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.
29872 Commits
32 Branches
283MB
Tree: 7cc8d6385a
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '7cc8d6385a'
${ noResults }
FFmpeg/libswscale/utils.c

1516 lines
54KB
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. #include <inttypes.h>

  22. #include <string.h>

  23. #include <math.h>

  24. #include <stdio.h>

  25. #include "config.h"

  26. #include <assert.h>

  27. #if HAVE_SYS_MMAN_H

  28. #include <sys/mman.h>

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

  30. #define MAP_ANONYMOUS MAP_ANON

  31. #endif

  32. #endif

  33. #if HAVE_VIRTUALALLOC

  34. #define WIN32_LEAN_AND_MEAN

  35. #include <windows.h>

  36. #endif

  37. #include "swscale.h"

  38. #include "swscale_internal.h"

  39. #include "rgb2rgb.h"

  40. #include "libavutil/intreadwrite.h"

  41. #include "libavutil/x86_cpu.h"

  42. #include "libavutil/cpu.h"

  43. #include "libavutil/avutil.h"

  44. #include "libavutil/bswap.h"

  45. #include "libavutil/opt.h"

  46. #include "libavutil/pixdesc.h"

  47. 

  48. unsigned swscale_version(void)

  49. {

  50.     return LIBSWSCALE_VERSION_INT;

  51. }

  52. 

  53. const char *swscale_configuration(void)

  54. {

  55.     return LIBAV_CONFIGURATION;

  56. }

  57. 

  58. const char *swscale_license(void)

  59. {

  60. #define LICENSE_PREFIX "libswscale license: "

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

  62. }

  63. 

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

  65. 

  66. typedef struct FormatEntry {

  67.     int is_supported_in, is_supported_out;

  68. } FormatEntry;

  69. 

  70. const static FormatEntry format_entries[PIX_FMT_NB] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  138. };

  139. 

  140. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  141. {

  142.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  143.         format_entries[pix_fmt].is_supported_in : 0;

  144. }

  145. 

  146. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  147. {

  148.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  149.         format_entries[pix_fmt].is_supported_out : 0;

  150. }

  151. 

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

  153. 

  154. const char *sws_format_name(enum PixelFormat format)

  155. {

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

  157.         return av_pix_fmt_descriptors[format].name;

  158.     else

  159.         return "Unknown format";

  160. }

  161. 

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

  163. {

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

  165.     else           return getSplineCoeff(        0.0,

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

  167.                                                  c + 3.0*d,

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

  169.                                          dist-1.0);

  170. }

  171. 

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

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

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

  175. {

  176.     int i;

  177.     int filterSize;

  178.     int filter2Size;

  179.     int minFilterSize;

  180.     int64_t *filter=NULL;

  181.     int64_t *filter2=NULL;

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

  183.     int ret= -1;

  184. 

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

  186. 

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

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

  189. 

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

  191.         int i;

  192.         filterSize= 1;

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

  194. 

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

  196.             filter[i*filterSize]= fone;

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

  198.         }

  199. 

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

  201.         int i;

  202.         int xDstInSrc;

  203.         filterSize= 1;

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

  205. 

  206.         xDstInSrc= xInc/2 - 0x8000;

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

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

  209. 

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

  211.             filter[i]= fone;

  212.             xDstInSrc+= xInc;

  213.         }

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

  215.         int i;

  216.         int xDstInSrc;

  217.         filterSize= 2;

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

  219. 

  220.         xDstInSrc= xInc/2 - 0x8000;

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

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

  223.             int j;

  224. 

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

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

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

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

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

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

  231.                 xx++;

  232.             }

  233.             xDstInSrc+= xInc;

  234.         }

  235.     } else {

  236.         int xDstInSrc;

  237.         int sizeFactor;

  238. 

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

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

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

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

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

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

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

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

  247.         else {

  248.             sizeFactor= 0; //GCC warning killer

  249.             assert(0);

  250.         }

  251. 

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

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

  254. 

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

  256. 

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

  258. 

  259.         xDstInSrc= xInc - 0x10000;

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

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

  262.             int j;

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

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

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

  266.                 double floatd;

  267.                 int64_t coeff;

  268. 

  269.                 if (xInc > 1<<16)

  270.                     d= d*dstW/srcW;

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

  272. 

  273.                 if (flags & SWS_BICUBIC) {

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

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

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

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

  278. 

