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.
35187 Commits
32 Branches
2.1GB
Tree: c1ae524e2b
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c1ae524e2b'
${ noResults }
FFmpeg/libswscale/utils.c

1545 lines
55KB
Raw Blame History 

  1. /*

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

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. #define _SVID_SOURCE //needed for MAP_ANONYMOUS

  22. #define _DARWIN_C_SOURCE // needed for MAP_ANON

  23. #include <inttypes.h>

  24. #include <string.h>

  25. #include <math.h>

  26. #include <stdio.h>

  27. #include "config.h"

  28. #include <assert.h>

  29. #if HAVE_SYS_MMAN_H

  30. #include <sys/mman.h>

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

  32. #define MAP_ANONYMOUS MAP_ANON

  33. #endif

  34. #endif

  35. #if HAVE_VIRTUALALLOC

  36. #define WIN32_LEAN_AND_MEAN

  37. #include <windows.h>

  38. #endif

  39. #include "swscale.h"

  40. #include "swscale_internal.h"

  41. #include "rgb2rgb.h"

  42. #include "libavutil/intreadwrite.h"

  43. #include "libavutil/x86_cpu.h"

  44. #include "libavutil/cpu.h"

  45. #include "libavutil/avutil.h"

  46. #include "libavutil/bswap.h"

  47. #include "libavutil/opt.h"

  48. #include "libavutil/pixdesc.h"

  49. #include "libavutil/avassert.h"

  50. 

  51. unsigned swscale_version(void)

  52. {

  53.     return LIBSWSCALE_VERSION_INT;

  54. }

  55. 

  56. const char *swscale_configuration(void)

  57. {

  58.     return FFMPEG_CONFIGURATION;

  59. }

  60. 

  61. const char *swscale_license(void)

  62. {

  63. #define LICENSE_PREFIX "libswscale license: "

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

  65. }

  66. 

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

  68. 

  69. typedef struct FormatEntry {

  70.     int is_supported_in, is_supported_out;

  71. } FormatEntry;

  72. 

  73. const static FormatEntry format_entries[PIX_FMT_NB] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  155. };

  156. 

  157. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  158. {

  159.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  160.         format_entries[pix_fmt].is_supported_in : 0;

  161. }

  162. 

  163. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  164. {

  165.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  166.         format_entries[pix_fmt].is_supported_out : 0;

  167. }

  168. 

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

  170. 

  171. #if FF_API_SWS_FORMAT_NAME

  172. const char *sws_format_name(enum PixelFormat format)

  173. {

  174.     return av_get_pix_fmt_name(format);

  175. }

  176. #endif

  177. 

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

  179. {

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

  181.     else           return getSplineCoeff(        0.0,

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

  183.                                                  c + 3.0*d,

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

  185.                                          dist-1.0);

  186. }

  187. 

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

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

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

  191. {

  192.     int i;

  193.     int filterSize;

  194.     int filter2Size;

  195.     int minFilterSize;

  196.     int64_t *filter=NULL;

  197.     int64_t *filter2=NULL;

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

  199.     int ret= -1;

  200. 

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

  202. 

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

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

  205. 

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

  207.         int i;

  208.         filterSize= 1;

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

  210. 

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

  212.             filter[i*filterSize]= fone;

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

  214.         }

  215. 

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

  217.         int i;

  218.         int xDstInSrc;

  219.         filterSize= 1;

  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. 

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

  227.             filter[i]= fone;

  228.             xDstInSrc+= xInc;

  229.         }

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

  231.         int i;

  232.         int xDstInSrc;

  233.         filterSize= 2;

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

  235. 

  236.         xDstInSrc= xInc/2 - 0x8000;

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

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

  239.             int j;

  240. 

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

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

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

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

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

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

  247.                 xx++;

  248.             }

  249.             xDstInSrc+= xInc;

  250.         }

  251.     } else {

  252.         int xDstInSrc;

  253.         int sizeFactor;

  254. 

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

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

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

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

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

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

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

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

  263.         else {

  264.             sizeFactor= 0; //GCC warning killer

  265.             assert(0);

  266.         }

  267. 

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

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

  270. 

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

  272. 

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

  274. 

  275.         xDstInSrc= xInc - 0x10000;

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

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

  278.             int j;

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

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

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

  282.                 double floatd;

  283.                 int64_t coeff;

  284. 

  285.                 if (xInc > 1<<16)

  286.                     d= d*dstW/srcW;

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

  288. 

