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.
32729 Commits
32 Branches
593MB
Tree: a8a9844a3c
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'a8a9844a3c'
${ noResults }
FFmpeg/libswscale/utils.c

1519 lines
54KB
Raw Blame History 

  1. /*

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

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. #define _SVID_SOURCE //needed for MAP_ANONYMOUS

  22. #define _DARWIN_C_SOURCE // needed for MAP_ANON

  23. #include <inttypes.h>

  24. #include <string.h>

  25. #include <math.h>

  26. #include <stdio.h>

  27. #include "config.h"

  28. #include <assert.h>

  29. #if HAVE_SYS_MMAN_H

  30. #include <sys/mman.h>

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

  32. #define MAP_ANONYMOUS MAP_ANON

  33. #endif

  34. #endif

  35. #if HAVE_VIRTUALALLOC

  36. #define WIN32_LEAN_AND_MEAN

  37. #include <windows.h>

  38. #endif

  39. #include "swscale.h"

  40. #include "swscale_internal.h"

  41. #include "rgb2rgb.h"

  42. #include "libavutil/intreadwrite.h"

  43. #include "libavutil/x86_cpu.h"

  44. #include "libavutil/cpu.h"

  45. #include "libavutil/avutil.h"

  46. #include "libavutil/bswap.h"

  47. #include "libavutil/opt.h"

  48. #include "libavutil/pixdesc.h"

  49. #include "libavutil/avassert.h"

  50. 

  51. unsigned swscale_version(void)

  52. {

  53.     return LIBSWSCALE_VERSION_INT;

  54. }

  55. 

  56. const char *swscale_configuration(void)

  57. {

  58.     return FFMPEG_CONFIGURATION;

  59. }

  60. 

  61. const char *swscale_license(void)

  62. {

  63. #define LICENSE_PREFIX "libswscale license: "

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

  65. }

  66. 

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

  68. 

  69. typedef struct FormatEntry {

  70.     int is_supported_in, is_supported_out;

  71. } FormatEntry;

  72. 

  73. const static FormatEntry format_entries[PIX_FMT_NB] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  137.     [PIX_FMT_YUV444P9BE]  = { 1 , 0 },

  138.     [PIX_FMT_YUV444P9LE]  = { 1 , 0 },

  139.     [PIX_FMT_YUV444P10BE] = { 1 , 0 },

  140.     [PIX_FMT_YUV444P10LE] = { 1 , 0 },

  141. };

  142. 

  143. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  144. {

  145.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  146.         format_entries[pix_fmt].is_supported_in : 0;

  147. }

  148. 

  149. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  150. {

  151.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  152.         format_entries[pix_fmt].is_supported_out : 0;

  153. }

  154. 

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

  156. 

  157. #if FF_API_SWS_FORMAT_NAME

  158. const char *sws_format_name(enum PixelFormat format)

  159. {

  160.     return av_get_pix_fmt_name(format);

  161. }

  162. #endif

  163. 

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

  165. {

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

  167.     else           return getSplineCoeff(        0.0,

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

  169.                                                  c + 3.0*d,

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

  171.                                          dist-1.0);

  172. }

  173. 

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

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

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

  177. {

  178.     int i;

  179.     int filterSize;

  180.     int filter2Size;

  181.     int minFilterSize;

  182.     int64_t *filter=NULL;

  183.     int64_t *filter2=NULL;

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

  185.     int ret= -1;

  186. 

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

  188. 

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

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

  191. 

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

  193.         int i;

  194.         filterSize= 1;

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

  196. 

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

  198.             filter[i*filterSize]= fone;

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

  200.         }

  201. 

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

  203.         int i;

  204.         int xDstInSrc;

  205.         filterSize= 1;

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

  207. 

  208.         xDstInSrc= xInc/2 - 0x8000;

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

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

  211. 

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

  213.             filter[i]= fone;

  214.             xDstInSrc+= xInc;

  215.         }

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

  217.         int i;

  218.         int xDstInSrc;

  219.         filterSize= 2;

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

  221. 

  222.         xDstInSrc= xInc/2 - 0x8000;

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

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

  225.             int j;

  226. 

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

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

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

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

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

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

  233.                 xx++;

  234.             }

  235.             xDstInSrc+= xInc;

  236.         }

  237.     } else {

  238.         int xDstInSrc;

  239.         int sizeFactor;

  240. 

