FFmpeg: libswscale/utils.c Source File

00001 /*
00002  * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
00003  *
00004  * This file is part of FFmpeg.
00005  *
00006  * FFmpeg 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  * FFmpeg 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 FFmpeg; if not, write to the Free Software
00018  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00019  */
00020 
00021 #define _SVID_SOURCE //needed for MAP_ANONYMOUS
00022 #define _DARWIN_C_SOURCE // needed for MAP_ANON
00023 #include <inttypes.h>
00024 #include <string.h>
00025 #include <math.h>
00026 #include <stdio.h>
00027 #include "config.h"
00028 #include <assert.h>
00029 #if HAVE_SYS_MMAN_H
00030 #include <sys/mman.h>
00031 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
00032 #define MAP_ANONYMOUS MAP_ANON
00033 #endif
00034 #endif
00035 #if HAVE_VIRTUALALLOC
00036 #define WIN32_LEAN_AND_MEAN
00037 #include <windows.h>
00038 #endif
00039 #include "swscale.h"
00040 #include "swscale_internal.h"
00041 #include "rgb2rgb.h"
00042 #include "libavutil/intreadwrite.h"
00043 #include "libavutil/x86_cpu.h"
00044 #include "libavutil/cpu.h"
00045 #include "libavutil/avutil.h"
00046 #include "libavutil/bswap.h"
00047 #include "libavutil/mathematics.h"
00048 #include "libavutil/opt.h"
00049 #include "libavutil/pixdesc.h"
00050 #include "libavutil/avassert.h"
00051 
00052 unsigned swscale_version(void)
00053 {
00054     return LIBSWSCALE_VERSION_INT;
00055 }
00056 
00057 const char *swscale_configuration(void)
00058 {
00059     return FFMPEG_CONFIGURATION;
00060 }
00061 
00062 const char *swscale_license(void)
00063 {
00064 #define LICENSE_PREFIX "libswscale license: "
00065     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
00066 }
00067 
00068 #define RET 0xC3 //near return opcode for x86
00069 
00070 typedef struct FormatEntry {
00071     int is_supported_in, is_supported_out;
00072 } FormatEntry;
00073 
00074 const static FormatEntry format_entries[PIX_FMT_NB] = {
00075     [PIX_FMT_YUV420P]     = { 1 , 1 },
00076     [PIX_FMT_YUYV422]     = { 1 , 1 },
00077     [PIX_FMT_RGB24]       = { 1 , 1 },
00078     [PIX_FMT_BGR24]       = { 1 , 1 },
00079     [PIX_FMT_YUV422P]     = { 1 , 1 },
00080     [PIX_FMT_YUV444P]     = { 1 , 1 },
00081     [PIX_FMT_YUV410P]     = { 1 , 1 },
00082     [PIX_FMT_YUV411P]     = { 1 , 1 },
00083     [PIX_FMT_GRAY8]       = { 1 , 1 },
00084     [PIX_FMT_MONOWHITE]   = { 1 , 1 },
00085     [PIX_FMT_MONOBLACK]   = { 1 , 1 },
00086     [PIX_FMT_PAL8]        = { 1 , 0 },
00087     [PIX_FMT_YUVJ420P]    = { 1 , 1 },
00088     [PIX_FMT_YUVJ422P]    = { 1 , 1 },
00089     [PIX_FMT_YUVJ444P]    = { 1 , 1 },
00090     [PIX_FMT_UYVY422]     = { 1 , 1 },
00091     [PIX_FMT_UYYVYY411]   = { 0 , 0 },
00092     [PIX_FMT_BGR8]        = { 1 , 1 },
00093     [PIX_FMT_BGR4]        = { 0 , 1 },
00094     [PIX_FMT_BGR4_BYTE]   = { 1 , 1 },
00095     [PIX_FMT_RGB8]        = { 1 , 1 },
00096     [PIX_FMT_RGB4]        = { 0 , 1 },
00097     [PIX_FMT_RGB4_BYTE]   = { 1 , 1 },
00098     [PIX_FMT_NV12]        = { 1 , 1 },
00099     [PIX_FMT_NV21]        = { 1 , 1 },
00100     [PIX_FMT_ARGB]        = { 1 , 1 },
00101     [PIX_FMT_RGBA]        = { 1 , 1 },
00102     [PIX_FMT_ABGR]        = { 1 , 1 },
00103     [PIX_FMT_BGRA]        = { 1 , 1 },
00104     [PIX_FMT_0RGB]        = { 1 , 1 },
00105     [PIX_FMT_RGB0]        = { 1 , 1 },
00106     [PIX_FMT_0BGR]        = { 1 , 1 },
00107     [PIX_FMT_BGR0]        = { 1 , 1 },
00108     [PIX_FMT_GRAY16BE]    = { 1 , 1 },
00109     [PIX_FMT_GRAY16LE]    = { 1 , 1 },
00110     [PIX_FMT_YUV440P]     = { 1 , 1 },
00111     [PIX_FMT_YUVJ440P]    = { 1 , 1 },
00112     [PIX_FMT_YUVA420P]    = { 1 , 1 },
00113     [PIX_FMT_RGB48BE]     = { 1 , 1 },
00114     [PIX_FMT_RGB48LE]     = { 1 , 1 },
00115     [PIX_FMT_RGBA64BE]    = { 0 , 0 },
00116     [PIX_FMT_RGBA64LE]    = { 0 , 0 },
00117     [PIX_FMT_RGB565BE]    = { 1 , 1 },
00118     [PIX_FMT_RGB565LE]    = { 1 , 1 },
00119     [PIX_FMT_RGB555BE]    = { 1 , 1 },
00120     [PIX_FMT_RGB555LE]    = { 1 , 1 },
00121     [PIX_FMT_BGR565BE]    = { 1 , 1 },
00122     [PIX_FMT_BGR565LE]    = { 1 , 1 },
00123     [PIX_FMT_BGR555BE]    = { 1 , 1 },
00124     [PIX_FMT_BGR555LE]    = { 1 , 1 },
00125     [PIX_FMT_YUV420P16LE] = { 1 , 1 },
00126     [PIX_FMT_YUV420P16BE] = { 1 , 1 },
00127     [PIX_FMT_YUV422P16LE] = { 1 , 1 },
00128     [PIX_FMT_YUV422P16BE] = { 1 , 1 },
00129     [PIX_FMT_YUV444P16LE] = { 1 , 1 },
00130     [PIX_FMT_YUV444P16BE] = { 1 , 1 },
00131     [PIX_FMT_RGB444LE]    = { 0 , 1 },
00132     [PIX_FMT_RGB444BE]    = { 0 , 1 },
00133     [PIX_FMT_BGR444LE]    = { 0 , 1 },
00134     [PIX_FMT_BGR444BE]    = { 0 , 1 },
00135     [PIX_FMT_Y400A]       = { 1 , 0 },
00136     [PIX_FMT_BGR48BE]     = { 1 , 1 },
00137     [PIX_FMT_BGR48LE]     = { 1 , 1 },
00138     [PIX_FMT_BGRA64BE]    = { 0 , 0 },
00139     [PIX_FMT_BGRA64LE]    = { 0 , 0 },
00140     [PIX_FMT_YUV420P9BE]  = { 1 , 1 },
00141     [PIX_FMT_YUV420P9LE]  = { 1 , 1 },
00142     [PIX_FMT_YUV420P10BE] = { 1 , 1 },
00143     [PIX_FMT_YUV420P10LE] = { 1 , 1 },
00144     [PIX_FMT_YUV422P9BE]  = { 1 , 1 },
00145     [PIX_FMT_YUV422P9LE]  = { 1 , 1 },
00146     [PIX_FMT_YUV422P10BE] = { 1 , 1 },
00147     [PIX_FMT_YUV422P10LE] = { 1 , 1 },
00148     [PIX_FMT_YUV444P9BE]  = { 1 , 1 },
00149     [PIX_FMT_YUV444P9LE]  = { 1 , 1 },
00150     [PIX_FMT_YUV444P10BE] = { 1 , 1 },
00151     [PIX_FMT_YUV444P10LE] = { 1 , 1 },
00152     [PIX_FMT_GBR24P]      = { 1 , 0 },
00153     [PIX_FMT_GBRP]        = { 1 , 0 },
00154     [PIX_FMT_GBRP9LE]     = { 1 , 0 },
00155     [PIX_FMT_GBRP9BE]     = { 1 , 0 },
00156     [PIX_FMT_GBRP10LE]    = { 1 , 0 },
00157     [PIX_FMT_GBRP10BE]    = { 1 , 0 },
00158     [PIX_FMT_GBRP16LE]    = { 1 , 0 },
00159     [PIX_FMT_GBRP16BE]    = { 1 , 0 },
00160 };
00161 
00162 int sws_isSupportedInput(enum PixelFormat pix_fmt)
00163 {
00164     return (unsigned)pix_fmt < PIX_FMT_NB ?