  289.                 if (flags & SWS_BICUBIC) {

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

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

  292. 

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

  294.                         coeff = 0.0;

  295.                     } else {

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

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

  298. 

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

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

  301.                         else

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

  303.                     }

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

  305.                 }

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

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

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

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

  310.                 }*/

  311.                 else if (flags & SWS_X) {

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

  313.                     double c;

  314. 

  315.                     if (floatd<1.0)

  316.                         c = cos(floatd*M_PI);

  317.                     else

  318.                         c=-1.0;

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

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

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

  322.                 } else if (flags & SWS_AREA) {

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

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

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

  326.                     else coeff=0.0;

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

  328.                 } else if (flags & SWS_GAUSS) {

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

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

  331.                 } else if (flags & SWS_SINC) {

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

  333.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  337.                 } else if (flags & SWS_BILINEAR) {

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

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

  340.                     coeff *= fone >> 30;

  341.                 } else if (flags & SWS_SPLINE) {

  342.                     double p=-2.196152422706632;

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

  344.                 } else {

  345.                     coeff= 0.0; //GCC warning killer

  346.                     assert(0);

  347.                 }

  348. 

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

  350.                 xx++;

  351.             }

  352.             xDstInSrc+= 2*xInc;

  353.         }

  354.     }

  355. 

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

  357.        av_free(filter);

  358.     */

  359.     assert(filterSize>0);

  360.     filter2Size= filterSize;

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

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

  363.     assert(filter2Size>0);

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

  365. 

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

  367.         int j, k;

  368. 

  369.         if(srcFilter) {

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

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

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

  373.             }

  374.         } else {

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

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

  377.         }

  378.         //FIXME dstFilter

  379. 

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

  381.     }

  382.     av_freep(&filter);

  383. 

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

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

  386.     minFilterSize= 0;

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

  388.         int min= filter2Size;

  389.         int j;

  390.         int64_t cutOff=0.0;

  391. 

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

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

  394.             int k;

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

  396. 

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

  398. 

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

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

  401. 

  402.             // move filter coefficients left

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

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

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

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

  407.         }

  408. 

  409.         cutOff=0;

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

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

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

  413. 

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

  415.             min--;

  416.         }

  417. 

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

  419.     }

  420. 

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

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

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

  424.         if (minFilterSize < 5)

  425.             filterAlign = 4;

  426. 

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

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

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

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

  431.         // decent-sized vector.

  432.         if (minFilterSize < 3)

  433.             filterAlign = 1;

  434.     }

  435. 

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

  437.         // special case for unscaled vertical filtering

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

  439.             filterAlign= 1;

  440.     }

  441. 

  442.     assert(minFilterSize > 0);

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

  444.     assert(filterSize > 0);

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

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

  447.         goto fail;

  448.     *outFilterSize= filterSize;

  449. 

  450.     if (flags&SWS_PRINT_INFO)

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

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

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

  454.         int j;

  455. 

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

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

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

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

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

  461.         }

  462.     }

  463. 

  464.     //FIXME try to align filterPos if possible

  465. 

  466.     //fix borders

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

  468.         int j;

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

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

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

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

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

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

  475.             }

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

  477.         }

  478. 

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

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

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

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

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

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

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

  486.             }

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

  488.         }

  489.     }

  490. 

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

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

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

  494. 

  495.     /* normalize & store in outFilter */

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

  497.         int j;

  498.         int64_t error=0;

  499.         int64_t sum=0;

  500. 

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

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

  503.         }

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

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

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

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

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

  509.             error= v - intV*sum;

  510.         }

  511.     }

  512. 

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

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

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

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

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

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

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

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

  521.     }

  522. 

  523.     ret=0;

  524. fail:

  525.     av_free(filter);

  526.     av_free(filter2);

  527.     return ret;

  528. }

  529. 

  530. #if HAVE_MMX2

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

  532. {

  533.     uint8_t *fragmentA;

  534.     x86_reg imm8OfPShufW1A;

  535.     x86_reg imm8OfPShufW2A;

  536.     x86_reg fragmentLengthA;

  537.     uint8_t *fragmentB;

  538.     x86_reg imm8OfPShufW1B;

  539.     x86_reg imm8OfPShufW2B;

  540.     x86_reg fragmentLengthB;

  541.     int fragmentPos;

  542. 

  543.     int xpos, i;

  544. 

  545.     // create an optimized horizontal scaling routine

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

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

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

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

  550.      * of fragmentA is used.

