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.
35286 Commits
32 Branches
552MB
Tree: b0a90c2004
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'b0a90c2004'
${ noResults }
FFmpeg/libswscale/utils.c

1553 lines
55KB
Raw Blame History 

  1. /*

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

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. #define _SVID_SOURCE //needed for MAP_ANONYMOUS

  22. #define _DARWIN_C_SOURCE // needed for MAP_ANON

  23. #include <inttypes.h>

  24. #include <string.h>

  25. #include <math.h>

  26. #include <stdio.h>

  27. #include "config.h"

  28. #include <assert.h>

  29. #if HAVE_SYS_MMAN_H

  30. #include <sys/mman.h>

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

  32. #define MAP_ANONYMOUS MAP_ANON

  33. #endif

  34. #endif

  35. #if HAVE_VIRTUALALLOC

  36. #define WIN32_LEAN_AND_MEAN

  37. #include <windows.h>

  38. #endif

  39. #include "swscale.h"

  40. #include "swscale_internal.h"

  41. #include "rgb2rgb.h"

  42. #include "libavutil/intreadwrite.h"

  43. #include "libavutil/x86_cpu.h"

  44. #include "libavutil/cpu.h"

  45. #include "libavutil/avutil.h"

  46. #include "libavutil/bswap.h"

  47. #include "libavutil/opt.h"

  48. #include "libavutil/pixdesc.h"

  49. #include "libavutil/avassert.h"

  50. 

  51. unsigned swscale_version(void)

  52. {

  53.     return LIBSWSCALE_VERSION_INT;

  54. }

  55. 

  56. const char *swscale_configuration(void)

  57. {

  58.     return FFMPEG_CONFIGURATION;

  59. }

  60. 

  61. const char *swscale_license(void)

  62. {

  63. #define LICENSE_PREFIX "libswscale license: "

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

  65. }

  66. 

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

  68. 

  69. typedef struct FormatEntry {

  70.     int is_supported_in, is_supported_out;

  71. } FormatEntry;

  72. 

  73. const static FormatEntry format_entries[PIX_FMT_NB] = {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  103.     [PIX_FMT_0RGB]        = { 1 , 1 },

  104.     [PIX_FMT_RGB0]        = { 1 , 1 },

  105.     [PIX_FMT_0BGR]        = { 1 , 1 },

  106.     [PIX_FMT_BGR0]        = { 1 , 1 },

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  159. };

  160. 

  161. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  162. {

  163.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  164.         format_entries[pix_fmt].is_supported_in : 0;

  165. }

  166. 

  167. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  168. {

  169.     return (unsigned)pix_fmt < PIX_FMT_NB ?

  170.         format_entries[pix_fmt].is_supported_out : 0;

  171. }

  172. 

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

  174. 

  175. #if FF_API_SWS_FORMAT_NAME

  176. const char *sws_format_name(enum PixelFormat format)

  177. {

  178.     return av_get_pix_fmt_name(format);

  179. }

  180. #endif

  181. 

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

  183. {

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

  185.     else           return getSplineCoeff(        0.0,

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

  187.                                                  c + 3.0*d,

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

  189.                                          dist-1.0);

  190. }

  191. 

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

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

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

  195. {

  196.     int i;

  197.     int filterSize;

  198.     int filter2Size;

  199.     int minFilterSize;

  200.     int64_t *filter=NULL;

  201.     int64_t *filter2=NULL;

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

  203.     int ret= -1;

  204. 

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

  206. 

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

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

  209. 

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

  211.         int i;

  212.         filterSize= 1;

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

  214. 

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

  216.             filter[i*filterSize]= fone;

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

  218.         }

  219. 

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

  221.         int i;

  222.         int xDstInSrc;

  223.         filterSize= 1;

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

  225. 

  226.         xDstInSrc= xInc/2 - 0x8000;

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

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

  229. 

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

  231.             filter[i]= fone;

  232.             xDstInSrc+= xInc;

  233.         }

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

  235.         int i;

  236.         int xDstInSrc;

  237.         filterSize= 2;

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

  239. 

  240.         xDstInSrc= xInc/2 - 0x8000;

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

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

  243.             int j;

  244. 

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

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

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

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

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

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

  251.                 xx++;

  252.             }

  253.             xDstInSrc+= xInc;

  254.         }

  255.     } else {

  256.         int xDstInSrc;

  257.         int sizeFactor;

  258. 

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

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

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

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

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

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

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

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

  267.         else {

  268.             sizeFactor= 0; //GCC warning killer

  269.             assert(0);

  270.         }

  271. 

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

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

  274. 

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

  276. 

