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.
26087 Commits
32 Branches
944MB
Tree: 9f694f0b3d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '9f694f0b3d'
${ noResults }
FFmpeg/libswscale/utils.c

1573 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

  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/avutil.h"

  45. #include "libavutil/bswap.h"

  46. #include "libavutil/pixdesc.h"

  47. 

  48. unsigned swscale_version(void)

  49. {

  50.     return LIBSWSCALE_VERSION_INT;

  51. }

  52. 

  53. const char *swscale_configuration(void)

  54. {

  55.     return FFMPEG_CONFIGURATION;

  56. }

  57. 

  58. const char *swscale_license(void)

  59. {

  60. #define LICENSE_PREFIX "libswscale license: "

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

  62. }

  63. 

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

  65. 

  66. #define isSupportedIn(x)    (       \

  67.            (x)==PIX_FMT_YUV420P     \

  68.         || (x)==PIX_FMT_YUVA420P    \

  69.         || (x)==PIX_FMT_YUYV422     \

  70.         || (x)==PIX_FMT_UYVY422     \

  71.         || (x)==PIX_FMT_RGB48BE     \

  72.         || (x)==PIX_FMT_RGB48LE     \

  73.         || (x)==PIX_FMT_RGB32       \

  74.         || (x)==PIX_FMT_RGB32_1     \

  75.         || (x)==PIX_FMT_BGR24       \

  76.         || (x)==PIX_FMT_BGR565      \

  77.         || (x)==PIX_FMT_BGR555      \

  78.         || (x)==PIX_FMT_BGR32       \

  79.         || (x)==PIX_FMT_BGR32_1     \

  80.         || (x)==PIX_FMT_RGB24       \

  81.         || (x)==PIX_FMT_RGB565      \

  82.         || (x)==PIX_FMT_RGB555      \

  83.         || (x)==PIX_FMT_GRAY8       \

  84.         || (x)==PIX_FMT_YUV410P     \

  85.         || (x)==PIX_FMT_YUV440P     \

  86.         || (x)==PIX_FMT_NV12        \

  87.         || (x)==PIX_FMT_NV21        \

  88.         || (x)==PIX_FMT_GRAY16BE    \

  89.         || (x)==PIX_FMT_GRAY16LE    \

  90.         || (x)==PIX_FMT_YUV444P     \

  91.         || (x)==PIX_FMT_YUV422P     \

  92.         || (x)==PIX_FMT_YUV411P     \

  93.         || (x)==PIX_FMT_YUVJ420P    \

  94.         || (x)==PIX_FMT_YUVJ422P    \

  95.         || (x)==PIX_FMT_YUVJ440P    \

  96.         || (x)==PIX_FMT_YUVJ444P    \

  97.         || (x)==PIX_FMT_PAL8        \

  98.         || (x)==PIX_FMT_BGR8        \

  99.         || (x)==PIX_FMT_RGB8        \

  100.         || (x)==PIX_FMT_BGR4_BYTE   \

  101.         || (x)==PIX_FMT_RGB4_BYTE   \

  102.         || (x)==PIX_FMT_YUV440P     \

  103.         || (x)==PIX_FMT_MONOWHITE   \

  104.         || (x)==PIX_FMT_MONOBLACK   \

  105.         || (x)==PIX_FMT_YUV420P16LE   \

  106.         || (x)==PIX_FMT_YUV422P16LE   \

  107.         || (x)==PIX_FMT_YUV444P16LE   \

  108.         || (x)==PIX_FMT_YUV420P16BE   \

  109.         || (x)==PIX_FMT_YUV422P16BE   \

  110.         || (x)==PIX_FMT_YUV444P16BE   \

  111.     )

  112. 

  113. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  114. {

  115.     return isSupportedIn(pix_fmt);

  116. }

  117. 

  118. #define isSupportedOut(x)   (       \

  119.            (x)==PIX_FMT_YUV420P     \

  120.         || (x)==PIX_FMT_YUVA420P    \

  121.         || (x)==PIX_FMT_YUYV422     \

  122.         || (x)==PIX_FMT_UYVY422     \

  123.         || (x)==PIX_FMT_YUV444P     \

  124.         || (x)==PIX_FMT_YUV422P     \

  125.         || (x)==PIX_FMT_YUV411P     \

  126.         || (x)==PIX_FMT_YUVJ420P    \

  127.         || (x)==PIX_FMT_YUVJ422P    \

  128.         || (x)==PIX_FMT_YUVJ440P    \

  129.         || (x)==PIX_FMT_YUVJ444P    \

  130.         || isAnyRGB(x)              \

  131.         || (x)==PIX_FMT_NV12        \

  132.         || (x)==PIX_FMT_NV21        \

  133.         || (x)==PIX_FMT_GRAY16BE    \

  134.         || (x)==PIX_FMT_GRAY16LE    \

  135.         || (x)==PIX_FMT_GRAY8       \

  136.         || (x)==PIX_FMT_YUV410P     \

  137.         || (x)==PIX_FMT_YUV440P     \

  138.         || (x)==PIX_FMT_YUV420P16LE   \

  139.         || (x)==PIX_FMT_YUV422P16LE   \

  140.         || (x)==PIX_FMT_YUV444P16LE   \

  141.         || (x)==PIX_FMT_YUV420P16BE   \

  142.         || (x)==PIX_FMT_YUV422P16BE   \

  143.         || (x)==PIX_FMT_YUV444P16BE   \

  144.     )

  145. 

