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 #include "config.h"
00022 
00023 #define _SVID_SOURCE // needed for MAP_ANONYMOUS
00024 #define _DARWIN_C_SOURCE // needed for MAP_ANON
00025 #include <inttypes.h>
00026 #include <math.h>
00027 #include <stdio.h>
00028 #include <string.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 
00040 #include "libavutil/attributes.h"
00041 #include "libavutil/avassert.h"
00042 #include "libavutil/avutil.h"
00043 #include "libavutil/bswap.h"
00044 #include "libavutil/cpu.h"
00045 #include "libavutil/intreadwrite.h"
00046 #include "libavutil/mathematics.h"
00047 #include "libavutil/opt.h"
00048 #include "libavutil/pixdesc.h"
00049 #include "libavutil/x86/asm.h"
00050 #include "libavutil/x86/cpu.h"
00051 #include "rgb2rgb.h"
00052 #include "swscale.h"
00053 #include "swscale_internal.h"
00054 
00055 unsigned swscale_version(void)
00056 {
00057     av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
00058     return LIBSWSCALE_VERSION_INT;
00059 }
00060 
00061 const char *swscale_configuration(void)
00062 {
00063     return FFMPEG_CONFIGURATION;
00064 }
00065 
00066 const char *swscale_license(void)
00067 {
00068 #define LICENSE_PREFIX "libswscale license: "
00069     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
00070 }
00071 
00072 #define RET 0xC3 // near return opcode for x86
00073 
00074 typedef struct FormatEntry {
00075     int is_supported_in, is_supported_out;
00076 } FormatEntry;
00077 
00078 static const FormatEntry format_entries[PIX_FMT_NB] = {
00079     [PIX_FMT_YUV420P]     = { 1, 1 },
00080     [PIX_FMT_YUYV422]     = { 1, 1 },
00081     [PIX_FMT_RGB24]       = { 1, 1 },
00082     [PIX_FMT_BGR24]       = { 1, 1 },
00083     [PIX_FMT_YUV422P]     = { 1, 1 },
00084     [PIX_FMT_YUV444P]     = { 1, 1 },
00085     [PIX_FMT_YUV410P]     = { 1, 1 },
00086     [PIX_FMT_YUV411P]     = { 1, 1 },
00087     [PIX_FMT_GRAY8]       = { 1, 1 },
00088     [PIX_FMT_MONOWHITE]   = { 1, 1 },
00089     [PIX_FMT_MONOBLACK]   = { 1, 1 },
00090     [PIX_FMT_PAL8]        = { 1, 0 },
00091     [PIX_FMT_YUVJ420P]    = { 1, 1 },
00092     [PIX_FMT_YUVJ422P]    = { 1, 1 },
00093     [PIX_FMT_YUVJ444P]    = { 1, 1 },
00094     [PIX_FMT_UYVY422]     = { 1, 1 },
00095     [PIX_FMT_UYYVYY411]   = { 0, 0 },
00096     [PIX_FMT_BGR8]        = { 1, 1 },
00097     [PIX_FMT_BGR4]        = { 0, 1 },
00098     [PIX_FMT_BGR4_BYTE]   = { 1, 1 },
00099     [PIX_FMT_RGB8]        = { 1, 1 },
00100     [PIX_FMT_RGB4]        = { 0, 1 },
00101     [PIX_FMT_RGB4_BYTE]   = { 1, 1 },
00102     [PIX_FMT_NV12]        = { 1, 1 },
00103     [PIX_FMT_NV21]        = { 1, 1 },
00104     [PIX_FMT_ARGB]        = { 1, 1 },
00105     [PIX_FMT_RGBA]        = { 1, 1 },
00106     [PIX_FMT_ABGR]        = { 1, 1 },
00107     [PIX_FMT_BGRA]        = { 1, 1 },
00108     [PIX_FMT_0RGB]        = { 1, 1 },
00109     [PIX_FMT_RGB0]        = { 1, 1 },
00110     [PIX_FMT_0BGR]        = { 1, 1 },
00111     [PIX_FMT_BGR0]        = { 1, 1 },
00112     [PIX_FMT_GRAY16BE]    = { 1, 1 },
00113     [PIX_FMT_GRAY16LE]    = { 1, 1 },
00114     [PIX_FMT_YUV440P]     = { 1, 1 },
00115     [PIX_FMT_YUVJ440P]    = { 1, 1 },
00116     [PIX_FMT_YUVA420P]    = { 1, 1 },
00117     [PIX_FMT_YUVA422P]    = { 1, 1 },
00118     [PIX_FMT_YUVA444P]    = { 1, 1 },
00119     [PIX_FMT_RGB48BE]     = { 1, 1 },
00120     [PIX_FMT_RGB48LE]     = { 1, 1 },
00121     [PIX_FMT_RGBA64BE]    = { 1, 0 },
00122     [PIX_FMT_RGBA64LE]    = { 1, 0 },
00123     [PIX_FMT_RGB565BE]    = { 1, 1 },
00124     [PIX_FMT_RGB565LE]    = { 1, 1 },
00125     [PIX_FMT_RGB555BE]    = { 1, 1 },
00126     [PIX_FMT_RGB555LE]    = { 1, 1 },
00127     [PIX_FMT_BGR565BE]    = { 1, 1 },
00128     [PIX_FMT_BGR565LE]    = { 1, 1 },
00129     [PIX_FMT_BGR555BE]    = { 1, 1 },
00130     [PIX_FMT_BGR555LE]    = { 1, 1 },
00131     [PIX_FMT_YUV420P16LE] = { 1, 1 },
00132     [PIX_FMT_YUV420P16BE] = { 1, 1 },
00133     [PIX_FMT_YUV422P16LE] = { 1, 1 },
00134     [PIX_FMT_YUV422P16BE] = { 1, 1 },
00135     [PIX_FMT_YUV444P16LE] = { 1, 1 },
00136     [PIX_FMT_YUV444P16BE] = { 1, 1 },
00137     [PIX_FMT_RGB444LE]    = { 1, 1 },
00138     [PIX_FMT_RGB444BE]    = { 1, 1 },
00139     [PIX_FMT_BGR444LE]    = { 1, 1 },
00140     [PIX_FMT_BGR444BE]    = { 1, 1 },
00141     [PIX_FMT_Y400A]       = { 1, 0 },
00142     [PIX_FMT_BGR48BE]     = { 1, 1 },
00143     [PIX_FMT_BGR48LE]     = { 1, 1 },
00144     [PIX_FMT_BGRA64BE]    = { 0, 0 },
00145     [PIX_FMT_BGRA64LE]    = { 0, 0 },
00146     [PIX_FMT_YUV420P9BE]  = { 1, 1 },
00147     [PIX_FMT_YUV420P9LE]  = { 1, 1 },
00148     [PIX_FMT_YUV420P10BE] = { 1, 1 },
00149     [PIX_FMT_YUV420P10LE] = { 1, 1 },
00150     [PIX_FMT_YUV420P12BE] = { 1, 1 },
00151     [PIX_FMT_YUV420P12LE] = { 1, 1 },
00152     [PIX_FMT_YUV420P14BE] = { 1, 1 },
00153     [PIX_FMT_YUV420P14LE] = { 1, 1 },
00154     [PIX_FMT_YUV422P9BE]  = { 1, 1 },
00155     [PIX_FMT_YUV422P9LE]  = { 1, 1 },
00156     [PIX_FMT_YUV422P10BE] = { 1, 1 },
00157     [PIX_FMT_YUV422P10LE] = { 1, 1 },
00158     [PIX_FMT_YUV422P12BE] = { 1, 1 },
00159     [PIX_FMT_YUV422P12LE] = { 1, 1 },
00160     [PIX_FMT_YUV422P14BE] = { 1, 1 },
00161     [PIX_FMT_YUV422P14LE] = { 1, 1 },
00162     [PIX_FMT_YUV444P9BE]  = { 1, 1 },
00163     [PIX_FMT_YUV444P9LE]  = { 1, 1 },
00164     [PIX_FMT_YUV444P10BE] = { 1, 1 },
00165     [PIX_FMT_YUV444P10LE] = { 1, 1 },
00166     [PIX_FMT_YUV444P12BE] = { 1, 1 },
00167     [PIX_FMT_YUV444P12LE] = { 1, 1 },
00168     [PIX_FMT_YUV444P14BE] = { 1, 1 },
00169     [PIX_FMT_YUV444P14LE] = { 1, 1 },
00170     [PIX_FMT_GBRP]        = { 1, 0 },
00171     [PIX_FMT_GBRP9LE]     = { 1, 0 },
00172     [PIX_FMT_GBRP9BE]     = { 1, 0 },
00173     [PIX_FMT_GBRP10LE]    = { 1, 0 },
00174     [PIX_FMT_GBRP10BE]    = { 1, 0 },
00175     [PIX_FMT_GBRP12LE]    = { 1, 0 },
00176     [PIX_FMT_GBRP12BE]    = { 1, 0 },
00177     [PIX_FMT_GBRP14LE]    = { 1, 0 },
00178     [PIX_FMT_GBRP14BE]    = { 1, 0 },
00179     [PIX_FMT_GBRP16LE]    = { 1, 0 },
00180     [PIX_FMT_GBRP16BE]    = { 1, 0 },
00181 };
00182 
00183 int sws_isSupportedInput(enum PixelFormat pix_fmt)
00184 {
00185     return (unsigned)pix_fmt < PIX_FMT_NB ?