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

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

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

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

  283.                     else

  284.                         coeff=0.0;

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

  286.                 }

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

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

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

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

  291.                 }*/

  292.                 else if (flags & SWS_X) {

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

  294.                     double c;

  295. 

  296.                     if (floatd<1.0)

  297.                         c = cos(floatd*M_PI);

  298.                     else

  299.                         c=-1.0;

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

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

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

  303.                 } else if (flags & SWS_AREA) {

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

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

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

  307.                     else coeff=0.0;

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

  309.                 } else if (flags & SWS_GAUSS) {

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

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

  312.                 } else if (flags & SWS_SINC) {

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

  314.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  318.                 } else if (flags & SWS_BILINEAR) {

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

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

  321.                     coeff *= fone >> 30;

  322.                 } else if (flags & SWS_SPLINE) {

  323.                     double p=-2.196152422706632;

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

  325.                 } else {

  326.                     coeff= 0.0; //GCC warning killer

  327.                     assert(0);

  328.                 }

  329. 

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

  331.                 xx++;

  332.             }

  333.             xDstInSrc+= 2*xInc;

  334.         }

  335.     }

  336. 

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

  338.        av_free(filter);

  339.     */

  340.     assert(filterSize>0);

  341.     filter2Size= filterSize;

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

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

  344.     assert(filter2Size>0);

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

  346. 

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

  348.         int j, k;

  349. 

  350.         if(srcFilter) {

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

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

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

  354.             }

  355.         } else {

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

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

  358.         }

  359.         //FIXME dstFilter

  360. 

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

  362.     }

  363.     av_freep(&filter);

  364. 

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

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

  367.     minFilterSize= 0;

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

  369.         int min= filter2Size;

  370.         int j;

  371.         int64_t cutOff=0.0;

  372. 

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

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

  375.             int k;

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

  377. 

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

  379. 

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

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

  382. 

  383.             // move filter coefficients left

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

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

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

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

  388.         }

  389. 

  390.         cutOff=0;

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

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

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

  394. 

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

  396.             min--;

  397.         }

  398. 

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

  400.     }

  401. 

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

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

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

  405.         if (minFilterSize < 5)

  406.             filterAlign = 4;

  407. 

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

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

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

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

  412.         // decent-sized vector.

  413.         if (minFilterSize < 3)

  414.             filterAlign = 1;

  415.     }

  416. 

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

  418.         // special case for unscaled vertical filtering

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

  420.             filterAlign= 1;

  421.     }

  422. 

  423.     assert(minFilterSize > 0);

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

  425.     assert(filterSize > 0);

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

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

  428.         goto fail;

  429.     *outFilterSize= filterSize;

  430. 

  431.     if (flags&SWS_PRINT_INFO)

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

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

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

  435.         int j;

  436. 

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

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

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

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

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

  442.         }

  443.     }

  444. 

  445.     //FIXME try to align filterPos if possible

  446. 

  447.     //fix borders

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

  449.         int j;

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

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

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

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

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

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

  456.             }

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

  458.         }

  459. 

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

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

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

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

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

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

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

  467.             }

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

  469.         }

  470.     }

  471. 

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

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

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

  475. 

  476.     /* normalize & store in outFilter */

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

  478.         int j;

  479.         int64_t error=0;

  480.         int64_t sum=0;

  481. 

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

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

  484.         }

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

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

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

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

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

  490.             error= v - intV*sum;

  491.         }

  492.     }

  493. 

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

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

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

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

  498.     }

  499. 

  500.     ret=0;

  501. fail:

  502.     av_free(filter);

  503.     av_free(filter2);

  504.     return ret;

  505. }

  506. 

  507. #if HAVE_MMX2

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

  509. {

  510.     uint8_t *fragmentA;

  511.     x86_reg imm8OfPShufW1A;

  512.     x86_reg imm8OfPShufW2A;

  513.     x86_reg fragmentLengthA;

  514.     uint8_t *fragmentB;

  515.     x86_reg imm8OfPShufW1B;

  516.     x86_reg imm8OfPShufW2B;

  517.     x86_reg fragmentLengthB;

  518.     int fragmentPos;

  519. 

  520.     int xpos, i;

  521. 

  522.     // create an optimized horizontal scaling routine

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

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

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

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

  527.      * of fragmentA is used.

