Libav 0.7.1
libswscale/utils.c
Go to the documentation of this file.
00001 /*
00002  * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
00003  *
00004  * This file is part of Libav.
00005  *
00006  * Libav is free software; you can redistribute it and/or
00007  * modify it under the terms of the GNU Lesser General Public
00008  * License as published by the Free Software Foundation; either
00009  * version 2.1 of the License, or (at your option) any later version.
00010  *
00011  * Libav is distributed in the hope that it will be useful,
00012  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00013  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00014  * Lesser General Public License for more details.
00015  *
00016  * You should have received a copy of the GNU Lesser General Public
00017  * License along with Libav; if not, write to the Free Software
00018  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00019  */
00020 
00021 #include <inttypes.h>
00022 #include <string.h>
00023 #include <math.h>
00024 #include <stdio.h>
00025 #include "config.h"
00026 #include <assert.h>
00027 #if HAVE_SYS_MMAN_H
00028 #include <sys/mman.h>
00029 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
00030 #define MAP_ANONYMOUS MAP_ANON
00031 #endif
00032 #endif
00033 #if HAVE_VIRTUALALLOC
00034 #define WIN32_LEAN_AND_MEAN
00035 #include <windows.h>
00036 #endif
00037 #include "swscale.h"
00038 #include "swscale_internal.h"
00039 #include "rgb2rgb.h"
00040 #include "libavutil/intreadwrite.h"
00041 #include "libavutil/x86_cpu.h"
00042 #include "libavutil/cpu.h"
00043 #include "libavutil/avutil.h"
00044 #include "libavutil/bswap.h"
00045 #include "libavutil/opt.h"
00046 #include "libavutil/pixdesc.h"
00047 
00048 unsigned swscale_version(void)
00049 {
00050     return LIBSWSCALE_VERSION_INT;
00051 }
00052 
00053 const char *swscale_configuration(void)
00054 {
00055     return LIBAV_CONFIGURATION;
00056 }
00057 
00058 const char *swscale_license(void)
00059 {
00060 #define LICENSE_PREFIX "libswscale license: "
00061     return LICENSE_PREFIX LIBAV_LICENSE + sizeof(LICENSE_PREFIX) - 1;
00062 }
00063 
00064 #define RET 0xC3 //near return opcode for x86
00065 
00066 #define isSupportedIn(x)    (       \
00067            (x)==PIX_FMT_YUV420P     \
00068         || (x)==PIX_FMT_YUVA420P    \
00069         || (x)==PIX_FMT_YUYV422     \
00070         || (x)==PIX_FMT_UYVY422     \
00071         || (x)==PIX_FMT_RGB48BE     \
00072         || (x)==PIX_FMT_RGB48LE     \
00073         || (x)==PIX_FMT_RGB32       \
00074         || (x)==PIX_FMT_RGB32_1     \
00075         || (x)==PIX_FMT_BGR48BE     \
00076         || (x)==PIX_FMT_BGR48LE     \
00077         || (x)==PIX_FMT_BGR24       \
00078         || (x)==PIX_FMT_BGR565LE    \
00079         || (x)==PIX_FMT_BGR565BE    \
00080         || (x)==PIX_FMT_BGR555LE    \
00081         || (x)==PIX_FMT_BGR555BE    \
00082         || (x)==PIX_FMT_BGR32       \
00083         || (x)==PIX_FMT_BGR32_1     \
00084         || (x)==PIX_FMT_RGB24       \
00085         || (x)==PIX_FMT_RGB565LE    \
00086         || (x)==PIX_FMT_RGB565BE    \
00087         || (x)==PIX_FMT_RGB555LE    \
00088         || (x)==PIX_FMT_RGB555BE    \
00089         || (x)==PIX_FMT_GRAY8       \
00090         || (x)==PIX_FMT_Y400A       \
00091         || (x)==PIX_FMT_YUV410P     \
00092         || (x)==PIX_FMT_YUV440P     \
00093         || (x)==PIX_FMT_NV12        \
00094         || (x)==PIX_FMT_NV21        \
00095         || (x)==PIX_FMT_GRAY16BE    \
00096         || (x)==PIX_FMT_GRAY16LE    \
00097         || (x)==PIX_FMT_YUV444P     \
00098         || (x)==PIX_FMT_YUV422P     \
00099         || (x)==PIX_FMT_YUV411P     \
00100         || (x)==PIX_FMT_YUVJ420P    \
00101         || (x)==PIX_FMT_YUVJ422P    \
00102         || (x)==PIX_FMT_YUVJ440P    \
00103         || (x)==PIX_FMT_YUVJ444P    \
00104         || (x)==PIX_FMT_PAL8        \
00105         || (x)==PIX_FMT_BGR8        \
00106         || (x)==PIX_FMT_RGB8        \
00107         || (x)==PIX_FMT_BGR4_BYTE   \
00108         || (x)==PIX_FMT_RGB4_BYTE   \
00109         || (x)==PIX_FMT_YUV440P     \
00110         || (x)==PIX_FMT_MONOWHITE   \
00111         || (x)==PIX_FMT_MONOBLACK   \
00112         || (x)==PIX_FMT_YUV420P9LE    \
00113         || (x)==PIX_FMT_YUV444P9LE    \
00114         || (x)==PIX_FMT_YUV420P10LE   \
00115         || (x)==PIX_FMT_YUV422P10LE   \
00116         || (x)==PIX_FMT_YUV444P10LE   \
00117         || (x)==PIX_FMT_YUV420P16LE   \
00118         || (x)==PIX_FMT_YUV422P16LE   \
00119         || (x)==PIX_FMT_YUV444P16LE   \
00120         || (x)==PIX_FMT_YUV420P9BE    \
00121         || (x)==PIX_FMT_YUV444P9BE    \
00122         || (x)==PIX_FMT_YUV420P10BE   \
00123         || (x)==PIX_FMT_YUV444P10BE   \
00124         || (x)==PIX_FMT_YUV422P10BE   \
00125         || (x)==PIX_FMT_YUV420P16BE   \
00126         || (x)==PIX_FMT_YUV422P16BE   \
00127         || (x)==PIX_FMT_YUV444P16BE   \
00128     )
00129 
00130 int sws_isSupportedInput(enum PixelFormat pix_fmt)
00131 {
00132     return isSupportedIn(pix_fmt);
00133 }
00134 
00135 #define isSupportedOut(x)   (       \
00136            (x)==PIX_FMT_YUV420P     \
00137         || (x)==PIX_FMT_YUVA420P    \
00138         || (x)==PIX_FMT_YUYV422     \
00139         || (x)==PIX_FMT_UYVY422     \
00140         || (x)==PIX_FMT_YUV444P     \
00141         || (x)==PIX_FMT_YUV422P     \
00142         || (x)==PIX_FMT_YUV411P     \
00143         || (x)==PIX_FMT_YUVJ420P    \
00144         || (x)==PIX_FMT_YUVJ422P    \
00145         || (x)==PIX_FMT_YUVJ440P    \
00146         || (x)==PIX_FMT_YUVJ444P    \
00147         || isRGBinBytes(x)          \
00148         || isBGRinBytes(x)          \
00149         || (x)==PIX_FMT_RGB565      \
00150         || (x)==PIX_FMT_RGB555      \
00151         || (x)==PIX_FMT_RGB444      \
00152         || (x)==PIX_FMT_BGR565      \
00153         || (x)==PIX_FMT_BGR555      \
00154         || (x)==PIX_FMT_BGR444      \
00155         || (x)==PIX_FMT_RGB8        \
00156         || (x)==PIX_FMT_BGR8        \
00157         || (x)==PIX_FMT_RGB4_BYTE   \
00158         || (x)==PIX_FMT_BGR4_BYTE   \
00159         || (x)==PIX_FMT_RGB4        \
00160         || (x)==PIX_FMT_BGR4        \
00161         || (x)==PIX_FMT_MONOBLACK   \
00162         || (x)==PIX_FMT_MONOWHITE   \
00163         || (x)==PIX_FMT_NV12        \
00164         || (x)==PIX_FMT_NV21        \
00165         || (x)==PIX_FMT_GRAY16BE    \
00166         || (x)==PIX_FMT_GRAY16LE    \
00167         || (x)==PIX_FMT_GRAY8       \
00168         || (x)==PIX_FMT_YUV410P     \
00169         || (x)==PIX_FMT_YUV440P     \
00170         || (x)==PIX_FMT_YUV420P9LE    \
00171         || (x)==PIX_FMT_YUV420P10LE   \
00172         || (x)==PIX_FMT_YUV420P16LE   \
00173         || (x)==PIX_FMT_YUV422P16LE   \
00174         || (x)==PIX_FMT_YUV444P16LE   \
00175         || (x)==PIX_FMT_YUV420P9BE    \
00176         || (x)==PIX_FMT_YUV420P10BE   \
00177         || (x)==PIX_FMT_YUV420P16BE   \
00178         || (x)==PIX_FMT_YUV422P16BE   \
00179         || (x)==PIX_FMT_YUV444P16BE   \
00180     )
00181 
00182 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
00183 {
00184     return isSupportedOut(pix_fmt);
00185 }
00186 
00187 extern const int32_t ff_yuv2rgb_coeffs[8][4];
00188 
00189 const char *sws_format_name(enum PixelFormat format)
00190 {
00191     if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
00192         return av_pix_fmt_descriptors[format].name;
00193     else
00194         return "Unknown format";
00195 }
00196 
00197 static double getSplineCoeff(double a, double b, double c, double d, double dist)
