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, {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\
 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         }\
 507
 508
 509 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
 510                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
 511                                     uint8_t *dest, int dstW, int y)
 512 {
 513         int i;
 514         switch(c->dstFormat)
 515         {
 516         case IMGFMT_RGB32:
 517         case IMGFMT_BGR32:
 518                 YSCALE_YUV_2_RGBX_C(uint32_t)
 519                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
 520                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
 521                 }
 522                 break;
 523         case IMGFMT_RGB24:
 524                 YSCALE_YUV_2_RGBX_C(uint8_t)
 525                         ((uint8_t*)dest)[0]= r[Y1];
 526                         ((uint8_t*)dest)[1]= g[Y1];
 527                         ((uint8_t*)dest)[2]= b[Y1];
 528                         ((uint8_t*)dest)[3]= r[Y2];
 529                         ((uint8_t*)dest)[4]= g[Y2];
 530                         ((uint8_t*)dest)[5]= b[Y2];
 531                         ((uint8_t*)dest)+=6;
 532                 }
 533                 break;
 534         case IMGFMT_BGR24:
 535                 YSCALE_YUV_2_RGBX_C(uint8_t)
 536                         ((uint8_t*)dest)[0]= b[Y1];
 537                         ((uint8_t*)dest)[1]= g[Y1];
 538                         ((uint8_t*)dest)[2]= r[Y1];
 539                         ((uint8_t*)dest)[3]= b[Y2];
 540                         ((uint8_t*)dest)[4]= g[Y2];
 541                         ((uint8_t*)dest)[5]= r[Y2];
 542                         ((uint8_t*)dest)+=6;
 543                 }
 544                 break;
 545         case IMGFMT_RGB16:
 546         case IMGFMT_BGR16:
 547                 {
 548                         const int dr1= dither_2x2_8[y&1    ][0];
 549                         const int dg1= dither_2x2_4[y&1    ][0];
 550                         const int db1= dither_2x2_8[(y&1)^1][0];
 551                         const int dr2= dither_2x2_8[y&1    ][1];
 552                         const int dg2= dither_2x2_4[y&1    ][1];
 553                         const int db2= dither_2x2_8[(y&1)^1][1];
 554                         YSCALE_YUV_2_RGBX_C(uint16_t)
 555                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
 556                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
 557                         }
 558                 }
 559                 break;
 560         case IMGFMT_RGB15:
 561         case IMGFMT_BGR15:
 562                 {
 563                         const int dr1= dither_2x2_8[y&1    ][0];
 564                         const int dg1= dither_2x2_8[y&1    ][1];
 565                         const int db1= dither_2x2_8[(y&1)^1][0];
 566                         const int dr2= dither_2x2_8[y&1    ][1];
 567                         const int dg2= dither_2x2_8[y&1    ][0];
 568                         const int db2= dither_2x2_8[(y&1)^1][1];
 569                         YSCALE_YUV_2_RGBX_C(uint16_t)
 570                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
 571                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
 572                         }
 573                 }
 574                 break;
 575         case IMGFMT_RGB8:
 576         case IMGFMT_BGR8:
 577                 {
 578                         const uint8_t * const d64= dither_8x8_73[y&7];
 579                         const uint8_t * const d32= dither_8x8_32[y&7];
 580                         YSCALE_YUV_2_RGBX_C(uint8_t)
 581                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
 582                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
 583                         }
 584                 }
 585                 break;
 586         case IMGFMT_RGB4:
 587         case IMGFMT_BGR4:
 588                 {
 589                         const uint8_t * const d64= dither_8x8_73 [y&7];
 590                         const uint8_t * const d128=dither_8x8_220[y&7];
 591                         YSCALE_YUV_2_RGBX_C(uint8_t)
 592                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
 593                                                   +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
 594                         }
 595                 }
 596                 break;
 597         case IMGFMT_RG4B:
 598         case IMGFMT_BG4B:
 599                 {
 600                         const uint8_t * const d64= dither_8x8_73 [y&7];
 601                         const uint8_t * const d128=dither_8x8_220[y&7];
 602                         YSCALE_YUV_2_RGBX_C(uint8_t)
 603                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
 604                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
 605                         }
 606                 }
 607                 break;
 608         case IMGFMT_RGB1:
 609         case IMGFMT_BGR1:
 610                 {
 611                         const uint8_t * const d128=dither_8x8_220[y&7];
 612                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
 613                         int acc=0;
 614                         for(i=0; i<dstW-1; i+=2){
 615                                 int j;
 616                                 int Y1=0;
 617                                 int Y2=0;
 618
 619                                 for(j=0; j<lumFilterSize; j++)
 620                                 {
 621                                         Y1 += lumSrc[j][i] * lumFilter[j];
 622                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
 623                                 }
 624                                 Y1>>=19;
 625                                 Y2>>=19;
 626                                 if((Y1|Y2)&256)
 627                                 {
 628                                         if(Y1>255)   Y1=255;
 629                                         else if(Y1<0)Y1=0;
 630                                         if(Y2>255)   Y2=255;
 631                                         else if(Y2<0)Y2=0;
 632                                 }
 633                                 acc+= acc + g[Y1+d128[(i+0)&7]];
 634                                 acc+= acc + g[Y2+d128[(i+1)&7]];
 635                                 if((i&7)==6){
 636                                         ((uint8_t*)dest)[0]= acc;
 637                                         ((uint8_t*)dest)++;
 638                                 }
 639                         }
 640                 }
 641                 break;
 642         case IMGFMT_YUY2:
 643                 YSCALE_YUV_2_PACKEDX_C(void)
 644                         ((uint8_t*)dest)[2*i2+0]= Y1;
 645                         ((uint8_t*)dest)[2*i2+1]= U;
 646                         ((uint8_t*)dest)[2*i2+2]= Y2;
 647                         ((uint8_t*)dest)[2*i2+3]= V;
 648                 }
 649                 break;
 650         }
 651 }
 652
 653
 654 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
 655 //Plain C versions
 656 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
 657 #define COMPILE_C
 658 #endif
 659
 660 #ifdef ARCH_X86
 661
 662 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
 663 #define COMPILE_MMX
 664 #endif
 665
 666 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
 667 #define COMPILE_MMX2
 668 #endif
 669
 670 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
 671 #define COMPILE_3DNOW
 672 #endif
 673 #endif //ARCH_X86
 674
 675 #undef HAVE_MMX
 676 #undef HAVE_MMX2
 677 #undef HAVE_3DNOW
 678
 679 #ifdef COMPILE_C
 680 #undef HAVE_MMX
 681 #undef HAVE_MMX2
 682 #undef HAVE_3DNOW
 683 #define RENAME(a) a ## _C
 684 #include "swscale_template.c"
 685 #endif
 686
 687 #ifdef ARCH_X86
 688
 689 //X86 versions
 690 /*
 691 #undef RENAME
 692 #undef HAVE_MMX
 693 #undef HAVE_MMX2
 694 #undef HAVE_3DNOW
 695 #define ARCH_X86
 696 #define RENAME(a) a ## _X86
 697 #include "swscale_template.c"
 698 */
 699 //MMX versions
 700 #ifdef COMPILE_MMX
 701 #undef RENAME
 702 #define HAVE_MMX
 703 #undef HAVE_MMX2
 704 #undef HAVE_3DNOW
 705 #define RENAME(a) a ## _MMX
 706 #include "swscale_template.c"
 707 #endif
 708
 709 //MMX2 versions
 710 #ifdef COMPILE_MMX2
 711 #undef RENAME
 712 #define HAVE_MMX
 713 #define HAVE_MMX2
 714 #undef HAVE_3DNOW
 715 #define RENAME(a) a ## _MMX2
 716 #include "swscale_template.c"
 717 #endif
 718
 719 //3DNOW versions
 720 #ifdef COMPILE_3DNOW
 721 #undef RENAME
 722 #define HAVE_MMX
 723 #undef HAVE_MMX2
 724 #define HAVE_3DNOW
 725 #define RENAME(a) a ## _3DNow
 726 #include "swscale_template.c"
 727 #endif
 728
 729 #endif //ARCH_X86
 730
 731 // minor note: the HAVE_xyz is messed up after that line so dont use it
 732
 733 static double getSplineCoeff(double a, double b, double c, double d, double dist)
 734 {
 735 //      printf("%f %f %f %f %f\n", a,b,c,d,dist);
 736         if(dist<=1.0)   return ((d*dist + c)*dist + b)*dist +a;
 737         else            return getSplineCoeff(  0.0,
 738                                                  b+ 2.0*c + 3.0*d,
 739                                                         c + 3.0*d,
 740                                                 -b- 3.0*c - 6.0*d,
 741                                                 dist-1.0);
 742 }
 743
 744 static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
 745                               int srcW, int dstW, int filterAlign, int one, int flags,
 746                               SwsVector *srcFilter, SwsVector *dstFilter)
 747 {
 748         int i;
 749         int filterSize;
 750         int filter2Size;
 751         int minFilterSize;