  551.      */

  552. 

  553.     //code fragment

  554. 

  555.     __asm__ volatile(

  556.         "jmp                         9f                 \n\t"

  557.     // Begin

  558.         "0:                                             \n\t"

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

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

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

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

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

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

  565.         "1:                                             \n\t"

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

  567.         "2:                                             \n\t"

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

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

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

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

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

  573. 

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

  575. 

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

  577.     // End

  578.         "9:                                             \n\t"

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

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

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

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

  583.         "dec                         %1                 \n\t"

  584.         "dec                         %2                 \n\t"

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

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

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

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

  589. 

  590. 

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

  592.         "=r" (fragmentLengthA)

  593.     );

  594. 

  595.     __asm__ volatile(

  596.         "jmp                         9f                 \n\t"

  597.     // Begin

  598.         "0:                                             \n\t"

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

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

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

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

  603.         "1:                                             \n\t"

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

  605.         "2:                                             \n\t"

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

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

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

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

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

  611. 

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

  613. 

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

  615.     // End

  616.         "9:                                             \n\t"

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

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

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

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

  621.         "dec                         %1                 \n\t"

  622.         "dec                         %2                 \n\t"

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

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

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

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

  627. 

  628. 

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

  630.         "=r" (fragmentLengthB)

  631.     );

  632. 

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

  634.     fragmentPos=0;

  635. 

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

  637.         int xx=xpos>>16;

  638. 

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

  640.             int a=0;

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

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

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

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

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

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

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

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

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

  650.             int shift=0;

  651. 

  652.             if (filterCode) {

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

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

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

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

  657.                 filterPos[i/2]= xx;

  658. 

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

  660. 

  661.                 filterCode[fragmentPos + imm8OfPShufW1]=

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

  663.                 filterCode[fragmentPos + imm8OfPShufW2]=

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

  665. 

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

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

  668. 

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

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

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

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

  673.                 }

  674.             }

  675. 

  676.             fragmentPos+= fragmentLength;

  677. 

  678.             if (filterCode)

  679.                 filterCode[fragmentPos]= RET;

  680.         }

  681.         xpos+=xInc;

  682.     }

  683.     if (filterCode)

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

  685. 

  686.     return fragmentPos + 1;

  687. }

  688. #endif /* HAVE_MMX2 */

  689. 

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

  691. {

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

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

  694. }

  695. 

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

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

  698.                              int brightness, int contrast, int saturation)

  699. {

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

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

  702. 

  703.     c->brightness= brightness;

  704.     c->contrast  = contrast;

  705.     c->saturation= saturation;

  706.     c->srcRange  = srcRange;

  707.     c->dstRange  = dstRange;

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

  709. 

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

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

  712. 

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

  714.     //FIXME factorize

  715. 

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

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

  718.     return 0;

  719. }

  720. 

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

  722.                              int *srcRange, int **table, int *dstRange,

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

  724. {

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

  726. 

  727.     *inv_table = c->srcColorspaceTable;

  728.     *table     = c->dstColorspaceTable;

  729.     *srcRange  = c->srcRange;

  730.     *dstRange  = c->dstRange;

  731.     *brightness= c->brightness;

  732.     *contrast  = c->contrast;

  733.     *saturation= c->saturation;

  734. 

  735.     return 0;

  736. }

  737. 

  738. static int handle_jpeg(enum PixelFormat *format)

  739. {

  740.     switch (*format) {

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

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

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

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

  745.     default:                                          return 0;

  746.     }

  747. }

  748. 

  749. SwsContext *sws_alloc_context(void)

  750. {

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

  752. 

  753.     c->av_class = &sws_context_class;

  754.     av_opt_set_defaults(c);

  755. 

  756.     return c;

  757. }

  758. 

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

  760. {

  761.     int i, j;

  762.     int usesVFilter, usesHFilter;

  763.     int unscaled;

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

  765.     int srcW= c->srcW;

  766.     int srcH= c->srcH;

  767.     int dstW= c->dstW;

  768.     int dstH= c->dstH;

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

  770.     int flags, cpu_flags;

  771.     enum PixelFormat srcFormat= c->srcFormat;

  772.     enum PixelFormat dstFormat= c->dstFormat;

  773. 

  774.     cpu_flags = av_get_cpu_flags();

  775.     flags     = c->flags;

  776.     emms_c();

  777.     if (!rgb15to16) sws_rgb2rgb_init();

  778. 