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

  278. 

  279.         xDstInSrc= xInc - 0x10000;

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

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

  282.             int j;

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

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

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

  286.                 double floatd;

  287.                 int64_t coeff;

  288. 

  289.                 if (xInc > 1<<16)

  290.                     d= d*dstW/srcW;

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

  292. 

  293.                 if (flags & SWS_BICUBIC) {

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

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

  296. 

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

  298.                         coeff = 0.0;

  299.                     } else {

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

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

  302. 

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

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

  305.                         else

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

  307.                     }

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

  309.                 }

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

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

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

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

  314.                 }*/

  315.                 else if (flags & SWS_X) {

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

  317.                     double c;

  318. 

  319.                     if (floatd<1.0)

  320.                         c = cos(floatd*M_PI);

  321.                     else

  322.                         c=-1.0;

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

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

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

  326.                 } else if (flags & SWS_AREA) {

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

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

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

  330.                     else coeff=0.0;

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

  332.                 } else if (flags & SWS_GAUSS) {

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

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

  335.                 } else if (flags & SWS_SINC) {

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

  337.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  341.                 } else if (flags & SWS_BILINEAR) {

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

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

  344.                     coeff *= fone >> 30;

  345.                 } else if (flags & SWS_SPLINE) {

  346.                     double p=-2.196152422706632;

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

  348.                 } else {

  349.                     coeff= 0.0; //GCC warning killer

  350.                     assert(0);

  351.                 }

  352. 

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

  354.                 xx++;

  355.             }

  356.             xDstInSrc+= 2*xInc;

  357.         }

  358.     }

  359. 

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

  361.        av_free(filter);

  362.     */

  363.     assert(filterSize>0);

  364.     filter2Size= filterSize;

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

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

  367.     assert(filter2Size>0);

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

  369. 

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

  371.         int j, k;

  372. 

  373.         if(srcFilter) {

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

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

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

  377.             }

  378.         } else {

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

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

  381.         }

  382.         //FIXME dstFilter

  383. 

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

  385.     }

  386.     av_freep(&filter);

  387. 

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

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

  390.     minFilterSize= 0;

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

  392.         int min= filter2Size;

  393.         int j;

  394.         int64_t cutOff=0.0;

  395. 

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

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

  398.             int k;

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

  400. 

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

  402. 

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

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

  405. 

  406.             // move filter coefficients left

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

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

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

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

  411.         }

  412. 

  413.         cutOff=0;

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

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

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

  417. 

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

  419.             min--;

  420.         }

  421. 

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

  423.     }

  424. 

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

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

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

  428.         if (minFilterSize < 5)

  429.             filterAlign = 4;

  430. 

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

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

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

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

  435.         // decent-sized vector.

  436.         if (minFilterSize < 3)

  437.             filterAlign = 1;

  438.     }

  439. 

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

  441.         // special case for unscaled vertical filtering

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

  443.             filterAlign= 1;

  444.     }

  445. 

  446.     assert(minFilterSize > 0);

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

  448.     assert(filterSize > 0);

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

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

  451.         goto fail;

  452.     *outFilterSize= filterSize;

  453. 

  454.     if (flags&SWS_PRINT_INFO)

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

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

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

  458.         int j;

  459. 

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

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

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

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

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

  465.         }

  466.     }

  467. 

  468.     //FIXME try to align filterPos if possible

  469. 

  470.     //fix borders

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

  472.         int j;

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

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

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

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

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

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

  479.             }

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

  481.         }

  482. 

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

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

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

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

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

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

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

  490.             }

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

  492.         }

  493.     }

  494. 

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

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

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

  498. 

  499.     /* normalize & store in outFilter */

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

  501.         int j;

  502.         int64_t error=0;

  503.         int64_t sum=0;

  504. 

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

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

  507.         }

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

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

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

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

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

  513.             error= v - intV*sum;

  514.         }

  515.     }

  516. 

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

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

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

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

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

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

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

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

  525.     }

  526. 

  527.     ret=0;

  528. fail:

  529.     av_free(filter);

  530.     av_free(filter2);

  531.     return ret;

  532. }

  533. 

  534. #if HAVE_MMX2

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

  536. {

  537.     uint8_t *fragmentA;

  538.     x86_reg imm8OfPShufW1A;

  539.     x86_reg imm8OfPShufW2A;

  540.     x86_reg fragmentLengthA;

  541.     uint8_t *fragmentB;

  542.     x86_reg imm8OfPShufW1B;

  543.     x86_reg imm8OfPShufW2B;

  544.     x86_reg fragmentLengthB;

  545.     int fragmentPos;

  546. 