  146. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  147. {

  148.     return isSupportedOut(pix_fmt);

  149. }

  150. 

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

  152. 

  153. const char *sws_format_name(enum PixelFormat format)

  154. {

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

  156.         return av_pix_fmt_descriptors[format].name;

  157.     else

  158.         return "Unknown format";

  159. }

  160. 

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

  162. {

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

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

  165.     else           return getSplineCoeff(        0.0,

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

  167.                                                  c + 3.0*d,

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

  169.                                          dist-1.0);

  170. }

  171. 

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

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

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

  175. {

  176.     int i;

  177.     int filterSize;

  178.     int filter2Size;

  179.     int minFilterSize;

  180.     int64_t *filter=NULL;

  181.     int64_t *filter2=NULL;

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

  183.     int ret= -1;

  184. #if ARCH_X86

  185.     if (flags & SWS_CPU_CAPS_MMX)

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

  187. #endif

  188. 

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

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

  191. 

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

  193.         int i;

  194.         filterSize= 1;

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

  196. 

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

  198.             filter[i*filterSize]= fone;

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

  200.         }

  201. 

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

  203.         int i;

  204.         int xDstInSrc;

  205.         filterSize= 1;

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

  207. 

  208.         xDstInSrc= xInc/2 - 0x8000;

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

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

  211. 

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

  213.             filter[i]= fone;

  214.             xDstInSrc+= xInc;

  215.         }

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

  217.         int i;

  218.         int xDstInSrc;

  219.         filterSize= 2;

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

  221. 

  222.         xDstInSrc= xInc/2 - 0x8000;

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

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

  225.             int j;

  226. 

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

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

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

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

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

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

  233.                 xx++;

  234.             }

  235.             xDstInSrc+= xInc;

  236.         }

  237.     } else {

  238.         int xDstInSrc;

  239.         int sizeFactor;

  240. 

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

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

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

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

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

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

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

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

  249.         else {

  250.             sizeFactor= 0; //GCC warning killer

  251.             assert(0);

  252.         }

  253. 

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

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

  256. 

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

  258. 

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

  260. 

  261.         xDstInSrc= xInc - 0x10000;

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

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

  264.             int j;

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

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

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

  268.                 double floatd;

  269.                 int64_t coeff;

  270. 

  271.                 if (xInc > 1<<16)

  272.                     d= d*dstW/srcW;

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

  274. 

  275.                 if (flags & SWS_BICUBIC) {

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

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

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

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

  280. 

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

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

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

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

  285.                     else

  286.                         coeff=0.0;

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

  288.                 }

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

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

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

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

  293.                 }*/

  294.                 else if (flags & SWS_X) {

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

  296.                     double c;

  297. 

  298.                     if (floatd<1.0)

  299.                         c = cos(floatd*M_PI);

  300.                     else

  301.                         c=-1.0;

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

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

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

  305.                 } else if (flags & SWS_AREA) {

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

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

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

  309.                     else coeff=0.0;

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

  311.                 } else if (flags & SWS_GAUSS) {

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

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

  314.                 } else if (flags & SWS_SINC) {

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

  316.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  320.                 } else if (flags & SWS_BILINEAR) {

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

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

  323.                     coeff *= fone >> 30;

  324.                 } else if (flags & SWS_SPLINE) {

  325.                     double p=-2.196152422706632;

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

  327.                 } else {

  328.                     coeff= 0.0; //GCC warning killer

  329.                     assert(0);

  330.                 }

  331. 

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

  333.                 xx++;

  334.             }

  335.             xDstInSrc+= 2*xInc;

  336.         }

  337.     }

  338. 

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

  340.        av_free(filter);

  341.     */

  342.     assert(filterSize>0);

  343.     filter2Size= filterSize;

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

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

  346.     assert(filter2Size>0);

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

  348. 

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

  350.         int j, k;

  351. 

  352.         if(srcFilter) {

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

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

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

  356.             }

  357.         } else {

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

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

  360.         }

  361.         //FIXME dstFilter

  362. 

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

  364.     }

  365.     av_freep(&filter);

  366. 

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

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

  369.     minFilterSize= 0;

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

  371.         int min= filter2Size;

  372.         int j;

  373.         int64_t cutOff=0.0;

  374. 

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

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

  377.             int k;

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

  379. 

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

  381. 

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

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

  384. 

  385.             // move filter coefficients left

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

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

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

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

  390.         }

  391. 

  392.         cutOff=0;

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

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

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

  396. 

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

  398.             min--;

  399.         }

  400. 