00165         format_entries[pix_fmt].is_supported_in : 0;
00166 }
00167 
00168 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
00169 {
00170     return (unsigned)pix_fmt < PIX_FMT_NB ?
00171         format_entries[pix_fmt].is_supported_out : 0;
00172 }
00173 
00174 extern const int32_t ff_yuv2rgb_coeffs[8][4];
00175 
00176 #if FF_API_SWS_FORMAT_NAME
00177 const char *sws_format_name(enum PixelFormat format)
00178 {
00179     return av_get_pix_fmt_name(format);
00180 }
00181 #endif
00182 
00183 static double getSplineCoeff(double a, double b, double c, double d, double dist)
00184 {
00185     if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
00186     else           return getSplineCoeff(        0.0,
00187                                           b+ 2.0*c + 3.0*d,
00188                                                  c + 3.0*d,
00189                                          -b- 3.0*c - 6.0*d,
00190                                          dist-1.0);
00191 }
00192 
00193 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
00194                       int srcW, int dstW, int filterAlign, int one, int flags, int cpu_flags,
00195                       SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
00196 {
00197     int i;
00198     int filterSize;
00199     int filter2Size;
00200     int minFilterSize;
00201     int64_t *filter=NULL;
00202     int64_t *filter2=NULL;
00203     const int64_t fone= 1LL<<54;
00204     int ret= -1;
00205 
00206     emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions)
00207 
00208     // NOTE: the +3 is for the MMX(+1)/SSE(+3) scaler which reads over the end
00209     FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+3)*sizeof(int16_t), fail);
00210 
00211     if (FFABS(xInc - 0x10000) <10) { // unscaled
00212         int i;
00213         filterSize= 1;
00214         FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
00215 
00216         for (i=0; i<dstW; i++) {
00217             filter[i*filterSize]= fone;
00218             (*filterPos)[i]=i;
00219         }
00220 
00221     } else if (flags&SWS_POINT) { // lame looking point sampling mode
00222         int i;
00223         int64_t xDstInSrc;
00224         filterSize= 1;
00225         FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
00226 
00227         xDstInSrc= xInc/2 - 0x8000;
00228         for (i=0; i<dstW; i++) {
00229             int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
00230 
00231             (*filterPos)[i]= xx;
00232             filter[i]= fone;
00233             xDstInSrc+= xInc;
00234         }
00235     } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
00236         int i;
00237         int64_t xDstInSrc;
00238         filterSize= 2;
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             int j;
00245 
00246             (*filterPos)[i]= xx;
00247             //bilinear upscale / linear interpolate / area averaging
00248             for (j=0; j<filterSize; j++) {
00249                 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
00250                 if (coeff<0) coeff=0;
00251                 filter[i*filterSize + j]= coeff;
00252                 xx++;
00253             }
00254             xDstInSrc+= xInc;
00255         }
00256     } else {
00257         int64_t xDstInSrc;
00258         int sizeFactor;
00259 
00260         if      (flags&SWS_BICUBIC)      sizeFactor=  4;
00261         else if (flags&SWS_X)            sizeFactor=  8;
00262         else if (flags&SWS_AREA)         sizeFactor=  1; //downscale only, for upscale it is bilinear
00263         else if (flags&SWS_GAUSS)        sizeFactor=  8;   // infinite ;)
00264         else if (flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
00265         else if (flags&SWS_SINC)         sizeFactor= 20; // infinite ;)
00266         else if (flags&SWS_SPLINE)       sizeFactor= 20;  // infinite ;)
00267         else if (flags&SWS_BILINEAR)     sizeFactor=  2;
00268         else {
00269             sizeFactor= 0; //GCC warning killer
00270             assert(0);
00271         }
00272 
00273         if (xInc <= 1<<16)      filterSize= 1 + sizeFactor; // upscale
00274         else                    filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
00275 
00276         if (filterSize > srcW-2) filterSize=srcW-2;
00277 
00278         FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
00279 
00280         xDstInSrc= xInc - 0x10000;
00281         for (i=0; i<dstW; i++) {
00282             int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
00283             int j;
00284             (*filterPos)[i]= xx;
00285             for (j=0; j<filterSize; j++) {
00286                 int64_t d= (FFABS(((int64_t)xx<<17) - xDstInSrc))<<13;
00287                 double floatd;
00288                 int64_t coeff;
00289 
00290                 if (xInc > 1<<16)
00291                     d= d*dstW/srcW;
00292                 floatd= d * (1.0/(1<<30));
00293 
00294                 if (flags & SWS_BICUBIC) {
00295                     int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] :   0) * (1<<24);
00296                     int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
00297 
00298                     if (d >= 1LL<<31) {
00299                         coeff = 0.0;
00300                     } else {
00301                         int64_t dd  = (d  * d) >> 30;
00302                         int64_t ddd = (dd * d) >> 30;
00303 
00304                         if (d < 1LL<<30)
00305                             coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
00306                         else
00307                             coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
00308                     }
00309                     coeff *= fone>>(30+24);
00310                 }
00311 /*                else if (flags & SWS_X) {
00312                     double p= param ? param*0.01 : 0.3;
00313                     coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
00314                     coeff*= pow(2.0, - p*d*d);
00315                 }*/
00316                 else if (flags & SWS_X) {
00317                     double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
00318                     double c;
00319 
00320                     if (floatd<1.0)
00321                         c = cos(floatd*M_PI);
00322                     else
00323                         c=-1.0;
00324                     if (c<0.0)      c= -pow(-c, A);
00325                     else            c=  pow( c, A);
00326                     coeff= (c*0.5 + 0.5)*fone;
00327                 } else if (flags & SWS_AREA) {
00328                     int64_t d2= d - (1<<29);
00329                     if      (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
00330                     else if (d2*xInc <  (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
00331                     else coeff=0.0;
00332                     coeff *= fone>>(30+16);
00333                 } else if (flags & SWS_GAUSS) {
00334                     double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
00335                     coeff = (pow(2.0, - p*floatd*floatd))*fone;
00336                 } else if (flags & SWS_SINC) {
00337                     coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
00338                 } else if (flags & SWS_LANCZOS) {
00339                     double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