  547.     int xpos, i;

  548. 

  549.     // create an optimized horizontal scaling routine

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

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

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

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

  554.      * of fragmentA is used.

  555.      */

  556. 

  557.     //code fragment

  558. 

  559.     __asm__ volatile(

  560.         "jmp                         9f                 \n\t"

  561.     // Begin

  562.         "0:                                             \n\t"

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

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

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

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

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

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

  569.         "1:                                             \n\t"

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

  571.         "2:                                             \n\t"

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

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

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

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

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

  577. 

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

  579. 

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

  581.     // End

  582.         "9:                                             \n\t"

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

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

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

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

  587.         "dec                         %1                 \n\t"

  588.         "dec                         %2                 \n\t"

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

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

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

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

  593. 

  594. 

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

  596.         "=r" (fragmentLengthA)

  597.     );

  598. 

  599.     __asm__ volatile(

  600.         "jmp                         9f                 \n\t"

  601.     // Begin

  602.         "0:                                             \n\t"

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

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

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

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

  607.         "1:                                             \n\t"

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

  609.         "2:                                             \n\t"

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

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

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

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

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

  615. 

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

  617. 

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

  619.     // End

  620.         "9:                                             \n\t"

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

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

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

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

  625.         "dec                         %1                 \n\t"

  626.         "dec                         %2                 \n\t"

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

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

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

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

  631. 

  632. 

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

  634.         "=r" (fragmentLengthB)

  635.     );

  636. 

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

  638.     fragmentPos=0;

  639. 

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

  641.         int xx=xpos>>16;

  642. 

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

  644.             int a=0;

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

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

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

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

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

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

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

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

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

  654.             int shift=0;

  655. 

  656.             if (filterCode) {

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

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

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

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

  661.                 filterPos[i/2]= xx;

  662. 

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

  664. 

  665.                 filterCode[fragmentPos + imm8OfPShufW1]=

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

  667.                 filterCode[fragmentPos + imm8OfPShufW2]=

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

  669. 

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

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

  672. 

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

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

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

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

  677.                 }

  678.             }

  679. 

  680.             fragmentPos+= fragmentLength;

  681. 

  682.             if (filterCode)

  683.                 filterCode[fragmentPos]= RET;

  684.         }

  685.         xpos+=xInc;

  686.     }

  687.     if (filterCode)

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

  689. 

  690.     return fragmentPos + 1;

  691. }

  692. #endif /* HAVE_MMX2 */

  693. 

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

  695. {

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

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

  698. }

  699. 

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

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

  702.                              int brightness, int contrast, int saturation)

  703. {

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

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

  706. 

  707.     c->brightness= brightness;

  708.     c->contrast  = contrast;

  709.     c->saturation= saturation;

  710.     c->srcRange  = srcRange;

  711.     c->dstRange  = dstRange;

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

  713. 

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

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

  716. 

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

  718.     //FIXME factorize

  719. 

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

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

  722.     return 0;

  723. }

  724. 

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

  726.                              int *srcRange, int **table, int *dstRange,

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

  728. {

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

  730. 

  731.     *inv_table = c->srcColorspaceTable;

  732.     *table     = c->dstColorspaceTable;

  733.     *srcRange  = c->srcRange;

  734.     *dstRange  = c->dstRange;

  735.     *brightness= c->brightness;

  736.     *contrast  = c->contrast;

  737.     *saturation= c->saturation;

  738. 

  739.     return 0;

  740. }

  741. 

  742. static int handle_jpeg(enum PixelFormat *format)

  743. {

  744.     switch (*format) {

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

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

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

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

  749.     case PIX_FMT_0BGR    : *format = PIX_FMT_ABGR   ; return 0;

  750.     case PIX_FMT_BGR0    : *format = PIX_FMT_BGRA   ; return 0;

  751.     case PIX_FMT_0RGB    : *format = PIX_FMT_ARGB   ; return 0;

  752.     case PIX_FMT_RGB0    : *format = PIX_FMT_RGBA   ; return 0;

  753.     default:                                          return 0;

  754.     }

  755. }

  756. 

  757. SwsContext *sws_alloc_context(void)

  758. {

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

  760. 

  761.     c->av_class = &sws_context_class;

  762.     av_opt_set_defaults(c);

  763. 

  764.     return c;

  765. }

  766. 

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

  768. {

  769.     int i, j;

  770.     int usesVFilter, usesHFilter;

  771.     int unscaled;

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

  773.     int srcW= c->srcW;

  774.     int srcH= c->srcH;

  775.     int dstW= c->dstW;

  776.     int dstH= c->dstH;

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

  778.     int flags, cpu_flags;

  779.     enum PixelFormat srcFormat= c->srcFormat;

  780.     enum PixelFormat dstFormat= c->dstFormat;

  781. 

