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.
22948 Commits
32 Branches
283MB
Tree: 8ef4e65e94
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '8ef4e65e94'
${ noResults }
FFmpeg/libswscale/utils.c

1591 lines
56KB
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 modify

  7.  * it under the terms of the GNU General Public License as published by

  8.  * the Free Software Foundation; either version 2 of the License, or

  9.  * (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

  14.  * GNU General Public License for more details.

  15.  *

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

  17.  * 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.  * the C code (not assembly, mmx, ...) of this file can be used

  21.  * under the LGPL license too

  22.  */

  23. 

  24. #define _SVID_SOURCE //needed for MAP_ANONYMOUS

  25. #include <inttypes.h>

  26. #include <string.h>

  27. #include <math.h>

  28. #include <stdio.h>

  29. #include "config.h"

  30. #include <assert.h>

  31. #if HAVE_SYS_MMAN_H

  32. #include <sys/mman.h>

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

  34. #define MAP_ANONYMOUS MAP_ANON

  35. #endif

  36. #endif

  37. #if HAVE_VIRTUALALLOC

  38. #define WIN32_LEAN_AND_MEAN

  39. #include <windows.h>

  40. #endif

  41. #include "swscale.h"

  42. #include "swscale_internal.h"

  43. #include "rgb2rgb.h"

  44. #include "libavutil/intreadwrite.h"

  45. #include "libavutil/x86_cpu.h"

  46. #include "libavutil/avutil.h"

  47. #include "libavutil/bswap.h"

  48. #include "libavutil/pixdesc.h"

  49. 

  50. unsigned swscale_version(void)

  51. {

  52.     return LIBSWSCALE_VERSION_INT;

  53. }

  54. 

  55. const char *swscale_configuration(void)

  56. {

  57.     return FFMPEG_CONFIGURATION;

  58. }

  59. 

  60. const char *swscale_license(void)

  61. {

  62. #define LICENSE_PREFIX "libswscale license: "

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

  64. }

  65. 

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

  67. 

  68. #define isSupportedIn(x)    (       \

  69.            (x)==PIX_FMT_YUV420P     \

  70.         || (x)==PIX_FMT_YUVA420P    \

  71.         || (x)==PIX_FMT_YUYV422     \

  72.         || (x)==PIX_FMT_UYVY422     \

  73.         || (x)==PIX_FMT_RGB48BE     \

  74.         || (x)==PIX_FMT_RGB48LE     \

  75.         || (x)==PIX_FMT_RGB32       \

  76.         || (x)==PIX_FMT_RGB32_1     \

  77.         || (x)==PIX_FMT_BGR24       \

  78.         || (x)==PIX_FMT_BGR565      \

  79.         || (x)==PIX_FMT_BGR555      \

  80.         || (x)==PIX_FMT_BGR32       \

  81.         || (x)==PIX_FMT_BGR32_1     \

  82.         || (x)==PIX_FMT_RGB24       \

  83.         || (x)==PIX_FMT_RGB565      \

  84.         || (x)==PIX_FMT_RGB555      \

  85.         || (x)==PIX_FMT_GRAY8       \

  86.         || (x)==PIX_FMT_YUV410P     \

  87.         || (x)==PIX_FMT_YUV440P     \

  88.         || (x)==PIX_FMT_NV12        \

  89.         || (x)==PIX_FMT_NV21        \

  90.         || (x)==PIX_FMT_GRAY16BE    \

  91.         || (x)==PIX_FMT_GRAY16LE    \

  92.         || (x)==PIX_FMT_YUV444P     \

  93.         || (x)==PIX_FMT_YUV422P     \

  94.         || (x)==PIX_FMT_YUV411P     \

  95.         || (x)==PIX_FMT_YUVJ420P    \

  96.         || (x)==PIX_FMT_YUVJ422P    \

  97.         || (x)==PIX_FMT_YUVJ440P    \

  98.         || (x)==PIX_FMT_YUVJ444P    \

  99.         || (x)==PIX_FMT_PAL8        \

  100.         || (x)==PIX_FMT_BGR8        \

  101.         || (x)==PIX_FMT_RGB8        \

  102.         || (x)==PIX_FMT_BGR4_BYTE   \

  103.         || (x)==PIX_FMT_RGB4_BYTE   \

  104.         || (x)==PIX_FMT_YUV440P     \

  105.         || (x)==PIX_FMT_MONOWHITE   \

  106.         || (x)==PIX_FMT_MONOBLACK   \

  107.         || (x)==PIX_FMT_YUV420P16LE   \

  108.         || (x)==PIX_FMT_YUV422P16LE   \

  109.         || (x)==PIX_FMT_YUV444P16LE   \

  110.         || (x)==PIX_FMT_YUV420P16BE   \

  111.         || (x)==PIX_FMT_YUV422P16BE   \

  112.         || (x)==PIX_FMT_YUV444P16BE   \

  113.     )

  114. 

  115. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  116. {

  117.     return isSupportedIn(pix_fmt);

  118. }

  119. 

