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.
31653 Commits
32 Branches
484MB
Tree: f234cd0f37
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'f234cd0f37'
${ noResults }
FFmpeg/libswscale/utils.c

1570 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. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  144. {

  145.     return format_entries[pix_fmt].is_supported_in;

  146. }

  147. 

  148. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  149. {

  150.     return format_entries[pix_fmt].is_supported_out;

  151. }

  152. 

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

  154. 

  155. #if FF_API_SWS_FORMAT_NAME

  156. const char *sws_format_name(enum PixelFormat format)

  157. {

  158.     return av_get_pix_fmt_name(format);

  159. }

  160. #endif

  161. 

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

  163. {

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

  165.     else           return getSplineCoeff(        0.0,

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

  167.                                                  c + 3.0*d,

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

  169.                                          dist-1.0);

  170. }

  171. 

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

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

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

  175. {

  176.     int i;

  177.     int filterSize;

  178.     int filter2Size;

  179.     int minFilterSize;

  180.     int64_t *filter=NULL;

  181.     int64_t *filter2=NULL;

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

  183.     int ret= -1;

  184. 

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

  186. 

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

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

  189. 

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

  191.         int i;

  192.         filterSize= 1;

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

  194. 

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

  196.             filter[i*filterSize]= fone;

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

  198.         }

  199. 

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

  201.         int i;

  202.         int xDstInSrc;

  203.         filterSize= 1;

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

  205. 

  206.         xDstInSrc= xInc/2 - 0x8000;

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

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

  209. 

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

  211.             filter[i]= fone;

  212.             xDstInSrc+= xInc;

  213.         }

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

  215.         int i;

  216.         int xDstInSrc;

  217.         filterSize= 2;

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

  219. 

  220.         xDstInSrc= xInc/2 - 0x8000;

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

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

  223.             int j;

  224. 

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

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

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

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

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

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

  231.                 xx++;

  232.             }

  233.             xDstInSrc+= xInc;

  234.         }

  235.     } else {

  236.         int xDstInSrc;

  237.         int sizeFactor;

  238. 

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

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

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

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

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

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

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

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

  247.         else {

  248.             sizeFactor= 0; //GCC warning killer

  249.             assert(0);

  250.         }

  251. 

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

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

  254. 

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

  256. 

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

  258. 

  259.         xDstInSrc= xInc - 0x10000;

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

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

  262.             int j;

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

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

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

  266.                 double floatd;

  267.                 int64_t coeff;

  268. 

  269.                 if (xInc > 1<<16)

  270.                     d= d*dstW/srcW;

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

  272. 

  273.                 if (flags & SWS_BICUBIC) {

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

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

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

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

  278. 

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

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

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

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

  283.                     else

  284.                         coeff=0.0;

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

  286.                 }

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

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

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

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

  291.                 }*/

  292.                 else if (flags & SWS_X) {

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

  294.                     double c;

  295. 

  296.                     if (floatd<1.0)

  297.                         c = cos(floatd*M_PI);

  298.                     else

  299.                         c=-1.0;

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

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

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

  303.                 } else if (flags & SWS_AREA) {

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

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

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

  307.                     else coeff=0.0;

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

  309.                 } else if (flags & SWS_GAUSS) {

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

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

  312.                 } else if (flags & SWS_SINC) {

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

  314.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  318.                 } else if (flags & SWS_BILINEAR) {

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

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

  321.                     coeff *= fone >> 30;

  322.                 } else if (flags & SWS_SPLINE) {

  323.                     double p=-2.196152422706632;

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

  325.                 } else {

  326.                     coeff= 0.0; //GCC warning killer

  327.                     assert(0);

  328.                 }

  329. 

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

  331.                 xx++;

  332.             }

  333.             xDstInSrc+= 2*xInc;

  334.         }

  335.     }

  336. 

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

  338.        av_free(filter);

  339.     */

  340.     assert(filterSize>0);

  341.     filter2Size= filterSize;

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

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

  344.     assert(filter2Size>0);

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

  346. 

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

  348.         int j, k;

  349. 

  350.         if(srcFilter) {

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

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

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

  354.             }

  355.         } else {

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

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

  358.         }

  359.         //FIXME dstFilter

  360. 

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

  362.     }

  363.     av_freep(&filter);

  364. 