00198 {
00199     if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
00200     else           return getSplineCoeff(        0.0,
00201                                           b+ 2.0*c + 3.0*d,
00202                                                  c + 3.0*d,
00203                                          -b- 3.0*c - 6.0*d,
00204                                          dist-1.0);
00205 }
00206 
00207 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
00208                       int srcW, int dstW, int filterAlign, int one, int flags, int cpu_flags,
00209                       SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
00210 {
00211     int i;
00212     int filterSize;
00213     int filter2Size;
00214     int minFilterSize;
00215     int64_t *filter=NULL;
00216     int64_t *filter2=NULL;
00217     const int64_t fone= 1LL<<54;
00218     int ret= -1;
00219 
00220     emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions)
00221 
00222     // NOTE: the +1 is for the MMX scaler which reads over the end
00223     FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
00224 
00225     if (FFABS(xInc - 0x10000) <10) { // unscaled
00226         int i;
00227         filterSize= 1;
00228         FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
00229 
00230         for (i=0; i<dstW; i++) {
00231             filter[i*filterSize]= fone;
00232             (*filterPos)[i]=i;
00233         }
00234 
00235     } else if (flags&SWS_POINT) { // lame looking point sampling mode
00236         int i;
00237         int xDstInSrc;
00238         filterSize= 1;
00239         FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
00240 
00241         xDstInSrc= xInc/2 - 0x8000;
00242         for (i=0; i<dstW; i++) {
00243             int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
00244 
00245             (*filterPos)[i]= xx;
00246             filter[i]= fone;
00247             xDstInSrc+= xInc;
00248         }
00249     } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
00250         int i;
00251         int xDstInSrc;
00252         filterSize= 2;
00253         FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
00254 
00255         xDstInSrc= xInc/2 - 0x8000;
00256         for (i=0; i<dstW; i++) {
00257             int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
00258             int j;
00259 
00260             (*filterPos)[i]= xx;
00261             //bilinear upscale / linear interpolate / area averaging
00262             for (j=0; j<filterSize; j++) {
00263                 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
00264                 if (coeff<0) coeff=0;
00265                 filter[i*filterSize + j]= coeff;
00266                 xx++;
00267             }
00268             xDstInSrc+= xInc;
00269         }
00270     } else {
00271         int xDstInSrc;
00272         int sizeFactor;
00273 
00274         if      (flags&SWS_BICUBIC)      sizeFactor=  4;
00275         else if (flags&SWS_X)            sizeFactor=  8;
00276         else if (flags&SWS_AREA)         sizeFactor=  1; //downscale only, for upscale it is bilinear
00277         else if (flags&SWS_GAUSS)        sizeFactor=  8;   // infinite ;)
00278         else if (flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
00279         else if (flags&SWS_SINC)         sizeFactor= 20; // infinite ;)
00280         else if (flags&SWS_SPLINE)       sizeFactor= 20;  // infinite ;)
00281         else if (flags&SWS_BILINEAR)     sizeFactor=  2;
00282         else {
00283             sizeFactor= 0; //GCC warning killer
00284             assert(0);
00285         }
00286 
00287         if (xInc <= 1<<16)      filterSize= 1 + sizeFactor; // upscale
00288         else                    filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
00289 
00290         if (filterSize > srcW-2) filterSize=srcW-2;
00291 
00292         FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
00293 
00294         xDstInSrc= xInc - 0x10000;
00295         for (i=0; i<dstW; i++) {
00296             int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
00297             int j;
00298             (*filterPos)[i]= xx;
00299             for (j=0; j<filterSize; j++) {
00300                 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
00301                 double floatd;
00302                 int64_t coeff;
00303 
00304                 if (xInc > 1<<16)
00305                     d= d*dstW/srcW;
00306                 floatd= d * (1.0/(1<<30));
00307 
00308                 if (flags & SWS_BICUBIC) {
00309                     int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] :   0) * (1<<24);
00310                     int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
00311                     int64_t dd = ( d*d)>>30;
00312                     int64_t ddd= (dd*d)>>30;
00313 
00314                     if      (d < 1LL<<30)
00315                         coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
00316                     else if (d < 1LL<<31)
00317                         coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
00318                     else
00319                         coeff=0.0;
00320                     coeff *= fone>>(30+24);
00321                 }
00322 /*                else if (flags & SWS_X) {
00323                     double p= param ? param*0.01 : 0.3;
00324                     coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
00325                     coeff*= pow(2.0, - p*d*d);
00326                 }*/
00327                 else if (flags & SWS_X) {
00328                     double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
00329                     double c;
00330 
00331                     if (floatd<1.0)
00332                         c = cos(floatd*M_PI);
00333                     else
00334                         c=-1.0;
00335                     if (c<0.0)      c= -pow(-c, A);
00336                     else            c=  pow( c, A);
00337                     coeff= (c*0.5 + 0.5)*fone;
00338                 } else if (flags & SWS_AREA) {
00339                     int64_t d2= d - (1<<29);
00340                     if      (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
00341                     else if (d2*xInc <  (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
00342                     else coeff=0.0;
00343                     coeff *= fone>>(30+16);
00344                 } else if (flags & SWS_GAUSS) {
00345                     double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
00346                     coeff = (pow(2.0, - p*floatd*floatd))*fone;
00347                 } else if (flags & SWS_SINC) {
00348                     coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
00349                 } else if (flags & SWS_LANCZOS) {
00350                     double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
00351                     coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
00352                     if (floatd>p) coeff=0;
00353                 } else if (flags & SWS_BILINEAR) {
00354                     coeff= (1<<30) - d;
00355                     if (coeff<0) coeff=0;
00356                     coeff *= fone >> 30;
00357                 } else if (flags & SWS_SPLINE) {
00358                     double p=-2.196152422706632;
00359                     coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
00360                 } else {
00361                     coeff= 0.0; //GCC warning killer
00362                     assert(0);
