git.sesse.net Git - ffmpeg/blob - libswscale/swscale.c

   1 /*
   2  * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
   3  *
   4  * This file is part of FFmpeg.
   5  *
   6  * FFmpeg is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * FFmpeg is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with FFmpeg; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  19  *
  20  * the C code (not assembly, mmx, ...) of this file can be used
  21  * under the LGPL license too
  22  */
  23
  24 /*
  25   supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
  26   supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
  27   {BGR,RGB}{1,4,8,15,16} support dithering
  28
  29   unscaled special converters (YV12=I420=IYUV, Y800=Y8)
  30   YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
  31   x -> x
  32   YUV9 -> YV12
  33   YUV9/YV12 -> Y800
  34   Y800 -> YUV9/YV12
  35   BGR24 -> BGR32 & RGB24 -> RGB32
  36   BGR32 -> BGR24 & RGB32 -> RGB24
  37   BGR15 -> BGR16
  38 */
  39
  40 /*
  41 tested special converters (most are tested actually but i didnt write it down ...)
  42  YV12 -> BGR16
  43  YV12 -> YV12
  44  BGR15 -> BGR16
  45  BGR16 -> BGR16
  46  YVU9 -> YV12
  47
  48 untested special converters
  49   YV12/I420 -> BGR15/BGR24/BGR32 (its the yuv2rgb stuff, so it should be ok)
  50   YV12/I420 -> YV12/I420
  51   YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
  52   BGR24 -> BGR32 & RGB24 -> RGB32
  53   BGR32 -> BGR24 & RGB32 -> RGB24
  54   BGR24 -> YV12
  55 */
  56
  57 #include <inttypes.h>
  58 #include <string.h>
  59 #include <math.h>
  60 #include <stdio.h>
  61 #include <unistd.h>
  62 #include "config.h"
  63 #include <assert.h>
  64 #ifdef HAVE_SYS_MMAN_H
  65 #include <sys/mman.h>
  66 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
  67 #define MAP_ANONYMOUS MAP_ANON
  68 #endif
  69 #endif
  70 #include "swscale.h"
  71 #include "swscale_internal.h"
  72 #include "x86_cpu.h"
  73 #include "bswap.h"
  74 #include "rgb2rgb.h"
  75 #ifdef USE_FASTMEMCPY
  76 #include "libvo/fastmemcpy.h"
  77 #endif
  78
  79 #undef MOVNTQ
  80 #undef PAVGB
  81
  82 //#undef HAVE_MMX2
  83 //#define HAVE_3DNOW
  84 //#undef HAVE_MMX
  85 //#undef ARCH_X86
  86 //#define WORDS_BIGENDIAN
  87 #define DITHER1XBPP
  88
  89 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
  90
  91 #define RET 0xC3 //near return opcode for X86
  92
  93 #ifdef MP_DEBUG
  94 #define ASSERT(x) assert(x);
  95 #else
  96 #define ASSERT(x) ;
  97 #endif
  98
  99 #ifdef M_PI
 100 #define PI M_PI
 101 #else
 102 #define PI 3.14159265358979323846
 103 #endif
 104
 105 #define isSupportedIn(x)  ((x)==PIX_FMT_YUV420P || (x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422\
 106                         || (x)==PIX_FMT_RGB32|| (x)==PIX_FMT_BGR24|| (x)==PIX_FMT_BGR565|| (x)==PIX_FMT_BGR555\
 107                         || (x)==PIX_FMT_BGR32|| (x)==PIX_FMT_RGB24|| (x)==PIX_FMT_RGB565|| (x)==PIX_FMT_RGB555\
 108                         || (x)==PIX_FMT_GRAY8 || (x)==PIX_FMT_YUV410P\
 109                         || (x)==PIX_FMT_GRAY16BE || (x)==PIX_FMT_GRAY16LE\
 110                         || (x)==PIX_FMT_YUV444P || (x)==PIX_FMT_YUV422P || (x)==PIX_FMT_YUV411P)
 111 #define isSupportedOut(x) ((x)==PIX_FMT_YUV420P || (x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422\
 112                         || (x)==PIX_FMT_YUV444P || (x)==PIX_FMT_YUV422P || (x)==PIX_FMT_YUV411P\
 113                         || isRGB(x) || isBGR(x)\
 114                         || (x)==PIX_FMT_NV12 || (x)==PIX_FMT_NV21\
 115                         || (x)==PIX_FMT_GRAY16BE || (x)==PIX_FMT_GRAY16LE\
 116                         || (x)==PIX_FMT_GRAY8 || (x)==PIX_FMT_YUV410P)
 117 #define isPacked(x)    ((x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422 ||isRGB(x) || isBGR(x))
 118
 119 #define RGB2YUV_SHIFT 16
 120 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
 121 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
 122 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
 123 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
 124 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
 125 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
 126 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
 127 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
 128 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
 129
 130 extern const int32_t Inverse_Table_6_9[8][4];
 131
 132 /*
 133 NOTES
 134 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
 135
 136 TODO
 137 more intelligent missalignment avoidance for the horizontal scaler
 138 write special vertical cubic upscale version
 139 Optimize C code (yv12 / minmax)
 140 add support for packed pixel yuv input & output
 141 add support for Y8 output
 142 optimize bgr24 & bgr32
 143 add BGR4 output support
 144 write special BGR->BGR scaler
 145 */
 146
 147 #if defined(ARCH_X86) && defined (CONFIG_GPL)
 148 static uint64_t attribute_used __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
 149 static uint64_t attribute_used __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
 150 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
 151 static uint64_t attribute_used __attribute__((aligned(8))) w02=       0x0002000200020002LL;
 152 static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
 153 static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
 154 static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
 155 static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
 156
 157 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
 158 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
 159 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
 160 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
 161
 162 static uint64_t __attribute__((aligned(8))) dither4[2]={
 163         0x0103010301030103LL,
 164         0x0200020002000200LL,};
 165
 166 static uint64_t __attribute__((aligned(8))) dither8[2]={
 167         0x0602060206020602LL,
 168         0x0004000400040004LL,};
 169
 170 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
 171 static uint64_t attribute_used __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
 172 static uint64_t attribute_used __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
 173 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
 174 static uint64_t attribute_used __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
 175 static uint64_t attribute_used __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
 176
 177 static uint64_t attribute_used __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
 178 static uint64_t attribute_used __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
 179 static uint64_t attribute_used __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
 180
 181 #ifdef FAST_BGR2YV12
 182 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
 183 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
 184 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
 185 #else
 186 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
 187 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
 188 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
 189 #endif /* FAST_BGR2YV12 */
 190 static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
 191 static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
 192 static const uint64_t w1111       attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
 193 #endif /* defined(ARCH_X86) */
 194
 195 // clipping helper table for C implementations:
 196 static unsigned char clip_table[768];
 197
 198 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
 199
 200 extern const uint8_t dither_2x2_4[2][8];
 201 extern const uint8_t dither_2x2_8[2][8];
 202 extern const uint8_t dither_8x8_32[8][8];
 203 extern const uint8_t dither_8x8_73[8][8];
 204 extern const uint8_t dither_8x8_220[8][8];
 205
 206 char *sws_format_name(enum PixelFormat format)
 207 {
 208     switch (format) {
 209         case PIX_FMT_YUV420P:
 210             return "yuv420p";
 211         case PIX_FMT_YUYV422:
 212             return "yuyv422";
 213         case PIX_FMT_RGB24:
 214             return "rgb24";
 215         case PIX_FMT_BGR24:
 216             return "bgr24";
 217         case PIX_FMT_YUV422P:
 218             return "yuv422p";
 219         case PIX_FMT_YUV444P:
 220             return "yuv444p";
 221         case PIX_FMT_RGB32:
 222             return "rgb32";
 223         case PIX_FMT_YUV410P:
 224             return "yuv410p";
 225         case PIX_FMT_YUV411P:
 226             return "yuv411p";
 227         case PIX_FMT_RGB565:
 228             return "rgb565";
 229         case PIX_FMT_RGB555:
 230             return "rgb555";
 231         case PIX_FMT_GRAY16BE:
 232             return "gray16be";
 233         case PIX_FMT_GRAY16LE:
 234             return "gray16le";
 235         case PIX_FMT_GRAY8:
 236             return "gray8";
 237         case PIX_FMT_MONOWHITE:
 238             return "mono white";
 239         case PIX_FMT_MONOBLACK:
 240             return "mono black";
 241         case PIX_FMT_PAL8:
 242             return "Palette";
 243         case PIX_FMT_YUVJ420P:
 244             return "yuvj420p";
 245         case PIX_FMT_YUVJ422P:
 246             return "yuvj422p";
 247         case PIX_FMT_YUVJ444P:
 248             return "yuvj444p";
 249         case PIX_FMT_XVMC_MPEG2_MC:
 250             return "xvmc_mpeg2_mc";
 251         case PIX_FMT_XVMC_MPEG2_IDCT:
 252             return "xvmc_mpeg2_idct";
 253         case PIX_FMT_UYVY422:
 254             return "uyvy422";
 255         case PIX_FMT_UYYVYY411:
 256             return "uyyvyy411";
 257         case PIX_FMT_RGB32_1:
 258             return "rgb32x";
 259         case PIX_FMT_BGR32_1:
 260             return "bgr32x";
 261         case PIX_FMT_BGR32:
 262             return "bgr32";
 263         case PIX_FMT_BGR565:
 264             return "bgr565";
 265         case PIX_FMT_BGR555:
 266             return "bgr555";
 267         case PIX_FMT_BGR8:
 268             return "bgr8";
 269         case PIX_FMT_BGR4:
 270             return "bgr4";
 271         case PIX_FMT_BGR4_BYTE:
 272             return "bgr4 byte";
 273         case PIX_FMT_RGB8:
 274             return "rgb8";
 275         case PIX_FMT_RGB4:
 276             return "rgb4";
 277         case PIX_FMT_RGB4_BYTE:
 278             return "rgb4 byte";
 279         case PIX_FMT_NV12:
 280             return "nv12";
 281         case PIX_FMT_NV21:
 282             return "nv21";
 283         default:
 284             return "Unknown format";
 285     }
 286 }
 287
 288 #if defined(ARCH_X86) && defined (CONFIG_GPL)
 289 void in_asm_used_var_warning_killer()
 290 {
 291  volatile int i= bF8+bFC+w10+
 292  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
 293  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
 294  if(i) i=0;
 295 }
 296 #endif
 297
 298 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
 299                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
 300                                     uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
 301 {
 302         //FIXME Optimize (just quickly writen not opti..)
 303         int i;
 304         for(i=0; i<dstW; i++)
 305         {
 306                 int val=1<<18;
 307                 int j;
 308                 for(j=0; j<lumFilterSize; j++)
 309                         val += lumSrc[j][i] * lumFilter[j];
 310
 311                 dest[i]= clip_uint8(val>>19);
 312         }
 313
 314         if(uDest != NULL)
 315                 for(i=0; i<chrDstW; i++)
 316                 {
 317                         int u=1<<18;
 318                         int v=1<<18;
 319                         int j;
 320                         for(j=0; j<chrFilterSize; j++)
 321                         {
 322                                 u += chrSrc[j][i] * chrFilter[j];
 323                                 v += chrSrc[j][i + 2048] * chrFilter[j];
 324                         }
 325
 326                         uDest[i]= clip_uint8(u>>19);
 327                         vDest[i]= clip_uint8(v>>19);
 328                 }
 329 }
 330
 331 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
 332                                 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
 333                                 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
 334 {
 335         //FIXME Optimize (just quickly writen not opti..)
