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.
31644 Commits
32 Branches
552MB
Tree: 64bb656326
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '64bb656326'
${ noResults }
FFmpeg/libswscale/utils.c

1574 lines
57KB
Raw Blame History 

  1. /*

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

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. #define _SVID_SOURCE //needed for MAP_ANONYMOUS

  22. #define _DARWIN_C_SOURCE // needed for MAP_ANON

  23. #include <inttypes.h>

  24. #include <string.h>

  25. #include <math.h>

  26. #include <stdio.h>

  27. #include "config.h"

  28. #include <assert.h>

  29. #if HAVE_SYS_MMAN_H

  30. #include <sys/mman.h>

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

  32. #define MAP_ANONYMOUS MAP_ANON

  33. #endif

  34. #endif

  35. #if HAVE_VIRTUALALLOC

  36. #define WIN32_LEAN_AND_MEAN

  37. #include <windows.h>

  38. #endif

  39. #include "swscale.h"

  40. #include "swscale_internal.h"

  41. #include "rgb2rgb.h"

  42. #include "libavutil/intreadwrite.h"

  43. #include "libavutil/x86_cpu.h"

  44. #include "libavutil/cpu.h"

  45. #include "libavutil/avutil.h"

  46. #include "libavutil/bswap.h"

  47. #include "libavutil/opt.h"

  48. #include "libavutil/pixdesc.h"

  49. #include "libavutil/avassert.h"

  50. 

  51. unsigned swscale_version(void)

  52. {

  53.     return LIBSWSCALE_VERSION_INT;

  54. }

  55. 

  56. const char *swscale_configuration(void)

  57. {

  58.     return FFMPEG_CONFIGURATION;

  59. }

  60. 

  61. const char *swscale_license(void)

  62. {

  63. #define LICENSE_PREFIX "libswscale license: "

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

  65. }

  66. 

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

  68. 

  69. typedef struct FormatEntry {

  70.     int is_supported_in, is_supported_out;

  71. } FormatEntry;

  72. 

  73. const static FormatEntry format_entries[PIX_FMT_NB] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  128.     [PIX_FMT_GRAY8A]      = { 1 , 0 },

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

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

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

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

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

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

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

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

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

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

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

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

  141. };

  142. 

  143. #define isSupportedIn(x) (format_entries[x].is_supported_in)

  144. 

  145. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  146. {

  147.     return format_entries[pix_fmt].is_supported_in;

  148. }

  149. 

  150. #define isSupportedOut(x) (format_entries[x].is_supported_out)

  151. 

  152. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  153. {

  154.     return format_entries[pix_fmt].is_supported_out;

  155. }

  156. 

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

  158. 

  159. #if FF_API_SWS_FORMAT_NAME

  160. const char *sws_format_name(enum PixelFormat format)

  161. {

  162.     return av_get_pix_fmt_name(format);

  163. }

  164. #endif

  165. 

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

  167. {

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

  169.     else           return getSplineCoeff(        0.0,

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

  171.                                                  c + 3.0*d,

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

  173.                                          dist-1.0);

  174. }

  175. 

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

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

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

  179. {

  180.     int i;

  181.     int filterSize;

  182.     int filter2Size;

  183.     int minFilterSize;

  184.     int64_t *filter=NULL;

  185.     int64_t *filter2=NULL;

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

  187.     int ret= -1;

  188. 

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

  190. 

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

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

  193. 

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

  195.         int i;

  196.         filterSize= 1;

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

  198. 

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

  200.             filter[i*filterSize]= fone;

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

  202.         }

  203. 

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

  205.         int i;

  206.         int xDstInSrc;

  207.         filterSize= 1;

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

  209. 

  210.         xDstInSrc= xInc/2 - 0x8000;

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

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

  213. 

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

  215.             filter[i]= fone;

  216.             xDstInSrc+= xInc;

  217.         }

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

  219.         int i;

  220.         int xDstInSrc;

  221.         filterSize= 2;

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

  223. 

  224.         xDstInSrc= xInc/2 - 0x8000;

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

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

  227.             int j;

  228. 

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

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

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

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

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

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

  235.                 xx++;

  236.             }

  237.             xDstInSrc+= xInc;

  238.         }

  239.     } else {

  240.         int xDstInSrc;

  241.         int sizeFactor;

  242. 

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

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

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

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

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

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

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

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

  251.         else {

  252.             sizeFactor= 0; //GCC warning killer

  253.             assert(0);

  254.         }

  255. 

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

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

  258. 

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

  260. 

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

  262. 

  263.         xDstInSrc= xInc - 0x10000;

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

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

  266.             int j;

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

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

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

  270.                 double floatd;

  271.                 int64_t coeff;

  272. 