  528.      */

  529. 

  530.     //code fragment

  531. 

  532.     __asm__ volatile(

  533.         "jmp                         9f                 \n\t"

  534.     // Begin

  535.         "0:                                             \n\t"

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

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

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

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

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

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

  542.         "1:                                             \n\t"

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

  544.         "2:                                             \n\t"

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

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

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

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

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

  550. 

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

  552. 

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

  554.     // End

  555.         "9:                                             \n\t"

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

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

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

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

  560.         "dec                         %1                 \n\t"

  561.         "dec                         %2                 \n\t"

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

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

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

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

  566. 

  567. 

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

  569.         "=r" (fragmentLengthA)

  570.     );

  571. 

  572.     __asm__ volatile(

  573.         "jmp                         9f                 \n\t"

  574.     // Begin

  575.         "0:                                             \n\t"

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

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

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

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

  580.         "1:                                             \n\t"

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

  582.         "2:                                             \n\t"

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

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

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

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

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

  588. 

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

  590. 

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

  592.     // End

  593.         "9:                                             \n\t"

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

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

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

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

  598.         "dec                         %1                 \n\t"

  599.         "dec                         %2                 \n\t"

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

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

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

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

  604. 

  605. 

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

  607.         "=r" (fragmentLengthB)

  608.     );

  609. 

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

  611.     fragmentPos=0;

  612. 

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

  614.         int xx=xpos>>16;

  615. 

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

  617.             int a=0;

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

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

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

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

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

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

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

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

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

  627.             int shift=0;

  628. 

  629.             if (filterCode) {

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

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

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

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

  634.                 filterPos[i/2]= xx;

  635. 

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

  637. 

  638.                 filterCode[fragmentPos + imm8OfPShufW1]=

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

  640.                 filterCode[fragmentPos + imm8OfPShufW2]=

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

  642. 

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

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

  645. 

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

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

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

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

  650.                 }

  651.             }

  652. 

  653.             fragmentPos+= fragmentLength;

  654. 

  655.             if (filterCode)

  656.                 filterCode[fragmentPos]= RET;

  657.         }

  658.         xpos+=xInc;

  659.     }

  660.     if (filterCode)

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

  662. 

  663.     return fragmentPos + 1;

  664. }

  665. #endif /* HAVE_MMX2 */

  666. 

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

  668. {

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

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

  671. }

  672. 

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

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

  675.                              int brightness, int contrast, int saturation)

  676. {

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

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

  679. 

  680.     c->brightness= brightness;

  681.     c->contrast  = contrast;

  682.     c->saturation= saturation;

  683.     c->srcRange  = srcRange;

  684.     c->dstRange  = dstRange;

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

  686. 

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

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

  689. 

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

  691.     //FIXME factorize

  692. 

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

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

  695.     return 0;

  696. }

  697. 

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

  699.                              int *srcRange, int **table, int *dstRange,

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

  701. {

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

  703. 

  704.     *inv_table = c->srcColorspaceTable;

  705.     *table     = c->dstColorspaceTable;

  706.     *srcRange  = c->srcRange;

  707.     *dstRange  = c->dstRange;

  708.     *brightness= c->brightness;

  709.     *contrast  = c->contrast;

  710.     *saturation= c->saturation;

  711. 

  712.     return 0;

  713. }

  714. 

  715. static int handle_jpeg(enum PixelFormat *format)

  716. {

  717.     switch (*format) {

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

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

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

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

  722.     default:                                          return 0;

  723.     }

  724. }

  725. 

  726. SwsContext *sws_alloc_context(void)

  727. {

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

  729. 

  730.     c->av_class = &sws_context_class;

  731.     av_opt_set_defaults(c);

  732. 

  733.     return c;

  734. }

  735. 

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

  737. {

  738.     int i;

  739.     int usesVFilter, usesHFilter;

  740.     int unscaled;

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

  742.     int srcW= c->srcW;

  743.     int srcH= c->srcH;

  744.     int dstW= c->dstW;

  745.     int dstH= c->dstH;

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

  747.     int flags, cpu_flags;

  748.     enum PixelFormat srcFormat= c->srcFormat;

  749.     enum PixelFormat dstFormat= c->dstFormat;

  750. 

  751.     cpu_flags = av_get_cpu_flags();

  752.     flags     = c->flags;

  753.     emms_c();

  754.     if (!rgb15to16) sws_rgb2rgb_init();

  755. 

