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

   1 /*
   2     Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
   3
   4     This program is free software; you can redistribute it and/or modify
   5     it under the terms of the GNU General Public License as published by
   6     the Free Software Foundation; either version 2 of the License, or
   7     (at your option) any later version.
   8
   9     This program is distributed in the hope that it will be useful,
  10     but WITHOUT ANY WARRANTY; without even the implied warranty of
  11     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12     GNU General Public License for more details.
  13
  14     You should have received a copy of the GNU General Public License
  15     along with this program; if not, write to the Free Software
  16     Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  17 */
  18
  19 /*
  20   supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
  21   supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
  22   {BGR,RGB}{1,4,8,15,16} support dithering
  23
  24   unscaled special converters (YV12=I420=IYUV, Y800=Y8)
  25   YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
  26   x -> x
  27   YUV9 -> YV12
  28   YUV9/YV12 -> Y800
  29   Y800 -> YUV9/YV12
  30   BGR24 -> BGR32 & RGB24 -> RGB32
  31   BGR32 -> BGR24 & RGB32 -> RGB24
  32   BGR15 -> BGR16
  33 */
  34
  35 /*
  36 tested special converters (most are tested actually but i didnt write it down ...)
  37  YV12 -> BGR16
  38  YV12 -> YV12
  39  BGR15 -> BGR16
  40  BGR16 -> BGR16
  41  YVU9 -> YV12
  42
  43 untested special converters
  44   YV12/I420 -> BGR15/BGR24/BGR32 (its the yuv2rgb stuff, so it should be ok)
  45   YV12/I420 -> YV12/I420
  46   YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
  47   BGR24 -> BGR32 & RGB24 -> RGB32
  48   BGR32 -> BGR24 & RGB32 -> RGB24
  49   BGR24 -> YV12
  50 */
  51
  52 #include <inttypes.h>
  53 #include <string.h>
  54 #include <math.h>
  55 #include <stdio.h>
  56 #include "../config.h"
  57 #include "../mangle.h"
  58 #include <assert.h>
  59 #ifdef HAVE_MALLOC_H
  60 #include <malloc.h>
  61 #else
  62 #include <stdlib.h>
  63 #endif
  64 #include "swscale.h"
  65 #include "swscale_internal.h"
  66 #include "../cpudetect.h"
  67 #include "../bswap.h"
  68 #include "../libvo/img_format.h"
  69 #include "rgb2rgb.h"
  70 #include "../libvo/fastmemcpy.h"
  71
  72 #undef MOVNTQ
  73 #undef PAVGB
  74
  75 //#undef HAVE_MMX2
  76 //#define HAVE_3DNOW
  77 //#undef HAVE_MMX
  78 //#undef ARCH_X86
  79 //#define WORDS_BIGENDIAN
  80 #define DITHER1XBPP
  81
  82 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
  83
  84 #define RET 0xC3 //near return opcode for X86
  85
  86 #ifdef MP_DEBUG
  87 #define ASSERT(x) assert(x);
  88 #else
  89 #define ASSERT(x) ;
  90 #endif
  91
  92 #ifdef M_PI
  93 #define PI M_PI
  94 #else
  95 #define PI 3.14159265358979323846
  96 #endif
  97
  98 //FIXME replace this with something faster
  99 #define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YVU9 \
 100                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
 101 #define isYUV(x)       ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x))
 102 #define isGray(x)      ((x)==IMGFMT_Y800)
 103 #define isRGB(x)       (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
 104 #define isBGR(x)       (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
 105 #define isSupportedIn(x)  ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
 106                         || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
 107                         || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
 108                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\
 109                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
 110 #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
 111                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
 112                         || isRGB(x) || isBGR(x)\
 113                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
 114 #define isPacked(x)    ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x))
 115
 116 #define RGB2YUV_SHIFT 16
 117 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
 118 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
 119 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
 120 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
 121 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
 122 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
 123 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
 124 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
 125 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
 126
 127 extern const int32_t Inverse_Table_6_9[8][4];
 128
 129 /*
 130 NOTES
 131 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
 132
 133 TODO
 134 more intelligent missalignment avoidance for the horizontal scaler
 135 write special vertical cubic upscale version
 136 Optimize C code (yv12 / minmax)
 137 add support for packed pixel yuv input & output
 138 add support for Y8 output
 139 optimize bgr24 & bgr32
 140 add BGR4 output support
 141 write special BGR->BGR scaler
 142 */
 143
 144 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
 145 #define MIN(a,b) ((a) > (b) ? (b) : (a))
 146 #define MAX(a,b) ((a) < (b) ? (b) : (a))
 147
 148 #ifdef ARCH_X86
 149 static uint64_t __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
 150 static uint64_t __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
 151 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
 152 static uint64_t __attribute__((aligned(8))) w02=       0x0002000200020002LL;
 153 static uint64_t __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
 154 static uint64_t __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
 155 static uint64_t __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
 156 static uint64_t __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
 157
 158 static volatile uint64_t __attribute__((aligned(8))) b5Dither;
 159 static volatile uint64_t __attribute__((aligned(8))) g5Dither;
 160 static volatile uint64_t __attribute__((aligned(8))) g6Dither;
 161 static volatile uint64_t __attribute__((aligned(8))) r5Dither;
 162
 163 static uint64_t __attribute__((aligned(8))) dither4[2]={
 164         0x0103010301030103LL,
 165         0x0200020002000200LL,};
 166
 167 static uint64_t __attribute__((aligned(8))) dither8[2]={
 168         0x0602060206020602LL,
 169         0x0004000400040004LL,};
 170
 171 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
 172 static uint64_t __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
 173 static uint64_t __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
 174 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
 175 static uint64_t __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
 176 static uint64_t __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
 177
 178 static uint64_t __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
 179 static uint64_t __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
 180 static uint64_t __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
 181
 182 #ifdef FAST_BGR2YV12
 183 static const uint64_t bgr2YCoeff  __attribute__((aligned(8))) = 0x000000210041000DULL;
 184 static const uint64_t bgr2UCoeff  __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
 185 static const uint64_t bgr2VCoeff  __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
 186 #else
 187 static const uint64_t bgr2YCoeff  __attribute__((aligned(8))) = 0x000020E540830C8BULL;
 188 static const uint64_t bgr2UCoeff  __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
 189 static const uint64_t bgr2VCoeff  __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
 190 #endif
 191 static const uint64_t bgr2YOffset __attribute__((aligned(8))) = 0x1010101010101010ULL;
 192 static const uint64_t bgr2UVOffset __attribute__((aligned(8)))= 0x8080808080808080ULL;
 193 static const uint64_t w1111       __attribute__((aligned(8))) = 0x0001000100010001ULL;
 194 #endif
 195
 196 // clipping helper table for C implementations:
 197 static unsigned char clip_table[768];
 198
 199 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
 200
 201 extern const uint8_t dither_2x2_4[2][8];
 202 extern const uint8_t dither_2x2_8[2][8];
 203 extern const uint8_t dither_8x8_32[8][8];
 204 extern const uint8_t dither_8x8_73[8][8];
 205 extern const uint8_t dither_8x8_220[8][8];
 206
 207 #ifdef ARCH_X86
 208 void in_asm_used_var_warning_killer()
 209 {
 210  volatile int i= bF8+bFC+w10+
 211  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
 212  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
 213  if(i) i=0;
 214 }
 215 #endif
 216
 217 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
 218                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
 219                                     uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
 220 {
 221         //FIXME Optimize (just quickly writen not opti..)