00340                     coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
00341                     if (floatd>p) coeff=0;
00342                 } else if (flags & SWS_BILINEAR) {
00343                     coeff= (1<<30) - d;
00344                     if (coeff<0) coeff=0;
00345                     coeff *= fone >> 30;
00346                 } else if (flags & SWS_SPLINE) {
00347                     double p=-2.196152422706632;
00348                     coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
00349                 } else {
00350                     coeff= 0.0; //GCC warning killer
00351                     assert(0);
00352                 }
00353 
00354                 filter[i*filterSize + j]= coeff;
00355                 xx++;
00356             }
00357             xDstInSrc+= 2*xInc;
00358         }
00359     }
00360 
00361     /* apply src & dst Filter to filter -> filter2
00362        av_free(filter);
00363     */
00364     assert(filterSize>0);
00365     filter2Size= filterSize;
00366     if (srcFilter) filter2Size+= srcFilter->length - 1;
00367     if (dstFilter) filter2Size+= dstFilter->length - 1;
00368     assert(filter2Size>0);
00369     FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
00370 
00371     for (i=0; i<dstW; i++) {
00372         int j, k;
00373 
00374         if(srcFilter) {
00375             for (k=0; k<srcFilter->length; k++) {
00376                 for (j=0; j<filterSize; j++)
00377                     filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
00378             }
00379         } else {
00380             for (j=0; j<filterSize; j++)
00381                 filter2[i*filter2Size + j]= filter[i*filterSize + j];
00382         }
00383         //FIXME dstFilter
00384 
00385         (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
00386     }
00387     av_freep(&filter);
00388 
00389     /* try to reduce the filter-size (step1 find size and shift left) */
00390     // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
00391     minFilterSize= 0;
00392     for (i=dstW-1; i>=0; i--) {
00393         int min= filter2Size;
00394         int j;
00395         int64_t cutOff=0.0;
00396 
00397         /* get rid of near zero elements on the left by shifting left */
00398         for (j=0; j<filter2Size; j++) {
00399             int k;
00400             cutOff += FFABS(filter2[i*filter2Size]);
00401 
00402             if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
00403 
00404             /* preserve monotonicity because the core can't handle the filter otherwise */
00405             if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
00406 
00407             // move filter coefficients left
00408             for (k=1; k<filter2Size; k++)
00409                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
00410             filter2[i*filter2Size + k - 1]= 0;
00411             (*filterPos)[i]++;
00412         }
00413 
00414         cutOff=0;
00415         /* count near zeros on the right */
00416         for (j=filter2Size-1; j>0; j--) {
00417             cutOff += FFABS(filter2[i*filter2Size + j]);
00418 
00419             if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
00420             min--;
00421         }
00422 
00423         if (min>minFilterSize) minFilterSize= min;
00424     }
00425 
00426     if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
00427         // we can handle the special case 4,
00428         // so we don't want to go to the full 8
00429         if (minFilterSize < 5)
00430             filterAlign = 4;
00431 
00432         // We really don't want to waste our time
00433         // doing useless computation, so fall back on
00434         // the scalar C code for very small filters.
00435         // Vectorizing is worth it only if you have a
00436         // decent-sized vector.
00437         if (minFilterSize < 3)
00438             filterAlign = 1;
00439     }
00440 
00441     if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
00442         // special case for unscaled vertical filtering
00443         if (minFilterSize == 1 && filterAlign == 2)
00444             filterAlign= 1;
00445     }
00446 
00447     assert(minFilterSize > 0);
00448     filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
00449     assert(filterSize > 0);
00450     filter= av_malloc(filterSize*dstW*sizeof(*filter));
00451     if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
00452         goto fail;
00453     *outFilterSize= filterSize;
00454 
00455     if (flags&SWS_PRINT_INFO)
00456         av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
00457     /* try to reduce the filter-size (step2 reduce it) */
00458     for (i=0; i<dstW; i++) {
00459         int j;
00460 
00461         for (j=0; j<filterSize; j++) {
00462             if (j>=filter2Size) filter[i*filterSize + j]= 0;
00463             else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
00464             if((flags & SWS_BITEXACT) && j>=minFilterSize)
00465                 filter[i*filterSize + j]= 0;
00466         }
00467     }
00468 
00469     //FIXME try to align filterPos if possible
00470 
00471     //fix borders
00472     for (i=0; i<dstW; i++) {
00473         int j;
00474         if ((*filterPos)[i] < 0) {
00475             // move filter coefficients left to compensate for filterPos
00476             for (j=1; j<filterSize; j++) {
00477                 int left= FFMAX(j + (*filterPos)[i], 0);
00478                 filter[i*filterSize + left] += filter[i*filterSize + j];
00479                 filter[i*filterSize + j]=0;
00480             }
00481             (*filterPos)[i]= 0;
00482         }
00483 
00484         if ((*filterPos)[i] + filterSize > srcW) {
00485             int shift= (*filterPos)[i] + filterSize - srcW;
00486             // move filter coefficients right to compensate for filterPos
00487             for (j=filterSize-2; j>=0; j--) {
00488                 int right= FFMIN(j + shift, filterSize-1);
00489                 filter[i*filterSize +right] += filter[i*filterSize +j];
00490                 filter[i*filterSize +j]=0;
00491             }
00492             (*filterPos)[i]= srcW - filterSize;
00493         }
00494     }
00495 
00496     // Note the +1 is for the MMX scaler which reads over the end
00497     /* align at 16 for AltiVec (needed by hScale_altivec_real) */
00498     FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+3)*sizeof(int16_t), fail);
00499 
00500     /* normalize & store in outFilter */
00501     for (i=0; i<dstW; i++) {
00502         int j;
00503         int64_t error=0;
00504         int64_t sum=0;
00505 
00506         for (j=0; j<filterSize; j++) {
00507             sum+= filter[i*filterSize + j];
00508         }
00509         sum= (sum + one/2)/ one;
00510         for (j=0; j<*outFilterSize; j++) {
00511             int64_t v= filter[i*filterSize + j] + error;
00512             int intV= ROUNDED_DIV(v, sum);
00513             (*outFilter)[i*(*outFilterSize) + j]= intV;
00514             error= v - intV*sum;
00515         }
00516     }
00517 
00518     (*filterPos)[dstW+0] =
00519     (*filterPos)[dstW+1] =
00520     (*filterPos)[dstW+2] = (*filterPos)[dstW-1]; // the MMX/SSE scaler will read over the end
00521     for (i=0; i<*outFilterSize; i++) {
00522         int k= (dstW - 1) * (*outFilterSize) + i;
00523         (*outFilter)[k + 1 * (*outFilterSize)] =
00524         (*outFilter)[k + 2 * (*outFilterSize)] =
