utils.c 75.8 KB
Newer Older
1 2 3 4 5
/*
 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
 *
 * This file is part of FFmpeg.
 *
6 7 8 9
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
10 11 12
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
15
 *
16 17
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
18 19 20
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

21 22 23
#include "config.h"

#define _SVID_SOURCE // needed for MAP_ANONYMOUS
24
#define _DARWIN_C_SOURCE // needed for MAP_ANON
25 26 27
#include <inttypes.h>
#include <math.h>
#include <stdio.h>
28
#include <string.h>
29 30 31 32 33 34 35 36 37 38
#if HAVE_SYS_MMAN_H
#include <sys/mman.h>
#if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
#define MAP_ANONYMOUS MAP_ANON
#endif
#endif
#if HAVE_VIRTUALALLOC
#define WIN32_LEAN_AND_MEAN
#include <windows.h>
#endif
39

40
#include "libavutil/attributes.h"
41
#include "libavutil/avassert.h"
42 43
#include "libavutil/avutil.h"
#include "libavutil/bswap.h"
44 45
#include "libavutil/cpu.h"
#include "libavutil/intreadwrite.h"
46
#include "libavutil/mathematics.h"
47
#include "libavutil/opt.h"
48
#include "libavutil/pixdesc.h"
49
#include "libavutil/ppc/cpu.h"
50
#include "libavutil/x86/asm.h"
51
#include "libavutil/x86/cpu.h"
52 53 54
#include "rgb2rgb.h"
#include "swscale.h"
#include "swscale_internal.h"
55

56 57
static void handle_formats(SwsContext *c);

58 59
unsigned swscale_version(void)
{
60
    av_assert0(LIBSWSCALE_VERSION_MICRO >= 100);
61 62 63 64 65
    return LIBSWSCALE_VERSION_INT;
}

const char *swscale_configuration(void)
{
66
    return FFMPEG_CONFIGURATION;
67 68 69 70 71
}

const char *swscale_license(void)
{
#define LICENSE_PREFIX "libswscale license: "
72
    return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
73 74
}

75
#define RET 0xC3 // near return opcode for x86
76

77
typedef struct FormatEntry {
78 79 80
    uint8_t is_supported_in         :1;
    uint8_t is_supported_out        :1;
    uint8_t is_supported_endianness :1;
81 82
} FormatEntry;

83 84 85 86 87 88 89 90 91 92 93 94 95 96
static const FormatEntry format_entries[AV_PIX_FMT_NB] = {
    [AV_PIX_FMT_YUV420P]     = { 1, 1 },
    [AV_PIX_FMT_YUYV422]     = { 1, 1 },
    [AV_PIX_FMT_RGB24]       = { 1, 1 },
    [AV_PIX_FMT_BGR24]       = { 1, 1 },
    [AV_PIX_FMT_YUV422P]     = { 1, 1 },
    [AV_PIX_FMT_YUV444P]     = { 1, 1 },
    [AV_PIX_FMT_YUV410P]     = { 1, 1 },
    [AV_PIX_FMT_YUV411P]     = { 1, 1 },
    [AV_PIX_FMT_GRAY8]       = { 1, 1 },
    [AV_PIX_FMT_MONOWHITE]   = { 1, 1 },
    [AV_PIX_FMT_MONOBLACK]   = { 1, 1 },
    [AV_PIX_FMT_PAL8]        = { 1, 0 },
    [AV_PIX_FMT_YUVJ420P]    = { 1, 1 },
Michael Niedermayer's avatar
Michael Niedermayer committed
97
    [AV_PIX_FMT_YUVJ411P]    = { 1, 1 },
98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113
    [AV_PIX_FMT_YUVJ422P]    = { 1, 1 },
    [AV_PIX_FMT_YUVJ444P]    = { 1, 1 },
    [AV_PIX_FMT_UYVY422]     = { 1, 1 },
    [AV_PIX_FMT_UYYVYY411]   = { 0, 0 },
    [AV_PIX_FMT_BGR8]        = { 1, 1 },
    [AV_PIX_FMT_BGR4]        = { 0, 1 },
    [AV_PIX_FMT_BGR4_BYTE]   = { 1, 1 },
    [AV_PIX_FMT_RGB8]        = { 1, 1 },
    [AV_PIX_FMT_RGB4]        = { 0, 1 },
    [AV_PIX_FMT_RGB4_BYTE]   = { 1, 1 },
    [AV_PIX_FMT_NV12]        = { 1, 1 },
    [AV_PIX_FMT_NV21]        = { 1, 1 },
    [AV_PIX_FMT_ARGB]        = { 1, 1 },
    [AV_PIX_FMT_RGBA]        = { 1, 1 },
    [AV_PIX_FMT_ABGR]        = { 1, 1 },
    [AV_PIX_FMT_BGRA]        = { 1, 1 },
114 115 116 117
    [AV_PIX_FMT_0RGB]        = { 1, 1 },
    [AV_PIX_FMT_RGB0]        = { 1, 1 },
    [AV_PIX_FMT_0BGR]        = { 1, 1 },
    [AV_PIX_FMT_BGR0]        = { 1, 1 },
118 119 120 121 122
    [AV_PIX_FMT_GRAY16BE]    = { 1, 1 },
    [AV_PIX_FMT_GRAY16LE]    = { 1, 1 },
    [AV_PIX_FMT_YUV440P]     = { 1, 1 },
    [AV_PIX_FMT_YUVJ440P]    = { 1, 1 },
    [AV_PIX_FMT_YUVA420P]    = { 1, 1 },
123 124
    [AV_PIX_FMT_YUVA422P]    = { 1, 1 },
    [AV_PIX_FMT_YUVA444P]    = { 1, 1 },
125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142
    [AV_PIX_FMT_YUVA420P9BE] = { 1, 1 },
    [AV_PIX_FMT_YUVA420P9LE] = { 1, 1 },
    [AV_PIX_FMT_YUVA422P9BE] = { 1, 1 },
    [AV_PIX_FMT_YUVA422P9LE] = { 1, 1 },
    [AV_PIX_FMT_YUVA444P9BE] = { 1, 1 },
    [AV_PIX_FMT_YUVA444P9LE] = { 1, 1 },
    [AV_PIX_FMT_YUVA420P10BE]= { 1, 1 },
    [AV_PIX_FMT_YUVA420P10LE]= { 1, 1 },
    [AV_PIX_FMT_YUVA422P10BE]= { 1, 1 },
    [AV_PIX_FMT_YUVA422P10LE]= { 1, 1 },
    [AV_PIX_FMT_YUVA444P10BE]= { 1, 1 },
    [AV_PIX_FMT_YUVA444P10LE]= { 1, 1 },
    [AV_PIX_FMT_YUVA420P16BE]= { 1, 1 },
    [AV_PIX_FMT_YUVA420P16LE]= { 1, 1 },
    [AV_PIX_FMT_YUVA422P16BE]= { 1, 1 },
    [AV_PIX_FMT_YUVA422P16LE]= { 1, 1 },
    [AV_PIX_FMT_YUVA444P16BE]= { 1, 1 },
    [AV_PIX_FMT_YUVA444P16LE]= { 1, 1 },
143 144
    [AV_PIX_FMT_RGB48BE]     = { 1, 1 },
    [AV_PIX_FMT_RGB48LE]     = { 1, 1 },
Paul B Mahol's avatar
Paul B Mahol committed
145 146
    [AV_PIX_FMT_RGBA64BE]    = { 1, 1 },
    [AV_PIX_FMT_RGBA64LE]    = { 1, 1 },
147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167
    [AV_PIX_FMT_RGB565BE]    = { 1, 1 },
    [AV_PIX_FMT_RGB565LE]    = { 1, 1 },
    [AV_PIX_FMT_RGB555BE]    = { 1, 1 },
    [AV_PIX_FMT_RGB555LE]    = { 1, 1 },
    [AV_PIX_FMT_BGR565BE]    = { 1, 1 },
    [AV_PIX_FMT_BGR565LE]    = { 1, 1 },
    [AV_PIX_FMT_BGR555BE]    = { 1, 1 },
    [AV_PIX_FMT_BGR555LE]    = { 1, 1 },
    [AV_PIX_FMT_YUV420P16LE] = { 1, 1 },
    [AV_PIX_FMT_YUV420P16BE] = { 1, 1 },
    [AV_PIX_FMT_YUV422P16LE] = { 1, 1 },
    [AV_PIX_FMT_YUV422P16BE] = { 1, 1 },
    [AV_PIX_FMT_YUV444P16LE] = { 1, 1 },
    [AV_PIX_FMT_YUV444P16BE] = { 1, 1 },
    [AV_PIX_FMT_RGB444LE]    = { 1, 1 },
    [AV_PIX_FMT_RGB444BE]    = { 1, 1 },
    [AV_PIX_FMT_BGR444LE]    = { 1, 1 },
    [AV_PIX_FMT_BGR444BE]    = { 1, 1 },
    [AV_PIX_FMT_Y400A]       = { 1, 0 },
    [AV_PIX_FMT_BGR48BE]     = { 1, 1 },
    [AV_PIX_FMT_BGR48LE]     = { 1, 1 },
168 169
    [AV_PIX_FMT_BGRA64BE]    = { 0, 0 },
    [AV_PIX_FMT_BGRA64LE]    = { 0, 0 },
170 171 172 173
    [AV_PIX_FMT_YUV420P9BE]  = { 1, 1 },
    [AV_PIX_FMT_YUV420P9LE]  = { 1, 1 },
    [AV_PIX_FMT_YUV420P10BE] = { 1, 1 },
    [AV_PIX_FMT_YUV420P10LE] = { 1, 1 },
174 175 176 177
    [AV_PIX_FMT_YUV420P12BE] = { 1, 1 },
    [AV_PIX_FMT_YUV420P12LE] = { 1, 1 },
    [AV_PIX_FMT_YUV420P14BE] = { 1, 1 },
    [AV_PIX_FMT_YUV420P14LE] = { 1, 1 },
178 179 180 181
    [AV_PIX_FMT_YUV422P9BE]  = { 1, 1 },
    [AV_PIX_FMT_YUV422P9LE]  = { 1, 1 },
    [AV_PIX_FMT_YUV422P10BE] = { 1, 1 },
    [AV_PIX_FMT_YUV422P10LE] = { 1, 1 },
182 183 184 185
    [AV_PIX_FMT_YUV422P12BE] = { 1, 1 },
    [AV_PIX_FMT_YUV422P12LE] = { 1, 1 },
    [AV_PIX_FMT_YUV422P14BE] = { 1, 1 },
    [AV_PIX_FMT_YUV422P14LE] = { 1, 1 },
186 187 188 189
    [AV_PIX_FMT_YUV444P9BE]  = { 1, 1 },
    [AV_PIX_FMT_YUV444P9LE]  = { 1, 1 },
    [AV_PIX_FMT_YUV444P10BE] = { 1, 1 },
    [AV_PIX_FMT_YUV444P10LE] = { 1, 1 },
190 191 192 193
    [AV_PIX_FMT_YUV444P12BE] = { 1, 1 },
    [AV_PIX_FMT_YUV444P12LE] = { 1, 1 },
    [AV_PIX_FMT_YUV444P14BE] = { 1, 1 },
    [AV_PIX_FMT_YUV444P14LE] = { 1, 1 },
194
    [AV_PIX_FMT_GBRP]        = { 1, 1 },
195 196 197 198 199 200 201 202
    [AV_PIX_FMT_GBRP9LE]     = { 1, 1 },
    [AV_PIX_FMT_GBRP9BE]     = { 1, 1 },
    [AV_PIX_FMT_GBRP10LE]    = { 1, 1 },
    [AV_PIX_FMT_GBRP10BE]    = { 1, 1 },
    [AV_PIX_FMT_GBRP12LE]    = { 1, 1 },
    [AV_PIX_FMT_GBRP12BE]    = { 1, 1 },
    [AV_PIX_FMT_GBRP14LE]    = { 1, 1 },
    [AV_PIX_FMT_GBRP14BE]    = { 1, 1 },
203 204
    [AV_PIX_FMT_GBRP16LE]    = { 1, 0 },
    [AV_PIX_FMT_GBRP16BE]    = { 1, 0 },
205 206
    [AV_PIX_FMT_XYZ12BE]     = { 1, 1, 1 },
    [AV_PIX_FMT_XYZ12LE]     = { 1, 1, 1 },
207 208 209
    [AV_PIX_FMT_GBRAP]       = { 1, 1 },
    [AV_PIX_FMT_GBRAP16LE]   = { 1, 0 },
    [AV_PIX_FMT_GBRAP16BE]   = { 1, 0 },
210 211
};