 222         int i;
 223         for(i=0; i<dstW; i++)
 224         {
 225                 int val=0;
 226                 int j;
 227                 for(j=0; j<lumFilterSize; j++)
 228                         val += lumSrc[j][i] * lumFilter[j];
 229
 230                 dest[i]= MIN(MAX(val>>19, 0), 255);
 231         }
 232
 233         if(uDest != NULL)
 234                 for(i=0; i<chrDstW; i++)
 235                 {
 236                         int u=0;
 237                         int v=0;
 238                         int j;
 239                         for(j=0; j<chrFilterSize; j++)
 240                         {
 241                                 u += chrSrc[j][i] * chrFilter[j];
 242                                 v += chrSrc[j][i + 2048] * chrFilter[j];
 243                         }
 244
 245                         uDest[i]= MIN(MAX(u>>19, 0), 255);
 246                         vDest[i]= MIN(MAX(v>>19, 0), 255);
 247                 }
 248 }
 249
 250
 251 #define YSCALE_YUV_2_PACKEDX_C(type) \
 252                 for(i=0; i<(dstW>>1); i++){\
 253                         int j;\
 254                         int Y1=0;\
 255                         int Y2=0;\
 256                         int U=0;\
 257                         int V=0;\
 258                         type *r, *b, *g;\
 259                         const int i2= 2*i;\
 260                         \
 261                         for(j=0; j<lumFilterSize; j++)\
 262                         {\
 263                                 Y1 += lumSrc[j][i2] * lumFilter[j];\
 264                                 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
 265                         }\
 266                         for(j=0; j<chrFilterSize; j++)\
 267                         {\
 268                                 U += chrSrc[j][i] * chrFilter[j];\
 269                                 V += chrSrc[j][i+2048] * chrFilter[j];\
 270                         }\
 271                         Y1>>=19;\
 272                         Y2>>=19;\
 273                         U >>=19;\
 274                         V >>=19;\
 275                         if((Y1|Y2|U|V)&256)\
 276                         {\
 277                                 if(Y1>255)   Y1=255;\
 278                                 else if(Y1<0)Y1=0;\
 279                                 if(Y2>255)   Y2=255;\
 280                                 else if(Y2<0)Y2=0;\
 281                                 if(U>255)    U=255;\
 282                                 else if(U<0) U=0;\
 283                                 if(V>255)    V=255;\
 284                                 else if(V<0) V=0;\
 285                         }
 286
 287 #define YSCALE_YUV_2_RGBX_C(type) \
 288                         YSCALE_YUV_2_PACKEDX_C(type)\
 289                         r = c->table_rV[V];\
 290                         g = c->table_gU[U] + c->table_gV[V];\
 291                         b = c->table_bU[U];\
 292
 293 #define YSCALE_YUV_2_PACKED2_C \
 294                 for(i=0; i<(dstW>>1); i++){\
 295                         const int i2= 2*i;\
 296                         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
 297                         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
 298                         int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
 299                         int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
 300
 301 #define YSCALE_YUV_2_RGB2_C(type) \
 302                         YSCALE_YUV_2_PACKED2_C\
 303                         type *r, *b, *g;\
 304                         r = c->table_rV[V];\
 305                         g = c->table_gU[U] + c->table_gV[V];\
 306                         b = c->table_bU[U];\
 307
 308 #define YSCALE_YUV_2_PACKED1_C \
 309                 for(i=0; i<(dstW>>1); i++){\
 310                         const int i2= 2*i;\
 311                         int Y1= buf0[i2  ]>>7;\
 312                         int Y2= buf0[i2+1]>>7;\
 313                         int U= (uvbuf1[i     ])>>7;\
 314                         int V= (uvbuf1[i+2048])>>7;\
 315
 316 #define YSCALE_YUV_2_RGB1_C(type) \
 317                         YSCALE_YUV_2_PACKED1_C\
 318                         type *r, *b, *g;\
 319                         r = c->table_rV[V];\
 320                         g = c->table_gU[U] + c->table_gV[V];\
 321                         b = c->table_bU[U];\
 322
 323 #define YSCALE_YUV_2_PACKED1B_C \
 324                 for(i=0; i<(dstW>>1); i++){\
 325                         const int i2= 2*i;\
 326                         int Y1= buf0[i2  ]>>7;\
 327                         int Y2= buf0[i2+1]>>7;\
 328                         int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
 329                         int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
 330
 331 #define YSCALE_YUV_2_RGB1B_C(type) \
 332                         YSCALE_YUV_2_PACKED1B_C\
 333                         type *r, *b, *g;\
 334                         r = c->table_rV[V];\
 335                         g = c->table_gU[U] + c->table_gV[V];\
 336                         b = c->table_bU[U];\
 337
 338 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
 339         switch(c->dstFormat)\
 340         {\
 341         case IMGFMT_BGR32:\
 342         case IMGFMT_RGB32:\
 343                 func(uint32_t)\
 344                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
 345                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
 346                 }               \
 347                 break;\
 348         case IMGFMT_RGB24:\
 349                 func(uint8_t)\
 350                         ((uint8_t*)dest)[0]= r[Y1];\
 351                         ((uint8_t*)dest)[1]= g[Y1];\
 352                         ((uint8_t*)dest)[2]= b[Y1];\
 353                         ((uint8_t*)dest)[3]= r[Y2];\
 354                         ((uint8_t*)dest)[4]= g[Y2];\
 355                         ((uint8_t*)dest)[5]= b[Y2];\
 356                         ((uint8_t*)dest)+=6;\
 357                 }\
 358                 break;\
 359         case IMGFMT_BGR24:\
 360                 func(uint8_t)\
 361                         ((uint8_t*)dest)[0]= b[Y1];\
 362                         ((uint8_t*)dest)[1]= g[Y1];\
 363                         ((uint8_t*)dest)[2]= r[Y1];\
 364                         ((uint8_t*)dest)[3]= b[Y2];\
 365                         ((uint8_t*)dest)[4]= g[Y2];\
 366                         ((uint8_t*)dest)[5]= r[Y2];\
 367                         ((uint8_t*)dest)+=6;\
 368                 }\
 369                 break;\
 370         case IMGFMT_RGB16:\
 371         case IMGFMT_BGR16:\
 372                 {\
 373                         const int dr1= dither_2x2_8[y&1    ][0];\
 374                         const int dg1= dither_2x2_4[y&1    ][0];\
 375                         const int db1= dither_2x2_8[(y&1)^1][0];\
 376                         const int dr2= dither_2x2_8[y&1    ][1];\
 377                         const int dg2= dither_2x2_4[y&1    ][1];\
 378                         const int db2= dither_2x2_8[(y&1)^1][1];\
 379                         func(uint16_t)\
 380                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
 381                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
 382                         }\
 383                 }\
 384                 break;\
 385         case IMGFMT_RGB15:\
 386         case IMGFMT_BGR15:\
 387                 {\
 388                         const int dr1= dither_2x2_8[y&1    ][0];\
 389                         const int dg1= dither_2x2_8[y&1    ][1];\
 390                         const int db1= dither_2x2_8[(y&1)^1][0];\
 391                         const int dr2= dither_2x2_8[y&1    ][1];\
 392                         const int dg2= dither_2x2_8[y&1    ][0];\
 393                         const int db2= dither_2x2_8[(y&1)^1][1];\
 394                         func(uint16_t)\
 395                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
 396                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
 397                         }\
 398                 }\
 399                 break;\
 400         case IMGFMT_RGB8:\
 401         case IMGFMT_BGR8:\
 402                 {\
 403                         const uint8_t * const d64= dither_8x8_73[y&7];\
 404                         const uint8_t * const d32= dither_8x8_32[y&7];\
 405                         func(uint8_t)\
 406                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
 407                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
 408                         }\
 409                 }\
 410                 break;\
 411         case IMGFMT_RGB4:\
 412         case IMGFMT_BGR4:\
 413                 {\
 414                         const uint8_t * const d64= dither_8x8_73 [y&7];\
 415                         const uint8_t * const d128=dither_8x8_220[y&7];\
 416                         func(uint8_t)\
 417                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
 418                                                  + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
 419                         }\
 420                 }\
 421                 break;\
 422         case IMGFMT_RG4B:\
 423         case IMGFMT_BG4B:\
 424                 {\
 425                         const uint8_t * const d64= dither_8x8_73 [y&7];\
 426                         const uint8_t * const d128=dither_8x8_220[y&7];\
 427                         func(uint8_t)\
 428                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
 429                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
 430                         }\
 431                 }\
 432                 break;\
 433         case IMGFMT_RGB1:\
 434         case IMGFMT_BGR1:\
 435                 {\
 436                         const uint8_t * const d128=dither_8x8_220[y&7];\
 437                         uint8_t *g= c->table_gU[128] + c->table_gV[128];\
 438                         for(i=0; i<dstW-7; i+=8){\
 439                                 int acc;\
 440                                 acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
 441                                 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
 442                                 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
 443                                 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
 444                                 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
 445                                 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
 446                                 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
 447                                 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
 448                                 ((uint8_t*)dest)[0]= acc;\
 449                                 ((uint8_t*)dest)++;\
 450                         }\
 451 \
 452 /*\
 453 ((uint8_t*)dest)-= dstW>>4;\
 454 {\
 455                         int acc=0;\
 456                         int left=0;\
 457                         static int top[1024];\
 458                         static int last_new[1024][1024];\
 459                         static int last_in3[1024][1024];\
 460                         static int drift[1024][1024];\
 461                         int topLeft=0;\
 462                         int shift=0;\
 463                         int count=0;\
 464                         const uint8_t * const d128=dither_8x8_220[y&7];\
 465                         int error_new=0;\
 466                         int error_in3=0;\
 467                         int f=0;\
 468                         \
 469                         for(i=dstW>>1; i<dstW; i++){\
 470                                 int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
 471                                 int in2 = (76309 * (in - 16) + 32768) >> 16;\
 472                                 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
 473                                 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
 474                                         + (last_new[y][i] - in3)*f/256;\
 475                                 int new= old> 128 ? 255 : 0;\
 476 \
 477                                 error_new+= ABS(last_new[y][i] - new);\
 478                                 error_in3+= ABS(last_in3[y][i] - in3);\
 479                                 f= error_new - error_in3*4;\
 480                                 if(f<0) f=0;\
 481                                 if(f>256) f=256;\
 482 \
 483                                 topLeft= top[i];\
 484                                 left= top[i]= old - new;\
 485                                 last_new[y][i]= new;\
 486                                 last_in3[y][i]= in3;\
 487 \
 488                                 acc+= acc + (new&1);\
 489                                 if((i&7)==6){\
 490                                         ((uint8_t*)dest)[0]= acc;\
 491                                         ((uint8_t*)dest)++;\
 492                                 }\
 493                         }\
 494 }\
 495 */\
 496                 }\
 497                 break;\
 498         case IMGFMT_YUY2:\
 499                 func2\
 500                         ((uint8_t*)dest)[2*i2+0]= Y1;\
 501                         ((uint8_t*)dest)[2*i2+1]= U;\
 502                         ((uint8_t*)dest)[2*i2+2]= Y2;\
 503                         ((uint8_t*)dest)[2*i2+3]= V;\
 504                 }               \
 505                 break;\
 506         case IMGFMT_UYVY:\
 507                 func2\
 508                         ((uint8_t*)dest)[2*i2+0]= U;\
 509                         ((uint8_t*)dest)[2*i2+1]= Y1;\
 510                         ((uint8_t*)dest)[2*i2+2]= V;\
 511                         ((uint8_t*)dest)[2*i2+3]= Y2;\
 512                 }               \
 513                 break;\
 514         }\
 515
 516
 517 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
 518                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
 519                                     uint8_t *dest, int dstW, int y)
 520 {
 521         int i;
 522         switch(c->dstFormat)
 523         {
 524         case IMGFMT_RGB32:
 525         case IMGFMT_BGR32:
 526                 YSCALE_YUV_2_RGBX_C(uint32_t)
 527                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
 528                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
 529                 }
 530                 break;
 531         case IMGFMT_RGB24:
 532                 YSCALE_YUV_2_RGBX_C(uint8_t)