00525         (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
00526     }
00527 
00528     ret=0;
00529 fail:
00530     av_free(filter);
00531     av_free(filter2);
00532     return ret;
00533 }
00534 
00535 #if HAVE_MMX2
00536 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
00537 {
00538     uint8_t *fragmentA;
00539     x86_reg imm8OfPShufW1A;
00540     x86_reg imm8OfPShufW2A;
00541     x86_reg fragmentLengthA;
00542     uint8_t *fragmentB;
00543     x86_reg imm8OfPShufW1B;
00544     x86_reg imm8OfPShufW2B;
00545     x86_reg fragmentLengthB;
00546     int fragmentPos;
00547 
00548     int xpos, i;
00549 
00550     // create an optimized horizontal scaling routine
00551     /* This scaler is made of runtime-generated MMX2 code using specially
00552      * tuned pshufw instructions. For every four output pixels, if four
00553      * input pixels are enough for the fast bilinear scaling, then a chunk
00554      * of fragmentB is used. If five input pixels are needed, then a chunk
00555      * of fragmentA is used.
00556      */
00557 
00558     //code fragment
00559 
00560     __asm__ volatile(
00561         "jmp                         9f                 \n\t"
00562     // Begin
00563         "0:                                             \n\t"
00564         "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
00565         "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
00566         "movd   1(%%"REG_c", %%"REG_S"), %%mm1          \n\t"
00567         "punpcklbw                %%mm7, %%mm1          \n\t"
00568         "punpcklbw                %%mm7, %%mm0          \n\t"
00569         "pshufw                   $0xFF, %%mm1, %%mm1   \n\t"
00570         "1:                                             \n\t"
00571         "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
00572         "2:                                             \n\t"
00573         "psubw                    %%mm1, %%mm0          \n\t"
00574         "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
00575         "pmullw                   %%mm3, %%mm0          \n\t"
00576         "psllw                       $7, %%mm1          \n\t"
00577         "paddw                    %%mm1, %%mm0          \n\t"
00578 
00579         "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"
00580 
00581         "add                         $8, %%"REG_a"      \n\t"
00582     // End
00583         "9:                                             \n\t"
00584 //        "int $3                                         \n\t"
00585         "lea                 " LOCAL_MANGLE(0b) ", %0   \n\t"
00586         "lea                 " LOCAL_MANGLE(1b) ", %1   \n\t"
00587         "lea                 " LOCAL_MANGLE(2b) ", %2   \n\t"
00588         "dec                         %1                 \n\t"
00589         "dec                         %2                 \n\t"
00590         "sub                         %0, %1             \n\t"
00591         "sub                         %0, %2             \n\t"
00592         "lea                 " LOCAL_MANGLE(9b) ", %3   \n\t"
00593         "sub                         %0, %3             \n\t"
00594 
00595 
00596         :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
00597         "=r" (fragmentLengthA)
00598     );
00599 
00600     __asm__ volatile(
00601         "jmp                         9f                 \n\t"
00602     // Begin
00603         "0:                                             \n\t"
00604         "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
00605         "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
00606         "punpcklbw                %%mm7, %%mm0          \n\t"
00607         "pshufw                   $0xFF, %%mm0, %%mm1   \n\t"
00608         "1:                                             \n\t"
00609         "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
00610         "2:                                             \n\t"
00611         "psubw                    %%mm1, %%mm0          \n\t"
00612         "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
00613         "pmullw                   %%mm3, %%mm0          \n\t"
00614         "psllw                       $7, %%mm1          \n\t"
00615         "paddw                    %%mm1, %%mm0          \n\t"
00616 
00617         "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"
00618 
00619         "add                         $8, %%"REG_a"      \n\t"
00620     // End
00621         "9:                                             \n\t"
00622 //        "int                       $3                   \n\t"
00623         "lea                 " LOCAL_MANGLE(0b) ", %0   \n\t"
00624         "lea                 " LOCAL_MANGLE(1b) ", %1   \n\t"
00625         "lea                 " LOCAL_MANGLE(2b) ", %2   \n\t"
00626         "dec                         %1                 \n\t"
00627         "dec                         %2                 \n\t"
00628         "sub                         %0, %1             \n\t"
00629         "sub                         %0, %2             \n\t"
00630         "lea                 " LOCAL_MANGLE(9b) ", %3   \n\t"
00631         "sub                         %0, %3             \n\t"
00632 
00633 
00634         :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
00635         "=r" (fragmentLengthB)
00636     );
00637 
00638     xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
00639     fragmentPos=0;
00640 
00641     for (i=0; i<dstW/numSplits; i++) {
00642         int xx=xpos>>16;
00643 
00644         if ((i&3) == 0) {
00645             int a=0;
00646             int b=((xpos+xInc)>>16) - xx;
00647             int c=((xpos+xInc*2)>>16) - xx;
00648             int d=((xpos+xInc*3)>>16) - xx;
00649             int inc                = (d+1<4);
00650             uint8_t *fragment      = (d+1<4) ? fragmentB       : fragmentA;
00651             x86_reg imm8OfPShufW1  = (d+1<4) ? imm8OfPShufW1B  : imm8OfPShufW1A;
00652             x86_reg imm8OfPShufW2  = (d+1<4) ? imm8OfPShufW2B  : imm8OfPShufW2A;
00653             x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
00654             int maxShift= 3-(d+inc);
00655             int shift=0;
00656 
00657             if (filterCode) {
00658                 filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
00659                 filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
00660                 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
00661                 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
00662                 filterPos[i/2]= xx;
00663 
00664                 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
00665 
00666                 filterCode[fragmentPos + imm8OfPShufW1]=
00667                     (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
00668                 filterCode[fragmentPos + imm8OfPShufW2]=
00669                     a | (b<<2) | (c<<4) | (d<<6);
00670 
00671                 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
00672                 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
00673 
00674                 if (shift && i>=shift) {
00675                     filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
00676                     filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
00677                     filterPos[i/2]-=shift;
00678                 }
00679             }
00680 
00681             fragmentPos+= fragmentLength;