  120. #define isSupportedOut(x)   (       \

  121.            (x)==PIX_FMT_YUV420P     \

  122.         || (x)==PIX_FMT_YUVA420P    \

  123.         || (x)==PIX_FMT_YUYV422     \

  124.         || (x)==PIX_FMT_UYVY422     \

  125.         || (x)==PIX_FMT_YUV444P     \

  126.         || (x)==PIX_FMT_YUV422P     \

  127.         || (x)==PIX_FMT_YUV411P     \

  128.         || (x)==PIX_FMT_YUVJ420P    \

  129.         || (x)==PIX_FMT_YUVJ422P    \

  130.         || (x)==PIX_FMT_YUVJ440P    \

  131.         || (x)==PIX_FMT_YUVJ444P    \

  132.         || isAnyRGB(x)              \

  133.         || (x)==PIX_FMT_NV12        \

  134.         || (x)==PIX_FMT_NV21        \

  135.         || (x)==PIX_FMT_GRAY16BE    \

  136.         || (x)==PIX_FMT_GRAY16LE    \

  137.         || (x)==PIX_FMT_GRAY8       \

  138.         || (x)==PIX_FMT_YUV410P     \

  139.         || (x)==PIX_FMT_YUV440P     \

  140.         || (x)==PIX_FMT_YUV420P16LE   \

  141.         || (x)==PIX_FMT_YUV422P16LE   \

  142.         || (x)==PIX_FMT_YUV444P16LE   \

  143.         || (x)==PIX_FMT_YUV420P16BE   \

  144.         || (x)==PIX_FMT_YUV422P16BE   \

  145.         || (x)==PIX_FMT_YUV444P16BE   \

  146.     )

  147. 

  148. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  149. {

  150.     return isSupportedOut(pix_fmt);

  151. }

  152. 

  153. #define usePal(x) (av_pix_fmt_descriptors[x].flags & PIX_FMT_PAL)

  154. 

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

  156. 

  157. const char *sws_format_name(enum PixelFormat format)

  158. {

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

  160.         return av_pix_fmt_descriptors[format].name;

  161.     else

  162.         return "Unknown format";

  163. }

  164. 

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

  166. {

  167. //    printf("%f %f %f %f %f\n", a,b,c,d,dist);

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

  169.     else           return getSplineCoeff(        0.0,

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

  171.                                                  c + 3.0*d,

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

  173.                                          dist-1.0);

  174. }

  175. 

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

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

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

  179. {

  180.     int i;

  181.     int filterSize;

  182.     int filter2Size;

  183.     int minFilterSize;

  184.     int64_t *filter=NULL;

  185.     int64_t *filter2=NULL;

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

  187.     int ret= -1;

  188. #if ARCH_X86

  189.     if (flags & SWS_CPU_CAPS_MMX)

  190.         __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)

  191. #endif

  192. 

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

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

  195. 

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

  197.         int i;

  198.         filterSize= 1;

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

  200. 

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

  202.             filter[i*filterSize]= fone;

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

  204.         }

  205. 

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

  207.         int i;

  208.         int xDstInSrc;

  209.         filterSize= 1;

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

  211. 

  212.         xDstInSrc= xInc/2 - 0x8000;

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

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

  215. 

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

  217.             filter[i]= fone;

  218.             xDstInSrc+= xInc;

  219.         }

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

  221.         int i;

  222.         int xDstInSrc;

  223.         filterSize= 2;

  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.             int j;

  230. 

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

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

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

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

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

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

  237.                 xx++;

  238.             }

  239.             xDstInSrc+= xInc;

  240.         }

  241.     } else {

  242.         int xDstInSrc;

  243.         int sizeFactor;

  244. 

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

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

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

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

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

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

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

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

  253.         else {

  254.             sizeFactor= 0; //GCC warning killer

  255.             assert(0);

  256.         }

  257. 

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

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

  260. 

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

  262. 

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

  264. 

  265.         xDstInSrc= xInc - 0x10000;

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

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

  268.             int j;

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

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

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

  272.                 double floatd;

  273.                 int64_t coeff;

  274. 

  275.                 if (xInc > 1<<16)

  276.                     d= d*dstW/srcW;

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

  278. 

  279.                 if (flags & SWS_BICUBIC) {

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

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

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

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

  284. 

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

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

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

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

  289.                     else

  290.                         coeff=0.0;

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

  292.                 }

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

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

  295.                     coeff = d ? sin(d*PI)/(d*PI) : 1.0;

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

  297.                 }*/

  298.                 else if (flags & SWS_X) {

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

  300.                     double c;

  301. 

  302.                     if (floatd<1.0)

  303.                         c = cos(floatd*M_PI);

  304.                     else

  305.                         c=-1.0;

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

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

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

  309.                 } else if (flags & SWS_AREA) {

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

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

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

  313.                     else coeff=0.0;

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

  315.                 } else if (flags & SWS_GAUSS) {

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

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

  318.                 } else if (flags & SWS_SINC) {

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

  320.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  324.                 } else if (flags & SWS_BILINEAR) {

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

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

  327.                     coeff *= fone >> 30;

  328.                 } else if (flags & SWS_SPLINE) {

  329.                     double p=-2.196152422706632;

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

  331.                 } else {

  332.                     coeff= 0.0; //GCC warning killer

  333.                     assert(0);

  334.                 }

  335. 

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

  337.                 xx++;

  338.             }

  339.             xDstInSrc+= 2*xInc;

  340.         }

  341.     }

  342. 

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

  344.        av_free(filter);

  345.     */

  346.     assert(filterSize>0);

  347.     filter2Size= filterSize;

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

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

  350.     assert(filter2Size>0);

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

  352. 

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

  354.         int j, k;

  355. 

  356.         if(srcFilter) {

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

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

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

  360.             }

  361.         } else {

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

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

  364.         }

  365.         //FIXME dstFilter

  366. 

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

  368.     }

  369.     av_freep(&filter);

  370. 

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

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

  373.     minFilterSize= 0;

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

  375.         int min= filter2Size;

  376.         int j;

  377.         int64_t cutOff=0.0;

  378. 

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

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

  381.             int k;

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

  383. 

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

  385. 

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

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

  388. 

  389.             // move filter coefficients left

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

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

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

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

  394.         }

  395. 

  396.         cutOff=0;

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

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

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

  400. 