 533                         ((uint8_t*)dest)[0]= r[Y1];
 534                         ((uint8_t*)dest)[1]= g[Y1];
 535                         ((uint8_t*)dest)[2]= b[Y1];
 536                         ((uint8_t*)dest)[3]= r[Y2];
 537                         ((uint8_t*)dest)[4]= g[Y2];
 538                         ((uint8_t*)dest)[5]= b[Y2];
 539                         ((uint8_t*)dest)+=6;
 540                 }
 541                 break;
 542         case IMGFMT_BGR24:
 543                 YSCALE_YUV_2_RGBX_C(uint8_t)
 544                         ((uint8_t*)dest)[0]= b[Y1];
 545                         ((uint8_t*)dest)[1]= g[Y1];
 546                         ((uint8_t*)dest)[2]= r[Y1];
 547                         ((uint8_t*)dest)[3]= b[Y2];
 548                         ((uint8_t*)dest)[4]= g[Y2];
 549                         ((uint8_t*)dest)[5]= r[Y2];
 550                         ((uint8_t*)dest)+=6;
 551                 }
 552                 break;
 553         case IMGFMT_RGB16:
 554         case IMGFMT_BGR16:
 555                 {
 556                         const int dr1= dither_2x2_8[y&1    ][0];
 557                         const int dg1= dither_2x2_4[y&1    ][0];
 558                         const int db1= dither_2x2_8[(y&1)^1][0];
 559                         const int dr2= dither_2x2_8[y&1    ][1];
 560                         const int dg2= dither_2x2_4[y&1    ][1];
 561                         const int db2= dither_2x2_8[(y&1)^1][1];
 562                         YSCALE_YUV_2_RGBX_C(uint16_t)
 563                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
 564                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
 565                         }
 566                 }
 567                 break;
 568         case IMGFMT_RGB15:
 569         case IMGFMT_BGR15:
 570                 {
 571                         const int dr1= dither_2x2_8[y&1    ][0];
 572                         const int dg1= dither_2x2_8[y&1    ][1];
 573                         const int db1= dither_2x2_8[(y&1)^1][0];
 574                         const int dr2= dither_2x2_8[y&1    ][1];
 575                         const int dg2= dither_2x2_8[y&1    ][0];
 576                         const int db2= dither_2x2_8[(y&1)^1][1];
 577                         YSCALE_YUV_2_RGBX_C(uint16_t)
 578                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
 579                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
 580                         }
 581                 }
 582                 break;
 583         case IMGFMT_RGB8:
 584         case IMGFMT_BGR8:
 585                 {
 586                         const uint8_t * const d64= dither_8x8_73[y&7];
 587                         const uint8_t * const d32= dither_8x8_32[y&7];
 588                         YSCALE_YUV_2_RGBX_C(uint8_t)
 589                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
 590                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
 591                         }
 592                 }
 593                 break;
 594         case IMGFMT_RGB4:
 595         case IMGFMT_BGR4:
 596                 {
 597                         const uint8_t * const d64= dither_8x8_73 [y&7];
 598                         const uint8_t * const d128=dither_8x8_220[y&7];
 599                         YSCALE_YUV_2_RGBX_C(uint8_t)
 600                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
 601                                                   +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
 602                         }
 603                 }
 604                 break;
 605         case IMGFMT_RG4B:
 606         case IMGFMT_BG4B:
 607                 {
 608                         const uint8_t * const d64= dither_8x8_73 [y&7];
 609                         const uint8_t * const d128=dither_8x8_220[y&7];
 610                         YSCALE_YUV_2_RGBX_C(uint8_t)
 611                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
 612                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
 613                         }
 614                 }
 615                 break;
 616         case IMGFMT_RGB1:
 617         case IMGFMT_BGR1:
 618                 {
 619                         const uint8_t * const d128=dither_8x8_220[y&7];
 620                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
 621                         int acc=0;
 622                         for(i=0; i<dstW-1; i+=2){
 623                                 int j;
 624                                 int Y1=0;
 625                                 int Y2=0;
 626
 627                                 for(j=0; j<lumFilterSize; j++)
 628                                 {
 629                                         Y1 += lumSrc[j][i] * lumFilter[j];
 630                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
 631                                 }
 632                                 Y1>>=19;
 633                                 Y2>>=19;
 634                                 if((Y1|Y2)&256)
 635                                 {
 636                                         if(Y1>255)   Y1=255;
 637                                         else if(Y1<0)Y1=0;
 638                                         if(Y2>255)   Y2=255;
 639                                         else if(Y2<0)Y2=0;
 640                                 }
 641                                 acc+= acc + g[Y1+d128[(i+0)&7]];
 642                                 acc+= acc + g[Y2+d128[(i+1)&7]];
 643                                 if((i&7)==6){
 644                                         ((uint8_t*)dest)[0]= acc;
 645                                         ((uint8_t*)dest)++;
 646                                 }
 647                         }
 648                 }
 649                 break;
 650         case IMGFMT_YUY2:
 651                 YSCALE_YUV_2_PACKEDX_C(void)
 652                         ((uint8_t*)dest)[2*i2+0]= Y1;
 653                         ((uint8_t*)dest)[2*i2+1]= U;
 654                         ((uint8_t*)dest)[2*i2+2]= Y2;
 655                         ((uint8_t*)dest)[2*i2+3]= V;
 656                 }
 657                 break;
 658         case IMGFMT_UYVY:
 659                 YSCALE_YUV_2_PACKEDX_C(void)
 660                         ((uint8_t*)dest)[2*i2+0]= U;
 661                         ((uint8_t*)dest)[2*i2+1]= Y1;
 662                         ((uint8_t*)dest)[2*i2+2]= V;
 663                         ((uint8_t*)dest)[2*i2+3]= Y2;
 664                 }
 665                 break;
 666         }
 667 }
 668
 669
 670 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
 671 //Plain C versions
 672 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
 673 #define COMPILE_C
 674 #endif
 675
 676 #ifdef ARCH_X86
 677
 678 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
 679 #define COMPILE_MMX
 680 #endif
 681
 682 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
 683 #define COMPILE_MMX2
 684 #endif
 685
 686 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
 687 #define COMPILE_3DNOW
 688 #endif
 689 #endif //ARCH_X86
 690
 691 #undef HAVE_MMX
 692 #undef HAVE_MMX2
 693 #undef HAVE_3DNOW
 694
 695 #ifdef COMPILE_C
 696 #undef HAVE_MMX
 697 #undef HAVE_MMX2
 698 #undef HAVE_3DNOW
 699 #define RENAME(a) a ## _C
 700 #include "swscale_template.c"
 701 #endif
 702
 703 #ifdef ARCH_X86
 704
 705 //X86 versions
 706 /*
 707 #undef RENAME
 708 #undef HAVE_MMX
 709 #undef HAVE_MMX2
 710 #undef HAVE_3DNOW
 711 #define ARCH_X86
 712 #define RENAME(a) a ## _X86
 713 #include "swscale_template.c"
 714 */
 715 //MMX versions
 716 #ifdef COMPILE_MMX
 717 #undef RENAME
 718 #define HAVE_MMX
 719 #undef HAVE_MMX2
 720 #undef HAVE_3DNOW
 721 #define RENAME(a) a ## _MMX
 722 #include "swscale_template.c"
 723 #endif
 724
 725 //MMX2 versions
 726 #ifdef COMPILE_MMX2
 727 #undef RENAME
 728 #define HAVE_MMX
 729 #define HAVE_MMX2
 730 #undef HAVE_3DNOW
 731 #define RENAME(a) a ## _MMX2
 732 #include "swscale_template.c"
 733 #endif
 734
 735 //3DNOW versions
 736 #ifdef COMPILE_3DNOW
 737 #undef RENAME
 738 #define HAVE_MMX
 739 #undef HAVE_MMX2
 740 #define HAVE_3DNOW
 741 #define RENAME(a) a ## _3DNow
 742 #include "swscale_template.c"
 743 #endif
 744
 745 #endif //ARCH_X86
 746
 747 // minor note: the HAVE_xyz is messed up after that line so don't use it
 748
 749 static double getSplineCoeff(double a, double b, double c, double d, double dist)
 750 {
 751 //      printf("%f %f %f %f %f\n", a,b,c,d,dist);
 752         if(dist<=1.0)   return ((d*dist + c)*dist + b)*dist +a;
 753         else            return getSplineCoeff(  0.0,
 754                                                  b+ 2.0*c + 3.0*d,
 755                                                         c + 3.0*d,
 756                                                 -b- 3.0*c - 6.0*d,
 757                                                 dist-1.0);
 758 }
 759
 760 static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
 761                               int srcW, int dstW, int filterAlign, int one, int flags,
 762                               SwsVector *srcFilter, SwsVector *dstFilter)
 763 {
 764         int i;
 765         int filterSize;
 766         int filter2Size;
 767         int minFilterSize;
 768         double *filter=NULL;
 769         double *filter2=NULL;
 770 #ifdef ARCH_X86
 771         if(flags & SWS_CPU_CAPS_MMX)
 772                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
 773 #endif
 774
 775         // Note the +1 is for the MMXscaler which reads over the end
 776         *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
 777
 778         if(ABS(xInc - 0x10000) <10) // unscaled
 779         {
 780                 int i;
 781                 filterSize= 1;
 782                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
 783                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
 784
 785                 for(i=0; i<dstW; i++)
 786                 {
 787                         filter[i*filterSize]=1;
 788                         (*filterPos)[i]=i;
 789                 }
 790
 791         }
 792         else if(flags&SWS_POINT) // lame looking point sampling mode
 793         {
 794                 int i;
 795                 int xDstInSrc;
 796                 filterSize= 1;
 797                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
 798
 799                 xDstInSrc= xInc/2 - 0x8000;
 800                 for(i=0; i<dstW; i++)
 801                 {
 802                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
 803
 804                         (*filterPos)[i]= xx;
 805                         filter[i]= 1.0;
 806                         xDstInSrc+= xInc;
 807                 }
 808         }
 809         else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
 810         {
 811                 int i;
 812                 int xDstInSrc;
 813                 if     (flags&SWS_BICUBIC) filterSize= 4;
 814                 else if(flags&SWS_X      ) filterSize= 4;
 815                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA
 816                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
 817
 818                 xDstInSrc= xInc/2 - 0x8000;
 819                 for(i=0; i<dstW; i++)
 820                 {
 821                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
 822                         int j;
 823
 824                         (*filterPos)[i]= xx;
 825                                 //Bilinear upscale / linear interpolate / Area averaging
 826                                 for(j=0; j<filterSize; j++)
 827                                 {
 828                                         double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
 829                                         double coeff= 1.0 - d;
 830                                         if(coeff<0) coeff=0;
 831                                         filter[i*filterSize + j]= coeff;
 832                                         xx++;
 833                                 }
 834                         xDstInSrc+= xInc;
 835                 }
 836         }
 837         else
 838         {
 839                 double xDstInSrc;
 840                 double sizeFactor, filterSizeInSrc;
 841                 const double xInc1= (double)xInc / (double)(1<<16);
 842                 int param= (flags&SWS_PARAM_MASK)>>SWS_PARAM_SHIFT;
 843
 844                 if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
 845                 else if(flags&SWS_X)            sizeFactor= 8.0;
 846                 else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
 847                 else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
 848                 else if(flags&SWS_LANCZOS)      sizeFactor= param ? 2.0*param : 6.0;
 849                 else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
 850                 else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
 851                 else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
 852                 else {
 853                         sizeFactor= 0.0; //GCC warning killer
 854                         ASSERT(0)
 855                 }
 856
 857                 if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
 858                 else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
 859
 860                 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
 861                 if(filterSize > srcW-2) filterSize=srcW-2;
 862
 863                 filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
 864
 865                 xDstInSrc= xInc1 / 2.0 - 0.5;
 866                 for(i=0; i<dstW; i++)
 867                 {
 868                         int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
 869                         int j;
 870                         (*filterPos)[i]= xx;
 871                         for(j=0; j<filterSize; j++)
 872                         {
 873                                 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
 874                                 double coeff;
 875                                 if(flags & SWS_BICUBIC)
 876                                 {
 877                                         double A= param ? -param*0.01 : -0.60;
 878
 879                                         // Equation is from VirtualDub
 880                                         if(d<1.0)
 881                                                 coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
 882                                         else if(d<2.0)
 883                                                 coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