00186            format_entries[pix_fmt].is_supported_in : 0;
00187 }
00188 
00189 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
00190 {
00191     return (unsigned)pix_fmt < PIX_FMT_NB ?
00192            format_entries[pix_fmt].is_supported_out : 0;
00193 }
00194 
00195 extern const int32_t ff_yuv2rgb_coeffs[8][4];
00196 
00197 #if FF_API_SWS_FORMAT_NAME
00198 const char *sws_format_name(enum PixelFormat format)
00199 {
00200     return av_get_pix_fmt_name(format);
00201 }
00202 #endif
00203 
00204 static double getSplineCoeff(double a, double b, double c, double d,
00205                              double dist)
00206 {
00207     if (dist <= 1.0)
00208         return ((d * dist + c) * dist + b) * dist + a;
00209     else
00210         return getSplineCoeff(0.0,
00211                                b + 2.0 * c + 3.0 * d,
00212                                c + 3.0 * d,
00213                               -b - 3.0 * c - 6.0 * d,
00214                               dist - 1.0);
00215 }
00216 
00217 static int initFilter(int16_t **outFilter, int32_t **filterPos,
00218                       int *outFilterSize, int xInc, int srcW, int dstW,
00219                       int filterAlign, int one, int flags, int cpu_flags,
00220                       SwsVector *srcFilter, SwsVector *dstFilter,
00221                       double param[2])
00222 {
00223     int i;
00224     int filterSize;
00225     int filter2Size;
00226     int minFilterSize;
00227     int64_t *filter    = NULL;
00228     int64_t *filter2   = NULL;
00229     const int64_t fone = 1LL << 54;
00230     int ret            = -1;
00231 
00232     emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
00233 
00234     // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
00235     FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
00236 
00237     if (FFABS(xInc - 0x10000) < 10) { // unscaled
00238         int i;
00239         filterSize = 1;
00240         FF_ALLOCZ_OR_GOTO(NULL, filter,
00241                           dstW * sizeof(*filter) * filterSize, fail);
00242 
00243         for (i = 0; i < dstW; i++) {
00244             filter[i * filterSize] = fone;
00245             (*filterPos)[i]        = i;
00246         }
00247     } else if (flags & SWS_POINT) { // lame looking point sampling mode
00248         int i;
00249         int64_t xDstInSrc;
00250         filterSize = 1;
00251         FF_ALLOC_OR_GOTO(NULL, filter,
00252                          dstW * sizeof(*filter) * filterSize, fail);
00253 
00254         xDstInSrc = xInc / 2 - 0x8000;
00255         for (i = 0; i < dstW; i++) {
00256             int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
00257 
00258             (*filterPos)[i] = xx;
00259             filter[i]       = fone;
00260             xDstInSrc      += xInc;
00261         }
00262     } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
00263                (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
00264         int i;
00265         int64_t xDstInSrc;
00266         filterSize = 2;
00267         FF_ALLOC_OR_GOTO(NULL, filter,
00268                          dstW * sizeof(*filter) * filterSize, fail);
00269 
00270         xDstInSrc = xInc / 2 - 0x8000;
00271         for (i = 0; i < dstW; i++) {
00272             int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
00273             int j;
00274 
00275             (*filterPos)[i] = xx;
00276             // bilinear upscale / linear interpolate / area averaging
00277             for (j = 0; j < filterSize; j++) {
00278                 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
00279                 if (coeff < 0)
00280                     coeff = 0;
00281                 filter[i * filterSize + j] = coeff;
00282                 xx++;
00283             }
00284             xDstInSrc += xInc;
00285         }
00286     } else {
00287         int64_t xDstInSrc;
00288         int sizeFactor;
00289 
00290         if (flags & SWS_BICUBIC)
00291             sizeFactor = 4;
00292         else if (flags & SWS_X)
00293             sizeFactor = 8;
00294         else if (flags & SWS_AREA)
00295             sizeFactor = 1;     // downscale only, for upscale it is bilinear
00296         else if (flags & SWS_GAUSS)
00297             sizeFactor = 8;     // infinite ;)
00298         else if (flags & SWS_LANCZOS)
00299             sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
00300         else if (flags & SWS_SINC)
00301             sizeFactor = 20;    // infinite ;)
00302         else if (flags & SWS_SPLINE)
00303             sizeFactor = 20;    // infinite ;)
00304         else if (flags & SWS_BILINEAR)
00305             sizeFactor = 2;
00306         else {
00307             av_assert0(0);
00308         }
00309 
00310         if (xInc <= 1 << 16)
00311             filterSize = 1 + sizeFactor;    // upscale
00312         else
00313             filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
00314 
00315         filterSize = FFMIN(filterSize, srcW - 2);
00316         filterSize = FFMAX(filterSize, 1);
00317 
00318         FF_ALLOC_OR_GOTO(NULL, filter,
00319                          dstW * sizeof(*filter) * filterSize, fail);
00320 
00321         xDstInSrc = xInc - 0x10000;
00322         for (i = 0; i < dstW; i++) {
00323             int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
00324             int j;
00325             (*filterPos)[i] = xx;
00326             for (j = 0; j < filterSize; j++) {
00327                 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
00328                 double floatd;
00329                 int64_t coeff;
00330 
00331                 if (xInc > 1 << 16)
00332                     d = d * dstW / srcW;
00333                 floatd = d * (1.0 / (1 << 30));
00334 
00335                 if (flags & SWS_BICUBIC) {
00336                     int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] :   0) * (1 << 24);
00337                     int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
00338 
00339                     if (d >= 1LL << 31) {
00340                         coeff = 0.0;
00341                     } else {
00342                         int64_t dd  = (d  * d) >> 30;
00343                         int64_t ddd = (dd * d) >> 30;
00344 
00345                         if (d < 1LL << 30)
00346                             coeff =  (12 * (1 << 24) -  9 * B - 6 * C) * ddd +
00347                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +
00348                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);
00349                         else
00350                             coeff =      (-B -  6 * C) * ddd +
00351                                       (6 * B + 30 * C) * dd  +
00352                                     (-12 * B - 48 * C) * d   +
00353                                       (8 * B + 24 * C) * (1 << 30);
00354                     }
00355                     coeff *= fone >> (30 + 24);
00356                 }
00357 #if 0
00358                 else if (flags & SWS_X) {
00359                     double p  = param ? param * 0.01 : 0.3;
00360                     coeff     = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
00361                     coeff    *= pow(2.0, -p * d * d);
00362                 }
00363 #endif
00364                 else if (flags & SWS_X) {
00365                     double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
00366                     double c;
00367 
00368                     if (floatd < 1.0)
00369                         c = cos(floatd * M_PI);
00370                     else
00371                         c = -1.0;
00372                     if (c < 0.0)
00373                         c = -pow(-c, A);
00374                     else
00375                         c = pow(c, A);
00376                     coeff = (c * 0.5 + 0.5) * fone;
00377                 } else if (flags & SWS_AREA) {
00378                     int64_t d2 = d - (1 << 29);
00379                     if (d2 * xInc < -(1LL << (29 + 16)))
00380                         coeff = 1.0 * (1LL << (30 + 16));
00381                     else if (d2 * xInc < (1LL << (29 + 16)))
00382                         coeff = -d2 * xInc + (1LL << (29 + 16));
00383                     else
00384                         coeff = 0.0;
00385                     coeff *= fone >> (30 + 16);
00386                 } else if (flags & SWS_GAUSS) {
00387                     double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
00388                     coeff = (pow(2.0, -p * floatd * floatd)) * fone;
00389                 } else if (flags & SWS_SINC) {
00390                     coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
00391                 } else if (flags & SWS_LANCZOS) {
00392                     double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
00393                     coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
00394                              (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
00395                     if (floatd > p)
00396                         coeff = 0;