  273.                 if (xInc > 1<<16)

  274.                     d= d*dstW/srcW;

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

  276. 

  277.                 if (flags & SWS_BICUBIC) {

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

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

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

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

  282. 

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

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

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

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

  287.                     else

  288.                         coeff=0.0;

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

  290.                 }

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

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

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

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

  295.                 }*/

  296.                 else if (flags & SWS_X) {

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

  298.                     double c;

  299. 

  300.                     if (floatd<1.0)

  301.                         c = cos(floatd*M_PI);

  302.                     else

  303.                         c=-1.0;

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

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

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

  307.                 } else if (flags & SWS_AREA) {

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

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

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

  311.                     else coeff=0.0;

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

  313.                 } else if (flags & SWS_GAUSS) {

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

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

  316.                 } else if (flags & SWS_SINC) {

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

  318.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  322.                 } else if (flags & SWS_BILINEAR) {

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

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

  325.                     coeff *= fone >> 30;

  326.                 } else if (flags & SWS_SPLINE) {

  327.                     double p=-2.196152422706632;

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

  329.                 } else {

  330.                     coeff= 0.0; //GCC warning killer

  331.                     assert(0);

  332.                 }

  333. 

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

  335.                 xx++;

  336.             }

  337.             xDstInSrc+= 2*xInc;

  338.         }

  339.     }

  340. 

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

  342.        av_free(filter);

  343.     */

  344.     assert(filterSize>0);

  345.     filter2Size= filterSize;

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

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

  348.     assert(filter2Size>0);

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

  350. 

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

  352.         int j, k;

  353. 

  354.         if(srcFilter) {

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

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

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

  358.             }

  359.         } else {

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

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

  362.         }

  363.         //FIXME dstFilter

  364. 

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

  366.     }

  367.     av_freep(&filter);

  368. 

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

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

  371.     minFilterSize= 0;

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

  373.         int min= filter2Size;

  374.         int j;

  375.         int64_t cutOff=0.0;

  376. 

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

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

  379.             int k;

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

  381. 

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

  383. 

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

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

  386. 

  387.             // move filter coefficients left

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

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

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

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

  392.         }

  393. 

  394.         cutOff=0;

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

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

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

  398. 

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

  400.             min--;

  401.         }

  402. 

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

  404.     }

  405. 

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

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

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

  409.         if (minFilterSize < 5)

  410.             filterAlign = 4;

  411. 

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

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

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

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

  416.         // decent-sized vector.

  417.         if (minFilterSize < 3)

  418.             filterAlign = 1;

  419.     }

  420. 

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

  422.         // special case for unscaled vertical filtering

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

  424.             filterAlign= 1;

  425.     }

  426. 

  427.     assert(minFilterSize > 0);

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

  429.     assert(filterSize > 0);

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

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

  432.         goto fail;

  433.     *outFilterSize= filterSize;

  434. 

  435.     if (flags&SWS_PRINT_INFO)

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

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

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

  439.         int j;

  440. 

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

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

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

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

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

  446.         }

  447.     }

  448. 

  449.     //FIXME try to align filterPos if possible

  450. 

  451.     //fix borders

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

  453.         int j;

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

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

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

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

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

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

  460.             }

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

  462.         }

  463. 

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

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

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

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

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

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

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

  471.             }

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

  473.         }

  474.     }

  475. 

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

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

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

  479. 

  480.     /* normalize & store in outFilter */

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

  482.         int j;

  483.         int64_t error=0;

  484.         int64_t sum=0;

  485. 

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

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

  488.         }

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

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

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

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

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

  494.             error= v - intV*sum;

  495.         }

  496.     }

  497. 

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

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

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

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

  502.     }

  503. 

  504.     ret=0;

  505. fail:

  506.     av_free(filter);

  507.     av_free(filter2);

  508.     return ret;

  509. }

  510. 

  511. #if HAVE_MMX2

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

  513. {

  514.     uint8_t *fragmentA;

  515.     x86_reg imm8OfPShufW1A;

  516.     x86_reg imm8OfPShufW2A;

  517.     x86_reg fragmentLengthA;

  518.     uint8_t *fragmentB;

  519.     x86_reg imm8OfPShufW1B;

  520.     x86_reg imm8OfPShufW2B;

  521.     x86_reg fragmentLengthB;

  522.     int fragmentPos;

  523. 

  524.     int xpos, i;

  525. 

  526.     // create an optimized horizontal scaling routine

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

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

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

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

  531.      * of fragmentA is used.

  532.      */

  533. 

  534.     //code fragment

  535. 

