git.sesse.net Git - vlc/blob - modules/video_filter/swscale/swscale.c

   1 /*
   2     Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
   3
   4     This program is free software; you can redistribute it and/or modify
   5     it under the terms of the GNU General Public License as published by
   6     the Free Software Foundation; either version 2 of the License, or
   7     (at your option) any later version.
   8
   9     This program is distributed in the hope that it will be useful,
  10     but WITHOUT ANY WARRANTY; without even the implied warranty of
  11     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12     GNU General Public License for more details.
  13
  14     You should have received a copy of the GNU General Public License
  15     along with this program; if not, write to the Free Software
  16     Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  17 */
  18
  19 /*
  20   supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
  21   supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
  22   {BGR,RGB}{1,4,8,15,16} support dithering
  23
  24   unscaled special converters (YV12=I420=IYUV, Y800=Y8)
  25   YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
  26   x -> x
  27   YUV9 -> YV12
  28   YUV9/YV12 -> Y800
  29   Y800 -> YUV9/YV12
  30   BGR24 -> BGR32 & RGB24 -> RGB32
  31   BGR32 -> BGR24 & RGB32 -> RGB24
  32   BGR15 -> BGR16
  33 */
  34
  35 /*
  36 tested special converters (most are tested actually but i didnt write it down ...)
  37  YV12 -> BGR16
  38  YV12 -> YV12
  39  BGR15 -> BGR16
  40  BGR16 -> BGR16
  41  YVU9 -> YV12
  42
  43 untested special converters
  44   YV12/I420 -> BGR15/BGR24/BGR32 (its the yuv2rgb stuff, so it should be ok)
  45   YV12/I420 -> YV12/I420
  46   YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
  47   BGR24 -> BGR32 & RGB24 -> RGB32
  48   BGR32 -> BGR24 & RGB32 -> RGB24
  49   BGR24 -> YV12
  50 */
  51
  52 #include <inttypes.h>
  53 #include <string.h>
  54 #include <math.h>
  55 #include <stdio.h>
  56 #include "config.h"
  57 #include <assert.h>
  58 #ifdef HAVE_MALLOC_H
  59 #include <malloc.h>
  60 #else
  61 #include <stdlib.h>
  62 #endif
  63 #include "swscale.h"
  64 #include "swscale_internal.h"
  65 #include "common.h"
  66 #include "rgb2rgb.h"
  67 #define RUNTIME_CPUDETECT 1
  68
  69 #undef MOVNTQ
  70 #undef PAVGB
  71
  72 //#undef HAVE_MMX2
  73 //#define HAVE_3DNOW
  74 //#undef HAVE_MMX
  75 //#undef ARCH_X86
  76 //#define WORDS_BIGENDIAN
  77 #define DITHER1XBPP
  78
  79 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
  80
  81 #define RET 0xC3 //near return opcode for X86
  82
  83 #ifdef MP_DEBUG
  84 #define ASSERT(x) assert(x);
  85 #else
  86 #define ASSERT(x) ;
  87 #endif
  88
  89 #ifdef M_PI
  90 #define PI M_PI
  91 #else
  92 #define PI 3.14159265358979323846
  93 #endif
  94
  95 //FIXME replace this with something faster
  96 #define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YVU9 \
  97                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
  98 #define isYUV(x)       ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x))
  99 #define isGray(x)      ((x)==IMGFMT_Y800)
 100 #define isRGB(x)       (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
 101 #define isBGR(x)       (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
 102 #define isSupportedIn(x)  ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
 103                         || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
 104                         || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
 105                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\
 106                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
 107 #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
 108                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
 109                         || isRGB(x) || isBGR(x)\
 110                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
 111 #define isPacked(x)    ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x))
 112
 113 #define RGB2YUV_SHIFT 16
 114 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
 115 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
 116 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
 117 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
 118 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
 119 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
 120 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
 121 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
 122 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
 123
 124 extern const int32_t Inverse_Table_6_9[8][4];
 125
 126 /*
 127 NOTES
 128 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
 129
 130 TODO
 131 more intelligent missalignment avoidance for the horizontal scaler
 132 write special vertical cubic upscale version
 133 Optimize C code (yv12 / minmax)
 134 add support for packed pixel yuv input & output
 135 add support for Y8 output
 136 optimize bgr24 & bgr32
 137 add BGR4 output support
 138 write special BGR->BGR scaler
 139 */
 140
 141 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
 142 #define MIN(a,b) ((a) > (b) ? (b) : (a))
 143 #define MAX(a,b) ((a) < (b) ? (b) : (a))
 144
 145 #ifdef ARCH_X86
 146 static uint64_t attribute_used __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
 147 static uint64_t attribute_used __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
 148 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
 149 static uint64_t attribute_used __attribute__((aligned(8))) w02=       0x0002000200020002LL;
 150 static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
 151 static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
 152 static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
 153 static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
 154
 155 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
 156 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
 157 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
 158 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
 159
 160 static uint64_t __attribute__((aligned(8))) dither4[2]={
 161         0x0103010301030103LL,
 162         0x0200020002000200LL,};
 163
 164 static uint64_t __attribute__((aligned(8))) dither8[2]={
 165         0x0602060206020602LL,
 166         0x0004000400040004LL,};
 167
 168 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
 169 static uint64_t attribute_used __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
 170 static uint64_t attribute_used __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
 171 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
 172 static uint64_t attribute_used __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
 173 static uint64_t attribute_used __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
 174
 175 static uint64_t attribute_used __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
 176 static uint64_t attribute_used __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
 177 static uint64_t attribute_used __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
 178
 179 #ifdef FAST_BGR2YV12
 180 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
 181 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
 182 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
 183 #else
 184 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
 185 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
 186 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
 187 #endif
 188 static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
 189 static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
 190 static const uint64_t w1111       attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
 191 #endif
 192
 193 // clipping helper table for C implementations:
 194 static unsigned char clip_table[768];
 195
 196 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
 197
 198 extern const uint8_t dither_2x2_4[2][8];
 199 extern const uint8_t dither_2x2_8[2][8];
 200 extern const uint8_t dither_8x8_32[8][8];
 201 extern const uint8_t dither_8x8_73[8][8];
 202 extern const uint8_t dither_8x8_220[8][8];
 203
 204 #ifdef ARCH_X86
 205 void in_asm_used_var_warning_killer()
 206 {
 207  volatile int i= bF8+bFC+w10+
 208  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
 209  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
 210  if(i) i=0;
 211 }
 212 #endif
 213
 214 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
 215                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
 216                                     uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
 217 {
 218         //FIXME Optimize (just quickly writen not opti..)
 219         int i;
 220         for(i=0; i<dstW; i++)
 221         {
 222                 int val=1<<18;
 223                 int j;
 224                 for(j=0; j<lumFilterSize; j++)
 225                         val += lumSrc[j][i] * lumFilter[j];
 226
 227                 dest[i]= MIN(MAX(val>>19, 0), 255);
 228         }
 229
 230         if(uDest != NULL)
 231                 for(i=0; i<chrDstW; i++)
 232                 {
 233                         int u=1<<18;
 234                         int v=1<<18;
 235                         int j;
 236                         for(j=0; j<chrFilterSize; j++)
 237                         {
 238                                 u += chrSrc[j][i] * chrFilter[j];
 239                                 v += chrSrc[j][i + 2048] * chrFilter[j];
 240                         }
 241
 242                         uDest[i]= MIN(MAX(u>>19, 0), 255);
 243                         vDest[i]= MIN(MAX(v>>19, 0), 255);
 244                 }
 245 }
 246
 247
 248 #define YSCALE_YUV_2_PACKEDX_C(type) \
 249                 for(i=0; i<(dstW>>1); i++){\
 250                         int j;\
 251                         int Y1=1<<18;\
 252                         int Y2=1<<18;\
 253                         int U=1<<18;\
 254                         int V=1<<18;\
 255                         type *r, *b, *g;\
 256                         const int i2= 2*i;\
 257                         \
 258                         for(j=0; j<lumFilterSize; j++)\
 259                         {\
 260                                 Y1 += lumSrc[j][i2] * lumFilter[j];\
 261                                 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
 262                         }\
 263                         for(j=0; j<chrFilterSize; j++)\
 264                         {\
 265                                 U += chrSrc[j][i] * chrFilter[j];\
 266                                 V += chrSrc[j][i+2048] * chrFilter[j];\
 267                         }\
 268                         Y1>>=19;\
 269                         Y2>>=19;\
 270                         U >>=19;\
 271                         V >>=19;\
 272                         if((Y1|Y2|U|V)&256)\
 273                         {\
 274                                 if(Y1>255)   Y1=255;\
 275                                 else if(Y1<0)Y1=0;\
 276                                 if(Y2>255)   Y2=255;\
 277                                 else if(Y2<0)Y2=0;\
 278                                 if(U>255)    U=255;\
 279                                 else if(U<0) U=0;\
 280                                 if(V>255)    V=255;\
 281                                 else if(V<0) V=0;\
 282                         }
 283
 284 #define YSCALE_YUV_2_RGBX_C(type) \
 285                         YSCALE_YUV_2_PACKEDX_C(type)\
 286                         r = c->table_rV[V];\
 287                         g = c->table_gU[U] + c->table_gV[V];\
 288                         b = c->table_bU[U];\
 289
 290 #define YSCALE_YUV_2_PACKED2_C \
 291                 for(i=0; i<(dstW>>1); i++){\
 292                         const int i2= 2*i;\
 293                         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
 294                         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
 295                         int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
 296                         int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
 297
 298 #define YSCALE_YUV_2_RGB2_C(type) \
 299                         YSCALE_YUV_2_PACKED2_C\
 300                         type *r, *b, *g;\
 301                         r = c->table_rV[V];\
 302                         g = c->table_gU[U] + c->table_gV[V];\
 303                         b = c->table_bU[U];\
 304
 305 #define YSCALE_YUV_2_PACKED1_C \
 306                 for(i=0; i<(dstW>>1); i++){\
 307                         const int i2= 2*i;\
 308                         int Y1= buf0[i2  ]>>7;\
 309                         int Y2= buf0[i2+1]>>7;\
 310                         int U= (uvbuf1[i     ])>>7;\
 311                         int V= (uvbuf1[i+2048])>>7;\
 312
 313 #define YSCALE_YUV_2_RGB1_C(type) \
 314                         YSCALE_YUV_2_PACKED1_C\
 315                         type *r, *b, *g;\
 316                         r = c->table_rV[V];\
 317                         g = c->table_gU[U] + c->table_gV[V];\
 318                         b = c->table_bU[U];\
 319
 320 #define YSCALE_YUV_2_PACKED1B_C \
 321                 for(i=0; i<(dstW>>1); i++){\
 322                         const int i2= 2*i;\
 323                         int Y1= buf0[i2  ]>>7;\