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

  402.             min--;

  403.         }

  404. 

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

  406.     }

  407. 

  408.     if (flags & SWS_CPU_CAPS_ALTIVEC) {

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

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

  411.         if (minFilterSize < 5)

  412.             filterAlign = 4;

  413. 

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

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

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

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

  418.         // decent-sized vector.

  419.         if (minFilterSize < 3)

  420.             filterAlign = 1;

  421.     }

  422. 

  423.     if (flags & SWS_CPU_CAPS_MMX) {

  424.         // special case for unscaled vertical filtering

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

  426.             filterAlign= 1;

  427.     }

  428. 

  429.     assert(minFilterSize > 0);

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

  431.     assert(filterSize > 0);

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

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

  434.         goto fail;

  435.     *outFilterSize= filterSize;

  436. 

  437.     if (flags&SWS_PRINT_INFO)

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

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

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

  441.         int j;

  442. 

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

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

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

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

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

  448.         }

  449.     }

  450. 

  451.     //FIXME try to align filterPos if possible

  452. 

  453.     //fix borders

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

  455.         int j;

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

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

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

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

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

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

  462.             }

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

  464.         }

  465. 

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

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

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

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

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

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

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

  473.             }

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

  475.         }

  476.     }

  477. 

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

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

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

  481. 

  482.     /* normalize & store in outFilter */

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

  484.         int j;

  485.         int64_t error=0;

  486.         int64_t sum=0;

  487. 

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

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

  490.         }

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

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

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

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

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

  496.             error= v - intV*sum;

  497.         }

  498.     }

  499. 

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

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

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

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

  504.     }

  505. 

  506.     ret=0;

  507. fail:

  508.     av_free(filter);

  509.     av_free(filter2);

  510.     return ret;

  511. }

  512. 

  513. #if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL

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

  515. {

  516.     uint8_t *fragmentA;

  517.     x86_reg imm8OfPShufW1A;

  518.     x86_reg imm8OfPShufW2A;

  519.     x86_reg fragmentLengthA;

  520.     uint8_t *fragmentB;

  521.     x86_reg imm8OfPShufW1B;

  522.     x86_reg imm8OfPShufW2B;

  523.     x86_reg fragmentLengthB;

  524.     int fragmentPos;

  525. 

  526.     int xpos, i;

  527. 

  528.     // create an optimized horizontal scaling routine

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

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

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

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

  533.      * of fragmentA is used.

  534.      */

  535. 

  536.     //code fragment

  537. 

  538.     __asm__ volatile(

  539.         "jmp                         9f                 \n\t"

  540.     // Begin

  541.         "0:                                             \n\t"

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

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

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

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

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

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

  548.         "1:                                             \n\t"

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

  550.         "2:                                             \n\t"

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

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

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

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

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

  556. 

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

  558. 

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

  560.     // End

  561.         "9:                                             \n\t"

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

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

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

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

  566.         "dec                         %1                 \n\t"

  567.         "dec                         %2                 \n\t"

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

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

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

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

  572. 

  573. 

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

  575.         "=r" (fragmentLengthA)

  576.     );

  577. 

  578.     __asm__ volatile(

  579.         "jmp                         9f                 \n\t"

  580.     // Begin

  581.         "0:                                             \n\t"

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

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

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

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

  586.         "1:                                             \n\t"

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

  588.         "2:                                             \n\t"

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

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

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

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

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

  594. 

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

  596. 

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

  598.     // End

  599.         "9:                                             \n\t"

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

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

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

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

  604.         "dec                         %1                 \n\t"

  605.         "dec                         %2                 \n\t"

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

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

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

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

  610. 

  611. 

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

  613.         "=r" (fragmentLengthB)

  614.     );

  615. 

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

  617.     fragmentPos=0;

  618. 

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

  620.         int xx=xpos>>16;

  621. 

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

  623.             int a=0;

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

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

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

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

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

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

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

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

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

  633.             int shift=0;

  634. 

  635.             if (filterCode) {

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

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

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

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

  640.                 filterPos[i/2]= xx;

  641. 

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

  643. 

  644.                 filterCode[fragmentPos + imm8OfPShufW1]=

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

  646.                 filterCode[fragmentPos + imm8OfPShufW2]=

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

  648. 

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

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

  651. 

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

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

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

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

  656.                 }

  657.             }

  658. 

  659.             fragmentPos+= fragmentLength;

  660. 

  661.             if (filterCode)

  662.                 filterCode[fragmentPos]= RET;

  663.         }

  664.         xpos+=xInc;

  665.     }

  666.     if (filterCode)

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

  668. 

  669.     return fragmentPos + 1;

  670. }

  671. #endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL */

  672. 

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

  674. {

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

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

  677. }

  678. 

  679. static uint16_t roundToInt16(int64_t f)

  680. {

  681.     int r= (f + (1<<15))>>16;

  682.          if (r<-0x7FFF) return 0x8000;

  683.     else if (r> 0x7FFF) return 0x7FFF;

  684.     else                return r;

  685. }

  686. 

  687. int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)

  688. {

  689.     int64_t crv =  inv_table[0];

  690.     int64_t cbu =  inv_table[1];

  691.     int64_t cgu = -inv_table[2];

  692.     int64_t cgv = -inv_table[3];

  693.     int64_t cy  = 1<<16;

  694.     int64_t oy  = 0;

  695. 

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

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

  698. 

  699.     c->brightness= brightness;

  700.     c->contrast  = contrast;

  701.     c->saturation= saturation;

  702.     c->srcRange  = srcRange;

  703.     c->dstRange  = dstRange;

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

  705. 

  706.     c->uOffset=   0x0400040004000400LL;

  707.     c->vOffset=   0x0400040004000400LL;

  708. 