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

  780. 

  781.     handle_jpeg(&srcFormat);

  782.     handle_jpeg(&dstFormat);

  783. 

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

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

  786.         c->srcFormat= srcFormat;

  787.         c->dstFormat= dstFormat;

  788.     }

  789. 

  790.     if (!sws_isSupportedInput(srcFormat)) {

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

  792.         return AVERROR(EINVAL);

  793.     }

  794.     if (!sws_isSupportedOutput(dstFormat)) {

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

  796.         return AVERROR(EINVAL);

  797.     }

  798. 

  799.     i= flags & ( SWS_POINT

  800.                 |SWS_AREA

  801.                 |SWS_BILINEAR

  802.                 |SWS_FAST_BILINEAR

  803.                 |SWS_BICUBIC

  804.                 |SWS_X

  805.                 |SWS_GAUSS

  806.                 |SWS_LANCZOS

  807.                 |SWS_SINC

  808.                 |SWS_SPLINE

  809.                 |SWS_BICUBLIN);

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

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

  812.         return AVERROR(EINVAL);

  813.     }

  814.     /* sanity check */

  815.     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

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

  817.                srcW, srcH, dstW, dstH);

  818.         return AVERROR(EINVAL);

  819.     }

  820. 

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

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

  823. 

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

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

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

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

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

  829. 

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

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

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

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

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

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

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

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

  838. 

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

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

  841. 

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

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

  844. 

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

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

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

  848. 

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

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

  851.       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8

  852.       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4

  853.       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE

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

  855.         c->chrSrcHSubSample=1;

  856. 

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

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

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

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

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

  862. 

  863.     /* unscaled special cases */

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

  865.         ff_get_unscaled_swscale(c);

  866. 

  867.         if (c->swScale) {

  868.             if (flags&SWS_PRINT_INFO)

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

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

  871.             return 0;

  872.         }

  873.     }

  874. 

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

  876.     if (c->srcBpc < 8)

  877.         c->srcBpc = 8;

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

  879.     if (c->dstBpc < 8)

  880.         c->dstBpc = 8;

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

  882.         c->srcBpc = 16;

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

  884.         dst_stride <<= 1;

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

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

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

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

  889.             if (flags&SWS_PRINT_INFO)

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

  891.         }

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

  893.     }

  894.     else

  895.         c->canMMX2BeUsed=0;

  896. 

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

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

  899. 

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

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

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

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

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

  905.     // first and last pixel

  906.     if (flags&SWS_FAST_BILINEAR) {

  907.         if (c->canMMX2BeUsed) {

  908.             c->lumXInc+= 20;

  909.             c->chrXInc+= 20;

  910.         }

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

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

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

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

  915.         }

  916.     }

  917. 

  918.     /* precalculate horizontal scaler filter coefficients */

  919.     {

  920. #if HAVE_MMX2

  921. // can't downscale !!!

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

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

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

  925. 

  926. #ifdef MAP_ANONYMOUS

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

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

  929. #elif HAVE_VIRTUALALLOC

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

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

  932. #else

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

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

  935. #endif

  936. 

  937. #ifdef MAP_ANONYMOUS

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

  939. #else

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

  941. #endif

  942.                 return AVERROR(ENOMEM);

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

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

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

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

  947. 

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

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

  950. 

  951. #ifdef MAP_ANONYMOUS

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

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

  954. #endif

  955.         } else

  956. #endif /* HAVE_MMX2 */

  957.         {

  958.             const int filterAlign=

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

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

  961.                 1;

  962. 

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

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

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

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

  967.                 goto fail;

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

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

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

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

  972.                 goto fail;

  973.         }

  974.     } // initialize horizontal stuff

  975. 

  976.     /* precalculate vertical scaler filter coefficients */

  977.     {

  978.         const int filterAlign=

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

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

  981.             1;

  982. 

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

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

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

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

  987.             goto fail;

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

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

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

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

  992.             goto fail;

  993. 

  994. #if HAVE_ALTIVEC

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

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

  997. 

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

  999.             int j;

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

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

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

  1003.         }

  1004. 

  1005.         for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {

  1006.             int j;

  1007.             short *p = (short *)&c->vCCoeffsBank[i];

  1008.             for (j=0;j<8;j++)

  1009.                 p[j] = c->vChrFilter[i];

  1010.         }

  1011. #endif

  1012.     }

  1013. 

