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 <assert.h>
00026 #include <inttypes.h>
00027 #include <math.h>
00028 #include <stdio.h>
00029 #include <string.h>
00030 #if HAVE_SYS_MMAN_H
00031 #include <sys/mman.h>
00032 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
00033 #define MAP_ANONYMOUS MAP_ANON
00034 #endif
00035 #endif
00036 #if HAVE_VIRTUALALLOC
00037 #define WIN32_LEAN_AND_MEAN
00038 #include <windows.h>
00039 #endif
00040 
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_cpu.h"
00050 #include "rgb2rgb.h"
00051 #include "swscale.h"
00052 #include "swscale_internal.h"
00053 
00054 unsigned swscale_version(void)
00055 {
00056     av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
00057     return LIBSWSCALE_VERSION_INT;
00058 }
00059 
00060 const char *swscale_configuration(void)
00061 {
00062     return FFMPEG_CONFIGURATION;
00063 }
00064 
00065 const char *swscale_license(void)
00066 {
00067 #define LICENSE_PREFIX "libswscale license: "
00068     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
00069 }
00070 
00071 #define RET 0xC3 // near return opcode for x86
00072 
00073 typedef struct FormatEntry {
00074     int is_supported_in, is_supported_out;
00075 } FormatEntry;
00076 
00077 static const FormatEntry format_entries[PIX_FMT_NB] = {
00078     [PIX_FMT_YUV420P]     = { 1, 1 },
00079     [PIX_FMT_YUYV422]     = { 1, 1 },
00080     [PIX_FMT_RGB24]       = { 1, 1 },
00081     [PIX_FMT_BGR24]       = { 1, 1 },
00082     [PIX_FMT_YUV422P]     = { 1, 1 },
00083     [PIX_FMT_YUV444P]     = { 1, 1 },
00084     [PIX_FMT_YUV410P]     = { 1, 1 },
00085     [PIX_FMT_YUV411P]     = { 1, 1 },
00086     [PIX_FMT_GRAY8]       = { 1, 1 },
00087     [PIX_FMT_MONOWHITE]   = { 1, 1 },
00088     [PIX_FMT_MONOBLACK]   = { 1, 1 },
00089     [PIX_FMT_PAL8]        = { 1, 0 },
00090     [PIX_FMT_YUVJ420P]    = { 1, 1 },
00091     [PIX_FMT_YUVJ422P]    = { 1, 1 },
00092     [PIX_FMT_YUVJ444P]    = { 1, 1 },
00093     [PIX_FMT_UYVY422]     = { 1, 1 },
00094     [PIX_FMT_UYYVYY411]   = { 0, 0 },
00095     [PIX_FMT_BGR8]        = { 1, 1 },
00096     [PIX_FMT_BGR4]        = { 0, 1 },
00097     [PIX_FMT_BGR4_BYTE]   = { 1, 1 },
00098     [PIX_FMT_RGB8]        = { 1, 1 },
00099     [PIX_FMT_RGB4]        = { 0, 1 },
00100     [PIX_FMT_RGB4_BYTE]   = { 1, 1 },
00101     [PIX_FMT_NV12]        = { 1, 1 },
00102     [PIX_FMT_NV21]        = { 1, 1 },
00103     [PIX_FMT_ARGB]        = { 1, 1 },
00104     [PIX_FMT_RGBA]        = { 1, 1 },
00105     [PIX_FMT_ABGR]        = { 1, 1 },
00106     [PIX_FMT_BGRA]        = { 1, 1 },
00107     [PIX_FMT_0RGB]        = { 1, 1 },
00108     [PIX_FMT_RGB0]        = { 1, 1 },
00109     [PIX_FMT_0BGR]        = { 1, 1 },
00110     [PIX_FMT_BGR0]        = { 1, 1 },
00111     [PIX_FMT_GRAY16BE]    = { 1, 1 },
00112     [PIX_FMT_GRAY16LE]    = { 1, 1 },
00113     [PIX_FMT_YUV440P]     = { 1, 1 },
00114     [PIX_FMT_YUVJ440P]    = { 1, 1 },
00115     [PIX_FMT_YUVA420P]    = { 1, 1 },
00116     [PIX_FMT_YUVA422P]    = { 1, 1 },
00117     [PIX_FMT_YUVA444P]    = { 1, 1 },
00118     [PIX_FMT_RGB48BE]     = { 1, 1 },
00119     [PIX_FMT_RGB48LE]     = { 1, 1 },
00120     [PIX_FMT_RGBA64BE]    = { 1, 0 },
00121     [PIX_FMT_RGBA64LE]    = { 1, 0 },
00122     [PIX_FMT_RGB565BE]    = { 1, 1 },
00123     [PIX_FMT_RGB565LE]    = { 1, 1 },
00124     [PIX_FMT_RGB555BE]    = { 1, 1 },
00125     [PIX_FMT_RGB555LE]    = { 1, 1 },
00126     [PIX_FMT_BGR565BE]    = { 1, 1 },
00127     [PIX_FMT_BGR565LE]    = { 1, 1 },
00128     [PIX_FMT_BGR555BE]    = { 1, 1 },
00129     [PIX_FMT_BGR555LE]    = { 1, 1 },
00130     [PIX_FMT_YUV420P16LE] = { 1, 1 },
00131     [PIX_FMT_YUV420P16BE] = { 1, 1 },
00132     [PIX_FMT_YUV422P16LE] = { 1, 1 },
00133     [PIX_FMT_YUV422P16BE] = { 1, 1 },
00134     [PIX_FMT_YUV444P16LE] = { 1, 1 },
00135     [PIX_FMT_YUV444P16BE] = { 1, 1 },
00136     [PIX_FMT_RGB444LE]    = { 1, 1 },
00137     [PIX_FMT_RGB444BE]    = { 1, 1 },
00138     [PIX_FMT_BGR444LE]    = { 1, 1 },
00139     [PIX_FMT_BGR444BE]    = { 1, 1 },
00140     [PIX_FMT_Y400A]       = { 1, 0 },
00141     [PIX_FMT_BGR48BE]     = { 1, 1 },
00142     [PIX_FMT_BGR48LE]     = { 1, 1 },
00143     [PIX_FMT_BGRA64BE]    = { 0, 0 },
00144     [PIX_FMT_BGRA64LE]    = { 0, 0 },
00145     [PIX_FMT_YUV420P9BE]  = { 1, 1 },
00146     [PIX_FMT_YUV420P9LE]  = { 1, 1 },
00147     [PIX_FMT_YUV420P10BE] = { 1, 1 },
00148     [PIX_FMT_YUV420P10LE] = { 1, 1 },
00149     [PIX_FMT_YUV422P9BE]  = { 1, 1 },
00150     [PIX_FMT_YUV422P9LE]  = { 1, 1 },
00151     [PIX_FMT_YUV422P10BE] = { 1, 1 },
00152     [PIX_FMT_YUV422P10LE] = { 1, 1 },
00153     [PIX_FMT_YUV444P9BE]  = { 1, 1 },
00154     [PIX_FMT_YUV444P9LE]  = { 1, 1 },
00155     [PIX_FMT_YUV444P10BE] = { 1, 1 },
00156     [PIX_FMT_YUV444P10LE] = { 1, 1 },
00157     [PIX_FMT_GBRP]        = { 1, 0 },
00158     [PIX_FMT_GBRP9LE]     = { 1, 0 },
00159     [PIX_FMT_GBRP9BE]     = { 1, 0 },
00160     [PIX_FMT_GBRP10LE]    = { 1, 0 },
00161     [PIX_FMT_GBRP10BE]    = { 1, 0 },
00162     [PIX_FMT_GBRP16LE]    = { 1, 0 },
00163     [PIX_FMT_GBRP16BE]    = { 1, 0 },
00164 };
00165 
00166 int sws_isSupportedInput(enum PixelFormat pix_fmt)
00167 {
00168     return (unsigned)pix_fmt < PIX_FMT_NB ?
