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.
28498 Commits
32 Branches
552MB
Tree: 9aa91043f3
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '9aa91043f3'
${ noResults }
FFmpeg/libswscale/utils.c

1574 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_YUV420P16LE   \

  147.         || (x)==PIX_FMT_YUV422P16LE   \

  148.         || (x)==PIX_FMT_YUV444P16LE   \

  149.         || (x)==PIX_FMT_YUV420P16BE   \

  150.         || (x)==PIX_FMT_YUV422P16BE   \

  151.         || (x)==PIX_FMT_YUV444P16BE   \

  152.     )

  153. 

  154. int sws_isSupportedOutput(enum PixelFormat pix_fmt)

  155. {

  156.     return isSupportedOut(pix_fmt);

  157. }

  158. 

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

  160. 

  161. const char *sws_format_name(enum PixelFormat format)

  162. {

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

  164.         return av_pix_fmt_descriptors[format].name;

  165.     else

  166.         return "Unknown format";

  167. }

  168. 

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

  170. {

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

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

  173.     else           return getSplineCoeff(        0.0,

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

  175.                                                  c + 3.0*d,

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

  177.                                          dist-1.0);

  178. }

  179. 

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

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

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

  183. {

  184.     int i;

  185.     int filterSize;

  186.     int filter2Size;

  187.     int minFilterSize;

  188.     int64_t *filter=NULL;

  189.     int64_t *filter2=NULL;

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

  191.     int ret= -1;

  192. #if ARCH_X86

  193.     if (flags & SWS_CPU_CAPS_MMX)

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

  195. #endif

  196. 

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

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

  199. 

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

  201.         int i;

  202.         filterSize= 1;

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

  204. 

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

  206.             filter[i*filterSize]= fone;

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

  208.         }

  209. 

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

  211.         int i;

  212.         int xDstInSrc;

  213.         filterSize= 1;

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

  215. 

  216.         xDstInSrc= xInc/2 - 0x8000;

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

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

  219. 

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

  221.             filter[i]= fone;

  222.             xDstInSrc+= xInc;

  223.         }

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

  225.         int i;

  226.         int xDstInSrc;

  227.         filterSize= 2;

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

  229. 

  230.         xDstInSrc= xInc/2 - 0x8000;

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

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

  233.             int j;

  234. 

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

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

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

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

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

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

  241.                 xx++;

  242.             }

  243.             xDstInSrc+= xInc;

  244.         }

  245.     } else {

  246.         int xDstInSrc;

  247.         int sizeFactor;

  248. 

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

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

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

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

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

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

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

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

  257.         else {

  258.             sizeFactor= 0; //GCC warning killer

  259.             assert(0);

  260.         }

  261. 

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

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

  264. 

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

  266. 

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

  268. 

  269.         xDstInSrc= xInc - 0x10000;

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

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

  272.             int j;

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

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

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

  276.                 double floatd;

  277.                 int64_t coeff;

  278. 

  279.                 if (xInc > 1<<16)

  280.                     d= d*dstW/srcW;

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

  282. 

  283.                 if (flags & SWS_BICUBIC) {

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

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

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

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

  288. 

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

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

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

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

  293.                     else

  294.                         coeff=0.0;

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

  296.                 }

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

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

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

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

  301.                 }*/

  302.                 else if (flags & SWS_X) {

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

  304.                     double c;

  305. 

  306.                     if (floatd<1.0)

  307.                         c = cos(floatd*M_PI);

  308.                     else

  309.                         c=-1.0;

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

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

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

  313.                 } else if (flags & SWS_AREA) {

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

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

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

  317.                     else coeff=0.0;

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

  319.                 } else if (flags & SWS_GAUSS) {

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

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

  322.                 } else if (flags & SWS_SINC) {

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

  324.                 } else if (flags & SWS_LANCZOS) {

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

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

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

  328.                 } else if (flags & SWS_BILINEAR) {

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

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

  331.                     coeff *= fone >> 30;

  332.                 } else if (flags & SWS_SPLINE) {

  333.                     double p=-2.196152422706632;

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

  335.                 } else {

  336.                     coeff= 0.0; //GCC warning killer

  337.                     assert(0);

  338.                 }

  339. 

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

  341.                 xx++;

  342.             }

  343.             xDstInSrc+= 2*xInc;

  344.         }

  345.     }

  346. 

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

  348.        av_free(filter);

  349.     */

  350.     assert(filterSize>0);

  351.     filter2Size= filterSize;

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

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

  354.     assert(filter2Size>0);

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

  356. 

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

  358.         int j, k;

  359. 