 324                         int Y2= buf0[i2+1]>>7;\
 325                         int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
 326                         int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
 327
 328 #define YSCALE_YUV_2_RGB1B_C(type) \
 329                         YSCALE_YUV_2_PACKED1B_C\
 330                         type *r, *b, *g;\
 331                         r = c->table_rV[V];\
 332                         g = c->table_gU[U] + c->table_gV[V];\
 333                         b = c->table_bU[U];\
 334
 335 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
 336         switch(c->dstFormat)\
 337         {\
 338         case IMGFMT_BGR32:\
 339         case IMGFMT_RGB32:\
 340                 func(uint32_t)\
 341                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
 342                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
 343                 }               \
 344                 break;\
 345         case IMGFMT_RGB24:\
 346                 func(uint8_t)\
 347                         ((uint8_t*)dest)[0]= r[Y1];\
 348                         ((uint8_t*)dest)[1]= g[Y1];\
 349                         ((uint8_t*)dest)[2]= b[Y1];\
 350                         ((uint8_t*)dest)[3]= r[Y2];\
 351                         ((uint8_t*)dest)[4]= g[Y2];\
 352                         ((uint8_t*)dest)[5]= b[Y2];\
 353                         dest+=6;\
 354                 }\
 355                 break;\
 356         case IMGFMT_BGR24:\
 357                 func(uint8_t)\
 358                         ((uint8_t*)dest)[0]= b[Y1];\
 359                         ((uint8_t*)dest)[1]= g[Y1];\
 360                         ((uint8_t*)dest)[2]= r[Y1];\
 361                         ((uint8_t*)dest)[3]= b[Y2];\
 362                         ((uint8_t*)dest)[4]= g[Y2];\
 363                         ((uint8_t*)dest)[5]= r[Y2];\
 364                         dest+=6;\
 365                 }\
 366                 break;\
 367         case IMGFMT_RGB16:\
 368         case IMGFMT_BGR16:\
 369                 {\
 370                         const int dr1= dither_2x2_8[y&1    ][0];\
 371                         const int dg1= dither_2x2_4[y&1    ][0];\
 372                         const int db1= dither_2x2_8[(y&1)^1][0];\
 373                         const int dr2= dither_2x2_8[y&1    ][1];\
 374                         const int dg2= dither_2x2_4[y&1    ][1];\
 375                         const int db2= dither_2x2_8[(y&1)^1][1];\
 376                         func(uint16_t)\
 377                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
 378                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
 379                         }\
 380                 }\
 381                 break;\
 382         case IMGFMT_RGB15:\
 383         case IMGFMT_BGR15:\
 384                 {\
 385                         const int dr1= dither_2x2_8[y&1    ][0];\
 386                         const int dg1= dither_2x2_8[y&1    ][1];\
 387                         const int db1= dither_2x2_8[(y&1)^1][0];\
 388                         const int dr2= dither_2x2_8[y&1    ][1];\
 389                         const int dg2= dither_2x2_8[y&1    ][0];\
 390                         const int db2= dither_2x2_8[(y&1)^1][1];\
 391                         func(uint16_t)\
 392                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
 393                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
 394                         }\
 395                 }\
 396                 break;\
 397         case IMGFMT_RGB8:\
 398         case IMGFMT_BGR8:\
 399                 {\
 400                         const uint8_t * const d64= dither_8x8_73[y&7];\
 401                         const uint8_t * const d32= dither_8x8_32[y&7];\
 402                         func(uint8_t)\
 403                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
 404                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
 405                         }\
 406                 }\
 407                 break;\
 408         case IMGFMT_RGB4:\
 409         case IMGFMT_BGR4:\
 410                 {\
 411                         const uint8_t * const d64= dither_8x8_73 [y&7];\
 412                         const uint8_t * const d128=dither_8x8_220[y&7];\
 413                         func(uint8_t)\
 414                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
 415                                                  + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
 416                         }\
 417                 }\
 418                 break;\
 419         case IMGFMT_RG4B:\
 420         case IMGFMT_BG4B:\
 421                 {\
 422                         const uint8_t * const d64= dither_8x8_73 [y&7];\
 423                         const uint8_t * const d128=dither_8x8_220[y&7];\
 424                         func(uint8_t)\
 425                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
 426                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
 427                         }\
 428                 }\
 429                 break;\
 430         case IMGFMT_RGB1:\
 431         case IMGFMT_BGR1:\
 432                 {\
 433                         const uint8_t * const d128=dither_8x8_220[y&7];\
 434                         uint8_t *g= c->table_gU[128] + c->table_gV[128];\
 435                         for(i=0; i<dstW-7; i+=8){\
 436                                 int acc;\
 437                                 acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
 438                                 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
 439                                 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
 440                                 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
 441                                 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
 442                                 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
 443                                 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
 444                                 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
 445                                 ((uint8_t*)dest)[0]= acc;\
 446                                 dest++;\
 447                         }\
 448 \
 449 /*\
 450 ((uint8_t*)dest)-= dstW>>4;\
 451 {\
 452                         int acc=0;\
 453                         int left=0;\
 454                         static int top[1024];\
 455                         static int last_new[1024][1024];\
 456                         static int last_in3[1024][1024];\
 457                         static int drift[1024][1024];\
 458                         int topLeft=0;\
 459                         int shift=0;\
 460                         int count=0;\
 461                         const uint8_t * const d128=dither_8x8_220[y&7];\
 462                         int error_new=0;\
 463                         int error_in3=0;\
 464                         int f=0;\
 465                         \
 466                         for(i=dstW>>1; i<dstW; i++){\
 467                                 int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
 468                                 int in2 = (76309 * (in - 16) + 32768) >> 16;\
 469                                 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
 470                                 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
 471                                         + (last_new[y][i] - in3)*f/256;\
 472                                 int new= old> 128 ? 255 : 0;\
 473 \
 474                                 error_new+= ABS(last_new[y][i] - new);\
 475                                 error_in3+= ABS(last_in3[y][i] - in3);\
 476                                 f= error_new - error_in3*4;\
 477                                 if(f<0) f=0;\
 478                                 if(f>256) f=256;\
 479 \
 480                                 topLeft= top[i];\
 481                                 left= top[i]= old - new;\
 482                                 last_new[y][i]= new;\
 483                                 last_in3[y][i]= in3;\
 484 \
 485                                 acc+= acc + (new&1);\
 486                                 if((i&7)==6){\
 487                                         ((uint8_t*)dest)[0]= acc;\
 488                                         ((uint8_t*)dest)++;\
 489                                 }\
 490                         }\
 491 }\
 492 */\
 493                 }\
 494                 break;\
 495         case IMGFMT_YUY2:\
 496                 func2\
 497                         ((uint8_t*)dest)[2*i2+0]= Y1;\
 498                         ((uint8_t*)dest)[2*i2+1]= U;\
 499                         ((uint8_t*)dest)[2*i2+2]= Y2;\
 500                         ((uint8_t*)dest)[2*i2+3]= V;\
 501                 }               \
 502                 break;\
 503         case IMGFMT_UYVY:\
 504                 func2\
 505                         ((uint8_t*)dest)[2*i2+0]= U;\
 506                         ((uint8_t*)dest)[2*i2+1]= Y1;\
 507                         ((uint8_t*)dest)[2*i2+2]= V;\
 508                         ((uint8_t*)dest)[2*i2+3]= Y2;\
 509                 }               \
 510                 break;\
 511         }\
 512
 513
 514 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
 515                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
 516                                     uint8_t *dest, int dstW, int y)
 517 {
 518         int i;
 519         switch(c->dstFormat)
 520         {
 521         case IMGFMT_RGB32:
 522         case IMGFMT_BGR32:
 523                 YSCALE_YUV_2_RGBX_C(uint32_t)
 524                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
 525                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
 526                 }
 527                 break;
 528         case IMGFMT_RGB24:
 529                 YSCALE_YUV_2_RGBX_C(uint8_t)
 530                         ((uint8_t*)dest)[0]= r[Y1];
 531                         ((uint8_t*)dest)[1]= g[Y1];
 532                         ((uint8_t*)dest)[2]= b[Y1];
 533                         ((uint8_t*)dest)[3]= r[Y2];
 534                         ((uint8_t*)dest)[4]= g[Y2];
 535                         ((uint8_t*)dest)[5]= b[Y2];
 536                         dest+=6;
 537                 }
 538                 break;
 539         case IMGFMT_BGR24:
 540                 YSCALE_YUV_2_RGBX_C(uint8_t)
 541                         ((uint8_t*)dest)[0]= b[Y1];
 542                         ((uint8_t*)dest)[1]= g[Y1];
 543                         ((uint8_t*)dest)[2]= r[Y1];
 544                         ((uint8_t*)dest)[3]= b[Y2];
 545                         ((uint8_t*)dest)[4]= g[Y2];
 546                         ((uint8_t*)dest)[5]= r[Y2];
 547                         dest+=6;
 548                 }
 549                 break;
 550         case IMGFMT_RGB16:
 551         case IMGFMT_BGR16:
 552                 {
 553                         const int dr1= dither_2x2_8[y&1    ][0];
 554                         const int dg1= dither_2x2_4[y&1    ][0];
 555                         const int db1= dither_2x2_8[(y&1)^1][0];
 556                         const int dr2= dither_2x2_8[y&1    ][1];
 557                         const int dg2= dither_2x2_4[y&1    ][1];
 558                         const int db2= dither_2x2_8[(y&1)^1][1];
 559                         YSCALE_YUV_2_RGBX_C(uint16_t)
 560                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
 561                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
 562                         }
 563                 }
 564                 break;
 565         case IMGFMT_RGB15:
 566         case IMGFMT_BGR15:
 567                 {
 568                         const int dr1= dither_2x2_8[y&1    ][0];
 569                         const int dg1= dither_2x2_8[y&1    ][1];
 570                         const int db1= dither_2x2_8[(y&1)^1][0];
 571                         const int dr2= dither_2x2_8[y&1    ][1];
 572                         const int dg2= dither_2x2_8[y&1    ][0];
 573                         const int db2= dither_2x2_8[(y&1)^1][1];
 574                         YSCALE_YUV_2_RGBX_C(uint16_t)
 575                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
 576                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
 577                         }
 578                 }
 579                 break;
 580         case IMGFMT_RGB8:
 581         case IMGFMT_BGR8:
 582                 {
 583                         const uint8_t * const d64= dither_8x8_73[y&7];
 584                         const uint8_t * const d32= dither_8x8_32[y&7];
 585                         YSCALE_YUV_2_RGBX_C(uint8_t)
 586                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
 587                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
 588                         }
 589                 }
 590                 break;
 591         case IMGFMT_RGB4:
 592         case IMGFMT_BGR4:
 593                 {
 594                         const uint8_t * const d64= dither_8x8_73 [y&7];
 595                         const uint8_t * const d128=dither_8x8_220[y&7];
 596                         YSCALE_YUV_2_RGBX_C(uint8_t)
 597                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