00169            format_entries[pix_fmt].is_supported_in : 0;
00170 }
00171 
00172 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
00173 {
00174     return (unsigned)pix_fmt < PIX_FMT_NB ?
00175            format_entries[pix_fmt].is_supported_out : 0;
00176 }
00177 
00178 extern const int32_t ff_yuv2rgb_coeffs[8][4];
00179 
00180 #if FF_API_SWS_FORMAT_NAME
00181 const char *sws_format_name(enum PixelFormat format)
00182 {
00183     return av_get_pix_fmt_name(format);
00184 }
00185 #endif
00186 
00187 static double getSplineCoeff(double a, double b, double c, double d,
00188                              double dist)
00189 {
00190     if (dist <= 1.0)
00191         return ((d * dist + c) * dist + b) * dist + a;
00192     else
00193         return getSplineCoeff(0.0,
00194                                b + 2.0 * c + 3.0 * d,
00195                                c + 3.0 * d,
00196                               -b - 3.0 * c - 6.0 * d,
00197                               dist - 1.0);
00198 }
00199 
00200 static int initFilter(int16_t **outFilter, int32_t **filterPos,
00201                       int *outFilterSize, int xInc, int srcW, int dstW,
00202                       int filterAlign, int one, int flags, int cpu_flags,
00203                       SwsVector *srcFilter, SwsVector *dstFilter,
00204                       double param[2])
00205 {
00206     int i;
00207     int filterSize;
00208     int filter2Size;
00209     int minFilterSize;
00210     int64_t *filter    = NULL;
00211     int64_t *filter2   = NULL;
00212     const int64_t fone = 1LL << 54;
00213     int ret            = -1;
00214 
00215     emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
00216 
00217     // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
00218     FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
00219 
00220     if (FFABS(xInc - 0x10000) < 10) { // unscaled
00221         int i;
00222         filterSize = 1;
00223         FF_ALLOCZ_OR_GOTO(NULL, filter,
00224                           dstW * sizeof(*filter) * filterSize, fail);
00225 
00226         for (i = 0; i < dstW; i++) {
00227             filter[i * filterSize] = fone;
00228             (*filterPos)[i]        = i;
00229         }
00230     } else if (flags & SWS_POINT) { // lame looking point sampling mode
00231         int i;
00232         int64_t xDstInSrc;
00233         filterSize = 1;
00234         FF_ALLOC_OR_GOTO(NULL, filter,
00235                          dstW * sizeof(*filter) * filterSize, fail);
00236 
00237         xDstInSrc = xInc / 2 - 0x8000;
00238         for (i = 0; i < dstW; i++) {
00239             int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
00240 
00241             (*filterPos)[i] = xx;
00242             filter[i]       = fone;
00243             xDstInSrc      += xInc;
00244         }
00245     } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
00246                (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
00247         int i;
00248         int64_t xDstInSrc;
00249         filterSize = 2;
00250         FF_ALLOC_OR_GOTO(NULL, filter,
00251                          dstW * sizeof(*filter) * filterSize, fail);
00252 
00253         xDstInSrc = xInc / 2 - 0x8000;
00254         for (i = 0; i < dstW; i++) {
00255             int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
00256             int j;
00257 
00258             (*filterPos)[i] = xx;
00259             // bilinear upscale / linear interpolate / area averaging
00260             for (j = 0; j < filterSize; j++) {
00261                 int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
00262                 if (coeff < 0)
00263                     coeff = 0;
00264                 filter[i * filterSize + j] = coeff;
00265                 xx++;
00266             }
00267             xDstInSrc += xInc;
00268         }
00269     } else {
00270         int64_t xDstInSrc;
00271         int sizeFactor;
00272 
00273         if (flags & SWS_BICUBIC)
00274             sizeFactor = 4;
00275         else if (flags & SWS_X)
00276             sizeFactor = 8;
00277         else if (flags & SWS_AREA)
00278             sizeFactor = 1;     // downscale only, for upscale it is bilinear
00279         else if (flags & SWS_GAUSS)
00280             sizeFactor = 8;     // infinite ;)
00281         else if (flags & SWS_LANCZOS)
00282             sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
00283         else if (flags & SWS_SINC)
00284             sizeFactor = 20;    // infinite ;)
00285         else if (flags & SWS_SPLINE)
00286             sizeFactor = 20;    // infinite ;)
00287         else if (flags & SWS_BILINEAR)
00288             sizeFactor = 2;
00289         else {
00290             sizeFactor = 0;     // GCC warning killer
00291             assert(0);
00292         }
00293 
00294         if (xInc <= 1 << 16)
00295             filterSize = 1 + sizeFactor;    // upscale
00296         else
00297             filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
00298 
00299         filterSize = FFMIN(filterSize, srcW - 2);
00300         filterSize = FFMAX(filterSize, 1);
00301 
00302         FF_ALLOC_OR_GOTO(NULL, filter,
00303                          dstW * sizeof(*filter) * filterSize, fail);
00304 
00305         xDstInSrc = xInc - 0x10000;
00306         for (i = 0; i < dstW; i++) {
00307             int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
00308             int j;
00309             (*filterPos)[i] = xx;
00310             for (j = 0; j < filterSize; j++) {
00311                 int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
00312                 double floatd;
00313                 int64_t coeff;
00314 
00315                 if (xInc > 1 << 16)
00316                     d = d * dstW / srcW;
00317                 floatd = d * (1.0 / (1 << 30));
00318 
00319                 if (flags & SWS_BICUBIC) {
00320                     int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] :   0) * (1 << 24);
00321                     int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
00322 
00323                     if (d >= 1LL << 31) {
00324                         coeff = 0.0;
00325                     } else {
00326                         int64_t dd  = (d  * d) >> 30;
00327                         int64_t ddd = (dd * d) >> 30;
00328 
00329                         if (d < 1LL << 30)
00330                             coeff =  (12 * (1 << 24) -  9 * B - 6 * C) * ddd +
00331                                     (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +
00332                                       (6 * (1 << 24) -  2 * B)         * (1 << 30);
00333                         else
00334                             coeff =      (-B -  6 * C) * ddd +
00335                                       (6 * B + 30 * C) * dd  +
00336                                     (-12 * B - 48 * C) * d   +
00337                                       (8 * B + 24 * C) * (1 << 30);
00338                     }
00339                     coeff *= fone >> (30 + 24);
00340                 }
00341 #if 0
00342                 else if (flags & SWS_X) {
00343                     double p  = param ? param * 0.01 : 0.3;
00344                     coeff     = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
00345                     coeff    *= pow(2.0, -p * d * d);
00346                 }
00347 #endif
00348                 else if (flags & SWS_X) {
00349                     double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
00350                     double c;
00351 
00352                     if (floatd < 1.0)
00353                         c = cos(floatd * M_PI);
00354                     else
00355                         c = -1.0;
00356                     if (c < 0.0)
00357                         c = -pow(-c, A);
00358                     else
00359                         c = pow(c, A);
00360                     coeff = (c * 0.5 + 0.5) * fone;
00361                 } else if (flags & SWS_AREA) {
00362                     int64_t d2 = d - (1 << 29);
00363                     if (d2 * xInc < -(1LL << (29 + 16)))
00364                         coeff = 1.0 * (1LL << (30 + 16));
00365                     else if (d2 * xInc < (1LL << (29 + 16)))
00366                         coeff = -d2 * xInc + (1LL << (29 + 16));
00367                     else
00368                         coeff = 0.0;
00369                     coeff *= fone >> (30 + 16);
00370                 } else if (flags & SWS_GAUSS) {
00371                     double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
00372                     coeff = (pow(2.0, -p * floatd * floatd)) * fone;
00373                 } else if (flags & SWS_SINC) {
00374                     coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
00375                 } else if (flags & SWS_LANCZOS) {
00376                     double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
00377                     coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
00378                              (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
00379                     if (floatd > p)
00380                         coeff = 0;