  1014.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1015.     c->vLumBufSize= c->vLumFilterSize;

  1016.     c->vChrBufSize= c->vChrFilterSize;

  1017.     for (i=0; i<dstH; i++) {

  1018.         int chrI= (int64_t)i*c->chrDstH / dstH;

  1019.         int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,

  1020.                            ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));

  1021. 

  1022.         nextSlice>>= c->chrSrcVSubSample;

  1023.         nextSlice<<= c->chrSrcVSubSample;

  1024.         if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)

  1025.             c->vLumBufSize= nextSlice - c->vLumFilterPos[i];

  1026.         if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))

  1027.             c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];

  1028.     }

  1029. 

  1030.     // allocate pixbufs (we use dynamic allocation because otherwise we would need to

  1031.     // allocate several megabytes to handle all possible cases)

  1032.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1033.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1034.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1035.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1036.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1037.     //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)

  1038.     /* align at 16 bytes for AltiVec */

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

  1040.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1041.         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];

  1042.     }

  1043.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1044.     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);

  1045.     c->uv_offx2 = dst_stride + 16;

  1046.     for (i=0; i<c->vChrBufSize; i++) {

  1047.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);

  1048.         c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];

  1049.         c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1050.     }

  1051.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1052.         for (i=0; i<c->vLumBufSize; i++) {

  1053.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1054.             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];

  1055.         }

  1056. 

  1057.     //try to avoid drawing green stuff between the right end and the stride end

  1058.     for (i=0; i<c->vChrBufSize; i++)

  1059.         if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){

  1060.             av_assert0(c->dstBpc > 10);

  1061.             for(j=0; j<dst_stride/2+1; j++)

  1062.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1063.         } else

  1064.             for(j=0; j<dst_stride+1; j++)

  1065.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1066. 

  1067.     assert(c->chrDstH <= dstH);

  1068. 

  1069.     if (flags&SWS_PRINT_INFO) {

  1070.         if      (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1071.         else if (flags&SWS_BILINEAR)      av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1072.         else if (flags&SWS_BICUBIC)       av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1073.         else if (flags&SWS_X)             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1074.         else if (flags&SWS_POINT)         av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1075.         else if (flags&SWS_AREA)          av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1076.         else if (flags&SWS_BICUBLIN)      av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1077.         else if (flags&SWS_GAUSS)         av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1078.         else if (flags&SWS_SINC)          av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1079.         else if (flags&SWS_LANCZOS)       av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1080.         else if (flags&SWS_SPLINE)        av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1081.         else                              av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1082. 

  1083.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1084.                av_get_pix_fmt_name(srcFormat),

  1085. #ifdef DITHER1XBPP

  1086.                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||

  1087.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1088.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",

  1089. #else

  1090.                "",

  1091. #endif

  1092.                av_get_pix_fmt_name(dstFormat));

  1093. 

  1094.         if      (HAVE_MMX2     && cpu_flags & AV_CPU_FLAG_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");

  1095.         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1096.         else if (HAVE_MMX      && cpu_flags & AV_CPU_FLAG_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");

  1097.         else if (HAVE_ALTIVEC  && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1098.         else                                   av_log(c, AV_LOG_INFO, "using C\n");

  1099. 

  1100.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1101.         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1102.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1103.         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1104.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1105.     }

  1106. 

  1107.     c->swScale= ff_getSwsFunc(c);

  1108.     return 0;

  1109. fail: //FIXME replace things by appropriate error codes

  1110.     return -1;

  1111. }

  1112. 

  1113. #if FF_API_SWS_GETCONTEXT

  1114. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1115.                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1116.                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1117. {

  1118.     SwsContext *c;

  1119. 

  1120.     if(!(c=sws_alloc_context()))

  1121.         return NULL;

  1122. 

  1123.     c->flags= flags;

  1124.     c->srcW= srcW;

  1125.     c->srcH= srcH;

  1126.     c->dstW= dstW;

  1127.     c->dstH= dstH;

  1128.     c->srcRange = handle_jpeg(&srcFormat);

  1129.     c->dstRange = handle_jpeg(&dstFormat);

  1130.     c->srcFormat= srcFormat;

  1131.     c->dstFormat= dstFormat;

  1132. 

  1133.     if (param) {

  1134.         c->param[0] = param[0];

  1135.         c->param[1] = param[1];

  1136.     }

  1137.     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);

  1138. 

  1139.     if(sws_init_context(c, srcFilter, dstFilter) < 0){

  1140.         sws_freeContext(c);

  1141.         return NULL;

  1142.     }

  1143. 