 336         int i;
 337         for(i=0; i<dstW; i++)
 338         {
 339                 int val=1<<18;
 340                 int j;
 341                 for(j=0; j<lumFilterSize; j++)
 342                         val += lumSrc[j][i] * lumFilter[j];
 343
 344                 dest[i]= clip_uint8(val>>19);
 345         }
 346
 347         if(uDest == NULL)
 348                 return;
 349
 350         if(dstFormat == PIX_FMT_NV12)
 351                 for(i=0; i<chrDstW; i++)
 352                 {
 353                         int u=1<<18;
 354                         int v=1<<18;
 355                         int j;
 356                         for(j=0; j<chrFilterSize; j++)
 357                         {
 358                                 u += chrSrc[j][i] * chrFilter[j];
 359                                 v += chrSrc[j][i + 2048] * chrFilter[j];
 360                         }
 361
 362                         uDest[2*i]= clip_uint8(u>>19);
 363                         uDest[2*i+1]= clip_uint8(v>>19);
 364                 }
 365         else
 366                 for(i=0; i<chrDstW; i++)
 367                 {
 368                         int u=1<<18;
 369                         int v=1<<18;
 370                         int j;
 371                         for(j=0; j<chrFilterSize; j++)
 372                         {
 373                                 u += chrSrc[j][i] * chrFilter[j];
 374                                 v += chrSrc[j][i + 2048] * chrFilter[j];
 375                         }
 376
 377                         uDest[2*i]= clip_uint8(v>>19);
 378                         uDest[2*i+1]= clip_uint8(u>>19);
 379                 }
 380 }
 381
 382 #define YSCALE_YUV_2_PACKEDX_C(type) \
 383                 for(i=0; i<(dstW>>1); i++){\
 384                         int j;\
 385                         int Y1=1<<18;\
 386                         int Y2=1<<18;\
 387                         int U=1<<18;\
 388                         int V=1<<18;\
 389                         type attribute_unused *r, *b, *g;\
 390                         const int i2= 2*i;\
 391                         \
 392                         for(j=0; j<lumFilterSize; j++)\
 393                         {\
 394                                 Y1 += lumSrc[j][i2] * lumFilter[j];\
 395                                 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
 396                         }\
 397                         for(j=0; j<chrFilterSize; j++)\
 398                         {\
 399                                 U += chrSrc[j][i] * chrFilter[j];\
 400                                 V += chrSrc[j][i+2048] * chrFilter[j];\
 401                         }\
 402                         Y1>>=19;\
 403                         Y2>>=19;\
 404                         U >>=19;\
 405                         V >>=19;\
 406                         if((Y1|Y2|U|V)&256)\
 407                         {\
 408                                 if(Y1>255)   Y1=255;\
 409                                 else if(Y1<0)Y1=0;\
 410                                 if(Y2>255)   Y2=255;\
 411                                 else if(Y2<0)Y2=0;\
 412                                 if(U>255)    U=255;\
 413                                 else if(U<0) U=0;\
 414                                 if(V>255)    V=255;\
 415                                 else if(V<0) V=0;\
 416                         }
 417
 418 #define YSCALE_YUV_2_RGBX_C(type) \
 419                         YSCALE_YUV_2_PACKEDX_C(type)\
 420                         r = (type *)c->table_rV[V];\
 421                         g = (type *)(c->table_gU[U] + c->table_gV[V]);\
 422                         b = (type *)c->table_bU[U];\
 423
 424 #define YSCALE_YUV_2_PACKED2_C \
 425                 for(i=0; i<(dstW>>1); i++){\
 426                         const int i2= 2*i;\
 427                         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
 428                         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
 429                         int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
 430                         int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
 431
 432 #define YSCALE_YUV_2_RGB2_C(type) \
 433                         YSCALE_YUV_2_PACKED2_C\
 434                         type *r, *b, *g;\
 435                         r = (type *)c->table_rV[V];\
 436                         g = (type *)(c->table_gU[U] + c->table_gV[V]);\
 437                         b = (type *)c->table_bU[U];\
 438
 439 #define YSCALE_YUV_2_PACKED1_C \
 440                 for(i=0; i<(dstW>>1); i++){\
 441                         const int i2= 2*i;\
 442                         int Y1= buf0[i2  ]>>7;\
 443                         int Y2= buf0[i2+1]>>7;\
 444                         int U= (uvbuf1[i     ])>>7;\
 445                         int V= (uvbuf1[i+2048])>>7;\
 446
 447 #define YSCALE_YUV_2_RGB1_C(type) \
 448                         YSCALE_YUV_2_PACKED1_C\
 449                         type *r, *b, *g;\
 450                         r = (type *)c->table_rV[V];\
 451                         g = (type *)(c->table_gU[U] + c->table_gV[V]);\
 452                         b = (type *)c->table_bU[U];\
 453
 454 #define YSCALE_YUV_2_PACKED1B_C \
 455                 for(i=0; i<(dstW>>1); i++){\
 456                         const int i2= 2*i;\
 457                         int Y1= buf0[i2  ]>>7;\
 458                         int Y2= buf0[i2+1]>>7;\
 459                         int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
 460                         int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
 461
 462 #define YSCALE_YUV_2_RGB1B_C(type) \
 463                         YSCALE_YUV_2_PACKED1B_C\
 464                         type *r, *b, *g;\
 465                         r = (type *)c->table_rV[V];\
 466                         g = (type *)(c->table_gU[U] + c->table_gV[V]);\
 467                         b = (type *)c->table_bU[U];\
 468
 469 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
 470         switch(c->dstFormat)\
 471         {\
 472         case PIX_FMT_RGB32:\
 473         case PIX_FMT_BGR32:\
 474                 func(uint32_t)\
 475                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
 476                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
 477                 }               \
 478                 break;\
 479         case PIX_FMT_RGB24:\
 480                 func(uint8_t)\
 481                         ((uint8_t*)dest)[0]= r[Y1];\
 482                         ((uint8_t*)dest)[1]= g[Y1];\
 483                         ((uint8_t*)dest)[2]= b[Y1];\
 484                         ((uint8_t*)dest)[3]= r[Y2];\
 485                         ((uint8_t*)dest)[4]= g[Y2];\
 486                         ((uint8_t*)dest)[5]= b[Y2];\
 487                         dest+=6;\
 488                 }\
 489                 break;\
 490         case PIX_FMT_BGR24:\
 491                 func(uint8_t)\
 492                         ((uint8_t*)dest)[0]= b[Y1];\
 493                         ((uint8_t*)dest)[1]= g[Y1];\
 494                         ((uint8_t*)dest)[2]= r[Y1];\
 495                         ((uint8_t*)dest)[3]= b[Y2];\
 496                         ((uint8_t*)dest)[4]= g[Y2];\
 497                         ((uint8_t*)dest)[5]= r[Y2];\
 498                         dest+=6;\
 499                 }\
 500                 break;\
 501         case PIX_FMT_RGB565:\
 502         case PIX_FMT_BGR565:\
 503                 {\
 504                         const int dr1= dither_2x2_8[y&1    ][0];\
 505                         const int dg1= dither_2x2_4[y&1    ][0];\
 506                         const int db1= dither_2x2_8[(y&1)^1][0];\
 507                         const int dr2= dither_2x2_8[y&1    ][1];\
 508                         const int dg2= dither_2x2_4[y&1    ][1];\
 509                         const int db2= dither_2x2_8[(y&1)^1][1];\
 510                         func(uint16_t)\
 511                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
 512                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
 513                         }\
 514                 }\
 515                 break;\
 516         case PIX_FMT_RGB555:\
 517         case PIX_FMT_BGR555:\
 518                 {\
 519                         const int dr1= dither_2x2_8[y&1    ][0];\
 520                         const int dg1= dither_2x2_8[y&1    ][1];\
 521                         const int db1= dither_2x2_8[(y&1)^1][0];\
 522                         const int dr2= dither_2x2_8[y&1    ][1];\
 523                         const int dg2= dither_2x2_8[y&1    ][0];\
 524                         const int db2= dither_2x2_8[(y&1)^1][1];\
 525                         func(uint16_t)\
 526                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
 527                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
 528                         }\
 529                 }\
 530                 break;\
 531         case PIX_FMT_RGB8:\
 532         case PIX_FMT_BGR8:\
 533                 {\
 534                         const uint8_t * const d64= dither_8x8_73[y&7];\
 535                         const uint8_t * const d32= dither_8x8_32[y&7];\
 536                         func(uint8_t)\
 537                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
 538                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
 539                         }\
 540                 }\
 541                 break;\
 542         case PIX_FMT_RGB4:\
 543         case PIX_FMT_BGR4:\
 544                 {\
 545                         const uint8_t * const d64= dither_8x8_73 [y&7];\
 546                         const uint8_t * const d128=dither_8x8_220[y&7];\
 547                         func(uint8_t)\
 548                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
 549                                                  + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
 550                         }\
 551                 }\
 552                 break;\
 553         case PIX_FMT_RGB4_BYTE:\
 554         case PIX_FMT_BGR4_BYTE:\
 555                 {\
 556                         const uint8_t * const d64= dither_8x8_73 [y&7];\
 557                         const uint8_t * const d128=dither_8x8_220[y&7];\
 558                         func(uint8_t)\
 559                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
 560                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
 561                         }\
 562                 }\
 563                 break;\
 564         case PIX_FMT_MONOBLACK:\
 565                 {\
 566                         const uint8_t * const d128=dither_8x8_220[y&7];\
 567                         uint8_t *g= c->table_gU[128] + c->table_gV[128];\
 568                         for(i=0; i<dstW-7; i+=8){\
 569                                 int acc;\
 570                                 acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
 571                                 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
 572                                 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
 573                                 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
 574                                 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
 575                                 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
 576                                 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
 577                                 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
 578                                 ((uint8_t*)dest)[0]= acc;\
 579                                 dest++;\
 580                         }\
 581 \
 582 /*\
 583 ((uint8_t*)dest)-= dstW>>4;\
 584 {\
 585                         int acc=0;\
 586                         int left=0;\
 587                         static int top[1024];\
 588                         static int last_new[1024][1024];\
 589                         static int last_in3[1024][1024];\
 590                         static int drift[1024][1024];\
 591                         int topLeft=0;\
 592                         int shift=0;\
 593                         int count=0;\
 594                         const uint8_t * const d128=dither_8x8_220[y&7];\
 595                         int error_new=0;\
 596                         int error_in3=0;\
 597                         int f=0;\
 598                         \
 599                         for(i=dstW>>1; i<dstW; i++){\
 600                                 int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
 601                                 int in2 = (76309 * (in - 16) + 32768) >> 16;\
 602                                 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
 603                                 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
 604                                         + (last_new[y][i] - in3)*f/256;\
 605                                 int new= old> 128 ? 255 : 0;\
 606 \
 607                                 error_new+= FFABS(last_new[y][i] - new);\
 608                                 error_in3+= FFABS(last_in3[y][i] - in3);\
 609                                 f= error_new - error_in3*4;\
 610                                 if(f<0) f=0;\
 611                                 if(f>256) f=256;\
 612 \
 613                                 topLeft= top[i];\
 614                                 left= top[i]= old - new;\
 615                                 last_new[y][i]= new;\
 616                                 last_in3[y][i]= in3;\
 617 \
 618                                 acc+= acc + (new&1);\
 619                                 if((i&7)==6){\
 620                                         ((uint8_t*)dest)[0]= acc;\
 621                                         ((uint8_t*)dest)++;\
 622                                 }\
 623                         }\
 624 }\
 625 */\
 626                 }\
 627                 break;\
 628         case PIX_FMT_YUYV422:\
 629                 func2\
 630                         ((uint8_t*)dest)[2*i2+0]= Y1;\
 631                         ((uint8_t*)dest)[2*i2+1]= U;\
 632                         ((uint8_t*)dest)[2*i2+2]= Y2;\
 633                         ((uint8_t*)dest)[2*i2+3]= V;\
 634                 }               \
 635                 break;\
 636         case PIX_FMT_UYVY422:\
 637                 func2\
 638                         ((uint8_t*)dest)[2*i2+0]= U;\
 639                         ((uint8_t*)dest)[2*i2+1]= Y1;\
 640                         ((uint8_t*)dest)[2*i2+2]= V;\
 641                         ((uint8_t*)dest)[2*i2+3]= Y2;\
 642                 }               \
 643                 break;\
 644         }\
 645
 646
 647 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
 648                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
 649                                     uint8_t *dest, int dstW, int y)
 650 {
 651         int i;
 652         switch(c->dstFormat)
 653         {
 654         case PIX_FMT_BGR32:
 655         case PIX_FMT_RGB32:
 656                 YSCALE_YUV_2_RGBX_C(uint32_t)
 657                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
 658                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
 659                 }
 660                 break;
 661         case PIX_FMT_RGB24:
 662                 YSCALE_YUV_2_RGBX_C(uint8_t)
 663                         ((uint8_t*)dest)[0]= r[Y1];
 664                         ((uint8_t*)dest)[1]= g[Y1];
 665                         ((uint8_t*)dest)[2]= b[Y1];
 666                         ((uint8_t*)dest)[3]= r[Y2];
 667                         ((uint8_t*)dest)[4]= g[Y2];
 668                         ((uint8_t*)dest)[5]= b[Y2];
 669                         dest+=6;
 670                 }
 671                 break;
 672         case PIX_FMT_BGR24:
 673                 YSCALE_YUV_2_RGBX_C(uint8_t)