  709.     if (!srcRange) {

  710.         cy= (cy*255) / 219;

  711.         oy= 16<<16;

  712.     } else {

  713.         crv= (crv*224) / 255;

  714.         cbu= (cbu*224) / 255;

  715.         cgu= (cgu*224) / 255;

  716.         cgv= (cgv*224) / 255;

  717.     }

  718. 

  719.     cy = (cy *contrast             )>>16;

  720.     crv= (crv*contrast * saturation)>>32;

  721.     cbu= (cbu*contrast * saturation)>>32;

  722.     cgu= (cgu*contrast * saturation)>>32;

  723.     cgv= (cgv*contrast * saturation)>>32;

  724. 

  725.     oy -= 256*brightness;

  726. 

  727.     c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;

  728.     c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;

  729.     c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;

  730.     c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;

  731.     c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;

  732.     c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;

  733. 

  734.     c->yuv2rgb_y_coeff  = (int16_t)roundToInt16(cy <<13);

  735.     c->yuv2rgb_y_offset = (int16_t)roundToInt16(oy << 9);

  736.     c->yuv2rgb_v2r_coeff= (int16_t)roundToInt16(crv<<13);

  737.     c->yuv2rgb_v2g_coeff= (int16_t)roundToInt16(cgv<<13);

  738.     c->yuv2rgb_u2g_coeff= (int16_t)roundToInt16(cgu<<13);

  739.     c->yuv2rgb_u2b_coeff= (int16_t)roundToInt16(cbu<<13);

  740. 

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

  742.     //FIXME factorize

  743. 

  744. #if ARCH_PPC && (HAVE_ALTIVEC || CONFIG_RUNTIME_CPUDETECT)

  745.     if (c->flags & SWS_CPU_CAPS_ALTIVEC)

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

  747. #endif

  748.     return 0;

  749. }

  750. 

  751. int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)

  752. {

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

  754. 

  755.     *inv_table = c->srcColorspaceTable;

  756.     *table     = c->dstColorspaceTable;

  757.     *srcRange  = c->srcRange;

  758.     *dstRange  = c->dstRange;

  759.     *brightness= c->brightness;

  760.     *contrast  = c->contrast;

  761.     *saturation= c->saturation;

  762. 

  763.     return 0;

  764. }

  765. 

  766. static int handle_jpeg(enum PixelFormat *format)

  767. {

  768.     switch (*format) {

  769.     case PIX_FMT_YUVJ420P:

  770.         *format = PIX_FMT_YUV420P;

  771.         return 1;

  772.     case PIX_FMT_YUVJ422P:

  773.         *format = PIX_FMT_YUV422P;

  774.         return 1;

  775.     case PIX_FMT_YUVJ444P:

  776.         *format = PIX_FMT_YUV444P;

  777.         return 1;

  778.     case PIX_FMT_YUVJ440P:

  779.         *format = PIX_FMT_YUV440P;

  780.         return 1;

  781.     default:

  782.         return 0;

  783.     }

  784. }

  785. 

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

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

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

  789. {

  790. 

  791.     SwsContext *c;

  792.     int i;

  793.     int usesVFilter, usesHFilter;

  794.     int unscaled;

  795.     int srcRange, dstRange;

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

  797. #if ARCH_X86

  798.     if (flags & SWS_CPU_CAPS_MMX)

  799.         __asm__ volatile("emms\n\t"::: "memory");

  800. #endif

  801. 

  802. #if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off

  803.     flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);

  804.     flags |= ff_hardcodedcpuflags();

  805. #endif /* CONFIG_RUNTIME_CPUDETECT */

  806.     if (!rgb15to16) sws_rgb2rgb_init(flags);

  807. 

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

  809. 

  810.     srcRange = handle_jpeg(&srcFormat);

  811.     dstRange = handle_jpeg(&dstFormat);

  812. 

  813.     if (!isSupportedIn(srcFormat)) {

  814.         av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));

  815.         return NULL;

  816.     }

  817.     if (!isSupportedOut(dstFormat)) {

  818.         av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));

  819.         return NULL;

  820.     }

  821. 

  822.     i= flags & ( SWS_POINT

  823.                 |SWS_AREA

  824.                 |SWS_BILINEAR

  825.                 |SWS_FAST_BILINEAR

  826.                 |SWS_BICUBIC

  827.                 |SWS_X

  828.                 |SWS_GAUSS

  829.                 |SWS_LANCZOS

  830.                 |SWS_SINC

  831.                 |SWS_SPLINE

  832.                 |SWS_BICUBLIN);

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

  834.         av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be chosen\n");

  835.         return NULL;

  836.     }

  837. 

  838.     /* sanity check */

  839.     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

  840.         av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",

  841.                srcW, srcH, dstW, dstH);

  842.         return NULL;

  843.     }

  844.     if(srcW > VOFW || dstW > VOFW) {

  845.         av_log(NULL, AV_LOG_ERROR, "swScaler: Compile-time maximum width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");

  846.         return NULL;

  847.     }

  848. 

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

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

  851. 

  852.     FF_ALLOCZ_OR_GOTO(NULL, c, sizeof(SwsContext), fail);

  853. 

  854.     c->av_class = &sws_context_class;

  855.     c->srcW= srcW;

  856.     c->srcH= srcH;

  857.     c->dstW= dstW;

  858.     c->dstH= dstH;

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

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

  861.     c->flags= flags;

  862.     c->dstFormat= dstFormat;

  863.     c->srcFormat= srcFormat;

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

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

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

  867. 

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

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

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

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

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

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

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

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

  876. 

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

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

  879. 

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

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

  882. 