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

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

  367.     minFilterSize= 0;

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

  369.         int min= filter2Size;

  370.         int j;

  371.         int64_t cutOff=0.0;

  372. 

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

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

  375.             int k;

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

  377. 

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

  379. 

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

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

  382. 

  383.             // move filter coefficients left

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

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

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

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

  388.         }

  389. 

  390.         cutOff=0;

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

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

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

  394. 

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

  396.             min--;

  397.         }

  398. 

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

  400.     }

  401. 

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

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

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

  405.         if (minFilterSize < 5)

  406.             filterAlign = 4;

  407. 

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

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

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

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

  412.         // decent-sized vector.

  413.         if (minFilterSize < 3)

  414.             filterAlign = 1;

  415.     }

  416. 

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

  418.         // special case for unscaled vertical filtering

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

  420.             filterAlign= 1;

  421.     }

  422. 

  423.     assert(minFilterSize > 0);

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

  425.     assert(filterSize > 0);

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

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

  428.         goto fail;

  429.     *outFilterSize= filterSize;

  430. 

  431.     if (flags&SWS_PRINT_INFO)

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

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

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

  435.         int j;

  436. 

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

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

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

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

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

  442.         }

  443.     }

  444. 

  445.     //FIXME try to align filterPos if possible

  446. 

  447.     //fix borders

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

  449.         int j;

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

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

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

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

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

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

  456.             }

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

  458.         }

  459. 

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

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

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

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

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

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

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

  467.             }

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

  469.         }

  470.     }

  471. 

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

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

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

  475. 

  476.     /* normalize & store in outFilter */

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

  478.         int j;

  479.         int64_t error=0;

  480.         int64_t sum=0;

  481. 

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

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

  484.         }

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

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

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

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

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

  490.             error= v - intV*sum;

  491.         }

  492.     }

  493. 

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

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

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

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

  498.     }

  499. 

  500.     ret=0;

  501. fail:

  502.     av_free(filter);

  503.     av_free(filter2);

  504.     return ret;

  505. }

  506. 

  507. #if HAVE_MMX2

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

  509. {

  510.     uint8_t *fragmentA;

  511.     x86_reg imm8OfPShufW1A;

  512.     x86_reg imm8OfPShufW2A;

  513.     x86_reg fragmentLengthA;

  514.     uint8_t *fragmentB;

  515.     x86_reg imm8OfPShufW1B;

  516.     x86_reg imm8OfPShufW2B;

  517.     x86_reg fragmentLengthB;

  518.     int fragmentPos;

  519. 

  520.     int xpos, i;

  521. 

  522.     // create an optimized horizontal scaling routine

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

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

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

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

  527.      * of fragmentA is used.

  528.      */

  529. 

  530.     //code fragment

  531. 

  532.     __asm__ volatile(

  533.         "jmp                         9f                 \n\t"

  534.     // Begin

  535.         "0:                                             \n\t"

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

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

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

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

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

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

  542.         "1:                                             \n\t"

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

  544.         "2:                                             \n\t"

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

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

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

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

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

  550. 

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

  552. 

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

  554.     // End

  555.         "9:                                             \n\t"

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

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

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

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

  560.         "dec                         %1                 \n\t"

  561.         "dec                         %2                 \n\t"

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

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

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

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

  566. 

  567. 

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

  569.         "=r" (fragmentLengthA)

  570.     );

  571. 

  572.     __asm__ volatile(

  573.         "jmp                         9f                 \n\t"

  574.     // Begin

  575.         "0:                                             \n\t"

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

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

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

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

  580.         "1:                                             \n\t"

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

  582.         "2:                                             \n\t"

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

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

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

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

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

  588. 

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

  590. 

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

  592.     // End

  593.         "9:                                             \n\t"

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

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

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

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

  598.         "dec                         %1                 \n\t"

  599.         "dec                         %2                 \n\t"

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

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

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

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

  604. 

  605. 

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

  607.         "=r" (fragmentLengthB)

  608.     );

  609. 

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

  611.     fragmentPos=0;

  612. 

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

  614.         int xx=xpos>>16;

  615. 

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

  617.             int a=0;

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

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

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

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

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

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

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

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

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

  627.             int shift=0;

  628. 

  629.             if (filterCode) {

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

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

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

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

  634.                 filterPos[i/2]= xx;

  635. 

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

  637. 