00381                 } else if (flags & SWS_BILINEAR) {
00382                     coeff = (1 << 30) - d;
00383                     if (coeff < 0)
00384                         coeff = 0;
00385                     coeff *= fone >> 30;
00386                 } else if (flags & SWS_SPLINE) {
00387                     double p = -2.196152422706632;
00388                     coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
00389                 } else {
00390                     coeff = 0.0; // GCC warning killer
00391                     assert(0);
00392                 }
00393 
00394                 filter[i * filterSize + j] = coeff;
00395                 xx++;
00396             }
00397             xDstInSrc += 2 * xInc;
00398         }
00399     }
00400 
00401     /* apply src & dst Filter to filter -> filter2
00402      * av_free(filter);
00403      */
00404     assert(filterSize > 0);
00405     filter2Size = filterSize;
00406     if (srcFilter)
00407         filter2Size += srcFilter->length - 1;
00408     if (dstFilter)
00409         filter2Size += dstFilter->length - 1;
00410     assert(filter2Size > 0);
00411     FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);
00412 
00413     for (i = 0; i < dstW; i++) {
00414         int j, k;
00415 
00416         if (srcFilter) {
00417             for (k = 0; k < srcFilter->length; k++) {
00418                 for (j = 0; j < filterSize; j++)
00419                     filter2[i * filter2Size + k + j] +=
00420                         srcFilter->coeff[k] * filter[i * filterSize + j];
00421             }
00422         } else {
00423             for (j = 0; j < filterSize; j++)
00424                 filter2[i * filter2Size + j] = filter[i * filterSize + j];
00425         }
00426         // FIXME dstFilter
00427 
00428         (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
00429     }
00430     av_freep(&filter);
00431 
00432     /* try to reduce the filter-size (step1 find size and shift left) */
00433     // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
00434     minFilterSize = 0;
00435     for (i = dstW - 1; i >= 0; i--) {
00436         int min = filter2Size;
00437         int j;
00438         int64_t cutOff = 0.0;
00439 
00440         /* get rid of near zero elements on the left by shifting left */
00441         for (j = 0; j < filter2Size; j++) {
00442             int k;
00443             cutOff += FFABS(filter2[i * filter2Size]);
00444 
00445             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
00446                 break;
00447 
00448             /* preserve monotonicity because the core can't handle the
00449              * filter otherwise */
00450             if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
00451                 break;
00452 
00453             // move filter coefficients left
00454             for (k = 1; k < filter2Size; k++)
00455                 filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
00456             filter2[i * filter2Size + k - 1] = 0;
00457             (*filterPos)[i]++;
00458         }
00459 
00460         cutOff = 0;
00461         /* count near zeros on the right */
00462         for (j = filter2Size - 1; j > 0; j--) {
00463             cutOff += FFABS(filter2[i * filter2Size + j]);
00464 
00465             if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
00466                 break;
00467             min--;
00468         }
00469 
00470         if (min > minFilterSize)
00471             minFilterSize = min;
00472     }
00473 
00474     if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
00475         // we can handle the special case 4, so we don't want to go the full 8
00476         if (minFilterSize < 5)
00477             filterAlign = 4;
00478 
00479         /* We really don't want to waste our time doing useless computation, so
00480          * fall back on the scalar C code for very small filters.
00481          * Vectorizing is worth it only if you have a decent-sized vector. */
00482         if (minFilterSize < 3)
00483             filterAlign = 1;
00484     }
00485 
00486     if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
00487         // special case for unscaled vertical filtering
00488         if (minFilterSize == 1 && filterAlign == 2)
00489             filterAlign = 1;
00490     }
00491 
00492     assert(minFilterSize > 0);
00493     filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
00494     assert(filterSize > 0);
00495     filter = av_malloc(filterSize * dstW * sizeof(*filter));
00496     if (filterSize >= MAX_FILTER_SIZE * 16 /
00497                       ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
00498         goto fail;
00499     *outFilterSize = filterSize;
00500 
00501     if (flags & SWS_PRINT_INFO)
00502         av_log(NULL, AV_LOG_VERBOSE,
00503                "SwScaler: reducing / aligning filtersize %d -> %d\n",
00504                filter2Size, filterSize);
00505     /* try to reduce the filter-size (step2 reduce it) */
00506     for (i = 0; i < dstW; i++) {
00507         int j;
00508 
00509         for (j = 0; j < filterSize; j++) {
00510             if (j >= filter2Size)
00511                 filter[i * filterSize + j] = 0;
00512             else
00513                 filter[i * filterSize + j] = filter2[i * filter2Size + j];
00514             if ((flags & SWS_BITEXACT) && j >= minFilterSize)
00515                 filter[i * filterSize + j] = 0;
00516         }
00517     }
00518 
00519     // FIXME try to align filterPos if possible
00520 
00521     // fix borders
00522     for (i = 0; i < dstW; i++) {
00523         int j;
00524         if ((*filterPos)[i] < 0) {
00525             // move filter coefficients left to compensate for filterPos
00526             for (j = 1; j < filterSize; j++) {
00527                 int left = FFMAX(j + (*filterPos)[i], 0);
00528                 filter[i * filterSize + left] += filter[i * filterSize + j];
00529                 filter[i * filterSize + j]     = 0;
00530             }
00531             (*filterPos)[i]= 0;
00532         }
00533 
00534         if ((*filterPos)[i] + filterSize > srcW) {
00535             int shift = (*filterPos)[i] + filterSize - srcW;
00536             // move filter coefficients right to compensate for filterPos
00537             for (j = filterSize - 2; j >= 0; j--) {
00538                 int right = FFMIN(j + shift, filterSize - 1);
00539                 filter[i * filterSize + right] += filter[i * filterSize + j];
00540                 filter[i * filterSize + j]      = 0;
00541             }
00542             (*filterPos)[i]= srcW - filterSize;
00543         }
00544     }
00545 
00546     // Note the +1 is for the MMX scaler which reads over the end
00547     /* align at 16 for AltiVec (needed by hScale_altivec_real) */
00548     FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
00549                       *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
00550 
00551     /* normalize & store in outFilter */
00552     for (i = 0; i < dstW; i++) {
00553         int j;
00554         int64_t error = 0;
00555         int64_t sum   = 0;
00556 
00557         for (j = 0; j < filterSize; j++) {
00558             sum += filter[i * filterSize + j];
00559         }
00560         sum = (sum + one / 2) / one;
00561         for (j = 0; j < *outFilterSize; j++) {
00562             int64_t v = filter[i * filterSize + j] + error;
00563             int intV  = ROUNDED_DIV(v, sum);
00564             (*outFilter)[i * (*outFilterSize) + j] = intV;
00565             error                                  = v - intV * sum;
00566         }
00567     }
00568 
00569     (*filterPos)[dstW + 0] =
00570     (*filterPos)[dstW + 1] =
00571     (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
00572                                                       * read over the end */
00573     for (i = 0; i < *outFilterSize; i++) {
00574         int k = (dstW - 1) * (*outFilterSize) + i;
00575         (*outFilter)[k + 1 * (*outFilterSize)] =
00576         (*outFilter)[k + 2 * (*outFilterSize)] =
00577         (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
00578     }
00579 
00580     ret = 0;
00581 
00582 fail:
00583     av_free(filter);
00584     av_free(filter2);
00585     return ret;
00586 }
00587 
00588 #if HAVE_MMX2
00589 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode,
00590                            int16_t *filter, int32_t *filterPos, int numSplits)
00591 {
00592     uint8_t *fragmentA;
00593     x86_reg imm8OfPShufW1A;
00594     x86_reg imm8OfPShufW2A;
00595     x86_reg fragmentLengthA;
00596     uint8_t *fragmentB;
00597     x86_reg imm8OfPShufW1B;
00598     x86_reg imm8OfPShufW2B;
00599     x86_reg fragmentLengthB;
00600     int fragmentPos;
00601 
00602     int xpos, i;
00603 
00604     // create an optimized horizontal scaling routine
00605     /* This scaler is made of runtime-generated MMX2 code using specially tuned
00606      * pshufw instructions. For every four output pixels, if four input pixels
00607      * are enough for the fast bilinear scaling, then a chunk of fragmentB is
00608      * used. If five input pixels are needed, then a chunk of fragmentA is used.