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

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

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

  886. 

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

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

  889.       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8

  890.       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4

  891.       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE

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

  893.         c->chrSrcHSubSample=1;

  894. 

  895.     if (param) {

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

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

  898.     } else {

  899.         c->param[0] =

  900.         c->param[1] = SWS_PARAM_DEFAULT;

  901.     }

  902. 

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

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

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

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

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

  908. 

  909.     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16);

  910. 

  911.     /* unscaled special cases */

  912.     if (unscaled && !usesHFilter && !usesVFilter && (srcRange == dstRange || isAnyRGB(dstFormat))) {

  913.         ff_get_unscaled_swscale(c);

  914. 

  915.         if (c->swScale) {

  916.             if (flags&SWS_PRINT_INFO)

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

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

  919.             return c;

  920.         }

  921.     }

  922. 

  923.     if (flags & SWS_CPU_CAPS_MMX2) {

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

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

  926.             if (flags&SWS_PRINT_INFO)

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

  928.         }

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

  930.     }

  931.     else

  932.         c->canMMX2BeUsed=0;

  933. 

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

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

  936. 

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

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

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

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

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

  942.     // first and last pixel

  943.     if (flags&SWS_FAST_BILINEAR) {

  944.         if (c->canMMX2BeUsed) {

  945.             c->lumXInc+= 20;

  946.             c->chrXInc+= 20;

  947.         }

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

  949.         else if (flags & SWS_CPU_CAPS_MMX) {

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

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

  952.         }

  953.     }

  954. 

  955.     /* precalculate horizontal scaler filter coefficients */

  956.     {

  957. #if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL

  958. // can't downscale !!!

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

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

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

  962. 

  963. #ifdef MAP_ANONYMOUS

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

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

  966. #elif HAVE_VIRTUALALLOC

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

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

  969. #else

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

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

  972. #endif

  973. 

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

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

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

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

  978. 

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

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

  981. 

  982. #ifdef MAP_ANONYMOUS

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

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

  985. #endif

  986.         } else

  987. #endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL */

  988.         {

  989.             const int filterAlign=

  990.                 (flags & SWS_CPU_CAPS_MMX) ? 4 :

  991.                 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :

  992.                 1;

  993. 

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

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

  996.                            (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,

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

  998.                 goto fail;

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

  1000.                            c->chrSrcW, c->chrDstW, filterAlign, 1<<14,

  1001.                            (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,

  1002.                            srcFilter->chrH, dstFilter->chrH, c->param) < 0)

  1003.                 goto fail;

  1004.         }

  1005.     } // initialize horizontal stuff

  1006. 

  1007.     /* precalculate vertical scaler filter coefficients */

  1008.     {

  1009.         const int filterAlign=

  1010.             (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :

  1011.             (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :

  1012.             1;

  1013. 

  1014.         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,

  1015.                        srcH      ,        dstH, filterAlign, (1<<12),

  1016.                        (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,

  1017.                        srcFilter->lumV, dstFilter->lumV, c->param) < 0)

  1018.             goto fail;

  1019.         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,

  1020.                        c->chrSrcH, c->chrDstH, filterAlign, (1<<12),

  1021.                        (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,

  1022.                        srcFilter->chrV, dstFilter->chrV, c->param) < 0)

  1023.             goto fail;

  1024. 

  1025. #if ARCH_PPC && (HAVE_ALTIVEC || CONFIG_RUNTIME_CPUDETECT)

  1026.         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);

  1027.         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);

  1028. 

  1029.         for (i=0;i<c->vLumFilterSize*c->dstH;i++) {

  1030.             int j;

  1031.             short *p = (short *)&c->vYCoeffsBank[i];

  1032.             for (j=0;j<8;j++)

  1033.                 p[j] = c->vLumFilter[i];

  1034.         }

  1035. 

  1036.         for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {

  1037.             int j;

  1038.             short *p = (short *)&c->vCCoeffsBank[i];

  1039.             for (j=0;j<8;j++)

  1040.                 p[j] = c->vChrFilter[i];

  1041.         }

  1042. #endif

  1043.     }

  1044. 

  1045.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1046.     c->vLumBufSize= c->vLumFilterSize;

  1047.     c->vChrBufSize= c->vChrFilterSize;

  1048.     for (i=0; i<dstH; i++) {

  1049.         int chrI= i*c->chrDstH / dstH;

  1050.         int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,

  1051.                            ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));

  1052. 

  1053.         nextSlice>>= c->chrSrcVSubSample;

  1054.         nextSlice<<= c->chrSrcVSubSample;

  1055.         if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)

  1056.             c->vLumBufSize= nextSlice - c->vLumFilterPos[i];

  1057.         if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))

  1058.             c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];

  1059.     }

  1060. 

  1061.     // allocate pixbufs (we use dynamic allocation because otherwise we would need to

  1062.     // allocate several megabytes to handle all possible cases)

  1063.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1064.     FF_ALLOC_OR_GOTO(c, c->chrPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1065.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1066.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1067.     //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)

  1068.     /* align at 16 bytes for AltiVec */

  1069.     for (i=0; i<c->vLumBufSize; i++) {

  1070.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], VOF+1, fail);

  1071.         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];

  1072.     }

  1073.     for (i=0; i<c->vChrBufSize; i++) {

  1074.         FF_ALLOC_OR_GOTO(c, c->chrPixBuf[i+c->vChrBufSize], (VOF+1)*2, fail);

  1075.         c->chrPixBuf[i] = c->chrPixBuf[i+c->vChrBufSize];

  1076.     }

  1077.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1078.         for (i=0; i<c->vLumBufSize; i++) {

  1079.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], VOF+1, fail);

  1080.             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];

  1081.         }

  1082. 

