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.
28500 Commits
32 Branches
293MB
Tree: c8f487deae
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c8f487deae'
${ noResults }
FFmpeg/libswscale/utils.c

1578 lines
57KB
Raw Blame History 

  1. /*

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

  3.  *

  4.  * This file is part of Libav.

  5.  *

  6.  * Libav 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.  * Libav 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 Libav; 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/opt.h"

  47. #include "libavutil/pixdesc.h"

  48. 

  49. unsigned swscale_version(void)

  50. {

  51.     return LIBSWSCALE_VERSION_INT;

  52. }

  53. 

  54. const char *swscale_configuration(void)

  55. {

  56.     return LIBAV_CONFIGURATION;

  57. }

  58. 

  59. const char *swscale_license(void)

  60. {

  61. #define LICENSE_PREFIX "libswscale license: "

  62.     return LICENSE_PREFIX LIBAV_LICENSE + sizeof(LICENSE_PREFIX) - 1;

  63. }

  64. 

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

  66. 

  67. #define isSupportedIn(x)    (       \

  68.            (x)==PIX_FMT_YUV420P     \

  69.         || (x)==PIX_FMT_YUVA420P    \

  70.         || (x)==PIX_FMT_YUYV422     \

  71.         || (x)==PIX_FMT_UYVY422     \

  72.         || (x)==PIX_FMT_RGB48BE     \

  73.         || (x)==PIX_FMT_RGB48LE     \

  74.         || (x)==PIX_FMT_RGB32       \

  75.         || (x)==PIX_FMT_RGB32_1     \

  76.         || (x)==PIX_FMT_BGR48BE     \

  77.         || (x)==PIX_FMT_BGR48LE     \

  78.         || (x)==PIX_FMT_BGR24       \

  79.         || (x)==PIX_FMT_BGR565      \

  80.         || (x)==PIX_FMT_BGR555      \

  81.         || (x)==PIX_FMT_BGR32       \

  82.         || (x)==PIX_FMT_BGR32_1     \

  83.         || (x)==PIX_FMT_RGB24       \

  84.         || (x)==PIX_FMT_RGB565      \

  85.         || (x)==PIX_FMT_RGB555      \

  86.         || (x)==PIX_FMT_GRAY8       \

  87.         || (x)==PIX_FMT_Y400A       \

  88.         || (x)==PIX_FMT_YUV410P     \

  89.         || (x)==PIX_FMT_YUV440P     \

  90.         || (x)==PIX_FMT_NV12        \

  91.         || (x)==PIX_FMT_NV21        \

  92.         || (x)==PIX_FMT_GRAY16BE    \

  93.         || (x)==PIX_FMT_GRAY16LE    \

  94.         || (x)==PIX_FMT_YUV444P     \

  95.         || (x)==PIX_FMT_YUV422P     \

  96.         || (x)==PIX_FMT_YUV411P     \

  97.         || (x)==PIX_FMT_YUVJ420P    \

  98.         || (x)==PIX_FMT_YUVJ422P    \

  99.         || (x)==PIX_FMT_YUVJ440P    \

  100.         || (x)==PIX_FMT_YUVJ444P    \

  101.         || (x)==PIX_FMT_PAL8        \

  102.         || (x)==PIX_FMT_BGR8        \

  103.         || (x)==PIX_FMT_RGB8        \

  104.         || (x)==PIX_FMT_BGR4_BYTE   \

  105.         || (x)==PIX_FMT_RGB4_BYTE   \

  106.         || (x)==PIX_FMT_YUV440P     \

  107.         || (x)==PIX_FMT_MONOWHITE   \

  108.         || (x)==PIX_FMT_MONOBLACK   \

  109.         || (x)==PIX_FMT_YUV420P9LE    \

  110.         || (x)==PIX_FMT_YUV420P10LE   \

  111.         || (x)==PIX_FMT_YUV420P16LE   \

  112.         || (x)==PIX_FMT_YUV422P16LE   \

  113.         || (x)==PIX_FMT_YUV444P16LE   \

  114.         || (x)==PIX_FMT_YUV420P9BE    \

  115.         || (x)==PIX_FMT_YUV420P10BE   \

  116.         || (x)==PIX_FMT_YUV420P16BE   \

  117.         || (x)==PIX_FMT_YUV422P16BE   \

  118.         || (x)==PIX_FMT_YUV444P16BE   \

  119.     )

  120. 

  121. int sws_isSupportedInput(enum PixelFormat pix_fmt)

  122. {

  123.     return isSupportedIn(pix_fmt);

  124. }

  125. 