  638.                 filterCode[fragmentPos + imm8OfPShufW1]=

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

  640.                 filterCode[fragmentPos + imm8OfPShufW2]=

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

  642. 

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

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

  645. 

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

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

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

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

  650.                 }

  651.             }

  652. 

  653.             fragmentPos+= fragmentLength;

  654. 

  655.             if (filterCode)

  656.                 filterCode[fragmentPos]= RET;

  657.         }

  658.         xpos+=xInc;

  659.     }

  660.     if (filterCode)

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

  662. 

  663.     return fragmentPos + 1;

  664. }

  665. #endif /* HAVE_MMX2 */

  666. 

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

  668. {

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

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

  671. }

  672. 

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

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

  675.                              int brightness, int contrast, int saturation)

  676. {

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

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

  679. 

  680.     c->brightness= brightness;

  681.     c->contrast  = contrast;

  682.     c->saturation= saturation;

  683.     c->srcRange  = srcRange;

  684.     c->dstRange  = dstRange;

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

  686. 

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

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

  689. 

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

  691.     //FIXME factorize

  692. 

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

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

  695.     return 0;

  696. }

  697. 

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

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

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

  701. {

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

  703. 

  704.     *inv_table = c->srcColorspaceTable;

  705.     *table     = c->dstColorspaceTable;

  706.     *srcRange  = c->srcRange;

  707.     *dstRange  = c->dstRange;

  708.     *brightness= c->brightness;

  709.     *contrast  = c->contrast;

  710.     *saturation= c->saturation;

  711. 

  712.     return 0;

  713. }

  714. 

  715. static int handle_jpeg(enum PixelFormat *format)

  716. {

  717.     switch (*format) {

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

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

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

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

  722.     default:                                          return 0;

  723.     }

  724. }

  725. 

  726. SwsContext *sws_alloc_context(void)

  727. {

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

  729. 

  730.     c->av_class = &sws_context_class;

  731.     av_opt_set_defaults(c);

  732. 

  733.     return c;

  734. }

  735. 

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

  737. {

  738.     int i, j;

  739.     int usesVFilter, usesHFilter;

  740.     int unscaled;

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

  742.     int srcW= c->srcW;

  743.     int srcH= c->srcH;

  744.     int dstW= c->dstW;

  745.     int dstH= c->dstH;

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

  747.     int flags, cpu_flags;

  748.     enum PixelFormat srcFormat= c->srcFormat;

  749.     enum PixelFormat dstFormat= c->dstFormat;

  750. 

  751.     cpu_flags = av_get_cpu_flags();

  752.     flags     = c->flags;

  753.     emms_c();

  754.     if (!rgb15to16) sws_rgb2rgb_init();

  755. 

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

  757. 

  758.     if (!sws_isSupportedInput(srcFormat)) {

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

  760.         return AVERROR(EINVAL);

  761.     }

  762.     if (!sws_isSupportedOutput(dstFormat)) {

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

  764.         return AVERROR(EINVAL);

  765.     }

  766. 

  767.     i= flags & ( SWS_POINT

  768.                 |SWS_AREA

  769.                 |SWS_BILINEAR

  770.                 |SWS_FAST_BILINEAR

  771.                 |SWS_BICUBIC

  772.                 |SWS_X

  773.                 |SWS_GAUSS

  774.                 |SWS_LANCZOS

  775.                 |SWS_SINC

  776.                 |SWS_SPLINE

  777.                 |SWS_BICUBLIN);

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

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

  780.         return AVERROR(EINVAL);

  781.     }

  782.     /* sanity check */

  783.     if (srcW<4 || srcH<1 || dstW<8 || dstH<1) { //FIXME check if these are enough and try to lowwer them after fixing the relevant parts of the code

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

  785.                srcW, srcH, dstW, dstH);

  786.         return AVERROR(EINVAL);

  787.     }

  788. 

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

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

  791. 

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

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

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

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

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

  797. 

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

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

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

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

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

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

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

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

  806. 

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

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

  809. 

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

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

  812. 

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

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

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

  816. 

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

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

  819.       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8

  820.       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4

  821.       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE

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

  823.         c->chrSrcHSubSample=1;

  824. 

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

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

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

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

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

  830. 

  831.     /* unscaled special cases */

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

  833.         ff_get_unscaled_swscale(c);

  834. 