00397                 } else if (flags & SWS_BILINEAR) {
00398                     coeff = (1 << 30) - d;
00399                     if (coeff < 0)
00400                         coeff = 0;
00401                     coeff *= fone >> 30;
00402                 } else if (flags & SWS_SPLINE) {
00403                     double p = -2.196152422706632;
00404                     coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
00405                 } else {
00406                     av_assert0(0);
00407                 }
00408 
00409                 filter[i * filterSize + j] = coeff;
00410                 xx++;
00411             }
00412             xDstInSrc += 2 * xInc;
00413         }
00414     }
00415 
00416     /* apply src & dst Filter to filter -> filter2
00417      * av_free(filter);
00418      */
00419     av_assert0(filterSize > 0);
00420     filter2Size = filterSize;
00421     if (srcFilter)
00422         filter2Size += srcFilter->length - 1;
00423     if (dstFilter)
00424         filter2Size += dstFilter->length - 1;
00425     av_assert0(filter2Size > 0);
00426     FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);
00427 
00428     for (i = 0; i < dstW; i++) {
00429         int j, k;
00430 
00431         if (srcFilter) {
00432             for (k = 0; k < srcFilter->length; k++) {
00433                 for (j = 0; j < filterSize; j++)
00434                     filter2[i * filter2Size + k + j] +=
00435                         srcFilter->coeff[k] * filter[i * filterSize + j];
00436             }
00437         } else {
00438             for (j = 0; j < filterSize; j++)
00439                 filter2[i * filter2Size + j] = filter[i * filterSize + j];
00440         }
00441         // FIXME dstFilter
00442 
00443         (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
00444     }
00445     av_freep(&filter);
00446 
00447     /* try to reduce the filter-size (step1 find size and shift left) */
00448     // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
00449     minFilterSize = 0;
00450     for (i = dstW - 1; i >= 0; i--) {
00451         int min = filter2Size;
00452         int j;
00453         int64_t cutOff = 0.0;
00454 
00455         /* get rid of near zero elements on the left by shifting left */
00456         for (j = 0; j < filter2Size; j++) {
00457             int k;
00458             cutOff += FFABS(filter2[i * filter2Size]);
00459 
00460             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
00461                 break;
00462 
00463             /* preserve monotonicity because the core can't handle the
00464              * filter otherwise */
00465             if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
00466                 break;
00467 
00468             // move filter coefficients left
00469             for (k = 1; k < filter2Size; k++)
00470                 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
00471             filter2[i * filter2Size + k - 1] = 0;
00472             (*filterPos)[i]++;
00473         }
00474 
00475         cutOff = 0;
00476         /* count near zeros on the right */
00477         for (j = filter2Size - 1; j > 0; j--) {
00478             cutOff += FFABS(filter2[i * filter2Size + j]);
00479 
00480             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
00481                 break;
00482             min--;
00483         }
00484 
00485         if (min > minFilterSize)
00486             minFilterSize = min;
00487     }
00488 
00489     if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
00490         // we can handle the special case 4, so we don't want to go the full 8
00491         if (minFilterSize < 5)
00492             filterAlign = 4;
00493 
00494         /* We really don't want to waste our time doing useless computation, so
00495          * fall back on the scalar C code for very small filters.
00496          * Vectorizing is worth it only if you have a decent-sized vector. */
00497         if (minFilterSize < 3)
00498             filterAlign = 1;
00499     }
00500 
00501     if (INLINE_MMX(cpu_flags)) {
00502         // special case for unscaled vertical filtering
00503         if (minFilterSize == 1 && filterAlign == 2)
00504             filterAlign = 1;
00505     }
00506 
00507     av_assert0(minFilterSize > 0);
00508     filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
00509     av_assert0(filterSize > 0);
00510     filter = av_malloc(filterSize * dstW * sizeof(*filter));
00511     if (filterSize >= MAX_FILTER_SIZE * 16 /
00512                       ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
00513         goto fail;
00514     *outFilterSize = filterSize;
00515 
00516     if (flags & SWS_PRINT_INFO)
00517         av_log(NULL, AV_LOG_VERBOSE,
00518                "SwScaler: reducing / aligning filtersize %d -> %d\n",
00519                filter2Size, filterSize);
00520     /* try to reduce the filter-size (step2 reduce it) */
00521     for (i = 0; i < dstW; i++) {
00522         int j;
00523 
00524         for (j = 0; j < filterSize; j++) {
00525             if (j >= filter2Size)
00526                 filter[i * filterSize + j] = 0;
00527             else
00528                 filter[i * filterSize + j] = filter2[i * filter2Size + j];
00529             if ((flags & SWS_BITEXACT) && j >= minFilterSize)
00530                 filter[i * filterSize + j] = 0;
00531         }
00532     }
00533 
00534     // FIXME try to align filterPos if possible
00535 
00536     // fix borders
00537     for (i = 0; i < dstW; i++) {
00538         int j;
00539         if ((*filterPos)[i] < 0) {
00540             // move filter coefficients left to compensate for filterPos
00541             for (j = 1; j < filterSize; j++) {
00542                 int left = FFMAX(j + (*filterPos)[i], 0);
00543                 filter[i * filterSize + left] += filter[i * filterSize + j];
00544                 filter[i * filterSize + j]     = 0;
00545             }
00546             (*filterPos)[i]= 0;
00547         }
00548 
00549         if ((*filterPos)[i] + filterSize > srcW) {
00550             int shift = (*filterPos)[i] + filterSize - srcW;
00551             // move filter coefficients right to compensate for filterPos
00552             for (j = filterSize - 2; j >= 0; j--) {
00553                 int right = FFMIN(j + shift, filterSize - 1);
00554                 filter[i * filterSize + right] += filter[i * filterSize + j];
00555                 filter[i * filterSize + j]      = 0;
00556             }
00557             (*filterPos)[i]= srcW - filterSize;
00558         }
00559     }
00560 
00561     // Note the +1 is for the MMX scaler which reads over the end
00562     /* align at 16 for AltiVec (needed by hScale_altivec_real) */
00563     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
00564                       *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
00565 
00566     /* normalize & store in outFilter */
00567     for (i = 0; i < dstW; i++) {
00568         int j;
00569         int64_t error = 0;
00570         int64_t sum   = 0;
00571 
00572         for (j = 0; j < filterSize; j++) {
00573             sum += filter[i * filterSize + j];
00574         }
00575         sum = (sum + one / 2) / one;
00576         for (j = 0; j < *outFilterSize; j++) {
00577             int64_t v = filter[i * filterSize + j] + error;
00578             int intV  = ROUNDED_DIV(v, sum);
00579             (*outFilter)[i * (*outFilterSize) + j] = intV;
00580             error                                  = v - intV * sum;
00581         }
00582     }
00583 
00584     (*filterPos)[dstW + 0] =
00585     (*filterPos)[dstW + 1] =
00586     (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
00587                                                       * read over the end */
00588     for (i = 0; i < *outFilterSize; i++) {
00589         int k = (dstW - 1) * (*outFilterSize) + i;
00590         (*outFilter)[k + 1 * (*outFilterSize)] =
00591         (*outFilter)[k + 2 * (*outFilterSize)] =
00592         (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
00593     }
00594 
00595     ret = 0;
00596 
00597 fail:
00598     av_free(filter);
00599     av_free(filter2);
00600     return ret;
00601 }
00602 
00603 #if HAVE_MMXEXT_INLINE
00604 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode,
00605                            int16_t *filter, int32_t *filterPos, int numSplits)
00606 {
00607     uint8_t *fragmentA;
00608     x86_reg imm8OfPShufW1A;
00609     x86_reg imm8OfPShufW2A;
00610     x86_reg fragmentLengthA;
00611     uint8_t *fragmentB;
00612     x86_reg imm8OfPShufW1B;
00613     x86_reg imm8OfPShufW2B;
00614     x86_reg fragmentLengthB;
00615     int fragmentPos;
00616 
00617     int xpos, i;
00618 
00619     // create an optimized horizontal scaling routine
00620     /* This scaler is made of runtime-generated MMX2 code using specially tuned
00621      * pshufw instructions. For every four output pixels, if four input pixels
00622      * are enough for the fast bilinear scaling, then a chunk of fragmentB is
00623      * used. If five input pixels are needed, then a chunk of fragmentA is used.