  126. #define isSupportedOut(x)   (       \

  127.            (x)==PIX_FMT_YUV420P     \

  128.         || (x)==PIX_FMT_YUVA420P    \

  129.         || (x)==PIX_FMT_YUYV422     \

  130.         || (x)==PIX_FMT_UYVY422     \

  131.         || (x)==PIX_FMT_YUV444P     \

  132.         || (x)==PIX_FMT_YUV422P     \

  133.         || (x)==PIX_FMT_YUV411P     \

  134.         || (x)==PIX_FMT_YUVJ420P    \

  135.         || (x)==PIX_FMT_YUVJ422P    \

  136.         || (x)==PIX_FMT_YUVJ440P    \

  137.         || (x)==PIX_FMT_YUVJ444P    \

  138.         || isAnyRGB(x)              \

  139.         || (x)==PIX_FMT_NV12        \

  140.         || (x)==PIX_FMT_NV21        \

  141.         || (x)==PIX_FMT_GRAY16BE    \

  142.         || (x)==PIX_FMT_GRAY16LE    \

  143.         || (x)==PIX_FMT_GRAY8       \

  144.         || (x)==PIX_FMT_YUV410P     \

  145.         || (x)==PIX_FMT_YUV440P     \

  146.         || (x)==PIX_FMT_YUV420P9LE    \

  147.         || (x)==PIX_FMT_YUV420P10LE   \

  148.         || (x)==PIX_FMT_YUV420P16LE   \

  149.         || (x)==PIX_FMT_YUV422P16LE   \

  150.         || (x)==PIX_FMT_YUV444P16LE   \

  151.         || (x)==PIX_FMT_YUV420P9BE    \

  152.         || (x)==PIX_FMT_YUV420P10BE   \

  153.         || (x)==PIX_FMT_YUV420P16BE   \

  154.         || (x)==PIX_FMT_YUV422P16BE   \

  155.         || (x)==PIX_FMT_YUV444P16BE   \

  156.     )

  157. 

  158. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  159. {

  160.     return isSupportedOut(pix_fmt);

  161. }

  162. 

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

  164. 

  165. const char *sws_format_name(enum PixelFormat format)

  166. {

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

  168.         return av_pix_fmt_descriptors[format].name;

  169.     else

  170.         return "Unknown format";

  171. }

  172. 

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

  174. {

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

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

  177.     else           return getSplineCoeff(        0.0,

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

  179.                                                  c + 3.0*d,

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

  181.                                          dist-1.0);

  182. }

  183. 

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

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

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

  187. {

  188.     int i;

  189.     int filterSize;

  190.     int filter2Size;

  191.     int minFilterSize;

  192.     int64_t *filter=NULL;

  193.     int64_t *filter2=NULL;

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

  195.     int ret= -1;

  196. #if ARCH_X86

  197.     if (flags & SWS_CPU_CAPS_MMX)

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

  199. #endif

  200. 

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

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

  203. 

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

  205.         int i;

  206.         filterSize= 1;

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

  208. 

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

  210.             filter[i*filterSize]= fone;

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

  212.         }

  213. 

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

  215.         int i;

  216.         int xDstInSrc;

  217.         filterSize= 1;

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

  219. 

  220.         xDstInSrc= xInc/2 - 0x8000;

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

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

  223. 

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

  225.             filter[i]= fone;

  226.             xDstInSrc+= xInc;

  227.         }

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

  229.         int i;

  230.         int xDstInSrc;

  231.         filterSize= 2;

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

  233. 

  234.         xDstInSrc= xInc/2 - 0x8000;

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

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

  237.             int j;

  238. 

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

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

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

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

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

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

  245.                 xx++;

  246.             }

  247.             xDstInSrc+= xInc;

  248.         }

  249.     } else {

  250.         int xDstInSrc;

  251.         int sizeFactor;

  252. 

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

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

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

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

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

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

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

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

  261.         else {

  262.             sizeFactor= 0; //GCC warning killer

  263.             assert(0);

  264.         }

  265. 

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

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

  268. 

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

  270. 

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

  272. 

  273.         xDstInSrc= xInc - 0x10000;

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

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

  276.             int j;

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

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

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

  280.                 double floatd;

  281.                 int64_t coeff;

  282. 

  283.                 if (xInc > 1<<16)

  284.                     d= d*dstW/srcW;

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

  286. 

  287.                 if (flags & SWS_BICUBIC) {

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

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

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

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

  292. 

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

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

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

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

  297.                     else

  298.                         coeff=0.0;

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

  300.                 }

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

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

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

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

  305.                 }*/

  306.                 else if (flags & SWS_X) {

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

  308.                     double c;

  309. 

  310.                     if (floatd<1.0)

  311.                         c = cos(floatd*M_PI);

  312.                     else

  313.                         c=-1.0;

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

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

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

  317.                 } else if (flags & SWS_AREA) {

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

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

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

  321.                     else coeff=0.0;

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

  323.                 } else if (flags & SWS_GAUSS) {

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

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

  326.                 } else if (flags & SWS_SINC) {

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

  328.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  332.                 } else if (flags & SWS_BILINEAR) {

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

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

  335.                     coeff *= fone >> 30;

  336.                 } else if (flags & SWS_SPLINE) {

  337.                     double p=-2.196152422706632;

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

  339.                 } else {

  340.                     coeff= 0.0; //GCC warning killer

  341.                     assert(0);

  342.                 }

  343. 

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

  345.                 xx++;

  346.             }

  347.             xDstInSrc+= 2*xInc;

  348.         }

  349.     }

  350. 

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

  352.        av_free(filter);

  353.     */

  354.     assert(filterSize>0);

  355.     filter2Size= filterSize;

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

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

  358.     assert(filter2Size>0);

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

  360. 