212
int sws_isSupportedInput(enum AVPixelFormat pix_fmt)
213
{
214
    return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
215
           format_entries[pix_fmt].is_supported_in : 0;
216 217
}

218
int sws_isSupportedOutput(enum AVPixelFormat pix_fmt)
219
{
220
    return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
221
           format_entries[pix_fmt].is_supported_out : 0;
222 223
}

224 225 226 227 228 229
int sws_isSupportedEndiannessConversion(enum AVPixelFormat pix_fmt)
{
    return (unsigned)pix_fmt < AV_PIX_FMT_NB ?
           format_entries[pix_fmt].is_supported_endianness : 0;
}

230
#if FF_API_SWS_FORMAT_NAME
231
const char *sws_format_name(enum AVPixelFormat format)
232
{
233 234 235
    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(format);
    if (desc)
        return desc->name;
236 237
    else
        return "Unknown format";
238 239 240
}
#endif

241 242
static double getSplineCoeff(double a, double b, double c, double d,
                             double dist)
243
{
244 245 246 247 248 249 250 251
    if (dist <= 1.0)
        return ((d * dist + c) * dist + b) * dist + a;
    else
        return getSplineCoeff(0.0,
                               b + 2.0 * c + 3.0 * d,
                               c + 3.0 * d,
                              -b - 3.0 * c - 6.0 * d,
                              dist - 1.0);
252 253
}

254 255 256 257 258 259 260 261 262
static av_cold int get_local_pos(SwsContext *s, int chr_subsample, int pos, int dir)
{
    if (pos < 0) {
        pos = (128 << chr_subsample) - 128;
    }
    pos += 128; // relative to ideal left edge
    return pos >> chr_subsample;
}

263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282
typedef struct {
    int flag;                   ///< flag associated to the algorithm
    const char *description;    ///< human-readable description
    int size_factor;            ///< size factor used when initing the filters
} ScaleAlgorithm;

static const ScaleAlgorithm scale_algorithms[] = {
    { SWS_AREA,          "area averaging",                  1 /* downscale only, for upscale it is bilinear */ },
    { SWS_BICUBIC,       "bicubic",                         4 },
    { SWS_BICUBLIN,      "luma bicubic / chroma bilinear", -1 },
    { SWS_BILINEAR,      "bilinear",                        2 },
    { SWS_FAST_BILINEAR, "fast bilinear",                  -1 },
    { SWS_GAUSS,         "Gaussian",                        8 /* infinite ;) */ },
    { SWS_LANCZOS,       "Lanczos",                        -1 /* custom */ },
    { SWS_POINT,         "nearest neighbor / point",       -1 },
    { SWS_SINC,          "sinc",                           20 /* infinite ;) */ },
    { SWS_SPLINE,        "bicubic spline",                 20 /* infinite :)*/ },
    { SWS_X,             "experimental",                    8 },
};

283 284 285 286 287
static av_cold int initFilter(int16_t **outFilter, int32_t **filterPos,
                              int *outFilterSize, int xInc, int srcW,
                              int dstW, int filterAlign, int one,
                              int flags, int cpu_flags,
                              SwsVector *srcFilter, SwsVector *dstFilter,
288
                              double param[2], int srcPos, int dstPos)
289 290 291 292 293
{
    int i;
    int filterSize;
    int filter2Size;
    int minFilterSize;
294 295
    int64_t *filter    = NULL;
    int64_t *filter2   = NULL;
296
    const int64_t fone = 1LL << (54 - FFMIN(av_log2(srcW/dstW), 8));
297
    int ret            = -1;
298

299
    emms_c(); // FIXME should not be required but IS (even for non-MMX versions)
300

301 302
    // NOTE: the +3 is for the MMX(+1) / SSE(+3) scaler which reads over the end
    FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW + 3) * sizeof(**filterPos), fail);
303

304
    if (FFABS(xInc - 0x10000) < 10 && srcPos == dstPos) { // unscaled
305
        int i;
306 307 308
        filterSize = 1;
        FF_ALLOCZ_OR_GOTO(NULL, filter,
                          dstW * sizeof(*filter) * filterSize, fail);
309

310 311 312
        for (i = 0; i < dstW; i++) {
            filter[i * filterSize] = fone;
            (*filterPos)[i]        = i;
313
        }
314
    } else if (flags & SWS_POINT) { // lame looking point sampling mode
315
        int i;
316
        int64_t xDstInSrc;
317 318 319
        filterSize = 1;
        FF_ALLOC_OR_GOTO(NULL, filter,
                         dstW * sizeof(*filter) * filterSize, fail);
320

321
        xDstInSrc = ((dstPos*(int64_t)xInc)>>8) - ((srcPos*0x8000LL)>>7);
322 323
        for (i = 0; i < dstW; i++) {
            int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
324

325 326 327
            (*filterPos)[i] = xx;
            filter[i]       = fone;
            xDstInSrc      += xInc;
328
        }
329 330
    } else if ((xInc <= (1 << 16) && (flags & SWS_AREA)) ||
               (flags & SWS_FAST_BILINEAR)) { // bilinear upscale
331
        int i;
332
        int64_t xDstInSrc;
333 334 335
        filterSize = 2;
        FF_ALLOC_OR_GOTO(NULL, filter,
                         dstW * sizeof(*filter) * filterSize, fail);
336

337
        xDstInSrc = ((dstPos*(int64_t)xInc)>>8) - ((srcPos*0x8000LL)>>7);
338 339
        for (i = 0; i < dstW; i++) {
            int xx = (xDstInSrc - ((filterSize - 1) << 15) + (1 << 15)) >> 16;
340 341
            int j;

342 343 344
            (*filterPos)[i] = xx;
            // bilinear upscale / linear interpolate / area averaging
            for (j = 0; j < filterSize; j++) {
345
                int64_t coeff= fone - FFABS(((int64_t)xx<<16) - xDstInSrc)*(fone>>16);
346 347 348
                if (coeff < 0)
                    coeff = 0;
                filter[i * filterSize + j] = coeff;
349 350
                xx++;
            }
351
            xDstInSrc += xInc;
352 353
        }
    } else {
354
        int64_t xDstInSrc;
355 356 357 358 359 360 361
        int sizeFactor = -1;

        for (i = 0; i < FF_ARRAY_ELEMS(scale_algorithms); i++) {
            if (flags & scale_algorithms[i].flag) {
                sizeFactor = scale_algorithms[i].size_factor;
                break;
            }
362
        }
363 364 365
        if (flags & SWS_LANCZOS)
            sizeFactor = param[0] != SWS_PARAM_DEFAULT ? ceil(2 * param[0]) : 6;
        av_assert0(sizeFactor > 0);
366

367 368 369 370
        if (xInc <= 1 << 16)
            filterSize = 1 + sizeFactor;    // upscale
        else
            filterSize = 1 + (sizeFactor * srcW + dstW - 1) / dstW;
371

372 373
        filterSize = FFMIN(filterSize, srcW - 2);
        filterSize = FFMAX(filterSize, 1);
374

375 376
        FF_ALLOC_OR_GOTO(NULL, filter,
                         dstW * sizeof(*filter) * filterSize, fail);
377

378
        xDstInSrc = ((dstPos*(int64_t)xInc)>>7) - ((srcPos*0x10000LL)>>7);
379 380
        for (i = 0; i < dstW; i++) {
            int xx = (xDstInSrc - ((filterSize - 2) << 16)) / (1 << 17);
381
            int j;
382 383 384
            (*filterPos)[i] = xx;
            for (j = 0; j < filterSize; j++) {
                int64_t d = (FFABS(((int64_t)xx << 17) - xDstInSrc)) << 13;
385 386 387
                double floatd;
                int64_t coeff;

388 389 390
                if (xInc > 1 << 16)
                    d = d * dstW / srcW;
                floatd = d * (1.0 / (1 << 30));
391 392

                if (flags & SWS_BICUBIC) {
393 394
                    int64_t B = (param[0] != SWS_PARAM_DEFAULT ? param[0] :   0) * (1 << 24);
                    int64_t C = (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1 << 24);
395

396
                    if (d >= 1LL << 31) {
397 398 399 400 401
                        coeff = 0.0;
                    } else {
                        int64_t dd  = (d  * d) >> 30;
                        int64_t ddd = (dd * d) >> 30;

402 403 404 405
                        if (d < 1LL << 30)
                            coeff =  (12 * (1 << 24) -  9 * B - 6 * C) * ddd +
                                    (-18 * (1 << 24) + 12 * B + 6 * C) *  dd +
                                      (6 * (1 << 24) -  2 * B)         * (1 << 30);
406
                        else
407 408 409 410
                            coeff =      (-B -  6 * C) * ddd +
                                      (6 * B + 30 * C) * dd  +
                                    (-12 * B - 48 * C) * d   +
                                      (8 * B + 24 * C) * (1 << 30);
411
                    }
412
                    coeff /= (1LL<<54)/fone;
413
                }
414 415 416 417 418
#if 0
                else if (flags & SWS_X) {
                    double p  = param ? param * 0.01 : 0.3;
                    coeff     = d ? sin(d * M_PI) / (d * M_PI) : 1.0;
                    coeff    *= pow(2.0, -p * d * d);
419
                }
420
#endif
421
                else if (flags & SWS_X) {
422
                    double A = param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
423 424
                    double c;

425 426
                    if (floatd < 1.0)
                        c = cos(floatd * M_PI);
427
                    else
428 429 430
                        c = -1.0;
                    if (c < 0.0)
                        c = -pow(-c, A);
431
                    else
432 433
                        c = pow(c, A);
                    coeff = (c * 0.5 + 0.5) * fone;
434
                } else if (flags & SWS_AREA) {
435 436 437 438 439 440 441 442
                    int64_t d2 = d - (1 << 29);
                    if (d2 * xInc < -(1LL << (29 + 16)))
                        coeff = 1.0 * (1LL << (30 + 16));
                    else if (d2 * xInc < (1LL << (29 + 16)))
                        coeff = -d2 * xInc + (1LL << (29 + 16));
                    else
                        coeff = 0.0;
                    coeff *= fone >> (30 + 16);
443
                } else if (flags & SWS_GAUSS) {
444 445
                    double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
                    coeff = (pow(2.0, -p * floatd * floatd)) * fone;
446
                } else if (flags & SWS_SINC) {
447
                    coeff = (d ? sin(floatd * M_PI) / (floatd * M_PI) : 1.0) * fone;
448
                } else if (flags & SWS_LANCZOS) {
449 450 451 452 453
                    double p = param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
                    coeff = (d ? sin(floatd * M_PI) * sin(floatd * M_PI / p) /
                             (floatd * floatd * M_PI * M_PI / p) : 1.0) * fone;
                    if (floatd > p)
                        coeff = 0;
454
                } else if (flags & SWS_BILINEAR) {
455 456 457
                    coeff = (1 << 30) - d;
                    if (coeff < 0)
                        coeff = 0;
458 459
                    coeff *= fone >> 30;
                } else if (flags & SWS_SPLINE) {
460 461
                    double p = -2.196152422706632;
                    coeff = getSplineCoeff(1.0, 0.0, p, -p - 1.0, floatd) * fone;
462
                } else {
463
                    av_assert0(0);
464 465
                }

466
                filter[i * filterSize + j] = coeff;
467 468
                xx++;
            }
469
            xDstInSrc += 2 * xInc;
470 471 472 473
        }
    }

