. #endif

  38. #include "swscale.h"

  39. #include "swscale_internal.h"

  40. #include "rgb2rgb.h"

  41. #include "libavutil/intreadwrite.h"

  42. #include "libavutil/x86_cpu.h"

  43. #include "libavutil/cpu.h"

  44. #include "libavutil/avutil.h"

  45. #include "libavutil/bswap.h"

  46. #include "libavutil/mathematics.h"

  47. #include "libavutil/opt.h"

  48. #include "libavutil/pixdesc.h"

  49. 

  50. unsigned swscale_version(void)

  51. {

  52.     return LIBSWSCALE_VERSION_INT;

  53. }

  54. 

  55. const char *swscale_configuration(void)

  56. {

  57.     return LIBAV_CONFIGURATION;

  58. }

  59. 

  60. const char *swscale_license(void)

  61. {

  62. #define LICENSE_PREFIX "libswscale license: "

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

  64. }

  65. 

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

  67. 

  68. typedef struct FormatEntry {

  69.     int is_supported_in, is_supported_out;

  70. } FormatEntry;

  71. 

  72. const static FormatEntry format_entries[PIX_FMT_NB] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  134.     [PIX_FMT_YUV422P9BE]  = { 1 , 1 },

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

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

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

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

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

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

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

  142.     [PIX_FMT_GBRP]        = { 1 , 0 },

  143.     [PIX_FMT_GBRP9LE]     = { 1 , 0 },

  144.     [PIX_FMT_GBRP9BE]     = { 1 , 0 },

  145.     [PIX_FMT_GBRP10LE]    = { 1 , 0 },

  146.     [PIX_FMT_GBRP10BE]    = { 1 , 0 },

  147.     [PIX_FMT_GBRP16LE]    = { 1 , 0 },

  148.     [PIX_FMT_GBRP16BE]    = { 1 , 0 },

  149. };

  150. 

  151. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  152. {

  153.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  154.         format_entries[pix_fmt].is_supported_in : 0;

  155. }

  156. 

  157. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  158. {

  159.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  160.         format_entries[pix_fmt].is_supported_out : 0;

  161. }

  162. 

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

  164. 

  165. const char *sws_format_name(enum PixelFormat format)

  166. {

  167.     if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)

  168.         return av_pix_fmt_descriptors[format].name;

  169.     else

  170.         return "Unknown format";

  171. }

  172. 

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

  174. {

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

  176.     else           return getSplineCoeff(        0.0,

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

  178.                                                  c + 3.0*d,

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

  180.                                          dist-1.0);

  181. }

  182. 

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

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

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

  186. {

  187.     int i;

  188.     int filterSize;

  189.     int filter2Size;

  190.     int minFilterSize;

  191.     int64_t *filter=NULL;

  192.     int64_t *filter2=NULL;

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

  194.     int ret= -1;

  195. 

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

  197. 

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

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

  200. 

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

  202.         int i;

  203.         filterSize= 1;

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

  205. 

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

  207.             filter[i*filterSize]= fone;

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

  209.         }

  210. 

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

  212.         int i;

  213.         int xDstInSrc;

  214.         filterSize= 1;

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

  216. 

  217.         xDstInSrc= xInc/2 - 0x8000;

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

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

  220. 

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

  222.             filter[i]= fone;

  223.             xDstInSrc+= xInc;

  224.         }

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

  226.         int i;

  227.         int xDstInSrc;

  228.         filterSize= 2;

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

  230. 

  231.         xDstInSrc= xInc/2 - 0x8000;

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

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

  234.             int j;

  235. 

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

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

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

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

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

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

  242.                 xx++;

  243.             }

  244.             xDstInSrc+= xInc;

  245.         }

  246.     } else {

  247.         int xDstInSrc;

  248.         int sizeFactor;

  249. 

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

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

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

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

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

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

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

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

  258.         else {

  259.             sizeFactor= 0; //GCC warning killer

  260.             assert(0);

  261.         }

  262. 

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

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

  265. 

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

  267. 

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

  269. 

  270.         xDstInSrc= xInc - 0x10000;

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

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

  273.             int j;

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

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

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

  277.                 double floatd;

  278.                 int64_t coeff;

  279. 

  280.                 if (xInc > 1<<16)

  281.                     d= d*dstW/srcW;

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

  283. 

  284.                 if (flags & SWS_BICUBIC) {

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

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

  287. 

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

  289.                         coeff = 0.0;

  290.                     } else {

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

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

  293. 