  536.     __asm__ volatile(

  537.         "jmp                         9f                 \n\t"

  538.     // Begin

  539.         "0:                                             \n\t"

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

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

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

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

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

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

  546.         "1:                                             \n\t"

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

  548.         "2:                                             \n\t"

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

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

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

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

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

  554. 

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

  556. 

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

  558.     // End

  559.         "9:                                             \n\t"

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

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

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

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

  564.         "dec                         %1                 \n\t"

  565.         "dec                         %2                 \n\t"

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

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

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

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

  570. 

  571. 

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

  573.         "=r" (fragmentLengthA)

  574.     );

  575. 

  576.     __asm__ volatile(

  577.         "jmp                         9f                 \n\t"

  578.     // Begin

  579.         "0:                                             \n\t"

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

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

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

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

  584.         "1:                                             \n\t"

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

  586.         "2:                                             \n\t"

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

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

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

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

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

  592. 

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

  594. 

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

  596.     // End

  597.         "9:                                             \n\t"

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

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

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

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

  602.         "dec                         %1                 \n\t"

  603.         "dec                         %2                 \n\t"

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

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

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

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

  608. 

  609. 

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

  611.         "=r" (fragmentLengthB)

  612.     );

  613. 

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

  615.     fragmentPos=0;

  616. 

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

  618.         int xx=xpos>>16;

  619. 

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

  621.             int a=0;

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

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

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

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

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

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

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

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

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

  631.             int shift=0;

  632. 

  633.             if (filterCode) {

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

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

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

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

  638.                 filterPos[i/2]= xx;

  639. 

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

  641. 

  642.                 filterCode[fragmentPos + imm8OfPShufW1]=

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

  644.                 filterCode[fragmentPos + imm8OfPShufW2]=

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

  646. 

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

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

  649. 

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

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

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

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

  654.                 }

  655.             }

  656. 

  657.             fragmentPos+= fragmentLength;

  658. 

  659.             if (filterCode)

  660.                 filterCode[fragmentPos]= RET;

  661.         }

  662.         xpos+=xInc;

  663.     }

  664.     if (filterCode)

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

  666. 

  667.     return fragmentPos + 1;

  668. }

  669. #endif /* HAVE_MMX2 */

  670. 

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

  672. {

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

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

  675. }

  676. 

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

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

  679.                              int brightness, int contrast, int saturation)

  680. {

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

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

  683. 

  684.     c->brightness= brightness;

  685.     c->contrast  = contrast;

  686.     c->saturation= saturation;

  687.     c->srcRange  = srcRange;

  688.     c->dstRange  = dstRange;

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

  690. 

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

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

  693. 

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

  695.     //FIXME factorize

  696. 

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

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

  699.     return 0;

  700. }

  701. 

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

  703.                              int *srcRange, int **table, int *dstRange,

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

  705. {

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

  707. 

  708.     *inv_table = c->srcColorspaceTable;

  709.     *table     = c->dstColorspaceTable;

  710.     *srcRange  = c->srcRange;

  711.     *dstRange  = c->dstRange;

  712.     *brightness= c->brightness;

  713.     *contrast  = c->contrast;

  714.     *saturation= c->saturation;

  715. 

  716.     return 0;

  717. }

  718. 

  719. static int handle_jpeg(enum PixelFormat *format)

  720. {

  721.     switch (*format) {

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

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

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

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

  726.     default:                                          return 0;

  727.     }

  728. }

  729. 

  730. SwsContext *sws_alloc_context(void)

  731. {

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

  733. 

  734.     c->av_class = &sws_context_class;

  735.     av_opt_set_defaults(c);

  736. 

  737.     return c;

  738. }

  739. 

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

  741. {

  742.     int i, j;

  743.     int usesVFilter, usesHFilter;

  744.     int unscaled;

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

  746.     int srcW= c->srcW;

  747.     int srcH= c->srcH;

  748.     int dstW= c->dstW;

  749.     int dstH= c->dstH;

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

  751.     int flags, cpu_flags;

  752.     enum PixelFormat srcFormat= c->srcFormat;

  753.     enum PixelFormat dstFormat= c->dstFormat;

  754. 

  755.     cpu_flags = av_get_cpu_flags();

  756.     flags     = c->flags;

  757.     emms_c();

  758.     if (!rgb15to16) sws_rgb2rgb_init();

  759. 

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

  761. 

  762.     if (!isSupportedIn(srcFormat)) {

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

  764.         return AVERROR(EINVAL);

  765.     }

  766.     if (!isSupportedOut(dstFormat)) {

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

  768.         return AVERROR(EINVAL);

  769.     }

  770. 