    /* apply src & dst Filter to filter -> filter2
474 475
     * av_free(filter);
     */
476
    av_assert0(filterSize > 0);
477 478 479 480 481
    filter2Size = filterSize;
    if (srcFilter)
        filter2Size += srcFilter->length - 1;
    if (dstFilter)
        filter2Size += dstFilter->length - 1;
482
    av_assert0(filter2Size > 0);
483 484 485
    FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size * dstW * sizeof(*filter2), fail);

    for (i = 0; i < dstW; i++) {
486 487
        int j, k;

488 489 490 491 492
        if (srcFilter) {
            for (k = 0; k < srcFilter->length; k++) {
                for (j = 0; j < filterSize; j++)
                    filter2[i * filter2Size + k + j] +=
                        srcFilter->coeff[k] * filter[i * filterSize + j];
493 494
            }
        } else {
495 496
            for (j = 0; j < filterSize; j++)
                filter2[i * filter2Size + j] = filter[i * filterSize + j];
497
        }
498
        // FIXME dstFilter
499

500
        (*filterPos)[i] += (filterSize - 1) / 2 - (filter2Size - 1) / 2;
501 502 503 504 505
    }
    av_freep(&filter);

    /* try to reduce the filter-size (step1 find size and shift left) */
    // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
506 507 508
    minFilterSize = 0;
    for (i = dstW - 1; i >= 0; i--) {
        int min = filter2Size;
509
        int j;
510
        int64_t cutOff = 0.0;
511 512

        /* get rid of near zero elements on the left by shifting left */
513
        for (j = 0; j < filter2Size; j++) {
514
            int k;
515
            cutOff += FFABS(filter2[i * filter2Size]);
516

517 518
            if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
                break;
519

520 521 522 523
            /* preserve monotonicity because the core can't handle the
             * filter otherwise */
            if (i < dstW - 1 && (*filterPos)[i] >= (*filterPos)[i + 1])
                break;
524 525

            // move filter coefficients left
526 527 528
            for (k = 1; k < filter2Size; k++)
                filter2[i * filter2Size + k - 1] = filter2[i * filter2Size + k];
            filter2[i * filter2Size + k - 1] = 0;
529 530 531
            (*filterPos)[i]++;
        }

532
        cutOff = 0;
533
        /* count near zeros on the right */
534 535
        for (j = filter2Size - 1; j > 0; j--) {
            cutOff += FFABS(filter2[i * filter2Size + j]);
536

537 538
            if (cutOff > SWS_MAX_REDUCE_CUTOFF * fone)
                break;
539 540 541
            min--;
        }

542 543
        if (min > minFilterSize)
            minFilterSize = min;
544 545
    }

546
    if (PPC_ALTIVEC(cpu_flags)) {
547
        // we can handle the special case 4, so we don't want to go the full 8
548 549 550
        if (minFilterSize < 5)
            filterAlign = 4;

551 552 553
        /* We really don't want to waste our time doing useless computation, so
         * fall back on the scalar C code for very small filters.
         * Vectorizing is worth it only if you have a decent-sized vector. */
554 555 556 557
        if (minFilterSize < 3)
            filterAlign = 1;
    }

558
    if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
559 560
        // special case for unscaled vertical filtering
        if (minFilterSize == 1 && filterAlign == 2)
561
            filterAlign = 1;
562 563
    }

564
    av_assert0(minFilterSize > 0);
565
    filterSize = (minFilterSize + (filterAlign - 1)) & (~(filterAlign - 1));
566
    av_assert0(filterSize > 0);
567 568
    filter = av_malloc(filterSize * dstW * sizeof(*filter));
    if (filterSize >= MAX_FILTER_SIZE * 16 /
569
                      ((flags & SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter) {
Stefano Sabatini's avatar
Stefano Sabatini committed
570
        av_log(NULL, AV_LOG_ERROR, "sws: filterSize %d is too large, try less extreme scaling or increase MAX_FILTER_SIZE and recompile\n", filterSize);
571
        goto fail;
572
    }
573
    *outFilterSize = filterSize;
574

575 576 577 578
    if (flags & SWS_PRINT_INFO)
        av_log(NULL, AV_LOG_VERBOSE,
               "SwScaler: reducing / aligning filtersize %d -> %d\n",
               filter2Size, filterSize);
579
    /* try to reduce the filter-size (step2 reduce it) */
580
    for (i = 0; i < dstW; i++) {
581 582
        int j;

583 584 585 586 587 588 589
        for (j = 0; j < filterSize; j++) {
            if (j >= filter2Size)
                filter[i * filterSize + j] = 0;
            else
                filter[i * filterSize + j] = filter2[i * filter2Size + j];
            if ((flags & SWS_BITEXACT) && j >= minFilterSize)
                filter[i * filterSize + j] = 0;
590 591 592
        }
    }

593
    // FIXME try to align filterPos if possible
594

595
    // fix borders
596
    for (i = 0; i < dstW; i++) {
597 598 599
        int j;
        if ((*filterPos)[i] < 0) {
            // move filter coefficients left to compensate for filterPos
600 601 602 603
            for (j = 1; j < filterSize; j++) {
                int left = FFMAX(j + (*filterPos)[i], 0);
                filter[i * filterSize + left] += filter[i * filterSize + j];
                filter[i * filterSize + j]     = 0;
604 605 606 607 608
            }
            (*filterPos)[i]= 0;
        }

        if ((*filterPos)[i] + filterSize > srcW) {
609
            int shift = (*filterPos)[i] + filterSize - srcW;
610
            // move filter coefficients right to compensate for filterPos
611 612 613 614
            for (j = filterSize - 2; j >= 0; j--) {
                int right = FFMIN(j + shift, filterSize - 1);
                filter[i * filterSize + right] += filter[i * filterSize + j];
                filter[i * filterSize + j]      = 0;
615 616 617 618 619 620 621
            }
            (*filterPos)[i]= srcW - filterSize;
        }
    }

    // Note the +1 is for the MMX scaler which reads over the end
    /* align at 16 for AltiVec (needed by hScale_altivec_real) */
622 623
    FF_ALLOCZ_OR_GOTO(NULL, *outFilter,
                      *outFilterSize * (dstW + 3) * sizeof(int16_t), fail);
624 625

    /* normalize & store in outFilter */
626
    for (i = 0; i < dstW; i++) {
627
        int j;
628 629
        int64_t error = 0;
        int64_t sum   = 0;
630

631 632
        for (j = 0; j < filterSize; j++) {
            sum += filter[i * filterSize + j];
633
        }
634
        sum = (sum + one / 2) / one;
635 636 637 638
        if (!sum) {
            av_log(NULL, AV_LOG_WARNING, "SwScaler: zero vector in scaling\n");
            sum = 1;
        }
639 640 641 642 643
        for (j = 0; j < *outFilterSize; j++) {
            int64_t v = filter[i * filterSize + j] + error;
            int intV  = ROUNDED_DIV(v, sum);
            (*outFilter)[i * (*outFilterSize) + j] = intV;
            error                                  = v - intV * sum;
644 645 646
        }
    }

647 648 649 650 651 652
    (*filterPos)[dstW + 0] =
    (*filterPos)[dstW + 1] =
    (*filterPos)[dstW + 2] = (*filterPos)[dstW - 1]; /* the MMX/SSE scaler will
                                                      * read over the end */
    for (i = 0; i < *outFilterSize; i++) {
        int k = (dstW - 1) * (*outFilterSize) + i;
653 654 655
        (*outFilter)[k + 1 * (*outFilterSize)] =
        (*outFilter)[k + 2 * (*outFilterSize)] =
        (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
656 657
    }

658 659
    ret = 0;

660
fail:
661 662
    if(ret < 0)
        av_log(NULL, AV_LOG_ERROR, "sws: initFilter failed\n");
663 664 665 666 667
    av_free(filter);
    av_free(filter2);
    return ret;
}

668
#if HAVE_MMXEXT_INLINE
669 670 671
static av_cold int init_hscaler_mmxext(int dstW, int xInc, uint8_t *filterCode,
                                       int16_t *filter, int32_t *filterPos,
                                       int numSplits)
672 673 674 675 676 677 678 679 680 681 682 683 684 685
{
    uint8_t *fragmentA;
    x86_reg imm8OfPShufW1A;
    x86_reg imm8OfPShufW2A;
    x86_reg fragmentLengthA;
    uint8_t *fragmentB;
    x86_reg imm8OfPShufW1B;
    x86_reg imm8OfPShufW2B;
    x86_reg fragmentLengthB;
    int fragmentPos;

    int xpos, i;

    // create an optimized horizontal scaling routine
686
    /* This scaler is made of runtime-generated MMXEXT code using specially tuned
687 688 689
     * pshufw instructions. For every four output pixels, if four input pixels
     * are enough for the fast bilinear scaling, then a chunk of fragmentB is
     * used. If five input pixels are needed, then a chunk of fragmentA is used.
690 691
     */

692
    // code fragment
693

694
    __asm__ volatile (
695
        "jmp                         9f                 \n\t"
696
        // Begin
697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715
        "0:                                             \n\t"
        "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
        "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
        "movd   1(%%"REG_c", %%"REG_S"), %%mm1          \n\t"
        "punpcklbw                %%mm7, %%mm1          \n\t"
        "punpcklbw                %%mm7, %%mm0          \n\t"
        "pshufw                   $0xFF, %%mm1, %%mm1   \n\t"
        "1:                                             \n\t"
        "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
        "2:                                             \n\t"
        "psubw                    %%mm1, %%mm0          \n\t"
        "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
        "pmullw                   %%mm3, %%mm0          \n\t"
        "psllw                       $7, %%mm1          \n\t"
        "paddw                    %%mm1, %%mm0          \n\t"

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

        "add                         $8, %%"REG_a"      \n\t"
716
        // End
717
        "9:                                             \n\t"
718 719 720 721
        // "int $3                                         \n\t"
        "lea       " LOCAL_MANGLE(0b) ", %0             \n\t"
        "lea       " LOCAL_MANGLE(1b) ", %1             \n\t"
        "lea       " LOCAL_MANGLE(2b) ", %2             \n\t"
722 723 724 725
        "dec                         %1                 \n\t"
        "dec                         %2                 \n\t"
        "sub                         %0, %1             \n\t"
        "sub                         %0, %2             \n\t"
726
        "lea       " LOCAL_MANGLE(9b) ", %3             \n\t"
727 728 729
        "sub                         %0, %3             \n\t"


730 731 732
        : "=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
          "=r" (fragmentLengthA)
        );
733

734
    __asm__ volatile (
735
        "jmp                         9f                 \n\t"
736
        // Begin
737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753
        "0:                                             \n\t"
        "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
        "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
        "punpcklbw                %%mm7, %%mm0          \n\t"
        "pshufw                   $0xFF, %%mm0, %%mm1   \n\t"
        "1:                                             \n\t"
        "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
        "2:                                             \n\t"
        "psubw                    %%mm1, %%mm0          \n\t"
        "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
        "pmullw                   %%mm3, %%mm0          \n\t"
        "psllw                       $7, %%mm1          \n\t"
        "paddw                    %%mm1, %%mm0          \n\t"

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

        "add                         $8, %%"REG_a"      \n\t"
754
        // End
755
        "9:                                             \n\t"
756 757 758 759
        // "int                       $3                   \n\t"
        "lea       " LOCAL_MANGLE(0b) ", %0             \n\t"
        "lea       " LOCAL_MANGLE(1b) ", %1             \n\t"
        "lea       " LOCAL_MANGLE(2b) ", %2             \n\t"
760 761 762 763
        "dec                         %1                 \n\t"
        "dec                         %2                 \n\t"
        "sub                         %0, %1             \n\t"
        "sub                         %0, %2             \n\t"
764
        "lea       " LOCAL_MANGLE(9b) ", %3             \n\t"
765 766 767
        "sub                         %0, %3             \n\t"


768 769 770
        : "=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
          "=r" (fragmentLengthB)
        );
771

772 773
    xpos        = 0; // lumXInc/2 - 0x8000; // difference between pixel centers
    fragmentPos = 0;
774