 674                         ((uint8_t*)dest)[0]= b[Y1];
 675                         ((uint8_t*)dest)[1]= g[Y1];
 676                         ((uint8_t*)dest)[2]= r[Y1];
 677                         ((uint8_t*)dest)[3]= b[Y2];
 678                         ((uint8_t*)dest)[4]= g[Y2];
 679                         ((uint8_t*)dest)[5]= r[Y2];
 680                         dest+=6;
 681                 }
 682                 break;
 683         case PIX_FMT_RGB565:
 684         case PIX_FMT_BGR565:
 685                 {
 686                         const int dr1= dither_2x2_8[y&1    ][0];
 687                         const int dg1= dither_2x2_4[y&1    ][0];
 688                         const int db1= dither_2x2_8[(y&1)^1][0];
 689                         const int dr2= dither_2x2_8[y&1    ][1];
 690                         const int dg2= dither_2x2_4[y&1    ][1];
 691                         const int db2= dither_2x2_8[(y&1)^1][1];
 692                         YSCALE_YUV_2_RGBX_C(uint16_t)
 693                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
 694                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
 695                         }
 696                 }
 697                 break;
 698         case PIX_FMT_RGB555:
 699         case PIX_FMT_BGR555:
 700                 {
 701                         const int dr1= dither_2x2_8[y&1    ][0];
 702                         const int dg1= dither_2x2_8[y&1    ][1];
 703                         const int db1= dither_2x2_8[(y&1)^1][0];
 704                         const int dr2= dither_2x2_8[y&1    ][1];
 705                         const int dg2= dither_2x2_8[y&1    ][0];
 706                         const int db2= dither_2x2_8[(y&1)^1][1];
 707                         YSCALE_YUV_2_RGBX_C(uint16_t)
 708                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
 709                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
 710                         }
 711                 }
 712                 break;
 713         case PIX_FMT_RGB8:
 714         case PIX_FMT_BGR8:
 715                 {
 716                         const uint8_t * const d64= dither_8x8_73[y&7];
 717                         const uint8_t * const d32= dither_8x8_32[y&7];
 718                         YSCALE_YUV_2_RGBX_C(uint8_t)
 719                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
 720                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
 721                         }
 722                 }
 723                 break;
 724         case PIX_FMT_RGB4:
 725         case PIX_FMT_BGR4:
 726                 {
 727                         const uint8_t * const d64= dither_8x8_73 [y&7];
 728                         const uint8_t * const d128=dither_8x8_220[y&7];
 729                         YSCALE_YUV_2_RGBX_C(uint8_t)
 730                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
 731                                                   +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
 732                         }
 733                 }
 734                 break;
 735         case PIX_FMT_RGB4_BYTE:
 736         case PIX_FMT_BGR4_BYTE:
 737                 {
 738                         const uint8_t * const d64= dither_8x8_73 [y&7];
 739                         const uint8_t * const d128=dither_8x8_220[y&7];
 740                         YSCALE_YUV_2_RGBX_C(uint8_t)
 741                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
 742                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
 743                         }
 744                 }
 745                 break;
 746         case PIX_FMT_MONOBLACK:
 747                 {
 748                         const uint8_t * const d128=dither_8x8_220[y&7];
 749                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
 750                         int acc=0;
 751                         for(i=0; i<dstW-1; i+=2){
 752                                 int j;
 753                                 int Y1=1<<18;
 754                                 int Y2=1<<18;
 755
 756                                 for(j=0; j<lumFilterSize; j++)
 757                                 {
 758                                         Y1 += lumSrc[j][i] * lumFilter[j];
 759                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
 760                                 }
 761                                 Y1>>=19;
 762                                 Y2>>=19;
 763                                 if((Y1|Y2)&256)
 764                                 {
 765                                         if(Y1>255)   Y1=255;
 766                                         else if(Y1<0)Y1=0;
 767                                         if(Y2>255)   Y2=255;
 768                                         else if(Y2<0)Y2=0;
 769                                 }
 770                                 acc+= acc + g[Y1+d128[(i+0)&7]];
 771                                 acc+= acc + g[Y2+d128[(i+1)&7]];
 772                                 if((i&7)==6){
 773                                         ((uint8_t*)dest)[0]= acc;
 774                                         dest++;
 775                                 }
 776                         }
 777                 }
 778                 break;
 779         case PIX_FMT_YUYV422:
 780                 YSCALE_YUV_2_PACKEDX_C(void)
 781                         ((uint8_t*)dest)[2*i2+0]= Y1;
 782                         ((uint8_t*)dest)[2*i2+1]= U;
 783                         ((uint8_t*)dest)[2*i2+2]= Y2;
 784                         ((uint8_t*)dest)[2*i2+3]= V;
 785                 }
 786                 break;
 787         case PIX_FMT_UYVY422:
 788                 YSCALE_YUV_2_PACKEDX_C(void)
 789                         ((uint8_t*)dest)[2*i2+0]= U;
 790                         ((uint8_t*)dest)[2*i2+1]= Y1;
 791                         ((uint8_t*)dest)[2*i2+2]= V;
 792                         ((uint8_t*)dest)[2*i2+3]= Y2;
 793                 }
 794                 break;
 795         }
 796 }
 797
 798
 799 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
 800 //Plain C versions
 801 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) || !defined(CONFIG_GPL)
 802 #define COMPILE_C
 803 #endif
 804
 805 #ifdef ARCH_POWERPC
 806 #if (defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
 807 #define COMPILE_ALTIVEC
 808 #endif //HAVE_ALTIVEC
 809 #endif //ARCH_POWERPC
 810
 811 #if defined(ARCH_X86)
 812
 813 #if ((defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
 814 #define COMPILE_MMX
 815 #endif
 816
 817 #if (defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
 818 #define COMPILE_MMX2
 819 #endif
 820
 821 #if ((defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
 822 #define COMPILE_3DNOW
 823 #endif
 824 #endif //ARCH_X86 || ARCH_X86_64
 825
 826 #undef HAVE_MMX
 827 #undef HAVE_MMX2
 828 #undef HAVE_3DNOW
 829
 830 #ifdef COMPILE_C
 831 #undef HAVE_MMX
 832 #undef HAVE_MMX2
 833 #undef HAVE_3DNOW
 834 #undef HAVE_ALTIVEC
 835 #define RENAME(a) a ## _C
 836 #include "swscale_template.c"
 837 #endif
 838
 839 #ifdef ARCH_POWERPC
 840 #ifdef COMPILE_ALTIVEC
 841 #undef RENAME
 842 #define HAVE_ALTIVEC
 843 #define RENAME(a) a ## _altivec
 844 #include "swscale_template.c"
 845 #endif
 846 #endif //ARCH_POWERPC
 847
 848 #if defined(ARCH_X86)
 849
 850 //X86 versions
 851 /*
 852 #undef RENAME
 853 #undef HAVE_MMX
 854 #undef HAVE_MMX2
 855 #undef HAVE_3DNOW
 856 #define ARCH_X86
 857 #define RENAME(a) a ## _X86
 858 #include "swscale_template.c"
 859 */
 860 //MMX versions
 861 #ifdef COMPILE_MMX
 862 #undef RENAME
 863 #define HAVE_MMX
 864 #undef HAVE_MMX2
 865 #undef HAVE_3DNOW
 866 #define RENAME(a) a ## _MMX
 867 #include "swscale_template.c"
 868 #endif
 869
 870 //MMX2 versions
 871 #ifdef COMPILE_MMX2
 872 #undef RENAME
 873 #define HAVE_MMX
 874 #define HAVE_MMX2
 875 #undef HAVE_3DNOW
 876 #define RENAME(a) a ## _MMX2
 877 #include "swscale_template.c"
 878 #endif
 879
 880 //3DNOW versions
 881 #ifdef COMPILE_3DNOW
 882 #undef RENAME
 883 #define HAVE_MMX
 884 #undef HAVE_MMX2
 885 #define HAVE_3DNOW
 886 #define RENAME(a) a ## _3DNow
 887 #include "swscale_template.c"
 888 #endif
 889
 890 #endif //ARCH_X86 || ARCH_X86_64
 891
 892 // minor note: the HAVE_xyz is messed up after that line so don't use it
 893
 894 static double getSplineCoeff(double a, double b, double c, double d, double dist)
 895 {
 896 //      printf("%f %f %f %f %f\n", a,b,c,d,dist);
 897         if(dist<=1.0)   return ((d*dist + c)*dist + b)*dist +a;
 898         else            return getSplineCoeff(  0.0,
 899                                                  b+ 2.0*c + 3.0*d,
 900                                                         c + 3.0*d,
 901                                                 -b- 3.0*c - 6.0*d,
 902                                                 dist-1.0);
 903 }
 904
 905 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
 906                               int srcW, int dstW, int filterAlign, int one, int flags,
 907                               SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
 908 {
 909         int i;
 910         int filterSize;
 911         int filter2Size;
 912         int minFilterSize;
 913         double *filter=NULL;
 914         double *filter2=NULL;
 915 #if defined(ARCH_X86)
 916         if(flags & SWS_CPU_CAPS_MMX)
 917                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
 918 #endif
 919
 920         // Note the +1 is for the MMXscaler which reads over the end
 921         *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
 922
 923         if(FFABS(xInc - 0x10000) <10) // unscaled
 924         {
 925                 int i;
 926                 filterSize= 1;
 927                 filter= av_malloc(dstW*sizeof(double)*filterSize);
 928                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
 929
 930                 for(i=0; i<dstW; i++)
 931                 {
 932                         filter[i*filterSize]=1;
 933                         (*filterPos)[i]=i;
 934                 }
 935
 936         }
 937         else if(flags&SWS_POINT) // lame looking point sampling mode
 938         {
 939                 int i;
 940                 int xDstInSrc;
 941                 filterSize= 1;
 942                 filter= av_malloc(dstW*sizeof(double)*filterSize);
 943
 944                 xDstInSrc= xInc/2 - 0x8000;
 945                 for(i=0; i<dstW; i++)
 946                 {
 947                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
 948
 949                         (*filterPos)[i]= xx;
 950                         filter[i]= 1.0;
 951                         xDstInSrc+= xInc;
 952                 }
 953         }
 954         else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
 955         {
 956                 int i;
 957                 int xDstInSrc;
 958                 if     (flags&SWS_BICUBIC) filterSize= 4;
 959                 else if(flags&SWS_X      ) filterSize= 4;
 960                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA
 961                 filter= av_malloc(dstW*sizeof(double)*filterSize);
 962
 963                 xDstInSrc= xInc/2 - 0x8000;
 964                 for(i=0; i<dstW; i++)
 965                 {
 966                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
 967                         int j;
 968
 969                         (*filterPos)[i]= xx;
 970                                 //Bilinear upscale / linear interpolate / Area averaging
 971                                 for(j=0; j<filterSize; j++)
 972                                 {
 973                                         double d= FFABS((xx<<16) - xDstInSrc)/(double)(1<<16);
 974                                         double coeff= 1.0 - d;
 975                                         if(coeff<0) coeff=0;
 976                                         filter[i*filterSize + j]= coeff;
 977                                         xx++;
 978                                 }
 979                         xDstInSrc+= xInc;
 980                 }
 981         }
 982         else
 983         {
 984                 double xDstInSrc;
 985                 double sizeFactor, filterSizeInSrc;
 986                 const double xInc1= (double)xInc / (double)(1<<16);
 987
 988                 if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
 989                 else if(flags&SWS_X)            sizeFactor= 8.0;
 990                 else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
 991                 else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
 992                 else if(flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
 993                 else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
 994                 else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
 995                 else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
 996                 else {
 997                         sizeFactor= 0.0; //GCC warning killer
 998                         ASSERT(0)
 999                 }
1000
1001                 if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
1002                 else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
1003
1004                 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1005                 if(filterSize > srcW-2) filterSize=srcW-2;
1006
1007                 filter= av_malloc(dstW*sizeof(double)*filterSize);
1008
1009                 xDstInSrc= xInc1 / 2.0 - 0.5;
1010                 for(i=0; i<dstW; i++)
1011                 {
1012                         int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
1013                         int j;
1014                         (*filterPos)[i]= xx;
1015                         for(j=0; j<filterSize; j++)
1016                         {
1017                                 double d= FFABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
1018                                 double coeff;
1019                                 if(flags & SWS_BICUBIC)
1020                                 {
1021                                         double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
1022                                         double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
1023
1024                                         if(d<1.0)
1025                                                 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
1026                                         else if(d<2.0)
1027                                                 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
1028                                         else
1029                                                 coeff=0.0;
1030                                 }
1031 /*                              else if(flags & SWS_X)
1032                                 {
1033                                         double p= param ? param*0.01 : 0.3;
1034                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1035                                         coeff*= pow(2.0, - p*d*d);
1036                                 }*/
1037                                 else if(flags & SWS_X)
1038                                 {
1039                                         double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1040
1041                                         if(d<1.0)
1042                                                 coeff = cos(d*PI);
1043                                         else
1044                                                 coeff=-1.0;
1045                                         if(coeff<0.0)   coeff= -pow(-coeff, A);
1046                                         else            coeff=  pow( coeff, A);
1047                                         coeff= coeff*0.5 + 0.5;
1048                                 }
1049                                 else if(flags & SWS_AREA)