00682 
00683             if (filterCode)
00684                 filterCode[fragmentPos]= RET;
00685         }
00686         xpos+=xInc;
00687     }
00688     if (filterCode)
00689         filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
00690 
00691     return fragmentPos + 1;
00692 }
00693 #endif /* HAVE_MMX2 */
00694 
00695 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
00696 {
00697     *h = av_pix_fmt_descriptors[format].log2_chroma_w;
00698     *v = av_pix_fmt_descriptors[format].log2_chroma_h;
00699 }
00700 
00701 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
00702                              int srcRange, const int table[4], int dstRange,
00703                              int brightness, int contrast, int saturation)
00704 {
00705     memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
00706     memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
00707 
00708     c->brightness= brightness;
00709     c->contrast  = contrast;
00710     c->saturation= saturation;
00711     c->srcRange  = srcRange;
00712     c->dstRange  = dstRange;
00713     if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
00714 
00715     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
00716     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
00717 
00718     ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
00719     //FIXME factorize
00720 
00721     if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
00722         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
00723     return 0;
00724 }
00725 
00726 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
00727                              int *srcRange, int **table, int *dstRange,
00728                              int *brightness, int *contrast, int *saturation)
00729 {
00730     if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
00731 
00732     *inv_table = c->srcColorspaceTable;
00733     *table     = c->dstColorspaceTable;
00734     *srcRange  = c->srcRange;
00735     *dstRange  = c->dstRange;
00736     *brightness= c->brightness;
00737     *contrast  = c->contrast;
00738     *saturation= c->saturation;
00739 
00740     return 0;
00741 }
00742 
00743 static int handle_jpeg(enum PixelFormat *format)
00744 {
00745     switch (*format) {
00746     case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
00747     case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
00748     case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
00749     case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
00750     case PIX_FMT_0BGR    : *format = PIX_FMT_ABGR   ; return 0;
00751     case PIX_FMT_BGR0    : *format = PIX_FMT_BGRA   ; return 0;
00752     case PIX_FMT_0RGB    : *format = PIX_FMT_ARGB   ; return 0;
00753     case PIX_FMT_RGB0    : *format = PIX_FMT_RGBA   ; return 0;
00754     default:                                          return 0;
00755     }
00756 }
00757 
00758 SwsContext *sws_alloc_context(void)
00759 {
00760     SwsContext *c= av_mallocz(sizeof(SwsContext));
00761 
00762     c->av_class = &sws_context_class;
00763     av_opt_set_defaults(c);
00764 
00765     return c;
00766 }
00767 
00768 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
00769 {
00770     int i, j;
00771     int usesVFilter, usesHFilter;
00772     int unscaled;
00773     SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
00774     int srcW= c->srcW;
00775     int srcH= c->srcH;
00776     int dstW= c->dstW;
00777     int dstH= c->dstH;
00778     int dst_stride = FFALIGN(dstW * sizeof(int16_t)+66, 16);
00779     int flags, cpu_flags;
00780     enum PixelFormat srcFormat= c->srcFormat;
00781     enum PixelFormat dstFormat= c->dstFormat;
00782 
00783     cpu_flags = av_get_cpu_flags();
00784     flags     = c->flags;
00785     emms_c();
00786     if (!rgb15to16) sws_rgb2rgb_init();
00787 
00788     unscaled = (srcW == dstW && srcH == dstH);
00789 
00790     handle_jpeg(&srcFormat);
00791     handle_jpeg(&dstFormat);
00792 
00793     if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){
00794         av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
00795         c->srcFormat= srcFormat;
00796         c->dstFormat= dstFormat;
00797     }
00798 
00799     if (!sws_isSupportedInput(srcFormat)) {
00800         av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n", av_get_pix_fmt_name(srcFormat));
00801         return AVERROR(EINVAL);
00802     }
00803     if (!sws_isSupportedOutput(dstFormat)) {
00804         av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n", av_get_pix_fmt_name(dstFormat));
00805         return AVERROR(EINVAL);
00806     }
00807 
00808     i= flags & ( SWS_POINT
00809                 |SWS_AREA
00810                 |SWS_BILINEAR
00811                 |SWS_FAST_BILINEAR
00812                 |SWS_BICUBIC
00813                 |SWS_X
00814                 |SWS_GAUSS
00815                 |SWS_LANCZOS
00816                 |SWS_SINC
00817                 |SWS_SPLINE
00818                 |SWS_BICUBLIN);
00819     if(!i || (i & (i-1))) {
00820         av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen\n");
00821         return AVERROR(EINVAL);
00822     }
00823     /* sanity check */
00824     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
00825         av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
00826                srcW, srcH, dstW, dstH);
00827         return AVERROR(EINVAL);
00828     }
00829 
00830     if (!dstFilter) dstFilter= &dummyFilter;
00831     if (!srcFilter) srcFilter= &dummyFilter;
00832 
00833     c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
00834     c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
00835     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
00836     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
00837     c->vRounder= 4* 0x0001000100010001ULL;
00838 
00839     usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
00840                   (srcFilter->chrV && srcFilter->chrV->length>1) ||
00841                   (dstFilter->lumV && dstFilter->lumV->length>1) ||
00842                   (dstFilter->chrV && dstFilter->chrV->length>1);
00843     usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
00844                   (srcFilter->chrH && srcFilter->chrH->length>1) ||
00845                   (dstFilter->lumH && dstFilter->lumH->length>1) ||
00846                   (dstFilter->chrH && dstFilter->chrH->length>1);
00847 
00848     getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
00849     getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
00850 
00851     // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
00852     if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
00853 
00854     // drop some chroma lines if the user wants it
00855     c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
00856     c->chrSrcVSubSample+= c->vChrDrop;
00857 
00858     // drop every other pixel for chroma calculation unless user wants full chroma
00859     if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
00860       && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8
00861       && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4
00862       && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
00863       && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
00864         c->chrSrcHSubSample=1;
00865 
00866     // Note the -((-x)>>y) is so that we always round toward +inf.