  782.     cpu_flags = av_get_cpu_flags();

  783.     flags     = c->flags;

  784.     emms_c();

  785.     if (!rgb15to16) sws_rgb2rgb_init();

  786. 

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

  788. 

  789.     handle_jpeg(&srcFormat);

  790.     handle_jpeg(&dstFormat);

  791. 

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

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

  794.         c->srcFormat= srcFormat;

  795.         c->dstFormat= dstFormat;

  796.     }

  797. 

  798.     if (!sws_isSupportedInput(srcFormat)) {

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

  800.         return AVERROR(EINVAL);

  801.     }

  802.     if (!sws_isSupportedOutput(dstFormat)) {

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

  804.         return AVERROR(EINVAL);

  805.     }

  806. 

  807.     i= flags & ( SWS_POINT

  808.                 |SWS_AREA

  809.                 |SWS_BILINEAR

  810.                 |SWS_FAST_BILINEAR

  811.                 |SWS_BICUBIC

  812.                 |SWS_X

  813.                 |SWS_GAUSS

  814.                 |SWS_LANCZOS

  815.                 |SWS_SINC

  816.                 |SWS_SPLINE

  817.                 |SWS_BICUBLIN);

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

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

  820.         return AVERROR(EINVAL);

  821.     }

  822.     /* sanity check */

  823.     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

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

  825.                srcW, srcH, dstW, dstH);

  826.         return AVERROR(EINVAL);

  827.     }

  828. 

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

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

  831. 

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

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

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

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

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

  837. 

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

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

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

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

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

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

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

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

  846. 

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

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

  849. 

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

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

  852. 

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

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

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

  856. 

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

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

  859.       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8

  860.       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4

  861.       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE

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

  863.         c->chrSrcHSubSample=1;

  864. 

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

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

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

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

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

  870. 

  871.     /* unscaled special cases */

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

  873.         ff_get_unscaled_swscale(c);

  874. 

  875.         if (c->swScale) {

  876.             if (flags&SWS_PRINT_INFO)

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

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

  879.             return 0;

  880.         }

  881.     }

  882. 

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

  884.     if (c->srcBpc < 8)

  885.         c->srcBpc = 8;

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

  887.     if (c->dstBpc < 8)

  888.         c->dstBpc = 8;

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

  890.         c->srcBpc = 16;

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

  892.         dst_stride <<= 1;

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

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

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

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

  897.             if (flags&SWS_PRINT_INFO)

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

  899.         }

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

  901.     }

  902.     else

  903.         c->canMMX2BeUsed=0;

  904. 

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

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

  907. 

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

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

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

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

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

  913.     // first and last pixel

  914.     if (flags&SWS_FAST_BILINEAR) {

  915.         if (c->canMMX2BeUsed) {

  916.             c->lumXInc+= 20;

  917.             c->chrXInc+= 20;

  918.         }

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

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

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

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

  923.         }

  924.     }

  925. 

  926.     /* precalculate horizontal scaler filter coefficients */

  927.     {

  928. #if HAVE_MMX2

  929. // can't downscale !!!

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

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

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

  933. 

  934. #ifdef MAP_ANONYMOUS

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

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

  937. #elif HAVE_VIRTUALALLOC

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

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

  940. #else

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

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

  943. #endif

  944. 

  945. #ifdef MAP_ANONYMOUS

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

  947. #else

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

  949. #endif

  950.                 return AVERROR(ENOMEM);

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

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

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

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

  955. 

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

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

  958. 

  959. #ifdef MAP_ANONYMOUS

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

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

  962. #endif

  963.         } else

  964. #endif /* HAVE_MMX2 */

  965.         {

  966.             const int filterAlign=

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

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

  969.                 1;

  970. 

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

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

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

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

  975.                 goto fail;

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

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

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

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

  980.                 goto fail;

  981.         }

  982.     } // initialize horizontal stuff

  983. 

  984.     /* precalculate vertical scaler filter coefficients */

  985.     {

  986.         const int filterAlign=

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

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

  989.             1;

  990. 

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

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

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

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

  995.             goto fail;

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

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

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

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

  1000.             goto fail;

  1001. 

  1002. #if HAVE_ALTIVEC

  1003.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);

  1004.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);

  1005. 