00624      */
00625 
00626     // code fragment
00627 
00628     __asm__ volatile (
00629         "jmp                         9f                 \n\t"
00630         // Begin
00631         "0:                                             \n\t"
00632         "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
00633         "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
00634         "movd   1(%%"REG_c", %%"REG_S"), %%mm1          \n\t"
00635         "punpcklbw                %%mm7, %%mm1          \n\t"
00636         "punpcklbw                %%mm7, %%mm0          \n\t"
00637         "pshufw                   $0xFF, %%mm1, %%mm1   \n\t"
00638         "1:                                             \n\t"
00639         "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
00640         "2:                                             \n\t"
00641         "psubw                    %%mm1, %%mm0          \n\t"
00642         "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
00643         "pmullw                   %%mm3, %%mm0          \n\t"
00644         "psllw                       $7, %%mm1          \n\t"
00645         "paddw                    %%mm1, %%mm0          \n\t"
00646 
00647         "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"
00648 
00649         "add                         $8, %%"REG_a"      \n\t"
00650         // End
00651         "9:                                             \n\t"
00652         // "int $3                                         \n\t"
00653         "lea       " LOCAL_MANGLE(0b) ", %0             \n\t"
00654         "lea       " LOCAL_MANGLE(1b) ", %1             \n\t"
00655         "lea       " LOCAL_MANGLE(2b) ", %2             \n\t"
00656         "dec                         %1                 \n\t"
00657         "dec                         %2                 \n\t"
00658         "sub                         %0, %1             \n\t"
00659         "sub                         %0, %2             \n\t"
00660         "lea       " LOCAL_MANGLE(9b) ", %3             \n\t"
00661         "sub                         %0, %3             \n\t"
00662 
00663 
00664         : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
00665           "=r" (fragmentLengthA)
00666         );
00667 
00668     __asm__ volatile (
00669         "jmp                         9f                 \n\t"
00670         // Begin
00671         "0:                                             \n\t"
00672         "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
00673         "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
00674         "punpcklbw                %%mm7, %%mm0          \n\t"
00675         "pshufw                   $0xFF, %%mm0, %%mm1   \n\t"
00676         "1:                                             \n\t"
00677         "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
00678         "2:                                             \n\t"
00679         "psubw                    %%mm1, %%mm0          \n\t"
00680         "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
00681         "pmullw                   %%mm3, %%mm0          \n\t"
00682         "psllw                       $7, %%mm1          \n\t"
00683         "paddw                    %%mm1, %%mm0          \n\t"
00684 
00685         "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"
00686 
00687         "add                         $8, %%"REG_a"      \n\t"
00688         // End
00689         "9:                                             \n\t"
00690         // "int                       $3                   \n\t"
00691         "lea       " LOCAL_MANGLE(0b) ", %0             \n\t"
00692         "lea       " LOCAL_MANGLE(1b) ", %1             \n\t"
00693         "lea       " LOCAL_MANGLE(2b) ", %2             \n\t"
00694         "dec                         %1                 \n\t"
00695         "dec                         %2                 \n\t"
00696         "sub                         %0, %1             \n\t"
00697         "sub                         %0, %2             \n\t"
00698         "lea       " LOCAL_MANGLE(9b) ", %3             \n\t"
00699         "sub                         %0, %3             \n\t"
00700 
00701 
00702         : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
00703           "=r" (fragmentLengthB)
00704         );
00705 
00706     xpos        = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
00707     fragmentPos = 0;
00708 
00709     for (i = 0; i < dstW / numSplits; i++) {
00710         int xx = xpos >> 16;
00711 
00712         if ((i & 3) == 0) {
00713             int a                  = 0;
00714             int b                  = ((xpos + xInc) >> 16) - xx;
00715             int c                  = ((xpos + xInc * 2) >> 16) - xx;
00716             int d                  = ((xpos + xInc * 3) >> 16) - xx;
00717             int inc                = (d + 1 < 4);
00718             uint8_t *fragment      = (d + 1 < 4) ? fragmentB : fragmentA;
00719             x86_reg imm8OfPShufW1  = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;
00720             x86_reg imm8OfPShufW2  = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;
00721             x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;
00722             int maxShift           = 3 - (d + inc);
00723             int shift              = 0;
00724 
00725             if (filterCode) {
00726                 filter[i]        = ((xpos              & 0xFFFF) ^ 0xFFFF) >> 9;
00727                 filter[i + 1]    = (((xpos + xInc)     & 0xFFFF) ^ 0xFFFF) >> 9;
00728                 filter[i + 2]    = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
00729                 filter[i + 3]    = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
00730                 filterPos[i / 2] = xx;
00731 
00732                 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
00733 
00734                 filterCode[fragmentPos + imm8OfPShufW1] =  (a + inc)       |
00735                                                           ((b + inc) << 2) |
00736                                                           ((c + inc) << 4) |
00737                                                           ((d + inc) << 6);
00738                 filterCode[fragmentPos + imm8OfPShufW2] =  a | (b << 2) |
00739                                                                (c << 4) |
00740                                                                (d << 6);
00741 
00742                 if (i + 4 - inc >= dstW)
00743                     shift = maxShift;               // avoid overread
00744                 else if ((filterPos[i / 2] & 3) <= maxShift)
00745                     shift = filterPos[i / 2] & 3;   // align
00746 
00747                 if (shift && i >= shift) {
00748                     filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
00749                     filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
00750                     filterPos[i / 2]                        -= shift;
00751                 }
00752             }
00753 
00754             fragmentPos += fragmentLength;
00755 
00756             if (filterCode)
00757                 filterCode[fragmentPos] = RET;
00758         }
00759         xpos += xInc;
00760     }
00761     if (filterCode)
00762         filterPos[((i / 2) + 1) & (~1)] = xpos >> 16;  // needed to jump to the next part
00763 
00764     return fragmentPos + 1;
00765 }
00766 #endif /* HAVE_MMXEXT_INLINE */
00767 
00768 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
00769 {
00770     *h = av_pix_fmt_descriptors[format].log2_chroma_w;
00771     *v = av_pix_fmt_descriptors[format].log2_chroma_h;
00772 }
00773 
00774 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
00775                              int srcRange, const int table[4], int dstRange,
00776                              int brightness, int contrast, int saturation)
00777 {
00778     memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
00779     memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);
00780 
00781     c->brightness = brightness;
00782     c->contrast   = contrast;
00783     c->saturation = saturation;
00784     c->srcRange   = srcRange;
00785     c->dstRange   = dstRange;
00786     if (isYUV(c->dstFormat) || isGray(c->dstFormat))
00787         return -1;
00788 
00789     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