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

  402.     }

  403. 

  404.     if (flags & SWS_CPU_CAPS_ALTIVEC) {

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

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

  407.         if (minFilterSize < 5)

  408.             filterAlign = 4;

  409. 

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

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

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

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

  414.         // decent-sized vector.

  415.         if (minFilterSize < 3)

  416.             filterAlign = 1;

  417.     }

  418. 

  419.     if (flags & SWS_CPU_CAPS_MMX) {

  420.         // special case for unscaled vertical filtering

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

  422.             filterAlign= 1;

  423.     }

  424. 

  425.     assert(minFilterSize > 0);

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

  427.     assert(filterSize > 0);

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

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

  430.         goto fail;

  431.     *outFilterSize= filterSize;

  432. 

  433.     if (flags&SWS_PRINT_INFO)

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

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

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

  437.         int j;

  438. 

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

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

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

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

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

  444.         }

  445.     }

  446. 

  447.     //FIXME try to align filterPos if possible

  448. 

  449.     //fix borders

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

  451.         int j;

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

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

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

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

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

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

  458.             }

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

  460.         }

  461. 

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

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

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

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

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

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

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

  469.             }

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

  471.         }

  472.     }

  473. 

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

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

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

  477. 

  478.     /* normalize & store in outFilter */

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

  480.         int j;

  481.         int64_t error=0;

  482.         int64_t sum=0;

  483. 

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

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

  486.         }

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

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

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

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

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

  492.             error= v - intV*sum;

  493.         }

  494.     }

  495. 

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

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

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

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

  500.     }

  501. 

  502.     ret=0;

  503. fail:

  504.     av_free(filter);

  505.     av_free(filter2);

  506.     return ret;

  507. }

  508. 

  509. #if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT)

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

  511. {

  512.     uint8_t *fragmentA;

  513.     x86_reg imm8OfPShufW1A;

  514.     x86_reg imm8OfPShufW2A;

  515.     x86_reg fragmentLengthA;

  516.     uint8_t *fragmentB;

  517.     x86_reg imm8OfPShufW1B;

  518.     x86_reg imm8OfPShufW2B;

  519.     x86_reg fragmentLengthB;

  520.     int fragmentPos;

  521. 

  522.     int xpos, i;

  523. 

  524.     // create an optimized horizontal scaling routine

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

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

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

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

  529.      * of fragmentA is used.

  530.      */

  531. 

  532.     //code fragment

  533. 

  534.     __asm__ volatile(

  535.         "jmp                         9f                 \n\t"

  536.     // Begin

  537.         "0:                                             \n\t"

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

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

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

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

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

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

  544.         "1:                                             \n\t"

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

  546.         "2:                                             \n\t"

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

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

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

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

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

  552. 

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

  554. 

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

  556.     // End

  557.         "9:                                             \n\t"

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

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

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

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

  562.         "dec                         %1                 \n\t"

  563.         "dec                         %2                 \n\t"

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

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

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

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

  568. 

  569. 

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

  571.         "=r" (fragmentLengthA)

  572.     );

  573. 

  574.     __asm__ volatile(

  575.         "jmp                         9f                 \n\t"

  576.     // Begin

  577.         "0:                                             \n\t"

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

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

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

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

  582.         "1:                                             \n\t"

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

  584.         "2:                                             \n\t"

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

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

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

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

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

  590. 

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

  592. 

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

  594.     // End

  595.         "9:                                             \n\t"

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

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

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

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

  600.         "dec                         %1                 \n\t"

  601.         "dec                         %2                 \n\t"

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

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

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

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

  606. 

  607. 

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

  609.         "=r" (fragmentLengthB)

  610.     );

  611. 

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

  613.     fragmentPos=0;

  614. 

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

  616.         int xx=xpos>>16;

  617. 

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

  619.             int a=0;

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

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

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

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

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

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

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

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

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

  629.             int shift=0;

  630. 

  631.             if (filterCode) {

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

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

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

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

  636.                 filterPos[i/2]= xx;

  637. 

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

  639. 

  640.                 filterCode[fragmentPos + imm8OfPShufW1]=

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

  642.                 filterCode[fragmentPos + imm8OfPShufW2]=

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

  644. 

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

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

  647. 

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

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

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

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

  652.                 }

  653.             }

  654. 

  655.             fragmentPos+= fragmentLength;

  656. 

  657.             if (filterCode)

  658.                 filterCode[fragmentPos]= RET;

  659.         }

  660.         xpos+=xInc;

  661.     }

  662.     if (filterCode)

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

  664. 

  665.     return fragmentPos + 1;

  666. }

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

  668. 

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

  670. {

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

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

  673. }

  674. 

  675. static int update_flags_cpu(int flags);

  676. 

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

  678. {

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

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

  681. 

  682.     c->brightness= brightness;

  683.     c->contrast  = contrast;

  684.     c->saturation= saturation;

  685.     c->srcRange  = srcRange;

  686.     c->dstRange  = dstRange;

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

  688. 