775 776
    for (i = 0; i < dstW / numSplits; i++) {
        int xx = xpos >> 16;
777

778 779 780 781 782 783 784 785 786 787 788 789
        if ((i & 3) == 0) {
            int a                  = 0;
            int b                  = ((xpos + xInc) >> 16) - xx;
            int c                  = ((xpos + xInc * 2) >> 16) - xx;
            int d                  = ((xpos + xInc * 3) >> 16) - xx;
            int inc                = (d + 1 < 4);
            uint8_t *fragment      = (d + 1 < 4) ? fragmentB : fragmentA;
            x86_reg imm8OfPShufW1  = (d + 1 < 4) ? imm8OfPShufW1B : imm8OfPShufW1A;
            x86_reg imm8OfPShufW2  = (d + 1 < 4) ? imm8OfPShufW2B : imm8OfPShufW2A;
            x86_reg fragmentLength = (d + 1 < 4) ? fragmentLengthB : fragmentLengthA;
            int maxShift           = 3 - (d + inc);
            int shift              = 0;
790 791

            if (filterCode) {
792 793 794 795 796
                filter[i]        = ((xpos              & 0xFFFF) ^ 0xFFFF) >> 9;
                filter[i + 1]    = (((xpos + xInc)     & 0xFFFF) ^ 0xFFFF) >> 9;
                filter[i + 2]    = (((xpos + xInc * 2) & 0xFFFF) ^ 0xFFFF) >> 9;
                filter[i + 3]    = (((xpos + xInc * 3) & 0xFFFF) ^ 0xFFFF) >> 9;
                filterPos[i / 2] = xx;
797 798 799

                memcpy(filterCode + fragmentPos, fragment, fragmentLength);

800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816
                filterCode[fragmentPos + imm8OfPShufW1] =  (a + inc)       |
                                                          ((b + inc) << 2) |
                                                          ((c + inc) << 4) |
                                                          ((d + inc) << 6);
                filterCode[fragmentPos + imm8OfPShufW2] =  a | (b << 2) |
                                                               (c << 4) |
                                                               (d << 6);

                if (i + 4 - inc >= dstW)
                    shift = maxShift;               // avoid overread
                else if ((filterPos[i / 2] & 3) <= maxShift)
                    shift = filterPos[i / 2] & 3;   // align

                if (shift && i >= shift) {
                    filterCode[fragmentPos + imm8OfPShufW1] += 0x55 * shift;
                    filterCode[fragmentPos + imm8OfPShufW2] += 0x55 * shift;
                    filterPos[i / 2]                        -= shift;
817 818 819
                }
            }

820
            fragmentPos += fragmentLength;
821 822

            if (filterCode)
823
                filterCode[fragmentPos] = RET;
824
        }
825
        xpos += xInc;
826 827
    }
    if (filterCode)
828
        filterPos[((i / 2) + 1) & (~1)] = xpos >> 16;  // needed to jump to the next part
829 830 831

    return fragmentPos + 1;
}
832
#endif /* HAVE_MMXEXT_INLINE */
833

834 835
static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)
{
836
    int64_t W, V, Z, Cy, Cu, Cv;
837 838 839 840 841 842
    int64_t vr =  table[0];
    int64_t ub =  table[1];
    int64_t ug = -table[2];
    int64_t vg = -table[3];
    int64_t ONE = 65536;
    int64_t cy = ONE;
843 844 845 846 847 848 849 850 851 852 853 854 855 856 857
    uint8_t *p = (uint8_t*)c->input_rgb2yuv_table;
    int i;
    static const int8_t map[] = {
    BY_IDX, GY_IDX, -1    , BY_IDX, BY_IDX, GY_IDX, -1    , BY_IDX,
    RY_IDX, -1    , GY_IDX, RY_IDX, RY_IDX, -1    , GY_IDX, RY_IDX,
    RY_IDX, GY_IDX, -1    , RY_IDX, RY_IDX, GY_IDX, -1    , RY_IDX,
    BY_IDX, -1    , GY_IDX, BY_IDX, BY_IDX, -1    , GY_IDX, BY_IDX,
    BU_IDX, GU_IDX, -1    , BU_IDX, BU_IDX, GU_IDX, -1    , BU_IDX,
    RU_IDX, -1    , GU_IDX, RU_IDX, RU_IDX, -1    , GU_IDX, RU_IDX,
    RU_IDX, GU_IDX, -1    , RU_IDX, RU_IDX, GU_IDX, -1    , RU_IDX,
    BU_IDX, -1    , GU_IDX, BU_IDX, BU_IDX, -1    , GU_IDX, BU_IDX,
    BV_IDX, GV_IDX, -1    , BV_IDX, BV_IDX, GV_IDX, -1    , BV_IDX,
    RV_IDX, -1    , GV_IDX, RV_IDX, RV_IDX, -1    , GV_IDX, RV_IDX,
    RV_IDX, GV_IDX, -1    , RV_IDX, RV_IDX, GV_IDX, -1    , RV_IDX,
    BV_IDX, -1    , GV_IDX, BV_IDX, BV_IDX, -1    , GV_IDX, BV_IDX,
858 859 860 861 862 863 864 865 866 867 868
    RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX,
    BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX, BY_IDX, RY_IDX,
    GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    ,
    -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX, -1    , GY_IDX,
    RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX,
    BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX, BU_IDX, RU_IDX,
    GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    ,
    -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX, -1    , GU_IDX,
    RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,
    BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,
    GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    ,
869 870 871 872 873 874 875 876 877 878 879 880
    -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, //23
    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //24
    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //25
    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //26
    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //27
    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //28
    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //29
    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //30
    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //31
    BY_IDX, GY_IDX, RY_IDX, -1    , -1    , -1    , -1    , -1    , //32
    BU_IDX, GU_IDX, RU_IDX, -1    , -1    , -1    , -1    , -1    , //33
    BV_IDX, GV_IDX, RV_IDX, -1    , -1    , -1    , -1    , -1    , //34
881
    };
882 883 884 885 886 887 888 889 890 891 892

    dstRange = 0; //FIXME range = 1 is handled elsewhere

    if (!dstRange) {
        cy = cy * 255 / 219;
    } else {
        vr = vr * 224 / 255;
        ub = ub * 224 / 255;
        ug = ug * 224 / 255;
        vg = vg * 224 / 255;
    }
893 894
    W = ROUNDED_DIV(ONE*ONE*ug, ub);
    V = ROUNDED_DIV(ONE*ONE*vg, vr);
895
    Z = ONE*ONE-W-V;
896

897 898 899
    Cy = ROUNDED_DIV(cy*Z, ONE);
    Cu = ROUNDED_DIV(ub*Z, ONE);
    Cv = ROUNDED_DIV(vr*Z, ONE);
900

901 902 903
    c->input_rgb2yuv_table[RY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V        , Cy);
    c->input_rgb2yuv_table[GY_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE  , Cy);
    c->input_rgb2yuv_table[BY_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W        , Cy);
904

905 906 907
    c->input_rgb2yuv_table[RU_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*V        , Cu);
    c->input_rgb2yuv_table[GU_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE  , Cu);
    c->input_rgb2yuv_table[BU_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(Z+W)    , Cu);
908

909 910 911
    c->input_rgb2yuv_table[RV_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*(V+Z)    , Cv);
    c->input_rgb2yuv_table[GV_IDX] = -ROUNDED_DIV((1 << RGB2YUV_SHIFT)*ONE*ONE  , Cv);
    c->input_rgb2yuv_table[BV_IDX] =  ROUNDED_DIV((1 << RGB2YUV_SHIFT)*W        , Cv);
912

913
    if(/*!dstRange && */!memcmp(table, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], sizeof(ff_yuv2rgb_coeffs[SWS_CS_DEFAULT]))) {
914 915 916 917 918 919 920 921 922 923
        c->input_rgb2yuv_table[BY_IDX] =  ((int)(0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
        c->input_rgb2yuv_table[BV_IDX] = (-(int)(0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
        c->input_rgb2yuv_table[BU_IDX] =  ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
        c->input_rgb2yuv_table[GY_IDX] =  ((int)(0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
        c->input_rgb2yuv_table[GV_IDX] = (-(int)(0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
        c->input_rgb2yuv_table[GU_IDX] = (-(int)(0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
        c->input_rgb2yuv_table[RY_IDX] =  ((int)(0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
        c->input_rgb2yuv_table[RV_IDX] =  ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
        c->input_rgb2yuv_table[RU_IDX] = (-(int)(0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
    }
924 925
    for(i=0; i<FF_ARRAY_ELEMS(map); i++)
        AV_WL16(p + 16*4 + 2*i, map[i] >= 0 ? c->input_rgb2yuv_table[map[i]] : 0);
926 927
}

928 929 930 931 932
static void fill_xyztables(struct SwsContext *c)
{
    int i;
    double xyzgamma = XYZ_GAMMA;
    double rgbgamma = 1.0 / RGB_GAMMA;
933 934
    double xyzgammainv = 1.0 / XYZ_GAMMA;
    double rgbgammainv = RGB_GAMMA;
935 936 937 938
    static const int16_t xyz2rgb_matrix[3][4] = {
        {13270, -6295, -2041},
        {-3969,  7682,   170},
        {  228,  -835,  4329} };
939 940 941 942 943
    static const int16_t rgb2xyz_matrix[3][4] = {
        {1689, 1464,  739},
        { 871, 2929,  296},
        {  79,  488, 3891} };
    static int16_t xyzgamma_tab[4096], rgbgamma_tab[4096], xyzgammainv_tab[4096], rgbgammainv_tab[4096];
944 945

    memcpy(c->xyz2rgb_matrix, xyz2rgb_matrix, sizeof(c->xyz2rgb_matrix));
946
    memcpy(c->rgb2xyz_matrix, rgb2xyz_matrix, sizeof(c->rgb2xyz_matrix));
947 948
    c->xyzgamma = xyzgamma_tab;
    c->rgbgamma = rgbgamma_tab;
949 950
    c->xyzgammainv = xyzgammainv_tab;
    c->rgbgammainv = rgbgammainv_tab;
951 952 953

    if (rgbgamma_tab[4095])
        return;
954 955 956

    /* set gamma vectors */
    for (i = 0; i < 4096; i++) {
957 958
        xyzgamma_tab[i] = lrint(pow(i / 4095.0, xyzgamma) * 4095.0);
        rgbgamma_tab[i] = lrint(pow(i / 4095.0, rgbgamma) * 4095.0);
959 960
        xyzgammainv_tab[i] = lrint(pow(i / 4095.0, xyzgammainv) * 4095.0);
        rgbgammainv_tab[i] = lrint(pow(i / 4095.0, rgbgammainv) * 4095.0);
961 962 963
    }
}

964 965 966
int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
                             int srcRange, const int table[4], int dstRange,
                             int brightness, int contrast, int saturation)
967
{
968 969
    const AVPixFmtDescriptor *desc_dst;
    const AVPixFmtDescriptor *desc_src;
970 971
    memmove(c->srcColorspaceTable, inv_table, sizeof(int) * 4);
    memmove(c->dstColorspaceTable, table, sizeof(int) * 4);
972

973 974 975 976
    handle_formats(c);
    desc_dst = av_pix_fmt_desc_get(c->dstFormat);
    desc_src = av_pix_fmt_desc_get(c->srcFormat);

977 978 979 980 981
    if(!isYUV(c->dstFormat) && !isGray(c->dstFormat))
        dstRange = 0;
    if(!isYUV(c->srcFormat) && !isGray(c->srcFormat))
        srcRange = 0;

982 983 984 985 986
    c->brightness = brightness;
    c->contrast   = contrast;
    c->saturation = saturation;
    c->srcRange   = srcRange;
    c->dstRange   = dstRange;
987

988 989
    fill_xyztables(c);

990
    if ((isYUV(c->dstFormat) || isGray(c->dstFormat)) && (isYUV(c->srcFormat) || isGray(c->srcFormat)))
991
        return -1;
992

993 994
    c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
    c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
995

996
    if (!isYUV(c->dstFormat) && !isGray(c->dstFormat)) {
997 998 999
        ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness,
                                 contrast, saturation);
        // FIXME factorize
1000