00790     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
00791 
00792     ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
00793                              contrast, saturation);
00794     // FIXME factorize
00795 
00796     if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
00797         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,
00798                                        contrast, saturation);
00799     return 0;
00800 }
00801 
00802 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
00803                              int *srcRange, int **table, int *dstRange,
00804                              int *brightness, int *contrast, int *saturation)
00805 {
00806     if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat))
00807         return -1;
00808 
00809     *inv_table  = c->srcColorspaceTable;
00810     *table      = c->dstColorspaceTable;
00811     *srcRange   = c->srcRange;
00812     *dstRange   = c->dstRange;
00813     *brightness = c->brightness;
00814     *contrast   = c->contrast;
00815     *saturation = c->saturation;
00816 
00817     return 0;
00818 }
00819 
00820 static int handle_jpeg(enum PixelFormat *format)
00821 {
00822     switch (*format) {
00823     case PIX_FMT_YUVJ420P:
00824         *format = PIX_FMT_YUV420P;
00825         return 1;
00826     case PIX_FMT_YUVJ422P:
00827         *format = PIX_FMT_YUV422P;
00828         return 1;
00829     case PIX_FMT_YUVJ444P:
00830         *format = PIX_FMT_YUV444P;
00831         return 1;
00832     case PIX_FMT_YUVJ440P:
00833         *format = PIX_FMT_YUV440P;
00834         return 1;
00835     default:
00836         return 0;
00837     }
00838 }
00839 
00840 static int handle_0alpha(enum PixelFormat *format)
00841 {
00842     switch (*format) {
00843     case PIX_FMT_0BGR    : *format = PIX_FMT_ABGR   ; return 1;
00844     case PIX_FMT_BGR0    : *format = PIX_FMT_BGRA   ; return 4;
00845     case PIX_FMT_0RGB    : *format = PIX_FMT_ARGB   ; return 1;
00846     case PIX_FMT_RGB0    : *format = PIX_FMT_RGBA   ; return 4;
00847     default:                                          return 0;
00848     }
00849 }
00850 
00851 SwsContext *sws_alloc_context(void)
00852 {
00853     SwsContext *c = av_mallocz(sizeof(SwsContext));
00854 
00855     c->av_class = &sws_context_class;
00856     av_opt_set_defaults(c);
00857 
00858     return c;
00859 }
00860 
00861 av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
00862                              SwsFilter *dstFilter)
00863 {
00864     int i, j;
00865     int usesVFilter, usesHFilter;
00866     int unscaled;
00867     SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
00868     int srcW              = c->srcW;
00869     int srcH              = c->srcH;
00870     int dstW              = c->dstW;
00871     int dstH              = c->dstH;
00872     int dst_stride        = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
00873     int flags, cpu_flags;
00874     enum PixelFormat srcFormat = c->srcFormat;
00875     enum PixelFormat dstFormat = c->dstFormat;
00876 
00877     cpu_flags = av_get_cpu_flags();
00878     flags     = c->flags;
00879     emms_c();
00880     if (!rgb15to16)
00881         sws_rgb2rgb_init();
00882 
00883     unscaled = (srcW == dstW && srcH == dstH);
00884 
00885     handle_jpeg(&srcFormat);
00886     handle_jpeg(&dstFormat);
00887     handle_0alpha(&srcFormat);
00888     handle_0alpha(&dstFormat);
00889 
00890     if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){
00891         av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
00892         c->srcFormat= srcFormat;
00893         c->dstFormat= dstFormat;
00894     }
00895 
00896     if (!sws_isSupportedInput(srcFormat)) {
00897         av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
00898                av_get_pix_fmt_name(srcFormat));
00899         return AVERROR(EINVAL);
00900     }
00901     if (!sws_isSupportedOutput(dstFormat)) {
00902         av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
00903                av_get_pix_fmt_name(dstFormat));
00904         return AVERROR(EINVAL);
00905     }
00906 
00907     i = flags & (SWS_POINT         |
00908                  SWS_AREA          |
00909                  SWS_BILINEAR      |
00910                  SWS_FAST_BILINEAR |
00911                  SWS_BICUBIC       |
00912                  SWS_X             |
00913                  SWS_GAUSS         |
00914                  SWS_LANCZOS       |
00915                  SWS_SINC          |
00916                  SWS_SPLINE        |
00917                  SWS_BICUBLIN);
00918     if (!i || (i & (i - 1))) {
00919         av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);
00920         return AVERROR(EINVAL);
00921     }
00922     /* sanity check */
00923     if (srcW < 4 || srcH < 1 || dstW < 8 || dstH < 1) {
00924         /* FIXME check if these are enough and try to lower them after
00925          * fixing the relevant parts of the code */
00926         av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
00927                srcW, srcH, dstW, dstH);
00928         return AVERROR(EINVAL);
00929     }
00930 
00931     if (!dstFilter)
00932         dstFilter = &dummyFilter;
00933     if (!srcFilter)
00934         srcFilter = &dummyFilter;
00935 
00936     c->lumXInc      = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
00937     c->lumYInc      = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
00938     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
00939     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
00940     c->vRounder     = 4 * 0x0001000100010001ULL;
00941 
00942     usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
00943                   (srcFilter->chrV && srcFilter->chrV->length > 1) ||
00944                   (dstFilter->lumV && dstFilter->lumV->length > 1) ||
00945                   (dstFilter->chrV && dstFilter->chrV->length > 1);
00946     usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
00947                   (srcFilter->chrH && srcFilter->chrH->length > 1) ||
00948                   (dstFilter->lumH && dstFilter->lumH->length > 1) ||
00949                   (dstFilter->chrH && dstFilter->chrH->length > 1);
00950 
00951     getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
00952     getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
00953 
00954 
00955     if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
00956         if (dstW&1) {
00957             av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
00958             flags |= SWS_FULL_CHR_H_INT;
00959             c->flags = flags;
00960         }
00961     }
00962     /* reuse chroma for 2 pixels RGB/BGR unless user wants full
00963      * chroma interpolation */
00964     if (flags & SWS_FULL_CHR_H_INT &&
00965         isAnyRGB(dstFormat)        &&
00966         dstFormat != PIX_FMT_RGBA  &&
00967         dstFormat != PIX_FMT_ARGB  &&
00968         dstFormat != PIX_FMT_BGRA  &&
00969         dstFormat != PIX_FMT_ABGR  &&
00970         dstFormat != PIX_FMT_RGB24 &&
00971         dstFormat != PIX_FMT_BGR24) {
00972         av_log(c, AV_LOG_WARNING,
00973                "full chroma interpolation for destination format '%s' not yet implemented\n",
00974                av_get_pix_fmt_name(dstFormat));
00975         flags   &= ~SWS_FULL_CHR_H_INT;
00976         c->flags = flags;
00977     }
00978     if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
00979         c->chrDstHSubSample = 1;
00980 
00981     // drop some chroma lines if the user wants it
00982     c->vChrDrop          = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
00983                            SWS_SRC_V_CHR_DROP_SHIFT;
00984     c->chrSrcVSubSample += c->vChrDrop;
00985 
00986     /* drop every other pixel for chroma calculation unless user
00987      * wants full chroma */
00988     if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP)   &&
00989         srcFormat != PIX_FMT_RGB8 && srcFormat != PIX_FMT_BGR8 &&
00990         srcFormat != PIX_FMT_RGB4 && srcFormat != PIX_FMT_BGR4 &&
00991         srcFormat != PIX_FMT_RGB4_BYTE && srcFormat != PIX_FMT_BGR4_BYTE &&
00992         ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
00993          (flags & SWS_FAST_BILINEAR)))
00994         c->chrSrcHSubSample = 1;
00995 
00996     // Note the -((-x)>>y) is so that we always round toward +inf.