00363                 }
00364 
00365                 filter[i*filterSize + j]= coeff;
00366                 xx++;
00367             }
00368             xDstInSrc+= 2*xInc;
00369         }
00370     }
00371 
00372     /* apply src & dst Filter to filter -> filter2
00373        av_free(filter);
00374     */
00375     assert(filterSize>0);
00376     filter2Size= filterSize;
00377     if (srcFilter) filter2Size+= srcFilter->length - 1;
00378     if (dstFilter) filter2Size+= dstFilter->length - 1;
00379     assert(filter2Size>0);
00380     FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
00381 
00382     for (i=0; i<dstW; i++) {
00383         int j, k;
00384 
00385         if(srcFilter) {
00386             for (k=0; k<srcFilter->length; k++) {
00387                 for (j=0; j<filterSize; j++)
00388                     filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
00389             }
00390         } else {
00391             for (j=0; j<filterSize; j++)
00392                 filter2[i*filter2Size + j]= filter[i*filterSize + j];
00393         }
00394         //FIXME dstFilter
00395 
00396         (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
00397     }
00398     av_freep(&filter);
00399 
00400     /* try to reduce the filter-size (step1 find size and shift left) */
00401     // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
00402     minFilterSize= 0;
00403     for (i=dstW-1; i>=0; i--) {
00404         int min= filter2Size;
00405         int j;
00406         int64_t cutOff=0.0;
00407 
00408         /* get rid of near zero elements on the left by shifting left */
00409         for (j=0; j<filter2Size; j++) {
00410             int k;
00411             cutOff += FFABS(filter2[i*filter2Size]);
00412 
00413             if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
00414 
00415             /* preserve monotonicity because the core can't handle the filter otherwise */
00416             if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
00417 
00418             // move filter coefficients left
00419             for (k=1; k<filter2Size; k++)
00420                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
00421             filter2[i*filter2Size + k - 1]= 0;
00422             (*filterPos)[i]++;
00423         }
00424 
00425         cutOff=0;
00426         /* count near zeros on the right */
00427         for (j=filter2Size-1; j>0; j--) {
00428             cutOff += FFABS(filter2[i*filter2Size + j]);
00429 
00430             if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
00431             min--;
00432         }
00433 
00434         if (min>minFilterSize) minFilterSize= min;
00435     }
00436 
00437     if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
00438         // we can handle the special case 4,
00439         // so we don't want to go to the full 8
00440         if (minFilterSize < 5)
00441             filterAlign = 4;
00442 
00443         // We really don't want to waste our time
00444         // doing useless computation, so fall back on
00445         // the scalar C code for very small filters.
00446         // Vectorizing is worth it only if you have a
00447         // decent-sized vector.
00448         if (minFilterSize < 3)
00449             filterAlign = 1;
00450     }
00451 
00452     if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
00453         // special case for unscaled vertical filtering
00454         if (minFilterSize == 1 && filterAlign == 2)
00455             filterAlign= 1;
00456     }
00457 
00458     assert(minFilterSize > 0);
00459     filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
00460     assert(filterSize > 0);
00461     filter= av_malloc(filterSize*dstW*sizeof(*filter));
00462     if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
00463         goto fail;
00464     *outFilterSize= filterSize;
00465 
00466     if (flags&SWS_PRINT_INFO)
00467         av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
00468     /* try to reduce the filter-size (step2 reduce it) */
00469     for (i=0; i<dstW; i++) {
00470         int j;
00471 
00472         for (j=0; j<filterSize; j++) {
00473             if (j>=filter2Size) filter[i*filterSize + j]= 0;
00474             else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
00475             if((flags & SWS_BITEXACT) && j>=minFilterSize)
00476                 filter[i*filterSize + j]= 0;
00477         }
00478     }
00479 
00480     //FIXME try to align filterPos if possible
00481 
00482     //fix borders
00483     for (i=0; i<dstW; i++) {
00484         int j;
00485         if ((*filterPos)[i] < 0) {
00486             // move filter coefficients left to compensate for filterPos
00487             for (j=1; j<filterSize; j++) {
00488                 int left= FFMAX(j + (*filterPos)[i], 0);
00489                 filter[i*filterSize + left] += filter[i*filterSize + j];
00490                 filter[i*filterSize + j]=0;
00491             }
00492             (*filterPos)[i]= 0;
00493         }
00494 
00495         if ((*filterPos)[i] + filterSize > srcW) {
00496             int shift= (*filterPos)[i] + filterSize - srcW;
00497             // move filter coefficients right to compensate for filterPos
00498             for (j=filterSize-2; j>=0; j--) {
00499                 int right= FFMIN(j + shift, filterSize-1);
00500                 filter[i*filterSize +right] += filter[i*filterSize +j];
00501                 filter[i*filterSize +j]=0;
00502             }
00503             (*filterPos)[i]= srcW - filterSize;
00504         }
00505     }
00506 
00507     // Note the +1 is for the MMX scaler which reads over the end
00508     /* align at 16 for AltiVec (needed by hScale_altivec_real) */
00509     FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
00510 
00511     /* normalize & store in outFilter */
00512     for (i=0; i<dstW; i++) {
00513         int j;
00514         int64_t error=0;
00515         int64_t sum=0;
00516 
00517         for (j=0; j<filterSize; j++) {
00518             sum+= filter[i*filterSize + j];
00519         }
00520         sum= (sum + one/2)/ one;
00521         for (j=0; j<*outFilterSize; j++) {
00522             int64_t v= filter[i*filterSize + j] + error;
00523             int intV= ROUNDED_DIV(v, sum);
00524             (*outFilter)[i*(*outFilterSize) + j]= intV;
00525             error= v - intV*sum;
00526         }
00527     }
00528 
00529     (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
00530     for (i=0; i<*outFilterSize; i++) {
00531         int j= dstW*(*outFilterSize);
00532         (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
00533     }
00534 
00535     ret=0;
00536 fail:
00537     av_free(filter);
00538     av_free(filter2);
00539     return ret;
00540 }
00541 
00542 #if HAVE_MMX2
00543 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
00544 {
00545     uint8_t *fragmentA;
00546     x86_reg imm8OfPShufW1A;
00547     x86_reg imm8OfPShufW2A;
00548     x86_reg fragmentLengthA;
00549     uint8_t *fragmentB;
00550     x86_reg imm8OfPShufW1B;
00551     x86_reg imm8OfPShufW2B;
00552     x86_reg fragmentLengthB;
00553     int fragmentPos;
00554 
00555     int xpos, i;
00556 
00557     // create an optimized horizontal scaling routine
00558     /* This scaler is made of runtime-generated MMX2 code using specially
00559      * tuned pshufw instructions. For every four output pixels, if four
00560      * input pixels are enough for the fast bilinear scaling, then a chunk
00561      * of fragmentB is used. If five input pixels are needed, then a chunk
00562      * of fragmentA is used.