  360.         if(srcFilter) {

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

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

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

  364.             }

  365.         } else {

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

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

  368.         }

  369.         //FIXME dstFilter

  370. 

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

  372.     }

  373.     av_freep(&filter);

  374. 

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

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

  377.     minFilterSize= 0;

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

  379.         int min= filter2Size;

  380.         int j;

  381.         int64_t cutOff=0.0;

  382. 

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

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

  385.             int k;

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

  387. 

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

  389. 

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

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

  392. 

  393.             // move filter coefficients left

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

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

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

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

  398.         }

  399. 

  400.         cutOff=0;

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

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

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

  404. 

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

  406.             min--;

  407.         }

  408. 

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

  410.     }

  411. 

  412.     if (flags & SWS_CPU_CAPS_ALTIVEC) {

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

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

  415.         if (minFilterSize < 5)

  416.             filterAlign = 4;

  417. 

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

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

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

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

  422.         // decent-sized vector.

  423.         if (minFilterSize < 3)

  424.             filterAlign = 1;

  425.     }

  426. 

  427.     if (flags & SWS_CPU_CAPS_MMX) {

  428.         // special case for unscaled vertical filtering

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

  430.             filterAlign= 1;

  431.     }

  432. 

  433.     assert(minFilterSize > 0);

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

  435.     assert(filterSize > 0);

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

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

  438.         goto fail;

  439.     *outFilterSize= filterSize;

  440. 

  441.     if (flags&SWS_PRINT_INFO)

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

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

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

  445.         int j;

  446. 

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

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

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

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

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

  452.         }

  453.     }

  454. 

  455.     //FIXME try to align filterPos if possible

  456. 

  457.     //fix borders

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

  459.         int j;

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

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

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

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

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

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

  466.             }

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

  468.         }

  469. 

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

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

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

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

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

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

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

  477.             }

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

  479.         }

  480.     }

  481. 

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

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

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

  485. 

  486.     /* normalize & store in outFilter */

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

  488.         int j;

  489.         int64_t error=0;

  490.         int64_t sum=0;

  491. 

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

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

  494.         }

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

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

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

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

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

  500.             error= v - intV*sum;

  501.         }

  502.     }

  503. 

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

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

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

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

  508.     }

  509. 

  510.     ret=0;

  511. fail:

  512.     av_free(filter);

  513.     av_free(filter2);

  514.     return ret;

  515. }

  516. 

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

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

  519. {

  520.     uint8_t *fragmentA;

  521.     x86_reg imm8OfPShufW1A;

  522.     x86_reg imm8OfPShufW2A;

  523.     x86_reg fragmentLengthA;

  524.     uint8_t *fragmentB;

  525.     x86_reg imm8OfPShufW1B;

  526.     x86_reg imm8OfPShufW2B;

  527.     x86_reg fragmentLengthB;

  528.     int fragmentPos;

  529. 

  530.     int xpos, i;

  531. 

  532.     // create an optimized horizontal scaling routine

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

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

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

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

  537.      * of fragmentA is used.

  538.      */

  539. 

  540.     //code fragment

  541. 

  542.     __asm__ volatile(

  543.         "jmp                         9f                 \n\t"

  544.     // Begin

  545.         "0:                                             \n\t"

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

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

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

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

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

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

  552.         "1:                                             \n\t"

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

  554.         "2:                                             \n\t"

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

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

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

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

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

  560. 

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

  562. 

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

  564.     // End

  565.         "9:                                             \n\t"

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

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

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

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

  570.         "dec                         %1                 \n\t"

  571.         "dec                         %2                 \n\t"

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

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

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

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

  576. 

  577. 

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

  579.         "=r" (fragmentLengthA)

  580.     );

  581. 

  582.     __asm__ volatile(

  583.         "jmp                         9f                 \n\t"

  584.     // Begin

  585.         "0:                                             \n\t"

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

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

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

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

  590.         "1:                                             \n\t"

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

  592.         "2:                                             \n\t"

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

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

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

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

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

  598. 

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

  600. 

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

  602.     // End

  603.         "9:                                             \n\t"

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

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

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

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

  608.         "dec                         %1                 \n\t"

  609.         "dec                         %2                 \n\t"

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

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

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

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

  614. 

  615. 

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

  617.         "=r" (fragmentLengthB)

  618.     );

  619. 

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

  621.     fragmentPos=0;

  622. 

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

  624.         int xx=xpos>>16;

  625. 

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

  627.             int a=0;

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

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

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

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

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

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

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

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

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

  637.             int shift=0;

  638. 

  639.             if (filterCode) {

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

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

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

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

  644.                 filterPos[i/2]= xx;

  645. 