00609      */
00610 
00611     // code fragment
00612 
00613     __asm__ volatile (
00614         "jmp                         9f                 \n\t"
00615         // Begin
00616         "0:                                             \n\t"
00617         "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
00618         "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
00619         "movd   1(%%"REG_c", %%"REG_S"), %%mm1          \n\t"
00620         "punpcklbw                %%mm7, %%mm1          \n\t"
00621         "punpcklbw                %%mm7, %%mm0          \n\t"
00622         "pshufw                   $0xFF, %%mm1, %%mm1   \n\t"
00623         "1:                                             \n\t"
00624         "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
00625         "2:                                             \n\t"
00626         "psubw                    %%mm1, %%mm0          \n\t"
00627         "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
00628         "pmullw                   %%mm3, %%mm0          \n\t"
00629         "psllw                       $7, %%mm1          \n\t"
00630         "paddw                    %%mm1, %%mm0          \n\t"
00631 
00632         "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"
00633 
00634         "add                         $8, %%"REG_a"      \n\t"
00635         // End
00636         "9:                                             \n\t"
00637         // "int $3                                         \n\t"
00638         "lea       " LOCAL_MANGLE(0b) ", %0             \n\t"
00639         "lea       " LOCAL_MANGLE(1b) ", %1             \n\t"
00640         "lea       " LOCAL_MANGLE(2b) ", %2             \n\t"
00641         "dec                         %1                 \n\t"
00642         "dec                         %2                 \n\t"
00643         "sub                         %0, %1             \n\t"
00644         "sub                         %0, %2             \n\t"
00645         "lea       " LOCAL_MANGLE(9b) ", %3             \n\t"
00646         "sub                         %0, %3             \n\t"
00647 
00648 
00649         : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
00650           "=r" (fragmentLengthA)
00651         );
00652 
00653     __asm__ volatile (
00654         "jmp                         9f                 \n\t"
00655         // Begin
00656         "0:                                             \n\t"
00657         "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
00658         "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
00659         "punpcklbw                %%mm7, %%mm0          \n\t"
00660         "pshufw                   $0xFF, %%mm0, %%mm1   \n\t"
00661         "1:                                             \n\t"
00662         "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
00663         "2:                                             \n\t"
00664         "psubw                    %%mm1, %%mm0          \n\t"
00665         "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
00666         "pmullw                   %%mm3, %%mm0          \n\t"
00667         "psllw                       $7, %%mm1          \n\t"
00668         "paddw                    %%mm1, %%mm0          \n\t"
00669 
00670         "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"
00671 
00672         "add                         $8, %%"REG_a"      \n\t"
00673         // End
00674         "9:                                             \n\t"
00675         // "int                       $3                   \n\t"
00676         "lea       " LOCAL_MANGLE(0b) ", %0             \n\t"
00677         "lea       " LOCAL_MANGLE(1b) ", %1             \n\t"
00678         "lea       " LOCAL_MANGLE(2b) ", %2             \n\t"
00679         "dec                         %1                 \n\t"
00680         "dec                         %2                 \n\t"
00681         "sub                         %0, %1             \n\t"
00682         "sub                         %0, %2             \n\t"
00683         "lea       " LOCAL_MANGLE(9b) ", %3             \n\t"
00684         "sub                         %0, %3             \n\t"
00685 
00686 
00687         : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
00688           "=r" (fragmentLengthB)
00689         );
00690 
00691     xpos        = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
00692     fragmentPos = 0;
00693 
00694     for (i = 0; i < dstW / numSplits; i++) {
00695         int xx = xpos >> 16;
00696 
00697         if ((i & 3) == 0) {
00698             int a                  = 0;
00699             int b                  = ((xpos + xInc) >> 16) - xx;
00700             int c                  = ((xpos + xInc * 2) >> 16) - xx;
00701             int d                  = ((xpos + xInc * 3) >> 16) - xx;
00702             int inc                = (d + 1 < 4);
00703             uint8_t *fragment      = (d + 1 < 4) ? fragmentB : fragmentA;
00704             x86_reg imm8OfPShufW1  = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;
00705             x86_reg imm8OfPShufW2  = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;
00706             x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;
00707             int maxShift           = 3 - (d + inc);
00708             int shift              = 0;
00709 
00710             if (filterCode) {
00711                 filter[i]        = ((xpos              & 0xFFFF) ^ 0xFFFF) >> 9;
00712                 filter[i + 1]    = (((xpos + xInc)     & 0xFFFF) ^ 0xFFFF) >> 9;
00713                 filter[i + 2]    = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
00714                 filter[i + 3]    = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
00715                 filterPos[i / 2] = xx;
00716 
00717                 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
00718 
00719                 filterCode[fragmentPos + imm8OfPShufW1] =  (a + inc)       |
00720                                                           ((b + inc) << 2) |
00721                                                           ((c + inc) << 4) |
00722                                                           ((d + inc) << 6);
00723                 filterCode[fragmentPos + imm8OfPShufW2] =  a | (b << 2) |
00724                                                                (c << 4) |
00725                                                                (d << 6);
00726 
00727                 if (i + 4 - inc >= dstW)
00728                     shift = maxShift;               // avoid overread
00729                 else if ((filterPos[i / 2] & 3) <= maxShift)
00730                     shift = filterPos[i / 2] & 3;   // align
00731 
00732                 if (shift && i >= shift) {
00733                     filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
00734                     filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
00735                     filterPos[i / 2]                        -= shift;
00736                 }
00737             }
00738 
00739             fragmentPos += fragmentLength;
00740 
00741             if (filterCode)
00742                 filterCode[fragmentPos] = RET;
00743         }
00744         xpos += xInc;
00745     }
00746     if (filterCode)
00747         filterPos[((i / 2) + 1) & (~1)] = xpos >> 16;  // needed to jump to the next part
00748 
00749     return fragmentPos + 1;
00750 }
00751 #endif /* HAVE_MMX2 */
00752 
00753 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
00754 {
00755     *h = av_pix_fmt_descriptors[format].log2_chroma_w;
00756     *v = av_pix_fmt_descriptors[format].log2_chroma_h;
00757 }
00758 
00759 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
00760                              int srcRange, const int table[4], int dstRange,
00761                              int brightness, int contrast, int saturation)
00762 {
00763     memcpy(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
00764     memcpy(c->dstColorspaceTable, table, sizeof(int) * 4);
00765 
00766     c->brightness = brightness;
00767     c->contrast   = contrast;
00768     c->saturation = saturation;
00769     c->srcRange   = srcRange;
00770     c->dstRange   = dstRange;
00771     if (isYUV(c->dstFormat) || isGray(c->dstFormat))
00772         return -1;
00773 
00774     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
00775     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
00776 
00777     ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
00778                              contrast, saturation);
00779     // FIXME factorize
00780 
00781     if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
00782         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness,
00783                                        contrast, saturation);
00784     return 0;
00785 }
00786 
00787 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
00788                              int *srcRange, int **table, int *dstRange,
00789                              int *brightness, int *contrast, int *saturation)
00790 {
00791     if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat))
00792         return -1;
00793 
00794     *inv_table  = c->srcColorspaceTable;
00795     *table      = c->dstColorspaceTable;
00796     *srcRange   = c->srcRange;
00797     *dstRange   = c->dstRange;
00798     *brightness = c->brightness;
00799     *contrast   = c->contrast;
00800     *saturation = c->saturation;
00801 
00802     return 0;
00803 }
00804 