  771.     i= flags & ( SWS_POINT

  772.                 |SWS_AREA

  773.                 |SWS_BILINEAR

  774.                 |SWS_FAST_BILINEAR

  775.                 |SWS_BICUBIC

  776.                 |SWS_X

  777.                 |SWS_GAUSS

  778.                 |SWS_LANCZOS

  779.                 |SWS_SINC

  780.                 |SWS_SPLINE

  781.                 |SWS_BICUBLIN);

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

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

  784.         return AVERROR(EINVAL);

  785.     }

  786.     /* sanity check */

  787.     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

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

  789.                srcW, srcH, dstW, dstH);

  790.         return AVERROR(EINVAL);

  791.     }

  792. 

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

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

  795. 

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

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

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

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

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

  801. 

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

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

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

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

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

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

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

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

  810. 

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

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

  813. 

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

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

  816. 

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

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

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

  820. 

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

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

  823.       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8

  824.       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4

  825.       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE

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

  827.         c->chrSrcHSubSample=1;

  828. 

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

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

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

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

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

  834. 

  835.     /* unscaled special cases */

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

  837.         ff_get_unscaled_swscale(c);

  838. 

  839.         if (c->swScale) {

  840.             if (flags&SWS_PRINT_INFO)

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

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

  843.             return 0;

  844.         }

  845.     }

  846. 

  847.     c->scalingBpp = FFMAX(av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1,

  848.                           av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1) >= 15 ? 16 : 8;

  849. 

  850.     if (c->scalingBpp == 16)

  851.         dst_stride <<= 1;

  852.     av_assert0(c->scalingBpp<=16);

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

  854.     if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->scalingBpp == 8) {

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

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

  857.             if (flags&SWS_PRINT_INFO)

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

  859.         }

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

  861.     }

  862.     else

  863.         c->canMMX2BeUsed=0;

  864. 

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

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

  867. 

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

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

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

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

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

  873.     // first and last pixel

  874.     if (flags&SWS_FAST_BILINEAR) {

  875.         if (c->canMMX2BeUsed) {

  876.             c->lumXInc+= 20;

  877.             c->chrXInc+= 20;

  878.         }

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

  880.         else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX && c->scalingBpp == 8) {

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

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

  883.         }

  884.     }

  885. 

  886.     /* precalculate horizontal scaler filter coefficients */

  887.     {

  888. #if HAVE_MMX2

  889. // can't downscale !!!

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

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

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

  893. 

  894. #ifdef MAP_ANONYMOUS

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

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

  897. #elif HAVE_VIRTUALALLOC

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

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

  900. #else

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

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

  903. #endif

  904. 

  905. #ifdef MAP_ANONYMOUS

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

  907. #else

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

  909. #endif

  910.                 return AVERROR(ENOMEM);

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

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

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

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

  915. 

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

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

  918. 

  919. #ifdef MAP_ANONYMOUS

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

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

  922. #endif

  923.         } else

  924. #endif /* HAVE_MMX2 */

  925.         {

  926.             const int filterAlign=

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

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

  929.                 1;

  930. 

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

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

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

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

  935.                 goto fail;

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

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

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

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

  940.                 goto fail;

  941.         }

  942.     } // initialize horizontal stuff

  943. 

  944.     /* precalculate vertical scaler filter coefficients */

  945.     {

  946.         const int filterAlign=

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

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

  949.             1;

  950. 

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

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

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

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

  955.             goto fail;

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

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

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

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

  960.             goto fail;

  961. 

  962. #if HAVE_ALTIVEC

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

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

  965. 

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

  967.             int j;

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

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

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

  971.         }

  972. 

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

  974.             int j;

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

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

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

  978.         }

  979. #endif

  980.     }

  981. 

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

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

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

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

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

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

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

  989. 

  990.         nextSlice>>= c->chrSrcVSubSample;

  991.         nextSlice<<= c->chrSrcVSubSample;

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

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

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

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

  996.     }

  997. 

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

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

  1000.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1001.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1002.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1003.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1004.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1005.     //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)

  1006.     /* align at 16 bytes for AltiVec */

  1007.     for (i=0; i<c->vLumBufSize; i++) {

  1008.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1009.         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];

  1010.     }

  1011.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1012.     c->uv_off   = (dst_stride>>1) + 64 / c->scalingBpp;

  1013.     c->uv_offx2 = dst_stride + 16;

  1014.     for (i=0; i<c->vChrBufSize; i++) {

  1015.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);

  1016.         c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];

  1017.         c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1018.     }

  1019.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

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

  1021.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1022.             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];

  1023.         }

  1024. 

  1025.     //try to avoid drawing green stuff between the right end and the stride end

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

  1027.         if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){

  1028.             av_assert0(c->scalingBpp == 16);

  1029.             for(j=0; j<dst_stride/2+1; j++)

  1030.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1031.         } else

  1032.             for(j=0; j<dst_stride+1; j++)

  1033.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1034. 