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

  362.         int j, k;

  363. 

  364.         if(srcFilter) {

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

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

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

  368.             }

  369.         } else {

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

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

  372.         }

  373.         //FIXME dstFilter

  374. 

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

  376.     }

  377.     av_freep(&filter);

  378. 

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

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

  381.     minFilterSize= 0;

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

  383.         int min= filter2Size;

  384.         int j;

  385.         int64_t cutOff=0.0;

  386. 

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

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

  389.             int k;

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

  391. 

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

  393. 

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

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

  396. 

  397.             // move filter coefficients left

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

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

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

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

  402.         }

  403. 

  404.         cutOff=0;

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

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

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

  408. 

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

  410.             min--;

  411.         }

  412. 

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

  414.     }

  415. 

  416.     if (flags & SWS_CPU_CAPS_ALTIVEC) {

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

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

  419.         if (minFilterSize < 5)

  420.             filterAlign = 4;

  421. 

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

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

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

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

  426.         // decent-sized vector.

  427.         if (minFilterSize < 3)

  428.             filterAlign = 1;

  429.     }

  430. 

  431.     if (flags & SWS_CPU_CAPS_MMX) {

  432.         // special case for unscaled vertical filtering

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

  434.             filterAlign= 1;

  435.     }

  436. 

  437.     assert(minFilterSize > 0);

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

  439.     assert(filterSize > 0);

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

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

  442.         goto fail;

  443.     *outFilterSize= filterSize;

  444. 

  445.     if (flags&SWS_PRINT_INFO)

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

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

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

  449.         int j;

  450. 

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

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

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

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

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

  456.         }

  457.     }

  458. 

  459.     //FIXME try to align filterPos if possible

  460. 

  461.     //fix borders

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

  463.         int j;

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

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

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

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

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

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

  470.             }

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

  472.         }

  473. 

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

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

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

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

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

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

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

  481.             }

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

  483.         }

  484.     }

  485. 

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

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

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

  489. 

  490.     /* normalize & store in outFilter */

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

  492.         int j;

  493.         int64_t error=0;

  494.         int64_t sum=0;

  495. 

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

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

  498.         }

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

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

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

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

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

  504.             error= v - intV*sum;

  505.         }

  506.     }

  507. 

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

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

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

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

  512.     }

  513. 

  514.     ret=0;

  515. fail:

  516.     av_free(filter);

  517.     av_free(filter2);

  518.     return ret;

  519. }

  520. 

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

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

  523. {

  524.     uint8_t *fragmentA;

  525.     x86_reg imm8OfPShufW1A;

  526.     x86_reg imm8OfPShufW2A;

  527.     x86_reg fragmentLengthA;

  528.     uint8_t *fragmentB;

  529.     x86_reg imm8OfPShufW1B;

  530.     x86_reg imm8OfPShufW2B;

  531.     x86_reg fragmentLengthB;

  532.     int fragmentPos;

  533. 

  534.     int xpos, i;

  535. 

  536.     // create an optimized horizontal scaling routine

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

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

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

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

  541.      * of fragmentA is used.

  542.      */

  543. 

  544.     //code fragment

  545. 

  546.     __asm__ volatile(

  547.         "jmp                         9f                 \n\t"

  548.     // Begin

  549.         "0:                                             \n\t"

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

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

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

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

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

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

  556.         "1:                                             \n\t"

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

  558.         "2:                                             \n\t"

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

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

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

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

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

  564. 

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

  566. 

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

  568.     // End

  569.         "9:                                             \n\t"

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

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

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

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

  574.         "dec                         %1                 \n\t"

  575.         "dec                         %2                 \n\t"

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

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

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

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

  580. 

  581. 

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

  583.         "=r" (fragmentLengthA)

  584.     );

  585. 

  586.     __asm__ volatile(

  587.         "jmp                         9f                 \n\t"

  588.     // Begin

  589.         "0:                                             \n\t"

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

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

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

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

  594.         "1:                                             \n\t"

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

  596.         "2:                                             \n\t"

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

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

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

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

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

  602. 

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

  604. 

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

  606.     // End

  607.         "9:                                             \n\t"

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

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

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

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

  612.         "dec                         %1                 \n\t"

  613.         "dec                         %2                 \n\t"

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

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

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

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

  618. 

  619. 

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

  621.         "=r" (fragmentLengthB)

  622.     );

  623. 

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

  625.     fragmentPos=0;

  626. 

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

  628.         int xx=xpos>>16;

  629. 

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

  631.             int a=0;

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

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

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

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

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

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

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

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

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

  641.             int shift=0;

  642. 

  643.             if (filterCode) {

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

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

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

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

  648.                 filterPos[i/2]= xx;

  649. 

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

  651. 

  652.                 filterCode[fragmentPos + imm8OfPShufW1]=

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

  654.                 filterCode[fragmentPos + imm8OfPShufW2]=

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

  656. 

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

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

  659. 

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

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

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

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

  664.                 }

  665.             }

  666. 

  667.             fragmentPos+= fragmentLength;

  668. 

  669.             if (filterCode)

  670.                 filterCode[fragmentPos]= RET;

  671.         }

  672.         xpos+=xInc;

  673.     }

  674.     if (filterCode)

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

  676. 