  1083.     //try to avoid drawing green stuff between the right end and the stride end

  1084.     for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);

  1085. 

  1086.     assert(2*VOFW == VOF);

  1087. 

  1088.     assert(c->chrDstH <= dstH);

  1089. 

  1090.     if (flags&SWS_PRINT_INFO) {

  1091.         if (flags&SWS_FAST_BILINEAR)

  1092.             av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1093.         else if (flags&SWS_BILINEAR)

  1094.             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1095.         else if (flags&SWS_BICUBIC)

  1096.             av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1097.         else if (flags&SWS_X)

  1098.             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1099.         else if (flags&SWS_POINT)

  1100.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1101.         else if (flags&SWS_AREA)

  1102.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1103.         else if (flags&SWS_BICUBLIN)

  1104.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1105.         else if (flags&SWS_GAUSS)

  1106.             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1107.         else if (flags&SWS_SINC)

  1108.             av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1109.         else if (flags&SWS_LANCZOS)

  1110.             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1111.         else if (flags&SWS_SPLINE)

  1112.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1113.         else

  1114.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1115. 

  1116.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1117.                sws_format_name(srcFormat),

  1118. #ifdef DITHER1XBPP

  1119.                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ? "dithered " : "",

  1120. #else

  1121.                "",

  1122. #endif

  1123.                sws_format_name(dstFormat));

  1124. 

  1125.         if (flags & SWS_CPU_CAPS_MMX2)

  1126.             av_log(c, AV_LOG_INFO, "using MMX2\n");

  1127.         else if (flags & SWS_CPU_CAPS_3DNOW)

  1128.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1129.         else if (flags & SWS_CPU_CAPS_MMX)

  1130.             av_log(c, AV_LOG_INFO, "using MMX\n");

  1131.         else if (flags & SWS_CPU_CAPS_ALTIVEC)

  1132.             av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1133.         else

  1134.             av_log(c, AV_LOG_INFO, "using C\n");

  1135. 

  1136.         if (flags & SWS_CPU_CAPS_MMX) {

  1137.             if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))

  1138.                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");

  1139.             else {

  1140.                 if (c->hLumFilterSize==4)

  1141.                     av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");

  1142.                 else if (c->hLumFilterSize==8)

  1143.                     av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");

  1144.                 else

  1145.                     av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");

  1146. 

  1147.                 if (c->hChrFilterSize==4)

  1148.                     av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");

  1149.                 else if (c->hChrFilterSize==8)

  1150.                     av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");

  1151.                 else

  1152.                     av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");

  1153.             }

  1154.         } else {

  1155. #if ARCH_X86

  1156.             av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");

  1157. #else

  1158.             if (flags & SWS_FAST_BILINEAR)

  1159.                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");

  1160.             else

  1161.                 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");

  1162. #endif

  1163.         }

  1164.         if (isPlanarYUV(dstFormat)) {

  1165.             if (c->vLumFilterSize==1)

  1166.                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1167.             else

  1168.                 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1169.         } else {

  1170.             if (c->vLumFilterSize==1 && c->vChrFilterSize==2)

  1171.                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"

  1172.                        "      2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1173.             else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)

  1174.                 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1175.             else

  1176.                 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1177.         }

  1178. 

  1179.         if (dstFormat==PIX_FMT_BGR24)

  1180.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",

  1181.                    (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));

  1182.         else if (dstFormat==PIX_FMT_RGB32)

  1183.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1184.         else if (dstFormat==PIX_FMT_BGR565)

  1185.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1186.         else if (dstFormat==PIX_FMT_BGR555)

  1187.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1188. 

  1189.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1190.         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1191.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1192.         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1193.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1194.     }

  1195. 

  1196.     c->swScale= ff_getSwsFunc(c);

  1197.     return c;

  1198. 

  1199. fail:

  1200.     sws_freeContext(c);

  1201.     return NULL;

  1202. }

  1203. 

  1204. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1205.                                 float lumaSharpen, float chromaSharpen,

  1206.                                 float chromaHShift, float chromaVShift,

  1207.                                 int verbose)

  1208. {

  1209.     SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  1210.     if (!filter)

  1211.         return NULL;

  1212. 

  1213.     if (lumaGBlur!=0.0) {

  1214.         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  1215.         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  1216.     } else {

  1217.         filter->lumH= sws_getIdentityVec();

  1218.         filter->lumV= sws_getIdentityVec();

  1219.     }

  1220. 

  1221.     if (chromaGBlur!=0.0) {

  1222.         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  1223.         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  1224.     } else {

  1225.         filter->chrH= sws_getIdentityVec();

  1226.         filter->chrV= sws_getIdentityVec();

  1227.     }

  1228. 

  1229.     if (chromaSharpen!=0.0) {

  1230.         SwsVector *id= sws_getIdentityVec();

  1231.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1232.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1233.         sws_addVec(filter->chrH, id);

  1234.         sws_addVec(filter->chrV, id);

  1235.         sws_freeVec(id);

  1236.     }

  1237. 

  1238.     if (lumaSharpen!=0.0) {

  1239.         SwsVector *id= sws_getIdentityVec();

  1240.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1241.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1242.         sws_addVec(filter->lumH, id);

  1243.         sws_addVec(filter->lumV, id);

  1244.         sws_freeVec(id);

  1245.     }

  1246. 

  1247.     if (chromaHShift != 0.0)

  1248.         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  1249. 

  1250.     if (chromaVShift != 0.0)

  1251.         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  1252. 

  1253.     sws_normalizeVec(filter->chrH, 1.0);

  1254.     sws_normalizeVec(filter->chrV, 1.0);

  1255.     sws_normalizeVec(filter->lumH, 1.0);

  1256.     sws_normalizeVec(filter->lumV, 1.0);

  1257. 