  1035.     assert(c->chrDstH <= dstH);

  1036. 

  1037.     if (flags&SWS_PRINT_INFO) {

  1038.         if      (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1039.         else if (flags&SWS_BILINEAR)      av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1040.         else if (flags&SWS_BICUBIC)       av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1041.         else if (flags&SWS_X)             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1042.         else if (flags&SWS_POINT)         av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1043.         else if (flags&SWS_AREA)          av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1044.         else if (flags&SWS_BICUBLIN)      av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1045.         else if (flags&SWS_GAUSS)         av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1046.         else if (flags&SWS_SINC)          av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1047.         else if (flags&SWS_LANCZOS)       av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1048.         else if (flags&SWS_SPLINE)        av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1049.         else                              av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1050. 

  1051.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1052.                av_get_pix_fmt_name(srcFormat),

  1053. #ifdef DITHER1XBPP

  1054.                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||

  1055.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1056.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",

  1057. #else

  1058.                "",

  1059. #endif

  1060.                av_get_pix_fmt_name(dstFormat));

  1061. 

  1062.         if      (HAVE_MMX2     && cpu_flags & AV_CPU_FLAG_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");

  1063.         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1064.         else if (HAVE_MMX      && cpu_flags & AV_CPU_FLAG_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");

  1065.         else if (HAVE_ALTIVEC  && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1066.         else                                   av_log(c, AV_LOG_INFO, "using C\n");

  1067. 

  1068.         if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {

  1069.             if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))

  1070.                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");

  1071.             else {

  1072.                 av_log(c, AV_LOG_VERBOSE, "using %s-tap MMX scaler for horizontal luminance scaling\n",

  1073.                        c->hLumFilterSize == 4 ? "4" :

  1074.                        c->hLumFilterSize == 8 ? "8" : "n");

  1075.                 av_log(c, AV_LOG_VERBOSE, "using %s-tap MMX scaler for horizontal chrominance scaling\n",

  1076.                        c->hChrFilterSize == 4 ? "4" :

  1077.                        c->hChrFilterSize == 8 ? "8" : "n");

  1078.             }

  1079.         } else {

  1080.             av_log(c, AV_LOG_VERBOSE, "using %s scaler for horizontal scaling\n",

  1081.                    HAVE_MMX ? "x86 asm" :

  1082.                    flags & SWS_FAST_BILINEAR ? "FAST_BILINEAR C" : "C");

  1083.         }

  1084.         if (isPlanarYUV(dstFormat)) {

  1085.             if (c->vLumFilterSize==1)

  1086.                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n",

  1087.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1088.             else

  1089.                 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n",

  1090.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1091.         } else {

  1092.             if (c->vLumFilterSize==1 && c->vChrFilterSize==2)

  1093.                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"

  1094.                        "      2-tap scaler for vertical chrominance scaling (BGR)\n",

  1095.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1096.             else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)

  1097.                 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n",

  1098.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1099.             else

  1100.                 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n",

  1101.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1102.         }

  1103. 

  1104.         if (dstFormat==PIX_FMT_BGR24)

  1105.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",

  1106.                    (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) ? "MMX2" :

  1107.                    ((HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"));

  1108.         else if (dstFormat==PIX_FMT_RGB32)

  1109.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n",

  1110.                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1111.         else if (dstFormat==PIX_FMT_BGR565)

  1112.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n",

  1113.                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1114.         else if (dstFormat==PIX_FMT_BGR555)

  1115.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n",

  1116.                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1117.         else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1118.                  dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)

  1119.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n",

  1120.                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1121. 

  1122.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1123.         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1124.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1125.         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1126.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1127.     }

  1128. 

  1129.     c->swScale= ff_getSwsFunc(c);

  1130.     return 0;

  1131. fail: //FIXME replace things by appropriate error codes

  1132.     return -1;

  1133. }

  1134. 

  1135. #if FF_API_SWS_GETCONTEXT

  1136. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1137.                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1138.                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1139. {

  1140.     SwsContext *c;

  1141. 

  1142.     if(!(c=sws_alloc_context()))

  1143.         return NULL;

  1144. 

  1145.     c->flags= flags;

  1146.     c->srcW= srcW;

  1147.     c->srcH= srcH;

  1148.     c->dstW= dstW;

  1149.     c->dstH= dstH;

  1150.     c->srcRange = handle_jpeg(&srcFormat);

  1151.     c->dstRange = handle_jpeg(&dstFormat);

  1152.     c->srcFormat= srcFormat;

  1153.     c->dstFormat= dstFormat;

  1154. 