  677.     return fragmentPos + 1;

  678. }

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

  680. 

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

  682. {

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

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

  685. }

  686. 

  687. static int update_flags_cpu(int flags);

  688. 

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

  690. {

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

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

  693. 

  694.     c->brightness= brightness;

  695.     c->contrast  = contrast;

  696.     c->saturation= saturation;

  697.     c->srcRange  = srcRange;

  698.     c->dstRange  = dstRange;

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

  700. 

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

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

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

  704. 

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

  706.     //FIXME factorize

  707. 

  708. #if HAVE_ALTIVEC

  709.     if (c->flags & SWS_CPU_CAPS_ALTIVEC)

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

  711. #endif

  712.     return 0;

  713. }

  714. 

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

  716. {

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

  718. 

  719.     *inv_table = c->srcColorspaceTable;

  720.     *table     = c->dstColorspaceTable;

  721.     *srcRange  = c->srcRange;

  722.     *dstRange  = c->dstRange;

  723.     *brightness= c->brightness;

  724.     *contrast  = c->contrast;

  725.     *saturation= c->saturation;

  726. 

  727.     return 0;

  728. }

  729. 

  730. static int handle_jpeg(enum PixelFormat *format)

  731. {

  732.     switch (*format) {

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

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

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

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

  737.     default:                                          return 0;

  738.     }

  739. }

  740. 

  741. static int update_flags_cpu(int flags)

  742. {

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

  744.     flags &= ~( SWS_CPU_CAPS_MMX

  745.                |SWS_CPU_CAPS_MMX2

  746.                |SWS_CPU_CAPS_3DNOW

  747.                |SWS_CPU_CAPS_SSE2

  748.                |SWS_CPU_CAPS_ALTIVEC

  749.                |SWS_CPU_CAPS_BFIN);

  750.     flags |= ff_hardcodedcpuflags();

  751. #endif /* CONFIG_RUNTIME_CPUDETECT */

  752.     return flags;

  753. }

  754. 

  755. SwsContext *sws_alloc_context(void)

  756. {

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

  758. 

  759.     c->av_class = &sws_context_class;

  760.     av_opt_set_defaults(c);

  761. 

  762.     return c;

  763. }

  764. 

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

  766. {

  767.     int i;

  768.     int usesVFilter, usesHFilter;

  769.     int unscaled;

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

  771.     int srcW= c->srcW;

  772.     int srcH= c->srcH;

  773.     int dstW= c->dstW;

  774.     int dstH= c->dstH;

  775.     int flags;

  776.     enum PixelFormat srcFormat= c->srcFormat;

  777.     enum PixelFormat dstFormat= c->dstFormat;

  778. 

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

  780. #if ARCH_X86

  781.     if (flags & SWS_CPU_CAPS_MMX)

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

  783. #endif

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

  785. 

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

  787. 

  788.     if (!isSupportedIn(srcFormat)) {

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

  790.         return AVERROR(EINVAL);

  791.     }

  792.     if (!isSupportedOut(dstFormat)) {

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

  794.         return AVERROR(EINVAL);

  795.     }

  796. 

  797.     i= flags & ( SWS_POINT

  798.                 |SWS_AREA

  799.                 |SWS_BILINEAR

  800.                 |SWS_FAST_BILINEAR

  801.                 |SWS_BICUBIC

  802.                 |SWS_X

  803.                 |SWS_GAUSS

  804.                 |SWS_LANCZOS

  805.                 |SWS_SINC

  806.                 |SWS_SPLINE

  807.                 |SWS_BICUBLIN);

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

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

  810.         return AVERROR(EINVAL);

  811.     }

  812.     /* sanity check */

  813.     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

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

  815.                srcW, srcH, dstW, dstH);

  816.         return AVERROR(EINVAL);

  817.     }

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

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

  820.         return AVERROR(EINVAL);

  821.     }

  822. 

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

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

  825. 

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

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

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

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

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

  831. 

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

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

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

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

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

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

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

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

  840. 

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

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

  843. 

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

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

  846. 

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

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

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

  850. 

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

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

  853.       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8

  854.       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4

  855.       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE

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

  857.         c->chrSrcHSubSample=1;

  858. 

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

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

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

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

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

  864. 

  865.     /* unscaled special cases */

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

  867.         ff_get_unscaled_swscale(c);

  868. 

  869.         if (c->swScale) {

  870.             if (flags&SWS_PRINT_INFO)

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

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

  873.             return 0;

  874.         }

  875.     }

  876. 

  877.     if (flags & SWS_CPU_CAPS_MMX2) {

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

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

  880.             if (flags&SWS_PRINT_INFO)

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

  882.         }

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

  884.     }

  885.     else

  886.         c->canMMX2BeUsed=0;

  887. 

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

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

  890. 

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

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

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

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

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

  896.     // first and last pixel

  897.     if (flags&SWS_FAST_BILINEAR) {

  898.         if (c->canMMX2BeUsed) {

  899.             c->lumXInc+= 20;

  900.             c->chrXInc+= 20;

  901.         }

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

  903.         else if (flags & SWS_CPU_CAPS_MMX) {

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

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

  906.         }

  907.     }

  908. 