1050                                 {
1051                                         double srcPixelSize= 1.0/xInc1;
1052                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
1053                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1054                                         else coeff=0.0;
1055                                 }
1056                                 else if(flags & SWS_GAUSS)
1057                                 {
1058                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1059                                         coeff = pow(2.0, - p*d*d);
1060                                 }
1061                                 else if(flags & SWS_SINC)
1062                                 {
1063                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1064                                 }
1065                                 else if(flags & SWS_LANCZOS)
1066                                 {
1067                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1068                                         coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1069                                         if(d>p) coeff=0;
1070                                 }
1071                                 else if(flags & SWS_BILINEAR)
1072                                 {
1073                                         coeff= 1.0 - d;
1074                                         if(coeff<0) coeff=0;
1075                                 }
1076                                 else if(flags & SWS_SPLINE)
1077                                 {
1078                                         double p=-2.196152422706632;
1079                                         coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1080                                 }
1081                                 else {
1082                                         coeff= 0.0; //GCC warning killer
1083                                         ASSERT(0)
1084                                 }
1085
1086                                 filter[i*filterSize + j]= coeff;
1087                                 xx++;
1088                         }
1089                         xDstInSrc+= xInc1;
1090                 }
1091         }
1092
1093         /* apply src & dst Filter to filter -> filter2
1094            av_free(filter);
1095         */
1096         ASSERT(filterSize>0)
1097         filter2Size= filterSize;
1098         if(srcFilter) filter2Size+= srcFilter->length - 1;
1099         if(dstFilter) filter2Size+= dstFilter->length - 1;
1100         ASSERT(filter2Size>0)
1101         filter2= av_malloc(filter2Size*dstW*sizeof(double));
1102
1103         for(i=0; i<dstW; i++)
1104         {
1105                 int j;
1106                 SwsVector scaleFilter;
1107                 SwsVector *outVec;
1108
1109                 scaleFilter.coeff= filter + i*filterSize;
1110                 scaleFilter.length= filterSize;
1111
1112                 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1113                 else          outVec= &scaleFilter;
1114
1115                 ASSERT(outVec->length == filter2Size)
1116                 //FIXME dstFilter
1117
1118                 for(j=0; j<outVec->length; j++)
1119                 {
1120                         filter2[i*filter2Size + j]= outVec->coeff[j];
1121                 }
1122
1123                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1124
1125                 if(outVec != &scaleFilter) sws_freeVec(outVec);
1126         }
1127         av_free(filter); filter=NULL;
1128
1129         /* try to reduce the filter-size (step1 find size and shift left) */
1130         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1131         minFilterSize= 0;
1132         for(i=dstW-1; i>=0; i--)
1133         {
1134                 int min= filter2Size;
1135                 int j;
1136                 double cutOff=0.0;
1137
1138                 /* get rid off near zero elements on the left by shifting left */
1139                 for(j=0; j<filter2Size; j++)
1140                 {
1141                         int k;
1142                         cutOff += FFABS(filter2[i*filter2Size]);
1143
1144                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1145
1146                         /* preserve Monotonicity because the core can't handle the filter otherwise */
1147                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1148
1149                         // Move filter coeffs left
1150                         for(k=1; k<filter2Size; k++)
1151                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1152                         filter2[i*filter2Size + k - 1]= 0.0;
1153                         (*filterPos)[i]++;
1154                 }
1155
1156                 cutOff=0.0;
1157                 /* count near zeros on the right */
1158                 for(j=filter2Size-1; j>0; j--)
1159                 {
1160                         cutOff += FFABS(filter2[i*filter2Size + j]);
1161
1162                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1163                         min--;
1164                 }
1165
1166                 if(min>minFilterSize) minFilterSize= min;
1167         }
1168
1169         if (flags & SWS_CPU_CAPS_ALTIVEC) {
1170           // we can handle the special case 4,
1171           // so we don't want to go to the full 8
1172           if (minFilterSize < 5)
1173             filterAlign = 4;
1174
1175           // we really don't want to waste our time
1176           // doing useless computation, so fall-back on
1177           // the scalar C code for very small filter.
1178           // vectorizing is worth it only if you have
1179           // decent-sized vector.
1180           if (minFilterSize < 3)
1181             filterAlign = 1;
1182         }
1183
1184         if (flags & SWS_CPU_CAPS_MMX) {
1185                 // special case for unscaled vertical filtering
1186                 if(minFilterSize == 1 && filterAlign == 2)
1187                         filterAlign= 1;
1188         }
1189
1190         ASSERT(minFilterSize > 0)
1191         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1192         ASSERT(filterSize > 0)
1193         filter= av_malloc(filterSize*dstW*sizeof(double));
1194         if(filterSize >= MAX_FILTER_SIZE)
1195                 return -1;
1196         *outFilterSize= filterSize;
1197
1198         if(flags&SWS_PRINT_INFO)
1199                 MSG_V("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1200         /* try to reduce the filter-size (step2 reduce it) */
1201         for(i=0; i<dstW; i++)
1202         {
1203                 int j;
1204
1205                 for(j=0; j<filterSize; j++)
1206                 {
1207                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1208                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1209                 }
1210         }
1211         av_free(filter2); filter2=NULL;
1212
1213
1214         //FIXME try to align filterpos if possible
1215
1216         //fix borders
1217         for(i=0; i<dstW; i++)
1218         {
1219                 int j;
1220                 if((*filterPos)[i] < 0)
1221                 {
1222                         // Move filter coeffs left to compensate for filterPos
1223                         for(j=1; j<filterSize; j++)
1224                         {
1225                                 int left= FFMAX(j + (*filterPos)[i], 0);
1226                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1227                                 filter[i*filterSize + j]=0;
1228                         }
1229                         (*filterPos)[i]= 0;
1230                 }
1231
1232                 if((*filterPos)[i] + filterSize > srcW)
1233                 {
1234                         int shift= (*filterPos)[i] + filterSize - srcW;
1235                         // Move filter coeffs right to compensate for filterPos
1236                         for(j=filterSize-2; j>=0; j--)
1237                         {
1238                                 int right= FFMIN(j + shift, filterSize-1);
1239                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1240                                 filter[i*filterSize +j]=0;
1241                         }
1242                         (*filterPos)[i]= srcW - filterSize;
1243                 }
1244         }
1245
1246         // Note the +1 is for the MMXscaler which reads over the end
1247         /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1248         *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1249
1250         /* Normalize & Store in outFilter */
1251         for(i=0; i<dstW; i++)
1252         {
1253                 int j;
1254                 double error=0;
1255                 double sum=0;
1256                 double scale= one;
1257
1258                 for(j=0; j<filterSize; j++)
1259                 {
1260                         sum+= filter[i*filterSize + j];
1261                 }
1262                 scale/= sum;
1263                 for(j=0; j<*outFilterSize; j++)
1264                 {
1265                         double v= filter[i*filterSize + j]*scale + error;
1266                         int intV= floor(v + 0.5);
1267                         (*outFilter)[i*(*outFilterSize) + j]= intV;
1268                         error = v - intV;
1269                 }
1270         }
1271
1272         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1273         for(i=0; i<*outFilterSize; i++)
1274         {
1275                 int j= dstW*(*outFilterSize);
1276                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1277         }
1278
1279         av_free(filter);
1280         return 0;
1281 }
1282
1283 #ifdef COMPILE_MMX2
1284 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1285 {
1286         uint8_t *fragmentA;
1287         long imm8OfPShufW1A;
1288         long imm8OfPShufW2A;
1289         long fragmentLengthA;
1290         uint8_t *fragmentB;
1291         long imm8OfPShufW1B;
1292         long imm8OfPShufW2B;
1293         long fragmentLengthB;
1294         int fragmentPos;
1295
1296         int xpos, i;
1297
1298         // create an optimized horizontal scaling routine
1299
1300         //code fragment
1301
1302         asm volatile(
1303                 "jmp 9f                         \n\t"
1304         // Begin
1305                 "0:                             \n\t"
1306                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t"
1307                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t"
1308                 "movd 1(%%"REG_c", %%"REG_S"), %%mm1\n\t"
1309                 "punpcklbw %%mm7, %%mm1         \n\t"
1310                 "punpcklbw %%mm7, %%mm0         \n\t"
1311                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1312                 "1:                             \n\t"
1313                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1314                 "2:                             \n\t"
1315                 "psubw %%mm1, %%mm0             \n\t"
1316                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1317                 "pmullw %%mm3, %%mm0            \n\t"
1318                 "psllw $7, %%mm1                \n\t"
1319                 "paddw %%mm1, %%mm0             \n\t"
1320
1321                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1322
1323                 "add $8, %%"REG_a"              \n\t"
1324         // End
1325                 "9:                             \n\t"
1326 //              "int $3\n\t"
1327                 "lea 0b, %0                     \n\t"
1328                 "lea 1b, %1                     \n\t"
1329                 "lea 2b, %2                     \n\t"
1330                 "dec %1                         \n\t"
1331                 "dec %2                         \n\t"
1332                 "sub %0, %1                     \n\t"
1333                 "sub %0, %2                     \n\t"
1334                 "lea 9b, %3                     \n\t"
1335                 "sub %0, %3                     \n\t"
1336
1337
1338                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1339                 "=r" (fragmentLengthA)
1340         );
1341
1342         asm volatile(
1343                 "jmp 9f                         \n\t"
1344         // Begin
1345                 "0:                             \n\t"
1346                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t"
1347                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t"
1348                 "punpcklbw %%mm7, %%mm0         \n\t"
1349                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1350                 "1:                             \n\t"
1351                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1352                 "2:                             \n\t"
1353                 "psubw %%mm1, %%mm0             \n\t"
1354                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1355                 "pmullw %%mm3, %%mm0            \n\t"
1356                 "psllw $7, %%mm1                \n\t"
1357                 "paddw %%mm1, %%mm0             \n\t"
1358
1359                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1360
1361                 "add $8, %%"REG_a"              \n\t"
1362         // End
1363                 "9:                             \n\t"
1364 //              "int $3\n\t"
1365                 "lea 0b, %0                     \n\t"
1366                 "lea 1b, %1                     \n\t"
1367                 "lea 2b, %2                     \n\t"
1368                 "dec %1                         \n\t"
1369                 "dec %2                         \n\t"
1370                 "sub %0, %1                     \n\t"
1371                 "sub %0, %2                     \n\t"
1372                 "lea 9b, %3                     \n\t"
1373                 "sub %0, %3                     \n\t"
1374
1375
1376                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1377                 "=r" (fragmentLengthB)
1378         );
1379
1380         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1381         fragmentPos=0;
1382
1383         for(i=0; i<dstW/numSplits; i++)
1384         {
1385                 int xx=xpos>>16;
1386
1387                 if((i&3) == 0)
1388                 {
1389                         int a=0;
1390                         int b=((xpos+xInc)>>16) - xx;
1391                         int c=((xpos+xInc*2)>>16) - xx;
1392                         int d=((xpos+xInc*3)>>16) - xx;
1393
1394                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1395                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1396                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1397                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1398                         filterPos[i/2]= xx;
1399
1400                         if(d+1<4)
1401                         {
1402                                 int maxShift= 3-(d+1);
1403                                 int shift=0;
1404
1405                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1406
1407                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1408                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1409                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1410                                         a | (b<<2) | (c<<4) | (d<<6);
1411
1412                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1413                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1414
1415                                 if(shift && i>=shift)
1416                                 {
1417                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1418                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1419                                         filterPos[i/2]-=shift;