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

  647. 

  648.                 filterCode[fragmentPos + imm8OfPShufW1]=

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

  650.                 filterCode[fragmentPos + imm8OfPShufW2]=

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

  652. 

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

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

  655. 

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

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

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

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

  660.                 }

  661.             }

  662. 

  663.             fragmentPos+= fragmentLength;

  664. 

  665.             if (filterCode)

  666.                 filterCode[fragmentPos]= RET;

  667.         }

  668.         xpos+=xInc;

  669.     }

  670.     if (filterCode)

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

  672. 

  673.     return fragmentPos + 1;

  674. }

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

  676. 

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

  678. {

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

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

  681. }

  682. 

  683. static int update_flags_cpu(int flags);

  684. 

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

  686. {

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

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

  689. 

  690.     c->brightness= brightness;

  691.     c->contrast  = contrast;

  692.     c->saturation= saturation;

  693.     c->srcRange  = srcRange;

  694.     c->dstRange  = dstRange;

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

  696. 

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

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

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

  700. 

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

  702.     //FIXME factorize

  703. 

  704. #if HAVE_ALTIVEC

  705.     if (c->flags & SWS_CPU_CAPS_ALTIVEC)

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

  707. #endif

  708.     return 0;

  709. }

  710. 

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

  712. {

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

  714. 

  715.     *inv_table = c->srcColorspaceTable;

  716.     *table     = c->dstColorspaceTable;

  717.     *srcRange  = c->srcRange;

  718.     *dstRange  = c->dstRange;

  719.     *brightness= c->brightness;

  720.     *contrast  = c->contrast;

  721.     *saturation= c->saturation;

  722. 

  723.     return 0;

  724. }

  725. 

  726. static int handle_jpeg(enum PixelFormat *format)

  727. {

  728.     switch (*format) {

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

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

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

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

  733.     default:                                          return 0;

  734.     }

  735. }

  736. 

  737. static int update_flags_cpu(int flags)

  738. {

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

  740.     flags &= ~( SWS_CPU_CAPS_MMX

  741.                |SWS_CPU_CAPS_MMX2

  742.                |SWS_CPU_CAPS_3DNOW

  743.                |SWS_CPU_CAPS_SSE2

  744.                |SWS_CPU_CAPS_ALTIVEC

  745.                |SWS_CPU_CAPS_BFIN);

  746.     flags |= ff_hardcodedcpuflags();

  747. #endif /* CONFIG_RUNTIME_CPUDETECT */

  748.     return flags;

  749. }

  750. 

  751. SwsContext *sws_alloc_context(void)

  752. {

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

  754. 

  755.     c->av_class = &sws_context_class;

  756.     av_opt_set_defaults(c);

  757. 

  758.     return c;

  759. }

  760. 

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

  762. {

  763.     int i;

  764.     int usesVFilter, usesHFilter;

  765.     int unscaled;

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

  767.     int srcW= c->srcW;

  768.     int srcH= c->srcH;

  769.     int dstW= c->dstW;

  770.     int dstH= c->dstH;

  771.     int flags;

  772.     enum PixelFormat srcFormat= c->srcFormat;

  773.     enum PixelFormat dstFormat= c->dstFormat;

  774. 

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

  776. #if ARCH_X86

  777.     if (flags & SWS_CPU_CAPS_MMX)

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

  779. #endif

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

  781. 

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

  783. 

  784.     if (!isSupportedIn(srcFormat)) {

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

  786.         return AVERROR(EINVAL);

  787.     }

  788.     if (!isSupportedOut(dstFormat)) {

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

  790.         return AVERROR(EINVAL);

  791.     }

  792. 

  793.     i= flags & ( SWS_POINT

  794.                 |SWS_AREA

  795.                 |SWS_BILINEAR

  796.                 |SWS_FAST_BILINEAR

  797.                 |SWS_BICUBIC

  798.                 |SWS_X

  799.                 |SWS_GAUSS

  800.                 |SWS_LANCZOS

  801.                 |SWS_SINC

  802.                 |SWS_SPLINE

  803.                 |SWS_BICUBLIN);

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

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

  806.         return AVERROR(EINVAL);

  807.     }

  808.     /* sanity check */

  809.     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

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

  811.                srcW, srcH, dstW, dstH);

  812.         return AVERROR(EINVAL);

  813.     }

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

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

  816.         return AVERROR(EINVAL);

  817.     }

  818. 

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

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

  821. 

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

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

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

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

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

  827. 

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

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

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

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

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

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

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

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

  836. 

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

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

  839. 

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

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

  842. 

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

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

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

  846. 