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

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

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

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

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

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

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

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

  249.         else {

  250.             sizeFactor= 0; //GCC warning killer

  251.             assert(0);

  252.         }

  253. 

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

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

  256. 

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

  258. 

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

  260. 

  261.         xDstInSrc= xInc - 0x10000;

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

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

  264.             int j;

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

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

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

  268.                 double floatd;

  269.                 int64_t coeff;

  270. 

  271.                 if (xInc > 1<<16)

  272.                     d= d*dstW/srcW;

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

  274. 

  275.                 if (flags & SWS_BICUBIC) {

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

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

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

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

  280. 

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

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

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

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

  285.                     else

  286.                         coeff=0.0;

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

  288.                 }

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

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

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

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

  293.                 }*/

  294.                 else if (flags & SWS_X) {

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

  296.                     double c;

  297. 

  298.                     if (floatd<1.0)

  299.                         c = cos(floatd*M_PI);

  300.                     else

  301.                         c=-1.0;

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

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

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

  305.                 } else if (flags & SWS_AREA) {

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

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

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

  309.                     else coeff=0.0;

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

  311.                 } else if (flags & SWS_GAUSS) {

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

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

  314.                 } else if (flags & SWS_SINC) {

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

  316.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  320.                 } else if (flags & SWS_BILINEAR) {

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

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

  323.                     coeff *= fone >> 30;

  324.                 } else if (flags & SWS_SPLINE) {

  325.                     double p=-2.196152422706632;

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

  327.                 } else {

  328.                     coeff= 0.0; //GCC warning killer

  329.                     assert(0);

  330.                 }

  331. 

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

  333.                 xx++;

  334.             }

  335.             xDstInSrc+= 2*xInc;

  336.         }

  337.     }

  338. 

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

  340.        av_free(filter);

  341.     */

  342.     assert(filterSize>0);

  343.     filter2Size= filterSize;

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

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

  346.     assert(filter2Size>0);

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

  348. 

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

  350.         int j, k;

  351. 

  352.         if(srcFilter) {

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

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

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

  356.             }

  357.         } else {

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

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

  360.         }

  361.         //FIXME dstFilter

  362. 

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

  364.     }

  365.     av_freep(&filter);

  366. 

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

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

  369.     minFilterSize= 0;

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

  371.         int min= filter2Size;

  372.         int j;

  373.         int64_t cutOff=0.0;

  374. 

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

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

  377.             int k;

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

  379. 

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

  381. 

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

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

  384. 

  385.             // move filter coefficients left

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

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

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

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

  390.         }

  391. 

  392.         cutOff=0;

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

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

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

  396. 

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

  398.             min--;

  399.         }

  400. 

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

  402.     }

  403. 

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

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

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

  407.         if (minFilterSize < 5)

  408.             filterAlign = 4;

  409. 

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

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

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

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

  414.         // decent-sized vector.

  415.         if (minFilterSize < 3)

  416.             filterAlign = 1;

  417.     }

  418. 

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

  420.         // special case for unscaled vertical filtering

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

  422.             filterAlign= 1;

  423.     }

  424. 

  425.     assert(minFilterSize > 0);

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

  427.     assert(filterSize > 0);

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

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

  430.         goto fail;

  431.     *outFilterSize= filterSize;

  432. 

  433.     if (flags&SWS_PRINT_INFO)

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

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

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

  437.         int j;

  438. 

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

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

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

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

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

  444.         }

  445.     }

  446. 

  447.     //FIXME try to align filterPos if possible

  448. 

  449.     //fix borders

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

  451.         int j;

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

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

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

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

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

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

  458.             }

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

  460.         }

  461. 

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

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

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

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

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

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

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

  469.             }

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

  471.         }

  472.     }

  473. 

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

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

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

  477. 

  478.     /* normalize & store in outFilter */

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

  480.         int j;

  481.         int64_t error=0;

  482.         int64_t sum=0;

  483. 

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

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

  486.         }

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

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

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

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

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

  492.             error= v - intV*sum;

  493.         }

  494.     }

  495. 

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

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

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

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

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

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

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

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

  504.     }

  505. 

  506.     ret=0;

  507. fail:

  508.     av_free(filter);

  509.     av_free(filter2);

  510.     return ret;

  511. }

  512. 