  909.     /* precalculate horizontal scaler filter coefficients */

  910.     {

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

  912. // can't downscale !!!

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

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

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

  916. 

  917. #ifdef MAP_ANONYMOUS

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

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

  920. #elif HAVE_VIRTUALALLOC

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

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

  923. #else

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

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

  926. #endif

  927. 

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

  929.                 return AVERROR(ENOMEM);

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

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

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

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

  934. 

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

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

  937. 

  938. #ifdef MAP_ANONYMOUS

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

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

  941. #endif

  942.         } else

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

  944.         {

  945.             const int filterAlign=

  946.                 (flags & SWS_CPU_CAPS_MMX) ? 4 :

  947.                 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :

  948.                 1;

  949. 

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

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

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

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

  954.                 goto fail;

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

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

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

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

  959.                 goto fail;

  960.         }

  961.     } // initialize horizontal stuff

  962. 

  963.     /* precalculate vertical scaler filter coefficients */

  964.     {

  965.         const int filterAlign=

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

  967.             (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :

  968.             1;

  969. 

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

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

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

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

  974.             goto fail;

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

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

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

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

  979.             goto fail;

  980. 

  981. #if HAVE_ALTIVEC

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

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

  984. 

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

  986.             int j;

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

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

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

  990.         }

  991. 

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

  993.             int j;

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

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

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

  997.         }

  998. #endif

  999.     }

  1000. 

  1001.     // calculate buffer sizes so that they won't run out while handling these damn slices

  1002.     c->vLumBufSize= c->vLumFilterSize;

  1003.     c->vChrBufSize= c->vChrFilterSize;

  1004.     for (i=0; i<dstH; i++) {

  1005.         int chrI= i*c->chrDstH / dstH;

  1006.         int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,

  1007.                            ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));

  1008. 

  1009.         nextSlice>>= c->chrSrcVSubSample;

  1010.         nextSlice<<= c->chrSrcVSubSample;

  1011.         if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)

  1012.             c->vLumBufSize= nextSlice - c->vLumFilterPos[i];

  1013.         if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))

  1014.             c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];

  1015.     }

  1016. 

  1017.     // allocate pixbufs (we use dynamic allocation because otherwise we would need to

  1018.     // allocate several megabytes to handle all possible cases)

  1019.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1020.     FF_ALLOC_OR_GOTO(c, c->chrPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1021.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1022.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1023.     //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)

  1024.     /* align at 16 bytes for AltiVec */

  1025.     for (i=0; i<c->vLumBufSize; i++) {

  1026.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], VOF+1, fail);

  1027.         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];

  1028.     }

  1029.     for (i=0; i<c->vChrBufSize; i++) {

  1030.         FF_ALLOC_OR_GOTO(c, c->chrPixBuf[i+c->vChrBufSize], (VOF+1)*2, fail);

  1031.         c->chrPixBuf[i] = c->chrPixBuf[i+c->vChrBufSize];

  1032.     }

  1033.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1034.         for (i=0; i<c->vLumBufSize; i++) {

  1035.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], VOF+1, fail);

  1036.             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];

  1037.         }

  1038. 

  1039.     //try to avoid drawing green stuff between the right end and the stride end

  1040.     for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);

  1041. 

  1042.     assert(2*VOFW == VOF);

  1043. 

  1044.     assert(c->chrDstH <= dstH);

  1045. 

  1046.     if (flags&SWS_PRINT_INFO) {

  1047.         if      (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1048.         else if (flags&SWS_BILINEAR)      av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1049.         else if (flags&SWS_BICUBIC)       av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1050.         else if (flags&SWS_X)             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1051.         else if (flags&SWS_POINT)         av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1052.         else if (flags&SWS_AREA)          av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1053.         else if (flags&SWS_BICUBLIN)      av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1054.         else if (flags&SWS_GAUSS)         av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1055.         else if (flags&SWS_SINC)          av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1056.         else if (flags&SWS_LANCZOS)       av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1057.         else if (flags&SWS_SPLINE)        av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1058.         else                              av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1059. 

  1060.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1061.                sws_format_name(srcFormat),

  1062. #ifdef DITHER1XBPP

  1063.                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||

  1064.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1065.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",

  1066. #else

  1067.                "",

  1068. #endif

  1069.                sws_format_name(dstFormat));

  1070. 

  1071.         if      (flags & SWS_CPU_CAPS_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");

  1072.         else if (flags & SWS_CPU_CAPS_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1073.         else if (flags & SWS_CPU_CAPS_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");

  1074.         else if (flags & SWS_CPU_CAPS_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");

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

  1076. 

  1077.         if (flags & SWS_CPU_CAPS_MMX) {

  1078.             if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))

  1079.                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");

  1080.             else {

  1081.                 if (c->hLumFilterSize==4)

  1082.                     av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");

  1083.                 else if (c->hLumFilterSize==8)

  1084.                     av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");

  1085.                 else

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

  1087. 