 752         double *filter=NULL;
 753         double *filter2=NULL;
 754 #ifdef ARCH_X86
 755         if(flags & SWS_CPU_CAPS_MMX)
 756                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
 757 #endif
 758
 759         // Note the +1 is for the MMXscaler which reads over the end
 760         *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
 761
 762         if(ABS(xInc - 0x10000) <10) // unscaled
 763         {
 764                 int i;
 765                 filterSize= 1;
 766                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
 767                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
 768
 769                 for(i=0; i<dstW; i++)
 770                 {
 771                         filter[i*filterSize]=1;
 772                         (*filterPos)[i]=i;
 773                 }
 774
 775         }
 776         else if(flags&SWS_POINT) // lame looking point sampling mode
 777         {
 778                 int i;
 779                 int xDstInSrc;
 780                 filterSize= 1;
 781                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
 782
 783                 xDstInSrc= xInc/2 - 0x8000;
 784                 for(i=0; i<dstW; i++)
 785                 {
 786                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
 787
 788                         (*filterPos)[i]= xx;
 789                         filter[i]= 1.0;
 790                         xDstInSrc+= xInc;
 791                 }
 792         }
 793         else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
 794         {
 795                 int i;
 796                 int xDstInSrc;
 797                 if     (flags&SWS_BICUBIC) filterSize= 4;
 798                 else if(flags&SWS_X      ) filterSize= 4;
 799                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA
 800                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
 801
 802                 xDstInSrc= xInc/2 - 0x8000;
 803                 for(i=0; i<dstW; i++)
 804                 {
 805                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
 806                         int j;
 807
 808                         (*filterPos)[i]= xx;
 809                                 //Bilinear upscale / linear interpolate / Area averaging
 810                                 for(j=0; j<filterSize; j++)
 811                                 {
 812                                         double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
 813                                         double coeff= 1.0 - d;
 814                                         if(coeff<0) coeff=0;
 815                                         filter[i*filterSize + j]= coeff;
 816                                         xx++;
 817                                 }
 818                         xDstInSrc+= xInc;
 819                 }
 820         }
 821         else
 822         {
 823                 double xDstInSrc;
 824                 double sizeFactor, filterSizeInSrc;
 825                 const double xInc1= (double)xInc / (double)(1<<16);
 826                 int param= (flags&SWS_PARAM_MASK)>>SWS_PARAM_SHIFT;
 827
 828                 if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
 829                 else if(flags&SWS_X)            sizeFactor= 8.0;
 830                 else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
 831                 else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
 832                 else if(flags&SWS_LANCZOS)      sizeFactor= param ? 2.0*param : 6.0;
 833                 else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
 834                 else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
 835                 else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
 836                 else {
 837                         sizeFactor= 0.0; //GCC warning killer
 838                         ASSERT(0)
 839                 }
 840
 841                 if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
 842                 else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
 843
 844                 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
 845                 if(filterSize > srcW-2) filterSize=srcW-2;
 846
 847                 filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
 848
 849                 xDstInSrc= xInc1 / 2.0 - 0.5;
 850                 for(i=0; i<dstW; i++)
 851                 {
 852                         int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
 853                         int j;
 854                         (*filterPos)[i]= xx;
 855                         for(j=0; j<filterSize; j++)
 856                         {
 857                                 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
 858                                 double coeff;
 859                                 if(flags & SWS_BICUBIC)
 860                                 {
 861                                         double A= param ? -param*0.01 : -0.60;
 862
 863                                         // Equation is from VirtualDub
 864                                         if(d<1.0)
 865                                                 coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
 866                                         else if(d<2.0)
 867                                                 coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
 868                                         else
 869                                                 coeff=0.0;
 870                                 }
 871 /*                              else if(flags & SWS_X)
 872                                 {
 873                                         double p= param ? param*0.01 : 0.3;
 874                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
 875                                         coeff*= pow(2.0, - p*d*d);
 876                                 }*/
 877                                 else if(flags & SWS_X)
 878                                 {
 879                                         double A= param ? param*0.1 : 1.0;
 880
 881                                         if(d<1.0)
 882                                                 coeff = cos(d*PI);
 883                                         else
 884                                                 coeff=-1.0;
 885                                         if(coeff<0.0)   coeff= -pow(-coeff, A);
 886                                         else            coeff=  pow( coeff, A);
 887                                         coeff= coeff*0.5 + 0.5;
 888                                 }
 889                                 else if(flags & SWS_AREA)
 890                                 {
 891                                         double srcPixelSize= 1.0/xInc1;
 892                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
 893                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
 894                                         else coeff=0.0;
 895                                 }
 896                                 else if(flags & SWS_GAUSS)
 897                                 {
 898                                         double p= param ? param*0.1 : 3.0;
 899                                         coeff = pow(2.0, - p*d*d);
 900                                 }
 901                                 else if(flags & SWS_SINC)
 902                                 {
 903                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
 904                                 }
 905                                 else if(flags & SWS_LANCZOS)
 906                                 {
 907                                         double p= param ? param : 3.0;
 908                                         coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
 909                                         if(d>p) coeff=0;
 910                                 }
 911                                 else if(flags & SWS_BILINEAR)
 912                                 {
 913                                         coeff= 1.0 - d;
 914                                         if(coeff<0) coeff=0;
 915                                 }
 916                                 else if(flags & SWS_SPLINE)
 917                                 {
 918                                         double p=-2.196152422706632;
 919                                         coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
 920                                 }
 921                                 else {
 922                                         coeff= 0.0; //GCC warning killer
 923                                         ASSERT(0)
 924                                 }
 925
 926                                 filter[i*filterSize + j]= coeff;
 927                                 xx++;
 928                         }
 929                         xDstInSrc+= xInc1;
 930                 }
 931         }
 932
 933         /* apply src & dst Filter to filter -> filter2
 934            free(filter);
 935         */
 936         ASSERT(filterSize>0)
 937         filter2Size= filterSize;
 938         if(srcFilter) filter2Size+= srcFilter->length - 1;
 939         if(dstFilter) filter2Size+= dstFilter->length - 1;
 940         ASSERT(filter2Size>0)
 941         filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
 942
 943         for(i=0; i<dstW; i++)
 944         {
 945                 int j;
 946                 SwsVector scaleFilter;
 947                 SwsVector *outVec;
 948
 949                 scaleFilter.coeff= filter + i*filterSize;
 950                 scaleFilter.length= filterSize;
 951
 952                 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