1420                                 }
1421
1422                                 fragmentPos+= fragmentLengthB;
1423                         }
1424                         else
1425                         {
1426                                 int maxShift= 3-d;
1427                                 int shift=0;
1428
1429                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1430
1431                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1432                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1433                                         a | (b<<2) | (c<<4) | (d<<6);
1434
1435                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1436                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1437
1438                                 if(shift && i>=shift)
1439                                 {
1440                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1441                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1442                                         filterPos[i/2]-=shift;
1443                                 }
1444
1445                                 fragmentPos+= fragmentLengthA;
1446                         }
1447
1448                         funnyCode[fragmentPos]= RET;
1449                 }
1450                 xpos+=xInc;
1451         }
1452         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1453 }
1454 #endif /* COMPILE_MMX2 */
1455
1456 static void globalInit(void){
1457     // generating tables:
1458     int i;
1459     for(i=0; i<768; i++){
1460         int c= clip_uint8(i-256);
1461         clip_table[i]=c;
1462     }
1463 }
1464
1465 static SwsFunc getSwsFunc(int flags){
1466
1467 #if defined(RUNTIME_CPUDETECT) && defined (CONFIG_GPL)
1468 #if defined(ARCH_X86)
1469         // ordered per speed fasterst first
1470         if(flags & SWS_CPU_CAPS_MMX2)
1471                 return swScale_MMX2;
1472         else if(flags & SWS_CPU_CAPS_3DNOW)
1473                 return swScale_3DNow;
1474         else if(flags & SWS_CPU_CAPS_MMX)
1475                 return swScale_MMX;
1476         else
1477                 return swScale_C;
1478
1479 #else
1480 #ifdef ARCH_POWERPC
1481         if(flags & SWS_CPU_CAPS_ALTIVEC)
1482           return swScale_altivec;
1483         else
1484           return swScale_C;
1485 #endif
1486         return swScale_C;
1487 #endif /* defined(ARCH_X86) */
1488 #else //RUNTIME_CPUDETECT
1489 #ifdef HAVE_MMX2
1490         return swScale_MMX2;
1491 #elif defined (HAVE_3DNOW)
1492         return swScale_3DNow;
1493 #elif defined (HAVE_MMX)
1494         return swScale_MMX;
1495 #elif defined (HAVE_ALTIVEC)
1496         return swScale_altivec;
1497 #else
1498         return swScale_C;
1499 #endif
1500 #endif //!RUNTIME_CPUDETECT
1501 }
1502
1503 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1504              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1505         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1506         /* Copy Y plane */
1507         if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1508                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1509         else
1510         {
1511                 int i;
1512                 uint8_t *srcPtr= src[0];
1513                 uint8_t *dstPtr= dst;
1514                 for(i=0; i<srcSliceH; i++)
1515                 {
1516                         memcpy(dstPtr, srcPtr, c->srcW);
1517                         srcPtr+= srcStride[0];
1518                         dstPtr+= dstStride[0];
1519                 }
1520         }
1521         dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1522         if (c->dstFormat == PIX_FMT_NV12)
1523                 interleaveBytes( src[1],src[2],dst,c->srcW/2,srcSliceH/2,srcStride[1],srcStride[2],dstStride[0] );
1524         else
1525                 interleaveBytes( src[2],src[1],dst,c->srcW/2,srcSliceH/2,srcStride[2],srcStride[1],dstStride[0] );
1526
1527         return srcSliceH;
1528 }
1529
1530 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1531              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1532         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1533
1534         yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1535
1536         return srcSliceH;
1537 }
1538
1539 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1540              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1541         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1542
1543         yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1544
1545         return srcSliceH;
1546 }
1547
1548 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1549 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1550                            int srcSliceH, uint8_t* dst[], int dstStride[]){
1551         const int srcFormat= c->srcFormat;
1552         const int dstFormat= c->dstFormat;
1553         const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1554         const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1555         const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1556         const int dstId= fmt_depth(dstFormat) >> 2;
1557         void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1558
1559         /* BGR -> BGR */
1560         if(   (isBGR(srcFormat) && isBGR(dstFormat))
1561            || (isRGB(srcFormat) && isRGB(dstFormat))){
1562                 switch(srcId | (dstId<<4)){
1563                 case 0x34: conv= rgb16to15; break;
1564                 case 0x36: conv= rgb24to15; break;
1565                 case 0x38: conv= rgb32to15; break;
1566                 case 0x43: conv= rgb15to16; break;
1567                 case 0x46: conv= rgb24to16; break;
1568                 case 0x48: conv= rgb32to16; break;
1569                 case 0x63: conv= rgb15to24; break;
1570                 case 0x64: conv= rgb16to24; break;
1571                 case 0x68: conv= rgb32to24; break;
1572                 case 0x83: conv= rgb15to32; break;
1573                 case 0x84: conv= rgb16to32; break;
1574                 case 0x86: conv= rgb24to32; break;
1575                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1576                                  sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1577                 }
1578         }else if(   (isBGR(srcFormat) && isRGB(dstFormat))
1579                  || (isRGB(srcFormat) && isBGR(dstFormat))){
1580                 switch(srcId | (dstId<<4)){
1581                 case 0x33: conv= rgb15tobgr15; break;
1582                 case 0x34: conv= rgb16tobgr15; break;
1583                 case 0x36: conv= rgb24tobgr15; break;
1584                 case 0x38: conv= rgb32tobgr15; break;
1585                 case 0x43: conv= rgb15tobgr16; break;
1586                 case 0x44: conv= rgb16tobgr16; break;
1587                 case 0x46: conv= rgb24tobgr16; break;
1588                 case 0x48: conv= rgb32tobgr16; break;
1589                 case 0x63: conv= rgb15tobgr24; break;
1590                 case 0x64: conv= rgb16tobgr24; break;
1591                 case 0x66: conv= rgb24tobgr24; break;
1592                 case 0x68: conv= rgb32tobgr24; break;
1593                 case 0x83: conv= rgb15tobgr32; break;
1594                 case 0x84: conv= rgb16tobgr32; break;
1595                 case 0x86: conv= rgb24tobgr32; break;
1596                 case 0x88: conv= rgb32tobgr32; break;
1597                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1598                                  sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1599                 }
1600         }else{
1601                 MSG_ERR("swScaler: internal error %s -> %s converter\n",
1602                          sws_format_name(srcFormat), sws_format_name(dstFormat));
1603         }
1604
1605         if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1606                 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1607         else
1608         {
1609                 int i;
1610                 uint8_t *srcPtr= src[0];
1611                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1612
1613                 for(i=0; i<srcSliceH; i++)
1614                 {
1615                         conv(srcPtr, dstPtr, c->srcW*srcBpp);
1616                         srcPtr+= srcStride[0];
1617                         dstPtr+= dstStride[0];
1618                 }
1619         }
1620         return srcSliceH;
1621 }
1622
1623 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1624              int srcSliceH, uint8_t* dst[], int dstStride[]){
1625
1626         rgb24toyv12(
1627                 src[0],
1628                 dst[0]+ srcSliceY    *dstStride[0],
1629                 dst[1]+(srcSliceY>>1)*dstStride[1],
1630                 dst[2]+(srcSliceY>>1)*dstStride[2],
1631                 c->srcW, srcSliceH,
1632                 dstStride[0], dstStride[1], srcStride[0]);
1633         return srcSliceH;
1634 }
1635
1636 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1637              int srcSliceH, uint8_t* dst[], int dstStride[]){
1638         int i;
1639
1640         /* copy Y */
1641         if(srcStride[0]==dstStride[0] && srcStride[0] > 0)
1642                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1643         else{
1644                 uint8_t *srcPtr= src[0];
1645                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1646
1647                 for(i=0; i<srcSliceH; i++)
1648                 {
1649                         memcpy(dstPtr, srcPtr, c->srcW);
1650                         srcPtr+= srcStride[0];
1651                         dstPtr+= dstStride[0];
1652                 }
1653         }
1654
1655         if(c->dstFormat==PIX_FMT_YUV420P){
1656                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1657                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1658         }else{
1659                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1660                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1661         }
1662         return srcSliceH;
1663 }
1664
1665 /* unscaled copy like stuff (assumes nearly identical formats) */
1666 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1667              int srcSliceH, uint8_t* dst[], int dstStride[]){
1668
1669         if(isPacked(c->srcFormat))
1670         {
1671                 if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1672                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1673                 else
1674                 {
1675                         int i;
1676                         uint8_t *srcPtr= src[0];
1677                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1678                         int length=0;
1679
1680                         /* universal length finder */
1681                         while(length+c->srcW <= FFABS(dstStride[0])
1682                            && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1683                         ASSERT(length!=0);
1684
1685                         for(i=0; i<srcSliceH; i++)
1686                         {
1687                                 memcpy(dstPtr, srcPtr, length);
1688                                 srcPtr+= srcStride[0];
1689                                 dstPtr+= dstStride[0];
1690                         }
1691                 }
1692         }
1693         else
1694         { /* Planar YUV or gray */
1695                 int plane;
1696                 for(plane=0; plane<3; plane++)
1697                 {
1698                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1699                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1700                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1701
1702                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1703                         {
1704                                 if(!isGray(c->dstFormat))
1705                                         memset(dst[plane], 128, dstStride[plane]*height);
1706                         }
1707                         else
1708                         {
1709                                 if(dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1710                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1711                                 else
1712                                 {
1713                                         int i;
1714                                         uint8_t *srcPtr= src[plane];
1715                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1716                                         for(i=0; i<height; i++)
1717                                         {
1718                                                 memcpy(dstPtr, srcPtr, length);
1719                                                 srcPtr+= srcStride[plane];
1720                                                 dstPtr+= dstStride[plane];
1721                                         }
1722                                 }
1723                         }
1724                 }
1725         }
1726         return srcSliceH;
1727 }
1728
1729 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1730              int srcSliceH, uint8_t* dst[], int dstStride[]){
1731
1732         int length= c->srcW;
1733         int y=      srcSliceY;
1734         int height= srcSliceH;
1735         int i, j;
1736         uint8_t *srcPtr= src[0];
1737         uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1738
1739         if(!isGray(c->dstFormat)){
1740                 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1741                 memset(dst[1], 128, dstStride[1]*height);
1742                 memset(dst[2], 128, dstStride[2]*height);
1743         }
1744         if(c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1745         for(i=0; i<height; i++)
1746         {
1747                 for(j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1748                 srcPtr+= srcStride[0];
1749                 dstPtr+= dstStride[0];
1750         }
1751         return srcSliceH;
1752 }
1753
1754 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1755              int srcSliceH, uint8_t* dst[], int dstStride[]){
1756
1757         int length= c->srcW;
1758         int y=      srcSliceY;
1759         int height= srcSliceH;
1760         int i, j;
1761         uint8_t *srcPtr= src[0];
1762         uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1763         for(i=0; i<height; i++)
1764         {
1765                 for(j=0; j<length; j++)
1766                 {
1767                         dstPtr[j<<1] = srcPtr[j];
1768                         dstPtr[(j<<1)+1] = srcPtr[j];
1769                 }
1770                 srcPtr+= srcStride[0];
1771                 dstPtr+= dstStride[0];
1772         }
1773         return srcSliceH;
1774 }
1775
1776 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1777              int srcSliceH, uint8_t* dst[], int dstStride[]){
1778
1779         int length= c->srcW;
1780         int y=      srcSliceY;
1781         int height= srcSliceH;
1782         int i, j;
1783         uint16_t *srcPtr= src[0];
1784         uint16_t *dstPtr= dst[0] + dstStride[0]*y/2;
1785         for(i=0; i<height; i++)
1786         {
1787                 for(j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
1788                 srcPtr+= srcStride[0]/2;
1789                 dstPtr+= dstStride[0]/2;
1790         }
1791         return srcSliceH;
1792 }
1793
1794
1795 static void getSubSampleFactors(int *h, int *v, int format){
1796         switch(format){
1797         case PIX_FMT_UYVY422:
1798         case PIX_FMT_YUYV422:
1799                 *h=1;
1800                 *v=0;
1801                 break;
1802         case PIX_FMT_YUV420P:
1803         case PIX_FMT_GRAY16BE:
1804         case PIX_FMT_GRAY16LE:
1805         case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
1806         case PIX_FMT_NV12:
1807         case PIX_FMT_NV21:
1808                 *h=1;