00805 static int handle_jpeg(enum PixelFormat *format)
00806 {
00807     switch (*format) {
00808     case PIX_FMT_YUVJ420P:
00809         *format = PIX_FMT_YUV420P;
00810         return 1;
00811     case PIX_FMT_YUVJ422P:
00812         *format = PIX_FMT_YUV422P;
00813         return 1;
00814     case PIX_FMT_YUVJ444P:
00815         *format = PIX_FMT_YUV444P;
00816         return 1;
00817     case PIX_FMT_YUVJ440P:
00818         *format = PIX_FMT_YUV440P;
00819         return 1;
00820     default:
00821         return 0;
00822     }
00823 }
00824 
00825 static int handle_0alpha(enum PixelFormat *format)
00826 {
00827     switch (*format) {
00828     case PIX_FMT_0BGR    : *format = PIX_FMT_ABGR   ; return 1;
00829     case PIX_FMT_BGR0    : *format = PIX_FMT_BGRA   ; return 4;
00830     case PIX_FMT_0RGB    : *format = PIX_FMT_ARGB   ; return 1;
00831     case PIX_FMT_RGB0    : *format = PIX_FMT_RGBA   ; return 4;
00832     default:                                          return 0;
00833     }
00834 }
00835 
00836 SwsContext *sws_alloc_context(void)
00837 {
00838     SwsContext *c = av_mallocz(sizeof(SwsContext));
00839 
00840     c->av_class = &sws_context_class;
00841     av_opt_set_defaults(c);
00842 
00843     return c;
00844 }
00845 
00846 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
00847 {
00848     int i, j;
00849     int usesVFilter, usesHFilter;
00850     int unscaled;
00851     SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
00852     int srcW              = c->srcW;
00853     int srcH              = c->srcH;
00854     int dstW              = c->dstW;
00855     int dstH              = c->dstH;
00856     int dst_stride        = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
00857     int flags, cpu_flags;
00858     enum PixelFormat srcFormat = c->srcFormat;
00859     enum PixelFormat dstFormat = c->dstFormat;
00860 
00861     cpu_flags = av_get_cpu_flags();
00862     flags     = c->flags;
00863     emms_c();
00864     if (!rgb15to16)
00865         sws_rgb2rgb_init();
00866 
00867     unscaled = (srcW == dstW && srcH == dstH);
00868 
00869     handle_jpeg(&srcFormat);
00870     handle_jpeg(&dstFormat);
00871     handle_0alpha(&srcFormat);
00872     handle_0alpha(&dstFormat);
00873 
00874     if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat){
00875         av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
00876         c->srcFormat= srcFormat;
00877         c->dstFormat= dstFormat;
00878     }
00879 
00880     if (!sws_isSupportedInput(srcFormat)) {
00881         av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
00882                av_get_pix_fmt_name(srcFormat));
00883         return AVERROR(EINVAL);
00884     }
00885     if (!sws_isSupportedOutput(dstFormat)) {
00886         av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
00887                av_get_pix_fmt_name(dstFormat));
00888         return AVERROR(EINVAL);
00889     }
00890 
00891     i = flags & (SWS_POINT         |
00892                  SWS_AREA          |
00893                  SWS_BILINEAR      |
00894                  SWS_FAST_BILINEAR |
00895                  SWS_BICUBIC       |
00896                  SWS_X             |
00897                  SWS_GAUSS         |
00898                  SWS_LANCZOS       |
00899                  SWS_SINC          |
00900                  SWS_SPLINE        |
00901                  SWS_BICUBLIN);
00902     if (!i || (i & (i - 1))) {
00903         av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen, got %X\n", i);
00904         return AVERROR(EINVAL);
00905     }
00906     /* sanity check */
00907     if (srcW < 4 || srcH < 1 || dstW < 8 || dstH < 1) {
00908         /* FIXME check if these are enough and try to lower them after
00909          * fixing the relevant parts of the code */
00910         av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
00911                srcW, srcH, dstW, dstH);
00912         return AVERROR(EINVAL);
00913     }
00914 
00915     if (!dstFilter)
00916         dstFilter = &dummyFilter;
00917     if (!srcFilter)
00918         srcFilter = &dummyFilter;
00919 
00920     c->lumXInc      = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
00921     c->lumYInc      = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
00922     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
00923     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
00924     c->vRounder     = 4 * 0x0001000100010001ULL;
00925 
00926     usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
00927                   (srcFilter->chrV && srcFilter->chrV->length > 1) ||
00928                   (dstFilter->lumV && dstFilter->lumV->length > 1) ||
00929                   (dstFilter->chrV && dstFilter->chrV->length > 1);
00930     usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
00931                   (srcFilter->chrH && srcFilter->chrH->length > 1) ||
00932                   (dstFilter->lumH && dstFilter->lumH->length > 1) ||
00933                   (dstFilter->chrH && dstFilter->chrH->length > 1);
00934 
00935     getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
00936     getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
00937 
00938 
00939     if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
00940         if (dstW&1) {
00941             av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
00942             flags |= SWS_FULL_CHR_H_INT;
00943             c->flags = flags;
00944         }
00945     }
00946     /* reuse chroma for 2 pixels RGB/BGR unless user wants full
00947      * chroma interpolation */
00948     if (flags & SWS_FULL_CHR_H_INT &&
00949         isAnyRGB(dstFormat)        &&
00950         dstFormat != PIX_FMT_RGBA  &&
00951         dstFormat != PIX_FMT_ARGB  &&
00952         dstFormat != PIX_FMT_BGRA  &&
00953         dstFormat != PIX_FMT_ABGR  &&
00954         dstFormat != PIX_FMT_RGB24 &&
00955         dstFormat != PIX_FMT_BGR24) {
00956         av_log(c, AV_LOG_WARNING,
00957                "full chroma interpolation for destination format '%s' not yet implemented\n",
00958                av_get_pix_fmt_name(dstFormat));
00959         flags   &= ~SWS_FULL_CHR_H_INT;
00960         c->flags = flags;
00961     }
00962     if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
00963         c->chrDstHSubSample = 1;
00964 
00965     // drop some chroma lines if the user wants it
00966     c->vChrDrop          = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
00967                            SWS_SRC_V_CHR_DROP_SHIFT;
00968     c->chrSrcVSubSample += c->vChrDrop;
00969 
00970     /* drop every other pixel for chroma calculation unless user
00971      * wants full chroma */
00972     if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP)   &&
00973         srcFormat != PIX_FMT_RGB8 && srcFormat != PIX_FMT_BGR8 &&
00974         srcFormat != PIX_FMT_RGB4 && srcFormat != PIX_FMT_BGR4 &&
00975         srcFormat != PIX_FMT_RGB4_BYTE && srcFormat != PIX_FMT_BGR4_BYTE &&
00976         ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
00977          (flags & SWS_FAST_BILINEAR)))
00978         c->chrSrcHSubSample = 1;
00979 
00980     // Note the -((-x)>>y) is so that we always round toward +inf.
00981     c->chrSrcW = -((-srcW) >> c->chrSrcHSubSample);
00982     c->chrSrcH = -((-srcH) >> c->chrSrcVSubSample);
00983     c->chrDstW = -((-dstW) >> c->chrDstHSubSample);
00984     c->chrDstH = -((-dstH) >> c->chrDstVSubSample);
00985 
00986     /* unscaled special cases */
00987     if (unscaled && !usesHFilter && !usesVFilter &&
00988         (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
00989         ff_get_unscaled_swscale(c);
00990 
00991         if (c->swScale) {
00992             if (flags & SWS_PRINT_INFO)
00993                 av_log(c, AV_LOG_INFO,
00994                        "using unscaled %s -> %s special converter\n",
00995                        av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
00996             return 0;
00997         }
00998     }
00999 
01000     c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;
01001     if (c->srcBpc < 8)
01002         c->srcBpc = 8;
01003     c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;
01004     if (c->dstBpc < 8)
01005         c->dstBpc = 8;
01006     if (isAnyRGB(srcFormat) || srcFormat == PIX_FMT_PAL8)
01007         c->srcBpc = 16;
01008     if (c->dstBpc == 16)
01009         dst_stride <<= 1;
01010     FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
01011     if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 &&
01012         c->srcBpc == 8 && c->dstBpc <= 10) {
01013         c->canMMX2BeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
01014                             (srcW & 15) == 0) ? 1 : 0;
01015         if (!c->canMMX2BeUsed && dstW >= srcW && (srcW & 15) == 0
01016             && (flags & SWS_FAST_BILINEAR)) {
01017             if (flags & SWS_PRINT_INFO)
01018                 av_log(c, AV_LOG_INFO,
01019                        "output width is not a multiple of 32 -> no MMX2 scaler\n");
01020         }
01021         if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
01022             c->canMMX2BeUsed=0;
01023     } else
01024         c->canMMX2BeUsed = 0;
01025 
01026     c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
01027     c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;
01028 
01029     /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
01030      * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
01031      * correct scaling.