  1088.                 if (c->hChrFilterSize==4)

  1089.                     av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");

  1090.                 else if (c->hChrFilterSize==8)

  1091.                     av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");

  1092.                 else

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

  1094.             }

  1095.         } else {

  1096. #if ARCH_X86

  1097.             av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");

  1098. #else

  1099.             if (flags & SWS_FAST_BILINEAR)

  1100.                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");

  1101.             else

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

  1103. #endif

  1104.         }

  1105.         if (isPlanarYUV(dstFormat)) {

  1106.             if (c->vLumFilterSize==1)

  1107.                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1108.             else

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

  1110.         } else {

  1111.             if (c->vLumFilterSize==1 && c->vChrFilterSize==2)

  1112.                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"

  1113.                        "      2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1114.             else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)

  1115.                 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1116.             else

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

  1118.         }

  1119. 

  1120.         if (dstFormat==PIX_FMT_BGR24)

  1121.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",

  1122.                    (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));

  1123.         else if (dstFormat==PIX_FMT_RGB32)

  1124.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1125.         else if (dstFormat==PIX_FMT_BGR565)

  1126.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1127.         else if (dstFormat==PIX_FMT_BGR555)

  1128.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1129.         else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1130.                  dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)

  1131.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1132. 

  1133.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1134.         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1135.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

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

  1137.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1138.     }

  1139. 

  1140.     c->swScale= ff_getSwsFunc(c);

  1141.     return 0;

  1142. fail: //FIXME replace things by appropriate error codes

  1143.     return -1;

  1144. }

  1145. 

  1146. #if FF_API_SWS_GETCONTEXT

  1147. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1148.                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1149.                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1150. {

  1151.     SwsContext *c;

  1152. 

  1153.     if(!(c=sws_alloc_context()))

  1154.         return NULL;

  1155. 

  1156.     c->flags= flags;

  1157.     c->srcW= srcW;

  1158.     c->srcH= srcH;

  1159.     c->dstW= dstW;

  1160.     c->dstH= dstH;

  1161.     c->srcRange = handle_jpeg(&srcFormat);

  1162.     c->dstRange = handle_jpeg(&dstFormat);

  1163.     c->srcFormat= srcFormat;

  1164.     c->dstFormat= dstFormat;

  1165. 

  1166.     if (param) {

  1167.         c->param[0] = param[0];

  1168.         c->param[1] = param[1];

  1169.     }

  1170.     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);

  1171. 

  1172.     if(sws_init_context(c, srcFilter, dstFilter) < 0){

  1173.         sws_freeContext(c);

  1174.         return NULL;

  1175.     }

  1176. 

  1177.     return c;

  1178. }

  1179. #endif

  1180. 

  1181. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1182.                                 float lumaSharpen, float chromaSharpen,

  1183.                                 float chromaHShift, float chromaVShift,

  1184.                                 int verbose)

  1185. {

  1186.     SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  1187.     if (!filter)

  1188.         return NULL;

  1189. 

  1190.     if (lumaGBlur!=0.0) {

  1191.         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  1192.         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  1193.     } else {

  1194.         filter->lumH= sws_getIdentityVec();

  1195.         filter->lumV= sws_getIdentityVec();

  1196.     }

  1197. 

  1198.     if (chromaGBlur!=0.0) {

  1199.         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  1200.         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  1201.     } else {

  1202.         filter->chrH= sws_getIdentityVec();

  1203.         filter->chrV= sws_getIdentityVec();

  1204.     }

  1205. 

  1206.     if (chromaSharpen!=0.0) {

  1207.         SwsVector *id= sws_getIdentityVec();

  1208.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1209.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1210.         sws_addVec(filter->chrH, id);

  1211.         sws_addVec(filter->chrV, id);

  1212.         sws_freeVec(id);

  1213.     }

  1214. 

  1215.     if (lumaSharpen!=0.0) {

  1216.         SwsVector *id= sws_getIdentityVec();

  1217.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1218.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1219.         sws_addVec(filter->lumH, id);

  1220.         sws_addVec(filter->lumV, id);

  1221.         sws_freeVec(id);

  1222.     }

  1223. 

  1224.     if (chromaHShift != 0.0)

  1225.         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  1226. 

  1227.     if (chromaVShift != 0.0)

  1228.         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  1229. 

  1230.     sws_normalizeVec(filter->chrH, 1.0);

  1231.     sws_normalizeVec(filter->chrV, 1.0);

  1232.     sws_normalizeVec(filter->lumH, 1.0);

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

  1234. 

  1235.     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1236.     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1237. 

  1238.     return filter;

  1239. }

  1240. 

  1241. SwsVector *sws_allocVec(int length)

  1242. {

  1243.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1244.     if (!vec)

  1245.         return NULL;

  1246.     vec->length = length;

  1247.     vec->coeff  = av_malloc(sizeof(double) * length);

  1248.     if (!vec->coeff)

  1249.         av_freep(&vec);

  1250.     return vec;

  1251. }

  1252. 

  1253. SwsVector *sws_getGaussianVec(double variance, double quality)

  1254. {

  1255.     const int length= (int)(variance*quality + 0.5) | 1;

  1256.     int i;

  1257.     double middle= (length-1)*0.5;

  1258.     SwsVector *vec= sws_allocVec(length);

  1259. 

  1260.     if (!vec)

  1261.         return NULL;

  1262. 