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

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

  691.     c->flags = update_flags_cpu(c->flags);

  692. 

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

  694.     //FIXME factorize

  695. 

  696. #if HAVE_ALTIVEC

  697.     if (c->flags & SWS_CPU_CAPS_ALTIVEC)

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

  699. #endif

  700.     return 0;

  701. }

  702. 

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

  704. {

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

  706. 

  707.     *inv_table = c->srcColorspaceTable;

  708.     *table     = c->dstColorspaceTable;

  709.     *srcRange  = c->srcRange;

  710.     *dstRange  = c->dstRange;

  711.     *brightness= c->brightness;

  712.     *contrast  = c->contrast;

  713.     *saturation= c->saturation;

  714. 

  715.     return 0;

  716. }

  717. 

  718. static int handle_jpeg(enum PixelFormat *format)

  719. {

  720.     switch (*format) {

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

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

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

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

  725.     default:                                          return 0;

  726.     }

  727. }

  728. 

  729. static int update_flags_cpu(int flags)

  730. {

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

  732.     flags &= ~( SWS_CPU_CAPS_MMX

  733.                |SWS_CPU_CAPS_MMX2

  734.                |SWS_CPU_CAPS_3DNOW

  735.                |SWS_CPU_CAPS_SSE2

  736.                |SWS_CPU_CAPS_ALTIVEC

  737.                |SWS_CPU_CAPS_BFIN);

  738.     flags |= ff_hardcodedcpuflags();

  739. #endif /* CONFIG_RUNTIME_CPUDETECT */

  740.     return flags;

  741. }

  742. 

  743. SwsContext *sws_alloc_context(void){

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

  745. 

  746.     c->av_class = &sws_context_class;

  747. 

  748.     return c;

  749. }

  750. 

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

  752.     int i;

  753.     int usesVFilter, usesHFilter;

  754.     int unscaled;

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

  756.     int srcW= c->srcW;

  757.     int srcH= c->srcH;

  758.     int dstW= c->dstW;

  759.     int dstH= c->dstH;

  760.     int flags;

  761.     enum PixelFormat srcFormat= c->srcFormat;

  762.     enum PixelFormat dstFormat= c->dstFormat;

  763. 

  764.     flags= c->flags = update_flags_cpu(c->flags);

  765. #if ARCH_X86

  766.     if (flags & SWS_CPU_CAPS_MMX)

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

  768. #endif

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

  770. 

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

  772. 

  773.     if (!isSupportedIn(srcFormat)) {

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

  775.         return AVERROR(EINVAL);

  776.     }

  777.     if (!isSupportedOut(dstFormat)) {

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

  779.         return AVERROR(EINVAL);

  780.     }

  781. 

  782.     i= flags & ( SWS_POINT

  783.                 |SWS_AREA

  784.                 |SWS_BILINEAR

  785.                 |SWS_FAST_BILINEAR

  786.                 |SWS_BICUBIC

  787.                 |SWS_X

  788.                 |SWS_GAUSS

  789.                 |SWS_LANCZOS

  790.                 |SWS_SINC

  791.                 |SWS_SPLINE

  792.                 |SWS_BICUBLIN);

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

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

  795.         return AVERROR(EINVAL);

  796.     }

  797.     /* sanity check */

  798.     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

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

  800.                srcW, srcH, dstW, dstH);

  801.         return AVERROR(EINVAL);

  802.     }

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

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

  805.         return AVERROR(EINVAL);

  806.     }

  807. 

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

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

  810. 

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

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

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

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

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

  816. 

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

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

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

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

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

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

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

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

  825. 

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

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

  828. 

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

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

  831. 

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

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

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

  835. 

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

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

  838.       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8

  839.       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4

  840.       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE

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

  842.         c->chrSrcHSubSample=1;

  843. 

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

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

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

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

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

  849. 

  850.     /* unscaled special cases */

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

  852.         ff_get_unscaled_swscale(c);

  853. 

  854.         if (c->swScale) {

  855.             if (flags&SWS_PRINT_INFO)

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

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

  858.             return 0;

  859.         }

  860.     }

  861. 

  862.     if (flags & SWS_CPU_CAPS_MMX2) {

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

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

  865.             if (flags&SWS_PRINT_INFO)

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

  867.         }

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

  869.     }

  870.     else

  871.         c->canMMX2BeUsed=0;

  872. 

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

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

  875. 

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

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

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

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

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

  881.     // first and last pixel

  882.     if (flags&SWS_FAST_BILINEAR) {

  883.         if (c->canMMX2BeUsed) {

  884.             c->lumXInc+= 20;

  885.             c->chrXInc+= 20;

  886.         }

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

  888.         else if (flags & SWS_CPU_CAPS_MMX) {

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

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

  891.         }

  892.     }

  893. 

  894.     /* precalculate horizontal scaler filter coefficients */

  895.     {

  896. #if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT)

  897. // can't downscale !!!

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

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

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

  901. 