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

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

  296.                         else

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

  298.                     }

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

  300.                 }

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

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

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

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

  305.                 }*/

  306.                 else if (flags & SWS_X) {

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

  308.                     double c;

  309. 

  310.                     if (floatd<1.0)

  311.                         c = cos(floatd*M_PI);

  312.                     else

  313.                         c=-1.0;

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

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

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

  317.                 } else if (flags & SWS_AREA) {

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

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

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

  321.                     else coeff=0.0;

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

  323.                 } else if (flags & SWS_GAUSS) {

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

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

  326.                 } else if (flags & SWS_SINC) {

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

  328.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  332.                 } else if (flags & SWS_BILINEAR) {

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

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

  335.                     coeff *= fone >> 30;

  336.                 } else if (flags & SWS_SPLINE) {

  337.                     double p=-2.196152422706632;

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

  339.                 } else {

  340.                     coeff= 0.0; //GCC warning killer

  341.                     assert(0);

  342.                 }

  343. 

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

  345.                 xx++;

  346.             }

  347.             xDstInSrc+= 2*xInc;

  348.         }

  349.     }

  350. 

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

  352.        av_free(filter);

  353.     */

  354.     assert(filterSize>0);

  355.     filter2Size= filterSize;

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

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

  358.     assert(filter2Size>0);

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

  360. 

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

  362.         int j, k;

  363. 

  364.         if(srcFilter) {

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

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

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

  368.             }

  369.         } else {

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

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

  372.         }

  373.         //FIXME dstFilter

  374. 

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

  376.     }

  377.     av_freep(&filter);

  378. 

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

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

  381.     minFilterSize= 0;

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

  383.         int min= filter2Size;

  384.         int j;

  385.         int64_t cutOff=0.0;

  386. 

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

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

  389.             int k;

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

  391. 

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

  393. 

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

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

  396. 

  397.             // move filter coefficients left

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

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

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

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

  402.         }

  403. 

  404.         cutOff=0;

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

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

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

  408. 

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

  410.             min--;

  411.         }

  412. 

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

  414.     }

  415. 

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

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

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

  419.         if (minFilterSize < 5)

  420.             filterAlign = 4;

  421. 

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

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

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

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

  426.         // decent-sized vector.

  427.         if (minFilterSize < 3)

  428.             filterAlign = 1;

  429.     }

  430. 

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

  432.         // special case for unscaled vertical filtering

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

  434.             filterAlign= 1;

  435.     }

  436. 

  437.     assert(minFilterSize > 0);

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

  439.     assert(filterSize > 0);

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

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

  442.         goto fail;

  443.     *outFilterSize= filterSize;

  444. 

  445.     if (flags&SWS_PRINT_INFO)

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

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

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

  449.         int j;

  450. 

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

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

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

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

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

  456.         }

  457.     }

  458. 

  459.     //FIXME try to align filterPos if possible

  460. 

  461.     //fix borders

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

  463.         int j;

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

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

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

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

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

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

  470.             }

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

  472.         }

  473. 

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

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

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

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

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

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

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

  481.             }

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

  483.         }

  484.     }

  485. 

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

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

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

  489. 

  490.     /* normalize & store in outFilter */

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

  492.         int j;

  493.         int64_t error=0;

  494.         int64_t sum=0;

  495. 

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

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

  498.         }

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

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

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

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

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

  504.             error= v - intV*sum;

  505.         }

  506.     }

  507. 

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

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

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

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

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

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

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

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

  516.     }

  517. 

  518.     ret=0;

  519. fail:

  520.     av_free(filter);

  521.     av_free(filter2);

  522.     return ret;

  523. }

  524. 

  525. #if HAVE_MMX2

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

  527. {

  528.     uint8_t *fragmentA;

  529.     x86_reg imm8OfPShufW1A;

  530.     x86_reg imm8OfPShufW2A;

  531.     x86_reg fragmentLengthA;

  532.     uint8_t *fragmentB;

  533.     x86_reg imm8OfPShufW1B;

  534.     x86_reg imm8OfPShufW2B;

  535.     x86_reg fragmentLengthB;

  536.     int fragmentPos;

  537. 

  538.     int xpos, i;

  539. 

  540.     // create an optimized horizontal scaling routine

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

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

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

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

  545.      * of fragmentA is used.

  546.      */

  547. 

  548.     //code fragment

  549. 