  1263.     for (i=0; i<length; i++) {

  1264.         double dist= i-middle;

  1265.         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);

  1266.     }

  1267. 

  1268.     sws_normalizeVec(vec, 1.0);

  1269. 

  1270.     return vec;

  1271. }

  1272. 

  1273. SwsVector *sws_getConstVec(double c, int length)

  1274. {

  1275.     int i;

  1276.     SwsVector *vec= sws_allocVec(length);

  1277. 

  1278.     if (!vec)

  1279.         return NULL;

  1280. 

  1281.     for (i=0; i<length; i++)

  1282.         vec->coeff[i]= c;

  1283. 

  1284.     return vec;

  1285. }

  1286. 

  1287. SwsVector *sws_getIdentityVec(void)

  1288. {

  1289.     return sws_getConstVec(1.0, 1);

  1290. }

  1291. 

  1292. static double sws_dcVec(SwsVector *a)

  1293. {

  1294.     int i;

  1295.     double sum=0;

  1296. 

  1297.     for (i=0; i<a->length; i++)

  1298.         sum+= a->coeff[i];

  1299. 

  1300.     return sum;

  1301. }

  1302. 

  1303. void sws_scaleVec(SwsVector *a, double scalar)

  1304. {

  1305.     int i;

  1306. 

  1307.     for (i=0; i<a->length; i++)

  1308.         a->coeff[i]*= scalar;

  1309. }

  1310. 

  1311. void sws_normalizeVec(SwsVector *a, double height)

  1312. {

  1313.     sws_scaleVec(a, height/sws_dcVec(a));

  1314. }

  1315. 

  1316. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1317. {

  1318.     int length= a->length + b->length - 1;

  1319.     int i, j;

  1320.     SwsVector *vec= sws_getConstVec(0.0, length);

  1321. 

  1322.     if (!vec)

  1323.         return NULL;

  1324. 

  1325.     for (i=0; i<a->length; i++) {

  1326.         for (j=0; j<b->length; j++) {

  1327.             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1328.         }

  1329.     }

  1330. 

  1331.     return vec;

  1332. }

  1333. 

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

  1335. {

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

  1337.     int i;

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

  1339. 

  1340.     if (!vec)

  1341.         return NULL;

  1342. 

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

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

  1345. 

  1346.     return vec;

  1347. }

  1348. 

  1349. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1350. {

  1351.     int length= FFMAX(a->length, b->length);

  1352.     int i;

  1353.     SwsVector *vec= sws_getConstVec(0.0, length);

  1354. 

  1355.     if (!vec)

  1356.         return NULL;

  1357. 

  1358.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1359.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1360. 

  1361.     return vec;

  1362. }

  1363. 

  1364. /* shift left / or right if "shift" is negative */

  1365. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1366. {

  1367.     int length= a->length + FFABS(shift)*2;

  1368.     int i;

  1369.     SwsVector *vec= sws_getConstVec(0.0, length);

  1370. 

  1371.     if (!vec)

  1372.         return NULL;

  1373. 

  1374.     for (i=0; i<a->length; i++) {

  1375.         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1376.     }

  1377. 

  1378.     return vec;

  1379. }

  1380. 

  1381. void sws_shiftVec(SwsVector *a, int shift)

  1382. {

  1383.     SwsVector *shifted= sws_getShiftedVec(a, shift);

  1384.     av_free(a->coeff);

  1385.     a->coeff= shifted->coeff;

  1386.     a->length= shifted->length;

  1387.     av_free(shifted);

  1388. }

  1389. 

  1390. void sws_addVec(SwsVector *a, SwsVector *b)

  1391. {

  1392.     SwsVector *sum= sws_sumVec(a, b);

  1393.     av_free(a->coeff);

  1394.     a->coeff= sum->coeff;

  1395.     a->length= sum->length;

  1396.     av_free(sum);

  1397. }

  1398. 

  1399. void sws_subVec(SwsVector *a, SwsVector *b)

  1400. {

  1401.     SwsVector *diff= sws_diffVec(a, b);

  1402.     av_free(a->coeff);

  1403.     a->coeff= diff->coeff;

  1404.     a->length= diff->length;

  1405.     av_free(diff);

  1406. }

  1407. 

  1408. void sws_convVec(SwsVector *a, SwsVector *b)

  1409. {

  1410.     SwsVector *conv= sws_getConvVec(a, b);

  1411.     av_free(a->coeff);

  1412.     a->coeff= conv->coeff;

  1413.     a->length= conv->length;

  1414.     av_free(conv);

  1415. }

  1416. 

  1417. SwsVector *sws_cloneVec(SwsVector *a)

  1418. {

  1419.     int i;

  1420.     SwsVector *vec= sws_allocVec(a->length);

  1421. 

  1422.     if (!vec)

  1423.         return NULL;

  1424. 

  1425.     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];

  1426. 

  1427.     return vec;

  1428. }

  1429. 

  1430. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1431. {

  1432.     int i;

  1433.     double max=0;

  1434.     double min=0;

  1435.     double range;

  1436. 

  1437.     for (i=0; i<a->length; i++)

  1438.         if (a->coeff[i]>max) max= a->coeff[i];

  1439. 