1809                 *v=1;
1810                 break;
1811         case PIX_FMT_YUV410P:
1812                 *h=2;
1813                 *v=2;
1814                 break;
1815         case PIX_FMT_YUV444P:
1816                 *h=0;
1817                 *v=0;
1818                 break;
1819         case PIX_FMT_YUV422P:
1820                 *h=1;
1821                 *v=0;
1822                 break;
1823         case PIX_FMT_YUV411P:
1824                 *h=2;
1825                 *v=0;
1826                 break;
1827         default:
1828                 *h=0;
1829                 *v=0;
1830                 break;
1831         }
1832 }
1833
1834 static uint16_t roundToInt16(int64_t f){
1835         int r= (f + (1<<15))>>16;
1836              if(r<-0x7FFF) return 0x8000;
1837         else if(r> 0x7FFF) return 0x7FFF;
1838         else               return r;
1839 }
1840
1841 /**
1842  * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1843  * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1844  * @return -1 if not supported
1845  */
1846 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1847         int64_t crv =  inv_table[0];
1848         int64_t cbu =  inv_table[1];
1849         int64_t cgu = -inv_table[2];
1850         int64_t cgv = -inv_table[3];
1851         int64_t cy  = 1<<16;
1852         int64_t oy  = 0;
1853
1854         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1855         memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1856         memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
1857
1858         c->brightness= brightness;
1859         c->contrast  = contrast;
1860         c->saturation= saturation;
1861         c->srcRange  = srcRange;
1862         c->dstRange  = dstRange;
1863
1864         c->uOffset=   0x0400040004000400LL;
1865         c->vOffset=   0x0400040004000400LL;
1866
1867         if(!srcRange){
1868                 cy= (cy*255) / 219;
1869                 oy= 16<<16;
1870         }
1871
1872         cy = (cy *contrast             )>>16;
1873         crv= (crv*contrast * saturation)>>32;
1874         cbu= (cbu*contrast * saturation)>>32;
1875         cgu= (cgu*contrast * saturation)>>32;
1876         cgv= (cgv*contrast * saturation)>>32;
1877
1878         oy -= 256*brightness;
1879
1880         c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
1881         c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
1882         c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1883         c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1884         c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1885         c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
1886
1887         yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1888         //FIXME factorize
1889
1890 #ifdef COMPILE_ALTIVEC
1891         if (c->flags & SWS_CPU_CAPS_ALTIVEC)
1892             yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
1893 #endif
1894         return 0;
1895 }
1896
1897 /**
1898  * @return -1 if not supported
1899  */
1900 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1901         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1902
1903         *inv_table = c->srcColorspaceTable;
1904         *table     = c->dstColorspaceTable;
1905         *srcRange  = c->srcRange;
1906         *dstRange  = c->dstRange;
1907         *brightness= c->brightness;
1908         *contrast  = c->contrast;
1909         *saturation= c->saturation;
1910
1911         return 0;
1912 }
1913
1914 static int handle_jpeg(int *format)
1915 {
1916         switch (*format) {
1917                 case PIX_FMT_YUVJ420P:
1918                         *format = PIX_FMT_YUV420P;
1919                         return 1;
1920                 case PIX_FMT_YUVJ422P:
1921                         *format = PIX_FMT_YUV422P;
1922                         return 1;
1923                 case PIX_FMT_YUVJ444P:
1924                         *format = PIX_FMT_YUV444P;
1925                         return 1;
1926                 default:
1927                         return 0;
1928         }
1929 }
1930
1931 SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
1932                          SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
1933
1934         SwsContext *c;
1935         int i;
1936         int usesVFilter, usesHFilter;
1937         int unscaled, needsDither;
1938         int srcRange, dstRange;
1939         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1940 #if defined(ARCH_X86)
1941         if(flags & SWS_CPU_CAPS_MMX)
1942                 asm volatile("emms\n\t"::: "memory");
1943 #endif
1944
1945 #if !defined(RUNTIME_CPUDETECT) || !defined (CONFIG_GPL) //ensure that the flags match the compiled variant if cpudetect is off
1946         flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1947 #ifdef HAVE_MMX2
1948         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1949 #elif defined (HAVE_3DNOW)
1950         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1951 #elif defined (HAVE_MMX)
1952         flags |= SWS_CPU_CAPS_MMX;
1953 #elif defined (HAVE_ALTIVEC)
1954         flags |= SWS_CPU_CAPS_ALTIVEC;
1955 #endif
1956 #endif /* RUNTIME_CPUDETECT */
1957         if(clip_table[512] != 255) globalInit();
1958         if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1959
1960         unscaled = (srcW == dstW && srcH == dstH);
1961         needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
1962                      && (fmt_depth(dstFormat))<24
1963                      && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1964
1965         srcRange = handle_jpeg(&srcFormat);
1966         dstRange = handle_jpeg(&dstFormat);
1967
1968         if(!isSupportedIn(srcFormat))
1969         {
1970                 MSG_ERR("swScaler: %s is not supported as input format\n", sws_format_name(srcFormat));
1971                 return NULL;
1972         }
1973         if(!isSupportedOut(dstFormat))
1974         {
1975                 MSG_ERR("swScaler: %s is not supported as output format\n", sws_format_name(dstFormat));
1976                 return NULL;
1977         }
1978
1979         /* sanity check */
1980         if(srcW<4 || srcH<1 || dstW<8 || dstH<1) //FIXME check if these are enough and try to lowwer them after fixing the relevant parts of the code
1981         {
1982                  MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
1983                         srcW, srcH, dstW, dstH);
1984                 return NULL;
1985         }
1986
1987         if(!dstFilter) dstFilter= &dummyFilter;
1988         if(!srcFilter) srcFilter= &dummyFilter;
1989
1990         c= av_mallocz(sizeof(SwsContext));
1991
1992         c->srcW= srcW;
1993         c->srcH= srcH;
1994         c->dstW= dstW;
1995         c->dstH= dstH;
1996         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1997         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1998         c->flags= flags;
1999         c->dstFormat= dstFormat;
2000         c->srcFormat= srcFormat;
2001         c->vRounder= 4* 0x0001000100010001ULL;
2002
2003         usesHFilter= usesVFilter= 0;
2004         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
2005         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
2006         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
2007         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
2008         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
2009         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
2010         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
2011         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
2012
2013         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2014         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2015
2016         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2017         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2018
2019         // drop some chroma lines if the user wants it
2020         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2021         c->chrSrcVSubSample+= c->vChrDrop;
2022
2023         // drop every 2. pixel for chroma calculation unless user wants full chroma
2024         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
2025                 c->chrSrcHSubSample=1;
2026
2027         if(param){
2028                 c->param[0] = param[0];
2029                 c->param[1] = param[1];
2030         }else{
2031                 c->param[0] =
2032                 c->param[1] = SWS_PARAM_DEFAULT;
2033         }
2034
2035         c->chrIntHSubSample= c->chrDstHSubSample;
2036         c->chrIntVSubSample= c->chrSrcVSubSample;
2037
2038         // note the -((-x)>>y) is so that we allways round toward +inf
2039         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2040         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2041         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2042         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2043
2044         sws_setColorspaceDetails(c, Inverse_Table_6_9[SWS_CS_DEFAULT], srcRange, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16);
2045
2046         /* unscaled special Cases */
2047         if(unscaled && !usesHFilter && !usesVFilter)
2048         {
2049                 /* yv12_to_nv12 */
2050                 if(srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2051                 {
2052                         c->swScale= PlanarToNV12Wrapper;
2053                 }
2054 #ifdef CONFIG_GPL
2055                 /* yuv2bgr */
2056                 if((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
2057                 {
2058                         c->swScale= yuv2rgb_get_func_ptr(c);
2059                 }
2060 #endif
2061
2062                 if( srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P )
2063                 {
2064                         c->swScale= yvu9toyv12Wrapper;
2065                 }
2066
2067                 /* bgr24toYV12 */
2068                 if(srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P)
2069                         c->swScale= bgr24toyv12Wrapper;
2070
2071                 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2072                 if(   (isBGR(srcFormat) || isRGB(srcFormat))
2073                    && (isBGR(dstFormat) || isRGB(dstFormat))
2074                    && !needsDither)
2075                         c->swScale= rgb2rgbWrapper;
2076
2077                 /* LQ converters if -sws 0 or -sws 4*/
2078                 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2079                         /* rgb/bgr -> rgb/bgr (dither needed forms) */
2080                         if(  (isBGR(srcFormat) || isRGB(srcFormat))
2081                           && (isBGR(dstFormat) || isRGB(dstFormat))
2082                           && needsDither)
2083                                 c->swScale= rgb2rgbWrapper;
2084
2085                         /* yv12_to_yuy2 */
2086                         if(srcFormat == PIX_FMT_YUV420P &&
2087                             (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422))
2088                         {
2089                                 if (dstFormat == PIX_FMT_YUYV422)
2090                                     c->swScale= PlanarToYuy2Wrapper;
2091                                 else
2092                                     c->swScale= PlanarToUyvyWrapper;
2093                         }
2094                 }
2095
2096 #ifdef COMPILE_ALTIVEC
2097                 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2098                     ((srcFormat == PIX_FMT_YUV420P &&
2099                       (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422)))) {
2100                   // unscaled YV12 -> packed YUV, we want speed
2101                   if (dstFormat == PIX_FMT_YUYV422)
2102                     c->swScale= yv12toyuy2_unscaled_altivec;
2103                   else
2104                     c->swScale= yv12touyvy_unscaled_altivec;
2105                 }
2106 #endif
2107
2108                 /* simple copy */
2109                 if(   srcFormat == dstFormat
2110                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2111                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2112                   )
2113                 {
2114                         c->swScale= simpleCopy;
2115                 }
2116
2117                 /* gray16{le,be} conversions */
2118                 if(isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2119                 {
2120                         c->swScale= gray16togray;
2121                 }
2122                 if((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2123                 {
2124                         c->swScale= graytogray16;
2125                 }
2126                 if(srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2127                 {
2128                         c->swScale= gray16swap;
2129                 }
2130
2131                 if(c->swScale){
2132                         if(flags&SWS_PRINT_INFO)
2133                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2134                                         sws_format_name(srcFormat), sws_format_name(dstFormat));
2135                         return c;
2136                 }
2137         }
2138
2139         if(flags & SWS_CPU_CAPS_MMX2)
2140         {
2141                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2142                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2143                 {
2144                         if(flags&SWS_PRINT_INFO)
2145                                 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
2146                 }
2147                 if(usesHFilter) c->canMMX2BeUsed=0;
2148         }
2149         else
2150                 c->canMMX2BeUsed=0;
2151
2152         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2153         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2154
2155         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2156         // but only for the FAST_BILINEAR mode otherwise do correct scaling
2157         // n-2 is the last chrominance sample available
2158         // this is not perfect, but noone shuld notice the difference, the more correct variant
2159         // would be like the vertical one, but that would require some special code for the
2160         // first and last pixel
2161         if(flags&SWS_FAST_BILINEAR)
2162         {
2163                 if(c->canMMX2BeUsed)
2164                 {
2165                         c->lumXInc+= 20;
2166                         c->chrXInc+= 20;
2167                 }
2168                 //we don't use the x86asm scaler if mmx is available
2169                 else if(flags & SWS_CPU_CAPS_MMX)
2170                 {
2171                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2172                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2173                 }
2174         }
2175
2176         /* precalculate horizontal scaler filter coefficients */
2177         {
2178                 const int filterAlign=
2179                   (flags & SWS_CPU_CAPS_MMX) ? 4 :
2180                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2181                   1;
2182
2183                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2184                                  srcW      ,       dstW, filterAlign, 1<<14,
2185                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2186                                  srcFilter->lumH, dstFilter->lumH, c->param);
2187                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2188                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2189                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2190                                  srcFilter->chrH, dstFilter->chrH, c->param);
2191
2192 #define MAX_FUNNY_CODE_SIZE 10000
2193 #if defined(COMPILE_MMX2)
2194 // can't downscale !!!