00563      */
00564 
00565     //code fragment
00566 
00567     __asm__ volatile(
00568         "jmp                         9f                 \n\t"
00569     // Begin
00570         "0:                                             \n\t"
00571         "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
00572         "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
00573         "movd   1(%%"REG_c", %%"REG_S"), %%mm1          \n\t"
00574         "punpcklbw                %%mm7, %%mm1          \n\t"
00575         "punpcklbw                %%mm7, %%mm0          \n\t"
00576         "pshufw                   $0xFF, %%mm1, %%mm1   \n\t"
00577         "1:                                             \n\t"
00578         "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
00579         "2:                                             \n\t"
00580         "psubw                    %%mm1, %%mm0          \n\t"
00581         "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
00582         "pmullw                   %%mm3, %%mm0          \n\t"
00583         "psllw                       $7, %%mm1          \n\t"
00584         "paddw                    %%mm1, %%mm0          \n\t"
00585 
00586         "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"
00587 
00588         "add                         $8, %%"REG_a"      \n\t"
00589     // End
00590         "9:                                             \n\t"
00591 //        "int $3                                         \n\t"
00592         "lea                 " LOCAL_MANGLE(0b) ", %0   \n\t"
00593         "lea                 " LOCAL_MANGLE(1b) ", %1   \n\t"
00594         "lea                 " LOCAL_MANGLE(2b) ", %2   \n\t"
00595         "dec                         %1                 \n\t"
00596         "dec                         %2                 \n\t"
00597         "sub                         %0, %1             \n\t"
00598         "sub                         %0, %2             \n\t"
00599         "lea                 " LOCAL_MANGLE(9b) ", %3   \n\t"
00600         "sub                         %0, %3             \n\t"
00601 
00602 
00603         :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
00604         "=r" (fragmentLengthA)
00605     );
00606 
00607     __asm__ volatile(
00608         "jmp                         9f                 \n\t"
00609     // Begin
00610         "0:                                             \n\t"
00611         "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
00612         "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
00613         "punpcklbw                %%mm7, %%mm0          \n\t"
00614         "pshufw                   $0xFF, %%mm0, %%mm1   \n\t"
00615         "1:                                             \n\t"
00616         "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
00617         "2:                                             \n\t"
00618         "psubw                    %%mm1, %%mm0          \n\t"
00619         "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
00620         "pmullw                   %%mm3, %%mm0          \n\t"
00621         "psllw                       $7, %%mm1          \n\t"
00622         "paddw                    %%mm1, %%mm0          \n\t"
00623 
00624         "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"
00625 
00626         "add                         $8, %%"REG_a"      \n\t"
00627     // End
00628         "9:                                             \n\t"
00629 //        "int                       $3                   \n\t"
00630         "lea                 " LOCAL_MANGLE(0b) ", %0   \n\t"
00631         "lea                 " LOCAL_MANGLE(1b) ", %1   \n\t"
00632         "lea                 " LOCAL_MANGLE(2b) ", %2   \n\t"
00633         "dec                         %1                 \n\t"
00634         "dec                         %2                 \n\t"
00635         "sub                         %0, %1             \n\t"
00636         "sub                         %0, %2             \n\t"
00637         "lea                 " LOCAL_MANGLE(9b) ", %3   \n\t"
00638         "sub                         %0, %3             \n\t"
00639 
00640 
00641         :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
00642         "=r" (fragmentLengthB)
00643     );
00644 
00645     xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
00646     fragmentPos=0;
00647 
00648     for (i=0; i<dstW/numSplits; i++) {
00649         int xx=xpos>>16;
00650 
00651         if ((i&3) == 0) {
00652             int a=0;
00653             int b=((xpos+xInc)>>16) - xx;
00654             int c=((xpos+xInc*2)>>16) - xx;
00655             int d=((xpos+xInc*3)>>16) - xx;
00656             int inc                = (d+1<4);
00657             uint8_t *fragment      = (d+1<4) ? fragmentB       : fragmentA;
00658             x86_reg imm8OfPShufW1  = (d+1<4) ? imm8OfPShufW1B  : imm8OfPShufW1A;
00659             x86_reg imm8OfPShufW2  = (d+1<4) ? imm8OfPShufW2B  : imm8OfPShufW2A;
00660             x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
00661             int maxShift= 3-(d+inc);
00662             int shift=0;
00663 
00664             if (filterCode) {
00665                 filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
00666                 filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
00667                 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
00668                 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
00669                 filterPos[i/2]= xx;
00670 
00671                 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
00672 
00673                 filterCode[fragmentPos + imm8OfPShufW1]=
00674                     (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
00675                 filterCode[fragmentPos + imm8OfPShufW2]=
00676                     a | (b<<2) | (c<<4) | (d<<6);
00677 
00678                 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
00679                 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
00680 
00681                 if (shift && i>=shift) {
00682                     filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
00683                     filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
00684                     filterPos[i/2]-=shift;
00685                 }
00686             }
00687 
00688             fragmentPos+= fragmentLength;
00689 
00690             if (filterCode)
00691                 filterCode[fragmentPos]= RET;
00692         }
00693         xpos+=xInc;
00694     }
00695     if (filterCode)
00696         filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
00697 
00698     return fragmentPos + 1;
00699 }
00700 #endif /* HAVE_MMX2 */
00701 
00702 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
00703 {
00704     *h = av_pix_fmt_descriptors[format].log2_chroma_w;
00705     *v = av_pix_fmt_descriptors[format].log2_chroma_h;
00706 }
00707 
00708 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)
00709 {
00710     memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
00711     memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
00712 
00713     c->brightness= brightness;
00714     c->contrast  = contrast;
00715     c->saturation= saturation;
00716     c->srcRange  = srcRange;
00717     c->dstRange  = dstRange;
00718     if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
00719 
00720     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
00721     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
00722 
00723     ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
00724     //FIXME factorize
00725 
00726     if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