 953                 else          outVec= &scaleFilter;
 954
 955                 ASSERT(outVec->length == filter2Size)
 956                 //FIXME dstFilter
 957
 958                 for(j=0; j<outVec->length; j++)
 959                 {
 960                         filter2[i*filter2Size + j]= outVec->coeff[j];
 961                 }
 962
 963                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
 964
 965                 if(outVec != &scaleFilter) sws_freeVec(outVec);
 966         }
 967         free(filter); filter=NULL;
 968
 969         /* try to reduce the filter-size (step1 find size and shift left) */
 970         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
 971         minFilterSize= 0;
 972         for(i=dstW-1; i>=0; i--)
 973         {
 974                 int min= filter2Size;
 975                 int j;
 976                 double cutOff=0.0;
 977
 978                 /* get rid off near zero elements on the left by shifting left */
 979                 for(j=0; j<filter2Size; j++)
 980                 {
 981                         int k;
 982                         cutOff += ABS(filter2[i*filter2Size]);
 983
 984                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
 985
 986                         /* preserve Monotonicity because the core cant handle the filter otherwise */
 987                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
 988
 989                         // Move filter coeffs left
 990                         for(k=1; k<filter2Size; k++)
 991                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
 992                         filter2[i*filter2Size + k - 1]= 0.0;
 993                         (*filterPos)[i]++;
 994                 }
 995
 996                 cutOff=0.0;
 997                 /* count near zeros on the right */
 998                 for(j=filter2Size-1; j>0; j--)
 999                 {
1000                         cutOff += ABS(filter2[i*filter2Size + j]);
1001
1002                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1003                         min--;
1004                 }
1005
1006                 if(min>minFilterSize) minFilterSize= min;
1007         }
1008
1009         ASSERT(minFilterSize > 0)
1010         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1011         ASSERT(filterSize > 0)
1012         filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1013         *outFilterSize= filterSize;
1014
1015         if(flags&SWS_PRINT_INFO)
1016                 MSG_INFO("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1017         /* try to reduce the filter-size (step2 reduce it) */
1018         for(i=0; i<dstW; i++)
1019         {
1020                 int j;
1021
1022                 for(j=0; j<filterSize; j++)
1023                 {
1024                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1025                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1026                 }
1027         }
1028         free(filter2); filter2=NULL;
1029
1030
1031         //FIXME try to align filterpos if possible
1032
1033         //fix borders
1034         for(i=0; i<dstW; i++)
1035         {
1036                 int j;
1037                 if((*filterPos)[i] < 0)
1038                 {
1039                         // Move filter coeffs left to compensate for filterPos
1040                         for(j=1; j<filterSize; j++)
1041                         {
1042                                 int left= MAX(j + (*filterPos)[i], 0);
1043                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1044                                 filter[i*filterSize + j]=0;
1045                         }
1046                         (*filterPos)[i]= 0;
1047                 }
1048
1049                 if((*filterPos)[i] + filterSize > srcW)
1050                 {
1051                         int shift= (*filterPos)[i] + filterSize - srcW;
1052                         // Move filter coeffs right to compensate for filterPos
1053                         for(j=filterSize-2; j>=0; j--)
1054                         {
1055                                 int right= MIN(j + shift, filterSize-1);
1056                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1057                                 filter[i*filterSize +j]=0;
1058                         }
1059                         (*filterPos)[i]= srcW - filterSize;
1060                 }
1061         }
1062
1063         // Note the +1 is for the MMXscaler which reads over the end
1064         *outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));
1065         memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1066
1067         /* Normalize & Store in outFilter */
1068         for(i=0; i<dstW; i++)
1069         {
1070                 int j;
1071                 double sum=0;
1072                 double scale= one;
1073                 for(j=0; j<filterSize; j++)
1074                 {
1075                         sum+= filter[i*filterSize + j];
1076                 }
1077                 scale/= sum;
1078                 for(j=0; j<*outFilterSize; j++)
1079                 {
1080                         (*outFilter)[i*(*outFilterSize) + j]= (int)(filter[i*filterSize + j]*scale);
1081                 }
1082         }
1083
1084         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1085         for(i=0; i<*outFilterSize; i++)
1086         {
1087                 int j= dstW*(*outFilterSize);
1088                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1089         }
1090
1091         free(filter);
1092 }
1093
1094 #ifdef ARCH_X86
1095 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1096 {
1097         uint8_t *fragmentA;
1098         int imm8OfPShufW1A;
1099         int imm8OfPShufW2A;
1100         int fragmentLengthA;
1101         uint8_t *fragmentB;
1102         int imm8OfPShufW1B;
1103         int imm8OfPShufW2B;
1104         int fragmentLengthB;
1105         int fragmentPos;
1106
1107         int xpos, i;
1108
1109         // create an optimized horizontal scaling routine
1110
1111         //code fragment
1112
1113         asm volatile(
1114                 "jmp 9f                         \n\t"
1115         // Begin
1116                 "0:                             \n\t"
1117                 "movq (%%edx, %%eax), %%mm3     \n\t"
1118                 "movd (%%ecx, %%esi), %%mm0     \n\t"
1119                 "movd 1(%%ecx, %%esi), %%mm1    \n\t"
1120                 "punpcklbw %%mm7, %%mm1         \n\t"
1121                 "punpcklbw %%mm7, %%mm0         \n\t"
1122                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1123                 "1:                             \n\t"
1124                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1125                 "2:                             \n\t"
1126                 "psubw %%mm1, %%mm0             \n\t"
1127                 "movl 8(%%ebx, %%eax), %%esi    \n\t"
1128                 "pmullw %%mm3, %%mm0            \n\t"
1129                 "psllw $7, %%mm1                \n\t"
1130                 "paddw %%mm1, %%mm0             \n\t"
1131
1132                 "movq %%mm0, (%%edi, %%eax)     \n\t"
1133
1134                 "addl $8, %%eax                 \n\t"
1135         // End
1136                 "9:                             \n\t"
1137 //              "int $3\n\t"
1138                 "leal 0b, %0                    \n\t"
1139                 "leal 1b, %1                    \n\t"
1140                 "leal 2b, %2                    \n\t"
1141                 "decl %1                        \n\t"
1142                 "decl %2                        \n\t"
1143                 "subl %0, %1                    \n\t"
1144                 "subl %0, %2                    \n\t"
1145                 "leal 9b, %3                    \n\t"
1146                 "subl %0, %3                    \n\t"
1147
1148
1149                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1150                 "=r" (fragmentLengthA)
1151         );
1152
1153         asm volatile(
1154                 "jmp 9f                         \n\t"
1155         // Begin
1156                 "0:                             \n\t"
1157                 "movq (%%edx, %%eax), %%mm3     \n\t"
1158                 "movd (%%ecx, %%esi), %%mm0     \n\t"
1159                 "punpcklbw %%mm7, %%mm0         \n\t"
1160                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1161                 "1:                             \n\t"
1162                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1163                 "2:                             \n\t"
1164                 "psubw %%mm1, %%mm0             \n\t"
1165                 "movl 8(%%ebx, %%eax), %%esi    \n\t"
1166                 "pmullw %%mm3, %%mm0            \n\t"
1167                 "psllw $7, %%mm1                \n\t"
1168                 "paddw %%mm1, %%mm0             \n\t"
1169
1170                 "movq %%mm0, (%%edi, %%eax)     \n\t"
1171
1172                 "addl $8, %%eax                 \n\t"
1173         // End
1174                 "9:                             \n\t"
1175 //              "int $3\n\t"
1176                 "leal 0b, %0                    \n\t"
1177                 "leal 1b, %1                    \n\t"
1178                 "leal 2b, %2                    \n\t"
1179                 "decl %1                        \n\t"
1180                 "decl %2                        \n\t"
1181                 "subl %0, %1                    \n\t"
1182                 "subl %0, %2                    \n\t"
1183                 "leal 9b, %3                    \n\t"
1184                 "subl %0, %3                    \n\t"
1185
1186