  550.     __asm__ volatile(

  551.         "jmp                         9f                 \n\t"

  552.     // Begin

  553.         "0:                                             \n\t"

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

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

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

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

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

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

  560.         "1:                                             \n\t"

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

  562.         "2:                                             \n\t"

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

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

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

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

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

  568. 

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

  570. 

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

  572.     // End

  573.         "9:                                             \n\t"

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

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

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

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

  578.         "dec                         %1                 \n\t"

  579.         "dec                         %2                 \n\t"

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

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

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

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

  584. 

  585. 

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

  587.         "=r" (fragmentLengthA)

  588.     );

  589. 

  590.     __asm__ volatile(

  591.         "jmp                         9f                 \n\t"

  592.     // Begin

  593.         "0:                                             \n\t"

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

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

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

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

  598.         "1:                                             \n\t"

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

  600.         "2:                                             \n\t"

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

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

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

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

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

  606. 

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

  608. 

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

  610.     // End

  611.         "9:                                             \n\t"

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

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

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

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

  616.         "dec                         %1                 \n\t"

  617.         "dec                         %2                 \n\t"

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

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

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

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

  622. 

  623. 

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

  625.         "=r" (fragmentLengthB)

  626.     );

  627. 

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

  629.     fragmentPos=0;

  630. 

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

  632.         int xx=xpos>>16;

  633. 

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

  635.             int a=0;

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

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

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

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

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

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

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

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

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

  645.             int shift=0;

  646. 

  647.             if (filterCode) {

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

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

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

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

  652.                 filterPos[i/2]= xx;

  653. 

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

  655. 

  656.                 filterCode[fragmentPos + imm8OfPShufW1]=

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

  658.                 filterCode[fragmentPos + imm8OfPShufW2]=

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

  660. 

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

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

  663. 

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

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

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

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

  668.                 }

  669.             }

  670. 

  671.             fragmentPos+= fragmentLength;

  672. 

  673.             if (filterCode)

  674.                 filterCode[fragmentPos]= RET;

  675.         }

  676.         xpos+=xInc;

  677.     }

  678.     if (filterCode)

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

  680. 

  681.     return fragmentPos + 1;

  682. }

  683. #endif /* HAVE_MMX2 */

  684. 

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

  686. {

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

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

  689. }

  690. 

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

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

  693.                              int brightness, int contrast, int saturation)

  694. {

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

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

  697. 

  698.     c->brightness= brightness;

  699.     c->contrast  = contrast;

  700.     c->saturation= saturation;

  701.     c->srcRange  = srcRange;

  702.     c->dstRange  = dstRange;

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

  704. 

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

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

  707. 

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

  709.     //FIXME factorize

  710. 

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

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

  713.     return 0;

  714. }

  715. 

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

  717.                              int *srcRange, int **table, int *dstRange,

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

  719. {

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

  721. 

  722.     *inv_table = c->srcColorspaceTable;

  723.     *table     = c->dstColorspaceTable;

  724.     *srcRange  = c->srcRange;

  725.     *dstRange  = c->dstRange;

  726.     *brightness= c->brightness;

  727.     *contrast  = c->contrast;

  728.     *saturation= c->saturation;

  729. 

  730.     return 0;

  731. }

  732. 

  733. static int handle_jpeg(enum PixelFormat *format)

  734. {

  735.     switch (*format) {

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

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

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

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

  740.     default:                                          return 0;

  741.     }

  742. }

  743. 

  744. SwsContext *sws_alloc_context(void)

  745. {

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

  747. 

  748.     c->av_class = &sws_context_class;

  749.     av_opt_set_defaults(c);

  750. 

  751.     return c;

  752. }

  753. 

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

  755. {

  756.     int i;

  757.     int usesVFilter, usesHFilter;

  758.     int unscaled;

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

  760.     int srcW= c->srcW;

  761.     int srcH= c->srcH;

  762.     int dstW= c->dstW;

  763.     int dstH= c->dstH;

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

  765.     int flags, cpu_flags;

  766.     enum PixelFormat srcFormat= c->srcFormat;

  767.     enum PixelFormat dstFormat= c->dstFormat;

  768. 

  769.     cpu_flags = av_get_cpu_flags();

  770.     flags     = c->flags;

  771.     emms_c();

  772.     if (!rgb15to16) sws_rgb2rgb_init();

  773. 

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

  775. 

  776.     if (!sws_isSupportedInput(srcFormat)) {

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

  778.         return AVERROR(EINVAL);

  779.     }

  780.     if (!sws_isSupportedOutput(dstFormat)) {

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

  782.         return AVERROR(EINVAL);

  783.     }

  784. 