00727         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
00728     return 0;
00729 }
00730 
00731 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)
00732 {
00733     if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
00734 
00735     *inv_table = c->srcColorspaceTable;
00736     *table     = c->dstColorspaceTable;
00737     *srcRange  = c->srcRange;
00738     *dstRange  = c->dstRange;
00739     *brightness= c->brightness;
00740     *contrast  = c->contrast;
00741     *saturation= c->saturation;
00742 
00743     return 0;
00744 }
00745 
00746 static int handle_jpeg(enum PixelFormat *format)
00747 {
00748     switch (*format) {
00749     case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
00750     case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
00751     case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
00752     case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
00753     default:                                          return 0;
00754     }
00755 }
00756 
00757 SwsContext *sws_alloc_context(void)
00758 {
00759     SwsContext *c= av_mallocz(sizeof(SwsContext));
00760 
00761     c->av_class = &sws_context_class;
00762     av_opt_set_defaults(c);
00763 
00764     return c;
00765 }
00766 
00767 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
00768 {
00769     int i;
00770     int usesVFilter, usesHFilter;
00771     int unscaled;
00772     SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
00773     int srcW= c->srcW;
00774     int srcH= c->srcH;
00775     int dstW= c->dstW;
00776     int dstH= c->dstH;
00777     int dst_stride = FFALIGN(dstW * sizeof(int16_t) + 16, 16), dst_stride_px = dst_stride >> 1;
00778     int flags, cpu_flags;
00779     enum PixelFormat srcFormat= c->srcFormat;
00780     enum PixelFormat dstFormat= c->dstFormat;
00781 
00782     cpu_flags = av_get_cpu_flags();
00783     flags     = c->flags;
00784     emms_c();
00785     if (!rgb15to16) sws_rgb2rgb_init();
00786 
00787     unscaled = (srcW == dstW && srcH == dstH);
00788 
00789     if (!isSupportedIn(srcFormat)) {
00790         av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n", sws_format_name(srcFormat));
00791         return AVERROR(EINVAL);
00792     }
00793     if (!isSupportedOut(dstFormat)) {
00794         av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n", sws_format_name(dstFormat));
00795         return AVERROR(EINVAL);
00796     }
00797 
00798     i= flags & ( SWS_POINT
00799                 |SWS_AREA
00800                 |SWS_BILINEAR
00801                 |SWS_FAST_BILINEAR
00802                 |SWS_BICUBIC
00803                 |SWS_X
00804                 |SWS_GAUSS
00805                 |SWS_LANCZOS
00806                 |SWS_SINC
00807                 |SWS_SPLINE
00808                 |SWS_BICUBLIN);
00809     if(!i || (i & (i-1))) {
00810         av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen\n");
00811         return AVERROR(EINVAL);
00812     }
00813     /* sanity check */
00814     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
00815         av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
00816                srcW, srcH, dstW, dstH);
00817         return AVERROR(EINVAL);
00818     }
00819 
00820     if (!dstFilter) dstFilter= &dummyFilter;
00821     if (!srcFilter) srcFilter= &dummyFilter;
00822 
00823     c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
00824     c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
00825     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
00826     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
00827     c->vRounder= 4* 0x0001000100010001ULL;
00828 
00829     usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
00830                   (srcFilter->chrV && srcFilter->chrV->length>1) ||
00831                   (dstFilter->lumV && dstFilter->lumV->length>1) ||
00832                   (dstFilter->chrV && dstFilter->chrV->length>1);
00833     usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
00834                   (srcFilter->chrH && srcFilter->chrH->length>1) ||
00835                   (dstFilter->lumH && dstFilter->lumH->length>1) ||
00836                   (dstFilter->chrH && dstFilter->chrH->length>1);
00837 
00838     getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
00839     getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
00840 
00841     // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
00842     if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
00843 
00844     // drop some chroma lines if the user wants it
00845     c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
00846     c->chrSrcVSubSample+= c->vChrDrop;
00847 
00848     // drop every other pixel for chroma calculation unless user wants full chroma
00849     if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
00850       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8
00851       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4
00852       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
00853       && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
00854         c->chrSrcHSubSample=1;
00855 
00856     // Note the -((-x)>>y) is so that we always round toward +inf.
00857     c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
00858     c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
00859     c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
00860     c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
00861 
00862     /* unscaled special cases */
00863     if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
00864         ff_get_unscaled_swscale(c);
00865 
00866         if (c->swScale) {
00867             if (flags&SWS_PRINT_INFO)
00868                 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
00869                        sws_format_name(srcFormat), sws_format_name(dstFormat));
00870             return 0;
00871         }
00872     }
00873 
00874     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW, 16) * 2, fail);
00875     if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) {
00876         c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
00877         if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
00878             if (flags&SWS_PRINT_INFO)
00879                 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
00880         }
00881         if (usesHFilter) c->canMMX2BeUsed=0;
00882     }
00883     else
00884         c->canMMX2BeUsed=0;
00885 
00886     c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
00887     c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
00888 
00889     // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
00890     // but only for the FAST_BILINEAR mode otherwise do correct scaling
00891     // n-2 is the last chrominance sample available
00892     // this is not perfect, but no one should notice the difference, the more correct variant
00893     // would be like the vertical one, but that would require some special code for the
00894     // first and last pixel
00895     if (flags&SWS_FAST_BILINEAR) {
00896         if (c->canMMX2BeUsed) {
00897             c->lumXInc+= 20;
00898             c->chrXInc+= 20;
00899         }
00900         //we don't use the x86 asm scaler if MMX is available
00901         else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
00902             c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
00903             c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
00904         }
00905     }
00906 
00907     /* precalculate horizontal scaler filter coefficients */
00908     {
00909 #if HAVE_MMX2
00910 // can't downscale !!!