  513. #if HAVE_MMX2

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

  515. {

  516.     uint8_t *fragmentA;

  517.     x86_reg imm8OfPShufW1A;

  518.     x86_reg imm8OfPShufW2A;

  519.     x86_reg fragmentLengthA;

  520.     uint8_t *fragmentB;

  521.     x86_reg imm8OfPShufW1B;

  522.     x86_reg imm8OfPShufW2B;

  523.     x86_reg fragmentLengthB;

  524.     int fragmentPos;

  525. 

  526.     int xpos, i;

  527. 

  528.     // create an optimized horizontal scaling routine

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

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

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

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

  533.      * of fragmentA is used.

  534.      */

  535. 

  536.     //code fragment

  537. 

  538.     __asm__ volatile(

  539.         "jmp                         9f                 \n\t"

  540.     // Begin

  541.         "0:                                             \n\t"

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

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

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

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

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

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

  548.         "1:                                             \n\t"

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

  550.         "2:                                             \n\t"

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

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

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

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

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

  556. 

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

  558. 

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

  560.     // End

  561.         "9:                                             \n\t"

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

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

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

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

  566.         "dec                         %1                 \n\t"

  567.         "dec                         %2                 \n\t"

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

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

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

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

  572. 

  573. 

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

  575.         "=r" (fragmentLengthA)

  576.     );

  577. 

  578.     __asm__ volatile(

  579.         "jmp                         9f                 \n\t"

  580.     // Begin

  581.         "0:                                             \n\t"

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

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

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

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

  586.         "1:                                             \n\t"

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

  588.         "2:                                             \n\t"

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

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

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

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

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

  594. 

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

  596. 

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

  598.     // End

  599.         "9:                                             \n\t"

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

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

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

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

  604.         "dec                         %1                 \n\t"

  605.         "dec                         %2                 \n\t"

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

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

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

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

  610. 

  611. 

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

  613.         "=r" (fragmentLengthB)

  614.     );

  615. 

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

  617.     fragmentPos=0;

  618. 

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

  620.         int xx=xpos>>16;

  621. 

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

  623.             int a=0;

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

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

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

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

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

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

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

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

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

  633.             int shift=0;

  634. 

  635.             if (filterCode) {

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

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

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

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

  640.                 filterPos[i/2]= xx;

  641. 

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

  643. 

  644.                 filterCode[fragmentPos + imm8OfPShufW1]=

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

  646.                 filterCode[fragmentPos + imm8OfPShufW2]=

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

  648. 

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

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

  651. 

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

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

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

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

  656.                 }

  657.             }

  658. 

  659.             fragmentPos+= fragmentLength;

  660. 

  661.             if (filterCode)

  662.                 filterCode[fragmentPos]= RET;

  663.         }

  664.         xpos+=xInc;

  665.     }

  666.     if (filterCode)

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

  668. 

  669.     return fragmentPos + 1;

  670. }

  671. #endif /* HAVE_MMX2 */

  672. 

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

  674. {

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

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

  677. }

  678. 

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

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

  681.                              int brightness, int contrast, int saturation)

  682. {

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

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

  685. 

  686.     c->brightness= brightness;

  687.     c->contrast  = contrast;

  688.     c->saturation= saturation;

  689.     c->srcRange  = srcRange;

  690.     c->dstRange  = dstRange;

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

  692. 

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

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

  695. 

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

  697.     //FIXME factorize

  698. 

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

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

  701.     return 0;

  702. }

  703. 

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

  705.                              int *srcRange, int **table, int *dstRange,

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

  707. {

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

  709. 

  710.     *inv_table = c->srcColorspaceTable;

  711.     *table     = c->dstColorspaceTable;

  712.     *srcRange  = c->srcRange;

  713.     *dstRange  = c->dstRange;

  714.     *brightness= c->brightness;

  715.     *contrast  = c->contrast;

  716.     *saturation= c->saturation;

  717. 

  718.     return 0;

  719. }

  720. 

  721. static int handle_jpeg(enum PixelFormat *format)

  722. {

  723.     switch (*format) {

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

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

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

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

  728.     default:                                          return 0;

  729.     }

  730. }

  731. 

  732. SwsContext *sws_alloc_context(void)

  733. {

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

  735. 

  736.     c->av_class = &sws_context_class;

  737.     av_opt_set_defaults(c);

  738. 

  739.     return c;

  740. }

  741. 