 884                                         else
 885                                                 coeff=0.0;
 886                                 }
 887 /*                              else if(flags & SWS_X)
 888                                 {
 889                                         double p= param ? param*0.01 : 0.3;
 890                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
 891                                         coeff*= pow(2.0, - p*d*d);
 892                                 }*/
 893                                 else if(flags & SWS_X)
 894                                 {
 895                                         double A= param ? param*0.1 : 1.0;
 896
 897                                         if(d<1.0)
 898                                                 coeff = cos(d*PI);
 899                                         else
 900                                                 coeff=-1.0;
 901                                         if(coeff<0.0)   coeff= -pow(-coeff, A);
 902                                         else            coeff=  pow( coeff, A);
 903                                         coeff= coeff*0.5 + 0.5;
 904                                 }
 905                                 else if(flags & SWS_AREA)
 906                                 {
 907                                         double srcPixelSize= 1.0/xInc1;
 908                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
 909                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
 910                                         else coeff=0.0;
 911                                 }
 912                                 else if(flags & SWS_GAUSS)
 913                                 {
 914                                         double p= param ? param*0.1 : 3.0;
 915                                         coeff = pow(2.0, - p*d*d);
 916                                 }
 917                                 else if(flags & SWS_SINC)
 918                                 {
 919                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
 920                                 }
 921                                 else if(flags & SWS_LANCZOS)
 922                                 {
 923                                         double p= param ? param : 3.0;
 924                                         coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
 925                                         if(d>p) coeff=0;
 926                                 }
 927                                 else if(flags & SWS_BILINEAR)
 928                                 {
 929                                         coeff= 1.0 - d;
 930                                         if(coeff<0) coeff=0;
 931                                 }
 932                                 else if(flags & SWS_SPLINE)
 933                                 {
 934                                         double p=-2.196152422706632;
 935                                         coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
 936                                 }
 937                                 else {
 938                                         coeff= 0.0; //GCC warning killer
 939                                         ASSERT(0)
 940                                 }
 941
 942                                 filter[i*filterSize + j]= coeff;
 943                                 xx++;
 944                         }
 945                         xDstInSrc+= xInc1;
 946                 }
 947         }
 948
 949         /* apply src & dst Filter to filter -> filter2
 950            free(filter);
 951         */
 952         ASSERT(filterSize>0)
 953         filter2Size= filterSize;
 954         if(srcFilter) filter2Size+= srcFilter->length - 1;
 955         if(dstFilter) filter2Size+= dstFilter->length - 1;
 956         ASSERT(filter2Size>0)
 957         filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
 958
 959         for(i=0; i<dstW; i++)
 960         {
 961                 int j;
 962                 SwsVector scaleFilter;
 963                 SwsVector *outVec;
 964
 965                 scaleFilter.coeff= filter + i*filterSize;
 966                 scaleFilter.length= filterSize;
 967
 968                 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
 969                 else          outVec= &scaleFilter;
 970
 971                 ASSERT(outVec->length == filter2Size)
 972                 //FIXME dstFilter
 973
 974                 for(j=0; j<outVec->length; j++)
 975                 {
 976                         filter2[i*filter2Size + j]= outVec->coeff[j];
 977                 }
 978
 979                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
 980
 981                 if(outVec != &scaleFilter) sws_freeVec(outVec);
 982         }
 983         free(filter); filter=NULL;
 984
 985         /* try to reduce the filter-size (step1 find size and shift left) */
 986         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
 987         minFilterSize= 0;
 988         for(i=dstW-1; i>=0; i--)
 989         {
 990                 int min= filter2Size;
 991                 int j;
 992                 double cutOff=0.0;
 993
 994                 /* get rid off near zero elements on the left by shifting left */
 995                 for(j=0; j<filter2Size; j++)
 996                 {
 997                         int k;
 998                         cutOff += ABS(filter2[i*filter2Size]);
 999
1000                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1001
1002                         /* preserve Monotonicity because the core can't handle the filter otherwise */
1003                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1004
1005                         // Move filter coeffs left
1006                         for(k=1; k<filter2Size; k++)
1007                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1008                         filter2[i*filter2Size + k - 1]= 0.0;
1009                         (*filterPos)[i]++;
1010                 }
1011
1012                 cutOff=0.0;
1013                 /* count near zeros on the right */
1014                 for(j=filter2Size-1; j>0; j--)
1015                 {
1016                         cutOff += ABS(filter2[i*filter2Size + j]);
1017
1018                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1019                         min--;
1020                 }
1021
1022                 if(min>minFilterSize) minFilterSize= min;
1023         }
1024
1025         ASSERT(minFilterSize > 0)
1026         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1027         ASSERT(filterSize > 0)
1028         filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1029         *outFilterSize= filterSize;
1030
1031         if(flags&SWS_PRINT_INFO)
1032                 MSG_INFO("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1033         /* try to reduce the filter-size (step2 reduce it) */
1034         for(i=0; i<dstW; i++)
1035         {
1036                 int j;
1037
1038                 for(j=0; j<filterSize; j++)
1039                 {
1040                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1041                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1042                 }
1043         }
1044         free(filter2); filter2=NULL;
1045
1046
1047         //FIXME try to align filterpos if possible
1048
1049         //fix borders
1050         for(i=0; i<dstW; i++)
1051         {
1052                 int j;
1053                 if((*filterPos)[i] < 0)
1054                 {
1055                         // Move filter coeffs left to compensate for filterPos
1056                         for(j=1; j<filterSize; j++)
1057                         {
1058                                 int left= MAX(j + (*filterPos)[i], 0);
1059                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1060                                 filter[i*filterSize + j]=0;
1061                         }
1062                         (*filterPos)[i]= 0;
1063                 }
1064
1065                 if((*filterPos)[i] + filterSize > srcW)
1066                 {
1067                         int shift= (*filterPos)[i] + filterSize - srcW;
1068                         // Move filter coeffs right to compensate for filterPos
1069                         for(j=filterSize-2; j>=0; j--)
1070                         {
1071                                 int right= MIN(j + shift, filterSize-1);
1072                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1073                                 filter[i*filterSize +j]=0;
1074                         }
1075                         (*filterPos)[i]= srcW - filterSize;
1076                 }
1077         }
1078
1079         // Note the +1 is for the MMXscaler which reads over the end
1080         *outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));
1081         memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1082
1083         /* Normalize & Store in outFilter */
1084         for(i=0; i<dstW; i++)
1085         {
1086                 int j;
1087                 double sum=0;
1088                 double scale= one;
1089                 for(j=0; j<filterSize; j++)
1090                 {
1091                         sum+= filter[i*filterSize + j];
1092                 }
1093                 scale/= sum;
1094                 for(j=0; j<*outFilterSize; j++)
1095                 {
1096                         (*outFilter)[i*(*outFilterSize) + j]= (int)(filter[i*filterSize + j]*scale);
1097                 }
1098         }
1099
1100         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1101         for(i=0; i<*outFilterSize; i++)
1102         {
1103                 int j= dstW*(*outFilterSize);
1104                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1105         }
1106
1107         free(filter);
1108 }
1109
1110 #ifdef ARCH_X86
1111 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1112 {
1113         uint8_t *fragmentA;
1114         int imm8OfPShufW1A;
1115         int imm8OfPShufW2A;
1116         int fragmentLengthA;
1117         uint8_t *fragmentB;
1118         int imm8OfPShufW1B;
1119         int imm8OfPShufW2B;
1120         int fragmentLengthB;
1121         int fragmentPos;
1122
1123         int xpos, i;
1124
1125         // create an optimized horizontal scaling routine
1126
1127         //code fragment
1128
1129         asm volatile(
1130                 "jmp 9f                         \n\t"
1131         // Begin
1132                 "0:                             \n\t"
1133                 "movq (%%edx, %%eax), %%mm3     \n\t"
1134                 "movd (%%ecx, %%esi), %%mm0     \n\t"
1135                 "movd 1(%%ecx, %%esi), %%mm1    \n\t"
1136                 "punpcklbw %%mm7, %%mm1         \n\t"
1137                 "punpcklbw %%mm7, %%mm0         \n\t"
1138                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1139                 "1:                             \n\t"
1140                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1141                 "2:                             \n\t"
1142                 "psubw %%mm1, %%mm0             \n\t"
1143                 "movl 8(%%ebx, %%eax), %%esi    \n\t"
1144                 "pmullw %%mm3, %%mm0            \n\t"
1145                 "psllw $7, %%mm1                \n\t"
1146                 "paddw %%mm1, %%mm0             \n\t"
1147
1148                 "movq %%mm0, (%%edi, %%eax)     \n\t"
1149
1150                 "addl $8, %%eax                 \n\t"
1151         // End
1152                 "9:                             \n\t"
1153 //              "int $3\n\t"
1154                 "leal 0b, %0                    \n\t"
1155                 "leal 1b, %1                    \n\t"
1156                 "leal 2b, %2                    \n\t"
1157                 "decl %1                        \n\t"
1158                 "decl %2                        \n\t"
1159                 "subl %0, %1                    \n\t"
1160                 "subl %0, %2                    \n\t"
1161                 "leal 9b, %3                    \n\t"
1162                 "subl %0, %3                    \n\t"
1163
1164
1165                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1166                 "=r" (fragmentLengthA)
1167         );
1168
1169         asm volatile(
1170                 "jmp 9f                         \n\t"
1171         // Begin
1172                 "0:                             \n\t"
1173                 "movq (%%edx, %%eax), %%mm3     \n\t"
1174                 "movd (%%ecx, %%esi), %%mm0     \n\t"
1175                 "punpcklbw %%mm7, %%mm0         \n\t"
1176                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1177                 "1:                             \n\t"
1178                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1179                 "2:                             \n\t"
1180                 "psubw %%mm1, %%mm0             \n\t"
1181                 "movl 8(%%ebx, %%eax), %%esi    \n\t"
1182                 "pmullw %%mm3, %%mm0            \n\t"
1183                 "psllw $7, %%mm1                \n\t"
1184                 "paddw %%mm1, %%mm0             \n\t"
1185
1186                 "movq %%mm0, (%%edi, %%eax)     \n\t"
1187
1188                 "addl $8, %%eax                 \n\t"
1189         // End
1190                 "9:                             \n\t"
1191 //              "int $3\n\t"
1192                 "leal 0b, %0                    \n\t"
1193                 "leal 1b, %1                    \n\t"
1194                 "leal 2b, %2                    \n\t"
1195                 "decl %1                        \n\t"
1196                 "decl %2                        \n\t"
1197                 "subl %0, %1                    \n\t"
1198                 "subl %0, %2                    \n\t"
1199                 "leal 9b, %3                    \n\t"
1200                 "subl %0, %3                    \n\t"
1201
1202
1203                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1204                 "=r" (fragmentLengthB)
1205         );
1206
1207         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1208         fragmentPos=0;
1209
1210         for(i=0; i<dstW/numSplits; i++)
1211         {
1212                 int xx=xpos>>16;
1213
1214                 if((i&3) == 0)
1215                 {
1216                         int a=0;
1217                         int b=((xpos+xInc)>>16) - xx;
1218                         int c=((xpos+xInc*2)>>16) - xx;
1219                         int d=((xpos+xInc*3)>>16) - xx;