00911         if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
00912             c->lumMmx2FilterCodeSize = initMMX2HScaler(      dstW, c->lumXInc, NULL, NULL, NULL, 8);
00913             c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
00914 
00915 #ifdef MAP_ANONYMOUS
00916             c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
00917             c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
00918 #elif HAVE_VIRTUALALLOC
00919             c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
00920             c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
00921 #else
00922             c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
00923             c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
00924 #endif
00925 
00926             if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
00927                 return AVERROR(ENOMEM);
00928             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter   , (dstW        /8+8)*sizeof(int16_t), fail);
00929             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter   , (c->chrDstW  /4+8)*sizeof(int16_t), fail);
00930             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW      /2/8+8)*sizeof(int32_t), fail);
00931             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
00932 
00933             initMMX2HScaler(      dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
00934             initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
00935 
00936 #ifdef MAP_ANONYMOUS
00937             mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
00938             mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
00939 #endif
00940         } else
00941 #endif /* HAVE_MMX2 */
00942         {
00943             const int filterAlign=
00944                 (HAVE_MMX     && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
00945                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
00946                 1;
00947 
00948             if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
00949                            srcW      ,       dstW, filterAlign, 1<<14,
00950                            (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags, cpu_flags,
00951                            srcFilter->lumH, dstFilter->lumH, c->param) < 0)
00952                 goto fail;
00953             if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
00954                            c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
00955                            (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
00956                            srcFilter->chrH, dstFilter->chrH, c->param) < 0)
00957                 goto fail;
00958         }
00959     } // initialize horizontal stuff
00960 
00961     /* precalculate vertical scaler filter coefficients */
00962     {
00963         const int filterAlign=
00964             (HAVE_MMX     && cpu_flags & AV_CPU_FLAG_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
00965             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
00966             1;
00967 
00968         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
00969                        srcH      ,        dstH, filterAlign, (1<<12),
00970                        (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags, cpu_flags,
00971                        srcFilter->lumV, dstFilter->lumV, c->param) < 0)
00972             goto fail;
00973         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
00974                        c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
00975                        (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
00976                        srcFilter->chrV, dstFilter->chrV, c->param) < 0)
00977             goto fail;
00978 
00979 #if HAVE_ALTIVEC
00980         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
00981         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
00982 
00983         for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
00984             int j;
00985             short *p = (short *)&c->vYCoeffsBank[i];
00986             for (j=0;j<8;j++)
00987                 p[j] = c->vLumFilter[i];
00988         }
00989 
00990         for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
00991             int j;
00992             short *p = (short *)&c->vCCoeffsBank[i];
00993             for (j=0;j<8;j++)
00994                 p[j] = c->vChrFilter[i];
00995         }
00996 #endif
00997     }
00998 
00999     // calculate buffer sizes so that they won't run out while handling these damn slices
01000     c->vLumBufSize= c->vLumFilterSize;
01001     c->vChrBufSize= c->vChrFilterSize;
01002     for (i=0; i<dstH; i++) {
01003         int chrI= i*c->chrDstH / dstH;
01004         int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
01005                            ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
01006 
01007         nextSlice>>= c->chrSrcVSubSample;
01008         nextSlice<<= c->chrSrcVSubSample;
01009         if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
01010             c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
01011         if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
01012             c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
01013     }
01014 
01015     // allocate pixbufs (we use dynamic allocation because otherwise we would need to
01016     // allocate several megabytes to handle all possible cases)
01017     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
01018     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
01019     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
01020     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
01021         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
01022     //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)
01023     /* align at 16 bytes for AltiVec */
01024     for (i=0; i<c->vLumBufSize; i++) {
01025         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+1, fail);
01026         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
01027     }
01028     c->uv_off = dst_stride_px;
01029     c->uv_offx2 = dst_stride;
01030     for (i=0; i<c->vChrBufSize; i++) {
01031         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+1, fail);
01032         c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
01033         c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + dst_stride_px;
01034     }
01035     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
01036         for (i=0; i<c->vLumBufSize; i++) {
01037             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+1, fail);
01038             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
01039         }
01040 
01041     //try to avoid drawing green stuff between the right end and the stride end
01042     for (i=0; i<c->vChrBufSize; i++)
01043         memset(c->chrUPixBuf[i], 64, dst_stride*2+1);
01044 
01045     assert(c->chrDstH <= dstH);
01046 
01047     if (flags&SWS_PRINT_INFO) {
01048         if      (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
01049         else if (flags&SWS_BILINEAR)      av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
01050         else if (flags&SWS_BICUBIC)       av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
01051         else if (flags&SWS_X)             av_log(c, AV_LOG_INFO, "Experimental scaler, ");
01052         else if (flags&SWS_POINT)         av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
01053         else if (flags&SWS_AREA)          av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
01054         else if (flags&SWS_BICUBLIN)      av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
01055         else if (flags&SWS_GAUSS)         av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
01056         else if (flags&SWS_SINC)          av_log(c, AV_LOG_INFO, "Sinc scaler, ");
01057         else if (flags&SWS_LANCZOS)       av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
01058         else if (flags&SWS_SPLINE)        av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
01059         else                              av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
01060 
01061         av_log(c, AV_LOG_INFO, "from %s to %s%s ",
01062                sws_format_name(srcFormat),
01063 #ifdef DITHER1XBPP
01064                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
01065                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
01066                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
01067 #else
01068                "",
01069 #endif
01070                sws_format_name(dstFormat));
01071 
01072         if      (HAVE_MMX2     && cpu_flags & AV_CPU_FLAG_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");