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

  743. {

  744.     int i, j;

  745.     int usesVFilter, usesHFilter;

  746.     int unscaled;

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

  748.     int srcW= c->srcW;

  749.     int srcH= c->srcH;

  750.     int dstW= c->dstW;

  751.     int dstH= c->dstH;

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

  753.     int flags, cpu_flags;

  754.     enum PixelFormat srcFormat= c->srcFormat;

  755.     enum PixelFormat dstFormat= c->dstFormat;

  756. 

  757.     cpu_flags = av_get_cpu_flags();

  758.     flags     = c->flags;

  759.     emms_c();

  760.     if (!rgb15to16) sws_rgb2rgb_init();

  761. 

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

  763. 

  764.     if (!sws_isSupportedInput(srcFormat)) {

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

  766.         return AVERROR(EINVAL);

  767.     }

  768.     if (!sws_isSupportedOutput(dstFormat)) {

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

  770.         return AVERROR(EINVAL);

  771.     }

  772. 

  773.     i= flags & ( SWS_POINT

  774.                 |SWS_AREA

  775.                 |SWS_BILINEAR

  776.                 |SWS_FAST_BILINEAR

  777.                 |SWS_BICUBIC

  778.                 |SWS_X

  779.                 |SWS_GAUSS

  780.                 |SWS_LANCZOS

  781.                 |SWS_SINC

  782.                 |SWS_SPLINE

  783.                 |SWS_BICUBLIN);

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

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

  786.         return AVERROR(EINVAL);

  787.     }

  788.     /* sanity check */

  789.     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

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

  791.                srcW, srcH, dstW, dstH);

  792.         return AVERROR(EINVAL);

  793.     }

  794. 

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

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

  797. 

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

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

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

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

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

  803. 

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

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

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

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

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

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

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

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

  812. 

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

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

  815. 

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

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

  818. 

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

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

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

  822. 

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

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

  825.       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8

  826.       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4

  827.       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE

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

  829.         c->chrSrcHSubSample=1;

  830. 

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

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

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

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

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

  836. 

  837.     /* unscaled special cases */

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

  839.         ff_get_unscaled_swscale(c);

  840. 

  841.         if (c->swScale) {

  842.             if (flags&SWS_PRINT_INFO)

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

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

  845.             return 0;

  846.         }

  847.     }

  848. 

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

  850.     if (c->srcBpc < 8)

  851.         c->srcBpc = 8;

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

  853.     if (c->dstBpc < 8)

  854.         c->dstBpc = 8;

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

  856.         c->srcBpc = 16;

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

  858.         dst_stride <<= 1;

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

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

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

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

  863.             if (flags&SWS_PRINT_INFO)

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

  865.         }

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

  867.     }

  868.     else

  869.         c->canMMX2BeUsed=0;

  870. 

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

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

  873. 

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

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

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

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

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

  879.     // first and last pixel

  880.     if (flags&SWS_FAST_BILINEAR) {

  881.         if (c->canMMX2BeUsed) {

  882.             c->lumXInc+= 20;

  883.             c->chrXInc+= 20;

  884.         }

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

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

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

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

  889.         }

  890.     }

  891. 

  892.     /* precalculate horizontal scaler filter coefficients */

  893.     {

  894. #if HAVE_MMX2

  895. // can't downscale !!!

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

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

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

  899. 

  900. #ifdef MAP_ANONYMOUS

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

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

  903. #elif HAVE_VIRTUALALLOC

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

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

  906. #else

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

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

  909. #endif

  910. 

  911. #ifdef MAP_ANONYMOUS

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

  913. #else

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

  915. #endif

  916.                 return AVERROR(ENOMEM);

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

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

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

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

  921. 

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

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

  924. 

  925. #ifdef MAP_ANONYMOUS

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

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

  928. #endif

  929.         } else

  930. #endif /* HAVE_MMX2 */

  931.         {

  932.             const int filterAlign=

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

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

  935.                 1;

  936. 

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

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

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

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

  941.                 goto fail;

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

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

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

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

  946.                 goto fail;

  947.         }

  948.     } // initialize horizontal stuff

  949. 

  950.     /* precalculate vertical scaler filter coefficients */

  951.     {

  952.         const int filterAlign=

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

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

  955.             1;

  956. 

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

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

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

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

  961.             goto fail;

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

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

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

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

  966.             goto fail;

  967. 