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

  757. 

  758.     if (!sws_isSupportedInput(srcFormat)) {

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

  760.         return AVERROR(EINVAL);

  761.     }

  762.     if (!sws_isSupportedOutput(dstFormat)) {

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

  764.         return AVERROR(EINVAL);

  765.     }

  766. 

  767.     i= flags & ( SWS_POINT

  768.                 |SWS_AREA

  769.                 |SWS_BILINEAR

  770.                 |SWS_FAST_BILINEAR

  771.                 |SWS_BICUBIC

  772.                 |SWS_X

  773.                 |SWS_GAUSS

  774.                 |SWS_LANCZOS

  775.                 |SWS_SINC

  776.                 |SWS_SPLINE

  777.                 |SWS_BICUBLIN);

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

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

  780.         return AVERROR(EINVAL);

  781.     }

  782.     /* sanity check */

  783.     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

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

  785.                srcW, srcH, dstW, dstH);

  786.         return AVERROR(EINVAL);

  787.     }

  788. 

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

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

  791. 

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

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

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

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

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

  797. 

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

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

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

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

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

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

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

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

  806. 

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

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

  809. 

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

  811.     if (flags & SWS_FULL_CHR_H_INT &&

  812.         dstFormat != PIX_FMT_RGBA &&

  813.         dstFormat != PIX_FMT_ARGB &&

  814.         dstFormat != PIX_FMT_BGRA &&

  815.         dstFormat != PIX_FMT_ABGR &&

  816.         dstFormat != PIX_FMT_RGB24 &&

  817.         dstFormat != PIX_FMT_BGR24) {

  818.         av_log(c, AV_LOG_ERROR,

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

  820.                sws_format_name(dstFormat));

  821.         flags &= ~SWS_FULL_CHR_H_INT;

  822.         c->flags = flags;

  823.     }

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

  825. 

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

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

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

  829. 

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

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

  832.       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8

  833.       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4

  834.       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE

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

  836.         c->chrSrcHSubSample=1;

  837. 

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

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

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

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

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

  843. 

  844.     /* unscaled special cases */

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

  846.         ff_get_unscaled_swscale(c);

  847. 

  848.         if (c->swScale) {

  849.             if (flags&SWS_PRINT_INFO)

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

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

  852.             return 0;

  853.         }

  854.     }

  855. 

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

  857.     if (c->srcBpc < 8)

  858.         c->srcBpc = 8;

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

  860.     if (c->dstBpc < 8)

  861.         c->dstBpc = 8;

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

  863.         dst_stride <<= 1;

  864.     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,

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

  866.                      fail);

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

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

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

  870.             if (flags&SWS_PRINT_INFO)

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

  872.         }

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

  874.     }

  875.     else

  876.         c->canMMX2BeUsed=0;

  877. 

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

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

  880. 

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

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

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

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

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

  886.     // first and last pixel

  887.     if (flags&SWS_FAST_BILINEAR) {

  888.         if (c->canMMX2BeUsed) {

  889.             c->lumXInc+= 20;

  890.             c->chrXInc+= 20;

  891.         }

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

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

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

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

  896.         }

  897.     }

  898. 

  899.     /* precalculate horizontal scaler filter coefficients */

  900.     {

  901. #if HAVE_MMX2

  902. // can't downscale !!!

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

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

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

  906. 

  907. #ifdef MAP_ANONYMOUS

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

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

  910. #elif HAVE_VIRTUALALLOC

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

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

  913. #else

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

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

  916. #endif

  917. 

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

  919.                 return AVERROR(ENOMEM);

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

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

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

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

  924. 

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

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

  927. 

  928. #ifdef MAP_ANONYMOUS

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

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

  931. #endif

  932.         } else

  933. #endif /* HAVE_MMX2 */

  934.         {

  935.             const int filterAlign=

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

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

  938.                 1;

  939. 

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

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

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

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

  944.                 goto fail;

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

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

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

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

  949.                 goto fail;

  950.         }

  951.     } // initialize horizontal stuff

  952. 

  953.     /* precalculate vertical scaler filter coefficients */

  954.     {

  955.         const int filterAlign=

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

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

  958.             1;

  959. 

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

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

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

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

  964.             goto fail;

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

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

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

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

  969.             goto fail;

  970. 