01073         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");
01074         else if (HAVE_MMX      && cpu_flags & AV_CPU_FLAG_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");
01075         else if (HAVE_ALTIVEC  && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
01076         else                                   av_log(c, AV_LOG_INFO, "using C\n");
01077 
01078         if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
01079             if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
01080                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
01081             else {
01082                 if (c->hLumFilterSize==4)
01083                     av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
01084                 else if (c->hLumFilterSize==8)
01085                     av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
01086                 else
01087                     av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
01088 
01089                 if (c->hChrFilterSize==4)
01090                     av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
01091                 else if (c->hChrFilterSize==8)
01092                     av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
01093                 else
01094                     av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
01095             }
01096         } else {
01097 #if HAVE_MMX
01098             av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
01099 #else
01100             if (flags & SWS_FAST_BILINEAR)
01101                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
01102             else
01103                 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
01104 #endif
01105         }
01106         if (isPlanarYUV(dstFormat)) {
01107             if (c->vLumFilterSize==1)
01108                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n",
01109                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
01110             else
01111                 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n",
01112                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
01113         } else {
01114             if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
01115                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
01116                        "      2-tap scaler for vertical chrominance scaling (BGR)\n",
01117                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
01118             else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
01119                 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n",
01120                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
01121             else
01122                 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n",
01123                        (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
01124         }
01125 
01126         if (dstFormat==PIX_FMT_BGR24)
01127             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
01128                    (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) ? "MMX2" :
01129                    ((HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"));
01130         else if (dstFormat==PIX_FMT_RGB32)
01131             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n",
01132                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
01133         else if (dstFormat==PIX_FMT_BGR565)
01134             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n",
01135                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
01136         else if (dstFormat==PIX_FMT_BGR555)
01137             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n",
01138                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
01139         else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
01140                  dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)
01141             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n",
01142                    (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
01143 
01144         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
01145         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
01146                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
01147         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
01148                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
01149     }
01150 
01151     c->swScale= ff_getSwsFunc(c);
01152     return 0;
01153 fail: //FIXME replace things by appropriate error codes
01154     return -1;
01155 }
01156 
01157 #if FF_API_SWS_GETCONTEXT
01158 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
01159                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,
01160                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
01161 {
01162     SwsContext *c;
01163 
01164     if(!(c=sws_alloc_context()))
01165         return NULL;
01166 
01167     c->flags= flags;
01168     c->srcW= srcW;
01169     c->srcH= srcH;
01170     c->dstW= dstW;
01171     c->dstH= dstH;
01172     c->srcRange = handle_jpeg(&srcFormat);
01173     c->dstRange = handle_jpeg(&dstFormat);
01174     c->srcFormat= srcFormat;
01175     c->dstFormat= dstFormat;
01176 
01177     if (param) {
01178         c->param[0] = param[0];
01179         c->param[1] = param[1];
01180     }
01181     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);
01182 
01183     if(sws_init_context(c, srcFilter, dstFilter) < 0){
01184         sws_freeContext(c);
01185         return NULL;
01186     }
01187 
01188     return c;
01189 }
01190 #endif
01191 
01192 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
01193                                 float lumaSharpen, float chromaSharpen,
01194                                 float chromaHShift, float chromaVShift,
01195                                 int verbose)
01196 {
01197     SwsFilter *filter= av_malloc(sizeof(SwsFilter));
01198     if (!filter)
01199         return NULL;
01200 
01201     if (lumaGBlur!=0.0) {
01202         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
01203         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
01204     } else {
01205         filter->lumH= sws_getIdentityVec();
01206         filter->lumV= sws_getIdentityVec();
01207     }
01208 
01209     if (chromaGBlur!=0.0) {
01210         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
01211         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
01212     } else {
01213         filter->chrH= sws_getIdentityVec();
01214         filter->chrV= sws_getIdentityVec();
01215     }
01216 
01217     if (chromaSharpen!=0.0) {
01218         SwsVector *id= sws_getIdentityVec();
01219         sws_scaleVec(filter->chrH, -chromaSharpen);
01220         sws_scaleVec(filter->chrV, -chromaSharpen);
01221         sws_addVec(filter->chrH, id);
01222         sws_addVec(filter->chrV, id);
01223         sws_freeVec(id);
01224     }
01225 
01226     if (lumaSharpen!=0.0) {
01227         SwsVector *id= sws_getIdentityVec();
01228         sws_scaleVec(filter->lumH, -lumaSharpen);
01229         sws_scaleVec(filter->lumV, -lumaSharpen);
01230         sws_addVec(filter->lumH, id);
01231         sws_addVec(filter->lumV, id);
01232         sws_freeVec(id);
01233     }
01234 
01235     if (chromaHShift != 0.0)
01236         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
01237 
01238     if (chromaVShift != 0.0)
01239         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
01240 
01241     sws_normalizeVec(filter->chrH, 1.0);
01242     sws_normalizeVec(filter->chrV, 1.0);
01243     sws_normalizeVec(filter->lumH, 1.0);
01244     sws_normalizeVec(filter->lumV, 1.0);
01245 
01246     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
01247     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
01248 
01249     return filter;
01250 }
01251 
01252 SwsVector *sws_allocVec(int length)
01253 {
01254     SwsVector *vec = av_malloc(sizeof(SwsVector));
01255     if (!vec)
01256         return NULL;
01257     vec->length = length;
01258     vec->coeff  = av_malloc(sizeof(double) * length);
01259     if (!vec->coeff)
01260         av_freep(&vec);
01261     return vec;
01262 }
01263 
01264 SwsVector *sws_getGaussianVec(double variance, double quality)
01265 {
01266     const int length= (int)(variance*quality + 0.5) | 1;
01267     int i;
01268     double middle= (length-1)*0.5;
01269     SwsVector *vec= sws_allocVec(length);
01270 
01271     if (!vec)
01272         return NULL;
01273 
01274     for (i=0; i<length; i++) {
01275         double dist= i-middle;
01276         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
01277     }
01278 
01279     sws_normalizeVec(vec, 1.0);
01280 
01281     return vec;
01282 }
01283 
01284 SwsVector *sws_getConstVec(double c, int length)