  968. #if HAVE_ALTIVEC

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

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

  971. 

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

  973.             int j;

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

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

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

  977.         }

  978. 

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

  980.             int j;

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

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

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

  984.         }

  985. #endif

  986.     }

  987. 

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

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

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

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

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

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

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

  995. 

  996.         nextSlice>>= c->chrSrcVSubSample;

  997.         nextSlice<<= c->chrSrcVSubSample;

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

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

  1000.         if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))

  1001.             c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];

  1002.     }

  1003. 

  1004.     // allocate pixbufs (we use dynamic allocation because otherwise we would need to

  1005.     // allocate several megabytes to handle all possible cases)

  1006.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1007.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1008.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1009.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1010.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1011.     //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)

  1012.     /* align at 16 bytes for AltiVec */

  1013.     for (i=0; i<c->vLumBufSize; i++) {

  1014.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1015.         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];

  1016.     }

  1017.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1018.     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);

  1019.     c->uv_offx2 = dst_stride + 16;

  1020.     for (i=0; i<c->vChrBufSize; i++) {

  1021.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);

  1022.         c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];

  1023.         c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1024.     }

  1025.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1026.         for (i=0; i<c->vLumBufSize; i++) {

  1027.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1028.             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];

  1029.         }

  1030. 

  1031.     //try to avoid drawing green stuff between the right end and the stride end

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

  1033.         if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){

  1034.             av_assert0(c->dstBpc > 10);

  1035.             for(j=0; j<dst_stride/2+1; j++)

  1036.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1037.         } else

  1038.             for(j=0; j<dst_stride+1; j++)

  1039.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1040. 

  1041.     assert(c->chrDstH <= dstH);

  1042. 

  1043.     if (flags&SWS_PRINT_INFO) {

  1044.         if      (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1045.         else if (flags&SWS_BILINEAR)      av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1046.         else if (flags&SWS_BICUBIC)       av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1047.         else if (flags&SWS_X)             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1048.         else if (flags&SWS_POINT)         av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1049.         else if (flags&SWS_AREA)          av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1050.         else if (flags&SWS_BICUBLIN)      av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1051.         else if (flags&SWS_GAUSS)         av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1052.         else if (flags&SWS_SINC)          av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1053.         else if (flags&SWS_LANCZOS)       av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1054.         else if (flags&SWS_SPLINE)        av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1055.         else                              av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1056. 

  1057.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1058.                av_get_pix_fmt_name(srcFormat),

  1059. #ifdef DITHER1XBPP

  1060.                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||

  1061.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1062.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",

  1063. #else

  1064.                "",

  1065. #endif

  1066.                av_get_pix_fmt_name(dstFormat));

  1067. 

  1068.         if      (HAVE_MMX2     && cpu_flags & AV_CPU_FLAG_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");

  1069.         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1070.         else if (HAVE_MMX      && cpu_flags & AV_CPU_FLAG_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");

  1071.         else if (HAVE_ALTIVEC  && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1072.         else                                   av_log(c, AV_LOG_INFO, "using C\n");

  1073. 

  1074.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1075.         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1076.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1077.         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1078.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1079.     }

  1080. 

  1081.     c->swScale= ff_getSwsFunc(c);

  1082.     return 0;

  1083. fail: //FIXME replace things by appropriate error codes

  1084.     return -1;

  1085. }

  1086. 

  1087. #if FF_API_SWS_GETCONTEXT

  1088. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1089.                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1090.                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1091. {

  1092.     SwsContext *c;

  1093. 

  1094.     if(!(c=sws_alloc_context()))

  1095.         return NULL;

  1096. 

  1097.     c->flags= flags;

  1098.     c->srcW= srcW;

  1099.     c->srcH= srcH;

  1100.     c->dstW= dstW;

  1101.     c->dstH= dstH;

  1102.     c->srcRange = handle_jpeg(&srcFormat);

  1103.     c->dstRange = handle_jpeg(&dstFormat);

  1104.     c->srcFormat= srcFormat;

  1105.     c->dstFormat= dstFormat;

  1106. 

  1107.     if (param) {

  1108.         c->param[0] = param[0];

  1109.         c->param[1] = param[1];

  1110.     }

  1111.     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);

  1112. 

  1113.     if(sws_init_context(c, srcFilter, dstFilter) < 0){

  1114.         sws_freeContext(c);

  1115.         return NULL;

  1116.     }

  1117. 