1001 1002 1003
        if (ARCH_PPC)
            ff_yuv2rgb_init_tables_ppc(c, inv_table, brightness,
                                       contrast, saturation);
1004
    }
1005 1006 1007

    fill_rgb2yuv_table(c, table, dstRange);

1008 1009 1010
    return 0;
}

1011 1012 1013
int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
                             int *srcRange, int **table, int *dstRange,
                             int *brightness, int *contrast, int *saturation)
1014
{
1015
    if (!c )
1016
        return -1;
1017

1018 1019 1020 1021 1022 1023 1024
    *inv_table  = c->srcColorspaceTable;
    *table      = c->dstColorspaceTable;
    *srcRange   = c->srcRange;
    *dstRange   = c->dstRange;
    *brightness = c->brightness;
    *contrast   = c->contrast;
    *saturation = c->saturation;
1025 1026 1027 1028

    return 0;
}

1029
static int handle_jpeg(enum AVPixelFormat *format)
1030 1031
{
    switch (*format) {
1032 1033
    case AV_PIX_FMT_YUVJ420P:
        *format = AV_PIX_FMT_YUV420P;
1034
        return 1;
Michael Niedermayer's avatar
Michael Niedermayer committed
1035 1036 1037
    case AV_PIX_FMT_YUVJ411P:
        *format = AV_PIX_FMT_YUV411P;
        return 1;
1038 1039
    case AV_PIX_FMT_YUVJ422P:
        *format = AV_PIX_FMT_YUV422P;
1040
        return 1;
1041 1042
    case AV_PIX_FMT_YUVJ444P:
        *format = AV_PIX_FMT_YUV444P;
1043
        return 1;
1044 1045
    case AV_PIX_FMT_YUVJ440P:
        *format = AV_PIX_FMT_YUV440P;
1046
        return 1;
1047 1048
    case AV_PIX_FMT_GRAY8:
        return 1;
1049 1050
    default:
        return 0;
1051 1052 1053
    }
}

1054
static int handle_0alpha(enum AVPixelFormat *format)
1055 1056
{
    switch (*format) {
1057 1058 1059 1060
    case AV_PIX_FMT_0BGR    : *format = AV_PIX_FMT_ABGR   ; return 1;
    case AV_PIX_FMT_BGR0    : *format = AV_PIX_FMT_BGRA   ; return 4;
    case AV_PIX_FMT_0RGB    : *format = AV_PIX_FMT_ARGB   ; return 1;
    case AV_PIX_FMT_RGB0    : *format = AV_PIX_FMT_RGBA   ; return 4;
1061
    default:                                          return 0;
1062 1063 1064
    }
}

1065 1066 1067 1068 1069 1070 1071 1072 1073
static int handle_xyz(enum AVPixelFormat *format)
{
    switch (*format) {
    case AV_PIX_FMT_XYZ12BE : *format = AV_PIX_FMT_RGB48BE; return 1;
    case AV_PIX_FMT_XYZ12LE : *format = AV_PIX_FMT_RGB48LE; return 1;
    default:                                                return 0;
    }
}

1074 1075
static void handle_formats(SwsContext *c)
{
1076 1077 1078 1079
    c->src0Alpha |= handle_0alpha(&c->srcFormat);
    c->dst0Alpha |= handle_0alpha(&c->dstFormat);
    c->srcXYZ    |= handle_xyz(&c->srcFormat);
    c->dstXYZ    |= handle_xyz(&c->dstFormat);
1080 1081
}

1082 1083
SwsContext *sws_alloc_context(void)
{
1084
    SwsContext *c = av_mallocz(sizeof(SwsContext));
1085

1086 1087 1088 1089
    if (c) {
        c->av_class = &sws_context_class;
        av_opt_set_defaults(c);
    }
1090 1091 1092 1093

    return c;
}

1094 1095
av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
                             SwsFilter *dstFilter)
1096
{
1097
    int i, j;
1098 1099
    int usesVFilter, usesHFilter;
    int unscaled;
1100 1101 1102 1103 1104
    SwsFilter dummyFilter = { NULL, NULL, NULL, NULL };
    int srcW              = c->srcW;
    int srcH              = c->srcH;
    int dstW              = c->dstW;
    int dstH              = c->dstH;
1105
    int dst_stride        = FFALIGN(dstW * sizeof(int16_t) + 66, 16);
1106
    int flags, cpu_flags;
1107 1108
    enum AVPixelFormat srcFormat = c->srcFormat;
    enum AVPixelFormat dstFormat = c->dstFormat;
1109 1110
    const AVPixFmtDescriptor *desc_src;
    const AVPixFmtDescriptor *desc_dst;
1111

1112 1113
    cpu_flags = av_get_cpu_flags();
    flags     = c->flags;
1114
    emms_c();
1115 1116
    if (!rgb15to16)
        sws_rgb2rgb_init();
1117 1118 1119

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

1120 1121 1122 1123 1124 1125 1126 1127
    c->srcRange |= handle_jpeg(&c->srcFormat);
    c->dstRange |= handle_jpeg(&c->dstFormat);

    if (!c->contrast && !c->saturation && !c->dstFormatBpp)
        sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange,
                                 ff_yuv2rgb_coeffs[SWS_CS_DEFAULT],
                                 c->dstRange, 0, 1 << 16, 1 << 16);

1128 1129
    if(srcFormat!=c->srcFormat || dstFormat!=c->dstFormat)
        av_log(c, AV_LOG_WARNING, "deprecated pixel format used, make sure you did set range correctly\n");
1130 1131 1132 1133 1134
    handle_formats(c);
    srcFormat = c->srcFormat;
    dstFormat = c->dstFormat;
    desc_src = av_pix_fmt_desc_get(srcFormat);
    desc_dst = av_pix_fmt_desc_get(dstFormat);
1135

1136 1137
    if (!(unscaled && sws_isSupportedEndiannessConversion(srcFormat) &&
          av_pix_fmt_swap_endianness(srcFormat) == dstFormat)) {
1138
    if (!sws_isSupportedInput(srcFormat)) {
1139
        av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n",
1140
               av_get_pix_fmt_name(srcFormat));
1141
        return AVERROR(EINVAL);
1142
    }
1143
    if (!sws_isSupportedOutput(dstFormat)) {
1144
        av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n",
1145
               av_get_pix_fmt_name(dstFormat));
1146
        return AVERROR(EINVAL);
1147
    }
1148
    }
1149

1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160
    i = flags & (SWS_POINT         |
                 SWS_AREA          |
                 SWS_BILINEAR      |
                 SWS_FAST_BILINEAR |
                 SWS_BICUBIC       |
                 SWS_X             |
                 SWS_GAUSS         |
                 SWS_LANCZOS       |
                 SWS_SINC          |
                 SWS_SPLINE        |
                 SWS_BICUBLIN);
1161 1162 1163 1164

    /* provide a default scaler if not set by caller */
    if (!i) {
        if (dstW < srcW && dstH < srcH)
1165
            flags |= SWS_BICUBIC;
1166
        else if (dstW > srcW && dstH > srcH)
1167
            flags |= SWS_BICUBIC;
1168
        else
1169
            flags |= SWS_BICUBIC;
1170 1171
        c->flags = flags;
    } else if (i & (i - 1)) {
1172
        av_log(c, AV_LOG_ERROR,
1173
               "Exactly one scaler algorithm must be chosen, got %X\n", i);
1174
        return AVERROR(EINVAL);
1175 1176
    }
    /* sanity check */
1177
    if (srcW < 1 || srcH < 1 || dstW < 1 || dstH < 1) {
1178 1179
        /* FIXME check if these are enough and try to lower them after
         * fixing the relevant parts of the code */
1180
        av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
1181
               srcW, srcH, dstW, dstH);
1182
        return AVERROR(EINVAL);
1183 1184
    }

1185 1186 1187 1188
    if (!dstFilter)
        dstFilter = &dummyFilter;
    if (!srcFilter)
        srcFilter = &dummyFilter;
1189

1190 1191
    c->lumXInc      = (((int64_t)srcW << 16) + (dstW >> 1)) / dstW;
    c->lumYInc      = (((int64_t)srcH << 16) + (dstH >> 1)) / dstH;
1192 1193
    c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
    c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
1194
    c->vRounder     = 4 * 0x0001000100010001ULL;
1195

1196 1197 1198 1199 1200 1201 1202 1203
    usesVFilter = (srcFilter->lumV && srcFilter->lumV->length > 1) ||
                  (srcFilter->chrV && srcFilter->chrV->length > 1) ||
                  (dstFilter->lumV && dstFilter->lumV->length > 1) ||
                  (dstFilter->chrV && dstFilter->chrV->length > 1);
    usesHFilter = (srcFilter->lumH && srcFilter->lumH->length > 1) ||
                  (srcFilter->chrH && srcFilter->chrH->length > 1) ||
                  (dstFilter->lumH && dstFilter->lumH->length > 1) ||
                  (dstFilter->chrH && dstFilter->chrH->length > 1);
1204

1205 1206
    av_pix_fmt_get_chroma_sub_sample(srcFormat, &c->chrSrcHSubSample, &c->chrSrcVSubSample);
    av_pix_fmt_get_chroma_sub_sample(dstFormat, &c->chrDstHSubSample, &c->chrDstVSubSample);
1207

1208 1209 1210 1211 1212
    if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) {
        if (dstW&1) {
            av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to odd output size\n");
            flags |= SWS_FULL_CHR_H_INT;
            c->flags = flags;
1213
        }
1214 1215 1216 1217 1218 1219 1220 1221 1222 1223

        if (   c->chrSrcHSubSample == 0
            && c->chrSrcVSubSample == 0
            && c->dither != SWS_DITHER_BAYER //SWS_FULL_CHR_H_INT is currently not supported with SWS_DITHER_BAYER
            && !(c->flags & SWS_FAST_BILINEAR)
        ) {
            av_log(c, AV_LOG_DEBUG, "Forcing full internal H chroma due to input having non subsampled chroma\n");
            flags |= SWS_FULL_CHR_H_INT;
            c->flags = flags;
        }
1224
    }
1225

1226 1227 1228 1229 1230
    if (c->dither == SWS_DITHER_AUTO) {
        if (flags & SWS_ERROR_DIFFUSION)
            c->dither = SWS_DITHER_ED;
    }

1231 1232 1233 1234
    if(dstFormat == AV_PIX_FMT_BGR4_BYTE ||
       dstFormat == AV_PIX_FMT_RGB4_BYTE ||
       dstFormat == AV_PIX_FMT_BGR8 ||
       dstFormat == AV_PIX_FMT_RGB8) {
1235 1236
        if (c->dither == SWS_DITHER_AUTO)
            c->dither = (flags & SWS_FULL_CHR_H_INT) ? SWS_DITHER_ED : SWS_DITHER_BAYER;
1237 1238 1239 1240 1241 1242 1243 1244
        if (!(flags & SWS_FULL_CHR_H_INT)) {
            if (c->dither == SWS_DITHER_ED) {
                av_log(c, AV_LOG_DEBUG,
                    "Desired dithering only supported in full chroma interpolation for destination format '%s'\n",
                    av_get_pix_fmt_name(dstFormat));
                flags   |= SWS_FULL_CHR_H_INT;
                c->flags = flags;
            }
1245
        }
1246 1247 1248 1249 1250 1251 1252
        if (flags & SWS_FULL_CHR_H_INT) {
            if (c->dither == SWS_DITHER_BAYER) {
                av_log(c, AV_LOG_DEBUG,
                    "Ordered dither is not supported in full chroma interpolation for destination format '%s'\n",
                    av_get_pix_fmt_name(dstFormat));
                c->dither = SWS_DITHER_ED;
            }
1253
        }
1254
    }
1255
    if (isPlanarRGB(dstFormat)) {
1256 1257
        if (!(flags & SWS_FULL_CHR_H_INT)) {
            av_log(c, AV_LOG_DEBUG,
1258 1259
                   "%s output is not supported with half chroma resolution, switching to full\n",
                   av_get_pix_fmt_name(dstFormat));
1260 1261 1262 1263
            flags   |= SWS_FULL_CHR_H_INT;
            c->flags = flags;
        }
    }
1264