  902. #ifdef MAP_ANONYMOUS

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

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

  905. #elif HAVE_VIRTUALALLOC

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

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

  908. #else

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

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

  911. #endif

  912. 

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

  914.                 return AVERROR(ENOMEM);

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

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

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

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

  919. 

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

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

  922. 

  923. #ifdef MAP_ANONYMOUS

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

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

  926. #endif

  927.         } else

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

  929.         {

  930.             const int filterAlign=

  931.                 (flags & SWS_CPU_CAPS_MMX) ? 4 :

  932.                 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :

  933.                 1;

  934. 

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

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

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

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

  939.                 goto fail;

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

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

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

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

  944.                 goto fail;

  945.         }

  946.     } // initialize horizontal stuff

  947. 

  948.     /* precalculate vertical scaler filter coefficients */

  949.     {

  950.         const int filterAlign=

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

  952.             (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :

  953.             1;

  954. 

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

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

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

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

  959.             goto fail;

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

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

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

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

  964.             goto fail;

  965. 

  966. #if HAVE_ALTIVEC

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

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

  969. 

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

  971.             int j;

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

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

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

  975.         }

  976. 

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

  978.             int j;

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

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

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

  982.         }

  983. #endif

  984.     }

  985. 

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

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

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

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

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

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

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

  993. 

  994.         nextSlice>>= c->chrSrcVSubSample;

  995.         nextSlice<<= c->chrSrcVSubSample;

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

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

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

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

  1000.     }

  1001. 

  1002.     // allocate pixbufs (we use dynamic allocation because otherwise we would need to

  1003.     // allocate several megabytes to handle all possible cases)

  1004.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1005.     FF_ALLOC_OR_GOTO(c, c->chrPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1006.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1007.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1008.     //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)

  1009.     /* align at 16 bytes for AltiVec */

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

  1011.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], VOF+1, fail);

  1012.         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];

  1013.     }

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

  1015.         FF_ALLOC_OR_GOTO(c, c->chrPixBuf[i+c->vChrBufSize], (VOF+1)*2, fail);

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

  1017.     }

  1018.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1019.         for (i=0; i<c->vLumBufSize; i++) {

  1020.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], VOF+1, fail);

  1021.             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];

  1022.         }

  1023. 

  1024.     //try to avoid drawing green stuff between the right end and the stride end

  1025.     for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);

  1026. 

  1027.     assert(2*VOFW == VOF);

  1028. 

  1029.     assert(c->chrDstH <= dstH);

  1030. 

  1031.     if (flags&SWS_PRINT_INFO) {

  1032.         if (flags&SWS_FAST_BILINEAR)

  1033.             av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1034.         else if (flags&SWS_BILINEAR)

  1035.             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1036.         else if (flags&SWS_BICUBIC)

  1037.             av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1038.         else if (flags&SWS_X)

  1039.             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1040.         else if (flags&SWS_POINT)

  1041.             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1042.         else if (flags&SWS_AREA)

  1043.             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1044.         else if (flags&SWS_BICUBLIN)

  1045.             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1046.         else if (flags&SWS_GAUSS)

  1047.             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1048.         else if (flags&SWS_SINC)

  1049.             av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1050.         else if (flags&SWS_LANCZOS)

  1051.             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1052.         else if (flags&SWS_SPLINE)

  1053.             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1054.         else

  1055.             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1056. 

  1057.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1058.                sws_format_name(srcFormat),

  1059. #ifdef DITHER1XBPP

  1060.                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||

  1061.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1062.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",

  1063. #else

  1064.                "",

  1065. #endif

  1066.                sws_format_name(dstFormat));

  1067. 

  1068.         if (flags & SWS_CPU_CAPS_MMX2)

  1069.             av_log(c, AV_LOG_INFO, "using MMX2\n");

  1070.         else if (flags & SWS_CPU_CAPS_3DNOW)

  1071.             av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1072.         else if (flags & SWS_CPU_CAPS_MMX)

  1073.             av_log(c, AV_LOG_INFO, "using MMX\n");

  1074.         else if (flags & SWS_CPU_CAPS_ALTIVEC)

  1075.             av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1076.         else

  1077.             av_log(c, AV_LOG_INFO, "using C\n");

  1078. 

  1079.         if (flags & SWS_CPU_CAPS_MMX) {

  1080.             if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))

  1081.                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");

  1082.             else {

  1083.                 if (c->hLumFilterSize==4)

  1084.                     av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");

  1085.                 else if (c->hLumFilterSize==8)

  1086.                     av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");

  1087.                 else

  1088.                     av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");

  1089. 