  785.     i= flags & ( SWS_POINT

  786.                 |SWS_AREA

  787.                 |SWS_BILINEAR

  788.                 |SWS_FAST_BILINEAR

  789.                 |SWS_BICUBIC

  790.                 |SWS_X

  791.                 |SWS_GAUSS

  792.                 |SWS_LANCZOS

  793.                 |SWS_SINC

  794.                 |SWS_SPLINE

  795.                 |SWS_BICUBLIN);

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

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

  798.         return AVERROR(EINVAL);

  799.     }

  800.     /* sanity check */

  801.     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

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

  803.                srcW, srcH, dstW, dstH);

  804.         return AVERROR(EINVAL);

  805.     }

  806. 

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

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

  809. 

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

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

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

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

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

  815. 

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

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

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

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

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

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

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

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

  824. 

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

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

  827. 

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

  829.     if (flags & SWS_FULL_CHR_H_INT &&

  830.         dstFormat != PIX_FMT_RGBA &&

  831.         dstFormat != PIX_FMT_ARGB &&

  832.         dstFormat != PIX_FMT_BGRA &&

  833.         dstFormat != PIX_FMT_ABGR &&

  834.         dstFormat != PIX_FMT_RGB24 &&

  835.         dstFormat != PIX_FMT_BGR24) {

  836.         av_log(c, AV_LOG_ERROR,

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

  838.                sws_format_name(dstFormat));

  839.         flags &= ~SWS_FULL_CHR_H_INT;

  840.         c->flags = flags;

  841.     }

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

  843. 

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

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

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

  847. 

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

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

  850.       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8

  851.       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4

  852.       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE

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

  854.         c->chrSrcHSubSample=1;

  855. 

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

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

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

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

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

  861. 

  862.     /* unscaled special cases */

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

  864.         ff_get_unscaled_swscale(c);

  865. 

  866.         if (c->swScale) {

  867.             if (flags&SWS_PRINT_INFO)

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

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

  870.             return 0;

  871.         }

  872.     }

  873. 

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

  875.     if (c->srcBpc < 8)

  876.         c->srcBpc = 8;

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

  878.     if (c->dstBpc < 8)

  879.         c->dstBpc = 8;

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

  881.         dst_stride <<= 1;

  882.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,

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

  884.                      fail);

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

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

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

  888.             if (flags&SWS_PRINT_INFO)

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

  890.         }

  891.         if (usesHFilter) c->canMMX2BeUsed=0;

  892.     }

  893.     else

  894.         c->canMMX2BeUsed=0;

  895. 

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

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

  898. 

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

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

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

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

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

  904.     // first and last pixel

  905.     if (flags&SWS_FAST_BILINEAR) {

  906.         if (c->canMMX2BeUsed) {

  907.             c->lumXInc+= 20;

  908.             c->chrXInc+= 20;

  909.         }

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

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

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

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

  914.         }

  915.     }

  916. 

  917.     /* precalculate horizontal scaler filter coefficients */

  918.     {

  919. #if HAVE_MMX2

  920. // can't downscale !!!

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

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

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

  924. 

  925. #ifdef MAP_ANONYMOUS

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

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

  928. #elif HAVE_VIRTUALALLOC

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

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

  931. #else

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

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

  934. #endif

  935. 

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

  937.                 return AVERROR(ENOMEM);

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

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

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

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

  942. 

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

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

  945. 

  946. #ifdef MAP_ANONYMOUS

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

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

  949. #endif

  950.         } else

  951. #endif /* HAVE_MMX2 */

  952.         {

  953.             const int filterAlign=

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

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

  956.                 1;

  957. 

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

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

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

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

  962.                 goto fail;

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

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

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

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

  967.                 goto fail;

  968.         }

  969.     } // initialize horizontal stuff

  970. 

  971.     /* precalculate vertical scaler filter coefficients */

  972.     {

  973.         const int filterAlign=

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

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

  976.             1;

  977. 

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

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

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

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

  982.             goto fail;

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

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

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

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

  987.             goto fail;

  988. 

  989. #if HAVE_ALTIVEC

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

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

  992. 

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

  994.             int j;

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

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

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

  998.         }

  999. 

  1000.         for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {

  1001.             int j;

  1002.             short *p = (short *)&c->vCCoeffsBank[i];

  1003.             for (j=0;j<8;j++)

  1004.                 p[j] = c->vChrFilter[i];

  1005.         }

  1006. #endif

  1007.     }

  1008. 