  1006.         for (i=0;i<c->vLumFilterSize*c->dstH;i++) {

  1007.             int j;

  1008.             short *p = (short *)&c->vYCoeffsBank[i];

  1009.             for (j=0;j<8;j++)

  1010.                 p[j] = c->vLumFilter[i];

  1011.         }

  1012. 

  1013.         for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {

  1014.             int j;

  1015.             short *p = (short *)&c->vCCoeffsBank[i];

  1016.             for (j=0;j<8;j++)

  1017.                 p[j] = c->vChrFilter[i];

  1018.         }

  1019. #endif

  1020.     }

  1021. 

  1022.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1023.     c->vLumBufSize= c->vLumFilterSize;

  1024.     c->vChrBufSize= c->vChrFilterSize;

  1025.     for (i=0; i<dstH; i++) {

  1026.         int chrI= (int64_t)i*c->chrDstH / dstH;

  1027.         int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,

  1028.                            ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));

  1029. 

  1030.         nextSlice>>= c->chrSrcVSubSample;

  1031.         nextSlice<<= c->chrSrcVSubSample;

  1032.         if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)

  1033.             c->vLumBufSize= nextSlice - c->vLumFilterPos[i];

  1034.         if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))

  1035.             c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];

  1036.     }

  1037. 

  1038.     // allocate pixbufs (we use dynamic allocation because otherwise we would need to

  1039.     // allocate several megabytes to handle all possible cases)

  1040.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1041.     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1042.     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1043.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1044.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1045.     //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)

  1046.     /* align at 16 bytes for AltiVec */

  1047.     for (i=0; i<c->vLumBufSize; i++) {

  1048.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1049.         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];

  1050.     }

  1051.     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)

  1052.     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);

  1053.     c->uv_offx2 = dst_stride + 16;

  1054.     for (i=0; i<c->vChrBufSize; i++) {

  1055.         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);

  1056.         c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];

  1057.         c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;

  1058.     }

  1059.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1060.         for (i=0; i<c->vLumBufSize; i++) {

  1061.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);

  1062.             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];

  1063.         }

  1064. 

  1065.     //try to avoid drawing green stuff between the right end and the stride end

  1066.     for (i=0; i<c->vChrBufSize; i++)

  1067.         if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){

  1068.             av_assert0(c->dstBpc > 10);

  1069.             for(j=0; j<dst_stride/2+1; j++)

  1070.                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;

  1071.         } else

  1072.             for(j=0; j<dst_stride+1; j++)

  1073.                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;

  1074. 

  1075.     assert(c->chrDstH <= dstH);

  1076. 

  1077.     if (flags&SWS_PRINT_INFO) {

  1078.         if      (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1079.         else if (flags&SWS_BILINEAR)      av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1080.         else if (flags&SWS_BICUBIC)       av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1081.         else if (flags&SWS_X)             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1082.         else if (flags&SWS_POINT)         av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1083.         else if (flags&SWS_AREA)          av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1084.         else if (flags&SWS_BICUBLIN)      av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1085.         else if (flags&SWS_GAUSS)         av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1086.         else if (flags&SWS_SINC)          av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1087.         else if (flags&SWS_LANCZOS)       av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1088.         else if (flags&SWS_SPLINE)        av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1089.         else                              av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1090. 

  1091.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1092.                av_get_pix_fmt_name(srcFormat),

  1093. #ifdef DITHER1XBPP

  1094.                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||

  1095.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1096.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",

  1097. #else

  1098.                "",

  1099. #endif

  1100.                av_get_pix_fmt_name(dstFormat));

  1101. 

  1102.         if      (HAVE_MMX2     && cpu_flags & AV_CPU_FLAG_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");

  1103.         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1104.         else if (HAVE_MMX      && cpu_flags & AV_CPU_FLAG_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");

  1105.         else if (HAVE_ALTIVEC  && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1106.         else                                   av_log(c, AV_LOG_INFO, "using C\n");

  1107. 

  1108.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1109.         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1110.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1111.         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1112.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1113.     }

  1114. 

  1115.     c->swScale= ff_getSwsFunc(c);

  1116.     return 0;

  1117. fail: //FIXME replace things by appropriate error codes

  1118.     return -1;

  1119. }

  1120. 

  1121. #if FF_API_SWS_GETCONTEXT

  1122. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1123.                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1124.                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1125. {

  1126.     SwsContext *c;

  1127. 

  1128.     if(!(c=sws_alloc_context()))

  1129.         return NULL;

  1130. 

  1131.     c->flags= flags;

  1132.     c->srcW= srcW;

  1133.     c->srcH= srcH;

  1134.     c->dstW= dstW;

  1135.     c->dstH= dstH;

  1136.     c->srcRange = handle_jpeg(&srcFormat);

  1137.     c->dstRange = handle_jpeg(&dstFormat);

  1138.     c->srcFormat= srcFormat;

  1139.     c->dstFormat= dstFormat;

  1140. 