1220
1221                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1222                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1223                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1224                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1225                         filterPos[i/2]= xx;
1226
1227                         if(d+1<4)
1228                         {
1229                                 int maxShift= 3-(d+1);
1230                                 int shift=0;
1231
1232                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1233
1234                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1235                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1236                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1237                                         a | (b<<2) | (c<<4) | (d<<6);
1238
1239                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1240                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1241
1242                                 if(shift && i>=shift)
1243                                 {
1244                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1245                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1246                                         filterPos[i/2]-=shift;
1247                                 }
1248
1249                                 fragmentPos+= fragmentLengthB;
1250                         }
1251                         else
1252                         {
1253                                 int maxShift= 3-d;
1254                                 int shift=0;
1255
1256                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1257
1258                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1259                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1260                                         a | (b<<2) | (c<<4) | (d<<6);
1261
1262                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1263                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1264
1265                                 if(shift && i>=shift)
1266                                 {
1267                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1268                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1269                                         filterPos[i/2]-=shift;
1270                                 }
1271
1272                                 fragmentPos+= fragmentLengthA;
1273                         }
1274
1275                         funnyCode[fragmentPos]= RET;
1276                 }
1277                 xpos+=xInc;
1278         }
1279         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1280 }
1281 #endif // ARCH_X86
1282
1283 static void globalInit(){
1284     // generating tables:
1285     int i;
1286     for(i=0; i<768; i++){
1287         int c= MIN(MAX(i-256, 0), 255);
1288         clip_table[i]=c;
1289     }
1290 }
1291
1292 static SwsFunc getSwsFunc(int flags){
1293
1294 #ifdef RUNTIME_CPUDETECT
1295 #ifdef ARCH_X86
1296         // ordered per speed fasterst first
1297         if(flags & SWS_CPU_CAPS_MMX2)
1298                 return swScale_MMX2;
1299         else if(flags & SWS_CPU_CAPS_3DNOW)
1300                 return swScale_3DNow;
1301         else if(flags & SWS_CPU_CAPS_MMX)
1302                 return swScale_MMX;
1303         else
1304                 return swScale_C;
1305
1306 #else
1307         return swScale_C;
1308 #endif
1309 #else //RUNTIME_CPUDETECT
1310 #ifdef HAVE_MMX2
1311         return swScale_MMX2;
1312 #elif defined (HAVE_3DNOW)
1313         return swScale_3DNow;
1314 #elif defined (HAVE_MMX)
1315         return swScale_MMX;
1316 #else
1317         return swScale_C;
1318 #endif
1319 #endif //!RUNTIME_CPUDETECT
1320 }
1321
1322 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1323              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1324         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1325         /* Copy Y plane */
1326         if(dstStride[0]==srcStride[0])
1327                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1328         else
1329         {
1330                 int i;
1331                 uint8_t *srcPtr= src[0];
1332                 uint8_t *dstPtr= dst;
1333                 for(i=0; i<srcSliceH; i++)
1334                 {
1335                         memcpy(dstPtr, srcPtr, srcStride[0]);
1336                         srcPtr+= srcStride[0];
1337                         dstPtr+= dstStride[0];
1338                 }
1339         }
1340         dst = dstParam[1] + dstStride[1]*srcSliceY;
1341         interleaveBytes( src[1],src[2],dst,c->srcW,srcSliceH,srcStride[1],srcStride[2],dstStride[0] );
1342
1343         return srcSliceH;
1344 }
1345
1346 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1347              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1348         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1349
1350         yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1351
1352         return srcSliceH;
1353 }
1354
1355 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1356              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1357         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1358
1359         yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1360
1361         return srcSliceH;
1362 }
1363
1364 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1365 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1366                            int srcSliceH, uint8_t* dst[], int dstStride[]){
1367         const int srcFormat= c->srcFormat;
1368         const int dstFormat= c->dstFormat;
1369         const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1370         const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1371         const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8
1372         const int dstId= (dstFormat&0xFF)>>2;
1373         void (*conv)(const uint8_t *src, uint8_t *dst, unsigned src_size)=NULL;
1374
1375         /* BGR -> BGR */
1376         if(   (isBGR(srcFormat) && isBGR(dstFormat))
1377            || (isRGB(srcFormat) && isRGB(dstFormat))){
1378                 switch(srcId | (dstId<<4)){
1379                 case 0x34: conv= rgb16to15; break;
1380                 case 0x36: conv= rgb24to15; break;
1381                 case 0x38: conv= rgb32to15; break;
1382                 case 0x43: conv= rgb15to16; break;
1383                 case 0x46: conv= rgb24to16; break;
1384                 case 0x48: conv= rgb32to16; break;
1385                 case 0x63: conv= rgb15to24; break;
1386                 case 0x64: conv= rgb16to24; break;
1387                 case 0x68: conv= rgb32to24; break;
1388                 case 0x83: conv= rgb15to32; break;
1389                 case 0x84: conv= rgb16to32; break;
1390                 case 0x86: conv= rgb24to32; break;
1391                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1392                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1393                 }
1394         }else if(   (isBGR(srcFormat) && isRGB(dstFormat))
1395                  || (isRGB(srcFormat) && isBGR(dstFormat))){
1396                 switch(srcId | (dstId<<4)){
1397                 case 0x33: conv= rgb15tobgr15; break;
1398                 case 0x34: conv= rgb16tobgr15; break;
1399                 case 0x36: conv= rgb24tobgr15; break;
1400                 case 0x38: conv= rgb32tobgr15; break;
1401                 case 0x43: conv= rgb15tobgr16; break;
1402                 case 0x44: conv= rgb16tobgr16; break;
1403                 case 0x46: conv= rgb24tobgr16; break;
1404                 case 0x48: conv= rgb32tobgr16; break;
1405                 case 0x63: conv= rgb15tobgr24; break;
1406                 case 0x64: conv= rgb16tobgr24; break;
1407                 case 0x66: conv= rgb24tobgr24; break;
1408                 case 0x68: conv= rgb32tobgr24; break;
1409                 case 0x83: conv= rgb15tobgr32; break;
1410                 case 0x84: conv= rgb16tobgr32; break;
1411                 case 0x86: conv= rgb24tobgr32; break;
1412                 case 0x88: conv= rgb32tobgr32; break;
1413                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1414                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1415                 }
1416         }else{
1417                 MSG_ERR("swScaler: internal error %s -> %s converter\n",
1418                          vo_format_name(srcFormat), vo_format_name(dstFormat));
1419         }
1420
1421         if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1422                 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1423         else
1424         {
1425                 int i;
1426                 uint8_t *srcPtr= src[0];
1427                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1428
1429                 for(i=0; i<srcSliceH; i++)
1430                 {
1431                         conv(srcPtr, dstPtr, c->srcW*srcBpp);
1432                         srcPtr+= srcStride[0];
1433                         dstPtr+= dstStride[0];
1434                 }
1435         }
1436         return srcSliceH;
1437 }
1438
1439 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1440              int srcSliceH, uint8_t* dst[], int dstStride[]){
1441
1442         rgb24toyv12(
1443                 src[0],
1444                 dst[0]+ srcSliceY    *dstStride[0],
1445                 dst[1]+(srcSliceY>>1)*dstStride[1],
1446                 dst[2]+(srcSliceY>>1)*dstStride[2],
1447                 c->srcW, srcSliceH,
1448                 dstStride[0], dstStride[1], srcStride[0]);
1449         return srcSliceH;
1450 }
1451
1452 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1453              int srcSliceH, uint8_t* dst[], int dstStride[]){
1454         int i;
1455
1456         /* copy Y */
1457         if(srcStride[0]==dstStride[0])
1458                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1459         else{
1460                 uint8_t *srcPtr= src[0];
1461                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1462
1463                 for(i=0; i<srcSliceH; i++)
1464                 {
1465                         memcpy(dstPtr, srcPtr, c->srcW);
1466                         srcPtr+= srcStride[0];
1467                         dstPtr+= dstStride[0];
1468                 }
1469         }
1470
1471         if(c->dstFormat==IMGFMT_YV12){
1472                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1473                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1474         }else{
1475                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1476                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1477         }
1478         return srcSliceH;
1479 }
1480
1481 /**
1482  * bring pointers in YUV order instead of YVU
1483  */
1484 static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1485         if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
1486            || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
1487                 sortedP[0]= p[0];
1488                 sortedP[1]= p[2];
1489                 sortedP[2]= p[1];
1490                 sortedStride[0]= stride[0];
1491                 sortedStride[1]= stride[2];
1492                 sortedStride[2]= stride[1];
1493         }
1494         else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8)
1495         {
1496                 sortedP[0]= p[0];
1497                 sortedP[1]=
1498                 sortedP[2]= NULL;
1499                 sortedStride[0]= stride[0];
1500                 sortedStride[1]=
1501                 sortedStride[2]= 0;
1502         }
1503         else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)
1504         {
1505                 sortedP[0]= p[0];
1506                 sortedP[1]= p[1];
1507                 sortedP[2]= p[2];
1508                 sortedStride[0]= stride[0];
1509                 sortedStride[1]= stride[1];
1510                 sortedStride[2]= stride[2];
1511         }else{
1512                 MSG_ERR("internal error in orderYUV\n");
1513         }
1514 }
1515
1516 /* unscaled copy like stuff (assumes nearly identical formats) */
1517 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1518              int srcSliceH, uint8_t* dst[], int dstStride[]){
1519
1520         if(isPacked(c->srcFormat))
1521         {
1522                 if(dstStride[0]==srcStride[0])
1523                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1524                 else
1525                 {
1526                         int i;
1527                         uint8_t *srcPtr= src[0];
1528                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1529                         int length=0;
1530
1531                         /* universal length finder */
1532                         while(length+c->srcW <= ABS(dstStride[0])
1533                            && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
1534                         ASSERT(length!=0);
1535
1536                         for(i=0; i<srcSliceH; i++)
1537                         {
1538                                 memcpy(dstPtr, srcPtr, length);
1539                                 srcPtr+= srcStride[0];
1540                                 dstPtr+= dstStride[0];
1541                         }
1542                 }
1543         }
1544         else
1545         { /* Planar YUV or gray */
1546                 int plane;
1547                 for(plane=0; plane<3; plane++)
1548                 {
1549                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1550                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1551                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1552
1553                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1554                         {
1555                                 if(!isGray(c->dstFormat))
1556                                         memset(dst[plane], 128, dstStride[plane]*height);
1557                         }
1558                         else
1559                         {
1560                                 if(dstStride[plane]==srcStride[plane])
1561                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1562                                 else
1563                                 {
1564                                         int i;
1565                                         uint8_t *srcPtr= src[plane];
1566                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1567                                         for(i=0; i<height; i++)