 598                                                   +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
 599                         }
 600                 }
 601                 break;
 602         case IMGFMT_RG4B:
 603         case IMGFMT_BG4B:
 604                 {
 605                         const uint8_t * const d64= dither_8x8_73 [y&7];
 606                         const uint8_t * const d128=dither_8x8_220[y&7];
 607                         YSCALE_YUV_2_RGBX_C(uint8_t)
 608                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
 609                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
 610                         }
 611                 }
 612                 break;
 613         case IMGFMT_RGB1:
 614         case IMGFMT_BGR1:
 615                 {
 616                         const uint8_t * const d128=dither_8x8_220[y&7];
 617                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
 618                         int acc=0;
 619                         for(i=0; i<dstW-1; i+=2){
 620                                 int j;
 621                                 int Y1=1<<18;
 622                                 int Y2=1<<18;
 623
 624                                 for(j=0; j<lumFilterSize; j++)
 625                                 {
 626                                         Y1 += lumSrc[j][i] * lumFilter[j];
 627                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
 628                                 }
 629                                 Y1>>=19;
 630                                 Y2>>=19;
 631                                 if((Y1|Y2)&256)
 632                                 {
 633                                         if(Y1>255)   Y1=255;
 634                                         else if(Y1<0)Y1=0;
 635                                         if(Y2>255)   Y2=255;
 636                                         else if(Y2<0)Y2=0;
 637                                 }
 638                                 acc+= acc + g[Y1+d128[(i+0)&7]];
 639                                 acc+= acc + g[Y2+d128[(i+1)&7]];
 640                                 if((i&7)==6){
 641                                         ((uint8_t*)dest)[0]= acc;
 642                                         dest++;
 643                                 }
 644                         }
 645                 }
 646                 break;
 647         case IMGFMT_YUY2:
 648                 YSCALE_YUV_2_PACKEDX_C(void)
 649                         ((uint8_t*)dest)[2*i2+0]= Y1;
 650                         ((uint8_t*)dest)[2*i2+1]= U;
 651                         ((uint8_t*)dest)[2*i2+2]= Y2;
 652                         ((uint8_t*)dest)[2*i2+3]= V;
 653                 }
 654                 break;
 655         case IMGFMT_UYVY:
 656                 YSCALE_YUV_2_PACKEDX_C(void)
 657                         ((uint8_t*)dest)[2*i2+0]= U;
 658                         ((uint8_t*)dest)[2*i2+1]= Y1;
 659                         ((uint8_t*)dest)[2*i2+2]= V;
 660                         ((uint8_t*)dest)[2*i2+3]= Y2;
 661                 }
 662                 break;
 663         }
 664 }
 665
 666
 667 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
 668 //Plain C versions
 669 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
 670 #define COMPILE_C
 671 #endif
 672
 673 #ifdef ARCH_POWERPC
 674 #ifdef HAVE_ALTIVEC
 675 #define COMPILE_ALTIVEC
 676 #endif //HAVE_ALTIVEC
 677 #endif //ARCH_POWERPC
 678
 679 #ifdef ARCH_X86
 680
 681 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
 682 #define COMPILE_MMX
 683 #endif
 684
 685 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
 686 #define COMPILE_MMX2
 687 #endif
 688
 689 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
 690 #define COMPILE_3DNOW
 691 #endif
 692 #endif //ARCH_X86
 693
 694 #undef HAVE_MMX
 695 #undef HAVE_MMX2
 696 #undef HAVE_3DNOW
 697
 698 #ifdef COMPILE_C
 699 #undef HAVE_MMX
 700 #undef HAVE_MMX2
 701 #undef HAVE_3DNOW
 702 #undef HAVE_ALTIVEC
 703 #define RENAME(a) a ## _C
 704 #include "swscale_template.c"
 705 #endif
 706
 707 #ifdef ARCH_POWERPC
 708 #ifdef COMPILE_ALTIVEC
 709 #undef RENAME
 710 #define HAVE_ALTIVEC
 711 #define RENAME(a) a ## _altivec
 712 #include "swscale_template.c"
 713 #endif
 714 #endif //ARCH_POWERPC
 715
 716 #ifdef ARCH_X86
 717
 718 //X86 versions
 719 /*
 720 #undef RENAME
 721 #undef HAVE_MMX
 722 #undef HAVE_MMX2
 723 #undef HAVE_3DNOW
 724 #define ARCH_X86
 725 #define RENAME(a) a ## _X86
 726 #include "swscale_template.c"
 727 */
 728 //MMX versions
 729 #ifdef COMPILE_MMX
 730 #undef RENAME
 731 #define HAVE_MMX
 732 #undef HAVE_MMX2
 733 #undef HAVE_3DNOW
 734 #define RENAME(a) a ## _MMX
 735 #include "swscale_template.c"
 736 #endif
 737
 738 //MMX2 versions
 739 #ifdef COMPILE_MMX2
 740 #undef RENAME
 741 #define HAVE_MMX
 742 #define HAVE_MMX2
 743 #undef HAVE_3DNOW
 744 #define RENAME(a) a ## _MMX2
 745 #include "swscale_template.c"
 746 #endif
 747
 748 //3DNOW versions
 749 #ifdef COMPILE_3DNOW
 750 #undef RENAME
 751 #define HAVE_MMX
 752 #undef HAVE_MMX2
 753 #define HAVE_3DNOW
 754 #define RENAME(a) a ## _3DNow
 755 #include "swscale_template.c"
 756 #endif
 757
 758 #endif //ARCH_X86
 759
 760 // minor note: the HAVE_xyz is messed up after that line so don't use it
 761
 762 static double getSplineCoeff(double a, double b, double c, double d, double dist)
 763 {
 764 //      printf("%f %f %f %f %f\n", a,b,c,d,dist);
 765         if(dist<=1.0)   return ((d*dist + c)*dist + b)*dist +a;
 766         else            return getSplineCoeff(  0.0,
 767                                                  b+ 2.0*c + 3.0*d,
 768                                                         c + 3.0*d,
 769                                                 -b- 3.0*c - 6.0*d,
 770                                                 dist-1.0);
 771 }
 772
 773 static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
 774                               int srcW, int dstW, int filterAlign, int one, int flags,
 775                               SwsVector *srcFilter, SwsVector *dstFilter)
 776 {
 777         int i;
 778         int filterSize;
 779         int filter2Size;
 780         int minFilterSize;
 781         double *filter=NULL;
 782         double *filter2=NULL;
 783 #ifdef ARCH_X86
 784         if(flags & SWS_CPU_CAPS_MMX)
 785                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
 786 #endif
 787
 788         // Note the +1 is for the MMXscaler which reads over the end
 789         *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
 790
 791         if(ABS(xInc - 0x10000) <10) // unscaled
 792         {
 793                 int i;
 794                 filterSize= 1;
 795                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
 796                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
 797
 798                 for(i=0; i<dstW; i++)
 799                 {
 800                         filter[i*filterSize]=1;
 801                         (*filterPos)[i]=i;
 802                 }
 803
 804         }
 805         else if(flags&SWS_POINT) // lame looking point sampling mode
 806         {
 807                 int i;
 808                 int xDstInSrc;
 809                 filterSize= 1;
 810                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
 811
 812                 xDstInSrc= xInc/2 - 0x8000;
 813                 for(i=0; i<dstW; i++)
 814                 {
 815                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
 816
 817                         (*filterPos)[i]= xx;
 818                         filter[i]= 1.0;
 819                         xDstInSrc+= xInc;
 820                 }
 821         }
 822         else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
 823         {
 824                 int i;
 825                 int xDstInSrc;
 826                 if     (flags&SWS_BICUBIC) filterSize= 4;
 827                 else if(flags&SWS_X      ) filterSize= 4;
 828                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA
 829                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
 830
 831                 xDstInSrc= xInc/2 - 0x8000;
 832                 for(i=0; i<dstW; i++)
 833                 {
 834                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
 835                         int j;
 836
 837                         (*filterPos)[i]= xx;
 838                                 //Bilinear upscale / linear interpolate / Area averaging
 839                                 for(j=0; j<filterSize; j++)
 840                                 {
 841                                         double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
 842                                         double coeff= 1.0 - d;
 843                                         if(coeff<0) coeff=0;
 844                                         filter[i*filterSize + j]= coeff;
 845                                         xx++;
 846                                 }
 847                         xDstInSrc+= xInc;
 848                 }
 849         }
 850         else
 851         {
 852                 double xDstInSrc;
 853                 double sizeFactor, filterSizeInSrc;
 854                 const double xInc1= (double)xInc / (double)(1<<16);
 855                 int param= (flags&SWS_PARAM_MASK)>>SWS_PARAM_SHIFT;
 856
 857                 if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
 858                 else if(flags&SWS_X)            sizeFactor= 8.0;
 859                 else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
 860                 else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
 861                 else if(flags&SWS_LANCZOS)      sizeFactor= param ? 2.0*param : 6.0;
 862                 else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
 863                 else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
 864                 else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
 865                 else {
 866                         sizeFactor= 0.0; //GCC warning killer
 867                         ASSERT(0)
 868                 }
 869
 870                 if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
 871                 else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
 872
 873                 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
 874                 if(filterSize > srcW-2) filterSize=srcW-2;
 875
 876                 filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
 877
 878                 xDstInSrc= xInc1 / 2.0 - 0.5;
 879                 for(i=0; i<dstW; i++)
 880                 {
 881                         int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
 882                         int j;
 883                         (*filterPos)[i]= xx;
 884                         for(j=0; j<filterSize; j++)
 885                         {
 886                                 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
 887                                 double coeff;
 888                                 if(flags & SWS_BICUBIC)
 889                                 {
 890                                         double A= param ? -param*0.01 : -0.60;
 891
 892                                         // Equation is from VirtualDub
 893                                         if(d<1.0)
 894                                                 coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
 895                                         else if(d<2.0)
 896                                                 coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
 897                                         else
 898                                                 coeff=0.0;
 899                                 }
 900 /*                              else if(flags & SWS_X)
 901                                 {
 902                                         double p= param ? param*0.01 : 0.3;
 903                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
 904                                         coeff*= pow(2.0, - p*d*d);
 905                                 }*/
 906                                 else if(flags & SWS_X)
 907                                 {
 908                                         double A= param ? param*0.1 : 1.0;
 909
 910                                         if(d<1.0)
 911                                                 coeff = cos(d*PI);
 912                                         else
 913                                                 coeff=-1.0;
 914                                         if(coeff<0.0)   coeff= -pow(-coeff, A);
 915                                         else            coeff=  pow( coeff, A);
 916                                         coeff= coeff*0.5 + 0.5;
 917                                 }
 918                                 else if(flags & SWS_AREA)
 919                                 {
 920                                         double srcPixelSize= 1.0/xInc1;
 921                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
 922                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
 923                                         else coeff=0.0;
 924                                 }
 925                                 else if(flags & SWS_GAUSS)
 926                                 {