  1141.     if (param) {

  1142.         c->param[0] = param[0];

  1143.         c->param[1] = param[1];

  1144.     }

  1145.     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);

  1146. 

  1147.     if(sws_init_context(c, srcFilter, dstFilter) < 0){

  1148.         sws_freeContext(c);

  1149.         return NULL;

  1150.     }

  1151. 

  1152.     return c;

  1153. }

  1154. #endif

  1155. 

  1156. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1157.                                 float lumaSharpen, float chromaSharpen,

  1158.                                 float chromaHShift, float chromaVShift,

  1159.                                 int verbose)

  1160. {

  1161.     SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  1162.     if (!filter)

  1163.         return NULL;

  1164. 

  1165.     if (lumaGBlur!=0.0) {

  1166.         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  1167.         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  1168.     } else {

  1169.         filter->lumH= sws_getIdentityVec();

  1170.         filter->lumV= sws_getIdentityVec();

  1171.     }

  1172. 

  1173.     if (chromaGBlur!=0.0) {

  1174.         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  1175.         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  1176.     } else {

  1177.         filter->chrH= sws_getIdentityVec();

  1178.         filter->chrV= sws_getIdentityVec();

  1179.     }

  1180. 

  1181.     if (chromaSharpen!=0.0) {

  1182.         SwsVector *id= sws_getIdentityVec();

  1183.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1184.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1185.         sws_addVec(filter->chrH, id);

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

  1187.         sws_freeVec(id);

  1188.     }

  1189. 

  1190.     if (lumaSharpen!=0.0) {

  1191.         SwsVector *id= sws_getIdentityVec();

  1192.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1193.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1194.         sws_addVec(filter->lumH, id);

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

  1196.         sws_freeVec(id);

  1197.     }

  1198. 

  1199.     if (chromaHShift != 0.0)

  1200.         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  1201. 

  1202.     if (chromaVShift != 0.0)

  1203.         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  1204. 

  1205.     sws_normalizeVec(filter->chrH, 1.0);

  1206.     sws_normalizeVec(filter->chrV, 1.0);

  1207.     sws_normalizeVec(filter->lumH, 1.0);

  1208.     sws_normalizeVec(filter->lumV, 1.0);

  1209. 

  1210.     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1211.     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1212. 

  1213.     return filter;

  1214. }

  1215. 

  1216. SwsVector *sws_allocVec(int length)

  1217. {

  1218.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1219.     if (!vec)

  1220.         return NULL;

  1221.     vec->length = length;

  1222.     vec->coeff  = av_malloc(sizeof(double) * length);

  1223.     if (!vec->coeff)

  1224.         av_freep(&vec);

  1225.     return vec;

  1226. }

  1227. 

  1228. SwsVector *sws_getGaussianVec(double variance, double quality)

  1229. {

  1230.     const int length= (int)(variance*quality + 0.5) | 1;

  1231.     int i;

  1232.     double middle= (length-1)*0.5;

  1233.     SwsVector *vec= sws_allocVec(length);

  1234. 

  1235.     if (!vec)

  1236.         return NULL;

  1237. 

  1238.     for (i=0; i<length; i++) {

  1239.         double dist= i-middle;

  1240.         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);

  1241.     }

  1242. 

  1243.     sws_normalizeVec(vec, 1.0);

  1244. 

  1245.     return vec;

  1246. }

  1247. 

  1248. SwsVector *sws_getConstVec(double c, int length)

  1249. {

  1250.     int i;

  1251.     SwsVector *vec= sws_allocVec(length);

  1252. 

  1253.     if (!vec)

  1254.         return NULL;

  1255. 

  1256.     for (i=0; i<length; i++)

  1257.         vec->coeff[i]= c;

  1258. 

  1259.     return vec;

  1260. }

  1261. 

  1262. SwsVector *sws_getIdentityVec(void)

  1263. {

  1264.     return sws_getConstVec(1.0, 1);

  1265. }

  1266. 

  1267. static double sws_dcVec(SwsVector *a)

  1268. {

  1269.     int i;

  1270.     double sum=0;

  1271. 

  1272.     for (i=0; i<a->length; i++)

  1273.         sum+= a->coeff[i];

  1274. 

  1275.     return sum;

  1276. }

  1277. 

  1278. void sws_scaleVec(SwsVector *a, double scalar)

  1279. {

  1280.     int i;

  1281. 