00867     c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
00868     c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
00869     c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
00870     c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
00871 
00872     /* unscaled special cases */
00873     if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
00874         ff_get_unscaled_swscale(c);
00875 
00876         if (c->swScale) {
00877             if (flags&SWS_PRINT_INFO)
00878                 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
00879                        av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
00880             return 0;
00881         }
00882     }
00883 
00884     c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;
00885     if (c->srcBpc < 8)
00886         c->srcBpc = 8;
00887     c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;
00888     if (c->dstBpc < 8)
00889         c->dstBpc = 8;
00890     if (isAnyRGB(srcFormat) || srcFormat == PIX_FMT_PAL8)
00891         c->srcBpc = 16;
00892     if (c->dstBpc == 16)
00893         dst_stride <<= 1;
00894     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
00895     if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->srcBpc == 8 && c->dstBpc <= 10) {
00896         c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
00897         if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
00898             if (flags&SWS_PRINT_INFO)
00899                 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
00900         }
00901         if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat)) c->canMMX2BeUsed=0;
00902     }
00903     else
00904         c->canMMX2BeUsed=0;
00905 
00906     c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
00907     c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
00908 
00909     // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
00910     // but only for the FAST_BILINEAR mode otherwise do correct scaling
00911     // n-2 is the last chrominance sample available
00912     // this is not perfect, but no one should notice the difference, the more correct variant
00913     // would be like the vertical one, but that would require some special code for the
00914     // first and last pixel
00915     if (flags&SWS_FAST_BILINEAR) {
00916         if (c->canMMX2BeUsed) {
00917             c->lumXInc+= 20;
00918             c->chrXInc+= 20;
00919         }
00920         //we don't use the x86 asm scaler if MMX is available
00921         else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX && c->dstBpc <= 10) {
00922             c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
00923             c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
00924         }
00925     }
00926 
00927     /* precalculate horizontal scaler filter coefficients */
00928     {
00929 #if HAVE_MMX2
00930 // can't downscale !!!
00931         if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
00932             c->lumMmx2FilterCodeSize = initMMX2HScaler(      dstW, c->lumXInc, NULL, NULL, NULL, 8);
00933             c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
00934 
00935 #ifdef MAP_ANONYMOUS
00936             c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
00937             c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
00938 #elif HAVE_VIRTUALALLOC
00939             c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
00940             c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
00941 #else
00942             c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
00943             c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
00944 #endif
00945 
00946 #ifdef MAP_ANONYMOUS
00947             if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)
00948 #else
00949             if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
00950 #endif
00951                 return AVERROR(ENOMEM);
00952             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter   , (dstW        /8+8)*sizeof(int16_t), fail);
00953             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter   , (c->chrDstW  /4+8)*sizeof(int16_t), fail);
00954             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW      /2/8+8)*sizeof(int32_t), fail);
00955             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
00956 
00957             initMMX2HScaler(      dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
00958             initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
00959 
00960 #ifdef MAP_ANONYMOUS
00961             mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
00962             mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
00963 #endif
00964         } else
00965 #endif /* HAVE_MMX2 */
00966         {
00967             const int filterAlign=
00968                 (HAVE_MMX     && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
00969                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
00970                 1;
00971 
00972             if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
00973                            srcW      ,       dstW, filterAlign, 1<<14,
00974                            (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags, cpu_flags,
00975                            srcFilter->lumH, dstFilter->lumH, c->param) < 0)
00976                 goto fail;
00977             if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
00978                            c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
00979                            (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
00980                            srcFilter->chrH, dstFilter->chrH, c->param) < 0)
00981                 goto fail;
00982         }
00983     } // initialize horizontal stuff
00984 
00985     /* precalculate vertical scaler filter coefficients */
00986     {
00987         const int filterAlign=
00988             (HAVE_MMX     && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
00989             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
00990             1;
00991 
00992         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
00993                        srcH      ,        dstH, filterAlign, (1<<12),
00994                        (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags, cpu_flags,
00995                        srcFilter->lumV, dstFilter->lumV, c->param) < 0)
00996             goto fail;
00997         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
00998                        c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
00999                        (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
01000                        srcFilter->chrV, dstFilter->chrV, c->param) < 0)
01001             goto fail;
01002 
01003 #if HAVE_ALTIVEC
01004         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
01005         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
01006 
01007         for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
01008             int j;
01009             short *p = (short *)&c->vYCoeffsBank[i];
01010             for (j=0;j<8;j++)
01011                 p[j] = c->vLumFilter[i];
01012         }
01013 
01014         for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
01015             int j;
01016             short *p = (short *)&c->vCCoeffsBank[i];
01017             for (j=0;j<8;j++)
01018                 p[j] = c->vChrFilter[i];
01019         }
01020 #endif
01021     }
01022 
01023     // calculate buffer sizes so that they won't run out while handling these damn slices
01024     c->vLumBufSize= c->vLumFilterSize;
01025     c->vChrBufSize= c->vChrFilterSize;
01026     for (i=0; i<dstH; i++) {
01027         int chrI= (int64_t)i*c->chrDstH / dstH;
01028         int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
01029                            ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
01030 
01031         nextSlice>>= c->chrSrcVSubSample;
01032         nextSlice<<= c->chrSrcVSubSample;
01033         if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
01034             c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
01035         if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
01036             c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
01037     }
01038 
01039     // allocate pixbufs (we use dynamic allocation because otherwise we would need to
01040     // allocate several megabytes to handle all possible cases)
01041     FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
01042     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
01043     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
01044     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