 927                                         double p= param ? param*0.1 : 3.0;
 928                                         coeff = pow(2.0, - p*d*d);
 929                                 }
 930                                 else if(flags & SWS_SINC)
 931                                 {
 932                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
 933                                 }
 934                                 else if(flags & SWS_LANCZOS)
 935                                 {
 936                                         double p= param ? param : 3.0;
 937                                         coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
 938                                         if(d>p) coeff=0;
 939                                 }
 940                                 else if(flags & SWS_BILINEAR)
 941                                 {
 942                                         coeff= 1.0 - d;
 943                                         if(coeff<0) coeff=0;
 944                                 }
 945                                 else if(flags & SWS_SPLINE)
 946                                 {
 947                                         double p=-2.196152422706632;
 948                                         coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
 949                                 }
 950                                 else {
 951                                         coeff= 0.0; //GCC warning killer
 952                                         ASSERT(0)
 953                                 }
 954
 955                                 filter[i*filterSize + j]= coeff;
 956                                 xx++;
 957                         }
 958                         xDstInSrc+= xInc1;
 959                 }
 960         }
 961
 962         /* apply src & dst Filter to filter -> filter2
 963            free(filter);
 964         */
 965         ASSERT(filterSize>0)
 966         filter2Size= filterSize;
 967         if(srcFilter) filter2Size+= srcFilter->length - 1;
 968         if(dstFilter) filter2Size+= dstFilter->length - 1;
 969         ASSERT(filter2Size>0)
 970         filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
 971
 972         for(i=0; i<dstW; i++)
 973         {
 974                 int j;
 975                 SwsVector scaleFilter;
 976                 SwsVector *outVec;
 977
 978                 scaleFilter.coeff= filter + i*filterSize;
 979                 scaleFilter.length= filterSize;
 980
 981                 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
 982                 else          outVec= &scaleFilter;
 983
 984                 ASSERT(outVec->length == filter2Size)
 985                 //FIXME dstFilter
 986
 987                 for(j=0; j<outVec->length; j++)
 988                 {
 989                         filter2[i*filter2Size + j]= outVec->coeff[j];
 990                 }
 991
 992                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
 993
 994                 if(outVec != &scaleFilter) sws_freeVec(outVec);
 995         }
 996         free(filter); filter=NULL;
 997
 998         /* try to reduce the filter-size (step1 find size and shift left) */
 999         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1000         minFilterSize= 0;
1001         for(i=dstW-1; i>=0; i--)
1002         {
1003                 int min= filter2Size;
1004                 int j;
1005                 double cutOff=0.0;
1006
1007                 /* get rid off near zero elements on the left by shifting left */
1008                 for(j=0; j<filter2Size; j++)
1009                 {
1010                         int k;
1011                         cutOff += ABS(filter2[i*filter2Size]);
1012
1013                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1014
1015                         /* preserve Monotonicity because the core can't handle the filter otherwise */
1016                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1017
1018                         // Move filter coeffs left
1019                         for(k=1; k<filter2Size; k++)
1020                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1021                         filter2[i*filter2Size + k - 1]= 0.0;
1022                         (*filterPos)[i]++;
1023                 }
1024
1025                 cutOff=0.0;
1026                 /* count near zeros on the right */
1027                 for(j=filter2Size-1; j>0; j--)
1028                 {
1029                         cutOff += ABS(filter2[i*filter2Size + j]);
1030
1031                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1032                         min--;
1033                 }
1034
1035                 if(min>minFilterSize) minFilterSize= min;
1036         }
1037
1038         if (flags & SWS_CPU_CAPS_ALTIVEC) {
1039           // we can handle the special case 4,
1040           // so we don't want to go to the full 8
1041           if (minFilterSize < 5)
1042             filterAlign = 4;
1043
1044           // we really don't want to waste our time
1045           // doing useless computation, so fall-back on
1046           // the scalar C code for very small filter.
1047           // vectorizing is worth it only if you have
1048           // decent-sized vector.
1049           if (minFilterSize < 3)
1050             filterAlign = 1;
1051         }
1052
1053         ASSERT(minFilterSize > 0)
1054         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1055         ASSERT(filterSize > 0)
1056         filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1057         *outFilterSize= filterSize;
1058
1059         if(flags&SWS_PRINT_INFO)
1060                 MSG_INFO("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1061         /* try to reduce the filter-size (step2 reduce it) */
1062         for(i=0; i<dstW; i++)
1063         {
1064                 int j;
1065
1066                 for(j=0; j<filterSize; j++)
1067                 {
1068                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1069                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1070                 }
1071         }
1072         free(filter2); filter2=NULL;
1073
1074
1075         //FIXME try to align filterpos if possible
1076
1077         //fix borders
1078         for(i=0; i<dstW; i++)
1079         {
1080                 int j;
1081                 if((*filterPos)[i] < 0)
1082                 {
1083                         // Move filter coeffs left to compensate for filterPos
1084                         for(j=1; j<filterSize; j++)
1085                         {
1086                                 int left= MAX(j + (*filterPos)[i], 0);
1087                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1088                                 filter[i*filterSize + j]=0;
1089                         }
1090                         (*filterPos)[i]= 0;
1091                 }
1092
1093                 if((*filterPos)[i] + filterSize > srcW)
1094                 {
1095                         int shift= (*filterPos)[i] + filterSize - srcW;
1096                         // Move filter coeffs right to compensate for filterPos
1097                         for(j=filterSize-2; j>=0; j--)
1098                         {
1099                                 int right= MIN(j + shift, filterSize-1);
1100                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1101                                 filter[i*filterSize +j]=0;
1102                         }
1103                         (*filterPos)[i]= srcW - filterSize;
1104                 }
1105         }
1106
1107         // Note the +1 is for the MMXscaler which reads over the end
1108         *outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));
1109         memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1110
1111         /* Normalize & Store in outFilter */
1112         for(i=0; i<dstW; i++)
1113         {
1114                 int j;
1115                 double error=0;
1116                 double sum=0;
1117                 double scale= one;
1118
1119                 for(j=0; j<filterSize; j++)
1120                 {
1121                         sum+= filter[i*filterSize + j];
1122                 }
1123                 scale/= sum;
1124                 for(j=0; j<*outFilterSize; j++)
1125                 {
1126                         double v= filter[i*filterSize + j]*scale + error;
1127                         int intV= floor(v + 0.5);
1128                         (*outFilter)[i*(*outFilterSize) + j]= intV;
1129                         error = v - intV;
1130                 }
1131         }
1132
1133         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1134         for(i=0; i<*outFilterSize; i++)
1135         {
1136                 int j= dstW*(*outFilterSize);
1137                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1138         }
1139
1140         free(filter);
1141 }
1142
1143 #ifdef ARCH_X86
1144 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1145 {
1146         uint8_t *fragmentA;
1147         int imm8OfPShufW1A;
1148         int imm8OfPShufW2A;
1149         int fragmentLengthA;
1150         uint8_t *fragmentB;
1151         int imm8OfPShufW1B;
1152         int imm8OfPShufW2B;
1153         int fragmentLengthB;
1154         int fragmentPos;
1155
1156         int xpos, i;
1157
1158         // create an optimized horizontal scaling routine
1159
1160         //code fragment
1161
1162         asm volatile(
1163                 "jmp 9f                         \n\t"
1164         // Begin
1165                 "0:                             \n\t"
1166                 "movq (%%edx, %%eax), %%mm3     \n\t"
1167                 "movd (%%ecx, %%esi), %%mm0     \n\t"
1168                 "movd 1(%%ecx, %%esi), %%mm1    \n\t"
1169                 "punpcklbw %%mm7, %%mm1         \n\t"
1170                 "punpcklbw %%mm7, %%mm0         \n\t"
1171                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1172                 "1:                             \n\t"
1173                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1174                 "2:                             \n\t"
1175                 "psubw %%mm1, %%mm0             \n\t"
1176                 "movl 8(%%ebx, %%eax), %%esi    \n\t"
1177                 "pmullw %%mm3, %%mm0            \n\t"
1178                 "psllw $7, %%mm1                \n\t"
1179                 "paddw %%mm1, %%mm0             \n\t"
1180
1181                 "movq %%mm0, (%%edi, %%eax)     \n\t"
1182
1183                 "addl $8, %%eax                 \n\t"
1184         // End
1185                 "9:                             \n\t"
1186 //              "int $3\n\t"
1187                 "leal 0b, %0                    \n\t"
1188                 "leal 1b, %1                    \n\t"
1189                 "leal 2b, %2                    \n\t"
1190                 "decl %1                        \n\t"
1191                 "decl %2                        \n\t"
1192                 "subl %0, %1                    \n\t"
1193                 "subl %0, %2                    \n\t"
1194                 "leal 9b, %3                    \n\t"
1195                 "subl %0, %3                    \n\t"
1196
1197
1198                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1199                 "=r" (fragmentLengthA)
1200         );
1201
1202         asm volatile(
1203                 "jmp 9f                         \n\t"
1204         // Begin
1205                 "0:                             \n\t"
1206                 "movq (%%edx, %%eax), %%mm3     \n\t"
1207                 "movd (%%ecx, %%esi), %%mm0     \n\t"
1208                 "punpcklbw %%mm7, %%mm0         \n\t"
1209                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1210                 "1:                             \n\t"
1211                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1212                 "2:                             \n\t"
1213                 "psubw %%mm1, %%mm0             \n\t"
1214                 "movl 8(%%ebx, %%eax), %%esi    \n\t"
1215                 "pmullw %%mm3, %%mm0            \n\t"
1216                 "psllw $7, %%mm1                \n\t"
1217                 "paddw %%mm1, %%mm0             \n\t"
1218
1219                 "movq %%mm0, (%%edi, %%eax)     \n\t"
1220
1221                 "addl $8, %%eax                 \n\t"
1222         // End
1223                 "9:                             \n\t"
1224 //              "int $3\n\t"
1225                 "leal 0b, %0                    \n\t"
1226                 "leal 1b, %1                    \n\t"
1227                 "leal 2b, %2                    \n\t"
1228                 "decl %1                        \n\t"
1229                 "decl %2                        \n\t"
1230                 "subl %0, %1                    \n\t"
1231                 "subl %0, %2                    \n\t"
1232                 "leal 9b, %3                    \n\t"
1233                 "subl %0, %3                    \n\t"
1234
1235
1236                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1237                 "=r" (fragmentLengthB)
1238         );
1239
1240         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1241         fragmentPos=0;
1242
1243         for(i=0; i<dstW/numSplits; i++)
1244         {
1245                 int xx=xpos>>16;
1246
1247                 if((i&3) == 0)