  1118.     return c;

  1119. }

  1120. #endif

  1121. 

  1122. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1123.                                 float lumaSharpen, float chromaSharpen,

  1124.                                 float chromaHShift, float chromaVShift,

  1125.                                 int verbose)

  1126. {

  1127.     SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  1128.     if (!filter)

  1129.         return NULL;

  1130. 

  1131.     if (lumaGBlur!=0.0) {

  1132.         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  1133.         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  1134.     } else {

  1135.         filter->lumH= sws_getIdentityVec();

  1136.         filter->lumV= sws_getIdentityVec();

  1137.     }

  1138. 

  1139.     if (chromaGBlur!=0.0) {

  1140.         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  1141.         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  1142.     } else {

  1143.         filter->chrH= sws_getIdentityVec();

  1144.         filter->chrV= sws_getIdentityVec();

  1145.     }

  1146. 

  1147.     if (chromaSharpen!=0.0) {

  1148.         SwsVector *id= sws_getIdentityVec();

  1149.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1150.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1151.         sws_addVec(filter->chrH, id);

  1152.         sws_addVec(filter->chrV, id);

  1153.         sws_freeVec(id);

  1154.     }

  1155. 

  1156.     if (lumaSharpen!=0.0) {

  1157.         SwsVector *id= sws_getIdentityVec();

  1158.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1159.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1160.         sws_addVec(filter->lumH, id);

  1161.         sws_addVec(filter->lumV, id);

  1162.         sws_freeVec(id);

  1163.     }

  1164. 

  1165.     if (chromaHShift != 0.0)

  1166.         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  1167. 

  1168.     if (chromaVShift != 0.0)

  1169.         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  1170. 

  1171.     sws_normalizeVec(filter->chrH, 1.0);

  1172.     sws_normalizeVec(filter->chrV, 1.0);

  1173.     sws_normalizeVec(filter->lumH, 1.0);

  1174.     sws_normalizeVec(filter->lumV, 1.0);

  1175. 

  1176.     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1177.     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1178. 

  1179.     return filter;

  1180. }

  1181. 

  1182. SwsVector *sws_allocVec(int length)

  1183. {

  1184.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1185.     if (!vec)

  1186.         return NULL;

  1187.     vec->length = length;

  1188.     vec->coeff  = av_malloc(sizeof(double) * length);

  1189.     if (!vec->coeff)

  1190.         av_freep(&vec);

  1191.     return vec;

  1192. }

  1193. 

  1194. SwsVector *sws_getGaussianVec(double variance, double quality)

  1195. {

  1196.     const int length= (int)(variance*quality + 0.5) | 1;

  1197.     int i;

  1198.     double middle= (length-1)*0.5;

  1199.     SwsVector *vec= sws_allocVec(length);

  1200. 

  1201.     if (!vec)

  1202.         return NULL;

  1203. 

  1204.     for (i=0; i<length; i++) {

  1205.         double dist= i-middle;

  1206.         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);

  1207.     }

  1208. 

  1209.     sws_normalizeVec(vec, 1.0);

  1210. 

  1211.     return vec;

  1212. }

  1213. 

  1214. SwsVector *sws_getConstVec(double c, int length)

  1215. {

  1216.     int i;

  1217.     SwsVector *vec= sws_allocVec(length);

  1218. 

  1219.     if (!vec)

  1220.         return NULL;

  1221. 

  1222.     for (i=0; i<length; i++)

  1223.         vec->coeff[i]= c;

  1224. 

  1225.     return vec;

  1226. }

  1227. 

  1228. SwsVector *sws_getIdentityVec(void)

  1229. {

  1230.     return sws_getConstVec(1.0, 1);

  1231. }

  1232. 

  1233. static double sws_dcVec(SwsVector *a)

  1234. {

  1235.     int i;

  1236.     double sum=0;

  1237. 

  1238.     for (i=0; i<a->length; i++)

  1239.         sum+= a->coeff[i];

  1240. 

  1241.     return sum;

  1242. }

  1243. 

  1244. void sws_scaleVec(SwsVector *a, double scalar)

  1245. {

  1246.     int i;

  1247. 

  1248.     for (i=0; i<a->length; i++)

  1249.         a->coeff[i]*= scalar;

  1250. }

  1251. 