1568                                         {
1569                                                 memcpy(dstPtr, srcPtr, length);
1570                                                 srcPtr+= srcStride[plane];
1571                                                 dstPtr+= dstStride[plane];
1572                                         }
1573                                 }
1574                         }
1575                 }
1576         }
1577         return srcSliceH;
1578 }
1579
1580 static int remove_dup_fourcc(int fourcc)
1581 {
1582         switch(fourcc)
1583         {
1584             case IMGFMT_I420:
1585             case IMGFMT_IYUV: return IMGFMT_YV12;
1586             case IMGFMT_Y8  : return IMGFMT_Y800;
1587             case IMGFMT_IF09: return IMGFMT_YVU9;
1588             default: return fourcc;
1589         }
1590 }
1591
1592 static void getSubSampleFactors(int *h, int *v, int format){
1593         switch(format){
1594         case IMGFMT_UYVY:
1595         case IMGFMT_YUY2:
1596                 *h=1;
1597                 *v=0;
1598                 break;
1599         case IMGFMT_YV12:
1600         case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
1601                 *h=1;
1602                 *v=1;
1603                 break;
1604         case IMGFMT_YVU9:
1605                 *h=2;
1606                 *v=2;
1607                 break;
1608         case IMGFMT_444P:
1609                 *h=0;
1610                 *v=0;
1611                 break;
1612         case IMGFMT_422P:
1613                 *h=1;
1614                 *v=0;
1615                 break;
1616         case IMGFMT_411P:
1617                 *h=2;
1618                 *v=0;
1619                 break;
1620         default:
1621                 *h=0;
1622                 *v=0;
1623                 break;
1624         }
1625 }
1626
1627 static uint16_t roundToInt16(int64_t f){
1628         int r= (f + (1<<15))>>16;
1629              if(r<-0x7FFF) return 0x8000;
1630         else if(r> 0x7FFF) return 0x7FFF;
1631         else               return r;
1632 }
1633
1634 /**
1635  * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1636  * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1637  * @return -1 if not supported
1638  */
1639 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1640         int64_t crv =  inv_table[0];
1641         int64_t cbu =  inv_table[1];
1642         int64_t cgu = -inv_table[2];
1643         int64_t cgv = -inv_table[3];
1644         int64_t cy  = 1<<16;
1645         int64_t oy  = 0;
1646
1647         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1648         memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1649         memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
1650
1651         c->brightness= brightness;
1652         c->contrast  = contrast;
1653         c->saturation= saturation;
1654         c->srcRange  = srcRange;
1655         c->dstRange  = dstRange;
1656
1657         c->uOffset=   0x0400040004000400LL;
1658         c->vOffset=   0x0400040004000400LL;
1659
1660         if(!srcRange){
1661                 cy= (cy*255) / 219;
1662                 oy= 16<<16;
1663         }
1664
1665         cy = (cy *contrast             )>>16;
1666         crv= (crv*contrast * saturation)>>32;
1667         cbu= (cbu*contrast * saturation)>>32;
1668         cgu= (cgu*contrast * saturation)>>32;
1669         cgv= (cgv*contrast * saturation)>>32;
1670
1671         oy -= 256*brightness;
1672
1673         c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
1674         c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
1675         c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1676         c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1677         c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1678         c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
1679
1680         yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1681         //FIXME factorize
1682
1683         return 0;
1684 }
1685
1686 /**
1687  * @return -1 if not supported
1688  */
1689 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1690         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1691
1692         *inv_table = c->srcColorspaceTable;
1693         *table     = c->dstColorspaceTable;
1694         *srcRange  = c->srcRange;
1695         *dstRange  = c->dstRange;
1696         *brightness= c->brightness;
1697         *contrast  = c->contrast;
1698         *saturation= c->saturation;
1699
1700         return 0;
1701 }
1702
1703 SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,
1704                          SwsFilter *srcFilter, SwsFilter *dstFilter){
1705
1706         SwsContext *c;
1707         int i;
1708         int usesFilter;
1709         int unscaled, needsDither;
1710         int srcFormat, dstFormat;
1711         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1712 #ifdef ARCH_X86
1713         if(flags & SWS_CPU_CAPS_MMX)
1714                 asm volatile("emms\n\t"::: "memory");
1715 #endif
1716
1717 #ifndef RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
1718         flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW);
1719 #ifdef HAVE_MMX2
1720         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1721 #elif defined (HAVE_3DNOW)
1722         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1723 #elif defined (HAVE_MMX)
1724         flags |= SWS_CPU_CAPS_MMX;
1725 #endif
1726 #endif
1727         if(clip_table[512] != 255) globalInit();
1728         if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1729
1730         /* avoid duplicate Formats, so we don't need to check to much */
1731         srcFormat = remove_dup_fourcc(origSrcFormat);
1732         dstFormat = remove_dup_fourcc(origDstFormat);
1733
1734         unscaled = (srcW == dstW && srcH == dstH);
1735         needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
1736                      && (dstFormat&0xFF)<24
1737                      && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1738
1739         if(!isSupportedIn(srcFormat))
1740         {
1741                 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1742                 return NULL;
1743         }
1744         if(!isSupportedOut(dstFormat))
1745         {
1746                 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1747                 return NULL;
1748         }
1749
1750         /* sanity check */
1751         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
1752         {
1753                  MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
1754                         srcW, srcH, dstW, dstH);
1755                 return NULL;
1756         }
1757
1758         if(!dstFilter) dstFilter= &dummyFilter;
1759         if(!srcFilter) srcFilter= &dummyFilter;
1760
1761         c= memalign(64, sizeof(SwsContext));
1762         memset(c, 0, sizeof(SwsContext));
1763
1764         c->srcW= srcW;
1765         c->srcH= srcH;
1766         c->dstW= dstW;
1767         c->dstH= dstH;
1768         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1769         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1770         c->flags= flags;
1771         c->dstFormat= dstFormat;
1772         c->srcFormat= srcFormat;
1773         c->origDstFormat= origDstFormat;
1774         c->origSrcFormat= origSrcFormat;
1775
1776         usesFilter=0;
1777         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesFilter=1;
1778         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesFilter=1;
1779         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesFilter=1;
1780         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesFilter=1;
1781         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesFilter=1;
1782         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesFilter=1;
1783         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesFilter=1;
1784         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesFilter=1;
1785
1786         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1787         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1788
1789         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1790         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1791
1792         // drop some chroma lines if the user wants it
1793         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1794         c->chrSrcVSubSample+= c->vChrDrop;
1795
1796         // drop every 2. pixel for chroma calculation unless user wants full chroma
1797         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
1798                 c->chrSrcHSubSample=1;
1799
1800         c->chrIntHSubSample= c->chrDstHSubSample;
1801         c->chrIntVSubSample= c->chrSrcVSubSample;
1802
1803         // note the -((-x)>>y) is so that we allways round toward +inf
1804         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1805         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
1806         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
1807         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
1808
1809         sws_setColorspaceDetails(c, Inverse_Table_6_9[SWS_CS_DEFAULT], 0, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, 0, 0, 1<<16, 1<<16);
1810
1811         /* unscaled special Cases */
1812         if(unscaled && !usesFilter)
1813         {
1814                 /* yv12_to_nv12 */
1815                 if(srcFormat == IMGFMT_YV12 && dstFormat == IMGFMT_NV12)
1816                 {
1817                         c->swScale= PlanarToNV12Wrapper;
1818                 }
1819                 /* yuv2bgr */
1820                 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
1821                 {
1822                         c->swScale= yuv2rgb_get_func_ptr(c);
1823                 }
1824
1825                 if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 )
1826                 {
1827                         c->swScale= yvu9toyv12Wrapper;
1828                 }
1829
1830                 /* bgr24toYV12 */
1831                 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
1832                         c->swScale= bgr24toyv12Wrapper;
1833
1834                 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
1835                 if(   (isBGR(srcFormat) || isRGB(srcFormat))
1836                    && (isBGR(dstFormat) || isRGB(dstFormat))
1837                    && !needsDither)
1838                         c->swScale= rgb2rgbWrapper;
1839
1840                 /* LQ converters if -sws 0 or -sws 4*/
1841                 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
1842                         /* rgb/bgr -> rgb/bgr (dither needed forms) */
1843                         if(  (isBGR(srcFormat) || isRGB(srcFormat))
1844                           && (isBGR(dstFormat) || isRGB(dstFormat))
1845                           && needsDither)
1846                                 c->swScale= rgb2rgbWrapper;
1847
1848                         /* yv12_to_yuy2 */
1849                         if(srcFormat == IMGFMT_YV12 &&
1850                             (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY))
1851                         {
1852                                 if (dstFormat == IMGFMT_YUY2)
1853                                     c->swScale= PlanarToYuy2Wrapper;
1854                                 else
1855                                     c->swScale= PlanarToUyvyWrapper;
1856                         }
1857                 }
1858
1859                 /* simple copy */
1860                 if(   srcFormat == dstFormat
1861                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
1862                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
1863                   )
1864                 {
1865                         c->swScale= simpleCopy;
1866                 }
1867
1868                 if(c->swScale){
1869                         if(flags&SWS_PRINT_INFO)
1870                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
1871                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
1872                         return c;
1873                 }
1874         }
1875
1876         if(flags & SWS_CPU_CAPS_MMX2)
1877         {
1878                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
1879                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
1880                 {
1881                         if(flags&SWS_PRINT_INFO)
1882                                 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
1883                 }
1884         }
1885         else
1886                 c->canMMX2BeUsed=0;
1887
1888         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
1889         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
1890
1891         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
1892         // but only for the FAST_BILINEAR mode otherwise do correct scaling
1893         // n-2 is the last chrominance sample available
1894         // this is not perfect, but noone shuld notice the difference, the more correct variant
1895         // would be like the vertical one, but that would require some special code for the
1896         // first and last pixel
1897         if(flags&SWS_FAST_BILINEAR)
1898         {
1899                 if(c->canMMX2BeUsed)
1900                 {
1901                         c->lumXInc+= 20;
1902                         c->chrXInc+= 20;
1903                 }
1904                 //we don't use the x86asm scaler if mmx is available
1905                 else if(flags & SWS_CPU_CAPS_MMX)
1906                 {
1907                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
1908                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
1909                 }
1910         }
1911
1912         /* precalculate horizontal scaler filter coefficients */
1913         {
1914                 const int filterAlign= (flags & SWS_CPU_CAPS_MMX) ? 4 : 1;
1915
1916                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
1917                                  srcW      ,       dstW, filterAlign, 1<<14,
1918                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
1919                                  srcFilter->lumH, dstFilter->lumH);
1920                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
1921                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
1922                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
1923                                  srcFilter->chrH, dstFilter->chrH);
1924
1925 #ifdef ARCH_X86
1926 // can't downscale !!!