1248                 {
1249                         int a=0;
1250                         int b=((xpos+xInc)>>16) - xx;
1251                         int c=((xpos+xInc*2)>>16) - xx;
1252                         int d=((xpos+xInc*3)>>16) - xx;
1253
1254                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1255                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1256                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1257                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1258                         filterPos[i/2]= xx;
1259
1260                         if(d+1<4)
1261                         {
1262                                 int maxShift= 3-(d+1);
1263                                 int shift=0;
1264
1265                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1266
1267                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1268                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1269                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1270                                         a | (b<<2) | (c<<4) | (d<<6);
1271
1272                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1273                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1274
1275                                 if(shift && i>=shift)
1276                                 {
1277                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1278                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1279                                         filterPos[i/2]-=shift;
1280                                 }
1281
1282                                 fragmentPos+= fragmentLengthB;
1283                         }
1284                         else
1285                         {
1286                                 int maxShift= 3-d;
1287                                 int shift=0;
1288
1289                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1290
1291                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1292                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1293                                         a | (b<<2) | (c<<4) | (d<<6);
1294
1295                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1296                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1297
1298                                 if(shift && i>=shift)
1299                                 {
1300                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1301                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1302                                         filterPos[i/2]-=shift;
1303                                 }
1304
1305                                 fragmentPos+= fragmentLengthA;
1306                         }
1307
1308                         funnyCode[fragmentPos]= RET;
1309                 }
1310                 xpos+=xInc;
1311         }
1312         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1313 }
1314 #endif // ARCH_X86
1315
1316 static void globalInit(){
1317     // generating tables:
1318     int i;
1319     for(i=0; i<768; i++){
1320         int c= MIN(MAX(i-256, 0), 255);
1321         clip_table[i]=c;
1322     }
1323 }
1324
1325 static SwsFunc getSwsFunc(int flags){
1326
1327 #ifdef RUNTIME_CPUDETECT
1328 #ifdef ARCH_X86
1329         // ordered per speed fasterst first
1330         if(flags & SWS_CPU_CAPS_MMX2)
1331                 return swScale_MMX2;
1332         else if(flags & SWS_CPU_CAPS_3DNOW)
1333                 return swScale_3DNow;
1334         else if(flags & SWS_CPU_CAPS_MMX)
1335                 return swScale_MMX;
1336         else
1337                 return swScale_C;
1338
1339 #else
1340 #ifdef ARCH_POWERPC
1341         if(flags & SWS_CPU_CAPS_ALTIVEC)
1342           return swScale_altivec;
1343         else
1344           return swScale_C;
1345 #endif
1346         return swScale_C;
1347 #endif
1348 #else //RUNTIME_CPUDETECT
1349 #ifdef HAVE_MMX2
1350         return swScale_MMX2;
1351 #elif defined (HAVE_3DNOW)
1352         return swScale_3DNow;
1353 #elif defined (HAVE_MMX)
1354         return swScale_MMX;
1355 #elif defined (HAVE_ALTIVEC)
1356         return swScale_altivec;
1357 #else
1358         return swScale_C;
1359 #endif
1360 #endif //!RUNTIME_CPUDETECT
1361 }
1362
1363 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1364              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1365         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1366         /* Copy Y plane */
1367         if(dstStride[0]==srcStride[0])
1368                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1369         else
1370         {
1371                 int i;
1372                 uint8_t *srcPtr= src[0];
1373                 uint8_t *dstPtr= dst;
1374                 for(i=0; i<srcSliceH; i++)
1375                 {
1376                         memcpy(dstPtr, srcPtr, srcStride[0]);
1377                         srcPtr+= srcStride[0];
1378                         dstPtr+= dstStride[0];
1379                 }
1380         }
1381         dst = dstParam[1] + dstStride[1]*srcSliceY;
1382         interleaveBytes( src[1],src[2],dst,c->srcW,srcSliceH,srcStride[1],srcStride[2],dstStride[0] );
1383
1384         return srcSliceH;
1385 }
1386
1387 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1388              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1389         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1390
1391         yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1392
1393         return srcSliceH;
1394 }
1395
1396 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1397              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1398         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1399
1400         yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1401
1402         return srcSliceH;
1403 }
1404
1405 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1406 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1407                            int srcSliceH, uint8_t* dst[], int dstStride[]){
1408         const int srcFormat= c->srcFormat;
1409         const int dstFormat= c->dstFormat;
1410         const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1411         const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1412         const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8
1413         const int dstId= (dstFormat&0xFF)>>2;
1414         void (*conv)(const uint8_t *src, uint8_t *dst, unsigned src_size)=NULL;
1415
1416         /* BGR -> BGR */
1417         if(   (isBGR(srcFormat) && isBGR(dstFormat))
1418            || (isRGB(srcFormat) && isRGB(dstFormat))){
1419                 switch(srcId | (dstId<<4)){
1420                 case 0x34: conv= rgb16to15; break;
1421                 case 0x36: conv= rgb24to15; break;
1422                 case 0x38: conv= rgb32to15; break;
1423                 case 0x43: conv= rgb15to16; break;
1424                 case 0x46: conv= rgb24to16; break;
1425                 case 0x48: conv= rgb32to16; break;
1426                 case 0x63: conv= rgb15to24; break;
1427                 case 0x64: conv= rgb16to24; break;
1428                 case 0x68: conv= rgb32to24; break;
1429                 case 0x83: conv= rgb15to32; break;
1430                 case 0x84: conv= rgb16to32; break;
1431                 case 0x86: conv= rgb24to32; break;
1432                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1433                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1434                 }
1435         }else if(   (isBGR(srcFormat) && isRGB(dstFormat))
1436                  || (isRGB(srcFormat) && isBGR(dstFormat))){
1437                 switch(srcId | (dstId<<4)){
1438                 case 0x33: conv= rgb15tobgr15; break;
1439                 case 0x34: conv= rgb16tobgr15; break;
1440                 case 0x36: conv= rgb24tobgr15; break;
1441                 case 0x38: conv= rgb32tobgr15; break;
1442                 case 0x43: conv= rgb15tobgr16; break;
1443                 case 0x44: conv= rgb16tobgr16; break;
1444                 case 0x46: conv= rgb24tobgr16; break;
1445                 case 0x48: conv= rgb32tobgr16; break;
1446                 case 0x63: conv= rgb15tobgr24; break;
1447                 case 0x64: conv= rgb16tobgr24; break;
1448                 case 0x66: conv= rgb24tobgr24; break;
1449                 case 0x68: conv= rgb32tobgr24; break;
1450                 case 0x83: conv= rgb15tobgr32; break;
1451                 case 0x84: conv= rgb16tobgr32; break;
1452                 case 0x86: conv= rgb24tobgr32; break;
1453                 case 0x88: conv= rgb32tobgr32; break;
1454                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1455                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1456                 }
1457         }else{
1458                 MSG_ERR("swScaler: internal error %s -> %s converter\n",
1459                          vo_format_name(srcFormat), vo_format_name(dstFormat));
1460         }
1461
1462         if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1463                 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1464         else
1465         {
1466                 int i;
1467                 uint8_t *srcPtr= src[0];
1468                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1469
1470                 for(i=0; i<srcSliceH; i++)
1471                 {
1472                         conv(srcPtr, dstPtr, c->srcW*srcBpp);
1473                         srcPtr+= srcStride[0];
1474                         dstPtr+= dstStride[0];
1475                 }
1476         }
1477         return srcSliceH;
1478 }
1479
1480 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1481              int srcSliceH, uint8_t* dst[], int dstStride[]){
1482
1483         rgb24toyv12(
1484                 src[0],
1485                 dst[0]+ srcSliceY    *dstStride[0],
1486                 dst[1]+(srcSliceY>>1)*dstStride[1],
1487                 dst[2]+(srcSliceY>>1)*dstStride[2],
1488                 c->srcW, srcSliceH,
1489                 dstStride[0], dstStride[1], srcStride[0]);
1490         return srcSliceH;
1491 }
1492
1493 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1494              int srcSliceH, uint8_t* dst[], int dstStride[]){
1495         int i;
1496
1497         /* copy Y */
1498         if(srcStride[0]==dstStride[0])
1499                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1500         else{
1501                 uint8_t *srcPtr= src[0];
1502                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1503
1504                 for(i=0; i<srcSliceH; i++)
1505                 {
1506                         memcpy(dstPtr, srcPtr, c->srcW);
1507                         srcPtr+= srcStride[0];
1508                         dstPtr+= dstStride[0];
1509                 }
1510         }
1511
1512         if(c->dstFormat==IMGFMT_YV12){
1513                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1514                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1515         }else{
1516                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1517                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1518         }
1519         return srcSliceH;
1520 }
1521
1522 /**
1523  * bring pointers in YUV order instead of YVU
1524  */
1525 static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1526         if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
1527            || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
1528                 sortedP[0]= p[0];
1529                 sortedP[1]= p[2];
1530                 sortedP[2]= p[1];
1531                 sortedStride[0]= stride[0];
1532                 sortedStride[1]= stride[2];
1533                 sortedStride[2]= stride[1];
1534         }
1535         else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8)
1536         {
1537                 sortedP[0]= p[0];
1538                 sortedP[1]=
1539                 sortedP[2]= NULL;
1540                 sortedStride[0]= stride[0];
1541                 sortedStride[1]=
1542                 sortedStride[2]= 0;
1543         }
1544         else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)
1545         {
1546                 sortedP[0]= p[0];
1547                 sortedP[1]= p[1];
1548                 sortedP[2]= p[2];
1549                 sortedStride[0]= stride[0];
1550                 sortedStride[1]= stride[1];
1551                 sortedStride[2]= stride[2];
1552         }else{
1553                 MSG_ERR("internal error in orderYUV\n");
1554         }
1555 }
1556
1557 /* unscaled copy like stuff (assumes nearly identical formats) */
1558 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1559              int srcSliceH, uint8_t* dst[], int dstStride[]){
1560
1561         if(isPacked(c->srcFormat))
1562         {
1563                 if(dstStride[0]==srcStride[0])
1564                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1565                 else
1566                 {
1567                         int i;
1568                         uint8_t *srcPtr= src[0];
1569                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1570                         int length=0;
1571
1572                         /* universal length finder */