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

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

  849.       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8

  850.       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4

  851.       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE

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

  853.         c->chrSrcHSubSample=1;

  854. 

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

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

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

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

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

  860. 

  861.     /* unscaled special cases */

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

  863.         ff_get_unscaled_swscale(c);

  864. 

  865.         if (c->swScale) {

  866.             if (flags&SWS_PRINT_INFO)

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

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

  869.             return 0;

  870.         }

  871.     }

  872. 

  873.     if (flags & SWS_CPU_CAPS_MMX2) {

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

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

  876.             if (flags&SWS_PRINT_INFO)

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

  878.         }

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

  880.     }

  881.     else

  882.         c->canMMX2BeUsed=0;

  883. 

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

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

  886. 

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

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

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

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

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

  892.     // first and last pixel

  893.     if (flags&SWS_FAST_BILINEAR) {

  894.         if (c->canMMX2BeUsed) {

  895.             c->lumXInc+= 20;

  896.             c->chrXInc+= 20;

  897.         }

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

  899.         else if (flags & SWS_CPU_CAPS_MMX) {

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

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

  902.         }

  903.     }

  904. 

  905.     /* precalculate horizontal scaler filter coefficients */

  906.     {

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

  908. // can't downscale !!!

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

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

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

  912. 

  913. #ifdef MAP_ANONYMOUS

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

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

  916. #elif HAVE_VIRTUALALLOC

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

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

  919. #else

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

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

  922. #endif

  923. 

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

  925.                 return AVERROR(ENOMEM);

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

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

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

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

  930. 

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

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

  933. 

  934. #ifdef MAP_ANONYMOUS

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

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

  937. #endif

  938.         } else

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

  940.         {

  941.             const int filterAlign=

  942.                 (flags & SWS_CPU_CAPS_MMX) ? 4 :

  943.                 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :

  944.                 1;

  945. 

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

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

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

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

  950.                 goto fail;

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

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

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

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

  955.                 goto fail;

  956.         }

  957.     } // initialize horizontal stuff

  958. 

  959.     /* precalculate vertical scaler filter coefficients */

  960.     {

  961.         const int filterAlign=

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

  963.             (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :

  964.             1;

  965. 

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

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

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

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

  970.             goto fail;

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

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

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

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

  975.             goto fail;

  976. 

  977. #if HAVE_ALTIVEC

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

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

  980. 

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

  982.             int j;

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

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

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

  986.         }

  987. 

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

  989.             int j;

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

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

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

  993.         }

  994. #endif

  995.     }

  996. 

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

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

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

  1000.     for (i=0; i<dstH; i++) {

  1001.         int chrI= i*c->chrDstH / dstH;

  1002.         int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,

  1003.                            ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));

  1004. 

  1005.         nextSlice>>= c->chrSrcVSubSample;

  1006.         nextSlice<<= c->chrSrcVSubSample;

  1007.         if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)

  1008.             c->vLumBufSize= nextSlice - c->vLumFilterPos[i];

  1009.         if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))

  1010.             c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];

  1011.     }

  1012. 

  1013.     // allocate pixbufs (we use dynamic allocation because otherwise we would need to

  1014.     // allocate several megabytes to handle all possible cases)

  1015.     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1016.     FF_ALLOC_OR_GOTO(c, c->chrPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);

  1017.     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))

  1018.         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);

  1019.     //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)

  1020.     /* align at 16 bytes for AltiVec */

  1021.     for (i=0; i<c->vLumBufSize; i++) {

  1022.         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], VOF+1, fail);

  1023.         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];

  1024.     }

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

  1026.         FF_ALLOC_OR_GOTO(c, c->chrPixBuf[i+c->vChrBufSize], (VOF+1)*2, fail);

  1027.         c->chrPixBuf[i] = c->chrPixBuf[i+c->vChrBufSize];

  1028.     }

  1029.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)

  1030.         for (i=0; i<c->vLumBufSize; i++) {

  1031.             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], VOF+1, fail);

  1032.             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];

  1033.         }

  1034. 

  1035.     //try to avoid drawing green stuff between the right end and the stride end

  1036.     for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);

  1037. 

  1038.     assert(2*VOFW == VOF);

  1039. 

  1040.     assert(c->chrDstH <= dstH);

  1041. 