  1155.     if (param) {

  1156.         c->param[0] = param[0];

  1157.         c->param[1] = param[1];

  1158.     }

  1159.     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);

  1160. 

  1161.     if(sws_init_context(c, srcFilter, dstFilter) < 0){

  1162.         sws_freeContext(c);

  1163.         return NULL;

  1164.     }

  1165. 

  1166.     return c;

  1167. }

  1168. #endif

  1169. 

  1170. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1171.                                 float lumaSharpen, float chromaSharpen,

  1172.                                 float chromaHShift, float chromaVShift,

  1173.                                 int verbose)

  1174. {

  1175.     SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  1176.     if (!filter)

  1177.         return NULL;

  1178. 

  1179.     if (lumaGBlur!=0.0) {

  1180.         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  1181.         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  1182.     } else {

  1183.         filter->lumH= sws_getIdentityVec();

  1184.         filter->lumV= sws_getIdentityVec();

  1185.     }

  1186. 

  1187.     if (chromaGBlur!=0.0) {

  1188.         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  1189.         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  1190.     } else {

  1191.         filter->chrH= sws_getIdentityVec();

  1192.         filter->chrV= sws_getIdentityVec();

  1193.     }

  1194. 

  1195.     if (chromaSharpen!=0.0) {

  1196.         SwsVector *id= sws_getIdentityVec();

  1197.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1198.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1199.         sws_addVec(filter->chrH, id);

  1200.         sws_addVec(filter->chrV, id);

  1201.         sws_freeVec(id);

  1202.     }

  1203. 

  1204.     if (lumaSharpen!=0.0) {

  1205.         SwsVector *id= sws_getIdentityVec();

  1206.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1207.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1208.         sws_addVec(filter->lumH, id);

  1209.         sws_addVec(filter->lumV, id);

  1210.         sws_freeVec(id);

  1211.     }

  1212. 

  1213.     if (chromaHShift != 0.0)

  1214.         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  1215. 

  1216.     if (chromaVShift != 0.0)

  1217.         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  1218. 

  1219.     sws_normalizeVec(filter->chrH, 1.0);

  1220.     sws_normalizeVec(filter->chrV, 1.0);

  1221.     sws_normalizeVec(filter->lumH, 1.0);

  1222.     sws_normalizeVec(filter->lumV, 1.0);

  1223. 

  1224.     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1225.     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1226. 

  1227.     return filter;

  1228. }

  1229. 

  1230. SwsVector *sws_allocVec(int length)

  1231. {

  1232.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1233.     if (!vec)

  1234.         return NULL;

  1235.     vec->length = length;

  1236.     vec->coeff  = av_malloc(sizeof(double) * length);

  1237.     if (!vec->coeff)

  1238.         av_freep(&vec);

  1239.     return vec;

  1240. }

  1241. 

  1242. SwsVector *sws_getGaussianVec(double variance, double quality)

  1243. {

  1244.     const int length= (int)(variance*quality + 0.5) | 1;

  1245.     int i;

  1246.     double middle= (length-1)*0.5;

  1247.     SwsVector *vec= sws_allocVec(length);

  1248. 

  1249.     if (!vec)

  1250.         return NULL;

  1251. 

  1252.     for (i=0; i<length; i++) {

  1253.         double dist= i-middle;

  1254.         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);

  1255.     }

  1256. 

  1257.     sws_normalizeVec(vec, 1.0);

  1258. 

  1259.     return vec;

  1260. }

  1261. 

  1262. SwsVector *sws_getConstVec(double c, int length)

  1263. {

  1264.     int i;

  1265.     SwsVector *vec= sws_allocVec(length);

  1266. 

  1267.     if (!vec)

  1268.         return NULL;

  1269. 

  1270.     for (i=0; i<length; i++)

  1271.         vec->coeff[i]= c;

  1272. 

  1273.     return vec;

  1274. }

  1275. 

  1276. SwsVector *sws_getIdentityVec(void)

  1277. {

  1278.     return sws_getConstVec(1.0, 1);

  1279. }

  1280. 

  1281. static double sws_dcVec(SwsVector *a)

  1282. {

  1283.     int i;

  1284.     double sum=0;

  1285. 

  1286.     for (i=0; i<a->length; i++)

  1287.         sum+= a->coeff[i];

  1288. 

  1289.     return sum;

  1290. }

  1291. 

  1292. void sws_scaleVec(SwsVector *a, double scalar)

  1293. {

  1294.     int i;

  1295. 

  1296.     for (i=0; i<a->length; i++)

  1297.         a->coeff[i]*= scalar;

  1298. }

  1299. 