  1252. void sws_normalizeVec(SwsVector *a, double height)

  1253. {

  1254.     sws_scaleVec(a, height/sws_dcVec(a));

  1255. }

  1256. 

  1257. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1258. {

  1259.     int length= a->length + b->length - 1;

  1260.     int i, j;

  1261.     SwsVector *vec= sws_getConstVec(0.0, length);

  1262. 

  1263.     if (!vec)

  1264.         return NULL;

  1265. 

  1266.     for (i=0; i<a->length; i++) {

  1267.         for (j=0; j<b->length; j++) {

  1268.             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1269.         }

  1270.     }

  1271. 

  1272.     return vec;

  1273. }

  1274. 

  1275. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1276. {

  1277.     int length= FFMAX(a->length, b->length);

  1278.     int i;

  1279.     SwsVector *vec= sws_getConstVec(0.0, length);

  1280. 

  1281.     if (!vec)

  1282.         return NULL;

  1283. 

  1284.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1285.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];

  1286. 

  1287.     return vec;

  1288. }

  1289. 

  1290. static SwsVector *sws_diffVec(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. /* shift left / or right if "shift" is negative */

  1306. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1307. {

  1308.     int length= a->length + FFABS(shift)*2;

  1309.     int i;

  1310.     SwsVector *vec= sws_getConstVec(0.0, length);

  1311. 

  1312.     if (!vec)

  1313.         return NULL;

  1314. 

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

  1316.         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1317.     }

  1318. 

  1319.     return vec;

  1320. }

  1321. 

  1322. void sws_shiftVec(SwsVector *a, int shift)

  1323. {

  1324.     SwsVector *shifted= sws_getShiftedVec(a, shift);

  1325.     av_free(a->coeff);

  1326.     a->coeff= shifted->coeff;

  1327.     a->length= shifted->length;

  1328.     av_free(shifted);

  1329. }

  1330. 

  1331. void sws_addVec(SwsVector *a, SwsVector *b)

  1332. {

  1333.     SwsVector *sum= sws_sumVec(a, b);

  1334.     av_free(a->coeff);

  1335.     a->coeff= sum->coeff;

  1336.     a->length= sum->length;

  1337.     av_free(sum);

  1338. }

  1339. 

  1340. void sws_subVec(SwsVector *a, SwsVector *b)

  1341. {

  1342.     SwsVector *diff= sws_diffVec(a, b);

  1343.     av_free(a->coeff);

  1344.     a->coeff= diff->coeff;

  1345.     a->length= diff->length;

  1346.     av_free(diff);

  1347. }

  1348. 

  1349. void sws_convVec(SwsVector *a, SwsVector *b)

  1350. {

  1351.     SwsVector *conv= sws_getConvVec(a, b);

  1352.     av_free(a->coeff);

  1353.     a->coeff= conv->coeff;

  1354.     a->length= conv->length;

  1355.     av_free(conv);

  1356. }

  1357. 

  1358. SwsVector *sws_cloneVec(SwsVector *a)

  1359. {

  1360.     int i;

  1361.     SwsVector *vec= sws_allocVec(a->length);

  1362. 

  1363.     if (!vec)

  1364.         return NULL;

  1365. 

  1366.     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];

  1367. 

  1368.     return vec;

  1369. }

  1370. 

  1371. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1372. {

  1373.     int i;

  1374.     double max=0;

  1375.     double min=0;

  1376.     double range;

  1377. 

  1378.     for (i=0; i<a->length; i++)

  1379.         if (a->coeff[i]>max) max= a->coeff[i];

  1380. 

  1381.     for (i=0; i<a->length; i++)

  1382.         if (a->coeff[i]<min) min= a->coeff[i];

  1383. 

  1384.     range= max - min;

  1385. 

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

  1387.         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1388.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1389.         for (;x>0; x--) av_log(log_ctx, log_level, " ");

  1390.         av_log(log_ctx, log_level, "|\n");

  1391.     }

  1392. }

  1393. 

  1394. void sws_freeVec(SwsVector *a)

  1395. {

  1396.     if (!a) return;

  1397.     av_freep(&a->coeff);

  1398.     a->length=0;

  1399.     av_free(a);

  1400. }

  1401. 

  1402. void sws_freeFilter(SwsFilter *filter)

  1403. {

  1404.     if (!filter) return;

  1405. 