  1009.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1010.     c->vLumBufSize= c->vLumFilterSize;

  1011.     c->vChrBufSize= c->vChrFilterSize;

  1012.     for (i=0; i<dstH; i++) {

  1013.         int chrI= i*c->chrDstH / dstH;

  1014.         int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,

  1015.                            ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));

  1016. 

  1017.         nextSlice>>= c->chrSrcVSubSample;

  1018.         nextSlice<<= c->chrSrcVSubSample;

  1019.         if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)

  1020.             c->vLumBufSize= nextSlice - c->vLumFilterPos[i];

  1021.         if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))

  1022.             c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];

  1023.     }

  1024. 

  1025.     // allocate pixbufs (we use dynamic allocation because otherwise we would need to

  1026.     // allocate several megabytes to handle all possible cases)

  1027.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1028.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1029.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1030.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1031.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1032.     //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)

  1033.     /* align at 16 bytes for AltiVec */

  1034.     for (i=0; i<c->vLumBufSize; i++) {

  1035.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1036.         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];

  1037.     }

  1038.     // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)

  1039.     c->uv_off_px   = dst_stride_px + 64 / (c->dstBpc &~ 7);

  1040.     c->uv_off_byte = dst_stride + 16;

  1041.     for (i=0; i<c->vChrBufSize; i++) {

  1042.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);

  1043.         c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];

  1044.         c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1045.     }

  1046.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1047.         for (i=0; i<c->vLumBufSize; i++) {

  1048.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1049.             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];

  1050.         }

  1051. 

  1052.     //try to avoid drawing green stuff between the right end and the stride end

  1053.     for (i=0; i<c->vChrBufSize; i++)

  1054.         memset(c->chrUPixBuf[i], 64, dst_stride*2+1);

  1055. 

  1056.     assert(c->chrDstH <= dstH);

  1057. 

  1058.     if (flags&SWS_PRINT_INFO) {

  1059.         if      (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1060.         else if (flags&SWS_BILINEAR)      av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1061.         else if (flags&SWS_BICUBIC)       av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1062.         else if (flags&SWS_X)             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1063.         else if (flags&SWS_POINT)         av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1064.         else if (flags&SWS_AREA)          av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1065.         else if (flags&SWS_BICUBLIN)      av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1066.         else if (flags&SWS_GAUSS)         av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1067.         else if (flags&SWS_SINC)          av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1068.         else if (flags&SWS_LANCZOS)       av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1069.         else if (flags&SWS_SPLINE)        av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1070.         else                              av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1071. 

  1072.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1073.                sws_format_name(srcFormat),

  1074. #ifdef DITHER1XBPP

  1075.                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||

  1076.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1077.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",

  1078. #else

  1079.                "",

  1080. #endif

  1081.                sws_format_name(dstFormat));

  1082. 

  1083.         if      (HAVE_MMX2     && cpu_flags & AV_CPU_FLAG_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");

  1084.         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1085.         else if (HAVE_MMX      && cpu_flags & AV_CPU_FLAG_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");

  1086.         else if (HAVE_ALTIVEC  && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1087.         else                                   av_log(c, AV_LOG_INFO, "using C\n");

  1088. 

  1089.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1090.         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1091.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1092.         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1093.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1094.     }

  1095. 

  1096.     c->swScale= ff_getSwsFunc(c);

  1097.     return 0;

  1098. fail: //FIXME replace things by appropriate error codes

  1099.     return -1;

  1100. }

  1101. 

  1102. #if FF_API_SWS_GETCONTEXT

  1103. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1104.                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1105.                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1106. {

  1107.     SwsContext *c;

  1108. 

  1109.     if(!(c=sws_alloc_context()))

  1110.         return NULL;

  1111. 

  1112.     c->flags= flags;

  1113.     c->srcW= srcW;

  1114.     c->srcH= srcH;

  1115.     c->dstW= dstW;

  1116.     c->dstH= dstH;

  1117.     c->srcRange = handle_jpeg(&srcFormat);

  1118.     c->dstRange = handle_jpeg(&dstFormat);

  1119.     c->srcFormat= srcFormat;

  1120.     c->dstFormat= dstFormat;

  1121. 

  1122.     if (param) {

  1123.         c->param[0] = param[0];

  1124.         c->param[1] = param[1];

  1125.     }

  1126.     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);

  1127. 