  971. #if HAVE_ALTIVEC

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

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

  974. 

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

  976.             int j;

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

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

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

  980.         }

  981. 

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

  983.             int j;

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

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

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

  987.         }

  988. #endif

  989.     }

  990. 

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

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

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

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

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

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

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

  998. 

  999.         nextSlice>>= c->chrSrcVSubSample;

  1000.         nextSlice<<= c->chrSrcVSubSample;

  1001.         if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)

  1002.             c->vLumBufSize= nextSlice - c->vLumFilterPos[i];

  1003.         if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))

  1004.             c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];

  1005.     }

  1006. 

  1007.     // allocate pixbufs (we use dynamic allocation because otherwise we would need to

  1008.     // allocate several megabytes to handle all possible cases)

  1009.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1010.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1011.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1012.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1013.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1014.     //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)

  1015.     /* align at 16 bytes for AltiVec */

  1016.     for (i=0; i<c->vLumBufSize; i++) {

  1017.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1018.         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];

  1019.     }

  1020.     // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)

  1021.     c->uv_off_px   = dst_stride_px + 64 / (c->dstBpc &~ 7);

  1022.     c->uv_off_byte = dst_stride + 16;

  1023.     for (i=0; i<c->vChrBufSize; i++) {

  1024.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);

  1025.         c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];

  1026.         c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1027.     }

  1028.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1029.         for (i=0; i<c->vLumBufSize; i++) {

  1030.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1031.             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];

  1032.         }

  1033. 

  1034.     //try to avoid drawing green stuff between the right end and the stride end

  1035.     for (i=0; i<c->vChrBufSize; i++)

  1036.         memset(c->chrUPixBuf[i], 64, dst_stride*2+1);

  1037. 

  1038.     assert(c->chrDstH <= dstH);

  1039. 

  1040.     if (flags&SWS_PRINT_INFO) {

  1041.         if      (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1042.         else if (flags&SWS_BILINEAR)      av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1043.         else if (flags&SWS_BICUBIC)       av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1044.         else if (flags&SWS_X)             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1045.         else if (flags&SWS_POINT)         av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1046.         else if (flags&SWS_AREA)          av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1047.         else if (flags&SWS_BICUBLIN)      av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1048.         else if (flags&SWS_GAUSS)         av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1049.         else if (flags&SWS_SINC)          av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1050.         else if (flags&SWS_LANCZOS)       av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1051.         else if (flags&SWS_SPLINE)        av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1052.         else                              av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1053. 

  1054.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1055.                sws_format_name(srcFormat),

  1056. #ifdef DITHER1XBPP

  1057.                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||

  1058.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1059.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",

  1060. #else

  1061.                "",

  1062. #endif

  1063.                sws_format_name(dstFormat));

  1064. 

  1065.         if      (HAVE_MMX2     && cpu_flags & AV_CPU_FLAG_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");

  1066.         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1067.         else if (HAVE_MMX      && cpu_flags & AV_CPU_FLAG_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");

  1068.         else if (HAVE_ALTIVEC  && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1069.         else                                   av_log(c, AV_LOG_INFO, "using C\n");

  1070. 

  1071.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1072.         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1073.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1074.         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1075.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1076.     }

  1077. 

  1078.     c->swScale= ff_getSwsFunc(c);

  1079.     return 0;

  1080. fail: //FIXME replace things by appropriate error codes

  1081.     return -1;

  1082. }

  1083. 

  1084. #if FF_API_SWS_GETCONTEXT

  1085. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1086.                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1087.                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1088. {

  1089.     SwsContext *c;

  1090. 

  1091.     if(!(c=sws_alloc_context()))

  1092.         return NULL;

  1093. 

  1094.     c->flags= flags;

  1095.     c->srcW= srcW;

  1096.     c->srcH= srcH;

  1097.     c->dstW= dstW;

  1098.     c->dstH= dstH;

  1099.     c->srcRange = handle_jpeg(&srcFormat);

  1100.     c->dstRange = handle_jpeg(&dstFormat);

  1101.     c->srcFormat= srcFormat;

  1102.     c->dstFormat= dstFormat;

  1103. 

  1104.     if (param) {

  1105.         c->param[0] = param[0];

  1106.         c->param[1] = param[1];

  1107.     }

  1108.     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);

  1109. 