1927                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
1928                 {
1929                         c->lumMmx2Filter   = (int16_t*)memalign(8, (dstW        /8+8)*sizeof(int16_t));
1930                         c->chrMmx2Filter   = (int16_t*)memalign(8, (c->chrDstW  /4+8)*sizeof(int16_t));
1931                         c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW      /2/8+8)*sizeof(int32_t));
1932                         c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
1933
1934                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
1935                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
1936                 }
1937 #endif
1938         } // Init Horizontal stuff
1939
1940
1941
1942         /* precalculate vertical scaler filter coefficients */
1943         initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
1944                         srcH      ,        dstH, 1, (1<<12)-4,
1945                         (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
1946                         srcFilter->lumV, dstFilter->lumV);
1947         initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
1948                         c->chrSrcH, c->chrDstH, 1, (1<<12)-4,
1949                         (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
1950                         srcFilter->chrV, dstFilter->chrV);
1951
1952         // Calculate Buffer Sizes so that they won't run out while handling these damn slices
1953         c->vLumBufSize= c->vLumFilterSize;
1954         c->vChrBufSize= c->vChrFilterSize;
1955         for(i=0; i<dstH; i++)
1956         {
1957                 int chrI= i*c->chrDstH / dstH;
1958                 int nextSlice= MAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
1959                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
1960                 if(c->chrSrcVSubSample > 1)
1961                     nextSlice&= ~3; // Slices start at boundaries which are divisable through 4
1962                 else
1963                     nextSlice&= ~1; // Slices start at boundaries which are divisable through 2
1964                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
1965                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
1966                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1967                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1968         }
1969
1970         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1971         c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
1972         c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
1973         //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)
1974         for(i=0; i<c->vLumBufSize; i++)
1975                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
1976         for(i=0; i<c->vChrBufSize; i++)
1977                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
1978
1979         //try to avoid drawing green stuff between the right end and the stride end
1980         for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
1981         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
1982
1983         ASSERT(c->chrDstH <= dstH)
1984
1985         if(flags&SWS_PRINT_INFO)
1986         {
1987 #ifdef DITHER1XBPP
1988                 char *dither= " dithered";
1989 #else
1990                 char *dither= "";
1991 #endif
1992                 if(flags&SWS_FAST_BILINEAR)
1993                         MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
1994                 else if(flags&SWS_BILINEAR)
1995                         MSG_INFO("\nSwScaler: BILINEAR scaler, ");
1996                 else if(flags&SWS_BICUBIC)
1997                         MSG_INFO("\nSwScaler: BICUBIC scaler, ");
1998                 else if(flags&SWS_X)
1999                         MSG_INFO("\nSwScaler: Experimental scaler, ");
2000                 else if(flags&SWS_POINT)
2001                         MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
2002                 else if(flags&SWS_AREA)
2003                         MSG_INFO("\nSwScaler: Area Averageing scaler, ");
2004                 else if(flags&SWS_BICUBLIN)
2005                         MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2006                 else if(flags&SWS_GAUSS)
2007                         MSG_INFO("\nSwScaler: Gaussian scaler, ");
2008                 else if(flags&SWS_SINC)
2009                         MSG_INFO("\nSwScaler: Sinc scaler, ");
2010                 else if(flags&SWS_LANCZOS)
2011                         MSG_INFO("\nSwScaler: Lanczos scaler, ");
2012                 else if(flags&SWS_SPLINE)
2013                         MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
2014                 else
2015                         MSG_INFO("\nSwScaler: ehh flags invalid?! ");
2016
2017                 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2018                         MSG_INFO("from %s to%s %s ",
2019                                 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
2020                 else
2021                         MSG_INFO("from %s to %s ",
2022                                 vo_format_name(srcFormat), vo_format_name(dstFormat));
2023
2024                 if(flags & SWS_CPU_CAPS_MMX2)
2025                         MSG_INFO("using MMX2\n");
2026                 else if(flags & SWS_CPU_CAPS_3DNOW)
2027                         MSG_INFO("using 3DNOW\n");
2028                 else if(flags & SWS_CPU_CAPS_MMX)
2029                         MSG_INFO("using MMX\n");
2030                 else
2031                         MSG_INFO("using C\n");
2032         }
2033
2034         if(flags & SWS_PRINT_INFO)
2035         {
2036                 if(flags & SWS_CPU_CAPS_MMX)
2037                 {
2038                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2039                                 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2040                         else
2041                         {
2042                                 if(c->hLumFilterSize==4)
2043                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2044                                 else if(c->hLumFilterSize==8)
2045                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2046                                 else
2047                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2048
2049                                 if(c->hChrFilterSize==4)
2050                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2051                                 else if(c->hChrFilterSize==8)
2052                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2053                                 else
2054                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2055                         }
2056                 }
2057                 else
2058                 {
2059 #ifdef ARCH_X86
2060                         MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2061 #else
2062                         if(flags & SWS_FAST_BILINEAR)
2063                                 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2064                         else
2065                                 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2066 #endif
2067                 }
2068                 if(isPlanarYUV(dstFormat))
2069                 {
2070                         if(c->vLumFilterSize==1)
2071                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2072                         else
2073                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2074                 }
2075                 else
2076                 {
2077                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2078                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2079                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2080                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2081                                 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2082                         else
2083                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2084                 }
2085
2086                 if(dstFormat==IMGFMT_BGR24)
2087                         MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2088                                 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2089                 else if(dstFormat==IMGFMT_BGR32)
2090                         MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2091                 else if(dstFormat==IMGFMT_BGR16)
2092                         MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2093                 else if(dstFormat==IMGFMT_BGR15)
2094                         MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2095
2096                 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2097         }
2098         if(flags & SWS_PRINT_INFO)
2099         {
2100                 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2101                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2102                 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2103                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2104         }
2105
2106         c->swScale= getSwsFunc(flags);
2107         return c;
2108 }
2109
2110 /**
2111  * swscale warper, so we don't need to export the SwsContext.