  835.         if (c->swScale) {

  836.             if (flags&SWS_PRINT_INFO)

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

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

  839.             return 0;

  840.         }

  841.     }

  842. 

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

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

  845. 

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

  847.         dst_stride <<= 1;

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

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

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

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

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

  853.             if (flags&SWS_PRINT_INFO)

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

  855.         }

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

  857.     }

  858.     else

  859.         c->canMMX2BeUsed=0;

  860. 

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

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

  863. 

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

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

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

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

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

  869.     // first and last pixel

  870.     if (flags&SWS_FAST_BILINEAR) {

  871.         if (c->canMMX2BeUsed) {

  872.             c->lumXInc+= 20;

  873.             c->chrXInc+= 20;

  874.         }

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

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

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

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

  879.         }

  880.     }

  881. 

  882.     /* precalculate horizontal scaler filter coefficients */

  883.     {

  884. #if HAVE_MMX2

  885. // can't downscale !!!

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

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

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

  889. 

  890. #ifdef MAP_ANONYMOUS

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

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

  893. #elif HAVE_VIRTUALALLOC

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

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

  896. #else

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

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

  899. #endif

  900. 

  901. #ifdef MAP_ANONYMOUS

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

  903. #else

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

  905. #endif

  906.                 return AVERROR(ENOMEM);

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

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

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

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

  911. 

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

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

  914. 

  915. #ifdef MAP_ANONYMOUS

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

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

  918. #endif

  919.         } else

  920. #endif /* HAVE_MMX2 */

  921.         {

  922.             const int filterAlign=

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

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

  925.                 1;

  926. 

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

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

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

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

  931.                 goto fail;

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

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

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

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

  936.                 goto fail;

  937.         }

  938.     } // initialize horizontal stuff

  939. 

  940.     /* precalculate vertical scaler filter coefficients */

  941.     {

  942.         const int filterAlign=

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

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

  945.             1;

  946. 

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

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

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

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

  951.             goto fail;

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

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

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

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

  956.             goto fail;

  957. 

  958. #if HAVE_ALTIVEC

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

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

  961. 

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

  963.             int j;

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

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

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

  967.         }

  968. 

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

  970.             int j;

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

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

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

  974.         }

  975. #endif

  976.     }

  977. 

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

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

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

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

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

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

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

  985. 

  986.         nextSlice>>= c->chrSrcVSubSample;

  987.         nextSlice<<= c->chrSrcVSubSample;

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

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

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

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

  992.     }

  993. 

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

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

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

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

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

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

  1000.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1001.     //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)

  1002.     /* align at 16 bytes for AltiVec */

  1003.     for (i=0; i<c->vLumBufSize; i++) {

  1004.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1005.         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];

  1006.     }

  1007.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1008.     c->uv_off   = (dst_stride>>1) + 64 / c->scalingBpp;

  1009.     c->uv_offx2 = dst_stride + 16;

  1010.     for (i=0; i<c->vChrBufSize; i++) {

  1011.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);

  1012.         c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];

  1013.         c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1014.     }

  1015.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

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

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

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

  1019.         }

  1020. 

  1021.     //try to avoid drawing green stuff between the right end and the stride end

  1022.     for (i=0; i<c->vChrBufSize; i++)

  1023.         if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){

  1024.             av_assert0(c->scalingBpp == 16);

  1025.             for(j=0; j<dst_stride/2+1; j++)

  1026.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1027.         } else

  1028.             for(j=0; j<dst_stride+1; j++)

  1029.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1030. 

  1031.     assert(c->chrDstH <= dstH);

  1032. 

  1033.     if (flags&SWS_PRINT_INFO) {

  1034.         if      (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1035.         else if (flags&SWS_BILINEAR)      av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1036.         else if (flags&SWS_BICUBIC)       av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1037.         else if (flags&SWS_X)             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1038.         else if (flags&SWS_POINT)         av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1039.         else if (flags&SWS_AREA)          av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1040.         else if (flags&SWS_BICUBLIN)      av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1041.         else if (flags&SWS_GAUSS)         av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1042.         else if (flags&SWS_SINC)          av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1043.         else if (flags&SWS_LANCZOS)       av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1044.         else if (flags&SWS_SPLINE)        av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1045.         else                              av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1046. 