  1110.     if(sws_init_context(c, srcFilter, dstFilter) < 0){

  1111.         sws_freeContext(c);

  1112.         return NULL;

  1113.     }

  1114. 

  1115.     return c;

  1116. }

  1117. #endif

  1118. 

  1119. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1120.                                 float lumaSharpen, float chromaSharpen,

  1121.                                 float chromaHShift, float chromaVShift,

  1122.                                 int verbose)

  1123. {

  1124.     SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  1125.     if (!filter)

  1126.         return NULL;

  1127. 

  1128.     if (lumaGBlur!=0.0) {

  1129.         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  1130.         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  1131.     } else {

  1132.         filter->lumH= sws_getIdentityVec();

  1133.         filter->lumV= sws_getIdentityVec();

  1134.     }

  1135. 

  1136.     if (chromaGBlur!=0.0) {

  1137.         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  1138.         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  1139.     } else {

  1140.         filter->chrH= sws_getIdentityVec();

  1141.         filter->chrV= sws_getIdentityVec();

  1142.     }

  1143. 

  1144.     if (chromaSharpen!=0.0) {

  1145.         SwsVector *id= sws_getIdentityVec();

  1146.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1147.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1148.         sws_addVec(filter->chrH, id);

  1149.         sws_addVec(filter->chrV, id);

  1150.         sws_freeVec(id);

  1151.     }

  1152. 

  1153.     if (lumaSharpen!=0.0) {

  1154.         SwsVector *id= sws_getIdentityVec();

  1155.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1156.         sws_scaleVec(filter->lumV, -lumaSharpen);

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

  1158.         sws_addVec(filter->lumV, id);

  1159.         sws_freeVec(id);

  1160.     }

  1161. 

  1162.     if (chromaHShift != 0.0)

  1163.         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  1164. 

  1165.     if (chromaVShift != 0.0)

  1166.         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  1167. 

  1168.     sws_normalizeVec(filter->chrH, 1.0);

  1169.     sws_normalizeVec(filter->chrV, 1.0);

  1170.     sws_normalizeVec(filter->lumH, 1.0);

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

  1172. 

  1173.     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1174.     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1175. 

  1176.     return filter;

  1177. }

  1178. 

  1179. SwsVector *sws_allocVec(int length)

  1180. {

  1181.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1182.     if (!vec)

  1183.         return NULL;

  1184.     vec->length = length;

  1185.     vec->coeff  = av_malloc(sizeof(double) * length);

  1186.     if (!vec->coeff)

  1187.         av_freep(&vec);

  1188.     return vec;

  1189. }

  1190. 

  1191. SwsVector *sws_getGaussianVec(double variance, double quality)

  1192. {

  1193.     const int length= (int)(variance*quality + 0.5) | 1;

  1194.     int i;

  1195.     double middle= (length-1)*0.5;

  1196.     SwsVector *vec= sws_allocVec(length);

  1197. 

  1198.     if (!vec)

  1199.         return NULL;

  1200. 

  1201.     for (i=0; i<length; i++) {

  1202.         double dist= i-middle;

  1203.         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);

  1204.     }

  1205. 

  1206.     sws_normalizeVec(vec, 1.0);

  1207. 

  1208.     return vec;

  1209. }

  1210. 

  1211. SwsVector *sws_getConstVec(double c, int length)

  1212. {

  1213.     int i;

  1214.     SwsVector *vec= sws_allocVec(length);

  1215. 

  1216.     if (!vec)

  1217.         return NULL;

  1218. 

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

  1220.         vec->coeff[i]= c;

  1221. 

  1222.     return vec;

  1223. }

  1224. 

  1225. SwsVector *sws_getIdentityVec(void)

  1226. {

  1227.     return sws_getConstVec(1.0, 1);

  1228. }

  1229. 

  1230. static double sws_dcVec(SwsVector *a)

  1231. {

  1232.     int i;

  1233.     double sum=0;

  1234. 

  1235.     for (i=0; i<a->length; i++)

  1236.         sum+= a->coeff[i];

  1237. 

  1238.     return sum;

  1239. }

  1240. 

  1241. void sws_scaleVec(SwsVector *a, double scalar)

  1242. {

  1243.     int i;

  1244. 

  1245.     for (i=0; i<a->length; i++)

  1246.         a->coeff[i]*= scalar;

  1247. }

  1248. 