  1300. void sws_normalizeVec(SwsVector *a, double height)

  1301. {

  1302.     sws_scaleVec(a, height/sws_dcVec(a));

  1303. }

  1304. 

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

  1306. {

  1307.     int length= a->length + b->length - 1;

  1308.     int i, j;

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

  1310. 

  1311.     if (!vec)

  1312.         return NULL;

  1313. 

  1314.     for (i=0; i<a->length; i++) {

  1315.         for (j=0; j<b->length; j++) {

  1316.             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1317.         }

  1318.     }

  1319. 

  1320.     return vec;

  1321. }

  1322. 

  1323. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1324. {

  1325.     int length= FFMAX(a->length, b->length);

  1326.     int i;

  1327.     SwsVector *vec= sws_getConstVec(0.0, length);

  1328. 

  1329.     if (!vec)

  1330.         return NULL;

  1331. 

  1332.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1333.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];

  1334. 

  1335.     return vec;

  1336. }

  1337. 

  1338. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1339. {

  1340.     int length= FFMAX(a->length, b->length);

  1341.     int i;

  1342.     SwsVector *vec= sws_getConstVec(0.0, length);

  1343. 

  1344.     if (!vec)

  1345.         return NULL;

  1346. 

  1347.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1348.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1349. 

  1350.     return vec;

  1351. }

  1352. 

  1353. /* shift left / or right if "shift" is negative */

  1354. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1355. {

  1356.     int length= a->length + FFABS(shift)*2;

  1357.     int i;

  1358.     SwsVector *vec= sws_getConstVec(0.0, length);

  1359. 

  1360.     if (!vec)

  1361.         return NULL;

  1362. 

  1363.     for (i=0; i<a->length; i++) {

  1364.         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1365.     }

  1366. 

  1367.     return vec;

  1368. }

  1369. 

  1370. void sws_shiftVec(SwsVector *a, int shift)

  1371. {

  1372.     SwsVector *shifted= sws_getShiftedVec(a, shift);

  1373.     av_free(a->coeff);

  1374.     a->coeff= shifted->coeff;

  1375.     a->length= shifted->length;

  1376.     av_free(shifted);

  1377. }

  1378. 

  1379. void sws_addVec(SwsVector *a, SwsVector *b)

  1380. {

  1381.     SwsVector *sum= sws_sumVec(a, b);

  1382.     av_free(a->coeff);

  1383.     a->coeff= sum->coeff;

  1384.     a->length= sum->length;

  1385.     av_free(sum);

  1386. }

  1387. 

  1388. void sws_subVec(SwsVector *a, SwsVector *b)

  1389. {

  1390.     SwsVector *diff= sws_diffVec(a, b);

  1391.     av_free(a->coeff);

  1392.     a->coeff= diff->coeff;

  1393.     a->length= diff->length;

  1394.     av_free(diff);

  1395. }

  1396. 

  1397. void sws_convVec(SwsVector *a, SwsVector *b)

  1398. {

  1399.     SwsVector *conv= sws_getConvVec(a, b);

  1400.     av_free(a->coeff);

  1401.     a->coeff= conv->coeff;

  1402.     a->length= conv->length;

  1403.     av_free(conv);

  1404. }

  1405. 

  1406. SwsVector *sws_cloneVec(SwsVector *a)

  1407. {

  1408.     int i;

  1409.     SwsVector *vec= sws_allocVec(a->length);

  1410. 

  1411.     if (!vec)

  1412.         return NULL;

  1413. 

  1414.     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];

  1415. 

  1416.     return vec;

  1417. }

  1418. 

  1419. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1420. {

  1421.     int i;

  1422.     double max=0;

  1423.     double min=0;

  1424.     double range;

  1425. 

  1426.     for (i=0; i<a->length; i++)

  1427.         if (a->coeff[i]>max) max= a->coeff[i];

  1428. 

  1429.     for (i=0; i<a->length; i++)

  1430.         if (a->coeff[i]<min) min= a->coeff[i];

  1431. 

  1432.     range= max - min;

  1433. 

  1434.     for (i=0; i<a->length; i++) {

  1435.         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1436.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1437.         for (;x>0; x--) av_log(log_ctx, log_level, " ");

  1438.         av_log(log_ctx, log_level, "|\n");

  1439.     }

  1440. }

  1441. 

  1442. #if LIBSWSCALE_VERSION_MAJOR < 1

  1443. void sws_printVec(SwsVector *a)

  1444. {

  1445.     sws_printVec2(a, NULL, AV_LOG_DEBUG);

  1446. }

  1447. #endif

  1448. 

  1449. void sws_freeVec(SwsVector *a)

  1450. {

  1451.     if (!a) return;

  1452.     av_freep(&a->coeff);

  1453.     a->length=0;

  1454.     av_free(a);

  1455. }

  1456. 