  1047.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1048.                av_get_pix_fmt_name(srcFormat),

  1049. #ifdef DITHER1XBPP

  1050.                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||

  1051.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1052.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",

  1053. #else

  1054.                "",

  1055. #endif

  1056.                av_get_pix_fmt_name(dstFormat));

  1057. 

  1058.         if      (HAVE_MMX2     && cpu_flags & AV_CPU_FLAG_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");

  1059.         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1060.         else if (HAVE_MMX      && cpu_flags & AV_CPU_FLAG_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");

  1061.         else if (HAVE_ALTIVEC  && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1062.         else                                   av_log(c, AV_LOG_INFO, "using C\n");

  1063. 

  1064.         if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {

  1065.             if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))

  1066.                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");

  1067.             else {

  1068.                 av_log(c, AV_LOG_VERBOSE, "using %s-tap MMX scaler for horizontal luminance scaling\n",

  1069.                        c->hLumFilterSize == 4 ? "4" :

  1070.                        c->hLumFilterSize == 8 ? "8" : "n");

  1071.                 av_log(c, AV_LOG_VERBOSE, "using %s-tap MMX scaler for horizontal chrominance scaling\n",

  1072.                        c->hChrFilterSize == 4 ? "4" :

  1073.                        c->hChrFilterSize == 8 ? "8" : "n");

  1074.             }

  1075.         } else {

  1076.             av_log(c, AV_LOG_VERBOSE, "using %s scaler for horizontal scaling\n",

  1077.                    HAVE_MMX ? "x86 asm" :

  1078.                    flags & SWS_FAST_BILINEAR ? "FAST_BILINEAR C" : "C");

  1079.         }

  1080.         if (isPlanarYUV(dstFormat)) {

  1081.             if (c->vLumFilterSize==1)

  1082.                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n",

  1083.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1084.             else

  1085.                 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n",

  1086.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1087.         } else {

  1088.             if (c->vLumFilterSize==1 && c->vChrFilterSize==2)

  1089.                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"

  1090.                        "      2-tap scaler for vertical chrominance scaling (BGR)\n",

  1091.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

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

  1093.                 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n",

  1094.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1095.             else

  1096.                 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n",

  1097.                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1098.         }

  1099. 

  1100.         if (dstFormat==PIX_FMT_BGR24)

  1101.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",

  1102.                    (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) ? "MMX2" :

  1103.                    ((HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"));

  1104.         else if (dstFormat==PIX_FMT_RGB32)

  1105.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n",

  1106.                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1107.         else if (dstFormat==PIX_FMT_BGR565)

  1108.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n",

  1109.                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1110.         else if (dstFormat==PIX_FMT_BGR555)

  1111.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n",

  1112.                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");

  1113.         else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1114.                  dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)

  1115.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n",

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

  1117. 

  1118.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1119.         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1120.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1121.         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1122.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1123.     }

  1124. 

  1125.     c->swScale= ff_getSwsFunc(c);

  1126.     return 0;

  1127. fail: //FIXME replace things by appropriate error codes

  1128.     return -1;

  1129. }

  1130. 

  1131. #if FF_API_SWS_GETCONTEXT

  1132. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1133.                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1134.                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1135. {

  1136.     SwsContext *c;

  1137. 

  1138.     if(!(c=sws_alloc_context()))

  1139.         return NULL;

  1140. 

  1141.     c->flags= flags;

  1142.     c->srcW= srcW;

  1143.     c->srcH= srcH;

  1144.     c->dstW= dstW;

  1145.     c->dstH= dstH;

  1146.     c->srcRange = handle_jpeg(&srcFormat);

  1147.     c->dstRange = handle_jpeg(&dstFormat);

  1148.     c->srcFormat= srcFormat;

  1149.     c->dstFormat= dstFormat;

  1150. 

  1151.     if (param) {

  1152.         c->param[0] = param[0];

  1153.         c->param[1] = param[1];

  1154.     }

  1155.     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);

  1156. 

  1157.     if(sws_init_context(c, srcFilter, dstFilter) < 0){

  1158.         sws_freeContext(c);

  1159.         return NULL;

  1160.     }

  1161. 

  1162.     return c;

  1163. }

  1164. #endif

  1165. 

  1166. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1167.                                 float lumaSharpen, float chromaSharpen,

  1168.                                 float chromaHShift, float chromaVShift,

  1169.                                 int verbose)

  1170. {

  1171.     SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  1172.     if (!filter)

  1173.         return NULL;

  1174. 