  1090.                 if (c->hChrFilterSize==4)

  1091.                     av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");

  1092.                 else if (c->hChrFilterSize==8)

  1093.                     av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");

  1094.                 else

  1095.                     av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");

  1096.             }

  1097.         } else {

  1098. #if ARCH_X86

  1099.             av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");

  1100. #else

  1101.             if (flags & SWS_FAST_BILINEAR)

  1102.                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");

  1103.             else

  1104.                 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");

  1105. #endif

  1106.         }

  1107.         if (isPlanarYUV(dstFormat)) {

  1108.             if (c->vLumFilterSize==1)

  1109.                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1110.             else

  1111.                 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1112.         } else {

  1113.             if (c->vLumFilterSize==1 && c->vChrFilterSize==2)

  1114.                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"

  1115.                        "      2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1116.             else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)

  1117.                 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1118.             else

  1119.                 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1120.         }

  1121. 

  1122.         if (dstFormat==PIX_FMT_BGR24)

  1123.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",

  1124.                    (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));

  1125.         else if (dstFormat==PIX_FMT_RGB32)

  1126.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1127.         else if (dstFormat==PIX_FMT_BGR565)

  1128.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1129.         else if (dstFormat==PIX_FMT_BGR555)

  1130.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1131.         else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1132.                  dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)

  1133.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1134. 

  1135.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1136.         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1137.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1138.         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1139.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1140.     }

  1141. 

  1142.     c->swScale= ff_getSwsFunc(c);

  1143.     return 0;

  1144. fail: //FIXME replace things by appropriate error codes

  1145.     return -1;

  1146. }

  1147. 

  1148. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1149.                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1150.                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1151. {

  1152.     SwsContext *c;

  1153. 

  1154.     if(!(c=sws_alloc_context()))

  1155.         return NULL;

  1156. 

  1157.     c->flags= flags;

  1158.     c->srcW= srcW;

  1159.     c->srcH= srcH;

  1160.     c->dstW= dstW;

  1161.     c->dstH= dstH;

  1162.     c->srcRange = handle_jpeg(&srcFormat);

  1163.     c->dstRange = handle_jpeg(&dstFormat);

  1164.     c->srcFormat= srcFormat;

  1165.     c->dstFormat= dstFormat;

  1166. 

  1167.     if (param) {

  1168.         c->param[0] = param[0];

  1169.         c->param[1] = param[1];

  1170.     } else {

  1171.         c->param[0] =

  1172.         c->param[1] = SWS_PARAM_DEFAULT;

  1173.     }

  1174.     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);

  1175. 

  1176.     if(sws_init_context(c, srcFilter, dstFilter) < 0){

  1177.         sws_freeContext(c);

  1178.         return NULL;

  1179.     }

  1180. 

  1181.     return c;

  1182. }

  1183. 

  1184. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1185.                                 float lumaSharpen, float chromaSharpen,

  1186.                                 float chromaHShift, float chromaVShift,

  1187.                                 int verbose)

  1188. {

  1189.     SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  1190.     if (!filter)

  1191.         return NULL;

  1192. 

  1193.     if (lumaGBlur!=0.0) {

  1194.         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  1195.         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  1196.     } else {

  1197.         filter->lumH= sws_getIdentityVec();

  1198.         filter->lumV= sws_getIdentityVec();

  1199.     }

  1200. 

  1201.     if (chromaGBlur!=0.0) {

  1202.         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  1203.         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  1204.     } else {

  1205.         filter->chrH= sws_getIdentityVec();

  1206.         filter->chrV= sws_getIdentityVec();

  1207.     }

  1208. 

  1209.     if (chromaSharpen!=0.0) {

  1210.         SwsVector *id= sws_getIdentityVec();

  1211.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1212.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1213.         sws_addVec(filter->chrH, id);

  1214.         sws_addVec(filter->chrV, id);

  1215.         sws_freeVec(id);

  1216.     }

  1217. 

  1218.     if (lumaSharpen!=0.0) {

  1219.         SwsVector *id= sws_getIdentityVec();

  1220.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1221.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1222.         sws_addVec(filter->lumH, id);

  1223.         sws_addVec(filter->lumV, id);

  1224.         sws_freeVec(id);

  1225.     }

  1226. 

  1227.     if (chromaHShift != 0.0)

  1228.         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  1229. 

  1230.     if (chromaVShift != 0.0)

  1231.         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  1232. 

  1233.     sws_normalizeVec(filter->chrH, 1.0);

  1234.     sws_normalizeVec(filter->chrV, 1.0);

  1235.     sws_normalizeVec(filter->lumH, 1.0);

  1236.     sws_normalizeVec(filter->lumV, 1.0);

  1237. 

  1238.     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1239.     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1240. 

  1241.     return filter;

  1242. }

  1243. 

  1244. SwsVector *sws_allocVec(int length)

  1245. {

  1246.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1247.     if (!vec)

  1248.         return NULL;

  1249.     vec->length = length;

  1250.     vec->coeff  = av_malloc(sizeof(double) * length);

  1251.     if (!vec->coeff)

  1252.         av_freep(&vec);

  1253.     return vec;

  1254. }

  1255. 

  1256. SwsVector *sws_getGaussianVec(double variance, double quality)

  1257. {

  1258.     const int length= (int)(variance*quality + 0.5) | 1;

  1259.     int i;

  1260.     double middle= (length-1)*0.5;

  1261.     SwsVector *vec= sws_allocVec(length);

  1262. 