2195                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2196                 {
2197 #ifdef MAP_ANONYMOUS
2198                         c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2199                         c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2200 #else
2201                         c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2202                         c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2203 #endif
2204
2205                         c->lumMmx2Filter   = av_malloc((dstW        /8+8)*sizeof(int16_t));
2206                         c->chrMmx2Filter   = av_malloc((c->chrDstW  /4+8)*sizeof(int16_t));
2207                         c->lumMmx2FilterPos= av_malloc((dstW      /2/8+8)*sizeof(int32_t));
2208                         c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2209
2210                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2211                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2212                 }
2213 #endif /* defined(COMPILE_MMX2) */
2214         } // Init Horizontal stuff
2215
2216
2217
2218         /* precalculate vertical scaler filter coefficients */
2219         {
2220                 const int filterAlign=
2221                   (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2222                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2223                   1;
2224
2225                 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2226                                 srcH      ,        dstH, filterAlign, (1<<12)-4,
2227                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2228                                 srcFilter->lumV, dstFilter->lumV, c->param);
2229                 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2230                                 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2231                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2232                                 srcFilter->chrV, dstFilter->chrV, c->param);
2233
2234 #ifdef HAVE_ALTIVEC
2235                 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2236                 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2237
2238                 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2239                   int j;
2240                   short *p = (short *)&c->vYCoeffsBank[i];
2241                   for (j=0;j<8;j++)
2242                     p[j] = c->vLumFilter[i];
2243                 }
2244
2245                 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2246                   int j;
2247                   short *p = (short *)&c->vCCoeffsBank[i];
2248                   for (j=0;j<8;j++)
2249                     p[j] = c->vChrFilter[i];
2250                 }
2251 #endif
2252         }
2253
2254         // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2255         c->vLumBufSize= c->vLumFilterSize;
2256         c->vChrBufSize= c->vChrFilterSize;
2257         for(i=0; i<dstH; i++)
2258         {
2259                 int chrI= i*c->chrDstH / dstH;
2260                 int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
2261                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2262
2263                 nextSlice>>= c->chrSrcVSubSample;
2264                 nextSlice<<= c->chrSrcVSubSample;
2265                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
2266                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
2267                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2268                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2269         }
2270
2271         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2272         c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2273         c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2274         //Note we need at least one pixel more at the end because of the mmx code (just in case someone wanna replace the 4000/8000)
2275         /* align at 16 bytes for AltiVec */
2276         for(i=0; i<c->vLumBufSize; i++)
2277                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(4000);
2278         for(i=0; i<c->vChrBufSize; i++)
2279                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc(8000);
2280
2281         //try to avoid drawing green stuff between the right end and the stride end
2282         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2283
2284         ASSERT(c->chrDstH <= dstH)
2285
2286         if(flags&SWS_PRINT_INFO)
2287         {
2288 #ifdef DITHER1XBPP
2289                 char *dither= " dithered";
2290 #else
2291                 char *dither= "";
2292 #endif
2293                 if(flags&SWS_FAST_BILINEAR)
2294                         MSG_INFO("SwScaler: FAST_BILINEAR scaler, ");
2295                 else if(flags&SWS_BILINEAR)
2296                         MSG_INFO("SwScaler: BILINEAR scaler, ");
2297                 else if(flags&SWS_BICUBIC)
2298                         MSG_INFO("SwScaler: BICUBIC scaler, ");
2299                 else if(flags&SWS_X)
2300                         MSG_INFO("SwScaler: Experimental scaler, ");
2301                 else if(flags&SWS_POINT)
2302                         MSG_INFO("SwScaler: Nearest Neighbor / POINT scaler, ");
2303                 else if(flags&SWS_AREA)
2304                         MSG_INFO("SwScaler: Area Averageing scaler, ");
2305                 else if(flags&SWS_BICUBLIN)
2306                         MSG_INFO("SwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2307                 else if(flags&SWS_GAUSS)
2308                         MSG_INFO("SwScaler: Gaussian scaler, ");
2309                 else if(flags&SWS_SINC)
2310                         MSG_INFO("SwScaler: Sinc scaler, ");
2311                 else if(flags&SWS_LANCZOS)
2312                         MSG_INFO("SwScaler: Lanczos scaler, ");
2313                 else if(flags&SWS_SPLINE)
2314                         MSG_INFO("SwScaler: Bicubic spline scaler, ");
2315                 else
2316                         MSG_INFO("SwScaler: ehh flags invalid?! ");
2317
2318                 if(dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2319                         MSG_INFO("from %s to%s %s ",
2320                                 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2321                 else
2322                         MSG_INFO("from %s to %s ",
2323                                 sws_format_name(srcFormat), sws_format_name(dstFormat));
2324
2325                 if(flags & SWS_CPU_CAPS_MMX2)
2326                         MSG_INFO("using MMX2\n");
2327                 else if(flags & SWS_CPU_CAPS_3DNOW)
2328                         MSG_INFO("using 3DNOW\n");
2329                 else if(flags & SWS_CPU_CAPS_MMX)
2330                         MSG_INFO("using MMX\n");
2331                 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2332                         MSG_INFO("using AltiVec\n");
2333                 else
2334                         MSG_INFO("using C\n");
2335         }
2336
2337         if(flags & SWS_PRINT_INFO)
2338         {
2339                 if(flags & SWS_CPU_CAPS_MMX)
2340                 {
2341                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2342                                 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2343                         else
2344                         {
2345                                 if(c->hLumFilterSize==4)
2346                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2347                                 else if(c->hLumFilterSize==8)
2348                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2349                                 else
2350                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2351
2352                                 if(c->hChrFilterSize==4)
2353                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2354                                 else if(c->hChrFilterSize==8)
2355                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2356                                 else
2357                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2358                         }
2359                 }
2360                 else
2361                 {
2362 #if defined(ARCH_X86)
2363                         MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2364 #else
2365                         if(flags & SWS_FAST_BILINEAR)
2366                                 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2367                         else
2368                                 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2369 #endif
2370                 }
2371                 if(isPlanarYUV(dstFormat))
2372                 {
2373                         if(c->vLumFilterSize==1)
2374                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2375                         else
2376                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2377                 }
2378                 else
2379                 {
2380                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2381                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2382                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2383                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2384                                 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2385                         else
2386                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2387                 }
2388
2389                 if(dstFormat==PIX_FMT_BGR24)
2390                         MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2391                                 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2392                 else if(dstFormat==PIX_FMT_RGB32)
2393                         MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2394                 else if(dstFormat==PIX_FMT_BGR565)
2395                         MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2396                 else if(dstFormat==PIX_FMT_BGR555)
2397                         MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2398
2399                 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2400         }
2401         if(flags & SWS_PRINT_INFO)
2402         {
2403                 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2404                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2405                 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2406                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2407         }
2408
2409         c->swScale= getSwsFunc(flags);
2410         return c;
2411 }
2412
2413 /**
2414  * swscale warper, so we don't need to export the SwsContext.