1265 1266
    /* reuse chroma for 2 pixels RGB/BGR unless user wants full
     * chroma interpolation */
1267
    if (flags & SWS_FULL_CHR_H_INT &&
1268
        isAnyRGB(dstFormat)        &&
1269
        !isPlanarRGB(dstFormat)    &&
1270 1271 1272 1273 1274
        dstFormat != AV_PIX_FMT_RGBA  &&
        dstFormat != AV_PIX_FMT_ARGB  &&
        dstFormat != AV_PIX_FMT_BGRA  &&
        dstFormat != AV_PIX_FMT_ABGR  &&
        dstFormat != AV_PIX_FMT_RGB24 &&
1275 1276 1277 1278 1279 1280
        dstFormat != AV_PIX_FMT_BGR24 &&
        dstFormat != AV_PIX_FMT_BGR4_BYTE &&
        dstFormat != AV_PIX_FMT_RGB4_BYTE &&
        dstFormat != AV_PIX_FMT_BGR8 &&
        dstFormat != AV_PIX_FMT_RGB8
    ) {
1281 1282
        av_log(c, AV_LOG_WARNING,
               "full chroma interpolation for destination format '%s' not yet implemented\n",
1283
               av_get_pix_fmt_name(dstFormat));
1284
        flags   &= ~SWS_FULL_CHR_H_INT;
1285
        c->flags = flags;
1286
    }
1287 1288
    if (isAnyRGB(dstFormat) && !(flags & SWS_FULL_CHR_H_INT))
        c->chrDstHSubSample = 1;
1289 1290

    // drop some chroma lines if the user wants it
1291 1292 1293 1294 1295 1296 1297
    c->vChrDrop          = (flags & SWS_SRC_V_CHR_DROP_MASK) >>
                           SWS_SRC_V_CHR_DROP_SHIFT;
    c->chrSrcVSubSample += c->vChrDrop;

    /* drop every other pixel for chroma calculation unless user
     * wants full chroma */
    if (isAnyRGB(srcFormat) && !(flags & SWS_FULL_CHR_H_INP)   &&
1298 1299 1300
        srcFormat != AV_PIX_FMT_RGB8 && srcFormat != AV_PIX_FMT_BGR8 &&
        srcFormat != AV_PIX_FMT_RGB4 && srcFormat != AV_PIX_FMT_BGR4 &&
        srcFormat != AV_PIX_FMT_RGB4_BYTE && srcFormat != AV_PIX_FMT_BGR4_BYTE &&
1301 1302 1303 1304
        srcFormat != AV_PIX_FMT_GBRP9BE   && srcFormat != AV_PIX_FMT_GBRP9LE  &&
        srcFormat != AV_PIX_FMT_GBRP10BE  && srcFormat != AV_PIX_FMT_GBRP10LE &&
        srcFormat != AV_PIX_FMT_GBRP12BE  && srcFormat != AV_PIX_FMT_GBRP12LE &&
        srcFormat != AV_PIX_FMT_GBRP14BE  && srcFormat != AV_PIX_FMT_GBRP14LE &&
1305
        srcFormat != AV_PIX_FMT_GBRP16BE  && srcFormat != AV_PIX_FMT_GBRP16LE &&
1306 1307 1308
        ((dstW >> c->chrDstHSubSample) <= (srcW >> 1) ||
         (flags & SWS_FAST_BILINEAR)))
        c->chrSrcHSubSample = 1;
1309

1310 1311 1312 1313 1314
    // Note the FF_CEIL_RSHIFT is so that we always round toward +inf.
    c->chrSrcW = FF_CEIL_RSHIFT(srcW, c->chrSrcHSubSample);
    c->chrSrcH = FF_CEIL_RSHIFT(srcH, c->chrSrcVSubSample);
    c->chrDstW = FF_CEIL_RSHIFT(dstW, c->chrDstHSubSample);
    c->chrDstH = FF_CEIL_RSHIFT(dstH, c->chrDstVSubSample);
1315

1316 1317
    FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);

1318
    /* unscaled special cases */
1319 1320
    if (unscaled && !usesHFilter && !usesVFilter &&
        (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
1321 1322
        ff_get_unscaled_swscale(c);

1323
        if (c->swscale) {
1324 1325 1326
            if (flags & SWS_PRINT_INFO)
                av_log(c, AV_LOG_INFO,
                       "using unscaled %s -> %s special converter\n",
1327
                       av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
1328
            return 0;
1329 1330 1331
        }
    }

1332
    c->srcBpc = 1 + desc_src->comp[0].depth_minus1;
1333 1334
    if (c->srcBpc < 8)
        c->srcBpc = 8;
1335
    c->dstBpc = 1 + desc_dst->comp[0].depth_minus1;
1336 1337
    if (c->dstBpc < 8)
        c->dstBpc = 8;
1338
    if (isAnyRGB(srcFormat) || srcFormat == AV_PIX_FMT_PAL8)
1339
        c->srcBpc = 16;
1340
    if (c->dstBpc == 16)
1341
        dst_stride <<= 1;
1342

1343
    if (INLINE_MMXEXT(cpu_flags) && c->srcBpc == 8 && c->dstBpc <= 14) {
1344 1345 1346
        c->canMMXEXTBeUsed = (dstW >= srcW && (dstW & 31) == 0 &&
                              (srcW & 15) == 0) ? 1 : 0;
        if (!c->canMMXEXTBeUsed && dstW >= srcW && (srcW & 15) == 0
1347

1348 1349 1350
            && (flags & SWS_FAST_BILINEAR)) {
            if (flags & SWS_PRINT_INFO)
                av_log(c, AV_LOG_INFO,
1351
                       "output width is not a multiple of 32 -> no MMXEXT scaler\n");
1352
        }
1353
        if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat))
1354
            c->canMMXEXTBeUsed = 0;
1355
    } else
1356
        c->canMMXEXTBeUsed = 0;
1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368

    c->chrXInc = (((int64_t)c->chrSrcW << 16) + (c->chrDstW >> 1)) / c->chrDstW;
    c->chrYInc = (((int64_t)c->chrSrcH << 16) + (c->chrDstH >> 1)) / c->chrDstH;

    /* Match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src
     * to pixel n-2 of dst, but only for the FAST_BILINEAR mode otherwise do
     * correct scaling.
     * n-2 is the last chrominance sample available.
     * This is not perfect, but no one should notice the difference, the more
     * correct variant would be like the vertical one, but that would require
     * some special code for the first and last pixel */
    if (flags & SWS_FAST_BILINEAR) {
1369
        if (c->canMMXEXTBeUsed) {
1370 1371
            c->lumXInc += 20;
            c->chrXInc += 20;
1372
        }
1373
        // we don't use the x86 asm scaler if MMX is available
1374
        else if (INLINE_MMX(cpu_flags) && c->dstBpc <= 14) {
1375 1376
            c->lumXInc = ((int64_t)(srcW       - 2) << 16) / (dstW       - 2) - 20;
            c->chrXInc = ((int64_t)(c->chrSrcW - 2) << 16) / (c->chrDstW - 2) - 20;
1377 1378 1379
        }
    }

1380 1381
#define USE_MMAP (HAVE_MMAP && HAVE_MPROTECT && defined MAP_ANONYMOUS)

1382 1383
    /* precalculate horizontal scaler filter coefficients */
    {
1384
#if HAVE_MMXEXT_INLINE
1385
// can't downscale !!!
1386
        if (c->canMMXEXTBeUsed && (flags & SWS_FAST_BILINEAR)) {
1387 1388 1389 1390
            c->lumMmxextFilterCodeSize = init_hscaler_mmxext(dstW, c->lumXInc, NULL,
                                                             NULL, NULL, 8);
            c->chrMmxextFilterCodeSize = init_hscaler_mmxext(c->chrDstW, c->chrXInc,
                                                             NULL, NULL, NULL, 4);
1391

1392
#if USE_MMAP
1393 1394 1395 1396 1397 1398 1399 1400
            c->lumMmxextFilterCode = mmap(NULL, c->lumMmxextFilterCodeSize,
                                          PROT_READ | PROT_WRITE,
                                          MAP_PRIVATE | MAP_ANONYMOUS,
                                          -1, 0);
            c->chrMmxextFilterCode = mmap(NULL, c->chrMmxextFilterCodeSize,
                                          PROT_READ | PROT_WRITE,
                                          MAP_PRIVATE | MAP_ANONYMOUS,
                                          -1, 0);
1401
#elif HAVE_VIRTUALALLOC
1402 1403 1404 1405 1406 1407 1408 1409
            c->lumMmxextFilterCode = VirtualAlloc(NULL,
                                                  c->lumMmxextFilterCodeSize,
                                                  MEM_COMMIT,
                                                  PAGE_EXECUTE_READWRITE);
            c->chrMmxextFilterCode = VirtualAlloc(NULL,
                                                  c->chrMmxextFilterCodeSize,
                                                  MEM_COMMIT,
                                                  PAGE_EXECUTE_READWRITE);
1410
#else
1411 1412
            c->lumMmxextFilterCode = av_malloc(c->lumMmxextFilterCodeSize);
            c->chrMmxextFilterCode = av_malloc(c->chrMmxextFilterCodeSize);
1413 1414
#endif

1415
#ifdef MAP_ANONYMOUS
1416
            if (c->lumMmxextFilterCode == MAP_FAILED || c->chrMmxextFilterCode == MAP_FAILED)
1417
#else
1418
            if (!c->lumMmxextFilterCode || !c->chrMmxextFilterCode)
1419
#endif
1420 1421
            {
                av_log(c, AV_LOG_ERROR, "Failed to allocate MMX2FilterCode\n");
1422
                return AVERROR(ENOMEM);
1423
            }
1424

1425 1426 1427 1428
            FF_ALLOCZ_OR_GOTO(c, c->hLumFilter,    (dstW           / 8 + 8) * sizeof(int16_t), fail);
            FF_ALLOCZ_OR_GOTO(c, c->hChrFilter,    (c->chrDstW     / 4 + 8) * sizeof(int16_t), fail);
            FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW       / 2 / 8 + 8) * sizeof(int32_t), fail);
            FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW / 2 / 4 + 8) * sizeof(int32_t), fail);
1429

1430 1431
            init_hscaler_mmxext(      dstW, c->lumXInc, c->lumMmxextFilterCode,
                                c->hLumFilter, (uint32_t*)c->hLumFilterPos, 8);
1432
            init_hscaler_mmxext(c->chrDstW, c->chrXInc, c->chrMmxextFilterCode,
1433
                                c->hChrFilter, (uint32_t*)c->hChrFilterPos, 4);
1434

1435
#if USE_MMAP
1436 1437 1438 1439 1440
            if (   mprotect(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize, PROT_EXEC | PROT_READ) == -1
                || mprotect(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize, PROT_EXEC | PROT_READ) == -1) {
                av_log(c, AV_LOG_ERROR, "mprotect failed, cannot use fast bilinear scaler\n");
                goto fail;
            }
1441 1442
#endif
        } else
1443
#endif /* HAVE_MMXEXT_INLINE */
1444
        {
1445 1446
            const int filterAlign = X86_MMX(cpu_flags)     ? 4 :
                                    PPC_ALTIVEC(cpu_flags) ? 8 : 1;
1447

1448 1449 1450 1451 1452
            if (initFilter(&c->hLumFilter, &c->hLumFilterPos,
                           &c->hLumFilterSize, c->lumXInc,
                           srcW, dstW, filterAlign, 1 << 14,
                           (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
                           cpu_flags, srcFilter->lumH, dstFilter->lumH,
1453 1454 1455
                           c->param,
                           get_local_pos(c, 0, 0, 0),
                           get_local_pos(c, 0, 0, 0)) < 0)
1456
                goto fail;
1457 1458 1459 1460 1461
            if (initFilter(&c->hChrFilter, &c->hChrFilterPos,
                           &c->hChrFilterSize, c->chrXInc,
                           c->chrSrcW, c->chrDstW, filterAlign, 1 << 14,
                           (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
                           cpu_flags, srcFilter->chrH, dstFilter->chrH,
1462 1463 1464
                           c->param,
                           get_local_pos(c, c->chrSrcHSubSample, c->src_h_chr_pos, 0),
                           get_local_pos(c, c->chrDstHSubSample, c->dst_h_chr_pos, 0)) < 0)
1465 1466 1467 1468 1469 1470
                goto fail;
        }
    } // initialize horizontal stuff