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

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

  1442. 

  1443.     range= max - min;

  1444. 

  1445.     for (i=0; i<a->length; i++) {

  1446.         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1447.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1448.         for (;x>0; x--) av_log(log_ctx, log_level, " ");

  1449.         av_log(log_ctx, log_level, "|\n");

  1450.     }

  1451. }

  1452. 

  1453. void sws_freeVec(SwsVector *a)

  1454. {

  1455.     if (!a) return;

  1456.     av_freep(&a->coeff);

  1457.     a->length=0;

  1458.     av_free(a);

  1459. }

  1460. 

  1461. void sws_freeFilter(SwsFilter *filter)

  1462. {

  1463.     if (!filter) return;

  1464. 

  1465.     if (filter->lumH) sws_freeVec(filter->lumH);

  1466.     if (filter->lumV) sws_freeVec(filter->lumV);

  1467.     if (filter->chrH) sws_freeVec(filter->chrH);

  1468.     if (filter->chrV) sws_freeVec(filter->chrV);

  1469.     av_free(filter);

  1470. }

  1471. 

  1472. void sws_freeContext(SwsContext *c)

  1473. {

  1474.     int i;

  1475.     if (!c) return;

  1476. 

  1477.     if (c->lumPixBuf) {

  1478.         for (i=0; i<c->vLumBufSize; i++)

  1479.             av_freep(&c->lumPixBuf[i]);

  1480.         av_freep(&c->lumPixBuf);

  1481.     }

  1482. 

  1483.     if (c->chrPixBuf) {

  1484.         for (i=0; i<c->vChrBufSize; i++)

  1485.             av_freep(&c->chrPixBuf[i]);

  1486.         av_freep(&c->chrPixBuf);

  1487.     }

  1488. 

  1489.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1490.         for (i=0; i<c->vLumBufSize; i++)

  1491.             av_freep(&c->alpPixBuf[i]);

  1492.         av_freep(&c->alpPixBuf);

  1493.     }

  1494. 

  1495.     av_freep(&c->vLumFilter);

  1496.     av_freep(&c->vChrFilter);

  1497.     av_freep(&c->hLumFilter);

  1498.     av_freep(&c->hChrFilter);

  1499. #if HAVE_ALTIVEC

  1500.     av_freep(&c->vYCoeffsBank);

  1501.     av_freep(&c->vCCoeffsBank);

  1502. #endif

  1503. 

  1504.     av_freep(&c->vLumFilterPos);

  1505.     av_freep(&c->vChrFilterPos);

  1506.     av_freep(&c->hLumFilterPos);

  1507.     av_freep(&c->hChrFilterPos);

  1508. 

  1509. #if ARCH_X86

  1510. #ifdef MAP_ANONYMOUS

  1511.     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1512.     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1513. #elif HAVE_VIRTUALALLOC

  1514.     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1515.     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1516. #else

  1517.     av_free(c->lumMmx2FilterCode);

  1518.     av_free(c->chrMmx2FilterCode);

  1519. #endif

  1520.     c->lumMmx2FilterCode=NULL;

  1521.     c->chrMmx2FilterCode=NULL;

  1522. #endif /* ARCH_X86 */

  1523. 

  1524.     av_freep(&c->yuvTable);

  1525. 

  1526.     av_free(c);

  1527. }

  1528. 

  1529. struct SwsContext *sws_getCachedContext(struct SwsContext *context,

  1530.                                         int srcW, int srcH, enum PixelFormat srcFormat,

  1531.                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1532.                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1533. {

  1534.     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};

  1535. 

  1536.     if (!param)

  1537.         param = default_param;

  1538. 

  1539.     flags = update_flags_cpu(flags);

  1540. 

  1541.     if (context &&

  1542.         (context->srcW      != srcW      ||

  1543.          context->srcH      != srcH      ||

  1544.          context->srcFormat != srcFormat ||

  1545.          context->dstW      != dstW      ||

  1546.          context->dstH      != dstH      ||

  1547.          context->dstFormat != dstFormat ||

  1548.          context->flags     != flags     ||

  1549.          context->param[0]  != param[0]  ||

  1550.          context->param[1]  != param[1])) {

  1551.         sws_freeContext(context);

  1552.         context = NULL;

  1553.     }

  1554. 

  1555.     if (!context) {

  1556.         if (!(context = sws_alloc_context()))

  1557.             return NULL;

  1558.         context->srcW      = srcW;

  1559.         context->srcH      = srcH;

  1560.         context->srcRange  = handle_jpeg(&srcFormat);

  1561.         context->srcFormat = srcFormat;

  1562.         context->dstW      = dstW;

  1563.         context->dstH      = dstH;

  1564.         context->dstRange  = handle_jpeg(&dstFormat);

  1565.         context->dstFormat = dstFormat;

  1566.         context->flags     = flags;

  1567.         context->param[0]  = param[0];

  1568.         context->param[1]  = param[1];

  1569.         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], context->srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, context->dstRange, 0, 1<<16, 1<<16);

  1570.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1571.             sws_freeContext(context);

  1572.             return NULL;

  1573.         }

  1574.     }

  1575.     return context;

  1576. }