01045         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
01046     //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)
01047     /* align at 16 bytes for AltiVec */
01048     for (i=0; i<c->vLumBufSize; i++) {
01049         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
01050         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
01051     }
01052     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
01053     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
01054     c->uv_offx2 = dst_stride + 16;
01055     for (i=0; i<c->vChrBufSize; i++) {
01056         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);
01057         c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
01058         c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
01059     }
01060     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
01061         for (i=0; i<c->vLumBufSize; i++) {
01062             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
01063             c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
01064         }
01065 
01066     //try to avoid drawing green stuff between the right end and the stride end
01067     for (i=0; i<c->vChrBufSize; i++)
01068         if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){
01069             av_assert0(c->dstBpc > 10);
01070             for(j=0; j<dst_stride/2+1; j++)
01071                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
01072         } else
01073             for(j=0; j<dst_stride+1; j++)
01074                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
01075 
01076     assert(c->chrDstH <= dstH);
01077 
01078     if (flags&SWS_PRINT_INFO) {
01079         if      (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
01080         else if (flags&SWS_BILINEAR)      av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
01081         else if (flags&SWS_BICUBIC)       av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
01082         else if (flags&SWS_X)             av_log(c, AV_LOG_INFO, "Experimental scaler, ");
01083         else if (flags&SWS_POINT)         av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
01084         else if (flags&SWS_AREA)          av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
01085         else if (flags&SWS_BICUBLIN)      av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
01086         else if (flags&SWS_GAUSS)         av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
01087         else if (flags&SWS_SINC)          av_log(c, AV_LOG_INFO, "Sinc scaler, ");
01088         else if (flags&SWS_LANCZOS)       av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
01089         else if (flags&SWS_SPLINE)        av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
01090         else                              av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
01091 
01092         av_log(c, AV_LOG_INFO, "from %s to %s%s ",
01093                av_get_pix_fmt_name(srcFormat),
01094 #ifdef DITHER1XBPP
01095                dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
01096                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
01097                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
01098 #else
01099                "",
01100 #endif
01101                av_get_pix_fmt_name(dstFormat));
01102 
01103         if      (HAVE_MMX2     && cpu_flags & AV_CPU_FLAG_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");
01104         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");
01105         else if (HAVE_MMX      && cpu_flags & AV_CPU_FLAG_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");
01106         else if (HAVE_ALTIVEC  && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
01107         else                                   av_log(c, AV_LOG_INFO, "using C\n");
01108 
01109         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
01110         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
01111                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
01112         av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
01113                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
01114     }
01115 
01116     c->swScale= ff_getSwsFunc(c);
01117     return 0;
01118 fail: //FIXME replace things by appropriate error codes
01119     return -1;
01120 }
01121 
01122 #if FF_API_SWS_GETCONTEXT
01123 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
01124                            int dstW, int dstH, enum PixelFormat dstFormat, int flags,
01125                            SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
01126 {
01127     SwsContext *c;
01128 
01129     if(!(c=sws_alloc_context()))
01130         return NULL;
01131 
01132     c->flags= flags;
01133     c->srcW= srcW;
01134     c->srcH= srcH;
01135     c->dstW= dstW;
01136     c->dstH= dstH;
01137     c->srcRange = handle_jpeg(&srcFormat);
01138     c->dstRange = handle_jpeg(&dstFormat);
01139     c->srcFormat= srcFormat;
01140     c->dstFormat= dstFormat;
01141 
01142     if (param) {
01143         c->param[0] = param[0];
01144         c->param[1] = param[1];
01145     }
01146     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);
01147 
01148     if(sws_init_context(c, srcFilter, dstFilter) < 0){
01149         sws_freeContext(c);
01150         return NULL;
01151     }
01152 
01153     return c;
01154 }
01155 #endif
01156 
01157 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
01158                                 float lumaSharpen, float chromaSharpen,
01159                                 float chromaHShift, float chromaVShift,
01160                                 int verbose)
01161 {
01162     SwsFilter *filter= av_malloc(sizeof(SwsFilter));
01163     if (!filter)
01164         return NULL;
01165 
01166     if (lumaGBlur!=0.0) {
01167         filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
01168         filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
01169     } else {
01170         filter->lumH= sws_getIdentityVec();
01171         filter->lumV= sws_getIdentityVec();
01172     }
01173 
01174     if (chromaGBlur!=0.0) {
01175         filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
01176         filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
01177     } else {
01178         filter->chrH= sws_getIdentityVec();
01179         filter->chrV= sws_getIdentityVec();
01180     }
01181 
01182     if (chromaSharpen!=0.0) {
01183         SwsVector *id= sws_getIdentityVec();
01184         sws_scaleVec(filter->chrH, -chromaSharpen);
01185         sws_scaleVec(filter->chrV, -chromaSharpen);
01186         sws_addVec(filter->chrH, id);
01187         sws_addVec(filter->chrV, id);
01188         sws_freeVec(id);
01189     }
01190 
01191     if (lumaSharpen!=0.0) {
01192         SwsVector *id= sws_getIdentityVec();
01193         sws_scaleVec(filter->lumH, -lumaSharpen);
01194         sws_scaleVec(filter->lumV, -lumaSharpen);
01195         sws_addVec(filter->lumH, id);
01196         sws_addVec(filter->lumV, id);
01197         sws_freeVec(id);
01198     }
01199 
01200     if (chromaHShift != 0.0)
01201         sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
01202 
01203     if (chromaVShift != 0.0)
01204         sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
01205 
01206     sws_normalizeVec(filter->chrH, 1.0);
01207     sws_normalizeVec(filter->chrV, 1.0);
01208     sws_normalizeVec(filter->lumH, 1.0);
01209     sws_normalizeVec(filter->lumV, 1.0);
01210 
01211     if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
01212     if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
01213 
01214     return filter;
01215 }
01216 
01217 SwsVector *sws_allocVec(int length)
01218 {
01219     SwsVector *vec = av_malloc(sizeof(SwsVector));
01220     if (!vec)
01221         return NULL;
01222     vec->length = length;
01223     vec->coeff  = av_malloc(sizeof(double) * length);
01224     if (!vec->coeff)
01225         av_freep(&vec);
01226     return vec;
01227 }
01228 
01229 SwsVector *sws_getGaussianVec(double variance, double quality)
01230 {
01231     const int length= (int)(variance*quality + 0.5) | 1;
01232     int i;
01233     double middle= (length-1)*0.5;
01234     SwsVector *vec= sws_allocVec(length);
01235 
01236     if (!vec)
01237         return NULL;
01238 
01239     for (i=0; i<length; i++) {
01240         double dist= i-middle;
01241         vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
01242     }
01243 
01244     sws_normalizeVec(vec, 1.0);
01245 
01246     return vec;
01247 }
01248 
01249 SwsVector *sws_getConstVec(double c, int length)
01250 {
01251     int i;
01252     SwsVector *vec= sws_allocVec(length);
01253 
01254     if (!vec)
01255         return NULL;
01256 