2415  * assumes planar YUV to be in YUV order instead of YVU
2416  */
2417 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2418                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2419         if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2420             MSG_ERR("swScaler: slices start in the middle!\n");
2421             return 0;
2422         }
2423         if (c->sliceDir == 0) {
2424             if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2425         }
2426
2427         // copy strides, so they can safely be modified
2428         if (c->sliceDir == 1) {
2429             // slices go from top to bottom
2430             int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
2431             int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
2432             return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2433         } else {
2434             // slices go from bottom to top => we flip the image internally
2435             uint8_t* src2[3]= {src[0] + (srcSliceH-1)*srcStride[0],
2436                                src[1] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1],
2437                                src[2] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2]
2438             };
2439             uint8_t* dst2[3]= {dst[0] + (c->dstH-1)*dstStride[0],
2440                                dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2441                                dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2442             int srcStride2[3]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2443             int dstStride2[3]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2444
2445             return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2446         }
2447 }
2448
2449 /**
2450  * swscale warper, so we don't need to export the SwsContext
2451  */
2452 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2453                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2454         return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2455 }
2456
2457 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2458                                 float lumaSharpen, float chromaSharpen,
2459                                 float chromaHShift, float chromaVShift,
2460                                 int verbose)
2461 {
2462         SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2463
2464         if(lumaGBlur!=0.0){
2465                 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2466                 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2467         }else{
2468                 filter->lumH= sws_getIdentityVec();
2469                 filter->lumV= sws_getIdentityVec();
2470         }
2471
2472         if(chromaGBlur!=0.0){
2473                 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2474                 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2475         }else{
2476                 filter->chrH= sws_getIdentityVec();
2477                 filter->chrV= sws_getIdentityVec();
2478         }
2479
2480         if(chromaSharpen!=0.0){
2481                 SwsVector *id= sws_getIdentityVec();
2482                 sws_scaleVec(filter->chrH, -chromaSharpen);
2483                 sws_scaleVec(filter->chrV, -chromaSharpen);
2484                 sws_addVec(filter->chrH, id);
2485                 sws_addVec(filter->chrV, id);
2486                 sws_freeVec(id);
2487         }
2488
2489         if(lumaSharpen!=0.0){
2490                 SwsVector *id= sws_getIdentityVec();
2491                 sws_scaleVec(filter->lumH, -lumaSharpen);
2492                 sws_scaleVec(filter->lumV, -lumaSharpen);
2493                 sws_addVec(filter->lumH, id);
2494                 sws_addVec(filter->lumV, id);
2495                 sws_freeVec(id);
2496         }
2497
2498         if(chromaHShift != 0.0)
2499                 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2500
2501         if(chromaVShift != 0.0)
2502                 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2503
2504         sws_normalizeVec(filter->chrH, 1.0);
2505         sws_normalizeVec(filter->chrV, 1.0);
2506         sws_normalizeVec(filter->lumH, 1.0);
2507         sws_normalizeVec(filter->lumV, 1.0);
2508
2509         if(verbose) sws_printVec(filter->chrH);
2510         if(verbose) sws_printVec(filter->lumH);
2511
2512         return filter;
2513 }
2514
2515 /**
2516  * returns a normalized gaussian curve used to filter stuff
2517  * quality=3 is high quality, lowwer is lowwer quality
2518  */
2519 SwsVector *sws_getGaussianVec(double variance, double quality){
2520         const int length= (int)(variance*quality + 0.5) | 1;
2521         int i;
2522         double *coeff= av_malloc(length*sizeof(double));
2523         double middle= (length-1)*0.5;
2524         SwsVector *vec= av_malloc(sizeof(SwsVector));
2525
2526         vec->coeff= coeff;
2527         vec->length= length;
2528
2529         for(i=0; i<length; i++)
2530         {
2531                 double dist= i-middle;
2532                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2533         }
2534
2535         sws_normalizeVec(vec, 1.0);
2536
2537         return vec;
2538 }
2539
2540 SwsVector *sws_getConstVec(double c, int length){
2541         int i;
2542         double *coeff= av_malloc(length*sizeof(double));
2543         SwsVector *vec= av_malloc(sizeof(SwsVector));
2544
2545         vec->coeff= coeff;
2546         vec->length= length;
2547
2548         for(i=0; i<length; i++)
2549                 coeff[i]= c;
2550
2551         return vec;
2552 }
2553
2554
2555 SwsVector *sws_getIdentityVec(void){
2556         return sws_getConstVec(1.0, 1);
2557 }
2558
2559 double sws_dcVec(SwsVector *a){
2560         int i;
2561         double sum=0;
2562
2563         for(i=0; i<a->length; i++)
2564                 sum+= a->coeff[i];
2565
2566         return sum;
2567 }
2568
2569 void sws_scaleVec(SwsVector *a, double scalar){
2570         int i;
2571
2572         for(i=0; i<a->length; i++)
2573                 a->coeff[i]*= scalar;
2574 }
2575
2576 void sws_normalizeVec(SwsVector *a, double height){
2577         sws_scaleVec(a, height/sws_dcVec(a));
2578 }
2579
2580 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2581         int length= a->length + b->length - 1;
2582         double *coeff= av_malloc(length*sizeof(double));
2583         int i, j;
2584         SwsVector *vec= av_malloc(sizeof(SwsVector));
2585
2586         vec->coeff= coeff;
2587         vec->length= length;
2588
2589         for(i=0; i<length; i++) coeff[i]= 0.0;
2590
2591         for(i=0; i<a->length; i++)
2592         {
2593                 for(j=0; j<b->length; j++)
2594                 {
2595                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2596                 }
2597         }
2598
2599         return vec;
2600 }
2601
2602 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2603         int length= FFMAX(a->length, b->length);
2604         double *coeff= av_malloc(length*sizeof(double));
2605         int i;
2606         SwsVector *vec= av_malloc(sizeof(SwsVector));
2607
2608         vec->coeff= coeff;
2609         vec->length= length;
2610
2611         for(i=0; i<length; i++) coeff[i]= 0.0;
2612
2613         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2614         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2615
2616         return vec;
2617 }
2618
2619 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2620         int length= FFMAX(a->length, b->length);
2621         double *coeff= av_malloc(length*sizeof(double));
2622         int i;
2623         SwsVector *vec= av_malloc(sizeof(SwsVector));
2624
2625         vec->coeff= coeff;
2626         vec->length= length;
2627
2628         for(i=0; i<length; i++) coeff[i]= 0.0;
2629
2630         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2631         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2632
2633         return vec;
2634 }
2635
2636 /* shift left / or right if "shift" is negative */
2637 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2638         int length= a->length + FFABS(shift)*2;
2639         double *coeff= av_malloc(length*sizeof(double));
2640         int i;
2641         SwsVector *vec= av_malloc(sizeof(SwsVector));
2642
2643         vec->coeff= coeff;
2644         vec->length= length;
2645
2646         for(i=0; i<length; i++) coeff[i]= 0.0;
2647
2648         for(i=0; i<a->length; i++)
2649         {
2650                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2651         }
2652
2653         return vec;
2654 }
2655
2656 void sws_shiftVec(SwsVector *a, int shift){
2657         SwsVector *shifted= sws_getShiftedVec(a, shift);
2658         av_free(a->coeff);
2659         a->coeff= shifted->coeff;
2660         a->length= shifted->length;
2661         av_free(shifted);
2662 }
2663
2664 void sws_addVec(SwsVector *a, SwsVector *b){
2665         SwsVector *sum= sws_sumVec(a, b);
2666         av_free(a->coeff);
2667         a->coeff= sum->coeff;
2668         a->length= sum->length;
2669         av_free(sum);
2670 }
2671
2672 void sws_subVec(SwsVector *a, SwsVector *b){
2673         SwsVector *diff= sws_diffVec(a, b);
2674         av_free(a->coeff);
2675         a->coeff= diff->coeff;
2676         a->length= diff->length;
2677         av_free(diff);
2678 }
2679
2680 void sws_convVec(SwsVector *a, SwsVector *b){
2681         SwsVector *conv= sws_getConvVec(a, b);
2682         av_free(a->coeff);
2683         a->coeff= conv->coeff;
2684         a->length= conv->length;
2685         av_free(conv);
2686 }
2687
2688 SwsVector *sws_cloneVec(SwsVector *a){
2689         double *coeff= av_malloc(a->length*sizeof(double));
2690         int i;
2691         SwsVector *vec= av_malloc(sizeof(SwsVector));
2692
2693         vec->coeff= coeff;
2694         vec->length= a->length;
2695
2696         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2697
2698         return vec;
2699 }
2700
2701 void sws_printVec(SwsVector *a){
2702         int i;
2703         double max=0;
2704         double min=0;
2705         double range;
2706
2707         for(i=0; i<a->length; i++)
2708                 if(a->coeff[i]>max) max= a->coeff[i];
2709
2710         for(i=0; i<a->length; i++)
2711                 if(a->coeff[i]<min) min= a->coeff[i];
2712
2713         range= max - min;
2714
2715         for(i=0; i<a->length; i++)
2716         {
2717                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2718                 MSG_DBG2("%1.3f ", a->coeff[i]);
2719                 for(;x>0; x--) MSG_DBG2(" ");
2720                 MSG_DBG2("|\n");
2721         }
2722 }
2723
2724 void sws_freeVec(SwsVector *a){
2725         if(!a) return;
2726         av_free(a->coeff);
2727         a->coeff=NULL;
2728         a->length=0;
2729         av_free(a);
2730 }
2731
2732 void sws_freeFilter(SwsFilter *filter){
2733         if(!filter) return;
2734
2735         if(filter->lumH) sws_freeVec(filter->lumH);
2736         if(filter->lumV) sws_freeVec(filter->lumV);
2737         if(filter->chrH) sws_freeVec(filter->chrH);
2738         if(filter->chrV) sws_freeVec(filter->chrV);
2739         av_free(filter);
2740 }
2741
2742
2743 void sws_freeContext(SwsContext *c){
2744         int i;
2745         if(!c) return;
2746
2747         if(c->lumPixBuf)
2748         {
2749                 for(i=0; i<c->vLumBufSize; i++)
2750                 {
2751                         av_free(c->lumPixBuf[i]);
2752                         c->lumPixBuf[i]=NULL;
2753                 }
2754                 av_free(c->lumPixBuf);
2755                 c->lumPixBuf=NULL;
2756         }
2757
2758         if(c->chrPixBuf)
2759         {
2760                 for(i=0; i<c->vChrBufSize; i++)
2761                 {
2762                         av_free(c->chrPixBuf[i]);
2763                         c->chrPixBuf[i]=NULL;
2764                 }
2765                 av_free(c->chrPixBuf);
2766                 c->chrPixBuf=NULL;
2767         }
2768
2769         av_free(c->vLumFilter);
2770         c->vLumFilter = NULL;
2771         av_free(c->vChrFilter);
2772         c->vChrFilter = NULL;
2773         av_free(c->hLumFilter);
2774         c->hLumFilter = NULL;
2775         av_free(c->hChrFilter);
2776         c->hChrFilter = NULL;
2777 #ifdef HAVE_ALTIVEC
2778         av_free(c->vYCoeffsBank);
2779         c->vYCoeffsBank = NULL;
2780         av_free(c->vCCoeffsBank);
2781         c->vCCoeffsBank = NULL;
2782 #endif
2783
2784         av_free(c->vLumFilterPos);
2785         c->vLumFilterPos = NULL;
2786         av_free(c->vChrFilterPos);
2787         c->vChrFilterPos = NULL;
2788         av_free(c->hLumFilterPos);
2789         c->hLumFilterPos = NULL;
2790         av_free(c->hChrFilterPos);
2791         c->hChrFilterPos = NULL;
2792
2793 #if defined(ARCH_X86) && defined(CONFIG_GPL)
2794 #ifdef MAP_ANONYMOUS
2795         if(c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
2796         if(c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
2797 #else
2798         av_free(c->funnyYCode);
2799         av_free(c->funnyUVCode);
2800 #endif
2801         c->funnyYCode=NULL;
2802         c->funnyUVCode=NULL;
2803 #endif /* defined(ARCH_X86) */
2804
2805         av_free(c->lumMmx2Filter);
2806         c->lumMmx2Filter=NULL;
2807         av_free(c->chrMmx2Filter);
2808         c->chrMmx2Filter=NULL;
2809         av_free(c->lumMmx2FilterPos);
2810         c->lumMmx2FilterPos=NULL;
2811         av_free(c->chrMmx2FilterPos);
2812         c->chrMmx2FilterPos=NULL;
2813         av_free(c->yuvTable);
2814         c->yuvTable=NULL;
2815
2816         av_free(c);
2817 }
2818
2819 /**
2820  * Checks if context is valid or reallocs a new one instead.
2821  * If context is NULL, just calls sws_getContext() to get a new one.
2822  * Otherwise, checks if the parameters are the same already saved in context.
2823  * If that is the case, returns the current context.
2824  * Otherwise, frees context and gets a new one.
2825  *
2826  * Be warned that srcFilter, dstFilter are not checked, they are
2827  * asumed to remain valid.
2828  */
2829 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
2830                         int srcW, int srcH, int srcFormat,
2831                         int dstW, int dstH, int dstFormat, int flags,
2832                         SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
2833 {
2834     if (context != NULL) {
2835         if ((context->srcW != srcW) || (context->srcH != srcH) ||
2836             (context->srcFormat != srcFormat) ||
2837             (context->dstW != dstW) || (context->dstH != dstH) ||
2838             (context->dstFormat != dstFormat) || (context->flags != flags) ||
2839             (context->param != param))
2840         {
2841             sws_freeContext(context);
2842             context = NULL;
2843         }
2844     }
2845     if (context == NULL) {
2846         return sws_getContext(srcW, srcH, srcFormat,
2847                         dstW, dstH, dstFormat, flags,
2848                         srcFilter, dstFilter, param);
2849     }
2850     return context;
2851 }
2852