  1128.     if(sws_init_context(c, srcFilter, dstFilter) < 0){

  1129.         sws_freeContext(c);

  1130.         return NULL;

  1131.     }

  1132. 

  1133.     return c;

  1134. }

  1135. #endif

  1136. 

  1137. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1138.                                 float lumaSharpen, float chromaSharpen,

  1139.                                 float chromaHShift, float chromaVShift,

  1140.                                 int verbose)

  1141. {

  1142.     SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  1143.     if (!filter)

  1144.         return NULL;

  1145. 

  1146.     if (lumaGBlur!=0.0) {

  1147.         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  1148.         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  1149.     } else {

  1150.         filter->lumH= sws_getIdentityVec();

  1151.         filter->lumV= sws_getIdentityVec();

  1152.     }

  1153. 

  1154.     if (chromaGBlur!=0.0) {

  1155.         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  1156.         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  1157.     } else {

  1158.         filter->chrH= sws_getIdentityVec();

  1159.         filter->chrV= sws_getIdentityVec();

  1160.     }

  1161. 

  1162.     if (chromaSharpen!=0.0) {

  1163.         SwsVector *id= sws_getIdentityVec();

  1164.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1165.         sws_scaleVec(filter->chrV, -chromaSharpen);

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

  1167.         sws_addVec(filter->chrV, id);

  1168.         sws_freeVec(id);

  1169.     }

  1170. 

  1171.     if (lumaSharpen!=0.0) {

  1172.         SwsVector *id= sws_getIdentityVec();

  1173.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1174.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1175.         sws_addVec(filter->lumH, id);

  1176.         sws_addVec(filter->lumV, id);

  1177.         sws_freeVec(id);

  1178.     }

  1179. 

  1180.     if (chromaHShift != 0.0)

  1181.         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  1182. 

  1183.     if (chromaVShift != 0.0)

  1184.         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  1185. 

  1186.     sws_normalizeVec(filter->chrH, 1.0);

  1187.     sws_normalizeVec(filter->chrV, 1.0);

  1188.     sws_normalizeVec(filter->lumH, 1.0);

  1189.     sws_normalizeVec(filter->lumV, 1.0);

  1190. 

  1191.     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1192.     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1193. 

  1194.     return filter;

  1195. }

  1196. 

  1197. SwsVector *sws_allocVec(int length)

  1198. {

  1199.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1200.     if (!vec)

  1201.         return NULL;

  1202.     vec->length = length;

  1203.     vec->coeff  = av_malloc(sizeof(double) * length);

  1204.     if (!vec->coeff)

  1205.         av_freep(&vec);

  1206.     return vec;

  1207. }

  1208. 

  1209. SwsVector *sws_getGaussianVec(double variance, double quality)

  1210. {

  1211.     const int length= (int)(variance*quality + 0.5) | 1;

  1212.     int i;

  1213.     double middle= (length-1)*0.5;

  1214.     SwsVector *vec= sws_allocVec(length);

  1215. 

  1216.     if (!vec)

  1217.         return NULL;

  1218. 

  1219.     for (i=0; i<length; i++) {

  1220.         double dist= i-middle;

  1221.         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);

  1222.     }

  1223. 

  1224.     sws_normalizeVec(vec, 1.0);

  1225. 

  1226.     return vec;

  1227. }

  1228. 

  1229. SwsVector *sws_getConstVec(double c, int length)

  1230. {

  1231.     int i;

  1232.     SwsVector *vec= sws_allocVec(length);

  1233. 

  1234.     if (!vec)

  1235.         return NULL;

  1236. 

  1237.     for (i=0; i<length; i++)

  1238.         vec->coeff[i]= c;

  1239. 

  1240.     return vec;

  1241. }

  1242. 

  1243. SwsVector *sws_getIdentityVec(void)

  1244. {

  1245.     return sws_getConstVec(1.0, 1);

  1246. }

  1247. 

  1248. static double sws_dcVec(SwsVector *a)

  1249. {

  1250.     int i;

  1251.     double sum=0;

  1252. 

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

  1254.         sum+= a->coeff[i];

  1255. 

  1256.     return sum;

  1257. }

  1258. 

  1259. void sws_scaleVec(SwsVector *a, double scalar)

  1260. {

  1261.     int i;

  1262. 

  1263.     for (i=0; i<a->length; i++)

  1264.         a->coeff[i]*= scalar;

  1265. }

  1266. 

  1267. void sws_normalizeVec(SwsVector *a, double height)

  1268. {

  1269.     sws_scaleVec(a, height/sws_dcVec(a));

  1270. }

  1271. 