1573                         while(length+c->srcW <= ABS(dstStride[0])
1574                            && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
1575                         ASSERT(length!=0);
1576
1577                         for(i=0; i<srcSliceH; i++)
1578                         {
1579                                 memcpy(dstPtr, srcPtr, length);
1580                                 srcPtr+= srcStride[0];
1581                                 dstPtr+= dstStride[0];
1582                         }
1583                 }
1584         }
1585         else
1586         { /* Planar YUV or gray */
1587                 int plane;
1588                 for(plane=0; plane<3; plane++)
1589                 {
1590                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1591                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1592                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1593
1594                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1595                         {
1596                                 if(!isGray(c->dstFormat))
1597                                         memset(dst[plane], 128, dstStride[plane]*height);
1598                         }
1599                         else
1600                         {
1601                                 if(dstStride[plane]==srcStride[plane])
1602                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1603                                 else
1604                                 {
1605                                         int i;
1606                                         uint8_t *srcPtr= src[plane];
1607                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1608                                         for(i=0; i<height; i++)
1609                                         {
1610                                                 memcpy(dstPtr, srcPtr, length);
1611                                                 srcPtr+= srcStride[plane];
1612                                                 dstPtr+= dstStride[plane];
1613                                         }
1614                                 }
1615                         }
1616                 }
1617         }
1618         return srcSliceH;
1619 }
1620
1621 static int remove_dup_fourcc(int fourcc)
1622 {
1623         switch(fourcc)
1624         {
1625             case IMGFMT_I420:
1626             case IMGFMT_IYUV: return IMGFMT_YV12;
1627             case IMGFMT_Y8  : return IMGFMT_Y800;
1628             case IMGFMT_IF09: return IMGFMT_YVU9;
1629             default: return fourcc;
1630         }
1631 }
1632
1633 static void getSubSampleFactors(int *h, int *v, int format){
1634         switch(format){
1635         case IMGFMT_UYVY:
1636         case IMGFMT_YUY2:
1637                 *h=1;
1638                 *v=0;
1639                 break;
1640         case IMGFMT_YV12:
1641         case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
1642                 *h=1;
1643                 *v=1;
1644                 break;
1645         case IMGFMT_YVU9:
1646                 *h=2;
1647                 *v=2;
1648                 break;
1649         case IMGFMT_444P:
1650                 *h=0;
1651                 *v=0;
1652                 break;
1653         case IMGFMT_422P:
1654                 *h=1;
1655                 *v=0;
1656                 break;
1657         case IMGFMT_411P:
1658                 *h=2;
1659                 *v=0;
1660                 break;
1661         default:
1662                 *h=0;
1663                 *v=0;
1664                 break;
1665         }
1666 }
1667
1668 static uint16_t roundToInt16(int64_t f){
1669         int r= (f + (1<<15))>>16;
1670              if(r<-0x7FFF) return 0x8000;
1671         else if(r> 0x7FFF) return 0x7FFF;
1672         else               return r;
1673 }
1674
1675 /**
1676  * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1677  * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1678  * @return -1 if not supported
1679  */
1680 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1681         int64_t crv =  inv_table[0];
1682         int64_t cbu =  inv_table[1];
1683         int64_t cgu = -inv_table[2];
1684         int64_t cgv = -inv_table[3];
1685         int64_t cy  = 1<<16;
1686         int64_t oy  = 0;
1687
1688         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1689         memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1690         memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
1691
1692         c->brightness= brightness;
1693         c->contrast  = contrast;
1694         c->saturation= saturation;
1695         c->srcRange  = srcRange;
1696         c->dstRange  = dstRange;
1697
1698         c->uOffset=   0x0400040004000400LL;
1699         c->vOffset=   0x0400040004000400LL;
1700
1701         if(!srcRange){
1702                 cy= (cy*255) / 219;
1703                 oy= 16<<16;
1704         }
1705
1706         cy = (cy *contrast             )>>16;
1707         crv= (crv*contrast * saturation)>>32;
1708         cbu= (cbu*contrast * saturation)>>32;
1709         cgu= (cgu*contrast * saturation)>>32;
1710         cgv= (cgv*contrast * saturation)>>32;
1711
1712         oy -= 256*brightness;
1713
1714         c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
1715         c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
1716         c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1717         c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1718         c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1719         c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
1720
1721         yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1722         //FIXME factorize
1723
1724 #ifdef HAVE_ALTIVEC
1725         yuv2rgb_altivec_init_tables (c, inv_table);
1726 #endif
1727         return 0;
1728 }
1729
1730 /**
1731  * @return -1 if not supported
1732  */
1733 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1734         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1735
1736         *inv_table = c->srcColorspaceTable;
1737         *table     = c->dstColorspaceTable;
1738         *srcRange  = c->srcRange;
1739         *dstRange  = c->dstRange;
1740         *brightness= c->brightness;
1741         *contrast  = c->contrast;
1742         *saturation= c->saturation;
1743
1744         return 0;
1745 }
1746
1747 SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,
1748                          SwsFilter *srcFilter, SwsFilter *dstFilter){
1749
1750         SwsContext *c;
1751         int i;
1752         int usesVFilter, usesHFilter;
1753         int unscaled, needsDither;
1754         int srcFormat, dstFormat;
1755         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1756 #ifdef ARCH_X86
1757         if(flags & SWS_CPU_CAPS_MMX)
1758                 asm volatile("emms\n\t"::: "memory");
1759 #endif
1760
1761 #ifndef RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
1762         flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1763 #ifdef HAVE_MMX2
1764         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1765 #elif defined (HAVE_3DNOW)
1766         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1767 #elif defined (HAVE_MMX)
1768         flags |= SWS_CPU_CAPS_MMX;
1769 #elif defined (HAVE_ALTIVEC)
1770         flags |= SWS_CPU_CAPS_ALTIVEC;
1771 #endif
1772 #endif
1773         if(clip_table[512] != 255) globalInit();
1774         if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1775
1776         /* avoid duplicate Formats, so we don't need to check to much */
1777         srcFormat = remove_dup_fourcc(origSrcFormat);
1778         dstFormat = remove_dup_fourcc(origDstFormat);
1779
1780         unscaled = (srcW == dstW && srcH == dstH);
1781         needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
1782                      && (dstFormat&0xFF)<24
1783                      && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1784
1785         if(!isSupportedIn(srcFormat))
1786         {
1787                 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1788                 return NULL;
1789         }
1790         if(!isSupportedOut(dstFormat))
1791         {
1792                 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1793                 return NULL;
1794         }
1795
1796         /* sanity check */
1797         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
1798         {
1799                  MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
1800                         srcW, srcH, dstW, dstH);
1801                 return NULL;
1802         }
1803
1804         if(!dstFilter) dstFilter= &dummyFilter;
1805         if(!srcFilter) srcFilter= &dummyFilter;
1806
1807         c= memalign(64, sizeof(SwsContext));
1808         memset(c, 0, sizeof(SwsContext));
1809
1810         c->srcW= srcW;
1811         c->srcH= srcH;
1812         c->dstW= dstW;
1813         c->dstH= dstH;
1814         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1815         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1816         c->flags= flags;
1817         c->dstFormat= dstFormat;
1818         c->srcFormat= srcFormat;
1819         c->origDstFormat= origDstFormat;
1820         c->origSrcFormat= origSrcFormat;
1821         c->vRounder= 4* 0x0001000100010001ULL;
1822
1823         usesHFilter= usesVFilter= 0;
1824         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
1825         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
1826         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
1827         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
1828         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
1829         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
1830         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
1831         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
1832
1833         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1834         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1835
1836         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1837         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1838
1839         // drop some chroma lines if the user wants it
1840         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1841         c->chrSrcVSubSample+= c->vChrDrop;
1842
1843         // drop every 2. pixel for chroma calculation unless user wants full chroma
1844         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
1845                 c->chrSrcHSubSample=1;
1846
1847         c->chrIntHSubSample= c->chrDstHSubSample;
1848         c->chrIntVSubSample= c->chrSrcVSubSample;
1849
1850         // note the -((-x)>>y) is so that we allways round toward +inf
1851         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1852         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
1853         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
1854         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
1855
1856         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);
1857
1858         /* unscaled special Cases */
1859         if(unscaled && !usesHFilter && !usesVFilter)
1860         {
1861                 /* yv12_to_nv12 */
1862                 if(srcFormat == IMGFMT_YV12 && dstFormat == IMGFMT_NV12)
1863                 {
1864                         c->swScale= PlanarToNV12Wrapper;
1865                 }
1866                 /* yuv2bgr */
1867                 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
1868                 {
1869                         c->swScale= yuv2rgb_get_func_ptr(c);
1870                 }
1871
1872                 if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 )
1873                 {
1874                         c->swScale= yvu9toyv12Wrapper;
1875                 }
1876
1877                 /* bgr24toYV12 */
1878                 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
1879                         c->swScale= bgr24toyv12Wrapper;
1880
1881                 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
1882                 if(   (isBGR(srcFormat) || isRGB(srcFormat))
1883                    && (isBGR(dstFormat) || isRGB(dstFormat))
1884                    && !needsDither)
1885                         c->swScale= rgb2rgbWrapper;
1886
1887                 /* LQ converters if -sws 0 or -sws 4*/
1888                 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
1889                         /* rgb/bgr -> rgb/bgr (dither needed forms) */
1890                         if(  (isBGR(srcFormat) || isRGB(srcFormat))
1891                           && (isBGR(dstFormat) || isRGB(dstFormat))
1892                           && needsDither)
1893                                 c->swScale= rgb2rgbWrapper;
1894
1895                         /* yv12_to_yuy2 */
1896                         if(srcFormat == IMGFMT_YV12 &&
1897                             (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY))
1898                         {
1899                                 if (dstFormat == IMGFMT_YUY2)
1900                                     c->swScale= PlanarToYuy2Wrapper;
1901                                 else
1902                                     c->swScale= PlanarToUyvyWrapper;
1903                         }
1904                 }
1905
1906 #ifdef HAVE_ALTIVEC