  1457. void sws_freeFilter(SwsFilter *filter)

  1458. {

  1459.     if (!filter) return;

  1460. 

  1461.     if (filter->lumH) sws_freeVec(filter->lumH);

  1462.     if (filter->lumV) sws_freeVec(filter->lumV);

  1463.     if (filter->chrH) sws_freeVec(filter->chrH);

  1464.     if (filter->chrV) sws_freeVec(filter->chrV);

  1465.     av_free(filter);

  1466. }

  1467. 

  1468. void sws_freeContext(SwsContext *c)

  1469. {

  1470.     int i;

  1471.     if (!c) return;

  1472. 

  1473.     if (c->lumPixBuf) {

  1474.         for (i=0; i<c->vLumBufSize; i++)

  1475.             av_freep(&c->lumPixBuf[i]);

  1476.         av_freep(&c->lumPixBuf);

  1477.     }

  1478. 

  1479.     if (c->chrUPixBuf) {

  1480.         for (i=0; i<c->vChrBufSize; i++)

  1481.             av_freep(&c->chrUPixBuf[i]);

  1482.         av_freep(&c->chrUPixBuf);

  1483.         av_freep(&c->chrVPixBuf);

  1484.     }

  1485. 

  1486.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1487.         for (i=0; i<c->vLumBufSize; i++)

  1488.             av_freep(&c->alpPixBuf[i]);

  1489.         av_freep(&c->alpPixBuf);

  1490.     }

  1491. 

  1492.     av_freep(&c->vLumFilter);

  1493.     av_freep(&c->vChrFilter);

  1494.     av_freep(&c->hLumFilter);

  1495.     av_freep(&c->hChrFilter);

  1496. #if HAVE_ALTIVEC

  1497.     av_freep(&c->vYCoeffsBank);

  1498.     av_freep(&c->vCCoeffsBank);

  1499. #endif

  1500. 

  1501.     av_freep(&c->vLumFilterPos);

  1502.     av_freep(&c->vChrFilterPos);

  1503.     av_freep(&c->hLumFilterPos);

  1504.     av_freep(&c->hChrFilterPos);

  1505. 

  1506. #if HAVE_MMX

  1507. #ifdef MAP_ANONYMOUS

  1508.     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1509.     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1510. #elif HAVE_VIRTUALALLOC

  1511.     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1512.     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1513. #else

  1514.     av_free(c->lumMmx2FilterCode);

  1515.     av_free(c->chrMmx2FilterCode);

  1516. #endif

  1517.     c->lumMmx2FilterCode=NULL;

  1518.     c->chrMmx2FilterCode=NULL;

  1519. #endif /* HAVE_MMX */

  1520. 

  1521.     av_freep(&c->yuvTable);

  1522.     av_freep(&c->formatConvBuffer);

  1523. 

  1524.     av_free(c);

  1525. }

  1526. 

  1527. struct SwsContext *sws_getCachedContext(struct SwsContext *context,

  1528.                                         int srcW, int srcH, enum PixelFormat srcFormat,

  1529.                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1530.                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1531. {

  1532.     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};

  1533. 

  1534.     if (!param)

  1535.         param = default_param;

  1536. 

  1537.     if (context &&

  1538.         (context->srcW      != srcW      ||

  1539.          context->srcH      != srcH      ||

  1540.          context->srcFormat != srcFormat ||

  1541.          context->dstW      != dstW      ||

  1542.          context->dstH      != dstH      ||

  1543.          context->dstFormat != dstFormat ||

  1544.          context->flags     != flags     ||

  1545.          context->param[0]  != param[0]  ||

  1546.          context->param[1]  != param[1])) {

  1547.         sws_freeContext(context);

  1548.         context = NULL;

  1549.     }

  1550. 

  1551.     if (!context) {

  1552.         if (!(context = sws_alloc_context()))

  1553.             return NULL;

  1554.         context->srcW      = srcW;

  1555.         context->srcH      = srcH;

  1556.         context->srcRange  = handle_jpeg(&srcFormat);

  1557.         context->srcFormat = srcFormat;

  1558.         context->dstW      = dstW;

  1559.         context->dstH      = dstH;

  1560.         context->dstRange  = handle_jpeg(&dstFormat);

  1561.         context->dstFormat = dstFormat;

  1562.         context->flags     = flags;

  1563.         context->param[0]  = param[0];

  1564.         context->param[1]  = param[1];

  1565.         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);

  1566.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1567.             sws_freeContext(context);

  1568.             return NULL;

  1569.         }

  1570.     }

  1571.     return context;

  1572. }