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

  1283.         a->coeff[i]*= scalar;

  1284. }

  1285. 

  1286. void sws_normalizeVec(SwsVector *a, double height)

  1287. {

  1288.     sws_scaleVec(a, height/sws_dcVec(a));

  1289. }

  1290. 

  1291. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1292. {

  1293.     int length= a->length + b->length - 1;

  1294.     int i, j;

  1295.     SwsVector *vec= sws_getConstVec(0.0, length);

  1296. 

  1297.     if (!vec)

  1298.         return NULL;

  1299. 

  1300.     for (i=0; i<a->length; i++) {

  1301.         for (j=0; j<b->length; j++) {

  1302.             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1303.         }

  1304.     }

  1305. 

  1306.     return vec;

  1307. }

  1308. 

  1309. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1310. {

  1311.     int length= FFMAX(a->length, b->length);

  1312.     int i;

  1313.     SwsVector *vec= sws_getConstVec(0.0, length);

  1314. 

  1315.     if (!vec)

  1316.         return NULL;

  1317. 

  1318.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1319.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];

  1320. 

  1321.     return vec;

  1322. }

  1323. 

  1324. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1325. {

  1326.     int length= FFMAX(a->length, b->length);

  1327.     int i;

  1328.     SwsVector *vec= sws_getConstVec(0.0, length);

  1329. 

  1330.     if (!vec)

  1331.         return NULL;

  1332. 

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

  1334.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1335. 

  1336.     return vec;

  1337. }

  1338. 

  1339. /* shift left / or right if "shift" is negative */

  1340. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1341. {

  1342.     int length= a->length + FFABS(shift)*2;

  1343.     int i;

  1344.     SwsVector *vec= sws_getConstVec(0.0, length);

  1345. 

  1346.     if (!vec)

  1347.         return NULL;

  1348. 

  1349.     for (i=0; i<a->length; i++) {

  1350.         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1351.     }

  1352. 

  1353.     return vec;

  1354. }

  1355. 

  1356. void sws_shiftVec(SwsVector *a, int shift)

  1357. {

  1358.     SwsVector *shifted= sws_getShiftedVec(a, shift);

  1359.     av_free(a->coeff);

  1360.     a->coeff= shifted->coeff;

  1361.     a->length= shifted->length;

  1362.     av_free(shifted);

  1363. }

  1364. 

  1365. void sws_addVec(SwsVector *a, SwsVector *b)

  1366. {

  1367.     SwsVector *sum= sws_sumVec(a, b);

  1368.     av_free(a->coeff);

  1369.     a->coeff= sum->coeff;

  1370.     a->length= sum->length;

  1371.     av_free(sum);

  1372. }

  1373. 

  1374. void sws_subVec(SwsVector *a, SwsVector *b)

  1375. {

  1376.     SwsVector *diff= sws_diffVec(a, b);

  1377.     av_free(a->coeff);

  1378.     a->coeff= diff->coeff;

  1379.     a->length= diff->length;

  1380.     av_free(diff);

  1381. }

  1382. 

  1383. void sws_convVec(SwsVector *a, SwsVector *b)

  1384. {

  1385.     SwsVector *conv= sws_getConvVec(a, b);

  1386.     av_free(a->coeff);

  1387.     a->coeff= conv->coeff;

  1388.     a->length= conv->length;

  1389.     av_free(conv);

  1390. }

  1391. 

  1392. SwsVector *sws_cloneVec(SwsVector *a)

  1393. {

  1394.     int i;

  1395.     SwsVector *vec= sws_allocVec(a->length);

  1396. 

  1397.     if (!vec)

  1398.         return NULL;

  1399. 

  1400.     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];

  1401. 

  1402.     return vec;

  1403. }

  1404. 

  1405. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1406. {

  1407.     int i;

  1408.     double max=0;

  1409.     double min=0;

  1410.     double range;

  1411. 

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

  1413.         if (a->coeff[i]>max) max= a->coeff[i];

  1414. 

  1415.     for (i=0; i<a->length; i++)

  1416.         if (a->coeff[i]<min) min= a->coeff[i];

  1417. 

  1418.     range= max - min;

  1419. 

  1420.     for (i=0; i<a->length; i++) {

  1421.         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1422.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1423.         for (;x>0; x--) av_log(log_ctx, log_level, " ");

  1424.         av_log(log_ctx, log_level, "|\n");

  1425.     }

  1426. }

  1427. 

  1428. void sws_freeVec(SwsVector *a)

  1429. {

  1430.     if (!a) return;

  1431.     av_freep(&a->coeff);

  1432.     a->length=0;

  1433.     av_free(a);

  1434. }

  1435. 