01257     for (i=0; i<length; i++)
01258         vec->coeff[i]= c;
01259 
01260     return vec;
01261 }
01262 
01263 SwsVector *sws_getIdentityVec(void)
01264 {
01265     return sws_getConstVec(1.0, 1);
01266 }
01267 
01268 static double sws_dcVec(SwsVector *a)
01269 {
01270     int i;
01271     double sum=0;
01272 
01273     for (i=0; i<a->length; i++)
01274         sum+= a->coeff[i];
01275 
01276     return sum;
01277 }
01278 
01279 void sws_scaleVec(SwsVector *a, double scalar)
01280 {
01281     int i;
01282 
01283     for (i=0; i<a->length; i++)
01284         a->coeff[i]*= scalar;
01285 }
01286 
01287 void sws_normalizeVec(SwsVector *a, double height)
01288 {
01289     sws_scaleVec(a, height/sws_dcVec(a));
01290 }
01291 
01292 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
01293 {
01294     int length= a->length + b->length - 1;
01295     int i, j;
01296     SwsVector *vec= sws_getConstVec(0.0, length);
01297 
01298     if (!vec)
01299         return NULL;
01300 
01301     for (i=0; i<a->length; i++) {
01302         for (j=0; j<b->length; j++) {
01303             vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
01304         }
01305     }
01306 
01307     return vec;
01308 }
01309 
01310 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
01311 {
01312     int length= FFMAX(a->length, b->length);
01313     int i;
01314     SwsVector *vec= sws_getConstVec(0.0, length);
01315 
01316     if (!vec)
01317         return NULL;
01318 
01319     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
01320     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
01321 
01322     return vec;
01323 }
01324 
01325 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
01326 {
01327     int length= FFMAX(a->length, b->length);
01328     int i;
01329     SwsVector *vec= sws_getConstVec(0.0, length);
01330 
01331     if (!vec)
01332         return NULL;
01333 
01334     for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
01335     for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
01336 
01337     return vec;
01338 }
01339 
01340 /* shift left / or right if "shift" is negative */
01341 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
01342 {
01343     int length= a->length + FFABS(shift)*2;
01344     int i;
01345     SwsVector *vec= sws_getConstVec(0.0, length);
01346 
01347     if (!vec)
01348         return NULL;
01349 
01350     for (i=0; i<a->length; i++) {
01351         vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
01352     }
01353 
01354     return vec;
01355 }
01356 
01357 void sws_shiftVec(SwsVector *a, int shift)
01358 {
01359     SwsVector *shifted= sws_getShiftedVec(a, shift);
01360     av_free(a->coeff);
01361     a->coeff= shifted->coeff;
01362     a->length= shifted->length;
01363     av_free(shifted);
01364 }
01365 
01366 void sws_addVec(SwsVector *a, SwsVector *b)
01367 {
01368     SwsVector *sum= sws_sumVec(a, b);
01369     av_free(a->coeff);
01370     a->coeff= sum->coeff;
01371     a->length= sum->length;
01372     av_free(sum);
01373 }
01374 
01375 void sws_subVec(SwsVector *a, SwsVector *b)
01376 {
01377     SwsVector *diff= sws_diffVec(a, b);
01378     av_free(a->coeff);
01379     a->coeff= diff->coeff;
01380     a->length= diff->length;
01381     av_free(diff);
01382 }
01383 
01384 void sws_convVec(SwsVector *a, SwsVector *b)
01385 {
01386     SwsVector *conv= sws_getConvVec(a, b);
01387     av_free(a->coeff);
01388     a->coeff= conv->coeff;
01389     a->length= conv->length;
01390     av_free(conv);
01391 }
01392 
01393 SwsVector *sws_cloneVec(SwsVector *a)
01394 {
01395     int i;
01396     SwsVector *vec= sws_allocVec(a->length);
01397 
01398     if (!vec)
01399         return NULL;
01400 
01401     for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
01402 
01403     return vec;
01404 }
01405 
01406 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
01407 {
01408     int i;
01409     double max=0;
01410     double min=0;
01411     double range;
01412 
01413     for (i=0; i<a->length; i++)
01414         if (a->coeff[i]>max) max= a->coeff[i];
01415 
01416     for (i=0; i<a->length; i++)
01417         if (a->coeff[i]<min) min= a->coeff[i];
01418 
01419     range= max - min;
01420 
01421     for (i=0; i<a->length; i++) {
01422         int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
01423         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
01424         for (;x>0; x--) av_log(log_ctx, log_level, " ");
01425         av_log(log_ctx, log_level, "|\n");
01426     }
01427 }
01428 
01429 void sws_freeVec(SwsVector *a)
01430 {
01431     if (!a) return;
01432     av_freep(&a->coeff);
01433     a->length=0;
01434     av_free(a);
01435 }
01436 
01437 void sws_freeFilter(SwsFilter *filter)
01438 {
01439     if (!filter) return;
01440 
01441     if (filter->lumH) sws_freeVec(filter->lumH);
01442     if (filter->lumV) sws_freeVec(filter->lumV);
01443     if (filter->chrH) sws_freeVec(filter->chrH);
01444     if (filter->chrV) sws_freeVec(filter->chrV);
01445     av_free(filter);
01446 }
01447 
01448 void sws_freeContext(SwsContext *c)
01449 {
01450     int i;
01451     if (!c) return;
01452 
01453     if (c->lumPixBuf) {
01454         for (i=0; i<c->vLumBufSize; i++)
01455             av_freep(&c->lumPixBuf[i]);
01456         av_freep(&c->lumPixBuf);
01457     }
01458 
01459     if (c->chrUPixBuf) {
01460         for (i=0; i<c->vChrBufSize; i++)
01461             av_freep(&c->chrUPixBuf[i]);
01462         av_freep(&c->chrUPixBuf);
01463         av_freep(&c->chrVPixBuf);
01464     }
01465 
01466     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
01467         for (i=0; i<c->vLumBufSize; i++)
01468             av_freep(&c->alpPixBuf[i]);
01469         av_freep(&c->alpPixBuf);
01470     }
01471 
01472     av_freep(&c->vLumFilter);
01473     av_freep(&c->vChrFilter);
01474     av_freep(&c->hLumFilter);
01475     av_freep(&c->hChrFilter);
01476 #if HAVE_ALTIVEC
01477     av_freep(&c->vYCoeffsBank);
01478     av_freep(&c->vCCoeffsBank);
01479 #endif
01480 
01481     av_freep(&c->vLumFilterPos);
01482     av_freep(&c->vChrFilterPos);
01483     av_freep(&c->hLumFilterPos);
01484     av_freep(&c->hChrFilterPos);
01485 
01486 #if HAVE_MMX
01487 #ifdef MAP_ANONYMOUS
01488     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
01489     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
01490 #elif HAVE_VIRTUALALLOC
01491     if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
01492     if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
01493 #else
01494     av_free(c->lumMmx2FilterCode);
01495     av_free(c->chrMmx2FilterCode);
01496 #endif
01497     c->lumMmx2FilterCode=NULL;
01498     c->chrMmx2FilterCode=NULL;
01499 #endif /* HAVE_MMX */
01500 
01501     av_freep(&c->yuvTable);
01502     av_freep(&c->formatConvBuffer);
01503 
01504     av_free(c);
01505 }
01506 
01507 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
01508                                         int srcW, int srcH, enum PixelFormat srcFormat,
01509                                         int dstW, int dstH, enum PixelFormat dstFormat, int flags,
01510                                         SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
01511 {
01512     static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
01513 
01514     if (!param)
01515         param = default_param;
01516 
01517     if (context &&
01518         (context->srcW      != srcW      ||
01519          context->srcH      != srcH      ||
01520          context->srcFormat != srcFormat ||
01521          context->dstW      != dstW      ||
01522          context->dstH      != dstH      ||
01523          context->dstFormat != dstFormat ||
01524          context->flags     != flags     ||
01525          context->param[0]  != param[0]  ||
01526          context->param[1]  != param[1])) {
01527         sws_freeContext(context);
01528         context = NULL;
01529     }
01530 
01531     if (!context) {
01532         if (!(context = sws_alloc_context()))
01533             return NULL;
01534         context->srcW      = srcW;
01535         context->srcH      = srcH;
01536         context->srcRange  = handle_jpeg(&srcFormat);
01537         context->srcFormat = srcFormat;
01538         context->dstW      = dstW;
01539         context->dstH      = dstH;
01540         context->dstRange  = handle_jpeg(&dstFormat);
01541         context->dstFormat = dstFormat;
01542         context->flags     = flags;
01543         context->param[0]  = param[0];
01544         context->param[1]  = param[1];
01545         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);
01546         if (sws_init_context(context, srcFilter, dstFilter) < 0) {
01547             sws_freeContext(context);
01548             return NULL;
01549         }
01550     }
01551     return context;
01552 }
01553