    /* precalculate vertical scaler filter coefficients */
    {
1471 1472
        const int filterAlign = X86_MMX(cpu_flags)     ? 2 :
                                PPC_ALTIVEC(cpu_flags) ? 8 : 1;
1473

1474 1475 1476 1477
        if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize,
                       c->lumYInc, srcH, dstH, filterAlign, (1 << 12),
                       (flags & SWS_BICUBLIN) ? (flags | SWS_BICUBIC) : flags,
                       cpu_flags, srcFilter->lumV, dstFilter->lumV,
1478 1479 1480
                       c->param,
                       get_local_pos(c, 0, 0, 1),
                       get_local_pos(c, 0, 0, 1)) < 0)
1481
            goto fail;
1482 1483 1484 1485 1486
        if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize,
                       c->chrYInc, c->chrSrcH, c->chrDstH,
                       filterAlign, (1 << 12),
                       (flags & SWS_BICUBLIN) ? (flags | SWS_BILINEAR) : flags,
                       cpu_flags, srcFilter->chrV, dstFilter->chrV,
1487 1488 1489 1490
                       c->param,
                       get_local_pos(c, c->chrSrcVSubSample, c->src_v_chr_pos, 1),
                       get_local_pos(c, c->chrDstVSubSample, c->dst_v_chr_pos, 1)) < 0)

1491 1492
            goto fail;

1493
#if HAVE_ALTIVEC
1494 1495
        FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof(vector signed short) * c->vLumFilterSize * c->dstH,    fail);
        FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof(vector signed short) * c->vChrFilterSize * c->chrDstH, fail);
1496

1497
        for (i = 0; i < c->vLumFilterSize * c->dstH; i++) {
1498 1499
            int j;
            short *p = (short *)&c->vYCoeffsBank[i];
1500
            for (j = 0; j < 8; j++)
1501 1502 1503
                p[j] = c->vLumFilter[i];
        }

1504
        for (i = 0; i < c->vChrFilterSize * c->chrDstH; i++) {
1505 1506
            int j;
            short *p = (short *)&c->vCCoeffsBank[i];
1507
            for (j = 0; j < 8; j++)
1508 1509 1510 1511 1512 1513
                p[j] = c->vChrFilter[i];
        }
#endif
    }

    // calculate buffer sizes so that they won't run out while handling these damn slices
1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529
    c->vLumBufSize = c->vLumFilterSize;
    c->vChrBufSize = c->vChrFilterSize;
    for (i = 0; i < dstH; i++) {
        int chrI      = (int64_t)i * c->chrDstH / dstH;
        int nextSlice = FFMAX(c->vLumFilterPos[i] + c->vLumFilterSize - 1,
                              ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)
                               << c->chrSrcVSubSample));

        nextSlice >>= c->chrSrcVSubSample;
        nextSlice <<= c->chrSrcVSubSample;
        if (c->vLumFilterPos[i] + c->vLumBufSize < nextSlice)
            c->vLumBufSize = nextSlice - c->vLumFilterPos[i];
        if (c->vChrFilterPos[chrI] + c->vChrBufSize <
            (nextSlice >> c->chrSrcVSubSample))
            c->vChrBufSize = (nextSlice >> c->chrSrcVSubSample) -
                             c->vChrFilterPos[chrI];
1530 1531
    }

1532 1533 1534
    for (i = 0; i < 4; i++)
        FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);

1535 1536 1537 1538 1539
    /* Allocate pixbufs (we use dynamic allocation because otherwise we would
     * need to allocate several megabytes to handle all possible cases) */
    FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);
    FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
    FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize * 3 * sizeof(int16_t *), fail);
1540
    if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1541 1542 1543
        FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize * 3 * sizeof(int16_t *), fail);
    /* Note we need at least one pixel more at the end because of the MMX code
     * (just in case someone wants to replace the 4000/8000). */
1544
    /* align at 16 bytes for AltiVec */
1545 1546 1547 1548
    for (i = 0; i < c->vLumBufSize; i++) {
        FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i + c->vLumBufSize],
                          dst_stride + 16, fail);
        c->lumPixBuf[i] = c->lumPixBuf[i + c->vLumBufSize];
1549
    }
1550
    // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1551
    c->uv_off   = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1552
    c->uv_offx2 = dst_stride + 16;
1553 1554 1555 1556 1557 1558
    for (i = 0; i < c->vChrBufSize; i++) {
        FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i + c->vChrBufSize],
                         dst_stride * 2 + 32, fail);
        c->chrUPixBuf[i] = c->chrUPixBuf[i + c->vChrBufSize];
        c->chrVPixBuf[i] = c->chrVPixBuf[i + c->vChrBufSize]
                         = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1559 1560
    }
    if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1561 1562 1563 1564
        for (i = 0; i < c->vLumBufSize; i++) {
            FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i + c->vLumBufSize],
                              dst_stride + 16, fail);
            c->alpPixBuf[i] = c->alpPixBuf[i + c->vLumBufSize];
1565 1566
        }

1567 1568
    // try to avoid drawing green stuff between the right end and the stride end
    for (i = 0; i < c->vChrBufSize; i++)
1569
        if(desc_dst->comp[0].depth_minus1 == 15){
1570
            av_assert0(c->dstBpc > 14);
1571 1572 1573 1574 1575
            for(j=0; j<dst_stride/2+1; j++)
                ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
        } else
            for(j=0; j<dst_stride+1; j++)
                ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1576

1577
    av_assert0(c->chrDstH <= dstH);
1578

1579
    if (flags & SWS_PRINT_INFO) {
1580
        const char *scaler = NULL, *cpucaps;
1581

1582 1583 1584 1585 1586 1587 1588 1589 1590
        for (i = 0; i < FF_ARRAY_ELEMS(scale_algorithms); i++) {
            if (flags & scale_algorithms[i].flag) {
                scaler = scale_algorithms[i].description;
                break;
            }
        }
        if (!scaler)
            scaler =  "ehh flags invalid?!";
        av_log(c, AV_LOG_INFO, "%s scaler, from %s to %s%s ",
1591
               scaler,
1592
               av_get_pix_fmt_name(srcFormat),
1593
#ifdef DITHER1XBPP
1594 1595 1596
               dstFormat == AV_PIX_FMT_BGR555   || dstFormat == AV_PIX_FMT_BGR565   ||
               dstFormat == AV_PIX_FMT_RGB444BE || dstFormat == AV_PIX_FMT_RGB444LE ||
               dstFormat == AV_PIX_FMT_BGR444BE || dstFormat == AV_PIX_FMT_BGR444LE ?
1597
                                                             "dithered " : "",
1598 1599 1600
#else
               "",
#endif
1601
               av_get_pix_fmt_name(dstFormat));
1602

1603
        if (INLINE_MMXEXT(cpu_flags))
1604
            cpucaps = "MMXEXT";
1605
        else if (INLINE_AMD3DNOW(cpu_flags))
1606
            cpucaps = "3DNOW";
1607
        else if (INLINE_MMX(cpu_flags))
1608
            cpucaps = "MMX";
1609
        else if (PPC_ALTIVEC(cpu_flags))
1610
            cpucaps = "AltiVec";
1611
        else
1612 1613 1614
            cpucaps = "C";

        av_log(c, AV_LOG_INFO, "using %s\n", cpucaps);
1615 1616

        av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1617 1618
        av_log(c, AV_LOG_DEBUG,
               "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1619
               c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1620 1621 1622 1623
        av_log(c, AV_LOG_DEBUG,
               "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
               c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH,
               c->chrXInc, c->chrYInc);
1624 1625
    }

1626
    c->swscale = ff_getSwsFunc(c);
1627
    return 0;
1628
fail: // FIXME replace things by appropriate error codes
1629 1630
    return -1;
}
1631

1632
#if FF_API_SWS_GETCONTEXT
1633 1634
SwsContext *sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat,
                           int dstW, int dstH, enum AVPixelFormat dstFormat,
1635 1636
                           int flags, SwsFilter *srcFilter,
                           SwsFilter *dstFilter, const double *param)
1637 1638 1639
{
    SwsContext *c;

1640
    if (!(c = sws_alloc_context()))
1641 1642
        return NULL;

1643 1644 1645 1646 1647 1648 1649
    c->flags     = flags;
    c->srcW      = srcW;
    c->srcH      = srcH;
    c->dstW      = dstW;
    c->dstH      = dstH;
    c->srcFormat = srcFormat;
    c->dstFormat = dstFormat;
1650 1651 1652 1653 1654 1655

    if (param) {
        c->param[0] = param[0];
        c->param[1] = param[1];
    }

1656
    if (sws_init_context(c, srcFilter, dstFilter) < 0) {
1657 1658 1659 1660 1661
        sws_freeContext(c);
        return NULL;
    }

    return c;
1662
}
1663
#endif
1664 1665 1666 1667 1668 1669

SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
                                float lumaSharpen, float chromaSharpen,
                                float chromaHShift, float chromaVShift,
                                int verbose)
{
1670
    SwsFilter *filter = av_malloc(sizeof(SwsFilter));
1671 1672 1673
    if (!filter)
        return NULL;

1674 1675 1676
    if (lumaGBlur != 0.0) {
        filter->lumH = sws_getGaussianVec(lumaGBlur, 3.0);
        filter->lumV = sws_getGaussianVec(lumaGBlur, 3.0);
1677
    } else {
1678 1679
        filter->lumH = sws_getIdentityVec();
        filter->lumV = sws_getIdentityVec();
1680 1681
    }

1682 1683 1684
    if (chromaGBlur != 0.0) {
        filter->chrH = sws_getGaussianVec(chromaGBlur, 3.0);
        filter->chrV = sws_getGaussianVec(chromaGBlur, 3.0);
1685
    } else {
1686 1687
        filter->chrH = sws_getIdentityVec();
        filter->chrV = sws_getIdentityVec();
1688 1689
    }

1690 1691
    if (chromaSharpen != 0.0) {
        SwsVector *id = sws_getIdentityVec();
1692 1693 1694 1695 1696 1697 1698
        sws_scaleVec(filter->chrH, -chromaSharpen);
        sws_scaleVec(filter->chrV, -chromaSharpen);
        sws_addVec(filter->chrH, id);
        sws_addVec(filter->chrV, id);
        sws_freeVec(id);
    }

1699 1700
    if (lumaSharpen != 0.0) {
        SwsVector *id = sws_getIdentityVec();
1701 1702 1703 1704 1705 1706 1707 1708
        sws_scaleVec(filter->lumH, -lumaSharpen);
        sws_scaleVec(filter->lumV, -lumaSharpen);
        sws_addVec(filter->lumH, id);
        sws_addVec(filter->lumV, id);
        sws_freeVec(id);
    }

    if (chromaHShift != 0.0)
1709
        sws_shiftVec(filter->chrH, (int)(chromaHShift + 0.5));
1710 1711

    if (chromaVShift != 0.0)
1712
        sws_shiftVec(filter->chrV, (int)(chromaVShift + 0.5));
1713 1714 1715 1716 1717 1718

    sws_normalizeVec(filter->chrH, 1.0);
    sws_normalizeVec(filter->chrV, 1.0);
    sws_normalizeVec(filter->lumH, 1.0);
    sws_normalizeVec(filter->lumV, 1.0);