01285 {
01286     int i;
01287     SwsVector *vec= sws_allocVec(length);
01288 
01289     if (!vec)
01290         return NULL;
01291 
01292     for (i=0; i<length; i++)
01293         vec->coeff[i]= c;
01294 
01295     return vec;
01296 }
01297 
01298 SwsVector *sws_getIdentityVec(void)
01299 {
01300     return sws_getConstVec(1.0, 1);
01301 }
01302 
01303 static double sws_dcVec(SwsVector *a)
01304 {
01305     int i;
01306     double sum=0;
01307 
01308     for (i=0; i<a->length; i++)
01309         sum+= a->coeff[i];
01310 
01311     return sum;
01312 }
01313 
01314 void sws_scaleVec(SwsVector *a, double scalar)
01315 {
01316     int i;
01317 
01318     for (i=0; i<a->length; i++)
01319         a->coeff[i]*= scalar;
01320 }
01321 
01322 void sws_normalizeVec(SwsVector *a, double height)
01323 {
01324     sws_scaleVec(a, height/sws_dcVec(a));
01325 }
01326 
01327 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
01328 {
01329     int length= a->length + b->length - 1;
01330     int i, j;
01331     SwsVector *vec= sws_getConstVec(0.0, length);
01332 
01333     if (!vec)
01334         return NULL;
01335 
01336     for (i=0; i<a->length; i++) {
01337         for (j=0; j<b->length; j++) {
01338             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
01339         }
01340     }
01341 
01342     return vec;
01343 }
01344 
01345 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
01346 {
01347     int length= FFMAX(a->length, b->length);
01348     int i;
01349     SwsVector *vec= sws_getConstVec(0.0, length);
01350 
01351     if (!vec)
01352         return NULL;
01353 
01354     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
01355     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
01356 
01357     return vec;
01358 }
01359 
01360 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
01361 {
01362     int length= FFMAX(a->length, b->length);
01363     int i;
01364     SwsVector *vec= sws_getConstVec(0.0, length);
01365 
01366     if (!vec)
01367         return NULL;
01368 
01369     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
01370     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
01371 
01372     return vec;
01373 }
01374 
01375 /* shift left / or right if "shift" is negative */
01376 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
01377 {
01378     int length= a->length + FFABS(shift)*2;
01379     int i;
01380     SwsVector *vec= sws_getConstVec(0.0, length);
01381 
01382     if (!vec)
01383         return NULL;
01384 
01385     for (i=0; i<a->length; i++) {
01386         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
01387     }
01388 
01389     return vec;
01390 }
01391 
01392 void sws_shiftVec(SwsVector *a, int shift)
01393 {
01394     SwsVector *shifted= sws_getShiftedVec(a, shift);
01395     av_free(a->coeff);
01396     a->coeff= shifted->coeff;
01397     a->length= shifted->length;
01398     av_free(shifted);
01399 }
01400 
01401 void sws_addVec(SwsVector *a, SwsVector *b)
01402 {
01403     SwsVector *sum= sws_sumVec(a, b);
01404     av_free(a->coeff);
01405     a->coeff= sum->coeff;
01406     a->length= sum->length;
01407     av_free(sum);
01408 }
01409 
01410 void sws_subVec(SwsVector *a, SwsVector *b)
01411 {
01412     SwsVector *diff= sws_diffVec(a, b);
01413     av_free(a->coeff);
01414     a->coeff= diff->coeff;
01415     a->length= diff->length;
01416     av_free(diff);
01417 }
01418 
01419 void sws_convVec(SwsVector *a, SwsVector *b)
01420 {
01421     SwsVector *conv= sws_getConvVec(a, b);
01422     av_free(a->coeff);
01423     a->coeff= conv->coeff;
01424     a->length= conv->length;
01425     av_free(conv);
01426 }
01427 
01428 SwsVector *sws_cloneVec(SwsVector *a)
01429 {
01430     int i;
01431     SwsVector *vec= sws_allocVec(a->length);
01432 
01433     if (!vec)
01434         return NULL;
01435 
01436     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
01437 
01438     return vec;
01439 }
01440 
01441 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
01442 {
01443     int i;
01444     double max=0;
01445     double min=0;
01446     double range;
01447 
01448     for (i=0; i<a->length; i++)
01449         if (a->coeff[i]>max) max= a->coeff[i];
01450 
01451     for (i=0; i<a->length; i++)
01452         if (a->coeff[i]<min) min= a->coeff[i];
01453 
01454     range= max - min;
01455 
01456     for (i=0; i<a->length; i++) {
01457         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
01458         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
01459         for (;x>0; x--) av_log(log_ctx, log_level, " ");
01460         av_log(log_ctx, log_level, "|\n");
01461     }
01462 }
01463 
01464 void sws_freeVec(SwsVector *a)
01465 {
01466     if (!a) return;
01467     av_freep(&a->coeff);
01468     a->length=0;
01469     av_free(a);
01470 }
01471 
01472 void sws_freeFilter(SwsFilter *filter)
01473 {
01474     if (!filter) return;
01475 
01476     if (filter->lumH) sws_freeVec(filter->lumH);
01477     if (filter->lumV) sws_freeVec(filter->lumV);
01478     if (filter->chrH) sws_freeVec(filter->chrH);
01479     if (filter->chrV) sws_freeVec(filter->chrV);
01480     av_free(filter);
01481 }
01482 
01483 void sws_freeContext(SwsContext *c)
01484 {
01485     int i;
01486     if (!c) return;
01487 
01488     if (c->lumPixBuf) {
01489         for (i=0; i<c->vLumBufSize; i++)
01490             av_freep(&c->lumPixBuf[i]);
01491         av_freep(&c->lumPixBuf);
01492     }
01493 
01494     if (c->chrUPixBuf) {
01495         for (i=0; i<c->vChrBufSize; i++)
01496             av_freep(&c->chrUPixBuf[i]);
01497         av_freep(&c->chrUPixBuf);
01498         av_freep(&c->chrVPixBuf);
01499     }
01500 
01501     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
01502         for (i=0; i<c->vLumBufSize; i++)
01503             av_freep(&c->alpPixBuf[i]);
01504         av_freep(&c->alpPixBuf);
01505     }
01506 
01507     av_freep(&c->vLumFilter);
01508     av_freep(&c->vChrFilter);
01509     av_freep(&c->hLumFilter);
01510     av_freep(&c->hChrFilter);
01511 #if HAVE_ALTIVEC
01512     av_freep(&c->vYCoeffsBank);
01513     av_freep(&c->vCCoeffsBank);
01514 #endif
01515 
01516     av_freep(&c->vLumFilterPos);
01517     av_freep(&c->vChrFilterPos);
01518     av_freep(&c->hLumFilterPos);
01519     av_freep(&c->hChrFilterPos);
01520 
01521 #if HAVE_MMX
01522 #ifdef MAP_ANONYMOUS
01523     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
01524     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
01525 #elif HAVE_VIRTUALALLOC
01526     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
01527     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
01528 #else
01529     av_free(c->lumMmx2FilterCode);
01530     av_free(c->chrMmx2FilterCode);
01531 #endif
01532     c->lumMmx2FilterCode=NULL;
01533     c->chrMmx2FilterCode=NULL;
01534 #endif /* HAVE_MMX */
01535 
01536     av_freep(&c->yuvTable);
01537     av_free(c->formatConvBuffer);
01538 
01539     av_free(c);
01540 }
01541 
01542 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
01543                                         int srcW, int srcH, enum PixelFormat srcFormat,
01544                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,
01545                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
01546 {
01547     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
01548 
01549     if (!param)
01550         param = default_param;
01551 
01552     if (context &&
01553         (context->srcW      != srcW      ||
01554          context->srcH      != srcH      ||
01555          context->srcFormat != srcFormat ||
01556          context->dstW      != dstW      ||
01557          context->dstH      != dstH      ||
01558          context->dstFormat != dstFormat ||
01559          context->flags     != flags     ||
01560          context->param[0]  != param[0]  ||
01561          context->param[1]  != param[1])) {
01562         sws_freeContext(context);
01563         context = NULL;
01564     }
01565 
01566     if (!context) {
01567         if (!(context = sws_alloc_context()))
01568             return NULL;
01569         context->srcW      = srcW;
01570         context->srcH      = srcH;
01571         context->srcRange  = handle_jpeg(&srcFormat);
01572         context->srcFormat = srcFormat;
01573         context->dstW      = dstW;
01574         context->dstH      = dstH;
01575         context->dstRange  = handle_jpeg(&dstFormat);
01576         context->dstFormat = dstFormat;
01577         context->flags     = flags;
01578         context->param[0]  = param[0];
01579         context->param[1]  = param[1];
01580         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);
01581         if (sws_init_context(context, srcFilter, dstFilter) < 0) {
01582             sws_freeContext(context);
01583             return NULL;
01584         }
01585     }
01586     return context;
01587 }
01588