01032      * n-2 is the last chrominance sample available.
01033      * This is not perfect, but no one should notice the difference, the more
01034      * correct variant would be like the vertical one, but that would require
01035      * some special code for the first and last pixel */
01036     if (flags & SWS_FAST_BILINEAR) {
01037         if (c->canMMX2BeUsed) {
01038             c->lumXInc += 20;
01039             c->chrXInc += 20;
01040         }
01041         // we don't use the x86 asm scaler if MMX is available
01042         else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX && c->dstBpc <= 10) {
01043             c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;
01044             c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
01045         }
01046     }
01047 
01048     /* precalculate horizontal scaler filter coefficients */
01049     {
01050 #if HAVE_MMX2
01051 // can't downscale !!!
01052         if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
01053             c->lumMmx2FilterCodeSize = initMMX2HScaler(dstW, c->lumXInc, NULL,
01054                                                        NULL, NULL, 8);
01055             c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc,
01056                                                        NULL, NULL, NULL, 4);
01057 
01058 #ifdef MAP_ANONYMOUS
01059             c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
01060             c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
01061 #elif HAVE_VIRTUALALLOC
01062             c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
01063             c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
01064 #else
01065             c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
01066             c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
01067 #endif
01068 
01069 #ifdef MAP_ANONYMOUS
01070             if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)
01071 #else
01072             if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
01073 #endif
01074             {
01075                 av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
01076                 return AVERROR(ENOMEM);
01077             }
01078 
01079             FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);
01080             FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);
01081             FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);
01082             FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
01083 
01084             initMMX2HScaler(      dstW, c->lumXInc, c->lumMmx2FilterCode,
01085                             c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
01086             initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode,
01087                             c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
01088 
01089 #ifdef MAP_ANONYMOUS
01090             mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
01091             mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
01092 #endif
01093         } else
01094 #endif /* HAVE_MMX2 */
01095         {
01096             const int filterAlign =
01097                 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
01098                 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
01099                 1;
01100 
01101             if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
01102                            &c->hLumFilterSize, c->lumXInc,
01103                            srcW, dstW, filterAlign, 1 << 14,
01104                            (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
01105                            cpu_flags, srcFilter->lumH, dstFilter->lumH,
01106                            c->param) < 0)
01107                 goto fail;
01108             if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
01109                            &c->hChrFilterSize, c->chrXInc,
01110                            c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
01111                            (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
01112                            cpu_flags, srcFilter->chrH, dstFilter->chrH,
01113                            c->param) < 0)
01114                 goto fail;
01115         }
01116     } // initialize horizontal stuff
01117 
01118     /* precalculate vertical scaler filter coefficients */
01119     {
01120         const int filterAlign =
01121             (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 :
01122             (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
01123             1;
01124 
01125         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
01126                        c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
01127                        (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
01128                        cpu_flags, srcFilter->lumV, dstFilter->lumV,
01129                        c->param) < 0)
01130             goto fail;
01131         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
01132                        c->chrYInc, c->chrSrcH, c->chrDstH,
01133                        filterAlign, (1 << 12),
01134                        (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
01135                        cpu_flags, srcFilter->chrV, dstFilter->chrV,
01136                        c->param) < 0)
01137             goto fail;
01138 
01139 #if HAVE_ALTIVEC
01140         FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);
01141         FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
01142 
01143         for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
01144             int j;
01145             short *p = (short *)&c->vYCoeffsBank[i];
01146             for (j = 0; j < 8; j++)
01147                 p[j] = c->vLumFilter[i];
01148         }
01149 
01150         for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
01151             int j;
01152             short *p = (short *)&c->vCCoeffsBank[i];
01153             for (j = 0; j < 8; j++)
01154                 p[j] = c->vChrFilter[i];
01155         }
01156 #endif
01157     }
01158 
01159     // calculate buffer sizes so that they won't run out while handling these damn slices
01160     c->vLumBufSize = c->vLumFilterSize;
01161     c->vChrBufSize = c->vChrFilterSize;
01162     for (i = 0; i < dstH; i++) {
01163         int chrI      = (int64_t)i * c->chrDstH / dstH;
01164         int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
01165                               ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
01166                                << c->chrSrcVSubSample));
01167 
01168         nextSlice >>= c->chrSrcVSubSample;
01169         nextSlice <<= c->chrSrcVSubSample;
01170         if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
01171             c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
01172         if (c->vChrFilterPos[chrI] + c->vChrBufSize <
01173             (nextSlice >> c->chrSrcVSubSample))
01174             c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
01175                              c->vChrFilterPos[chrI];
01176     }
01177 
01178     /* Allocate pixbufs (we use dynamic allocation because otherwise we would
01179      * need to allocate several megabytes to handle all possible cases) */
01180     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);
01181     FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
01182     FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
01183     if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
01184         FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
01185     /* Note we need at least one pixel more at the end because of the MMX code
01186      * (just in case someone wants to replace the 4000/8000). */
01187     /* align at 16 bytes for AltiVec */
01188     for (i = 0; i < c->vLumBufSize; i++) {
01189         FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
01190                           dst_stride + 16, fail);
01191         c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
01192     }
01193     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
01194     c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
01195     c->uv_offx2 = dst_stride + 16;
01196     for (i = 0; i < c->vChrBufSize; i++) {
01197         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
01198                          dst_stride * 2 + 32, fail);
01199         c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
01200         c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
01201                          = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
01202     }
01203     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
01204         for (i = 0; i < c->vLumBufSize; i++) {
01205             FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
01206                               dst_stride + 16, fail);
01207             c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
01208         }
01209 
01210     // try to avoid drawing green stuff between the right end and the stride end
01211     for (i = 0; i < c->vChrBufSize; i++)
01212         if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){
01213             av_assert0(c->dstBpc > 10);
01214             for(j=0; j<dst_stride/2+1; j++)
01215                 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
01216         } else
01217             for(j=0; j<dst_stride+1; j++)
01218                 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
01219 
01220     assert(c->chrDstH <= dstH);
01221 
01222     if (flags & SWS_PRINT_INFO) {
01223         if (flags & SWS_FAST_BILINEAR)
01224             av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
01225         else if (flags & SWS_BILINEAR)
01226             av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
01227         else if (flags & SWS_BICUBIC)
01228             av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
01229         else if (flags & SWS_X)
01230             av_log(c, AV_LOG_INFO, "Experimental scaler, ");
01231         else if (flags & SWS_POINT)
01232             av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
01233         else if (flags & SWS_AREA)
01234             av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
01235         else if (flags & SWS_BICUBLIN)
01236             av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
01237         else if (flags & SWS_GAUSS)
01238             av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
01239         else if (flags & SWS_SINC)
01240             av_log(c, AV_LOG_INFO, "Sinc scaler, ");
01241         else if (flags & SWS_LANCZOS)
01242             av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
01243         else if (flags & SWS_SPLINE)
01244             av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
01245         else
01246             av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
01247 
01248         av_log(c, AV_LOG_INFO, "from %s to %s%s ",
01249                av_get_pix_fmt_name(srcFormat),
01250 #ifdef DITHER1XBPP
01251                dstFormat == PIX_FMT_BGR555   || dstFormat == PIX_FMT_BGR565   ||
01252                dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
01253                dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ?