  1042.     if (flags&SWS_PRINT_INFO) {

  1043.         if      (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");

  1044.         else if (flags&SWS_BILINEAR)      av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");

  1045.         else if (flags&SWS_BICUBIC)       av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");

  1046.         else if (flags&SWS_X)             av_log(c, AV_LOG_INFO, "Experimental scaler, ");

  1047.         else if (flags&SWS_POINT)         av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");

  1048.         else if (flags&SWS_AREA)          av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");

  1049.         else if (flags&SWS_BICUBLIN)      av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");

  1050.         else if (flags&SWS_GAUSS)         av_log(c, AV_LOG_INFO, "Gaussian scaler, ");

  1051.         else if (flags&SWS_SINC)          av_log(c, AV_LOG_INFO, "Sinc scaler, ");

  1052.         else if (flags&SWS_LANCZOS)       av_log(c, AV_LOG_INFO, "Lanczos scaler, ");

  1053.         else if (flags&SWS_SPLINE)        av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");

  1054.         else                              av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");

  1055. 

  1056.         av_log(c, AV_LOG_INFO, "from %s to %s%s ",

  1057.                sws_format_name(srcFormat),

  1058. #ifdef DITHER1XBPP

  1059.                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||

  1060.                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1061.                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",

  1062. #else

  1063.                "",

  1064. #endif

  1065.                sws_format_name(dstFormat));

  1066. 

  1067.         if      (flags & SWS_CPU_CAPS_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");

  1068.         else if (flags & SWS_CPU_CAPS_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");

  1069.         else if (flags & SWS_CPU_CAPS_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");

  1070.         else if (flags & SWS_CPU_CAPS_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");

  1071.         else                                   av_log(c, AV_LOG_INFO, "using C\n");

  1072. 

  1073.         if (flags & SWS_CPU_CAPS_MMX) {

  1074.             if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))

  1075.                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");

  1076.             else {

  1077.                 if (c->hLumFilterSize==4)

  1078.                     av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");

  1079.                 else if (c->hLumFilterSize==8)

  1080.                     av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");

  1081.                 else

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

  1083. 

  1084.                 if (c->hChrFilterSize==4)

  1085.                     av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");

  1086.                 else if (c->hChrFilterSize==8)

  1087.                     av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");

  1088.                 else

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

  1090.             }

  1091.         } else {

  1092. #if ARCH_X86

  1093.             av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");

  1094. #else

  1095.             if (flags & SWS_FAST_BILINEAR)

  1096.                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");

  1097.             else

  1098.                 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");

  1099. #endif

  1100.         }

  1101.         if (isPlanarYUV(dstFormat)) {

  1102.             if (c->vLumFilterSize==1)

  1103.                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1104.             else

  1105.                 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1106.         } else {

  1107.             if (c->vLumFilterSize==1 && c->vChrFilterSize==2)

  1108.                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"

  1109.                        "      2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1110.             else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)

  1111.                 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1112.             else

  1113.                 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1114.         }

  1115. 

  1116.         if (dstFormat==PIX_FMT_BGR24)

  1117.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",

  1118.                    (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));

  1119.         else if (dstFormat==PIX_FMT_RGB32)

  1120.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1121.         else if (dstFormat==PIX_FMT_BGR565)

  1122.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1123.         else if (dstFormat==PIX_FMT_BGR555)

  1124.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1125.         else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||

  1126.                  dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)

  1127.             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");

  1128. 

  1129.         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);

  1130.         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1131.                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);

  1132.         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",

  1133.                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);

  1134.     }

  1135. 

  1136.     c->swScale= ff_getSwsFunc(c);

  1137.     return 0;

  1138. fail: //FIXME replace things by appropriate error codes

  1139.     return -1;

  1140. }

  1141. 

  1142. #if FF_API_SWS_GETCONTEXT

  1143. SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,

  1144.                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1145.                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1146. {

  1147.     SwsContext *c;

  1148. 

  1149.     if(!(c=sws_alloc_context()))

  1150.         return NULL;

  1151. 

  1152.     c->flags= flags;

  1153.     c->srcW= srcW;

  1154.     c->srcH= srcH;

  1155.     c->dstW= dstW;

  1156.     c->dstH= dstH;

  1157.     c->srcRange = handle_jpeg(&srcFormat);

  1158.     c->dstRange = handle_jpeg(&dstFormat);

  1159.     c->srcFormat= srcFormat;

  1160.     c->dstFormat= dstFormat;

  1161. 

  1162.     if (param) {

  1163.         c->param[0] = param[0];

  1164.         c->param[1] = param[1];

  1165.     }

  1166.     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);

  1167. 

  1168.     if(sws_init_context(c, srcFilter, dstFilter) < 0){

  1169.         sws_freeContext(c);

  1170.         return NULL;

  1171.     }

  1172. 

  1173.     return c;

  1174. }

  1175. #endif

  1176. 