1187                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1188                 "=r" (fragmentLengthB)
1189         );
1190
1191         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1192         fragmentPos=0;
1193
1194         for(i=0; i<dstW/numSplits; i++)
1195         {
1196                 int xx=xpos>>16;
1197
1198                 if((i&3) == 0)
1199                 {
1200                         int a=0;
1201                         int b=((xpos+xInc)>>16) - xx;
1202                         int c=((xpos+xInc*2)>>16) - xx;
1203                         int d=((xpos+xInc*3)>>16) - xx;
1204
1205                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1206                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1207                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1208                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1209                         filterPos[i/2]= xx;
1210
1211                         if(d+1<4)
1212                         {
1213                                 int maxShift= 3-(d+1);
1214                                 int shift=0;
1215
1216                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1217
1218                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1219                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1220                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1221                                         a | (b<<2) | (c<<4) | (d<<6);
1222
1223                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1224                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1225
1226                                 if(shift && i>=shift)
1227                                 {
1228                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1229                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1230                                         filterPos[i/2]-=shift;
1231                                 }
1232
1233                                 fragmentPos+= fragmentLengthB;
1234                         }
1235                         else
1236                         {
1237                                 int maxShift= 3-d;
1238                                 int shift=0;
1239
1240                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1241
1242                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1243                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1244                                         a | (b<<2) | (c<<4) | (d<<6);
1245
1246                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1247                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1248
1249                                 if(shift && i>=shift)
1250                                 {
1251                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1252                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1253                                         filterPos[i/2]-=shift;
1254                                 }
1255
1256                                 fragmentPos+= fragmentLengthA;
1257                         }
1258
1259                         funnyCode[fragmentPos]= RET;
1260                 }
1261                 xpos+=xInc;
1262         }
1263         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1264 }
1265 #endif // ARCH_X86
1266
1267 static void globalInit(){
1268     // generating tables:
1269     int i;
1270     for(i=0; i<768; i++){
1271         int c= MIN(MAX(i-256, 0), 255);
1272         clip_table[i]=c;
1273     }
1274 }
1275
1276 static SwsFunc getSwsFunc(int flags){
1277
1278 #ifdef RUNTIME_CPUDETECT
1279 #ifdef ARCH_X86
1280         // ordered per speed fasterst first
1281         if(flags & SWS_CPU_CAPS_MMX2)
1282                 return swScale_MMX2;
1283         else if(flags & SWS_CPU_CAPS_3DNOW)
1284                 return swScale_3DNow;
1285         else if(flags & SWS_CPU_CAPS_MMX)
1286                 return swScale_MMX;
1287         else
1288                 return swScale_C;
1289
1290 #else
1291         return swScale_C;
1292 #endif
1293 #else //RUNTIME_CPUDETECT
1294 #ifdef HAVE_MMX2
1295         return swScale_MMX2;
1296 #elif defined (HAVE_3DNOW)
1297         return swScale_3DNow;
1298 #elif defined (HAVE_MMX)
1299         return swScale_MMX;
1300 #else
1301         return swScale_C;
1302 #endif
1303 #endif //!RUNTIME_CPUDETECT
1304 }
1305
1306 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1307              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1308         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1309         /* Copy Y plane */
1310         if(dstStride[0]==srcStride[0])
1311                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1312         else
1313         {
1314                 int i;
1315                 uint8_t *srcPtr= src[0];
1316                 uint8_t *dstPtr= dst;
1317                 for(i=0; i<srcSliceH; i++)
1318                 {
1319                         memcpy(dstPtr, srcPtr, srcStride[0]);
1320                         srcPtr+= srcStride[0];
1321                         dstPtr+= dstStride[0];
1322                 }
1323         }
1324         dst = dstParam[1] + dstStride[1]*srcSliceY;
1325         interleaveBytes( src[1],src[2],dst,c->srcW,srcSliceH,srcStride[1],srcStride[2],dstStride[0] );
1326
1327         return srcSliceH;
1328 }
1329
1330 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1331              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1332         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1333
1334         yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1335
1336         return srcSliceH;
1337 }
1338
1339 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1340 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1341                            int srcSliceH, uint8_t* dst[], int dstStride[]){
1342         const int srcFormat= c->srcFormat;
1343         const int dstFormat= c->dstFormat;
1344         const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1345         const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1346         const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8
1347         const int dstId= (dstFormat&0xFF)>>2;
1348         void (*conv)(const uint8_t *src, uint8_t *dst, unsigned src_size)=NULL;
1349
1350         /* BGR -> BGR */
1351         if(   (isBGR(srcFormat) && isBGR(dstFormat))
1352            || (isRGB(srcFormat) && isRGB(dstFormat))){
1353                 switch(srcId | (dstId<<4)){
1354                 case 0x34: conv= rgb16to15; break;
1355                 case 0x36: conv= rgb24to15; break;
1356                 case 0x38: conv= rgb32to15; break;
1357                 case 0x43: conv= rgb15to16; break;
1358                 case 0x46: conv= rgb24to16; break;
1359                 case 0x48: conv= rgb32to16; break;
1360                 case 0x63: conv= rgb15to24; break;
1361                 case 0x64: conv= rgb16to24; break;
1362                 case 0x68: conv= rgb32to24; break;
1363                 case 0x83: conv= rgb15to32; break;
1364                 case 0x84: conv= rgb16to32; break;
1365                 case 0x86: conv= rgb24to32; break;
1366                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1367                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1368                 }
1369         }else if(   (isBGR(srcFormat) && isRGB(dstFormat))
1370                  || (isRGB(srcFormat) && isBGR(dstFormat))){
1371                 switch(srcId | (dstId<<4)){
1372                 case 0x33: conv= rgb15tobgr15; break;
1373                 case 0x34: conv= rgb16tobgr15; break;
1374                 case 0x36: conv= rgb24tobgr15; break;
1375                 case 0x38: conv= rgb32tobgr15; break;
1376                 case 0x43: conv= rgb15tobgr16; break;
1377                 case 0x44: conv= rgb16tobgr16; break;
1378                 case 0x46: conv= rgb24tobgr16; break;
1379                 case 0x48: conv= rgb32tobgr16; break;
1380                 case 0x63: conv= rgb15tobgr24; break;
1381                 case 0x64: conv= rgb16tobgr24; break;
1382                 case 0x66: conv= rgb24tobgr24; break;
1383                 case 0x68: conv= rgb32tobgr24; break;
1384                 case 0x83: conv= rgb15tobgr32; break;
1385                 case 0x84: conv= rgb16tobgr32; break;
1386                 case 0x86: conv= rgb24tobgr32; break;
1387                 case 0x88: conv= rgb32tobgr32; break;
1388                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1389                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1390                 }
1391         }else{
1392                 MSG_ERR("swScaler: internal error %s -> %s converter\n",
1393                          vo_format_name(srcFormat), vo_format_name(dstFormat));
1394         }
1395
1396         if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1397                 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1398         else
1399         {
1400                 int i;
1401                 uint8_t *srcPtr= src[0];
1402                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1403
1404                 for(i=0; i<srcSliceH; i++)
1405                 {
1406                         conv(srcPtr, dstPtr, c->srcW*srcBpp);
1407                         srcPtr+= srcStride[0];
1408                         dstPtr+= dstStride[0];
1409                 }
1410         }
1411         return srcSliceH;
1412 }
1413
1414 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1415              int srcSliceH, uint8_t* dst[], int dstStride[]){
1416
1417         rgb24toyv12(
1418                 src[0],
1419                 dst[0]+ srcSliceY    *dstStride[0],