1719 1720 1721 1722
    if (verbose)
        sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
    if (verbose)
        sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1723 1724 1725 1726 1727 1728

    return filter;
}

SwsVector *sws_allocVec(int length)
{
1729 1730 1731 1732 1733 1734
    SwsVector *vec;

    if(length <= 0 || length > INT_MAX/ sizeof(double))
        return NULL;

    vec = av_malloc(sizeof(SwsVector));
1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745
    if (!vec)
        return NULL;
    vec->length = length;
    vec->coeff  = av_malloc(sizeof(double) * length);
    if (!vec->coeff)
        av_freep(&vec);
    return vec;
}

SwsVector *sws_getGaussianVec(double variance, double quality)
{
1746
    const int length = (int)(variance * quality + 0.5) | 1;
1747
    int i;
1748
    double middle  = (length - 1) * 0.5;
1749 1750 1751 1752 1753 1754
    SwsVector *vec;

    if(variance < 0 || quality < 0)
        return NULL;

    vec = sws_allocVec(length);
1755 1756 1757 1758

    if (!vec)
        return NULL;

1759 1760 1761 1762
    for (i = 0; i < length; i++) {
        double dist = i - middle;
        vec->coeff[i] = exp(-dist * dist / (2 * variance * variance)) /
                        sqrt(2 * variance * M_PI);
1763 1764 1765 1766 1767 1768 1769 1770 1771 1772
    }

    sws_normalizeVec(vec, 1.0);

    return vec;
}

SwsVector *sws_getConstVec(double c, int length)
{
    int i;
1773
    SwsVector *vec = sws_allocVec(length);
1774 1775 1776 1777

    if (!vec)
        return NULL;

1778 1779
    for (i = 0; i < length; i++)
        vec->coeff[i] = c;
1780 1781 1782 1783 1784 1785 1786 1787 1788

    return vec;
}

SwsVector *sws_getIdentityVec(void)
{
    return sws_getConstVec(1.0, 1);
}

1789
static double sws_dcVec(SwsVector *a)
1790 1791
{
    int i;
1792
    double sum = 0;
1793

1794 1795
    for (i = 0; i < a->length; i++)
        sum += a->coeff[i];
1796 1797 1798 1799 1800 1801 1802 1803

    return sum;
}

void sws_scaleVec(SwsVector *a, double scalar)
{
    int i;

1804 1805
    for (i = 0; i < a->length; i++)
        a->coeff[i] *= scalar;
1806 1807 1808 1809
}

void sws_normalizeVec(SwsVector *a, double height)
{
1810
    sws_scaleVec(a, height / sws_dcVec(a));
1811 1812 1813 1814
}

static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
{
1815
    int length = a->length + b->length - 1;
1816
    int i, j;
1817
    SwsVector *vec = sws_getConstVec(0.0, length);
1818 1819 1820 1821

    if (!vec)
        return NULL;

1822 1823 1824
    for (i = 0; i < a->length; i++) {
        for (j = 0; j < b->length; j++) {
            vec->coeff[i + j] += a->coeff[i] * b->coeff[j];
1825 1826 1827 1828 1829 1830 1831 1832
        }
    }

    return vec;
}

static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
{
1833
    int length = FFMAX(a->length, b->length);
1834
    int i;
1835
    SwsVector *vec = sws_getConstVec(0.0, length);
1836 1837 1838 1839

    if (!vec)
        return NULL;

1840 1841 1842 1843
    for (i = 0; i < a->length; i++)
        vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
    for (i = 0; i < b->length; i++)
        vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] += b->coeff[i];
1844 1845 1846 1847 1848 1849

    return vec;
}

static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
{
1850
    int length = FFMAX(a->length, b->length);
1851
    int i;
1852
    SwsVector *vec = sws_getConstVec(0.0, length);
1853 1854 1855 1856

    if (!vec)
        return NULL;

1857 1858 1859 1860
    for (i = 0; i < a->length; i++)
        vec->coeff[i + (length - 1) / 2 - (a->length - 1) / 2] += a->coeff[i];
    for (i = 0; i < b->length; i++)
        vec->coeff[i + (length - 1) / 2 - (b->length - 1) / 2] -= b->coeff[i];
1861 1862 1863 1864 1865 1866 1867

    return vec;
}

/* shift left / or right if "shift" is negative */
static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
{
1868
    int length = a->length + FFABS(shift) * 2;
1869
    int i;
1870
    SwsVector *vec = sws_getConstVec(0.0, length);
1871 1872 1873 1874

    if (!vec)
        return NULL;

1875 1876 1877
    for (i = 0; i < a->length; i++) {
        vec->coeff[i + (length    - 1) / 2 -
                       (a->length - 1) / 2 - shift] = a->coeff[i];
1878 1879 1880 1881 1882 1883 1884
    }

    return vec;
}

void sws_shiftVec(SwsVector *a, int shift)
{
1885
    SwsVector *shifted = sws_getShiftedVec(a, shift);
1886
    av_free(a->coeff);
1887 1888
    a->coeff  = shifted->coeff;
    a->length = shifted->length;
1889 1890 1891 1892 1893
    av_free(shifted);
}

void sws_addVec(SwsVector *a, SwsVector *b)
{
1894
    SwsVector *sum = sws_sumVec(a, b);
1895
    av_free(a->coeff);
1896 1897
    a->coeff  = sum->coeff;
    a->length = sum->length;
1898 1899 1900 1901 1902
    av_free(sum);
}

void sws_subVec(SwsVector *a, SwsVector *b)
{
1903
    SwsVector *diff = sws_diffVec(a, b);
1904
    av_free(a->coeff);
1905 1906
    a->coeff  = diff->coeff;
    a->length = diff->length;
1907 1908 1909 1910 1911
    av_free(diff);
}

void sws_convVec(SwsVector *a, SwsVector *b)
{
1912
    SwsVector *conv = sws_getConvVec(a, b);
1913
    av_free(a->coeff);
1914 1915
    a->coeff  = conv->coeff;
    a->length = conv->length;
1916 1917 1918 1919 1920
    av_free(conv);
}

SwsVector *sws_cloneVec(SwsVector *a)
{
1921
    SwsVector *vec = sws_allocVec(a->length);
1922 1923 1924 1925

    if (!vec)
        return NULL;

1926
    memcpy(vec->coeff, a->coeff, a->length * sizeof(*a->coeff));
1927 1928 1929 1930 1931 1932 1933

    return vec;
}

void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
{
    int i;
1934 1935
    double max = 0;
    double min = 0;
1936 1937
    double range;

1938 1939 1940
    for (i = 0; i < a->length; i++)
        if (a->coeff[i] > max)
            max = a->coeff[i];
1941

1942 1943 1944
    for (i = 0; i < a->length; i++)
        if (a->coeff[i] < min)
            min = a->coeff[i];
1945

1946
    range = max - min;
1947

1948 1949
    for (i = 0; i < a->length; i++) {
        int x = (int)((a->coeff[i] - min) * 60.0 / range + 0.5);
1950
        av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1951 1952
        for (; x > 0; x--)
            av_log(log_ctx, log_level, " ");
1953 1954 1955 1956 1957 1958
        av_log(log_ctx, log_level, "|\n");
    }
}

void sws_freeVec(SwsVector *a)
{
1959 1960
    if (!a)
        return;
1961
    av_freep(&a->coeff);
1962
    a->length = 0;
1963 1964 1965 1966 1967
    av_free(a);
}

void sws_freeFilter(SwsFilter *filter)
{
1968 1969 1970
    if (!filter)
        return;

1971 1972 1973 1974
    sws_freeVec(filter->lumH);
    sws_freeVec(filter->lumV);
    sws_freeVec(filter->chrH);
    sws_freeVec(filter->chrV);
1975 1976 1977 1978 1979 1980
    av_free(filter);
}

void sws_freeContext(SwsContext *c)
{
    int i;
1981 1982
    if (!c)
        return;
1983 1984

    if (c->lumPixBuf) {
1985
        for (i = 0; i < c->vLumBufSize; i++)
1986 1987 1988 1989
            av_freep(&c->lumPixBuf[i]);
        av_freep(&c->lumPixBuf);
    }

1990
    if (c->chrUPixBuf) {
1991
        for (i = 0; i < c->vChrBufSize; i++)
1992 1993 1994
            av_freep(&c->chrUPixBuf[i]);
        av_freep(&c->chrUPixBuf);
        av_freep(&c->chrVPixBuf);
1995 1996 1997
    }

    if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1998
        for (i = 0; i < c->vLumBufSize; i++)
1999 2000 2001 2002
            av_freep(&c->alpPixBuf[i]);
        av_freep(&c->alpPixBuf);
    }

2003 2004 2005
    for (i = 0; i < 4; i++)
        av_freep(&c->dither_error[i]);

2006 2007 2008 2009
    av_freep(&c->vLumFilter);
    av_freep(&c->vChrFilter);
    av_freep(&c->hLumFilter);
    av_freep(&c->hChrFilter);
2010
#if HAVE_ALTIVEC
2011 2012 2013 2014 2015 2016 2017 2018 2019
    av_freep(&c->vYCoeffsBank);
    av_freep(&c->vCCoeffsBank);
#endif

    av_freep(&c->vLumFilterPos);
    av_freep(&c->vChrFilterPos);
    av_freep(&c->hLumFilterPos);
    av_freep(&c->hChrFilterPos);

2020
#if HAVE_MMX_INLINE
2021
#if USE_MMAP
2022 2023 2024 2025
    if (c->lumMmxextFilterCode)
        munmap(c->lumMmxextFilterCode, c->lumMmxextFilterCodeSize);
    if (c->chrMmxextFilterCode)
        munmap(c->chrMmxextFilterCode, c->chrMmxextFilterCodeSize);
2026
#elif HAVE_VIRTUALALLOC
2027 2028 2029 2030
    if (c->lumMmxextFilterCode)
        VirtualFree(c->lumMmxextFilterCode, 0, MEM_RELEASE);
    if (c->chrMmxextFilterCode)
        VirtualFree(c->chrMmxextFilterCode, 0, MEM_RELEASE);
2031
#else
2032 2033
    av_free(c->lumMmxextFilterCode);
    av_free(c->chrMmxextFilterCode);
2034
#endif
2035 2036
    c->lumMmxextFilterCode = NULL;
    c->chrMmxextFilterCode = NULL;
2037
#endif /* HAVE_MMX_INLINE */
2038 2039

    av_freep(&c->yuvTable);
2040
    av_freep(&c->formatConvBuffer);
2041 2042 2043 2044

    av_free(c);
}

2045
struct SwsContext *sws_getCachedContext(struct SwsContext *context, int srcW,
2046
                                        int srcH, enum AVPixelFormat srcFormat,
2047
                                        int dstW, int dstH,
2048
                                        enum AVPixelFormat dstFormat, int flags,
2049 2050 2051
                                        SwsFilter *srcFilter,
                                        SwsFilter *dstFilter,
                                        const double *param)
2052
{
2053 2054
    static const double default_param[2] = { SWS_PARAM_DEFAULT,
                                             SWS_PARAM_DEFAULT };
2055 2056 2057 2058

    if (!param)
        param = default_param;

2059
    if (context &&
2060 2061 2062 2063 2064 2065 2066 2067
        (context->srcW      != srcW      ||
         context->srcH      != srcH      ||
         context->srcFormat != srcFormat ||
         context->dstW      != dstW      ||
         context->dstH      != dstH      ||
         context->dstFormat != dstFormat ||
         context->flags     != flags     ||
         context->param[0]  != param[0]  ||
2068 2069 2070 2071
         context->param[1]  != param[1])) {
        sws_freeContext(context);
        context = NULL;
    }
2072

2073
    if (!context) {
2074 2075 2076 2077 2078
        if (!(context = sws_alloc_context()))
            return NULL;
        context->srcW      = srcW;
        context->srcH      = srcH;
        context->srcFormat = srcFormat;
2079 2080
        context->dstW      = dstW;
        context->dstH      = dstH;
2081 2082 2083 2084 2085 2086 2087 2088
        context->dstFormat = dstFormat;
        context->flags     = flags;
        context->param[0]  = param[0];
        context->param[1]  = param[1];
        if (sws_init_context(context, srcFilter, dstFilter) < 0) {
            sws_freeContext(context);
            return NULL;
        }
2089 2090 2091
    }
    return context;
}