  1177. SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,

  1178.                                 float lumaSharpen, float chromaSharpen,

  1179.                                 float chromaHShift, float chromaVShift,

  1180.                                 int verbose)

  1181. {

  1182.     SwsFilter *filter= av_malloc(sizeof(SwsFilter));

  1183.     if (!filter)

  1184.         return NULL;

  1185. 

  1186.     if (lumaGBlur!=0.0) {

  1187.         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);

  1188.         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);

  1189.     } else {

  1190.         filter->lumH= sws_getIdentityVec();

  1191.         filter->lumV= sws_getIdentityVec();

  1192.     }

  1193. 

  1194.     if (chromaGBlur!=0.0) {

  1195.         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);

  1196.         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);

  1197.     } else {

  1198.         filter->chrH= sws_getIdentityVec();

  1199.         filter->chrV= sws_getIdentityVec();

  1200.     }

  1201. 

  1202.     if (chromaSharpen!=0.0) {

  1203.         SwsVector *id= sws_getIdentityVec();

  1204.         sws_scaleVec(filter->chrH, -chromaSharpen);

  1205.         sws_scaleVec(filter->chrV, -chromaSharpen);

  1206.         sws_addVec(filter->chrH, id);

  1207.         sws_addVec(filter->chrV, id);

  1208.         sws_freeVec(id);

  1209.     }

  1210. 

  1211.     if (lumaSharpen!=0.0) {

  1212.         SwsVector *id= sws_getIdentityVec();

  1213.         sws_scaleVec(filter->lumH, -lumaSharpen);

  1214.         sws_scaleVec(filter->lumV, -lumaSharpen);

  1215.         sws_addVec(filter->lumH, id);

  1216.         sws_addVec(filter->lumV, id);

  1217.         sws_freeVec(id);

  1218.     }

  1219. 

  1220.     if (chromaHShift != 0.0)

  1221.         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));

  1222. 

  1223.     if (chromaVShift != 0.0)

  1224.         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));

  1225. 

  1226.     sws_normalizeVec(filter->chrH, 1.0);

  1227.     sws_normalizeVec(filter->chrV, 1.0);

  1228.     sws_normalizeVec(filter->lumH, 1.0);

  1229.     sws_normalizeVec(filter->lumV, 1.0);

  1230. 

  1231.     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);

  1232.     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);

  1233. 

  1234.     return filter;

  1235. }

  1236. 

  1237. SwsVector *sws_allocVec(int length)

  1238. {

  1239.     SwsVector *vec = av_malloc(sizeof(SwsVector));

  1240.     if (!vec)

  1241.         return NULL;

  1242.     vec->length = length;

  1243.     vec->coeff  = av_malloc(sizeof(double) * length);

  1244.     if (!vec->coeff)

  1245.         av_freep(&vec);

  1246.     return vec;

  1247. }

  1248. 

  1249. SwsVector *sws_getGaussianVec(double variance, double quality)

  1250. {

  1251.     const int length= (int)(variance*quality + 0.5) | 1;

  1252.     int i;

  1253.     double middle= (length-1)*0.5;

  1254.     SwsVector *vec= sws_allocVec(length);

  1255. 

  1256.     if (!vec)

  1257.         return NULL;

  1258. 

  1259.     for (i=0; i<length; i++) {

  1260.         double dist= i-middle;

  1261.         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);

  1262.     }

  1263. 

  1264.     sws_normalizeVec(vec, 1.0);

  1265. 

  1266.     return vec;

  1267. }

  1268. 

  1269. SwsVector *sws_getConstVec(double c, int length)

  1270. {

  1271.     int i;

  1272.     SwsVector *vec= sws_allocVec(length);

  1273. 

  1274.     if (!vec)

  1275.         return NULL;

  1276. 

  1277.     for (i=0; i<length; i++)

  1278.         vec->coeff[i]= c;

  1279. 

  1280.     return vec;

  1281. }

  1282. 

  1283. SwsVector *sws_getIdentityVec(void)

  1284. {

  1285.     return sws_getConstVec(1.0, 1);

  1286. }

  1287. 

  1288. static double sws_dcVec(SwsVector *a)

  1289. {

  1290.     int i;

  1291.     double sum=0;

  1292. 

  1293.     for (i=0; i<a->length; i++)

  1294.         sum+= a->coeff[i];

  1295. 

  1296.     return sum;

  1297. }

  1298. 

  1299. void sws_scaleVec(SwsVector *a, double scalar)

  1300. {

  1301.     int i;

  1302. 

  1303.     for (i=0; i<a->length; i++)

  1304.         a->coeff[i]*= scalar;

  1305. }

  1306. 