  1175.     if (lumaGBlur!=0.0) {

  1176.         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  1177.         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  1178.     } else {

  1179.         filter->lumH= sws_getIdentityVec();

  1180.         filter->lumV= sws_getIdentityVec();

  1181.     }

  1182. 

  1183.     if (chromaGBlur!=0.0) {

  1184.         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  1185.         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  1186.     } else {

  1187.         filter->chrH= sws_getIdentityVec();

  1188.         filter->chrV= sws_getIdentityVec();

  1189.     }

  1190. 

  1191.     if (chromaSharpen!=0.0) {

  1192.         SwsVector *id= sws_getIdentityVec();

  1193.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1194.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1195.         sws_addVec(filter->chrH, id);

  1196.         sws_addVec(filter->chrV, id);

  1197.         sws_freeVec(id);

  1198.     }

  1199. 

  1200.     if (lumaSharpen!=0.0) {

  1201.         SwsVector *id= sws_getIdentityVec();

  1202.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1203.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1204.         sws_addVec(filter->lumH, id);

  1205.         sws_addVec(filter->lumV, id);

  1206.         sws_freeVec(id);

  1207.     }

  1208. 

  1209.     if (chromaHShift != 0.0)

  1210.         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  1211. 

  1212.     if (chromaVShift != 0.0)

  1213.         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  1214. 

  1215.     sws_normalizeVec(filter->chrH, 1.0);

  1216.     sws_normalizeVec(filter->chrV, 1.0);

  1217.     sws_normalizeVec(filter->lumH, 1.0);

  1218.     sws_normalizeVec(filter->lumV, 1.0);

  1219. 

  1220.     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1221.     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1222. 

  1223.     return filter;

  1224. }

  1225. 

  1226. SwsVector *sws_allocVec(int length)

  1227. {

  1228.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1229.     if (!vec)

  1230.         return NULL;

  1231.     vec->length = length;

  1232.     vec->coeff  = av_malloc(sizeof(double) * length);

  1233.     if (!vec->coeff)

  1234.         av_freep(&vec);

  1235.     return vec;

  1236. }

  1237. 

  1238. SwsVector *sws_getGaussianVec(double variance, double quality)

  1239. {

  1240.     const int length= (int)(variance*quality + 0.5) | 1;

  1241.     int i;

  1242.     double middle= (length-1)*0.5;

  1243.     SwsVector *vec= sws_allocVec(length);

  1244. 

  1245.     if (!vec)

  1246.         return NULL;

  1247. 

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

  1249.         double dist= i-middle;

  1250.         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);

  1251.     }

  1252. 

  1253.     sws_normalizeVec(vec, 1.0);

  1254. 

  1255.     return vec;

  1256. }

  1257. 

  1258. SwsVector *sws_getConstVec(double c, int length)

  1259. {

  1260.     int i;

  1261.     SwsVector *vec= sws_allocVec(length);

  1262. 

  1263.     if (!vec)

  1264.         return NULL;

  1265. 

  1266.     for (i=0; i<length; i++)

  1267.         vec->coeff[i]= c;

  1268. 

  1269.     return vec;

  1270. }

  1271. 

  1272. SwsVector *sws_getIdentityVec(void)

  1273. {

  1274.     return sws_getConstVec(1.0, 1);

  1275. }

  1276. 

  1277. static double sws_dcVec(SwsVector *a)

  1278. {

  1279.     int i;

  1280.     double sum=0;

  1281. 

  1282.     for (i=0; i<a->length; i++)

  1283.         sum+= a->coeff[i];

  1284. 

  1285.     return sum;

  1286. }

  1287. 

  1288. void sws_scaleVec(SwsVector *a, double scalar)

  1289. {

  1290.     int i;

  1291. 

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

  1293.         a->coeff[i]*= scalar;

  1294. }

  1295. 

  1296. void sws_normalizeVec(SwsVector *a, double height)

  1297. {

  1298.     sws_scaleVec(a, height/sws_dcVec(a));

  1299. }

  1300. 

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

  1302. {

  1303.     int length= a->length + b->length - 1;

  1304.     int i, j;

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

  1306. 

  1307.     if (!vec)

  1308.         return NULL;

  1309. 

  1310.     for (i=0; i<a->length; i++) {

  1311.         for (j=0; j<b->length; j++) {

  1312.             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1313.         }

  1314.     }

  1315. 