  1436. void sws_freeFilter(SwsFilter *filter)

  1437. {

  1438.     if (!filter) return;

  1439. 

  1440.     if (filter->lumH) sws_freeVec(filter->lumH);

  1441.     if (filter->lumV) sws_freeVec(filter->lumV);

  1442.     if (filter->chrH) sws_freeVec(filter->chrH);

  1443.     if (filter->chrV) sws_freeVec(filter->chrV);

  1444.     av_free(filter);

  1445. }

  1446. 

  1447. void sws_freeContext(SwsContext *c)

  1448. {

  1449.     int i;

  1450.     if (!c) return;

  1451. 

  1452.     if (c->lumPixBuf) {

  1453.         for (i=0; i<c->vLumBufSize; i++)

  1454.             av_freep(&c->lumPixBuf[i]);

  1455.         av_freep(&c->lumPixBuf);

  1456.     }

  1457. 

  1458.     if (c->chrUPixBuf) {

  1459.         for (i=0; i<c->vChrBufSize; i++)

  1460.             av_freep(&c->chrUPixBuf[i]);

  1461.         av_freep(&c->chrUPixBuf);

  1462.         av_freep(&c->chrVPixBuf);

  1463.     }

  1464. 

  1465.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1466.         for (i=0; i<c->vLumBufSize; i++)

  1467.             av_freep(&c->alpPixBuf[i]);

  1468.         av_freep(&c->alpPixBuf);

  1469.     }

  1470. 

  1471.     av_freep(&c->vLumFilter);

  1472.     av_freep(&c->vChrFilter);

  1473.     av_freep(&c->hLumFilter);

  1474.     av_freep(&c->hChrFilter);

  1475. #if HAVE_ALTIVEC

  1476.     av_freep(&c->vYCoeffsBank);

  1477.     av_freep(&c->vCCoeffsBank);

  1478. #endif

  1479. 

  1480.     av_freep(&c->vLumFilterPos);

  1481.     av_freep(&c->vChrFilterPos);

  1482.     av_freep(&c->hLumFilterPos);

  1483.     av_freep(&c->hChrFilterPos);

  1484. 

  1485. #if HAVE_MMX

  1486. #ifdef MAP_ANONYMOUS

  1487.     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1488.     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1489. #elif HAVE_VIRTUALALLOC

  1490.     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1491.     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1492. #else

  1493.     av_free(c->lumMmx2FilterCode);

  1494.     av_free(c->chrMmx2FilterCode);

  1495. #endif

  1496.     c->lumMmx2FilterCode=NULL;

  1497.     c->chrMmx2FilterCode=NULL;

  1498. #endif /* HAVE_MMX */

  1499. 

  1500.     av_freep(&c->yuvTable);

  1501.     av_freep(&c->formatConvBuffer);

  1502. 

  1503.     av_free(c);

  1504. }

  1505. 

  1506. struct SwsContext *sws_getCachedContext(struct SwsContext *context,

  1507.                                         int srcW, int srcH, enum PixelFormat srcFormat,

  1508.                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1509.                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1510. {

  1511.     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};

  1512. 

  1513.     if (!param)

  1514.         param = default_param;

  1515. 

  1516.     if (context &&

  1517.         (context->srcW      != srcW      ||

  1518.          context->srcH      != srcH      ||

  1519.          context->srcFormat != srcFormat ||

  1520.          context->dstW      != dstW      ||

  1521.          context->dstH      != dstH      ||

  1522.          context->dstFormat != dstFormat ||

  1523.          context->flags     != flags     ||

  1524.          context->param[0]  != param[0]  ||

  1525.          context->param[1]  != param[1])) {

  1526.         sws_freeContext(context);

  1527.         context = NULL;

  1528.     }

  1529. 

  1530.     if (!context) {

  1531.         if (!(context = sws_alloc_context()))

  1532.             return NULL;

  1533.         context->srcW      = srcW;

  1534.         context->srcH      = srcH;

  1535.         context->srcRange  = handle_jpeg(&srcFormat);

  1536.         context->srcFormat = srcFormat;

  1537.         context->dstW      = dstW;

  1538.         context->dstH      = dstH;

  1539.         context->dstRange  = handle_jpeg(&dstFormat);

  1540.         context->dstFormat = dstFormat;

  1541.         context->flags     = flags;

  1542.         context->param[0]  = param[0];

  1543.         context->param[1]  = param[1];

  1544.         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);

  1545.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1546.             sws_freeContext(context);

  1547.             return NULL;

  1548.         }

  1549.     }

  1550.     return context;

  1551. }