1420                 dst[1]+(srcSliceY>>1)*dstStride[1],
1421                 dst[2]+(srcSliceY>>1)*dstStride[2],
1422                 c->srcW, srcSliceH,
1423                 dstStride[0], dstStride[1], srcStride[0]);
1424         return srcSliceH;
1425 }
1426
1427 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1428              int srcSliceH, uint8_t* dst[], int dstStride[]){
1429         int i;
1430
1431         /* copy Y */
1432         if(srcStride[0]==dstStride[0])
1433                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1434         else{
1435                 uint8_t *srcPtr= src[0];
1436                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1437
1438                 for(i=0; i<srcSliceH; i++)
1439                 {
1440                         memcpy(dstPtr, srcPtr, c->srcW);
1441                         srcPtr+= srcStride[0];
1442                         dstPtr+= dstStride[0];
1443                 }
1444         }
1445
1446         if(c->dstFormat==IMGFMT_YV12){
1447                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1448                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1449         }else{
1450                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1451                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1452         }
1453         return srcSliceH;
1454 }
1455
1456 /**
1457  * bring pointers in YUV order instead of YVU
1458  */
1459 static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1460         if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
1461            || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
1462                 sortedP[0]= p[0];
1463                 sortedP[1]= p[2];
1464                 sortedP[2]= p[1];
1465                 sortedStride[0]= stride[0];
1466                 sortedStride[1]= stride[2];
1467                 sortedStride[2]= stride[1];
1468         }
1469         else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8)
1470         {
1471                 sortedP[0]= p[0];
1472                 sortedP[1]=
1473                 sortedP[2]= NULL;
1474                 sortedStride[0]= stride[0];
1475                 sortedStride[1]=
1476                 sortedStride[2]= 0;
1477         }
1478         else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)
1479         {
1480                 sortedP[0]= p[0];
1481                 sortedP[1]= p[1];
1482                 sortedP[2]= p[2];
1483                 sortedStride[0]= stride[0];
1484                 sortedStride[1]= stride[1];
1485                 sortedStride[2]= stride[2];
1486         }else{
1487                 MSG_ERR("internal error in orderYUV\n");
1488         }
1489 }
1490
1491 /* unscaled copy like stuff (assumes nearly identical formats) */
1492 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1493              int srcSliceH, uint8_t* dst[], int dstStride[]){
1494
1495         if(isPacked(c->srcFormat))
1496         {
1497                 if(dstStride[0]==srcStride[0])
1498                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1499                 else
1500                 {
1501                         int i;
1502                         uint8_t *srcPtr= src[0];
1503                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1504                         int length=0;
1505
1506                         /* universal length finder */
1507                         while(length+c->srcW <= ABS(dstStride[0])
1508                            && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
1509                         ASSERT(length!=0);
1510
1511                         for(i=0; i<srcSliceH; i++)
1512                         {
1513                                 memcpy(dstPtr, srcPtr, length);
1514                                 srcPtr+= srcStride[0];
1515                                 dstPtr+= dstStride[0];
1516                         }
1517                 }
1518         }
1519         else
1520         { /* Planar YUV or gray */
1521                 int plane;
1522                 for(plane=0; plane<3; plane++)
1523                 {
1524                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1525                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1526                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1527
1528                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1529                         {
1530                                 if(!isGray(c->dstFormat))
1531                                         memset(dst[plane], 128, dstStride[plane]*height);
1532                         }
1533                         else
1534                         {
1535                                 if(dstStride[plane]==srcStride[plane])
1536                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1537                                 else
1538                                 {
1539                                         int i;
1540                                         uint8_t *srcPtr= src[plane];
1541                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1542                                         for(i=0; i<height; i++)
1543                                         {
1544                                                 memcpy(dstPtr, srcPtr, length);
1545                                                 srcPtr+= srcStride[plane];
1546                                                 dstPtr+= dstStride[plane];
1547                                         }
1548                                 }
1549                         }
1550                 }
1551         }
1552         return srcSliceH;
1553 }
1554
1555 static int remove_dup_fourcc(int fourcc)
1556 {
1557         switch(fourcc)
1558         {
1559             case IMGFMT_I420:
1560             case IMGFMT_IYUV: return IMGFMT_YV12;
1561             case IMGFMT_Y8  : return IMGFMT_Y800;
1562             case IMGFMT_IF09: return IMGFMT_YVU9;
1563             default: return fourcc;
1564         }
1565 }
1566
1567 static void getSubSampleFactors(int *h, int *v, int format){
1568         switch(format){
1569         case IMGFMT_UYVY:
1570         case IMGFMT_YUY2:
1571                 *h=1;
1572                 *v=0;
1573                 break;
1574         case IMGFMT_YV12:
1575         case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
1576                 *h=1;
1577                 *v=1;
1578                 break;
1579         case IMGFMT_YVU9:
1580                 *h=2;
1581                 *v=2;
1582                 break;
1583         case IMGFMT_444P:
1584                 *h=0;
1585                 *v=0;
1586                 break;
1587         case IMGFMT_422P:
1588                 *h=1;
1589                 *v=0;
1590                 break;
1591         case IMGFMT_411P:
1592                 *h=2;
1593                 *v=0;
1594                 break;
1595         default:
1596                 *h=0;
1597                 *v=0;
1598                 break;
1599         }
1600 }
1601
1602 static uint16_t roundToInt16(int64_t f){
1603         int r= (f + (1<<15))>>16;
1604              if(r<-0x7FFF) return 0x8000;
1605         else if(r> 0x7FFF) return 0x7FFF;
1606         else               return r;
1607 }
1608
1609 /**
1610  * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1611  * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1612  * @return -1 if not supported
1613  */
1614 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1615         int64_t crv =  inv_table[0];
1616         int64_t cbu =  inv_table[1];
1617         int64_t cgu = -inv_table[2];
1618         int64_t cgv = -inv_table[3];
1619         int64_t cy  = 1<<16;
1620         int64_t oy  = 0;
1621
1622         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1623         memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1624         memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
1625
1626         c->brightness= brightness;
1627         c->contrast  = contrast;
1628         c->saturation= saturation;
1629         c->srcRange  = srcRange;
1630         c->dstRange  = dstRange;
1631
1632         c->uOffset=   0x0400040004000400LL;
1633         c->vOffset=   0x0400040004000400LL;
1634
1635         if(!srcRange){
1636                 cy= (cy*255) / 219;
1637                 oy= 16<<16;
1638         }
1639
1640         cy = (cy *contrast             )>>16;
1641         crv= (crv*contrast * saturation)>>32;
1642         cbu= (cbu*contrast * saturation)>>32;
1643         cgu= (cgu*contrast * saturation)>>32;
1644         cgv= (cgv*contrast * saturation)>>32;
1645
1646         oy -= 256*brightness;
1647
1648         c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
1649         c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
1650         c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1651         c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1652         c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1653         c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
1654
1655         yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1656         //FIXME factorize
1657
1658         return 0;
1659 }
1660
1661 /**
1662  * @return -1 if not supported
1663  */
1664 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1665         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1666
1667         *inv_table = c->srcColorspaceTable;
1668         *table     = c->dstColorspaceTable;
1669         *srcRange  = c->srcRange;
1670         *dstRange  = c->dstRange;
1671         *brightness= c->brightness;
1672         *contrast  = c->contrast;
1673         *saturation= c->saturation;
1674
1675         return 0;
1676 }
1677