  1316.     return vec;

  1317. }

  1318. 

  1319. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1320. {

  1321.     int length= FFMAX(a->length, b->length);

  1322.     int i;

  1323.     SwsVector *vec= sws_getConstVec(0.0, length);

  1324. 

  1325.     if (!vec)

  1326.         return NULL;

  1327. 

  1328.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1329.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];

  1330. 

  1331.     return vec;

  1332. }

  1333. 

  1334. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1335. {

  1336.     int length= FFMAX(a->length, b->length);

  1337.     int i;

  1338.     SwsVector *vec= sws_getConstVec(0.0, length);

  1339. 

  1340.     if (!vec)

  1341.         return NULL;

  1342. 

  1343.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1344.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1345. 

  1346.     return vec;

  1347. }

  1348. 

  1349. /* shift left / or right if "shift" is negative */

  1350. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1351. {

  1352.     int length= a->length + FFABS(shift)*2;

  1353.     int i;

  1354.     SwsVector *vec= sws_getConstVec(0.0, length);

  1355. 

  1356.     if (!vec)

  1357.         return NULL;

  1358. 

  1359.     for (i=0; i<a->length; i++) {

  1360.         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1361.     }

  1362. 

  1363.     return vec;

  1364. }

  1365. 

  1366. void sws_shiftVec(SwsVector *a, int shift)

  1367. {

  1368.     SwsVector *shifted= sws_getShiftedVec(a, shift);

  1369.     av_free(a->coeff);

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

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

  1372.     av_free(shifted);

  1373. }

  1374. 

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

  1376. {

  1377.     SwsVector *sum= sws_sumVec(a, b);

  1378.     av_free(a->coeff);

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

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

  1381.     av_free(sum);

  1382. }

  1383. 

  1384. void sws_subVec(SwsVector *a, SwsVector *b)

  1385. {

  1386.     SwsVector *diff= sws_diffVec(a, b);

  1387.     av_free(a->coeff);

  1388.     a->coeff= diff->coeff;

  1389.     a->length= diff->length;

  1390.     av_free(diff);

  1391. }

  1392. 

  1393. void sws_convVec(SwsVector *a, SwsVector *b)

  1394. {

  1395.     SwsVector *conv= sws_getConvVec(a, b);

  1396.     av_free(a->coeff);

  1397.     a->coeff= conv->coeff;

  1398.     a->length= conv->length;

  1399.     av_free(conv);

  1400. }

  1401. 

  1402. SwsVector *sws_cloneVec(SwsVector *a)

  1403. {

  1404.     int i;

  1405.     SwsVector *vec= sws_allocVec(a->length);

  1406. 

  1407.     if (!vec)

  1408.         return NULL;

  1409. 

  1410.     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];

  1411. 

  1412.     return vec;

  1413. }

  1414. 

  1415. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1416. {

  1417.     int i;

  1418.     double max=0;

  1419.     double min=0;

  1420.     double range;

  1421. 

  1422.     for (i=0; i<a->length; i++)

  1423.         if (a->coeff[i]>max) max= a->coeff[i];

  1424. 

  1425.     for (i=0; i<a->length; i++)

  1426.         if (a->coeff[i]<min) min= a->coeff[i];

  1427. 

  1428.     range= max - min;

  1429. 

  1430.     for (i=0; i<a->length; i++) {

  1431.         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1432.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1433.         for (;x>0; x--) av_log(log_ctx, log_level, " ");

  1434.         av_log(log_ctx, log_level, "|\n");

  1435.     }

  1436. }

  1437. 

  1438. #if LIBSWSCALE_VERSION_MAJOR < 1

  1439. void sws_printVec(SwsVector *a)

  1440. {

  1441.     sws_printVec2(a, NULL, AV_LOG_DEBUG);

  1442. }

  1443. #endif

  1444. 

  1445. void sws_freeVec(SwsVector *a)

  1446. {

  1447.     if (!a) return;

  1448.     av_freep(&a->coeff);

  1449.     a->length=0;

  1450.     av_free(a);

  1451. }

  1452. 

  1453. void sws_freeFilter(SwsFilter *filter)

  1454. {

  1455.     if (!filter) return;

  1456. 