  1406.     if (filter->lumH) sws_freeVec(filter->lumH);

  1407.     if (filter->lumV) sws_freeVec(filter->lumV);

  1408.     if (filter->chrH) sws_freeVec(filter->chrH);

  1409.     if (filter->chrV) sws_freeVec(filter->chrV);

  1410.     av_free(filter);

  1411. }

  1412. 

  1413. void sws_freeContext(SwsContext *c)

  1414. {

  1415.     int i;

  1416.     if (!c) return;

  1417. 

  1418.     if (c->lumPixBuf) {

  1419.         for (i=0; i<c->vLumBufSize; i++)

  1420.             av_freep(&c->lumPixBuf[i]);

  1421.         av_freep(&c->lumPixBuf);

  1422.     }

  1423. 

  1424.     if (c->chrUPixBuf) {

  1425.         for (i=0; i<c->vChrBufSize; i++)

  1426.             av_freep(&c->chrUPixBuf[i]);

  1427.         av_freep(&c->chrUPixBuf);

  1428.         av_freep(&c->chrVPixBuf);

  1429.     }

  1430. 

  1431.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1432.         for (i=0; i<c->vLumBufSize; i++)

  1433.             av_freep(&c->alpPixBuf[i]);

  1434.         av_freep(&c->alpPixBuf);

  1435.     }

  1436. 

  1437.     av_freep(&c->vLumFilter);

  1438.     av_freep(&c->vChrFilter);

  1439.     av_freep(&c->hLumFilter);

  1440.     av_freep(&c->hChrFilter);

  1441. #if HAVE_ALTIVEC

  1442.     av_freep(&c->vYCoeffsBank);

  1443.     av_freep(&c->vCCoeffsBank);

  1444. #endif

  1445. 

  1446.     av_freep(&c->vLumFilterPos);

  1447.     av_freep(&c->vChrFilterPos);

  1448.     av_freep(&c->hLumFilterPos);

  1449.     av_freep(&c->hChrFilterPos);

  1450. 

  1451. #if HAVE_MMX

  1452. #ifdef MAP_ANONYMOUS

  1453.     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1454.     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1455. #elif HAVE_VIRTUALALLOC

  1456.     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1457.     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1458. #else

  1459.     av_free(c->lumMmx2FilterCode);

  1460.     av_free(c->chrMmx2FilterCode);

  1461. #endif

  1462.     c->lumMmx2FilterCode=NULL;

  1463.     c->chrMmx2FilterCode=NULL;

  1464. #endif /* HAVE_MMX */

  1465. 

  1466.     av_freep(&c->yuvTable);

  1467.     av_freep(&c->formatConvBuffer);

  1468. 

  1469.     av_free(c);

  1470. }

  1471. 

  1472. struct SwsContext *sws_getCachedContext(struct SwsContext *context,

  1473.                                         int srcW, int srcH, enum PixelFormat srcFormat,

  1474.                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1475.                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1476. {

  1477.     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};

  1478. 

  1479.     if (!param)

  1480.         param = default_param;

  1481. 

  1482.     if (context &&

  1483.         (context->srcW      != srcW      ||

  1484.          context->srcH      != srcH      ||

  1485.          context->srcFormat != srcFormat ||

  1486.          context->dstW      != dstW      ||

  1487.          context->dstH      != dstH      ||

  1488.          context->dstFormat != dstFormat ||

  1489.          context->flags     != flags     ||

  1490.          context->param[0]  != param[0]  ||

  1491.          context->param[1]  != param[1])) {

  1492.         sws_freeContext(context);

  1493.         context = NULL;

  1494.     }

  1495. 

  1496.     if (!context) {

  1497.         if (!(context = sws_alloc_context()))

  1498.             return NULL;

  1499.         context->srcW      = srcW;

  1500.         context->srcH      = srcH;

  1501.         context->srcRange  = handle_jpeg(&srcFormat);

  1502.         context->srcFormat = srcFormat;

  1503.         context->dstW      = dstW;

  1504.         context->dstH      = dstH;

  1505.         context->dstRange  = handle_jpeg(&dstFormat);

  1506.         context->dstFormat = dstFormat;

  1507.         context->flags     = flags;

  1508.         context->param[0]  = param[0];

  1509.         context->param[1]  = param[1];

  1510.         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);

  1511.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1512.             sws_freeContext(context);

  1513.             return NULL;

  1514.         }

  1515.     }

  1516.     return context;

  1517. }