00997     c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);
00998     c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);
00999     c->chrDstW = -((-dstW) >> c->chrDstHSubSample);
01000     c->chrDstH = -((-dstH) >> c->chrDstVSubSample);
01001 
01002     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
01003 
01004     /* unscaled special cases */
01005     if (unscaled && !usesHFilter && !usesVFilter &&
01006         (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
01007         ff_get_unscaled_swscale(c);
01008 
01009         if (c->swScale) {
01010             if (flags & SWS_PRINT_INFO)
01011                 av_log(c, AV_LOG_INFO,
01012                        "using unscaled %s -> %s special converter\n",
01013                        av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
01014             return 0;
01015         }
01016     }
01017 
01018     c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;
01019     if (c->srcBpc < 8)
01020         c->srcBpc = 8;
01021     c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;
01022     if (c->dstBpc < 8)
01023         c->dstBpc = 8;
01024     if (isAnyRGB(srcFormat) || srcFormat == PIX_FMT_PAL8)
01025         c->srcBpc = 16;
01026     if (c->dstBpc == 16)
01027         dst_stride <<= 1;
01028     if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {
01029         c->canMMX2BeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
01030                             (srcW & 15) == 0) ? 1 : 0;
01031         if (!c->canMMX2BeUsed && dstW >= srcW && (srcW & 15) == 0
01032             && (flags & SWS_FAST_BILINEAR)) {
01033             if (flags & SWS_PRINT_INFO)
01034                 av_log(c, AV_LOG_INFO,
01035                        "output width is not a multiple of 32 -> no MMX2 scaler\n");
01036         }
01037         if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
01038             c->canMMX2BeUsed=0;
01039     } else
01040         c->canMMX2BeUsed = 0;
01041 
01042     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
01043     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
01044 
01045     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
01046      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
01047      * correct scaling.
01048      * n-2 is the last chrominance sample available.
01049      * This is not perfect, but no one should notice the difference, the more
01050      * correct variant would be like the vertical one, but that would require
01051      * some special code for the first and last pixel */
01052     if (flags & SWS_FAST_BILINEAR) {
01053         if (c->canMMX2BeUsed) {
01054             c->lumXInc += 20;
01055             c->chrXInc += 20;
01056         }
01057         // we don't use the x86 asm scaler if MMX is available
01058         else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {
01059             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;
01060             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
01061         }
01062     }
01063 
01064     /* precalculate horizontal scaler filter coefficients */
01065     {
01066 #if HAVE_MMXEXT_INLINE
01067 // can't downscale !!!
01068         if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
01069             c->lumMmx2FilterCodeSize = initMMX2HScaler(dstW, c->lumXInc, NULL,
01070                                                        NULL, NULL, 8);
01071             c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc,
01072                                                        NULL, NULL, NULL, 4);
01073 
01074 #ifdef MAP_ANONYMOUS
01075             c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
01076             c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
01077 #elif HAVE_VIRTUALALLOC
01078             c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
01079             c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
01080 #else
01081             c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
01082             c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
01083 #endif
01084 
01085 #ifdef MAP_ANONYMOUS
01086             if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)
01087 #else
01088             if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
01089 #endif
01090             {
01091                 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
01092                 return AVERROR(ENOMEM);
01093             }
01094 
01095             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);
01096             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);
01097             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);
01098             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
01099 
01100             initMMX2HScaler(      dstW, c->lumXInc, c->lumMmx2FilterCode,
01101                             c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
01102             initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode,
01103                             c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
01104 
01105 #ifdef MAP_ANONYMOUS
01106             mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
01107             mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
01108 #endif
01109         } else
01110 #endif /* HAVE_MMXEXT_INLINE */
01111         {
01112             const int filterAlign =
01113                 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
01114                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
01115                 1;
01116 
01117             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
01118                            &c->hLumFilterSize, c->lumXInc,
01119                            srcW, dstW, filterAlign, 1 << 14,
01120                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
01121                            cpu_flags, srcFilter->lumH, dstFilter->lumH,
01122                            c->param) < 0)
01123                 goto fail;
01124             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
01125                            &c->hChrFilterSize, c->chrXInc,
01126                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
01127                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
01128                            cpu_flags, srcFilter->chrH, dstFilter->chrH,
01129                            c->param) < 0)
01130                 goto fail;
01131         }
01132     } // initialize horizontal stuff
01133 
01134     /* precalculate vertical scaler filter coefficients */
01135     {
01136         const int filterAlign =
01137             (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
01138             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
01139             1;
01140 
01141         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
01142                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
01143                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
01144                        cpu_flags, srcFilter->lumV, dstFilter->lumV,
01145                        c->param) < 0)
01146             goto fail;
01147         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
01148                        c->chrYInc, c->chrSrcH, c->chrDstH,
01149                        filterAlign, (1 << 12),
01150                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
01151                        cpu_flags, srcFilter->chrV, dstFilter->chrV,
01152                        c->param) < 0)
01153             goto fail;
01154 
01155 #if HAVE_ALTIVEC
01156         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);
01157         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
01158 
01159         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
01160             int j;
01161             short *p = (short *)&c->vYCoeffsBank[i];
01162             for (j = 0; j < 8; j++)
01163                 p[j] = c->vLumFilter[i];
01164         }
01165 
01166         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
01167             int j;
01168             short *p = (short *)&c->vCCoeffsBank[i];
01169             for (j = 0; j < 8; j++)
01170                 p[j] = c->vChrFilter[i];
01171         }
01172 #endif
01173     }
01174 
01175     // calculate buffer sizes so that they won't run out while handling these damn slices
01176     c->vLumBufSize = c->vLumFilterSize;
01177     c->vChrBufSize = c->vChrFilterSize;
01178     for (i = 0; i < dstH; i++) {
01179         int chrI      = (int64_t)i * c->chrDstH / dstH;
01180         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
01181                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
01182                                << c->chrSrcVSubSample));
01183 
01184         nextSlice >>= c->chrSrcVSubSample;
01185         nextSlice <<= c->chrSrcVSubSample;
01186         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
01187             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
01188         if (c->vChrFilterPos[chrI] + c->vChrBufSize <
01189             (nextSlice >> c->chrSrcVSubSample))
01190             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
01191                              c->vChrFilterPos[chrI];
01192     }
01193 
01194     /* Allocate pixbufs (we use dynamic allocation because otherwise we would
01195      * need to allocate several megabytes to handle all possible cases) */
01196     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);
01197     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
01198     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
01199     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
01200         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
01201     /* Note we need at least one pixel more at the end because of the MMX code
01202      * (just in case someone wants to replace the 4000/8000). */
01203     /* align at 16 bytes for AltiVec */
01204     for (i = 0; i < c->vLumBufSize; i++) {
01205         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
01206                           dst_stride + 16, fail);
01207         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
01208     }
01209     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
01210     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
01211     c->uv_offx2 = dst_stride + 16;
01212     for (i = 0; i < c->vChrBufSize; i++) {
01213         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
01214                          dst_stride * 2 + 32, fail);
01215         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
01216         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
01217                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
01218     }
01219     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
01220         for (i = 0; i < c->vLumBufSize; i++) {
01221             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
01222                               dst_stride + 16, fail);
01223             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
01224         }
01225 
01226     // try to avoid drawing green stuff between the right end and the stride end
01227     for (i = 0; i < c->vChrBufSize; i++)
01228         if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){
01229             av_assert0(c->dstBpc > 14);
01230             for(j=0; j<dst_stride/2+1; j++)
01231                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
01232         } else
01233             for(j=0; j<dst_stride+1; j++)
01234                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
01235 
01236     av_assert0(c->chrDstH <= dstH);
01237 
01238     if (flags & SWS_PRINT_INFO) {
01239         if (flags & SWS_FAST_BILINEAR)
01240             av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
01241         else if (flags & SWS_BILINEAR)
01242             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
01243         else if (flags & SWS_BICUBIC)
01244             av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
01245         else if (flags & SWS_X)
01246             av_log(c, AV_LOG_INFO, "Experimental scaler, ");
01247         else if (flags & SWS_POINT)
01248             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
01249         else if (flags & SWS_AREA)
01250             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
01251         else if (flags & SWS_BICUBLIN)
01252             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
01253         else if (flags & SWS_GAUSS)
01254             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
01255         else if (flags & SWS_SINC)
01256             av_log(c, AV_LOG_INFO, "Sinc scaler, ");
01257         else if (flags & SWS_LANCZOS)
01258             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
01259         else if (flags & SWS_SPLINE)
01260             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
01261         else
01262             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
01263 
01264         av_log(c, AV_LOG_INFO, "from %s to %s%s ",
01265                av_get_pix_fmt_name(srcFormat),
01266 #ifdef DITHER1XBPP
01267                dstFormat == PIX_FMT_BGR555   || dstFormat == PIX_FMT_BGR565   ||
01268                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
01269                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ?