  1249. void sws_normalizeVec(SwsVector *a, double height)

  1250. {

  1251.     sws_scaleVec(a, height/sws_dcVec(a));

  1252. }

  1253. 

  1254. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1255. {

  1256.     int length= a->length + b->length - 1;

  1257.     int i, j;

  1258.     SwsVector *vec= sws_getConstVec(0.0, length);

  1259. 

  1260.     if (!vec)

  1261.         return NULL;

  1262. 

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

  1264.         for (j=0; j<b->length; j++) {

  1265.             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1266.         }

  1267.     }

  1268. 

  1269.     return vec;

  1270. }

  1271. 

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

  1273. {

  1274.     int length= FFMAX(a->length, b->length);

  1275.     int i;

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

  1277. 

  1278.     if (!vec)

  1279.         return NULL;

  1280. 

  1281.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1282.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];

  1283. 

  1284.     return vec;

  1285. }

  1286. 

  1287. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1288. {

  1289.     int length= FFMAX(a->length, b->length);

  1290.     int i;

  1291.     SwsVector *vec= sws_getConstVec(0.0, length);

  1292. 

  1293.     if (!vec)

  1294.         return NULL;

  1295. 

  1296.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1297.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1298. 

  1299.     return vec;

  1300. }

  1301. 

  1302. /* shift left / or right if "shift" is negative */

  1303. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1304. {

  1305.     int length= a->length + FFABS(shift)*2;

  1306.     int i;

  1307.     SwsVector *vec= sws_getConstVec(0.0, length);

  1308. 

  1309.     if (!vec)

  1310.         return NULL;

  1311. 

  1312.     for (i=0; i<a->length; i++) {

  1313.         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1314.     }

  1315. 

  1316.     return vec;

  1317. }

  1318. 

  1319. void sws_shiftVec(SwsVector *a, int shift)

  1320. {

  1321.     SwsVector *shifted= sws_getShiftedVec(a, shift);

  1322.     av_free(a->coeff);

  1323.     a->coeff= shifted->coeff;

  1324.     a->length= shifted->length;

  1325.     av_free(shifted);

  1326. }

  1327. 

  1328. void sws_addVec(SwsVector *a, SwsVector *b)

  1329. {

  1330.     SwsVector *sum= sws_sumVec(a, b);

  1331.     av_free(a->coeff);

  1332.     a->coeff= sum->coeff;

  1333.     a->length= sum->length;

  1334.     av_free(sum);

  1335. }

  1336. 

  1337. void sws_subVec(SwsVector *a, SwsVector *b)

  1338. {

  1339.     SwsVector *diff= sws_diffVec(a, b);

  1340.     av_free(a->coeff);

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

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

  1343.     av_free(diff);

  1344. }

  1345. 

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

  1347. {

  1348.     SwsVector *conv= sws_getConvVec(a, b);

  1349.     av_free(a->coeff);

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

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

  1352.     av_free(conv);

  1353. }

  1354. 

  1355. SwsVector *sws_cloneVec(SwsVector *a)

  1356. {

  1357.     int i;

  1358.     SwsVector *vec= sws_allocVec(a->length);

  1359. 

  1360.     if (!vec)

  1361.         return NULL;

  1362. 

  1363.     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];

  1364. 

  1365.     return vec;

  1366. }

  1367. 

  1368. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1369. {

  1370.     int i;

  1371.     double max=0;

  1372.     double min=0;

  1373.     double range;

  1374. 

  1375.     for (i=0; i<a->length; i++)

  1376.         if (a->coeff[i]>max) max= a->coeff[i];

  1377. 

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

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

  1380. 

  1381.     range= max - min;

  1382. 

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

  1384.         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1385.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1386.         for (;x>0; x--) av_log(log_ctx, log_level, " ");

  1387.         av_log(log_ctx, log_level, "|\n");

  1388.     }

  1389. }

  1390. 

  1391. void sws_freeVec(SwsVector *a)

  1392. {

  1393.     if (!a) return;

  1394.     av_freep(&a->coeff);

  1395.     a->length=0;

  1396.     av_free(a);

  1397. }

  1398. 

  1399. void sws_freeFilter(SwsFilter *filter)

  1400. {

  1401.     if (!filter) return;

  1402. 