1678 SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,
1679                          SwsFilter *srcFilter, SwsFilter *dstFilter){
1680
1681         SwsContext *c;
1682         int i;
1683         int usesFilter;
1684         int unscaled, needsDither;
1685         int srcFormat, dstFormat;
1686         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1687 #ifdef ARCH_X86
1688         if(flags & SWS_CPU_CAPS_MMX)
1689                 asm volatile("emms\n\t"::: "memory");
1690 #endif
1691
1692 #ifndef RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
1693         flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW);
1694 #ifdef HAVE_MMX2
1695         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1696 #elif defined (HAVE_3DNOW)
1697         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1698 #elif defined (HAVE_MMX)
1699         flags |= SWS_CPU_CAPS_MMX;
1700 #endif
1701 #endif
1702         if(clip_table[512] != 255) globalInit();
1703         if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1704
1705         /* avoid dupplicate Formats, so we dont need to check to much */
1706         srcFormat = remove_dup_fourcc(origSrcFormat);
1707         dstFormat = remove_dup_fourcc(origDstFormat);
1708
1709         unscaled = (srcW == dstW && srcH == dstH);
1710         needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
1711                      && (dstFormat&0xFF)<24
1712                      && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1713
1714         if(!isSupportedIn(srcFormat))
1715         {
1716                 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1717                 return NULL;
1718         }
1719         if(!isSupportedOut(dstFormat))
1720         {
1721                 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1722                 return NULL;
1723         }
1724
1725         /* sanity check */
1726         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
1727         {
1728                  MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
1729                         srcW, srcH, dstW, dstH);
1730                 return NULL;
1731         }
1732
1733         if(!dstFilter) dstFilter= &dummyFilter;
1734         if(!srcFilter) srcFilter= &dummyFilter;
1735
1736         c= memalign(64, sizeof(SwsContext));
1737         memset(c, 0, sizeof(SwsContext));
1738
1739         c->srcW= srcW;
1740         c->srcH= srcH;
1741         c->dstW= dstW;
1742         c->dstH= dstH;
1743         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1744         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1745         c->flags= flags;
1746         c->dstFormat= dstFormat;
1747         c->srcFormat= srcFormat;
1748         c->origDstFormat= origDstFormat;
1749         c->origSrcFormat= origSrcFormat;
1750
1751         usesFilter=0;
1752         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesFilter=1;
1753         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesFilter=1;
1754         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesFilter=1;
1755         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesFilter=1;
1756         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesFilter=1;
1757         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesFilter=1;
1758         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesFilter=1;
1759         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesFilter=1;
1760
1761         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1762         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1763
1764         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1765         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1766
1767         // drop some chroma lines if the user wants it
1768         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1769         c->chrSrcVSubSample+= c->vChrDrop;
1770
1771         // drop every 2. pixel for chroma calculation unless user wants full chroma
1772         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
1773                 c->chrSrcHSubSample=1;
1774
1775         c->chrIntHSubSample= c->chrDstHSubSample;
1776         c->chrIntVSubSample= c->chrSrcVSubSample;
1777
1778         // note the -((-x)>>y) is so that we allways round toward +inf
1779         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1780         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
1781         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
1782         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
1783
1784         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);
1785
1786         /* unscaled special Cases */
1787         if(unscaled && !usesFilter)
1788         {
1789                 /* yv12_to_nv12 */
1790                 if(srcFormat == IMGFMT_YV12 && dstFormat == IMGFMT_NV12)
1791                 {
1792                         c->swScale= PlanarToNV12Wrapper;
1793                 }
1794                 /* yuv2bgr */
1795                 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
1796                 {
1797                         c->swScale= yuv2rgb_get_func_ptr(c);
1798                 }
1799
1800                 if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 )
1801                 {
1802                         c->swScale= yvu9toyv12Wrapper;
1803                 }
1804
1805                 /* bgr24toYV12 */
1806                 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
1807                         c->swScale= bgr24toyv12Wrapper;
1808
1809                 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
1810                 if(   (isBGR(srcFormat) || isRGB(srcFormat))
1811                    && (isBGR(dstFormat) || isRGB(dstFormat))
1812                    && !needsDither)
1813                         c->swScale= rgb2rgbWrapper;
1814
1815                 /* LQ converters if -sws 0 or -sws 4*/
1816                 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
1817                         /* rgb/bgr -> rgb/bgr (dither needed forms) */
1818                         if(  (isBGR(srcFormat) || isRGB(srcFormat))
1819                           && (isBGR(dstFormat) || isRGB(dstFormat))
1820                           && needsDither)
1821                                 c->swScale= rgb2rgbWrapper;
1822
1823                         /* yv12_to_yuy2 */
1824                         if(srcFormat == IMGFMT_YV12 && dstFormat == IMGFMT_YUY2)
1825                         {
1826                                 c->swScale= PlanarToYuy2Wrapper;
1827                         }
1828                 }
1829
1830                 /* simple copy */
1831                 if(   srcFormat == dstFormat
1832                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
1833                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
1834                   )
1835                 {
1836                         c->swScale= simpleCopy;
1837                 }
1838
1839                 if(c->swScale){
1840                         if(flags&SWS_PRINT_INFO)
1841                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
1842                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
1843                         return c;
1844                 }
1845         }
1846
1847         if(flags & SWS_CPU_CAPS_MMX2)
1848         {
1849                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
1850                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
1851                 {
1852                         if(flags&SWS_PRINT_INFO)
1853                                 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
1854                 }
1855         }
1856         else
1857                 c->canMMX2BeUsed=0;
1858
1859         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
1860         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
1861
1862         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
1863         // but only for the FAST_BILINEAR mode otherwise do correct scaling
1864         // n-2 is the last chrominance sample available
1865         // this is not perfect, but noone shuld notice the difference, the more correct variant
1866         // would be like the vertical one, but that would require some special code for the
1867         // first and last pixel
1868         if(flags&SWS_FAST_BILINEAR)
1869         {
1870                 if(c->canMMX2BeUsed)
1871                 {
1872                         c->lumXInc+= 20;
1873                         c->chrXInc+= 20;
1874                 }
1875                 //we dont use the x86asm scaler if mmx is available
1876                 else if(flags & SWS_CPU_CAPS_MMX)
1877                 {
1878                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
1879                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
1880                 }
1881         }
1882
1883         /* precalculate horizontal scaler filter coefficients */
1884         {
1885                 const int filterAlign= (flags & SWS_CPU_CAPS_MMX) ? 4 : 1;
1886
1887                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
1888                                  srcW      ,       dstW, filterAlign, 1<<14,
1889                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
1890                                  srcFilter->lumH, dstFilter->lumH);
1891                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
1892                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
1893                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
1894                                  srcFilter->chrH, dstFilter->chrH);
1895
1896 #ifdef ARCH_X86
1897 // cant downscale !!!