  1258.     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1259.     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1260. 

  1261.     return filter;

  1262. }

  1263. 

  1264. SwsVector *sws_allocVec(int length)

  1265. {

  1266.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1267.     if (!vec)

  1268.         return NULL;

  1269.     vec->length = length;

  1270.     vec->coeff  = av_malloc(sizeof(double) * length);

  1271.     if (!vec->coeff)

  1272.         av_freep(&vec);

  1273.     return vec;

  1274. }

  1275. 

  1276. SwsVector *sws_getGaussianVec(double variance, double quality)

  1277. {

  1278.     const int length= (int)(variance*quality + 0.5) | 1;

  1279.     int i;

  1280.     double middle= (length-1)*0.5;

  1281.     SwsVector *vec= sws_allocVec(length);

  1282. 

  1283.     if (!vec)

  1284.         return NULL;

  1285. 

  1286.     for (i=0; i<length; i++) {

  1287.         double dist= i-middle;

  1288.         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);

  1289.     }

  1290. 

  1291.     sws_normalizeVec(vec, 1.0);

  1292. 

  1293.     return vec;

  1294. }

  1295. 

  1296. SwsVector *sws_getConstVec(double c, int length)

  1297. {

  1298.     int i;

  1299.     SwsVector *vec= sws_allocVec(length);

  1300. 

  1301.     if (!vec)

  1302.         return NULL;

  1303. 

  1304.     for (i=0; i<length; i++)

  1305.         vec->coeff[i]= c;

  1306. 

  1307.     return vec;

  1308. }

  1309. 

  1310. SwsVector *sws_getIdentityVec(void)

  1311. {

  1312.     return sws_getConstVec(1.0, 1);

  1313. }

  1314. 

  1315. double sws_dcVec(SwsVector *a)

  1316. {

  1317.     int i;

  1318.     double sum=0;

  1319. 

  1320.     for (i=0; i<a->length; i++)

  1321.         sum+= a->coeff[i];

  1322. 

  1323.     return sum;

  1324. }

  1325. 

  1326. void sws_scaleVec(SwsVector *a, double scalar)

  1327. {

  1328.     int i;

  1329. 

  1330.     for (i=0; i<a->length; i++)

  1331.         a->coeff[i]*= scalar;

  1332. }

  1333. 

  1334. void sws_normalizeVec(SwsVector *a, double height)

  1335. {

  1336.     sws_scaleVec(a, height/sws_dcVec(a));

  1337. }

  1338. 

  1339. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1340. {

  1341.     int length= a->length + b->length - 1;

  1342.     int i, j;

  1343.     SwsVector *vec= sws_getConstVec(0.0, length);

  1344. 

  1345.     if (!vec)

  1346.         return NULL;

  1347. 

  1348.     for (i=0; i<a->length; i++) {

  1349.         for (j=0; j<b->length; j++) {

  1350.             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1351.         }

  1352.     }

  1353. 

  1354.     return vec;

  1355. }

  1356. 

  1357. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1358. {

  1359.     int length= FFMAX(a->length, b->length);

  1360.     int i;

  1361.     SwsVector *vec= sws_getConstVec(0.0, length);

  1362. 

  1363.     if (!vec)

  1364.         return NULL;

  1365. 

  1366.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1367.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];

  1368. 

  1369.     return vec;

  1370. }

  1371. 

  1372. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1373. {

  1374.     int length= FFMAX(a->length, b->length);

  1375.     int i;

  1376.     SwsVector *vec= sws_getConstVec(0.0, length);

  1377. 

  1378.     if (!vec)

  1379.         return NULL;

  1380. 

  1381.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1382.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1383. 

  1384.     return vec;

  1385. }

  1386. 

  1387. /* shift left / or right if "shift" is negative */

  1388. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1389. {

  1390.     int length= a->length + FFABS(shift)*2;

  1391.     int i;

  1392.     SwsVector *vec= sws_getConstVec(0.0, length);

  1393. 

  1394.     if (!vec)

  1395.         return NULL;

  1396. 

  1397.     for (i=0; i<a->length; i++) {

  1398.         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1399.     }

  1400. 

  1401.     return vec;

  1402. }

  1403. 

  1404. void sws_shiftVec(SwsVector *a, int shift)

  1405. {

  1406.     SwsVector *shifted= sws_getShiftedVec(a, shift);

  1407.     av_free(a->coeff);

  1408.     a->coeff= shifted->coeff;

  1409.     a->length= shifted->length;

  1410.     av_free(shifted);

  1411. }

  1412. 

  1413. void sws_addVec(SwsVector *a, SwsVector *b)

  1414. {

  1415.     SwsVector *sum= sws_sumVec(a, b);

  1416.     av_free(a->coeff);

  1417.     a->coeff= sum->coeff;

  1418.     a->length= sum->length;

  1419.     av_free(sum);

  1420. }

  1421. 

  1422. void sws_subVec(SwsVector *a, SwsVector *b)

  1423. {

  1424.     SwsVector *diff= sws_diffVec(a, b);

  1425.     av_free(a->coeff);

  1426.     a->coeff= diff->coeff;

  1427.     a->length= diff->length;

  1428.     av_free(diff);

  1429. }

  1430. 

  1431. void sws_convVec(SwsVector *a, SwsVector *b)

  1432. {

  1433.     SwsVector *conv= sws_getConvVec(a, b);

  1434.     av_free(a->coeff);

  1435.     a->coeff= conv->coeff;

  1436.     a->length= conv->length;

  1437.     av_free(conv);

  1438. }

  1439. 