  1263.     if (!vec)

  1264.         return NULL;

  1265. 

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

  1267.         double dist= i-middle;

  1268.         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);

  1269.     }

  1270. 

  1271.     sws_normalizeVec(vec, 1.0);

  1272. 

  1273.     return vec;

  1274. }

  1275. 

  1276. SwsVector *sws_getConstVec(double c, int length)

  1277. {

  1278.     int i;

  1279.     SwsVector *vec= sws_allocVec(length);

  1280. 

  1281.     if (!vec)

  1282.         return NULL;

  1283. 

  1284.     for (i=0; i<length; i++)

  1285.         vec->coeff[i]= c;

  1286. 

  1287.     return vec;

  1288. }

  1289. 

  1290. SwsVector *sws_getIdentityVec(void)

  1291. {

  1292.     return sws_getConstVec(1.0, 1);

  1293. }

  1294. 

  1295. static double sws_dcVec(SwsVector *a)

  1296. {

  1297.     int i;

  1298.     double sum=0;

  1299. 

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

  1301.         sum+= a->coeff[i];

  1302. 

  1303.     return sum;

  1304. }

  1305. 

  1306. void sws_scaleVec(SwsVector *a, double scalar)

  1307. {

  1308.     int i;

  1309. 

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

  1311.         a->coeff[i]*= scalar;

  1312. }

  1313. 

  1314. void sws_normalizeVec(SwsVector *a, double height)

  1315. {

  1316.     sws_scaleVec(a, height/sws_dcVec(a));

  1317. }

  1318. 

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

  1320. {

  1321.     int length= a->length + b->length - 1;

  1322.     int i, j;

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

  1324. 

  1325.     if (!vec)

  1326.         return NULL;

  1327. 

  1328.     for (i=0; i<a->length; i++) {

  1329.         for (j=0; j<b->length; j++) {

  1330.             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1331.         }

  1332.     }

  1333. 

  1334.     return vec;

  1335. }

  1336. 

  1337. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1338. {

  1339.     int length= FFMAX(a->length, b->length);

  1340.     int i;

  1341.     SwsVector *vec= sws_getConstVec(0.0, length);

  1342. 

  1343.     if (!vec)

  1344.         return NULL;

  1345. 

  1346.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

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

  1348. 

  1349.     return vec;

  1350. }

  1351. 

  1352. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1353. {

  1354.     int length= FFMAX(a->length, b->length);

  1355.     int i;

  1356.     SwsVector *vec= sws_getConstVec(0.0, length);

  1357. 

  1358.     if (!vec)

  1359.         return NULL;

  1360. 

  1361.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1362.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1363. 

  1364.     return vec;

  1365. }

  1366. 

  1367. /* shift left / or right if "shift" is negative */

  1368. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1369. {

  1370.     int length= a->length + FFABS(shift)*2;

  1371.     int i;

  1372.     SwsVector *vec= sws_getConstVec(0.0, length);

  1373. 

  1374.     if (!vec)

  1375.         return NULL;

  1376. 

  1377.     for (i=0; i<a->length; i++) {

  1378.         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1379.     }

  1380. 

  1381.     return vec;

  1382. }

  1383. 

  1384. void sws_shiftVec(SwsVector *a, int shift)

  1385. {

  1386.     SwsVector *shifted= sws_getShiftedVec(a, shift);

  1387.     av_free(a->coeff);

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

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

  1390.     av_free(shifted);

  1391. }

  1392. 

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

  1394. {

  1395.     SwsVector *sum= sws_sumVec(a, b);

  1396.     av_free(a->coeff);

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

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

  1399.     av_free(sum);

  1400. }

  1401. 

  1402. void sws_subVec(SwsVector *a, SwsVector *b)

  1403. {

  1404.     SwsVector *diff= sws_diffVec(a, b);

  1405.     av_free(a->coeff);

  1406.     a->coeff= diff->coeff;

  1407.     a->length= diff->length;

  1408.     av_free(diff);

  1409. }

  1410. 

  1411. void sws_convVec(SwsVector *a, SwsVector *b)

  1412. {

  1413.     SwsVector *conv= sws_getConvVec(a, b);

  1414.     av_free(a->coeff);

  1415.     a->coeff= conv->coeff;

  1416.     a->length= conv->length;

  1417.     av_free(conv);

  1418. }

  1419. 

  1420. SwsVector *sws_cloneVec(SwsVector *a)

  1421. {

  1422.     int i;

  1423.     SwsVector *vec= sws_allocVec(a->length);

  1424. 

  1425.     if (!vec)

  1426.         return NULL;

  1427. 

  1428.     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];

  1429. 

  1430.     return vec;

  1431. }

  1432. 

  1433. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1434. {

  1435.     int i;

  1436.     double max=0;

  1437.     double min=0;

  1438.     double range;

  1439. 

  1440.     for (i=0; i<a->length; i++)

  1441.         if (a->coeff[i]>max) max= a->coeff[i];

  1442. 