01270                                                              "dithered " : "",
01271 #else
01272                "",
01273 #endif
01274                av_get_pix_fmt_name(dstFormat));
01275 
01276         if (INLINE_MMXEXT(cpu_flags))
01277             av_log(c, AV_LOG_INFO, "using MMX2\n");
01278         else if (INLINE_AMD3DNOW(cpu_flags))
01279             av_log(c, AV_LOG_INFO, "using 3DNOW\n");
01280         else if (INLINE_MMX(cpu_flags))
01281             av_log(c, AV_LOG_INFO, "using MMX\n");
01282         else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)
01283             av_log(c, AV_LOG_INFO, "using AltiVec\n");
01284         else
01285             av_log(c, AV_LOG_INFO, "using C\n");
01286 
01287         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
01288         av_log(c, AV_LOG_DEBUG,
01289                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
01290                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
01291         av_log(c, AV_LOG_DEBUG,
01292                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
01293                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
01294                c->chrXInc, c->chrYInc);
01295     }
01296 
01297     c->swScale = ff_getSwsFunc(c);
01298     return 0;
01299 fail: // FIXME replace things by appropriate error codes
01300     return -1;
01301 }
01302 
01303 #if FF_API_SWS_GETCONTEXT
01304 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
01305                            int dstW, int dstH, enum PixelFormat dstFormat,
01306                            int flags, SwsFilter *srcFilter,
01307                            SwsFilter *dstFilter, const double *param)
01308 {
01309     SwsContext *c;
01310 
01311     if (!(c = sws_alloc_context()))
01312         return NULL;
01313 
01314     c->flags     = flags;
01315     c->srcW      = srcW;
01316     c->srcH      = srcH;
01317     c->dstW      = dstW;
01318     c->dstH      = dstH;
01319     c->srcRange  = handle_jpeg(&srcFormat);
01320     c->dstRange  = handle_jpeg(&dstFormat);
01321     c->src0Alpha = handle_0alpha(&srcFormat);
01322     c->dst0Alpha = handle_0alpha(&dstFormat);
01323     c->srcFormat = srcFormat;
01324     c->dstFormat = dstFormat;
01325 
01326     if (param) {
01327         c->param[0] = param[0];
01328         c->param[1] = param[1];
01329     }
01330     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
01331                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
01332                              c->dstRange, 0, 1 << 16, 1 << 16);
01333 
01334     if (sws_init_context(c, srcFilter, dstFilter) < 0) {
01335         sws_freeContext(c);
01336         return NULL;
01337     }
01338 
01339     return c;
01340 }
01341 #endif
01342 
01343 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
01344                                 float lumaSharpen, float chromaSharpen,
01345                                 float chromaHShift, float chromaVShift,
01346                                 int verbose)
01347 {
01348     SwsFilter *filter = av_malloc(sizeof(SwsFilter));
01349     if (!filter)
01350         return NULL;
01351 
01352     if (lumaGBlur != 0.0) {
01353         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
01354         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
01355     } else {
01356         filter->lumH = sws_getIdentityVec();
01357         filter->lumV = sws_getIdentityVec();
01358     }
01359 
01360     if (chromaGBlur != 0.0) {
01361         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
01362         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
01363     } else {
01364         filter->chrH = sws_getIdentityVec();
01365         filter->chrV = sws_getIdentityVec();
01366     }
01367 
01368     if (chromaSharpen != 0.0) {
01369         SwsVector *id = sws_getIdentityVec();
01370         sws_scaleVec(filter->chrH, -chromaSharpen);
01371         sws_scaleVec(filter->chrV, -chromaSharpen);
01372         sws_addVec(filter->chrH, id);
01373         sws_addVec(filter->chrV, id);
01374         sws_freeVec(id);
01375     }
01376 
01377     if (lumaSharpen != 0.0) {
01378         SwsVector *id = sws_getIdentityVec();
01379         sws_scaleVec(filter->lumH, -lumaSharpen);
01380         sws_scaleVec(filter->lumV, -lumaSharpen);
01381         sws_addVec(filter->lumH, id);
01382         sws_addVec(filter->lumV, id);
01383         sws_freeVec(id);
01384     }
01385 
01386     if (chromaHShift != 0.0)
01387         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
01388 
01389     if (chromaVShift != 0.0)
01390         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
01391 
01392     sws_normalizeVec(filter->chrH, 1.0);
01393     sws_normalizeVec(filter->chrV, 1.0);
01394     sws_normalizeVec(filter->lumH, 1.0);
01395     sws_normalizeVec(filter->lumV, 1.0);
01396 
01397     if (verbose)
01398         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
01399     if (verbose)
01400         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
01401 
01402     return filter;
01403 }
01404 
01405 SwsVector *sws_allocVec(int length)
01406 {
01407     SwsVector *vec = av_malloc(sizeof(SwsVector));
01408     if (!vec)
01409         return NULL;
01410     vec->length = length;
01411     vec->coeff  = av_malloc(sizeof(double) * length);
01412     if (!vec->coeff)
01413         av_freep(&vec);
01414     return vec;
01415 }
01416 
01417 SwsVector *sws_getGaussianVec(double variance, double quality)
01418 {
01419     const int length = (int)(variance * quality + 0.5) | 1;
01420     int i;
01421     double middle  = (length - 1) * 0.5;
01422     SwsVector *vec = sws_allocVec(length);
01423 
01424     if (!vec)
01425         return NULL;
01426 
01427     for (i = 0; i < length; i++) {
01428         double dist = i - middle;
01429         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
01430                         sqrt(2 * variance * M_PI);
01431     }
01432 
01433     sws_normalizeVec(vec, 1.0);
01434 
01435     return vec;
01436 }
01437 
01438 SwsVector *sws_getConstVec(double c, int length)
01439 {
01440     int i;
01441     SwsVector *vec = sws_allocVec(length);
01442 
01443     if (!vec)
01444         return NULL;
01445 
01446     for (i = 0; i < length; i++)
01447         vec->coeff[i] = c;
01448 
01449     return vec;
01450 }
01451 
01452 SwsVector *sws_getIdentityVec(void)
01453 {
01454     return sws_getConstVec(1.0, 1);
01455 }
01456 
01457 static double sws_dcVec(SwsVector *a)
01458 {
01459     int i;
01460     double sum = 0;
01461 
01462     for (i = 0; i < a->length; i++)
01463         sum += a->coeff[i];
01464 
01465     return sum;
01466 }
01467 
01468 void sws_scaleVec(SwsVector *a, double scalar)
01469 {
01470     int i;
01471 
01472     for (i = 0; i < a->length; i++)
01473         a->coeff[i] *= scalar;
01474 }
01475 
01476 void sws_normalizeVec(SwsVector *a, double height)
01477 {
01478     sws_scaleVec(a, height / sws_dcVec(a));
01479 }
01480 
01481 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
01482 {
01483     int length = a->length + b->length - 1;
01484     int i, j;
01485     SwsVector *vec = sws_getConstVec(0.0, length);
01486 
01487     if (!vec)
01488         return NULL;
01489 
01490     for (i = 0; i < a->length; i++) {
01491         for (j = 0; j < b->length; j++) {
01492             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
01493         }
01494     }
01495 
01496     return vec;
01497 }
01498 
01499 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
01500 {
01501     int length = FFMAX(a->length, b->length);
01502     int i;
01503     SwsVector *vec = sws_getConstVec(0.0, length);
01504 
01505     if (!vec)
01506         return NULL;
01507 
01508     for (i = 0; i < a->length; i++)
01509         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
01510     for (i = 0; i < b->length; i++)
01511         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
01512 
01513     return vec;
01514 }
01515 
01516 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
01517 {
01518     int length = FFMAX(a->length, b->length);
01519     int i;