  1144.     return c;

  1145. }

  1146. #endif

  1147. 

  1148. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1149.                                 float lumaSharpen, float chromaSharpen,

  1150.                                 float chromaHShift, float chromaVShift,

  1151.                                 int verbose)

  1152. {

  1153.     SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  1154.     if (!filter)

  1155.         return NULL;

  1156. 

  1157.     if (lumaGBlur!=0.0) {

  1158.         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  1159.         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  1160.     } else {

  1161.         filter->lumH= sws_getIdentityVec();

  1162.         filter->lumV= sws_getIdentityVec();

  1163.     }

  1164. 

  1165.     if (chromaGBlur!=0.0) {

  1166.         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  1167.         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  1168.     } else {

  1169.         filter->chrH= sws_getIdentityVec();

  1170.         filter->chrV= sws_getIdentityVec();

  1171.     }

  1172. 

  1173.     if (chromaSharpen!=0.0) {

  1174.         SwsVector *id= sws_getIdentityVec();

  1175.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1176.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1177.         sws_addVec(filter->chrH, id);

  1178.         sws_addVec(filter->chrV, id);

  1179.         sws_freeVec(id);

  1180.     }

  1181. 

  1182.     if (lumaSharpen!=0.0) {

  1183.         SwsVector *id= sws_getIdentityVec();

  1184.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1185.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1186.         sws_addVec(filter->lumH, id);

  1187.         sws_addVec(filter->lumV, id);

  1188.         sws_freeVec(id);

  1189.     }

  1190. 

  1191.     if (chromaHShift != 0.0)

  1192.         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  1193. 

  1194.     if (chromaVShift != 0.0)

  1195.         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  1196. 

  1197.     sws_normalizeVec(filter->chrH, 1.0);

  1198.     sws_normalizeVec(filter->chrV, 1.0);

  1199.     sws_normalizeVec(filter->lumH, 1.0);

  1200.     sws_normalizeVec(filter->lumV, 1.0);

  1201. 

  1202.     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1203.     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1204. 

  1205.     return filter;

  1206. }

  1207. 

  1208. SwsVector *sws_allocVec(int length)

  1209. {

  1210.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1211.     if (!vec)

  1212.         return NULL;

  1213.     vec->length = length;

  1214.     vec->coeff  = av_malloc(sizeof(double) * length);

  1215.     if (!vec->coeff)

  1216.         av_freep(&vec);

  1217.     return vec;

  1218. }

  1219. 

  1220. SwsVector *sws_getGaussianVec(double variance, double quality)

  1221. {

  1222.     const int length= (int)(variance*quality + 0.5) | 1;

  1223.     int i;

  1224.     double middle= (length-1)*0.5;

  1225.     SwsVector *vec= sws_allocVec(length);

  1226. 

  1227.     if (!vec)

  1228.         return NULL;

  1229. 

  1230.     for (i=0; i<length; i++) {

  1231.         double dist= i-middle;

  1232.         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);

  1233.     }

  1234. 

  1235.     sws_normalizeVec(vec, 1.0);

  1236. 

  1237.     return vec;

  1238. }

  1239. 

  1240. SwsVector *sws_getConstVec(double c, int length)

  1241. {

  1242.     int i;

  1243.     SwsVector *vec= sws_allocVec(length);

  1244. 

  1245.     if (!vec)

  1246.         return NULL;

  1247. 

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

  1249.         vec->coeff[i]= c;

  1250. 

  1251.     return vec;

  1252. }

  1253. 

  1254. SwsVector *sws_getIdentityVec(void)

  1255. {

  1256.     return sws_getConstVec(1.0, 1);

  1257. }

  1258. 

  1259. static double sws_dcVec(SwsVector *a)

  1260. {

  1261.     int i;

  1262.     double sum=0;

  1263. 

  1264.     for (i=0; i<a->length; i++)

  1265.         sum+= a->coeff[i];

  1266. 

  1267.     return sum;

  1268. }

  1269. 

  1270. void sws_scaleVec(SwsVector *a, double scalar)

  1271. {

  1272.     int i;

  1273. 

  1274.     for (i=0; i<a->length; i++)

  1275.         a->coeff[i]*= scalar;

  1276. }

  1277. 

  1278. void sws_normalizeVec(SwsVector *a, double height)

  1279. {

  1280.     sws_scaleVec(a, height/sws_dcVec(a));

  1281. }

  1282. 