1898                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
1899                 {
1900                         c->lumMmx2Filter   = (int16_t*)memalign(8, (dstW        /8+8)*sizeof(int16_t));
1901                         c->chrMmx2Filter   = (int16_t*)memalign(8, (c->chrDstW  /4+8)*sizeof(int16_t));
1902                         c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW      /2/8+8)*sizeof(int32_t));
1903                         c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
1904
1905                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
1906                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
1907                 }
1908 #endif
1909         } // Init Horizontal stuff
1910
1911
1912
1913         /* precalculate vertical scaler filter coefficients */
1914         initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
1915                         srcH      ,        dstH, 1, (1<<12)-4,
1916                         (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
1917                         srcFilter->lumV, dstFilter->lumV);
1918         initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
1919                         c->chrSrcH, c->chrDstH, 1, (1<<12)-4,
1920                         (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
1921                         srcFilter->chrV, dstFilter->chrV);
1922
1923         // Calculate Buffer Sizes so that they wont run out while handling these damn slices
1924         c->vLumBufSize= c->vLumFilterSize;
1925         c->vChrBufSize= c->vChrFilterSize;
1926         for(i=0; i<dstH; i++)
1927         {
1928                 int chrI= i*c->chrDstH / dstH;
1929                 int nextSlice= MAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
1930                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
1931                 nextSlice&= ~3; // Slices start at boundaries which are divisable through 4
1932                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
1933                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
1934                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1935                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1936         }
1937
1938         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1939         c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
1940         c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
1941         //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)
1942         for(i=0; i<c->vLumBufSize; i++)
1943                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
1944         for(i=0; i<c->vChrBufSize; i++)
1945                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
1946
1947         //try to avoid drawing green stuff between the right end and the stride end
1948         for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
1949         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
1950
1951         ASSERT(c->chrDstH <= dstH)
1952
1953         if(flags&SWS_PRINT_INFO)
1954         {
1955 #ifdef DITHER1XBPP
1956                 char *dither= " dithered";
1957 #else
1958                 char *dither= "";
1959 #endif
1960                 if(flags&SWS_FAST_BILINEAR)
1961                         MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
1962                 else if(flags&SWS_BILINEAR)
1963                         MSG_INFO("\nSwScaler: BILINEAR scaler, ");
1964                 else if(flags&SWS_BICUBIC)
1965                         MSG_INFO("\nSwScaler: BICUBIC scaler, ");
1966                 else if(flags&SWS_X)
1967                         MSG_INFO("\nSwScaler: Experimental scaler, ");
1968                 else if(flags&SWS_POINT)
1969                         MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
1970                 else if(flags&SWS_AREA)
1971                         MSG_INFO("\nSwScaler: Area Averageing scaler, ");
1972                 else if(flags&SWS_BICUBLIN)
1973                         MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
1974                 else if(flags&SWS_GAUSS)
1975                         MSG_INFO("\nSwScaler: Gaussian scaler, ");
1976                 else if(flags&SWS_SINC)
1977                         MSG_INFO("\nSwScaler: Sinc scaler, ");
1978                 else if(flags&SWS_LANCZOS)
1979                         MSG_INFO("\nSwScaler: Lanczos scaler, ");
1980                 else if(flags&SWS_SPLINE)
1981                         MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
1982                 else
1983                         MSG_INFO("\nSwScaler: ehh flags invalid?! ");
1984
1985                 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
1986                         MSG_INFO("from %s to%s %s ",
1987                                 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
1988                 else
1989                         MSG_INFO("from %s to %s ",
1990                                 vo_format_name(srcFormat), vo_format_name(dstFormat));
1991
1992                 if(flags & SWS_CPU_CAPS_MMX2)
1993                         MSG_INFO("using MMX2\n");
1994                 else if(flags & SWS_CPU_CAPS_3DNOW)
1995                         MSG_INFO("using 3DNOW\n");
1996                 else if(flags & SWS_CPU_CAPS_MMX)
1997                         MSG_INFO("using MMX\n");
1998                 else
1999                         MSG_INFO("using C\n");
2000         }
2001
2002         if(flags & SWS_PRINT_INFO)
2003         {
2004                 if(flags & SWS_CPU_CAPS_MMX)
2005                 {
2006                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2007                                 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2008                         else
2009                         {
2010                                 if(c->hLumFilterSize==4)
2011                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2012                                 else if(c->hLumFilterSize==8)
2013                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2014                                 else
2015                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2016
2017                                 if(c->hChrFilterSize==4)
2018                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2019                                 else if(c->hChrFilterSize==8)
2020                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2021                                 else
2022                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2023                         }
2024                 }
2025                 else
2026                 {
2027 #ifdef ARCH_X86
2028                         MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2029 #else
2030                         if(flags & SWS_FAST_BILINEAR)
2031                                 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2032                         else
2033                                 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2034 #endif
2035                 }
2036                 if(isPlanarYUV(dstFormat))
2037                 {
2038                         if(c->vLumFilterSize==1)
2039                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2040                         else
2041                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2042                 }
2043                 else
2044                 {
2045                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2046                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2047                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2048                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2049                                 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2050                         else
2051                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2052                 }
2053
2054                 if(dstFormat==IMGFMT_BGR24)
2055                         MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2056                                 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2057                 else if(dstFormat==IMGFMT_BGR32)
2058                         MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2059                 else if(dstFormat==IMGFMT_BGR16)
2060                         MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2061                 else if(dstFormat==IMGFMT_BGR15)
2062                         MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2063
2064                 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2065         }
2066         if(flags & SWS_PRINT_INFO)
2067         {
2068                 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2069                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2070                 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2071                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2072         }
2073
2074         c->swScale= getSwsFunc(flags);
2075         return c;
2076 }
2077
2078 /**
2079  * swscale warper, so we dont need to export the SwsContext.
2080  * assumes planar YUV to be in YUV order instead of YVU
2081  */
2082 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2083                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2084         return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2085 }
2086
2087 /**
2088  * swscale warper, so we dont need to export the SwsContext
2089  */
2090 int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
2091                            int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
2092         int srcStride[3];
2093         int dstStride[3];
2094         uint8_t *src[3];
2095         uint8_t *dst[3];
2096         sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam);
2097         sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam);
2098 //printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
2099
2100         return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2101 }
2102
2103 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2104                                 float lumaSharpen, float chromaSharpen,
2105                                 float chromaHShift, float chromaVShift,
2106                                 int verbose)
2107 {
2108         SwsFilter *filter= malloc(sizeof(SwsFilter));
2109
2110         if(lumaGBlur!=0.0){
2111                 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2112                 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2113         }else{
2114                 filter->lumH= sws_getIdentityVec();
2115                 filter->lumV= sws_getIdentityVec();
2116         }
2117
2118         if(chromaGBlur!=0.0){