01254                                                              "dithered " : "",
01255 #else
01256                "",
01257 #endif
01258                av_get_pix_fmt_name(dstFormat));
01259 
01260         if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2)
01261             av_log(c, AV_LOG_INFO, "using MMX2\n");
01262         else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)
01263             av_log(c, AV_LOG_INFO, "using 3DNOW\n");
01264         else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX)
01265             av_log(c, AV_LOG_INFO, "using MMX\n");
01266         else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC)
01267             av_log(c, AV_LOG_INFO, "using AltiVec\n");
01268         else
01269             av_log(c, AV_LOG_INFO, "using C\n");
01270 
01271         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
01272         av_log(c, AV_LOG_DEBUG,
01273                "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
01274                c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
01275         av_log(c, AV_LOG_DEBUG,
01276                "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
01277                c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
01278                c->chrXInc, c->chrYInc);
01279     }
01280 
01281     c->swScale = ff_getSwsFunc(c);
01282     return 0;
01283 fail: // FIXME replace things by appropriate error codes
01284     return -1;
01285 }
01286 
01287 #if FF_API_SWS_GETCONTEXT
01288 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
01289                            int dstW, int dstH, enum PixelFormat dstFormat,
01290                            int flags, SwsFilter *srcFilter,
01291                            SwsFilter *dstFilter, const double *param)
01292 {
01293     SwsContext *c;
01294 
01295     if (!(c = sws_alloc_context()))
01296         return NULL;
01297 
01298     c->flags     = flags;
01299     c->srcW      = srcW;
01300     c->srcH      = srcH;
01301     c->dstW      = dstW;
01302     c->dstH      = dstH;
01303     c->srcRange  = handle_jpeg(&srcFormat);
01304     c->dstRange  = handle_jpeg(&dstFormat);
01305     c->src0Alpha = handle_0alpha(&srcFormat);
01306     c->dst0Alpha = handle_0alpha(&dstFormat);
01307     c->srcFormat = srcFormat;
01308     c->dstFormat = dstFormat;
01309 
01310     if (param) {
01311         c->param[0] = param[0];
01312         c->param[1] = param[1];
01313     }
01314     sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
01315                              ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
01316                              c->dstRange, 0, 1 << 16, 1 << 16);
01317 
01318     if (sws_init_context(c, srcFilter, dstFilter) < 0) {
01319         sws_freeContext(c);
01320         return NULL;
01321     }
01322 
01323     return c;
01324 }
01325 #endif
01326 
01327 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
01328                                 float lumaSharpen, float chromaSharpen,
01329                                 float chromaHShift, float chromaVShift,
01330                                 int verbose)
01331 {
01332     SwsFilter *filter = av_malloc(sizeof(SwsFilter));
01333     if (!filter)
01334         return NULL;
01335 
01336     if (lumaGBlur != 0.0) {
01337         filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
01338         filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
01339     } else {
01340         filter->lumH = sws_getIdentityVec();
01341         filter->lumV = sws_getIdentityVec();
01342     }
01343 
01344     if (chromaGBlur != 0.0) {
01345         filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
01346         filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
01347     } else {
01348         filter->chrH = sws_getIdentityVec();
01349         filter->chrV = sws_getIdentityVec();
01350     }
01351 
01352     if (chromaSharpen != 0.0) {
01353         SwsVector *id = sws_getIdentityVec();
01354         sws_scaleVec(filter->chrH, -chromaSharpen);
01355         sws_scaleVec(filter->chrV, -chromaSharpen);
01356         sws_addVec(filter->chrH, id);
01357         sws_addVec(filter->chrV, id);
01358         sws_freeVec(id);
01359     }
01360 
01361     if (lumaSharpen != 0.0) {
01362         SwsVector *id = sws_getIdentityVec();
01363         sws_scaleVec(filter->lumH, -lumaSharpen);
01364         sws_scaleVec(filter->lumV, -lumaSharpen);
01365         sws_addVec(filter->lumH, id);
01366         sws_addVec(filter->lumV, id);
01367         sws_freeVec(id);
01368     }
01369 
01370     if (chromaHShift != 0.0)
01371         sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
01372 
01373     if (chromaVShift != 0.0)
01374         sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
01375 
01376     sws_normalizeVec(filter->chrH, 1.0);
01377     sws_normalizeVec(filter->chrV, 1.0);
01378     sws_normalizeVec(filter->lumH, 1.0);
01379     sws_normalizeVec(filter->lumV, 1.0);
01380 
01381     if (verbose)
01382         sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
01383     if (verbose)
01384         sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
01385 
01386     return filter;
01387 }
01388 
01389 SwsVector *sws_allocVec(int length)
01390 {
01391     SwsVector *vec = av_malloc(sizeof(SwsVector));
01392     if (!vec)
01393         return NULL;
01394     vec->length = length;
01395     vec->coeff  = av_malloc(sizeof(double) * length);
01396     if (!vec->coeff)
01397         av_freep(&vec);
01398     return vec;
01399 }
01400 
01401 SwsVector *sws_getGaussianVec(double variance, double quality)
01402 {
01403     const int length = (int)(variance * quality + 0.5) | 1;
01404     int i;
01405     double middle  = (length - 1) * 0.5;
01406     SwsVector *vec = sws_allocVec(length);
01407 
01408     if (!vec)
01409         return NULL;
01410 
01411     for (i = 0; i < length; i++) {
01412         double dist = i - middle;
01413         vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
01414                         sqrt(2 * variance * M_PI);
01415     }
01416 
01417     sws_normalizeVec(vec, 1.0);
01418 
01419     return vec;
01420 }
01421 
01422 SwsVector *sws_getConstVec(double c, int length)
01423 {
01424     int i;
01425     SwsVector *vec = sws_allocVec(length);
01426 
01427     if (!vec)
01428         return NULL;
01429 
01430     for (i = 0; i < length; i++)
01431         vec->coeff[i] = c;
01432 
01433     return vec;
01434 }
01435 
01436 SwsVector *sws_getIdentityVec(void)
01437 {
01438     return sws_getConstVec(1.0, 1);
01439 }
01440 
01441 static double sws_dcVec(SwsVector *a)
01442 {
01443     int i;
01444     double sum = 0;
01445 
01446     for (i = 0; i < a->length; i++)
01447         sum += a->coeff[i];
01448 
01449     return sum;
01450 }
01451 
01452 void sws_scaleVec(SwsVector *a, double scalar)
01453 {
01454     int i;
01455 
01456     for (i = 0; i < a->length; i++)
01457         a->coeff[i] *= scalar;
01458 }
01459 
01460 void sws_normalizeVec(SwsVector *a, double height)
01461 {
01462     sws_scaleVec(a, height / sws_dcVec(a));
01463 }
01464 
01465 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
01466 {
01467     int length = a->length + b->length - 1;
01468     int i, j;
01469     SwsVector *vec = sws_getConstVec(0.0, length);
01470 
01471     if (!vec)
01472         return NULL;
01473 
01474     for (i = 0; i < a->length; i++) {
01475         for (j = 0; j < b->length; j++) {
01476             vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
01477         }
01478     }
01479 
01480     return vec;
01481 }
01482 
01483 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
01484 {
01485     int length = FFMAX(a->length, b->length);
01486     int i;
01487     SwsVector *vec = sws_getConstVec(0.0, length);
01488 
01489     if (!vec)
01490         return NULL;
01491 
01492     for (i = 0; i < a->length; i++)
01493         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
01494     for (i = 0; i < b->length; i++)
01495         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
01496 
01497     return vec;
01498 }
01499 
01500 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
01501 {