  1283. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1284. {

  1285.     int length= a->length + b->length - 1;

  1286.     int i, j;

  1287.     SwsVector *vec= sws_getConstVec(0.0, length);

  1288. 

  1289.     if (!vec)

  1290.         return NULL;

  1291. 

  1292.     for (i=0; i<a->length; i++) {

  1293.         for (j=0; j<b->length; j++) {

  1294.             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1295.         }

  1296.     }

  1297. 

  1298.     return vec;

  1299. }

  1300. 

  1301. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1302. {

  1303.     int length= FFMAX(a->length, b->length);

  1304.     int i;

  1305.     SwsVector *vec= sws_getConstVec(0.0, length);

  1306. 

  1307.     if (!vec)

  1308.         return NULL;

  1309. 

  1310.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1311.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];

  1312. 

  1313.     return vec;

  1314. }

  1315. 

  1316. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1317. {

  1318.     int length= FFMAX(a->length, b->length);

  1319.     int i;

  1320.     SwsVector *vec= sws_getConstVec(0.0, length);

  1321. 

  1322.     if (!vec)

  1323.         return NULL;

  1324. 

  1325.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1326.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1327. 

  1328.     return vec;

  1329. }

  1330. 

  1331. /* shift left / or right if "shift" is negative */

  1332. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1333. {

  1334.     int length= a->length + FFABS(shift)*2;

  1335.     int i;

  1336.     SwsVector *vec= sws_getConstVec(0.0, length);

  1337. 

  1338.     if (!vec)

  1339.         return NULL;

  1340. 

  1341.     for (i=0; i<a->length; i++) {

  1342.         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1343.     }

  1344. 

  1345.     return vec;

  1346. }

  1347. 

  1348. void sws_shiftVec(SwsVector *a, int shift)

  1349. {

  1350.     SwsVector *shifted= sws_getShiftedVec(a, shift);

  1351.     av_free(a->coeff);

  1352.     a->coeff= shifted->coeff;

  1353.     a->length= shifted->length;

  1354.     av_free(shifted);

  1355. }

  1356. 

  1357. void sws_addVec(SwsVector *a, SwsVector *b)

  1358. {

  1359.     SwsVector *sum= sws_sumVec(a, b);

  1360.     av_free(a->coeff);

  1361.     a->coeff= sum->coeff;

  1362.     a->length= sum->length;

  1363.     av_free(sum);

  1364. }

  1365. 

  1366. void sws_subVec(SwsVector *a, SwsVector *b)

  1367. {

  1368.     SwsVector *diff= sws_diffVec(a, b);

  1369.     av_free(a->coeff);

  1370.     a->coeff= diff->coeff;

  1371.     a->length= diff->length;

  1372.     av_free(diff);

  1373. }

  1374. 

  1375. void sws_convVec(SwsVector *a, SwsVector *b)

  1376. {

  1377.     SwsVector *conv= sws_getConvVec(a, b);

  1378.     av_free(a->coeff);

  1379.     a->coeff= conv->coeff;

  1380.     a->length= conv->length;

  1381.     av_free(conv);

  1382. }

  1383. 

  1384. SwsVector *sws_cloneVec(SwsVector *a)

  1385. {

  1386.     int i;

  1387.     SwsVector *vec= sws_allocVec(a->length);

  1388. 

  1389.     if (!vec)

  1390.         return NULL;

  1391. 

  1392.     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];

  1393. 

  1394.     return vec;

  1395. }

  1396. 

  1397. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1398. {

  1399.     int i;

  1400.     double max=0;

  1401.     double min=0;

  1402.     double range;

  1403. 

  1404.     for (i=0; i<a->length; i++)

  1405.         if (a->coeff[i]>max) max= a->coeff[i];

  1406. 

  1407.     for (i=0; i<a->length; i++)

  1408.         if (a->coeff[i]<min) min= a->coeff[i];

  1409. 

  1410.     range= max - min;

  1411. 

  1412.     for (i=0; i<a->length; i++) {

  1413.         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1414.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1415.         for (;x>0; x--) av_log(log_ctx, log_level, " ");

  1416.         av_log(log_ctx, log_level, "|\n");

  1417.     }

  1418. }

  1419. 

  1420. void sws_freeVec(SwsVector *a)

  1421. {

  1422.     if (!a) return;

  1423.     av_freep(&a->coeff);

  1424.     a->length=0;

  1425.     av_free(a);

  1426. }

  1427. 