2119                 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2120                 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2121         }else{
2122                 filter->chrH= sws_getIdentityVec();
2123                 filter->chrV= sws_getIdentityVec();
2124         }
2125
2126         if(chromaSharpen!=0.0){
2127                 SwsVector *g= sws_getConstVec(-1.0, 3);
2128                 SwsVector *id= sws_getConstVec(10.0/chromaSharpen, 1);
2129                 g->coeff[1]=2.0;
2130                 sws_addVec(id, g);
2131                 sws_convVec(filter->chrH, id);
2132                 sws_convVec(filter->chrV, id);
2133                 sws_freeVec(g);
2134                 sws_freeVec(id);
2135         }
2136
2137         if(lumaSharpen!=0.0){
2138                 SwsVector *g= sws_getConstVec(-1.0, 3);
2139                 SwsVector *id= sws_getConstVec(10.0/lumaSharpen, 1);
2140                 g->coeff[1]=2.0;
2141                 sws_addVec(id, g);
2142                 sws_convVec(filter->lumH, id);
2143                 sws_convVec(filter->lumV, id);
2144                 sws_freeVec(g);
2145                 sws_freeVec(id);
2146         }
2147
2148         if(chromaHShift != 0.0)
2149                 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2150
2151         if(chromaVShift != 0.0)
2152                 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2153
2154         sws_normalizeVec(filter->chrH, 1.0);
2155         sws_normalizeVec(filter->chrV, 1.0);
2156         sws_normalizeVec(filter->lumH, 1.0);
2157         sws_normalizeVec(filter->lumV, 1.0);
2158
2159         if(verbose) sws_printVec(filter->chrH);
2160         if(verbose) sws_printVec(filter->lumH);
2161
2162         return filter;
2163 }
2164
2165 /**
2166  * returns a normalized gaussian curve used to filter stuff
2167  * quality=3 is high quality, lowwer is lowwer quality
2168  */
2169 SwsVector *sws_getGaussianVec(double variance, double quality){
2170         const int length= (int)(variance*quality + 0.5) | 1;
2171         int i;
2172         double *coeff= memalign(sizeof(double), length*sizeof(double));
2173         double middle= (length-1)*0.5;
2174         SwsVector *vec= malloc(sizeof(SwsVector));
2175
2176         vec->coeff= coeff;
2177         vec->length= length;
2178
2179         for(i=0; i<length; i++)
2180         {
2181                 double dist= i-middle;
2182                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2183         }
2184
2185         sws_normalizeVec(vec, 1.0);
2186
2187         return vec;
2188 }
2189
2190 SwsVector *sws_getConstVec(double c, int length){
2191         int i;
2192         double *coeff= memalign(sizeof(double), length*sizeof(double));
2193         SwsVector *vec= malloc(sizeof(SwsVector));
2194
2195         vec->coeff= coeff;
2196         vec->length= length;
2197
2198         for(i=0; i<length; i++)
2199                 coeff[i]= c;
2200
2201         return vec;
2202 }
2203
2204
2205 SwsVector *sws_getIdentityVec(void){
2206         double *coeff= memalign(sizeof(double), sizeof(double));
2207         SwsVector *vec= malloc(sizeof(SwsVector));
2208         coeff[0]= 1.0;
2209
2210         vec->coeff= coeff;
2211         vec->length= 1;
2212
2213         return vec;
2214 }
2215
2216 void sws_normalizeVec(SwsVector *a, double height){
2217         int i;
2218         double sum=0;
2219         double inv;
2220
2221         for(i=0; i<a->length; i++)
2222                 sum+= a->coeff[i];
2223
2224         inv= height/sum;
2225
2226         for(i=0; i<a->length; i++)
2227                 a->coeff[i]*= inv;
2228 }
2229
2230 void sws_scaleVec(SwsVector *a, double scalar){
2231         int i;
2232
2233         for(i=0; i<a->length; i++)
2234                 a->coeff[i]*= scalar;
2235 }
2236
2237 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2238         int length= a->length + b->length - 1;
2239         double *coeff= memalign(sizeof(double), length*sizeof(double));
2240         int i, j;
2241         SwsVector *vec= malloc(sizeof(SwsVector));
2242
2243         vec->coeff= coeff;
2244         vec->length= length;
2245
2246         for(i=0; i<length; i++) coeff[i]= 0.0;
2247
2248         for(i=0; i<a->length; i++)
2249         {
2250                 for(j=0; j<b->length; j++)
2251                 {
2252                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2253                 }
2254         }
2255
2256         return vec;
2257 }
2258
2259 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2260         int length= MAX(a->length, b->length);
2261         double *coeff= memalign(sizeof(double), length*sizeof(double));
2262         int i;
2263         SwsVector *vec= malloc(sizeof(SwsVector));
2264
2265         vec->coeff= coeff;
2266         vec->length= length;
2267
2268         for(i=0; i<length; i++) coeff[i]= 0.0;
2269
2270         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2271         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2272
2273         return vec;
2274 }
2275
2276 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2277         int length= MAX(a->length, b->length);
2278         double *coeff= memalign(sizeof(double), length*sizeof(double));
2279         int i;
2280         SwsVector *vec= malloc(sizeof(SwsVector));
2281
2282         vec->coeff= coeff;
2283         vec->length= length;
2284
2285         for(i=0; i<length; i++) coeff[i]= 0.0;
2286
2287         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2288         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2289
2290         return vec;
2291 }
2292
2293 /* shift left / or right if "shift" is negative */
2294 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2295         int length= a->length + ABS(shift)*2;
2296         double *coeff= memalign(sizeof(double), length*sizeof(double));
2297         int i;
2298         SwsVector *vec= malloc(sizeof(SwsVector));
2299
2300         vec->coeff= coeff;
2301         vec->length= length;
2302
2303         for(i=0; i<length; i++) coeff[i]= 0.0;
2304
2305         for(i=0; i<a->length; i++)
2306         {
2307                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2308         }
2309
2310         return vec;
2311 }
2312
2313 void sws_shiftVec(SwsVector *a, int shift){
2314         SwsVector *shifted= sws_getShiftedVec(a, shift);
2315         free(a->coeff);
2316         a->coeff= shifted->coeff;
2317         a->length= shifted->length;
2318         free(shifted);
2319 }
2320
2321 void sws_addVec(SwsVector *a, SwsVector *b){
2322         SwsVector *sum= sws_sumVec(a, b);
2323         free(a->coeff);
2324         a->coeff= sum->coeff;
2325         a->length= sum->length;
2326         free(sum);
2327 }
2328
2329 void sws_subVec(SwsVector *a, SwsVector *b){
2330         SwsVector *diff= sws_diffVec(a, b);
2331         free(a->coeff);
2332         a->coeff= diff->coeff;
2333         a->length= diff->length;
2334         free(diff);
2335 }
2336
2337 void sws_convVec(SwsVector *a, SwsVector *b){
2338         SwsVector *conv= sws_getConvVec(a, b);
2339         free(a->coeff);
2340         a->coeff= conv->coeff;
2341         a->length= conv->length;
2342         free(conv);
2343 }
2344
2345 SwsVector *sws_cloneVec(SwsVector *a){
2346         double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2347         int i;
2348         SwsVector *vec= malloc(sizeof(SwsVector));
2349
2350         vec->coeff= coeff;
2351         vec->length= a->length;
2352
2353         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2354
2355         return vec;
2356 }
2357
2358 void sws_printVec(SwsVector *a){
2359         int i;
2360         double max=0;
2361         double min=0;
2362         double range;
2363
2364         for(i=0; i<a->length; i++)
2365                 if(a->coeff[i]>max) max= a->coeff[i];
2366
2367         for(i=0; i<a->length; i++)
2368                 if(a->coeff[i]<min) min= a->coeff[i];
2369
2370         range= max - min;
2371
2372         for(i=0; i<a->length; i++)
2373         {
2374                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2375                 MSG_DBG2("%1.3f ", a->coeff[i]);
2376                 for(;x>0; x--) MSG_DBG2(" ");
2377                 MSG_DBG2("|\n");
2378         }
2379 }
2380
2381 void sws_freeVec(SwsVector *a){
2382         if(!a) return;
2383         if(a->coeff) free(a->coeff);
2384         a->coeff=NULL;
2385         a->length=0;
2386         free(a);
2387 }
2388
2389 void sws_freeFilter(SwsFilter *filter){
2390         if(!filter) return;
2391
2392         if(filter->lumH) sws_freeVec(filter->lumH);
2393         if(filter->lumV) sws_freeVec(filter->lumV);
2394         if(filter->chrH) sws_freeVec(filter->chrH);
2395         if(filter->chrV) sws_freeVec(filter->chrV);
2396         free(filter);
2397 }
2398
2399
2400 void sws_freeContext(SwsContext *c){
2401         int i;
2402         if(!c) return;
2403
2404         if(c->lumPixBuf)
2405         {
2406                 for(i=0; i<c->vLumBufSize; i++)
2407                 {
2408                         if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2409                         c->lumPixBuf[i]=NULL;
2410                 }
2411                 free(c->lumPixBuf);
2412                 c->lumPixBuf=NULL;
2413         }
2414
2415         if(c->chrPixBuf)
2416         {
2417                 for(i=0; i<c->vChrBufSize; i++)
2418                 {
2419                         if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2420                         c->chrPixBuf[i]=NULL;
2421                 }
2422                 free(c->chrPixBuf);
2423                 c->chrPixBuf=NULL;
2424         }
2425
2426         if(c->vLumFilter) free(c->vLumFilter);
2427         c->vLumFilter = NULL;
2428         if(c->vChrFilter) free(c->vChrFilter);
2429         c->vChrFilter = NULL;
2430         if(c->hLumFilter) free(c->hLumFilter);
2431         c->hLumFilter = NULL;
2432         if(c->hChrFilter) free(c->hChrFilter);
2433         c->hChrFilter = NULL;
2434
2435         if(c->vLumFilterPos) free(c->vLumFilterPos);
2436         c->vLumFilterPos = NULL;
2437         if(c->vChrFilterPos) free(c->vChrFilterPos);
2438         c->vChrFilterPos = NULL;
2439         if(c->hLumFilterPos) free(c->hLumFilterPos);
2440         c->hLumFilterPos = NULL;
2441         if(c->hChrFilterPos) free(c->hChrFilterPos);
2442         c->hChrFilterPos = NULL;
2443
2444         if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2445         c->lumMmx2Filter=NULL;
2446         if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2447         c->chrMmx2Filter=NULL;
2448         if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2449         c->lumMmx2FilterPos=NULL;
2450         if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2451         c->chrMmx2FilterPos=NULL;
2452         if(c->yuvTable) free(c->yuvTable);
2453         c->yuvTable=NULL;
2454
2455         free(c);
2456 }
2457