01502     int length = FFMAX(a->length, b->length);
01503     int i;
01504     SwsVector *vec = sws_getConstVec(0.0, length);
01505 
01506     if (!vec)
01507         return NULL;
01508 
01509     for (i = 0; i < a->length; i++)
01510         vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
01511     for (i = 0; i < b->length; i++)
01512         vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
01513 
01514     return vec;
01515 }
01516 
01517 /* shift left / or right if "shift" is negative */
01518 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
01519 {
01520     int length = a->length + FFABS(shift) * 2;
01521     int i;
01522     SwsVector *vec = sws_getConstVec(0.0, length);
01523 
01524     if (!vec)
01525         return NULL;
01526 
01527     for (i = 0; i < a->length; i++) {
01528         vec->coeff[i + (length    - 1) / 2 -
01529                        (a->length - 1) / 2 - shift] = a->coeff[i];
01530     }
01531 
01532     return vec;
01533 }
01534 
01535 void sws_shiftVec(SwsVector *a, int shift)
01536 {
01537     SwsVector *shifted = sws_getShiftedVec(a, shift);
01538     av_free(a->coeff);
01539     a->coeff  = shifted->coeff;
01540     a->length = shifted->length;
01541     av_free(shifted);
01542 }
01543 
01544 void sws_addVec(SwsVector *a, SwsVector *b)
01545 {
01546     SwsVector *sum = sws_sumVec(a, b);
01547     av_free(a->coeff);
01548     a->coeff  = sum->coeff;
01549     a->length = sum->length;
01550     av_free(sum);
01551 }
01552 
01553 void sws_subVec(SwsVector *a, SwsVector *b)
01554 {
01555     SwsVector *diff = sws_diffVec(a, b);
01556     av_free(a->coeff);
01557     a->coeff  = diff->coeff;
01558     a->length = diff->length;
01559     av_free(diff);
01560 }
01561 
01562 void sws_convVec(SwsVector *a, SwsVector *b)
01563 {
01564     SwsVector *conv = sws_getConvVec(a, b);
01565     av_free(a->coeff);
01566     a->coeff  = conv->coeff;
01567     a->length = conv->length;
01568     av_free(conv);
01569 }
01570 
01571 SwsVector *sws_cloneVec(SwsVector *a)
01572 {
01573     int i;
01574     SwsVector *vec = sws_allocVec(a->length);
01575 
01576     if (!vec)
01577         return NULL;
01578 
01579     for (i = 0; i < a->length; i++)
01580         vec->coeff[i] = a->coeff[i];
01581 
01582     return vec;
01583 }
01584 
01585 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
01586 {
01587     int i;
01588     double max = 0;
01589     double min = 0;
01590     double range;
01591 
01592     for (i = 0; i < a->length; i++)
01593         if (a->coeff[i] > max)
01594             max = a->coeff[i];
01595 
01596     for (i = 0; i < a->length; i++)
01597         if (a->coeff[i] < min)
01598             min = a->coeff[i];
01599 
01600     range = max - min;
01601 
01602     for (i = 0; i < a->length; i++) {
01603         int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
01604         av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
01605         for (; x > 0; x--)
01606             av_log(log_ctx, log_level, " ");
01607         av_log(log_ctx, log_level, "|\n");
01608     }
01609 }
01610 
01611 void sws_freeVec(SwsVector *a)
01612 {
01613     if (!a)
01614         return;
01615     av_freep(&a->coeff);
01616     a->length = 0;
01617     av_free(a);
01618 }
01619 
01620 void sws_freeFilter(SwsFilter *filter)
01621 {
01622     if (!filter)
01623         return;
01624 
01625     if (filter->lumH)
01626         sws_freeVec(filter->lumH);
01627     if (filter->lumV)
01628         sws_freeVec(filter->lumV);
01629     if (filter->chrH)
01630         sws_freeVec(filter->chrH);
01631     if (filter->chrV)
01632         sws_freeVec(filter->chrV);
01633     av_free(filter);
01634 }
01635 
01636 void sws_freeContext(SwsContext *c)
01637 {
01638     int i;
01639     if (!c)
01640         return;
01641 
01642     if (c->lumPixBuf) {
01643         for (i = 0; i < c->vLumBufSize; i++)
01644             av_freep(&c->lumPixBuf[i]);
01645         av_freep(&c->lumPixBuf);
01646     }
01647 
01648     if (c->chrUPixBuf) {
01649         for (i = 0; i < c->vChrBufSize; i++)
01650             av_freep(&c->chrUPixBuf[i]);
01651         av_freep(&c->chrUPixBuf);
01652         av_freep(&c->chrVPixBuf);
01653     }
01654 
01655     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
01656         for (i = 0; i < c->vLumBufSize; i++)
01657             av_freep(&c->alpPixBuf[i]);
01658         av_freep(&c->alpPixBuf);
01659     }
01660 
01661     av_freep(&c->vLumFilter);
01662     av_freep(&c->vChrFilter);
01663     av_freep(&c->hLumFilter);
01664     av_freep(&c->hChrFilter);
01665 #if HAVE_ALTIVEC
01666     av_freep(&c->vYCoeffsBank);
01667     av_freep(&c->vCCoeffsBank);
01668 #endif
01669 
01670     av_freep(&c->vLumFilterPos);
01671     av_freep(&c->vChrFilterPos);
01672     av_freep(&c->hLumFilterPos);
01673     av_freep(&c->hChrFilterPos);
01674 
01675 #if HAVE_MMX
01676 #ifdef MAP_ANONYMOUS
01677     if (c->lumMmx2FilterCode)
01678         munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
01679     if (c->chrMmx2FilterCode)
01680         munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
01681 #elif HAVE_VIRTUALALLOC
01682     if (c->lumMmx2FilterCode)
01683         VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
01684     if (c->chrMmx2FilterCode)
01685         VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
01686 #else
01687     av_free(c->lumMmx2FilterCode);
01688     av_free(c->chrMmx2FilterCode);
01689 #endif
01690     c->lumMmx2FilterCode = NULL;
01691     c->chrMmx2FilterCode = NULL;
01692 #endif /* HAVE_MMX */
01693 
01694     av_freep(&c->yuvTable);
01695     av_freep(&c->formatConvBuffer);
01696 
01697     av_free(c);
01698 }
01699 
01700 struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
01701                                         int srcH, enum PixelFormat srcFormat,
01702                                         int dstW, int dstH,
01703                                         enum PixelFormat dstFormat, int flags,
01704                                         SwsFilter *srcFilter,
01705                                         SwsFilter *dstFilter,
01706                                         const double *param)
01707 {
01708     static const double default_param[2] = { SWS_PARAM_DEFAULT,
01709                                              SWS_PARAM_DEFAULT };
01710 
01711     if (!param)
01712         param = default_param;
01713 
01714     if (context &&
01715         (context->srcW      != srcW      ||
01716          context->srcH      != srcH      ||
01717          context->srcFormat != srcFormat ||
01718          context->dstW      != dstW      ||
01719          context->dstH      != dstH      ||
01720          context->dstFormat != dstFormat ||
01721          context->flags     != flags     ||
01722          context->param[0]  != param[0]  ||
01723          context->param[1]  != param[1])) {
01724         sws_freeContext(context);
01725         context = NULL;
01726     }
01727 
01728     if (!context) {
01729         if (!(context = sws_alloc_context()))
01730             return NULL;
01731         context->srcW      = srcW;
01732         context->srcH      = srcH;
01733         context->srcRange  = handle_jpeg(&srcFormat);
01734         context->src0Alpha = handle_0alpha(&srcFormat);
01735         context->srcFormat = srcFormat;
01736         context->dstW      = dstW;
01737         context->dstH      = dstH;
01738         context->dstRange  = handle_jpeg(&dstFormat);
01739         context->dst0Alpha = handle_0alpha(&dstFormat);
01740         context->dstFormat = dstFormat;
01741         context->flags     = flags;
01742         context->param[0]  = param[0];
01743         context->param[1]  = param[1];
01744         sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
01745                                  context->srcRange,
01746                                  ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/,
01747                                  context->dstRange, 0, 1 << 16, 1 << 16);
01748         if (sws_init_context(context, srcFilter, dstFilter) < 0) {
01749             sws_freeContext(context);
01750             return NULL;
01751         }
01752     }
01753     return context;
01754 }