1907                 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
1908                     ((srcFormat == IMGFMT_YV12 &&
1909                       (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY)))) {
1910                   // unscaled YV12 -> packed YUV, we want speed
1911                   if (dstFormat == IMGFMT_YUY2)
1912                     c->swScale= yv12toyuy2_unscaled_altivec;
1913                   else
1914                     c->swScale= yv12touyvy_unscaled_altivec;
1915                 }
1916 #endif
1917
1918                 /* simple copy */
1919                 if(   srcFormat == dstFormat
1920                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
1921                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
1922                   )
1923                 {
1924                         c->swScale= simpleCopy;
1925                 }
1926
1927                 if(c->swScale){
1928                         if(flags&SWS_PRINT_INFO)
1929                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
1930                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
1931                         return c;
1932                 }
1933         }
1934
1935         if(flags & SWS_CPU_CAPS_MMX2)
1936         {
1937                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
1938                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
1939                 {
1940                         if(flags&SWS_PRINT_INFO)
1941                                 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
1942                 }
1943                 if(usesHFilter) c->canMMX2BeUsed=0;
1944         }
1945         else
1946                 c->canMMX2BeUsed=0;
1947
1948         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
1949         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
1950
1951         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
1952         // but only for the FAST_BILINEAR mode otherwise do correct scaling
1953         // n-2 is the last chrominance sample available
1954         // this is not perfect, but noone shuld notice the difference, the more correct variant
1955         // would be like the vertical one, but that would require some special code for the
1956         // first and last pixel
1957         if(flags&SWS_FAST_BILINEAR)
1958         {
1959                 if(c->canMMX2BeUsed)
1960                 {
1961                         c->lumXInc+= 20;
1962                         c->chrXInc+= 20;
1963                 }
1964                 //we don't use the x86asm scaler if mmx is available
1965                 else if(flags & SWS_CPU_CAPS_MMX)
1966                 {
1967                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
1968                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
1969                 }
1970         }
1971
1972         /* precalculate horizontal scaler filter coefficients */
1973         {
1974                 const int filterAlign=
1975                   (flags & SWS_CPU_CAPS_MMX) ? 4 :
1976                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
1977                   1;
1978
1979                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
1980                                  srcW      ,       dstW, filterAlign, 1<<14,
1981                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
1982                                  srcFilter->lumH, dstFilter->lumH);
1983                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
1984                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
1985                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
1986                                  srcFilter->chrH, dstFilter->chrH);
1987
1988 #ifdef ARCH_X86
1989 // can't downscale !!!
1990                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
1991                 {
1992                         c->lumMmx2Filter   = (int16_t*)memalign(8, (dstW        /8+8)*sizeof(int16_t));
1993                         c->chrMmx2Filter   = (int16_t*)memalign(8, (c->chrDstW  /4+8)*sizeof(int16_t));
1994                         c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW      /2/8+8)*sizeof(int32_t));
1995                         c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
1996
1997                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
1998                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
1999                 }
2000 #endif
2001         } // Init Horizontal stuff
2002
2003
2004
2005         /* precalculate vertical scaler filter coefficients */
2006         {
2007                 const int filterAlign=
2008                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2009                   1;
2010
2011                 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2012                                 srcH      ,        dstH, filterAlign, (1<<12)-4,
2013                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2014                                 srcFilter->lumV, dstFilter->lumV);
2015                 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2016                                 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2017                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2018                                 srcFilter->chrV, dstFilter->chrV);
2019         }
2020
2021         // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2022         c->vLumBufSize= c->vLumFilterSize;
2023         c->vChrBufSize= c->vChrFilterSize;
2024         for(i=0; i<dstH; i++)
2025         {
2026                 int chrI= i*c->chrDstH / dstH;
2027                 int nextSlice= MAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
2028                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2029
2030                 nextSlice>>= c->chrSrcVSubSample;
2031                 nextSlice<<= c->chrSrcVSubSample;
2032                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
2033                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
2034                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2035                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2036         }
2037
2038         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2039         c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
2040         c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
2041         //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)
2042         for(i=0; i<c->vLumBufSize; i++)
2043                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
2044         for(i=0; i<c->vChrBufSize; i++)
2045                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
2046
2047         //try to avoid drawing green stuff between the right end and the stride end
2048         for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2049         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2050
2051         ASSERT(c->chrDstH <= dstH)
2052
2053         if(flags&SWS_PRINT_INFO)
2054         {
2055 #ifdef DITHER1XBPP
2056                 char *dither= " dithered";
2057 #else
2058                 char *dither= "";
2059 #endif
2060                 if(flags&SWS_FAST_BILINEAR)
2061                         MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
2062                 else if(flags&SWS_BILINEAR)
2063                         MSG_INFO("\nSwScaler: BILINEAR scaler, ");
2064                 else if(flags&SWS_BICUBIC)
2065                         MSG_INFO("\nSwScaler: BICUBIC scaler, ");
2066                 else if(flags&SWS_X)
2067                         MSG_INFO("\nSwScaler: Experimental scaler, ");
2068                 else if(flags&SWS_POINT)
2069                         MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
2070                 else if(flags&SWS_AREA)
2071                         MSG_INFO("\nSwScaler: Area Averageing scaler, ");
2072                 else if(flags&SWS_BICUBLIN)
2073                         MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2074                 else if(flags&SWS_GAUSS)
2075                         MSG_INFO("\nSwScaler: Gaussian scaler, ");
2076                 else if(flags&SWS_SINC)
2077                         MSG_INFO("\nSwScaler: Sinc scaler, ");
2078                 else if(flags&SWS_LANCZOS)
2079                         MSG_INFO("\nSwScaler: Lanczos scaler, ");
2080                 else if(flags&SWS_SPLINE)
2081                         MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
2082                 else
2083                         MSG_INFO("\nSwScaler: ehh flags invalid?! ");
2084
2085                 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2086                         MSG_INFO("from %s to%s %s ",
2087                                 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
2088                 else
2089                         MSG_INFO("from %s to %s ",
2090                                 vo_format_name(srcFormat), vo_format_name(dstFormat));
2091
2092                 if(flags & SWS_CPU_CAPS_MMX2)
2093                         MSG_INFO("using MMX2\n");
2094                 else if(flags & SWS_CPU_CAPS_3DNOW)
2095                         MSG_INFO("using 3DNOW\n");
2096                 else if(flags & SWS_CPU_CAPS_MMX)
2097                         MSG_INFO("using MMX\n");
2098                 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2099                         MSG_INFO("using AltiVec\n");
2100                 else
2101                         MSG_INFO("using C\n");
2102         }
2103
2104         if(flags & SWS_PRINT_INFO)
2105         {
2106                 if(flags & SWS_CPU_CAPS_MMX)
2107                 {
2108                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2109                                 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2110                         else
2111                         {
2112                                 if(c->hLumFilterSize==4)
2113                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2114                                 else if(c->hLumFilterSize==8)
2115                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2116                                 else
2117                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2118
2119                                 if(c->hChrFilterSize==4)
2120                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2121                                 else if(c->hChrFilterSize==8)
2122                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2123                                 else
2124                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2125                         }
2126                 }
2127                 else
2128                 {
2129 #ifdef ARCH_X86
2130                         MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2131 #else
2132                         if(flags & SWS_FAST_BILINEAR)
2133                                 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2134                         else
2135                                 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2136 #endif
2137                 }
2138                 if(isPlanarYUV(dstFormat))
2139                 {
2140                         if(c->vLumFilterSize==1)
2141                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2142                         else
2143                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2144                 }
2145                 else
2146                 {
2147                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2148                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2149                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2150                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2151                                 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2152                         else
2153                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2154                 }
2155
2156                 if(dstFormat==IMGFMT_BGR24)
2157                         MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2158                                 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2159                 else if(dstFormat==IMGFMT_BGR32)
2160                         MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2161                 else if(dstFormat==IMGFMT_BGR16)
2162                         MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2163                 else if(dstFormat==IMGFMT_BGR15)
2164                         MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2165
2166                 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2167         }
2168         if(flags & SWS_PRINT_INFO)
2169         {
2170                 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2171                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2172                 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2173                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2174         }
2175
2176         c->swScale= getSwsFunc(flags);
2177         return c;
2178 }
2179
2180 /**
2181  * swscale warper, so we don't need to export the SwsContext.