  1307. void sws_normalizeVec(SwsVector *a, double height)

  1308. {

  1309.     sws_scaleVec(a, height/sws_dcVec(a));

  1310. }

  1311. 

  1312. static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)

  1313. {

  1314.     int length= a->length + b->length - 1;

  1315.     int i, j;

  1316.     SwsVector *vec= sws_getConstVec(0.0, length);

  1317. 

  1318.     if (!vec)

  1319.         return NULL;

  1320. 

  1321.     for (i=0; i<a->length; i++) {

  1322.         for (j=0; j<b->length; j++) {

  1323.             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];

  1324.         }

  1325.     }

  1326. 

  1327.     return vec;

  1328. }

  1329. 

  1330. static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)

  1331. {

  1332.     int length= FFMAX(a->length, b->length);

  1333.     int i;

  1334.     SwsVector *vec= sws_getConstVec(0.0, length);

  1335. 

  1336.     if (!vec)

  1337.         return NULL;

  1338. 

  1339.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1340.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];

  1341. 

  1342.     return vec;

  1343. }

  1344. 

  1345. static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)

  1346. {

  1347.     int length= FFMAX(a->length, b->length);

  1348.     int i;

  1349.     SwsVector *vec= sws_getConstVec(0.0, length);

  1350. 

  1351.     if (!vec)

  1352.         return NULL;

  1353. 

  1354.     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];

  1355.     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];

  1356. 

  1357.     return vec;

  1358. }

  1359. 

  1360. /* shift left / or right if "shift" is negative */

  1361. static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)

  1362. {

  1363.     int length= a->length + FFABS(shift)*2;

  1364.     int i;

  1365.     SwsVector *vec= sws_getConstVec(0.0, length);

  1366. 

  1367.     if (!vec)

  1368.         return NULL;

  1369. 

  1370.     for (i=0; i<a->length; i++) {

  1371.         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];

  1372.     }

  1373. 

  1374.     return vec;

  1375. }

  1376. 

  1377. void sws_shiftVec(SwsVector *a, int shift)

  1378. {

  1379.     SwsVector *shifted= sws_getShiftedVec(a, shift);

  1380.     av_free(a->coeff);

  1381.     a->coeff= shifted->coeff;

  1382.     a->length= shifted->length;

  1383.     av_free(shifted);

  1384. }

  1385. 

  1386. void sws_addVec(SwsVector *a, SwsVector *b)

  1387. {

  1388.     SwsVector *sum= sws_sumVec(a, b);

  1389.     av_free(a->coeff);

  1390.     a->coeff= sum->coeff;

  1391.     a->length= sum->length;

  1392.     av_free(sum);

  1393. }

  1394. 

  1395. void sws_subVec(SwsVector *a, SwsVector *b)

  1396. {

  1397.     SwsVector *diff= sws_diffVec(a, b);

  1398.     av_free(a->coeff);

  1399.     a->coeff= diff->coeff;

  1400.     a->length= diff->length;

  1401.     av_free(diff);

  1402. }

  1403. 

  1404. void sws_convVec(SwsVector *a, SwsVector *b)

  1405. {

  1406.     SwsVector *conv= sws_getConvVec(a, b);

  1407.     av_free(a->coeff);

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

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

  1410.     av_free(conv);

  1411. }

  1412. 

  1413. SwsVector *sws_cloneVec(SwsVector *a)

  1414. {

  1415.     int i;

  1416.     SwsVector *vec= sws_allocVec(a->length);

  1417. 

  1418.     if (!vec)

  1419.         return NULL;

  1420. 

  1421.     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];

  1422. 

  1423.     return vec;

  1424. }

  1425. 

  1426. void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)

  1427. {

  1428.     int i;

  1429.     double max=0;

  1430.     double min=0;

  1431.     double range;

  1432. 

  1433.     for (i=0; i<a->length; i++)

  1434.         if (a->coeff[i]>max) max= a->coeff[i];

  1435. 

  1436.     for (i=0; i<a->length; i++)

  1437.         if (a->coeff[i]<min) min= a->coeff[i];

  1438. 

  1439.     range= max - min;

  1440. 

  1441.     for (i=0; i<a->length; i++) {

  1442.         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);

  1443.         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);

  1444.         for (;x>0; x--) av_log(log_ctx, log_level, " ");

  1445.         av_log(log_ctx, log_level, "|\n");

  1446.     }

  1447. }

  1448. 

  1449. void sws_freeVec(SwsVector *a)

  1450. {

  1451.     if (!a) return;

  1452.     av_freep(&a->coeff);

  1453.     a->length=0;

  1454.     av_free(a);

  1455. }

  1456. 