  1428. void sws_freeFilter(SwsFilter *filter)

  1429. {

  1430.     if (!filter) return;

  1431. 

  1432.     if (filter->lumH) sws_freeVec(filter->lumH);

  1433.     if (filter->lumV) sws_freeVec(filter->lumV);

  1434.     if (filter->chrH) sws_freeVec(filter->chrH);

  1435.     if (filter->chrV) sws_freeVec(filter->chrV);

  1436.     av_free(filter);

  1437. }

  1438. 

  1439. void sws_freeContext(SwsContext *c)

  1440. {

  1441.     int i;

  1442.     if (!c) return;

  1443. 

  1444.     if (c->lumPixBuf) {

  1445.         for (i=0; i<c->vLumBufSize; i++)

  1446.             av_freep(&c->lumPixBuf[i]);

  1447.         av_freep(&c->lumPixBuf);

  1448.     }

  1449. 

  1450.     if (c->chrUPixBuf) {

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

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

  1453.         av_freep(&c->chrUPixBuf);

  1454.         av_freep(&c->chrVPixBuf);

  1455.     }

  1456. 

  1457.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1458.         for (i=0; i<c->vLumBufSize; i++)

  1459.             av_freep(&c->alpPixBuf[i]);

  1460.         av_freep(&c->alpPixBuf);

  1461.     }

  1462. 

  1463.     av_freep(&c->vLumFilter);

  1464.     av_freep(&c->vChrFilter);

  1465.     av_freep(&c->hLumFilter);

  1466.     av_freep(&c->hChrFilter);

  1467. #if HAVE_ALTIVEC

  1468.     av_freep(&c->vYCoeffsBank);

  1469.     av_freep(&c->vCCoeffsBank);

  1470. #endif

  1471. 

  1472.     av_freep(&c->vLumFilterPos);

  1473.     av_freep(&c->vChrFilterPos);

  1474.     av_freep(&c->hLumFilterPos);

  1475.     av_freep(&c->hChrFilterPos);

  1476. 

  1477. #if HAVE_MMX

  1478. #ifdef MAP_ANONYMOUS

  1479.     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1480.     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1481. #elif HAVE_VIRTUALALLOC

  1482.     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1483.     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1484. #else

  1485.     av_free(c->lumMmx2FilterCode);

  1486.     av_free(c->chrMmx2FilterCode);

  1487. #endif

  1488.     c->lumMmx2FilterCode=NULL;

  1489.     c->chrMmx2FilterCode=NULL;

  1490. #endif /* HAVE_MMX */

  1491. 

  1492.     av_freep(&c->yuvTable);

  1493.     av_freep(&c->formatConvBuffer);

  1494. 

  1495.     av_free(c);

  1496. }

  1497. 

  1498. struct SwsContext *sws_getCachedContext(struct SwsContext *context,

  1499.                                         int srcW, int srcH, enum PixelFormat srcFormat,

  1500.                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1501.                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1502. {

  1503.     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};

  1504. 

  1505.     if (!param)

  1506.         param = default_param;

  1507. 

  1508.     if (context &&

  1509.         (context->srcW      != srcW      ||

  1510.          context->srcH      != srcH      ||

  1511.          context->srcFormat != srcFormat ||

  1512.          context->dstW      != dstW      ||

  1513.          context->dstH      != dstH      ||

  1514.          context->dstFormat != dstFormat ||

  1515.          context->flags     != flags     ||

  1516.          context->param[0]  != param[0]  ||

  1517.          context->param[1]  != param[1])) {

  1518.         sws_freeContext(context);

  1519.         context = NULL;

  1520.     }

  1521. 

  1522.     if (!context) {

  1523.         if (!(context = sws_alloc_context()))

  1524.             return NULL;

  1525.         context->srcW      = srcW;

  1526.         context->srcH      = srcH;

  1527.         context->srcRange  = handle_jpeg(&srcFormat);

  1528.         context->srcFormat = srcFormat;

  1529.         context->dstW      = dstW;

  1530.         context->dstH      = dstH;

  1531.         context->dstRange  = handle_jpeg(&dstFormat);

  1532.         context->dstFormat = dstFormat;

  1533.         context->flags     = flags;

  1534.         context->param[0]  = param[0];

  1535.         context->param[1]  = param[1];

  1536.         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);

  1537.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1538.             sws_freeContext(context);

  1539.             return NULL;

  1540.         }

  1541.     }

  1542.     return context;

  1543. }