  1272. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1273. {

  1274.     int length= a->length + b->length - 1;

  1275.     int i, j;

  1276.     SwsVector *vec= sws_getConstVec(0.0, length);

  1277. 

  1278.     if (!vec)

  1279.         return NULL;

  1280. 

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

  1282.         for (j=0; j<b->length; j++) {

  1283.             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1284.         }

  1285.     }

  1286. 

  1287.     return vec;

  1288. }

  1289. 

  1290. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1291. {

  1292.     int length= FFMAX(a->length, b->length);

  1293.     int i;

  1294.     SwsVector *vec= sws_getConstVec(0.0, length);

  1295. 

  1296.     if (!vec)

  1297.         return NULL;

  1298. 

  1299.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1300.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];

  1301. 

  1302.     return vec;

  1303. }

  1304. 

  1305. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1306. {

  1307.     int length= FFMAX(a->length, b->length);

  1308.     int i;

  1309.     SwsVector *vec= sws_getConstVec(0.0, length);

  1310. 

  1311.     if (!vec)

  1312.         return NULL;

  1313. 

  1314.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1315.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1316. 

  1317.     return vec;

  1318. }

  1319. 

  1320. /* shift left / or right if "shift" is negative */

  1321. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1322. {

  1323.     int length= a->length + FFABS(shift)*2;

  1324.     int i;

  1325.     SwsVector *vec= sws_getConstVec(0.0, length);

  1326. 

  1327.     if (!vec)

  1328.         return NULL;

  1329. 

  1330.     for (i=0; i<a->length; i++) {

  1331.         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1332.     }

  1333. 

  1334.     return vec;

  1335. }

  1336. 

  1337. void sws_shiftVec(SwsVector *a, int shift)

  1338. {

  1339.     SwsVector *shifted= sws_getShiftedVec(a, shift);

  1340.     av_free(a->coeff);

  1341.     a->coeff= shifted->coeff;

  1342.     a->length= shifted->length;

  1343.     av_free(shifted);

  1344. }

  1345. 

  1346. void sws_addVec(SwsVector *a, SwsVector *b)

  1347. {

  1348.     SwsVector *sum= sws_sumVec(a, b);

  1349.     av_free(a->coeff);

  1350.     a->coeff= sum->coeff;

  1351.     a->length= sum->length;

  1352.     av_free(sum);

  1353. }

  1354. 

  1355. void sws_subVec(SwsVector *a, SwsVector *b)

  1356. {

  1357.     SwsVector *diff= sws_diffVec(a, b);

  1358.     av_free(a->coeff);

  1359.     a->coeff= diff->coeff;

  1360.     a->length= diff->length;

  1361.     av_free(diff);

  1362. }

  1363. 

  1364. void sws_convVec(SwsVector *a, SwsVector *b)

  1365. {

  1366.     SwsVector *conv= sws_getConvVec(a, b);

  1367.     av_free(a->coeff);

  1368.     a->coeff= conv->coeff;

  1369.     a->length= conv->length;

  1370.     av_free(conv);

  1371. }

  1372. 

  1373. SwsVector *sws_cloneVec(SwsVector *a)

  1374. {

  1375.     int i;

  1376.     SwsVector *vec= sws_allocVec(a->length);

  1377. 

  1378.     if (!vec)

  1379.         return NULL;

  1380. 

  1381.     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];

  1382. 

  1383.     return vec;

  1384. }

  1385. 

  1386. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1387. {

  1388.     int i;

  1389.     double max=0;

  1390.     double min=0;

  1391.     double range;

  1392. 

  1393.     for (i=0; i<a->length; i++)

  1394.         if (a->coeff[i]>max) max= a->coeff[i];

  1395. 

  1396.     for (i=0; i<a->length; i++)

  1397.         if (a->coeff[i]<min) min= a->coeff[i];

  1398. 

  1399.     range= max - min;

  1400. 

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

  1402.         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1403.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1404.         for (;x>0; x--) av_log(log_ctx, log_level, " ");

  1405.         av_log(log_ctx, log_level, "|\n");

  1406.     }

  1407. }

  1408. 

  1409. void sws_freeVec(SwsVector *a)

  1410. {

  1411.     if (!a) return;

  1412.     av_freep(&a->coeff);

  1413.     a->length=0;

  1414.     av_free(a);

  1415. }

  1416. 

  1417. void sws_freeFilter(SwsFilter *filter)

  1418. {

  1419.     if (!filter) return;

  1420. 