  1457. void sws_freeFilter(SwsFilter *filter)

  1458. {

  1459.     if (!filter) return;

  1460. 

  1461.     if (filter->lumH) sws_freeVec(filter->lumH);

  1462.     if (filter->lumV) sws_freeVec(filter->lumV);

  1463.     if (filter->chrH) sws_freeVec(filter->chrH);

  1464.     if (filter->chrV) sws_freeVec(filter->chrV);

  1465.     av_free(filter);

  1466. }

  1467. 

  1468. void sws_freeContext(SwsContext *c)

  1469. {

  1470.     int i;

  1471.     if (!c) return;

  1472. 

  1473.     if (c->lumPixBuf) {

  1474.         for (i=0; i<c->vLumBufSize; i++)

  1475.             av_freep(&c->lumPixBuf[i]);

  1476.         av_freep(&c->lumPixBuf);

  1477.     }

  1478. 

  1479.     if (c->chrPixBuf) {

  1480.         for (i=0; i<c->vChrBufSize; i++)

  1481.             av_freep(&c->chrPixBuf[i]);

  1482.         av_freep(&c->chrPixBuf);

  1483.     }

  1484. 

  1485.     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {

  1486.         for (i=0; i<c->vLumBufSize; i++)

  1487.             av_freep(&c->alpPixBuf[i]);

  1488.         av_freep(&c->alpPixBuf);

  1489.     }

  1490. 

  1491.     av_freep(&c->vLumFilter);

  1492.     av_freep(&c->vChrFilter);

  1493.     av_freep(&c->hLumFilter);

  1494.     av_freep(&c->hChrFilter);

  1495. #if HAVE_ALTIVEC

  1496.     av_freep(&c->vYCoeffsBank);

  1497.     av_freep(&c->vCCoeffsBank);

  1498. #endif

  1499. 

  1500.     av_freep(&c->vLumFilterPos);

  1501.     av_freep(&c->vChrFilterPos);

  1502.     av_freep(&c->hLumFilterPos);

  1503.     av_freep(&c->hChrFilterPos);

  1504. 

  1505. #if ARCH_X86

  1506. #ifdef MAP_ANONYMOUS

  1507.     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);

  1508.     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);

  1509. #elif HAVE_VIRTUALALLOC

  1510.     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);

  1511.     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);

  1512. #else

  1513.     av_free(c->lumMmx2FilterCode);

  1514.     av_free(c->chrMmx2FilterCode);

  1515. #endif

  1516.     c->lumMmx2FilterCode=NULL;

  1517.     c->chrMmx2FilterCode=NULL;

  1518. #endif /* ARCH_X86 */

  1519. 

  1520.     av_freep(&c->yuvTable);

  1521. 

  1522.     av_free(c);

  1523. }

  1524. 

  1525. struct SwsContext *sws_getCachedContext(struct SwsContext *context,

  1526.                                         int srcW, int srcH, enum PixelFormat srcFormat,

  1527.                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,

  1528.                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)

  1529. {

  1530.     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};

  1531. 

  1532.     if (!param)

  1533.         param = default_param;

  1534. 

  1535.     flags = update_flags_cpu(flags);

  1536. 

  1537.     if (context &&

  1538.         (context->srcW      != srcW      ||

  1539.          context->srcH      != srcH      ||

  1540.          context->srcFormat != srcFormat ||

  1541.          context->dstW      != dstW      ||

  1542.          context->dstH      != dstH      ||

  1543.          context->dstFormat != dstFormat ||

  1544.          context->flags     != flags     ||

  1545.          context->param[0]  != param[0]  ||

  1546.          context->param[1]  != param[1])) {

  1547.         sws_freeContext(context);

  1548.         context = NULL;

  1549.     }

  1550. 

  1551.     if (!context) {

  1552.         if (!(context = sws_alloc_context()))

  1553.             return NULL;

  1554.         context->srcW      = srcW;

  1555.         context->srcH      = srcH;

  1556.         context->srcRange  = handle_jpeg(&srcFormat);

  1557.         context->srcFormat = srcFormat;

  1558.         context->dstW      = dstW;

  1559.         context->dstH      = dstH;

  1560.         context->dstRange  = handle_jpeg(&dstFormat);

  1561.         context->dstFormat = dstFormat;

  1562.         context->flags     = flags;

  1563.         context->param[0]  = param[0];

  1564.         context->param[1]  = param[1];

  1565.         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);

  1566.         if (sws_init_context(context, srcFilter, dstFilter) < 0) {

  1567.             sws_freeContext(context);

  1568.             return NULL;

  1569.         }

  1570.     }

  1571.     return context;

  1572. }