  1403.     if (filter->lumH) sws_freeVec(filter->lumH);

  1404.     if (filter->lumV) sws_freeVec(filter->lumV);

  1405.     if (filter->chrH) sws_freeVec(filter->chrH);

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

  1407.     av_free(filter);

  1408. }

  1409. 

  1410. void sws_freeContext(SwsContext *c)

  1411. {

  1412.     int i;

  1413.     if (!c) return;

  1414. 

  1415.     if (c->lumPixBuf) {

  1416.         for (i=0; i<c->vLumBufSize; i++)

  1417.             av_freep(&c->lumPixBuf[i]);

  1418.         av_freep(&c->lumPixBuf);

  1419.     }

  1420. 

  1421.     if (c->chrUPixBuf) {

  1422.         for (i=0; i<c->vChrBufSize; i++)

  1423.             av_freep(&c->chrUPixBuf[i]);

  1424.         av_freep(&c->chrUPixBuf);

  1425.         av_freep(&c->chrVPixBuf);

  1426.     }

  1427. 

  1428.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1429.         for (i=0; i<c->vLumBufSize; i++)

  1430.             av_freep(&c->alpPixBuf[i]);

  1431.         av_freep(&c->alpPixBuf);

  1432.     }

  1433. 

  1434.     av_freep(&c->vLumFilter);

  1435.     av_freep(&c->vChrFilter);

  1436.     av_freep(&c->hLumFilter);

  1437.     av_freep(&c->hChrFilter);

  1438. #if HAVE_ALTIVEC

  1439.     av_freep(&c->vYCoeffsBank);

  1440.     av_freep(&c->vCCoeffsBank);

  1441. #endif

  1442. 

  1443.     av_freep(&c->vLumFilterPos);

  1444.     av_freep(&c->vChrFilterPos);

  1445.     av_freep(&c->hLumFilterPos);

  1446.     av_freep(&c->hChrFilterPos);

  1447. 

  1448. #if HAVE_MMX

  1449. #ifdef MAP_ANONYMOUS

  1450.     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1451.     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1452. #elif HAVE_VIRTUALALLOC

  1453.     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1454.     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1455. #else

  1456.     av_free(c->lumMmx2FilterCode);

  1457.     av_free(c->chrMmx2FilterCode);

  1458. #endif

  1459.     c->lumMmx2FilterCode=NULL;

  1460.     c->chrMmx2FilterCode=NULL;

  1461. #endif /* HAVE_MMX */

  1462. 

  1463.     av_freep(&c->yuvTable);

  1464.     av_free(c->formatConvBuffer);

  1465. 

  1466.     av_free(c);

  1467. }

  1468. 

  1469. struct SwsContext *sws_getCachedContext(struct SwsContext *context,

  1470.                                         int srcW, int srcH, enum PixelFormat srcFormat,

  1471.                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1472.                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1473. {

  1474.     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};

  1475. 

  1476.     if (!param)

  1477.         param = default_param;

  1478. 

  1479.     if (context &&

  1480.         (context->srcW      != srcW      ||

  1481.          context->srcH      != srcH      ||

  1482.          context->srcFormat != srcFormat ||

  1483.          context->dstW      != dstW      ||

  1484.          context->dstH      != dstH      ||

  1485.          context->dstFormat != dstFormat ||

  1486.          context->flags     != flags     ||

  1487.          context->param[0]  != param[0]  ||

  1488.          context->param[1]  != param[1])) {

  1489.         sws_freeContext(context);

  1490.         context = NULL;

  1491.     }

  1492. 

  1493.     if (!context) {

  1494.         if (!(context = sws_alloc_context()))

  1495.             return NULL;

  1496.         context->srcW      = srcW;

  1497.         context->srcH      = srcH;

  1498.         context->srcRange  = handle_jpeg(&srcFormat);

  1499.         context->srcFormat = srcFormat;

  1500.         context->dstW      = dstW;

  1501.         context->dstH      = dstH;

  1502.         context->dstRange  = handle_jpeg(&dstFormat);

  1503.         context->dstFormat = dstFormat;

  1504.         context->flags     = flags;

  1505.         context->param[0]  = param[0];

  1506.         context->param[1]  = param[1];

  1507.         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);

  1508.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1509.             sws_freeContext(context);

  1510.             return NULL;

  1511.         }

  1512.     }

  1513.     return context;

  1514. }