2112  * assumes planar YUV to be in YUV order instead of YVU
2113  */
2114 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2115                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2116         return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2117 }
2118
2119 /**
2120  * swscale warper, so we don't need to export the SwsContext
2121  */
2122 int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
2123                            int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
2124         int srcStride[3];
2125         int dstStride[3];
2126         uint8_t *src[3];
2127         uint8_t *dst[3];
2128         sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam);
2129         sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam);
2130 //printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
2131
2132         return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2133 }
2134
2135 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2136                                 float lumaSharpen, float chromaSharpen,
2137                                 float chromaHShift, float chromaVShift,
2138                                 int verbose)
2139 {
2140         SwsFilter *filter= malloc(sizeof(SwsFilter));
2141
2142         if(lumaGBlur!=0.0){
2143                 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2144                 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2145         }else{
2146                 filter->lumH= sws_getIdentityVec();
2147                 filter->lumV= sws_getIdentityVec();
2148         }
2149
2150         if(chromaGBlur!=0.0){
2151                 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2152                 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2153         }else{
2154                 filter->chrH= sws_getIdentityVec();
2155                 filter->chrV= sws_getIdentityVec();
2156         }
2157
2158         if(chromaSharpen!=0.0){
2159                 SwsVector *g= sws_getConstVec(-1.0, 3);
2160                 SwsVector *id= sws_getConstVec(10.0/chromaSharpen, 1);
2161                 g->coeff[1]=2.0;
2162                 sws_addVec(id, g);
2163                 sws_convVec(filter->chrH, id);
2164                 sws_convVec(filter->chrV, id);
2165                 sws_freeVec(g);
2166                 sws_freeVec(id);
2167         }
2168
2169         if(lumaSharpen!=0.0){
2170                 SwsVector *g= sws_getConstVec(-1.0, 3);
2171                 SwsVector *id= sws_getConstVec(10.0/lumaSharpen, 1);
2172                 g->coeff[1]=2.0;
2173                 sws_addVec(id, g);
2174                 sws_convVec(filter->lumH, id);
2175                 sws_convVec(filter->lumV, id);
2176                 sws_freeVec(g);
2177                 sws_freeVec(id);
2178         }
2179
2180         if(chromaHShift != 0.0)
2181                 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2182
2183         if(chromaVShift != 0.0)
2184                 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2185
2186         sws_normalizeVec(filter->chrH, 1.0);
2187         sws_normalizeVec(filter->chrV, 1.0);
2188         sws_normalizeVec(filter->lumH, 1.0);
2189         sws_normalizeVec(filter->lumV, 1.0);
2190
2191         if(verbose) sws_printVec(filter->chrH);
2192         if(verbose) sws_printVec(filter->lumH);
2193
2194         return filter;
2195 }
2196
2197 /**
2198  * returns a normalized gaussian curve used to filter stuff
2199  * quality=3 is high quality, lowwer is lowwer quality
2200  */
2201 SwsVector *sws_getGaussianVec(double variance, double quality){
2202         const int length= (int)(variance*quality + 0.5) | 1;
2203         int i;
2204         double *coeff= memalign(sizeof(double), length*sizeof(double));
2205         double middle= (length-1)*0.5;
2206         SwsVector *vec= malloc(sizeof(SwsVector));
2207
2208         vec->coeff= coeff;
2209         vec->length= length;
2210
2211         for(i=0; i<length; i++)
2212         {
2213                 double dist= i-middle;
2214                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2215         }
2216
2217         sws_normalizeVec(vec, 1.0);
2218
2219         return vec;
2220 }
2221
2222 SwsVector *sws_getConstVec(double c, int length){
2223         int i;
2224         double *coeff= memalign(sizeof(double), length*sizeof(double));
2225         SwsVector *vec= malloc(sizeof(SwsVector));
2226
2227         vec->coeff= coeff;
2228         vec->length= length;
2229
2230         for(i=0; i<length; i++)
2231                 coeff[i]= c;
2232
2233         return vec;
2234 }
2235
2236
2237 SwsVector *sws_getIdentityVec(void){
2238         double *coeff= memalign(sizeof(double), sizeof(double));
2239         SwsVector *vec= malloc(sizeof(SwsVector));
2240         coeff[0]= 1.0;
2241
2242         vec->coeff= coeff;
2243         vec->length= 1;
2244
2245         return vec;
2246 }
2247
2248 void sws_normalizeVec(SwsVector *a, double height){
2249         int i;
2250         double sum=0;
2251         double inv;
2252
2253         for(i=0; i<a->length; i++)
2254                 sum+= a->coeff[i];
2255
2256         inv= height/sum;
2257
2258         for(i=0; i<a->length; i++)
2259                 a->coeff[i]*= inv;
2260 }
2261
2262 void sws_scaleVec(SwsVector *a, double scalar){
2263         int i;
2264
2265         for(i=0; i<a->length; i++)
2266                 a->coeff[i]*= scalar;
2267 }
2268
2269 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2270         int length= a->length + b->length - 1;
2271         double *coeff= memalign(sizeof(double), length*sizeof(double));
2272         int i, j;
2273         SwsVector *vec= malloc(sizeof(SwsVector));
2274
2275         vec->coeff= coeff;
2276         vec->length= length;
2277
2278         for(i=0; i<length; i++) coeff[i]= 0.0;
2279
2280         for(i=0; i<a->length; i++)
2281         {
2282                 for(j=0; j<b->length; j++)
2283                 {
2284                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2285                 }
2286         }
2287
2288         return vec;
2289 }
2290
2291 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2292         int length= MAX(a->length, b->length);
2293         double *coeff= memalign(sizeof(double), length*sizeof(double));
2294         int i;
2295         SwsVector *vec= malloc(sizeof(SwsVector));
2296
2297         vec->coeff= coeff;
2298         vec->length= length;
2299
2300         for(i=0; i<length; i++) coeff[i]= 0.0;
2301
2302         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2303         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2304
2305         return vec;
2306 }
2307
2308 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2309         int length= MAX(a->length, b->length);
2310         double *coeff= memalign(sizeof(double), length*sizeof(double));
2311         int i;
2312         SwsVector *vec= malloc(sizeof(SwsVector));
2313
2314         vec->coeff= coeff;
2315         vec->length= length;
2316
2317         for(i=0; i<length; i++) coeff[i]= 0.0;
2318
2319         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2320         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2321
2322         return vec;
2323 }
2324
2325 /* shift left / or right if "shift" is negative */
2326 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2327         int length= a->length + ABS(shift)*2;
2328         double *coeff= memalign(sizeof(double), length*sizeof(double));
2329         int i;
2330         SwsVector *vec= malloc(sizeof(SwsVector));
2331
2332         vec->coeff= coeff;
2333         vec->length= length;
2334
2335         for(i=0; i<length; i++) coeff[i]= 0.0;
2336
2337         for(i=0; i<a->length; i++)
2338         {
2339                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2340         }
2341
2342         return vec;
2343 }
2344
2345 void sws_shiftVec(SwsVector *a, int shift){
2346         SwsVector *shifted= sws_getShiftedVec(a, shift);
2347         free(a->coeff);
2348         a->coeff= shifted->coeff;
2349         a->length= shifted->length;
2350         free(shifted);
2351 }
2352
2353 void sws_addVec(SwsVector *a, SwsVector *b){
2354         SwsVector *sum= sws_sumVec(a, b);
2355         free(a->coeff);
2356         a->coeff= sum->coeff;
2357         a->length= sum->length;
2358         free(sum);
2359 }
2360
2361 void sws_subVec(SwsVector *a, SwsVector *b){
2362         SwsVector *diff= sws_diffVec(a, b);
2363         free(a->coeff);
2364         a->coeff= diff->coeff;
2365         a->length= diff->length;
2366         free(diff);
2367 }
2368
2369 void sws_convVec(SwsVector *a, SwsVector *b){
2370         SwsVector *conv= sws_getConvVec(a, b);
2371         free(a->coeff);
2372         a->coeff= conv->coeff;
2373         a->length= conv->length;
2374         free(conv);
2375 }
2376
2377 SwsVector *sws_cloneVec(SwsVector *a){
2378         double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2379         int i;
2380         SwsVector *vec= malloc(sizeof(SwsVector));
2381
2382         vec->coeff= coeff;
2383         vec->length= a->length;
2384
2385         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2386
2387         return vec;
2388 }
2389
2390 void sws_printVec(SwsVector *a){
2391         int i;
2392         double max=0;
2393         double min=0;
2394         double range;
2395
2396         for(i=0; i<a->length; i++)
2397                 if(a->coeff[i]>max) max= a->coeff[i];
2398
2399         for(i=0; i<a->length; i++)
2400                 if(a->coeff[i]<min) min= a->coeff[i];
2401
2402         range= max - min;
2403
2404         for(i=0; i<a->length; i++)
2405         {
2406                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2407                 MSG_DBG2("%1.3f ", a->coeff[i]);
2408                 for(;x>0; x--) MSG_DBG2(" ");
2409                 MSG_DBG2("|\n");
2410         }
2411 }
2412
2413 void sws_freeVec(SwsVector *a){
2414         if(!a) return;
2415         if(a->coeff) free(a->coeff);
2416         a->coeff=NULL;
2417         a->length=0;
2418         free(a);
2419 }
2420
2421 void sws_freeFilter(SwsFilter *filter){
2422         if(!filter) return;
2423
2424         if(filter->lumH) sws_freeVec(filter->lumH);
2425         if(filter->lumV) sws_freeVec(filter->lumV);
2426         if(filter->chrH) sws_freeVec(filter->chrH);
2427         if(filter->chrV) sws_freeVec(filter->chrV);
2428         free(filter);
2429 }
2430
2431
2432 void sws_freeContext(SwsContext *c){
2433         int i;
2434         if(!c) return;
2435
2436         if(c->lumPixBuf)
2437         {
2438                 for(i=0; i<c->vLumBufSize; i++)
2439                 {
2440                         if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2441                         c->lumPixBuf[i]=NULL;
2442                 }
2443                 free(c->lumPixBuf);
2444                 c->lumPixBuf=NULL;
2445         }
2446
2447         if(c->chrPixBuf)
2448         {
2449                 for(i=0; i<c->vChrBufSize; i++)
2450                 {
2451                         if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2452                         c->chrPixBuf[i]=NULL;
2453                 }
2454                 free(c->chrPixBuf);
2455                 c->chrPixBuf=NULL;
2456         }
2457
2458         if(c->vLumFilter) free(c->vLumFilter);
2459         c->vLumFilter = NULL;
2460         if(c->vChrFilter) free(c->vChrFilter);
2461         c->vChrFilter = NULL;
2462         if(c->hLumFilter) free(c->hLumFilter);
2463         c->hLumFilter = NULL;
2464         if(c->hChrFilter) free(c->hChrFilter);
2465         c->hChrFilter = NULL;
2466
2467         if(c->vLumFilterPos) free(c->vLumFilterPos);
2468         c->vLumFilterPos = NULL;
2469         if(c->vChrFilterPos) free(c->vChrFilterPos);
2470         c->vChrFilterPos = NULL;
2471         if(c->hLumFilterPos) free(c->hLumFilterPos);
2472         c->hLumFilterPos = NULL;
2473         if(c->hChrFilterPos) free(c->hChrFilterPos);
2474         c->hChrFilterPos = NULL;
2475
2476         if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2477         c->lumMmx2Filter=NULL;
2478         if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2479         c->chrMmx2Filter=NULL;
2480         if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2481         c->lumMmx2FilterPos=NULL;
2482         if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2483         c->chrMmx2FilterPos=NULL;
2484         if(c->yuvTable) free(c->yuvTable);
2485         c->yuvTable=NULL;
2486
2487         free(c);
2488 }
2489