  1421.     if (filter->lumH) sws_freeVec(filter->lumH);

  1422.     if (filter->lumV) sws_freeVec(filter->lumV);

  1423.     if (filter->chrH) sws_freeVec(filter->chrH);

  1424.     if (filter->chrV) sws_freeVec(filter->chrV);

  1425.     av_free(filter);

  1426. }

  1427. 

  1428. void sws_freeContext(SwsContext *c)

  1429. {

  1430.     int i;

  1431.     if (!c) return;

  1432. 

  1433.     if (c->lumPixBuf) {

  1434.         for (i=0; i<c->vLumBufSize; i++)

  1435.             av_freep(&c->lumPixBuf[i]);

  1436.         av_freep(&c->lumPixBuf);

  1437.     }

  1438. 

  1439.     if (c->chrUPixBuf) {

  1440.         for (i=0; i<c->vChrBufSize; i++)

  1441.             av_freep(&c->chrUPixBuf[i]);

  1442.         av_freep(&c->chrUPixBuf);

  1443.         av_freep(&c->chrVPixBuf);

  1444.     }

  1445. 

  1446.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1447.         for (i=0; i<c->vLumBufSize; i++)

  1448.             av_freep(&c->alpPixBuf[i]);

  1449.         av_freep(&c->alpPixBuf);

  1450.     }

  1451. 

  1452.     av_freep(&c->vLumFilter);

  1453.     av_freep(&c->vChrFilter);

  1454.     av_freep(&c->hLumFilter);

  1455.     av_freep(&c->hChrFilter);

  1456. #if HAVE_ALTIVEC

  1457.     av_freep(&c->vYCoeffsBank);

  1458.     av_freep(&c->vCCoeffsBank);

  1459. #endif

  1460. 

  1461.     av_freep(&c->vLumFilterPos);

  1462.     av_freep(&c->vChrFilterPos);

  1463.     av_freep(&c->hLumFilterPos);

  1464.     av_freep(&c->hChrFilterPos);

  1465. 

  1466. #if HAVE_MMX

  1467. #ifdef MAP_ANONYMOUS

  1468.     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1469.     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1470. #elif HAVE_VIRTUALALLOC

  1471.     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1472.     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1473. #else

  1474.     av_free(c->lumMmx2FilterCode);

  1475.     av_free(c->chrMmx2FilterCode);

  1476. #endif

  1477.     c->lumMmx2FilterCode=NULL;

  1478.     c->chrMmx2FilterCode=NULL;

  1479. #endif /* HAVE_MMX */

  1480. 

  1481.     av_freep(&c->yuvTable);

  1482.     av_free(c->formatConvBuffer);

  1483. 

  1484.     av_free(c);

  1485. }

  1486. 

  1487. struct SwsContext *sws_getCachedContext(struct SwsContext *context,

  1488.                                         int srcW, int srcH, enum PixelFormat srcFormat,

  1489.                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1490.                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1491. {

  1492.     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};

  1493. 

  1494.     if (!param)

  1495.         param = default_param;

  1496. 

  1497.     if (context &&

  1498.         (context->srcW      != srcW      ||

  1499.          context->srcH      != srcH      ||

  1500.          context->srcFormat != srcFormat ||

  1501.          context->dstW      != dstW      ||

  1502.          context->dstH      != dstH      ||

  1503.          context->dstFormat != dstFormat ||

  1504.          context->flags     != flags     ||

  1505.          context->param[0]  != param[0]  ||

  1506.          context->param[1]  != param[1])) {

  1507.         sws_freeContext(context);

  1508.         context = NULL;

  1509.     }

  1510. 

  1511.     if (!context) {

  1512.         if (!(context = sws_alloc_context()))

  1513.             return NULL;

  1514.         context->srcW      = srcW;

  1515.         context->srcH      = srcH;

  1516.         context->srcRange  = handle_jpeg(&srcFormat);

  1517.         context->srcFormat = srcFormat;

  1518.         context->dstW      = dstW;

  1519.         context->dstH      = dstH;

  1520.         context->dstRange  = handle_jpeg(&dstFormat);

  1521.         context->dstFormat = dstFormat;

  1522.         context->flags     = flags;

  1523.         context->param[0]  = param[0];

  1524.         context->param[1]  = param[1];

  1525.         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);

  1526.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1527.             sws_freeContext(context);

  1528.             return NULL;

  1529.         }

  1530.     }

  1531.     return context;

  1532. }