  1457.     if (filter->lumH) sws_freeVec(filter->lumH);

  1458.     if (filter->lumV) sws_freeVec(filter->lumV);

  1459.     if (filter->chrH) sws_freeVec(filter->chrH);

  1460.     if (filter->chrV) sws_freeVec(filter->chrV);

  1461.     av_free(filter);

  1462. }

  1463. 

  1464. void sws_freeContext(SwsContext *c)

  1465. {

  1466.     int i;

  1467.     if (!c) return;

  1468. 

  1469.     if (c->lumPixBuf) {

  1470.         for (i=0; i<c->vLumBufSize; i++)

  1471.             av_freep(&c->lumPixBuf[i]);

  1472.         av_freep(&c->lumPixBuf);

  1473.     }

  1474. 

  1475.     if (c->chrUPixBuf) {

  1476.         for (i=0; i<c->vChrBufSize; i++)

  1477.             av_freep(&c->chrUPixBuf[i]);

  1478.         av_freep(&c->chrUPixBuf);

  1479.         av_freep(&c->chrVPixBuf);

  1480.     }

  1481. 

  1482.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1483.         for (i=0; i<c->vLumBufSize; i++)

  1484.             av_freep(&c->alpPixBuf[i]);

  1485.         av_freep(&c->alpPixBuf);

  1486.     }

  1487. 

  1488.     av_freep(&c->vLumFilter);

  1489.     av_freep(&c->vChrFilter);

  1490.     av_freep(&c->hLumFilter);

  1491.     av_freep(&c->hChrFilter);

  1492. #if HAVE_ALTIVEC

  1493.     av_freep(&c->vYCoeffsBank);

  1494.     av_freep(&c->vCCoeffsBank);

  1495. #endif

  1496. 

  1497.     av_freep(&c->vLumFilterPos);

  1498.     av_freep(&c->vChrFilterPos);

  1499.     av_freep(&c->hLumFilterPos);

  1500.     av_freep(&c->hChrFilterPos);

  1501. 

  1502. #if HAVE_MMX

  1503. #ifdef MAP_ANONYMOUS

  1504.     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1505.     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1506. #elif HAVE_VIRTUALALLOC

  1507.     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1508.     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1509. #else

  1510.     av_free(c->lumMmx2FilterCode);

  1511.     av_free(c->chrMmx2FilterCode);

  1512. #endif

  1513.     c->lumMmx2FilterCode=NULL;

  1514.     c->chrMmx2FilterCode=NULL;

  1515. #endif /* HAVE_MMX */

  1516. 

  1517.     av_freep(&c->yuvTable);

  1518.     av_freep(&c->formatConvBuffer);

  1519. 

  1520.     av_free(c);

  1521. }

  1522. 

  1523. struct SwsContext *sws_getCachedContext(struct SwsContext *context,

  1524.                                         int srcW, int srcH, enum PixelFormat srcFormat,

  1525.                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1526.                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1527. {

  1528.     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};

  1529. 

  1530.     if (!param)

  1531.         param = default_param;

  1532. 

  1533.     if (context &&

  1534.         (context->srcW      != srcW      ||

  1535.          context->srcH      != srcH      ||

  1536.          context->srcFormat != srcFormat ||

  1537.          context->dstW      != dstW      ||

  1538.          context->dstH      != dstH      ||

  1539.          context->dstFormat != dstFormat ||

  1540.          context->flags     != flags     ||

  1541.          context->param[0]  != param[0]  ||

  1542.          context->param[1]  != param[1])) {

  1543.         sws_freeContext(context);

  1544.         context = NULL;

  1545.     }

  1546. 

  1547.     if (!context) {

  1548.         if (!(context = sws_alloc_context()))

  1549.             return NULL;

  1550.         context->srcW      = srcW;

  1551.         context->srcH      = srcH;

  1552.         context->srcRange  = handle_jpeg(&srcFormat);

  1553.         context->srcFormat = srcFormat;

  1554.         context->dstW      = dstW;

  1555.         context->dstH      = dstH;

  1556.         context->dstRange  = handle_jpeg(&dstFormat);

  1557.         context->dstFormat = dstFormat;

  1558.         context->flags     = flags;

  1559.         context->param[0]  = param[0];

  1560.         context->param[1]  = param[1];

  1561.         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);

  1562.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1563.             sws_freeContext(context);

  1564.             return NULL;

  1565.         }

  1566.     }

  1567.     return context;

  1568. }