  1440. SwsVector *sws_cloneVec(SwsVector *a)

  1441. {

  1442.     int i;

  1443.     SwsVector *vec= sws_allocVec(a->length);

  1444. 

  1445.     if (!vec)

  1446.         return NULL;

  1447. 

  1448.     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];

  1449. 

  1450.     return vec;

  1451. }

  1452. 

  1453. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1454. {

  1455.     int i;

  1456.     double max=0;

  1457.     double min=0;

  1458.     double range;

  1459. 

  1460.     for (i=0; i<a->length; i++)

  1461.         if (a->coeff[i]>max) max= a->coeff[i];

  1462. 

  1463.     for (i=0; i<a->length; i++)

  1464.         if (a->coeff[i]<min) min= a->coeff[i];

  1465. 

  1466.     range= max - min;

  1467. 

  1468.     for (i=0; i<a->length; i++) {

  1469.         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1470.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1471.         for (;x>0; x--) av_log(log_ctx, log_level, " ");

  1472.         av_log(log_ctx, log_level, "|\n");

  1473.     }

  1474. }

  1475. 

  1476. #if LIBSWSCALE_VERSION_MAJOR < 1

  1477. void sws_printVec(SwsVector *a)

  1478. {

  1479.     sws_printVec2(a, NULL, AV_LOG_DEBUG);

  1480. }

  1481. #endif

  1482. 

  1483. void sws_freeVec(SwsVector *a)

  1484. {

  1485.     if (!a) return;

  1486.     av_freep(&a->coeff);

  1487.     a->length=0;

  1488.     av_free(a);

  1489. }

  1490. 

  1491. void sws_freeFilter(SwsFilter *filter)

  1492. {

  1493.     if (!filter) return;

  1494. 

  1495.     if (filter->lumH) sws_freeVec(filter->lumH);

  1496.     if (filter->lumV) sws_freeVec(filter->lumV);

  1497.     if (filter->chrH) sws_freeVec(filter->chrH);

  1498.     if (filter->chrV) sws_freeVec(filter->chrV);

  1499.     av_free(filter);

  1500. }

  1501. 

  1502. void sws_freeContext(SwsContext *c)

  1503. {

  1504.     int i;

  1505.     if (!c) return;

  1506. 

  1507.     if (c->lumPixBuf) {

  1508.         for (i=0; i<c->vLumBufSize; i++)

  1509.             av_freep(&c->lumPixBuf[i]);

  1510.         av_freep(&c->lumPixBuf);

  1511.     }

  1512. 

  1513.     if (c->chrPixBuf) {

  1514.         for (i=0; i<c->vChrBufSize; i++)

  1515.             av_freep(&c->chrPixBuf[i]);

  1516.         av_freep(&c->chrPixBuf);

  1517.     }

  1518. 

  1519.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1520.         for (i=0; i<c->vLumBufSize; i++)

  1521.             av_freep(&c->alpPixBuf[i]);

  1522.         av_freep(&c->alpPixBuf);

  1523.     }

  1524. 

  1525.     av_freep(&c->vLumFilter);

  1526.     av_freep(&c->vChrFilter);

  1527.     av_freep(&c->hLumFilter);

  1528.     av_freep(&c->hChrFilter);

  1529. #if ARCH_PPC && (HAVE_ALTIVEC || CONFIG_RUNTIME_CPUDETECT)

  1530.     av_freep(&c->vYCoeffsBank);

  1531.     av_freep(&c->vCCoeffsBank);

  1532. #endif

  1533. 

  1534.     av_freep(&c->vLumFilterPos);

  1535.     av_freep(&c->vChrFilterPos);

  1536.     av_freep(&c->hLumFilterPos);

  1537.     av_freep(&c->hChrFilterPos);

  1538. 

  1539. #if ARCH_X86 && CONFIG_GPL

  1540. #ifdef MAP_ANONYMOUS

  1541.     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1542.     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1543. #elif HAVE_VIRTUALALLOC

  1544.     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, MEM_RELEASE);

  1545.     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, MEM_RELEASE);

  1546. #else

  1547.     av_free(c->lumMmx2FilterCode);

  1548.     av_free(c->chrMmx2FilterCode);

  1549. #endif

  1550.     c->lumMmx2FilterCode=NULL;

  1551.     c->chrMmx2FilterCode=NULL;

  1552. #endif /* ARCH_X86 && CONFIG_GPL */

  1553. 

  1554.     av_freep(&c->yuvTable);

  1555. 

  1556.     av_free(c);

  1557. }

  1558. 

  1559. struct SwsContext *sws_getCachedContext(struct SwsContext *context,

  1560.                                         int srcW, int srcH, enum PixelFormat srcFormat,

  1561.                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1562.                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1563. {

  1564.     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};

  1565. 

  1566.     if (!param)

  1567.         param = default_param;

  1568. 

  1569.     if (context &&

  1570.         (context->srcW      != srcW      ||

  1571.          context->srcH      != srcH      ||

  1572.          context->srcFormat != srcFormat ||

  1573.          context->dstW      != dstW      ||

  1574.          context->dstH      != dstH      ||

  1575.          context->dstFormat != dstFormat ||

  1576.          context->flags     != flags     ||

  1577.          context->param[0]  != param[0]  ||

  1578.          context->param[1]  != param[1])) {

  1579.         sws_freeContext(context);

  1580.         context = NULL;

  1581.     }

  1582. 

  1583.     if (!context) {

  1584.         return sws_getContext(srcW, srcH, srcFormat,

  1585.                               dstW, dstH, dstFormat, flags,

  1586.                               srcFilter, dstFilter, param);

  1587.     }

  1588.     return context;

  1589. }