  1443.     for (i=0; i<a->length; i++)

  1444.         if (a->coeff[i]<min) min= a->coeff[i];

  1445. 

  1446.     range= max - min;

  1447. 

  1448.     for (i=0; i<a->length; i++) {

  1449.         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1450.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1451.         for (;x>0; x--) av_log(log_ctx, log_level, " ");

  1452.         av_log(log_ctx, log_level, "|\n");

  1453.     }

  1454. }

  1455. 

  1456. #if LIBSWSCALE_VERSION_MAJOR < 1

  1457. void sws_printVec(SwsVector *a)

  1458. {

  1459.     sws_printVec2(a, NULL, AV_LOG_DEBUG);

  1460. }

  1461. #endif

  1462. 

  1463. void sws_freeVec(SwsVector *a)

  1464. {

  1465.     if (!a) return;

  1466.     av_freep(&a->coeff);

  1467.     a->length=0;

  1468.     av_free(a);

  1469. }

  1470. 

  1471. void sws_freeFilter(SwsFilter *filter)

  1472. {

  1473.     if (!filter) return;

  1474. 

  1475.     if (filter->lumH) sws_freeVec(filter->lumH);

  1476.     if (filter->lumV) sws_freeVec(filter->lumV);

  1477.     if (filter->chrH) sws_freeVec(filter->chrH);

  1478.     if (filter->chrV) sws_freeVec(filter->chrV);

  1479.     av_free(filter);

  1480. }

  1481. 

  1482. void sws_freeContext(SwsContext *c)

  1483. {

  1484.     int i;

  1485.     if (!c) return;

  1486. 

  1487.     if (c->lumPixBuf) {

  1488.         for (i=0; i<c->vLumBufSize; i++)

  1489.             av_freep(&c->lumPixBuf[i]);

  1490.         av_freep(&c->lumPixBuf);

  1491.     }

  1492. 

  1493.     if (c->chrPixBuf) {

  1494.         for (i=0; i<c->vChrBufSize; i++)

  1495.             av_freep(&c->chrPixBuf[i]);

  1496.         av_freep(&c->chrPixBuf);

  1497.     }

  1498. 

  1499.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1500.         for (i=0; i<c->vLumBufSize; i++)

  1501.             av_freep(&c->alpPixBuf[i]);

  1502.         av_freep(&c->alpPixBuf);

  1503.     }

  1504. 

  1505.     av_freep(&c->vLumFilter);

  1506.     av_freep(&c->vChrFilter);

  1507.     av_freep(&c->hLumFilter);

  1508.     av_freep(&c->hChrFilter);

  1509. #if HAVE_ALTIVEC

  1510.     av_freep(&c->vYCoeffsBank);

  1511.     av_freep(&c->vCCoeffsBank);

  1512. #endif

  1513. 

  1514.     av_freep(&c->vLumFilterPos);

  1515.     av_freep(&c->vChrFilterPos);

  1516.     av_freep(&c->hLumFilterPos);

  1517.     av_freep(&c->hChrFilterPos);

  1518. 

  1519. #if ARCH_X86

  1520. #ifdef MAP_ANONYMOUS

  1521.     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1522.     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1523. #elif HAVE_VIRTUALALLOC

  1524.     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1525.     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1526. #else

  1527.     av_free(c->lumMmx2FilterCode);

  1528.     av_free(c->chrMmx2FilterCode);

  1529. #endif

  1530.     c->lumMmx2FilterCode=NULL;

  1531.     c->chrMmx2FilterCode=NULL;

  1532. #endif /* ARCH_X86 */

  1533. 

  1534.     av_freep(&c->yuvTable);

  1535. 

  1536.     av_free(c);

  1537. }

  1538. 

  1539. struct SwsContext *sws_getCachedContext(struct SwsContext *context,

  1540.                                         int srcW, int srcH, enum PixelFormat srcFormat,

  1541.                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1542.                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1543. {

  1544.     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};

  1545. 

  1546.     if (!param)

  1547.         param = default_param;

  1548. 

  1549.     flags = update_flags_cpu(flags);

  1550. 

  1551.     if (context &&

  1552.         (context->srcW      != srcW      ||

  1553.          context->srcH      != srcH      ||

  1554.          context->srcFormat != srcFormat ||

  1555.          context->dstW      != dstW      ||

  1556.          context->dstH      != dstH      ||

  1557.          context->dstFormat != dstFormat ||

  1558.          context->flags     != flags     ||

  1559.          context->param[0]  != param[0]  ||

  1560.          context->param[1]  != param[1])) {

  1561.         sws_freeContext(context);

  1562.         context = NULL;

  1563.     }

  1564. 

  1565.     if (!context) {

  1566.         return sws_getContext(srcW, srcH, srcFormat,

  1567.                               dstW, dstH, dstFormat, flags,

  1568.                               srcFilter, dstFilter, param);

  1569.     }

  1570.     return context;

  1571. }