01520     SwsVector *vec = sws_getConstVec(0.0, length);
01521 
01522     if (!vec)
01523         return NULL;
01524 
01525     for (i = 0; i < a->length; i++)
01526         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
01527     for (i = 0; i < b->length; i++)
01528         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
01529 
01530     return vec;
01531 }
01532 
01533 /* shift left / or right if "shift" is negative */
01534 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
01535 {
01536     int length = a->length + FFABS(shift) * 2;
01537     int i;
01538     SwsVector *vec = sws_getConstVec(0.0, length);
01539 
01540     if (!vec)
01541         return NULL;
01542 
01543     for (i = 0; i < a->length; i++) {
01544         vec->coeff[i + (length    - 1) / 2 -
01545                        (a->length - 1) / 2 - shift] = a->coeff[i];
01546     }
01547 
01548     return vec;
01549 }
01550 
01551 void sws_shiftVec(SwsVector *a, int shift)
01552 {
01553     SwsVector *shifted = sws_getShiftedVec(a, shift);
01554     av_free(a->coeff);
01555     a->coeff  = shifted->coeff;
01556     a->length = shifted->length;
01557     av_free(shifted);
01558 }
01559 
01560 void sws_addVec(SwsVector *a, SwsVector *b)
01561 {
01562     SwsVector *sum = sws_sumVec(a, b);
01563     av_free(a->coeff);
01564     a->coeff  = sum->coeff;
01565     a->length = sum->length;
01566     av_free(sum);
01567 }
01568 
01569 void sws_subVec(SwsVector *a, SwsVector *b)
01570 {
01571     SwsVector *diff = sws_diffVec(a, b);
01572     av_free(a->coeff);
01573     a->coeff  = diff->coeff;
01574     a->length = diff->length;
01575     av_free(diff);
01576 }
01577 
01578 void sws_convVec(SwsVector *a, SwsVector *b)
01579 {
01580     SwsVector *conv = sws_getConvVec(a, b);
01581     av_free(a->coeff);
01582     a->coeff  = conv->coeff;
01583     a->length = conv->length;
01584     av_free(conv);
01585 }
01586 
01587 SwsVector *sws_cloneVec(SwsVector *a)
01588 {
01589     int i;
01590     SwsVector *vec = sws_allocVec(a->length);
01591 
01592     if (!vec)
01593         return NULL;
01594 
01595     for (i = 0; i < a->length; i++)
01596         vec->coeff[i] = a->coeff[i];
01597 
01598     return vec;
01599 }
01600 
01601 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
01602 {
01603     int i;
01604     double max = 0;
01605     double min = 0;
01606     double range;
01607 
01608     for (i = 0; i < a->length; i++)
01609         if (a->coeff[i] > max)
01610             max = a->coeff[i];
01611 
01612     for (i = 0; i < a->length; i++)
01613         if (a->coeff[i] < min)
01614             min = a->coeff[i];
01615 
01616     range = max - min;
01617 
01618     for (i = 0; i < a->length; i++) {
01619         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
01620         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
01621         for (; x > 0; x--)
01622             av_log(log_ctx, log_level, " ");
01623         av_log(log_ctx, log_level, "|\n");
01624     }
01625 }
01626 
01627 void sws_freeVec(SwsVector *a)
01628 {
01629     if (!a)
01630         return;
01631     av_freep(&a->coeff);
01632     a->length = 0;
01633     av_free(a);
01634 }
01635 
01636 void sws_freeFilter(SwsFilter *filter)
01637 {
01638     if (!filter)
01639         return;
01640 
01641     if (filter->lumH)
01642         sws_freeVec(filter->lumH);
01643     if (filter->lumV)
01644         sws_freeVec(filter->lumV);
01645     if (filter->chrH)
01646         sws_freeVec(filter->chrH);
01647     if (filter->chrV)
01648         sws_freeVec(filter->chrV);
01649     av_free(filter);
01650 }
01651 
01652 void sws_freeContext(SwsContext *c)
01653 {
01654     int i;
01655     if (!c)
01656         return;
01657 
01658     if (c->lumPixBuf) {
01659         for (i = 0; i < c->vLumBufSize; i++)
01660             av_freep(&c->lumPixBuf[i]);
01661         av_freep(&c->lumPixBuf);
01662     }
01663 
01664     if (c->chrUPixBuf) {
01665         for (i = 0; i < c->vChrBufSize; i++)
01666             av_freep(&c->chrUPixBuf[i]);
01667         av_freep(&c->chrUPixBuf);
01668         av_freep(&c->chrVPixBuf);
01669     }
01670 
01671     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
01672         for (i = 0; i < c->vLumBufSize; i++)
01673             av_freep(&c->alpPixBuf[i]);
01674         av_freep(&c->alpPixBuf);
01675     }
01676 
01677     av_freep(&c->vLumFilter);
01678     av_freep(&c->vChrFilter);
01679     av_freep(&c->hLumFilter);
01680     av_freep(&c->hChrFilter);
01681 #if HAVE_ALTIVEC
01682     av_freep(&c->vYCoeffsBank);
01683     av_freep(&c->vCCoeffsBank);
01684 #endif
01685 
01686     av_freep(&c->vLumFilterPos);
01687     av_freep(&c->vChrFilterPos);
01688     av_freep(&c->hLumFilterPos);
01689     av_freep(&c->hChrFilterPos);
01690 
01691 #if HAVE_MMX_INLINE
01692 #ifdef MAP_ANONYMOUS
01693     if (c->lumMmx2FilterCode)
01694         munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
01695     if (c->chrMmx2FilterCode)
01696         munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
01697 #elif HAVE_VIRTUALALLOC
01698     if (c->lumMmx2FilterCode)
01699         VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
01700     if (c->chrMmx2FilterCode)
01701         VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
01702 #else
01703     av_free(c->lumMmx2FilterCode);
01704     av_free(c->chrMmx2FilterCode);
01705 #endif
01706     c->lumMmx2FilterCode = NULL;
01707     c->chrMmx2FilterCode = NULL;
01708 #endif /* HAVE_MMX_INLINE */
01709 
01710     av_freep(&c->yuvTable);
01711     av_freep(&c->formatConvBuffer);
01712 
01713     av_free(c);
01714 }
01715 
01716 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
01717                                         int srcH, enum PixelFormat srcFormat,
01718                                         int dstW, int dstH,
01719                                         enum PixelFormat dstFormat, int flags,
01720                                         SwsFilter *srcFilter,
01721                                         SwsFilter *dstFilter,
01722                                         const double *param)
01723 {
01724     static const double default_param[2] = { SWS_PARAM_DEFAULT,
01725                                              SWS_PARAM_DEFAULT };
01726 
01727     if (!param)
01728         param = default_param;
01729 
01730     if (context &&
01731         (context->srcW      != srcW      ||
01732          context->srcH      != srcH      ||
01733          context->srcFormat != srcFormat ||
01734          context->dstW      != dstW      ||
01735          context->dstH      != dstH      ||
01736          context->dstFormat != dstFormat ||
01737          context->flags     != flags     ||
01738          context->param[0]  != param[0]  ||
01739          context->param[1]  != param[1])) {
01740         sws_freeContext(context);
01741         context = NULL;
01742     }
01743 
01744     if (!context) {
01745         if (!(context = sws_alloc_context()))
01746             return NULL;
01747         context->srcW      = srcW;
01748         context->srcH      = srcH;
01749         context->srcRange  = handle_jpeg(&srcFormat);
01750         context->src0Alpha = handle_0alpha(&srcFormat);
01751         context->srcFormat = srcFormat;
01752         context->dstW      = dstW;
01753         context->dstH      = dstH;
01754         context->dstRange  = handle_jpeg(&dstFormat);
01755         context->dst0Alpha = handle_0alpha(&dstFormat);
01756         context->dstFormat = dstFormat;
01757         context->flags     = flags;
01758         context->param[0]  = param[0];
01759         context->param[1]  = param[1];
01760         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
01761                                  context->srcRange,
01762                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
01763                                  context->dstRange, 0, 1 << 16, 1 << 16);
01764         if (sws_init_context(context, srcFilter, dstFilter) < 0) {
01765             sws_freeContext(context);
01766             return NULL;
01767         }
01768     }
01769     return context;
01770 }