2182  * assumes planar YUV to be in YUV order instead of YVU
2183  */
2184 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2185                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2186         //copy strides, so they can safely be modified
2187         int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
2188         int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
2189         return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2190 }
2191
2192 /**
2193  * swscale warper, so we don't need to export the SwsContext
2194  */
2195 int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
2196                            int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
2197         int srcStride[3];
2198         int dstStride[3];
2199         uint8_t *src[3];
2200         uint8_t *dst[3];
2201         sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam);
2202         sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam);
2203 //printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
2204
2205         return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2206 }
2207
2208 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2209                                 float lumaSharpen, float chromaSharpen,
2210                                 float chromaHShift, float chromaVShift,
2211                                 int verbose)
2212 {
2213         SwsFilter *filter= malloc(sizeof(SwsFilter));
2214
2215         if(lumaGBlur!=0.0){
2216                 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2217                 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2218         }else{
2219                 filter->lumH= sws_getIdentityVec();
2220                 filter->lumV= sws_getIdentityVec();
2221         }
2222
2223         if(chromaGBlur!=0.0){
2224                 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2225                 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2226         }else{
2227                 filter->chrH= sws_getIdentityVec();
2228                 filter->chrV= sws_getIdentityVec();
2229         }
2230
2231         if(chromaSharpen!=0.0){
2232                 SwsVector *g= sws_getConstVec(-1.0, 3);
2233                 SwsVector *id= sws_getConstVec(10.0/chromaSharpen, 1);
2234                 g->coeff[1]=2.0;
2235                 sws_addVec(id, g);
2236                 sws_convVec(filter->chrH, id);
2237                 sws_convVec(filter->chrV, id);
2238                 sws_freeVec(g);
2239                 sws_freeVec(id);
2240         }
2241
2242         if(lumaSharpen!=0.0){
2243                 SwsVector *g= sws_getConstVec(-1.0, 3);
2244                 SwsVector *id= sws_getConstVec(10.0/lumaSharpen, 1);
2245                 g->coeff[1]=2.0;
2246                 sws_addVec(id, g);
2247                 sws_convVec(filter->lumH, id);
2248                 sws_convVec(filter->lumV, id);
2249                 sws_freeVec(g);
2250                 sws_freeVec(id);
2251         }
2252
2253         if(chromaHShift != 0.0)
2254                 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2255
2256         if(chromaVShift != 0.0)
2257                 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2258
2259         sws_normalizeVec(filter->chrH, 1.0);
2260         sws_normalizeVec(filter->chrV, 1.0);
2261         sws_normalizeVec(filter->lumH, 1.0);
2262         sws_normalizeVec(filter->lumV, 1.0);
2263
2264         if(verbose) sws_printVec(filter->chrH);
2265         if(verbose) sws_printVec(filter->lumH);
2266
2267         return filter;
2268 }
2269
2270 /**
2271  * returns a normalized gaussian curve used to filter stuff
2272  * quality=3 is high quality, lowwer is lowwer quality
2273  */
2274 SwsVector *sws_getGaussianVec(double variance, double quality){
2275         const int length= (int)(variance*quality + 0.5) | 1;
2276         int i;
2277         double *coeff= memalign(sizeof(double), length*sizeof(double));
2278         double middle= (length-1)*0.5;
2279         SwsVector *vec= malloc(sizeof(SwsVector));
2280
2281         vec->coeff= coeff;
2282         vec->length= length;
2283
2284         for(i=0; i<length; i++)
2285         {
2286                 double dist= i-middle;
2287                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2288         }
2289
2290         sws_normalizeVec(vec, 1.0);
2291
2292         return vec;
2293 }
2294
2295 SwsVector *sws_getConstVec(double c, int length){
2296         int i;
2297         double *coeff= memalign(sizeof(double), length*sizeof(double));
2298         SwsVector *vec= malloc(sizeof(SwsVector));
2299
2300         vec->coeff= coeff;
2301         vec->length= length;
2302
2303         for(i=0; i<length; i++)
2304                 coeff[i]= c;
2305
2306         return vec;
2307 }
2308
2309
2310 SwsVector *sws_getIdentityVec(void){
2311         double *coeff= memalign(sizeof(double), sizeof(double));
2312         SwsVector *vec= malloc(sizeof(SwsVector));
2313         coeff[0]= 1.0;
2314
2315         vec->coeff= coeff;
2316         vec->length= 1;
2317
2318         return vec;
2319 }
2320
2321 void sws_normalizeVec(SwsVector *a, double height){
2322         int i;
2323         double sum=0;
2324         double inv;
2325
2326         for(i=0; i<a->length; i++)
2327                 sum+= a->coeff[i];
2328
2329         inv= height/sum;
2330
2331         for(i=0; i<a->length; i++)
2332                 a->coeff[i]*= inv;
2333 }
2334
2335 void sws_scaleVec(SwsVector *a, double scalar){
2336         int i;
2337
2338         for(i=0; i<a->length; i++)
2339                 a->coeff[i]*= scalar;
2340 }
2341
2342 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2343         int length= a->length + b->length - 1;
2344         double *coeff= memalign(sizeof(double), length*sizeof(double));
2345         int i, j;
2346         SwsVector *vec= malloc(sizeof(SwsVector));
2347
2348         vec->coeff= coeff;
2349         vec->length= length;
2350
2351         for(i=0; i<length; i++) coeff[i]= 0.0;
2352
2353         for(i=0; i<a->length; i++)
2354         {
2355                 for(j=0; j<b->length; j++)
2356                 {
2357                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2358                 }
2359         }
2360
2361         return vec;
2362 }
2363
2364 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2365         int length= MAX(a->length, b->length);
2366         double *coeff= memalign(sizeof(double), length*sizeof(double));
2367         int i;
2368         SwsVector *vec= malloc(sizeof(SwsVector));
2369
2370         vec->coeff= coeff;
2371         vec->length= length;
2372
2373         for(i=0; i<length; i++) coeff[i]= 0.0;
2374
2375         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2376         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2377
2378         return vec;
2379 }
2380
2381 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2382         int length= MAX(a->length, b->length);
2383         double *coeff= memalign(sizeof(double), length*sizeof(double));
2384         int i;
2385         SwsVector *vec= malloc(sizeof(SwsVector));
2386
2387         vec->coeff= coeff;
2388         vec->length= length;
2389
2390         for(i=0; i<length; i++) coeff[i]= 0.0;
2391
2392         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2393         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2394
2395         return vec;
2396 }
2397
2398 /* shift left / or right if "shift" is negative */
2399 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2400         int length= a->length + ABS(shift)*2;
2401         double *coeff= memalign(sizeof(double), length*sizeof(double));
2402         int i;
2403         SwsVector *vec= malloc(sizeof(SwsVector));
2404
2405         vec->coeff= coeff;
2406         vec->length= length;
2407
2408         for(i=0; i<length; i++) coeff[i]= 0.0;
2409
2410         for(i=0; i<a->length; i++)
2411         {
2412                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2413         }
2414
2415         return vec;
2416 }
2417
2418 void sws_shiftVec(SwsVector *a, int shift){
2419         SwsVector *shifted= sws_getShiftedVec(a, shift);
2420         free(a->coeff);
2421         a->coeff= shifted->coeff;
2422         a->length= shifted->length;
2423         free(shifted);
2424 }
2425
2426 void sws_addVec(SwsVector *a, SwsVector *b){
2427         SwsVector *sum= sws_sumVec(a, b);
2428         free(a->coeff);
2429         a->coeff= sum->coeff;
2430         a->length= sum->length;
2431         free(sum);
2432 }
2433
2434 void sws_subVec(SwsVector *a, SwsVector *b){
2435         SwsVector *diff= sws_diffVec(a, b);
2436         free(a->coeff);
2437         a->coeff= diff->coeff;
2438         a->length= diff->length;
2439         free(diff);
2440 }
2441
2442 void sws_convVec(SwsVector *a, SwsVector *b){
2443         SwsVector *conv= sws_getConvVec(a, b);
2444         free(a->coeff);
2445         a->coeff= conv->coeff;
2446         a->length= conv->length;
2447         free(conv);
2448 }
2449
2450 SwsVector *sws_cloneVec(SwsVector *a){
2451         double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2452         int i;
2453         SwsVector *vec= malloc(sizeof(SwsVector));
2454
2455         vec->coeff= coeff;
2456         vec->length= a->length;
2457
2458         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2459
2460         return vec;
2461 }
2462
2463 void sws_printVec(SwsVector *a){
2464         int i;
2465         double max=0;
2466         double min=0;
2467         double range;
2468
2469         for(i=0; i<a->length; i++)
2470                 if(a->coeff[i]>max) max= a->coeff[i];
2471
2472         for(i=0; i<a->length; i++)
2473                 if(a->coeff[i]<min) min= a->coeff[i];
2474
2475         range= max - min;
2476
2477         for(i=0; i<a->length; i++)
2478         {
2479                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2480                 MSG_DBG2("%1.3f ", a->coeff[i]);
2481                 for(;x>0; x--) MSG_DBG2(" ");
2482                 MSG_DBG2("|\n");
2483         }
2484 }
2485
2486 void sws_freeVec(SwsVector *a){
2487         if(!a) return;
2488         if(a->coeff) free(a->coeff);
2489         a->coeff=NULL;
2490         a->length=0;
2491         free(a);
2492 }
2493
2494 void sws_freeFilter(SwsFilter *filter){
2495         if(!filter) return;
2496
2497         if(filter->lumH) sws_freeVec(filter->lumH);
2498         if(filter->lumV) sws_freeVec(filter->lumV);
2499         if(filter->chrH) sws_freeVec(filter->chrH);
2500         if(filter->chrV) sws_freeVec(filter->chrV);
2501         free(filter);
2502 }
2503
2504
2505 void sws_freeContext(SwsContext *c){
2506         int i;
2507         if(!c) return;
2508
2509         if(c->lumPixBuf)
2510         {
2511                 for(i=0; i<c->vLumBufSize; i++)
2512                 {
2513                         if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2514                         c->lumPixBuf[i]=NULL;
2515                 }
2516                 free(c->lumPixBuf);
2517                 c->lumPixBuf=NULL;
2518         }
2519
2520         if(c->chrPixBuf)
2521         {
2522                 for(i=0; i<c->vChrBufSize; i++)
2523                 {
2524                         if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2525                         c->chrPixBuf[i]=NULL;
2526                 }
2527                 free(c->chrPixBuf);
2528                 c->chrPixBuf=NULL;
2529         }
2530
2531         if(c->vLumFilter) free(c->vLumFilter);
2532         c->vLumFilter = NULL;
2533         if(c->vChrFilter) free(c->vChrFilter);
2534         c->vChrFilter = NULL;
2535         if(c->hLumFilter) free(c->hLumFilter);
2536         c->hLumFilter = NULL;
2537         if(c->hChrFilter) free(c->hChrFilter);
2538         c->hChrFilter = NULL;
2539
2540         if(c->vLumFilterPos) free(c->vLumFilterPos);
2541         c->vLumFilterPos = NULL;
2542         if(c->vChrFilterPos) free(c->vChrFilterPos);
2543         c->vChrFilterPos = NULL;
2544         if(c->hLumFilterPos) free(c->hLumFilterPos);
2545         c->hLumFilterPos = NULL;
2546         if(c->hChrFilterPos) free(c->hChrFilterPos);
2547         c->hChrFilterPos = NULL;
2548
2549         if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2550         c->lumMmx2Filter=NULL;
2551         if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2552         c->chrMmx2Filter=NULL;
2553         if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2554         c->lumMmx2FilterPos=NULL;
2555         if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2556         c->chrMmx2FilterPos=NULL;
2557         if(c->yuvTable) free(c->yuvTable);
2558         c->yuvTable=NULL;
2559
2560         free(c);
2561 }
2562