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

   1 /*
   2     Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
   3
   4     This program is free software; you can redistribute it and/or modify
   5     it under the terms of the GNU General Public License as published by
   6     the Free Software Foundation; either version 2 of the License, or
   7     (at your option) any later version.
   8
   9     This program is distributed in the hope that it will be useful,
  10     but WITHOUT ANY WARRANTY; without even the implied warranty of
  11     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12     GNU General Public License for more details.
  13
  14     You should have received a copy of the GNU General Public License
  15     along with this program; if not, write to the Free Software
  16     Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 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 <unistd.h>
  57 #include "config.h"
  58 #include "mangle.h"
  59 #include <assert.h>
  60 #ifdef HAVE_MALLOC_H
  61 #include <malloc.h>
  62 #else
  63 #include <stdlib.h>
  64 #endif
  65 #ifdef HAVE_SYS_MMAN_H
  66 #include <sys/mman.h>
  67 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
  68 #define MAP_ANONYMOUS MAP_ANON
  69 #endif
  70 #endif
  71 #include "swscale.h"
  72 #include "swscale_internal.h"
  73 #include "cpudetect.h"
  74 #include "bswap.h"
  75 #include "libvo/img_format.h"
  76 #include "rgb2rgb.h"
  77 #include "libvo/fastmemcpy.h"
  78
  79 #undef MOVNTQ
  80 #undef PAVGB
  81
  82 //#undef HAVE_MMX2
  83 //#define HAVE_3DNOW
  84 //#undef HAVE_MMX
  85 //#undef ARCH_X86
  86 //#define WORDS_BIGENDIAN
  87 #define DITHER1XBPP
  88
  89 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
  90
  91 #define RET 0xC3 //near return opcode for X86
  92
  93 #ifdef MP_DEBUG
  94 #define ASSERT(x) assert(x);
  95 #else
  96 #define ASSERT(x) ;
  97 #endif
  98
  99 #ifdef M_PI
 100 #define PI M_PI
 101 #else
 102 #define PI 3.14159265358979323846
 103 #endif
 104
 105 //FIXME replace this with something faster
 106 #define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YVU9 \
 107                         || (x)==IMGFMT_NV12 || (x)==IMGFMT_NV21 \
 108                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
 109 #define isYUV(x)       ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x))
 110 #define isGray(x)      ((x)==IMGFMT_Y800)
 111 #define isRGB(x)       (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
 112 #define isBGR(x)       (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
 113 #define isSupportedIn(x)  ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
 114                         || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
 115                         || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
 116                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\
 117                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
 118 #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
 119                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
 120                         || isRGB(x) || isBGR(x)\
 121                         || (x)==IMGFMT_NV12 || (x)==IMGFMT_NV21\
 122                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
 123 #define isPacked(x)    ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x))
 124
 125 #define RGB2YUV_SHIFT 16
 126 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
 127 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
 128 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
 129 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
 130 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
 131 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
 132 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
 133 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
 134 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
 135
 136 extern const int32_t Inverse_Table_6_9[8][4];
 137
 138 /*
 139 NOTES
 140 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
 141
 142 TODO
 143 more intelligent missalignment avoidance for the horizontal scaler
 144 write special vertical cubic upscale version
 145 Optimize C code (yv12 / minmax)
 146 add support for packed pixel yuv input & output
 147 add support for Y8 output
 148 optimize bgr24 & bgr32
 149 add BGR4 output support
 150 write special BGR->BGR scaler
 151 */
 152
 153 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
 154 #define MIN(a,b) ((a) > (b) ? (b) : (a))
 155 #define MAX(a,b) ((a) < (b) ? (b) : (a))
 156
 157 #if defined(ARCH_X86) || defined(ARCH_X86_64)
 158 static uint64_t attribute_used __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
 159 static uint64_t attribute_used __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
 160 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
 161 static uint64_t attribute_used __attribute__((aligned(8))) w02=       0x0002000200020002LL;
 162 static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
 163 static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
 164 static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
 165 static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
 166
 167 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
 168 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
 169 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
 170 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
 171
 172 static uint64_t __attribute__((aligned(8))) dither4[2]={
 173         0x0103010301030103LL,
 174         0x0200020002000200LL,};
 175
 176 static uint64_t __attribute__((aligned(8))) dither8[2]={
 177         0x0602060206020602LL,
 178         0x0004000400040004LL,};
 179
 180 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
 181 static uint64_t attribute_used __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
 182 static uint64_t attribute_used __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
 183 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
 184 static uint64_t attribute_used __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
 185 static uint64_t attribute_used __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
 186
 187 static uint64_t attribute_used __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
 188 static uint64_t attribute_used __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
 189 static uint64_t attribute_used __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
 190
 191 #ifdef FAST_BGR2YV12
 192 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
 193 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
 194 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
 195 #else
 196 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
 197 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
 198 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
 199 #endif
 200 static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
 201 static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
 202 static const uint64_t w1111       attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
 203 #endif
 204
 205 // clipping helper table for C implementations:
 206 static unsigned char clip_table[768];
 207
 208 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
 209
 210 extern const uint8_t dither_2x2_4[2][8];
 211 extern const uint8_t dither_2x2_8[2][8];
 212 extern const uint8_t dither_8x8_32[8][8];
 213 extern const uint8_t dither_8x8_73[8][8];
 214 extern const uint8_t dither_8x8_220[8][8];
 215
 216 #if defined(ARCH_X86) || defined(ARCH_X86_64)
 217 void in_asm_used_var_warning_killer()
 218 {
 219  volatile int i= bF8+bFC+w10+
 220  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
 221  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
 222  if(i) i=0;
 223 }
 224 #endif
 225
 226 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
 227                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
 228                                     uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
 229 {
 230         //FIXME Optimize (just quickly writen not opti..)
 231         int i;
 232         for(i=0; i<dstW; i++)
 233         {
 234                 int val=1<<18;
 235                 int j;
 236                 for(j=0; j<lumFilterSize; j++)
 237                         val += lumSrc[j][i] * lumFilter[j];
 238
 239                 dest[i]= MIN(MAX(val>>19, 0), 255);
 240         }
 241
 242         if(uDest != NULL)
 243                 for(i=0; i<chrDstW; i++)
 244                 {
 245                         int u=1<<18;
 246                         int v=1<<18;
 247                         int j;
 248                         for(j=0; j<chrFilterSize; j++)
 249                         {
 250                                 u += chrSrc[j][i] * chrFilter[j];
 251                                 v += chrSrc[j][i + 2048] * chrFilter[j];
 252                         }
 253
 254                         uDest[i]= MIN(MAX(u>>19, 0), 255);
 255                         vDest[i]= MIN(MAX(v>>19, 0), 255);
 256                 }
 257 }
 258
 259 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
 260                                 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
 261                                 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
 262 {
 263         //FIXME Optimize (just quickly writen not opti..)
 264         int i;
 265         for(i=0; i<dstW; i++)
 266         {
 267                 int val=1<<18;
 268                 int j;
 269                 for(j=0; j<lumFilterSize; j++)
 270                         val += lumSrc[j][i] * lumFilter[j];
 271
 272                 dest[i]= MIN(MAX(val>>19, 0), 255);
 273         }
 274
 275         if(uDest == NULL)
 276                 return;
 277
 278         if(dstFormat == IMGFMT_NV12)
 279                 for(i=0; i<chrDstW; i++)
 280                 {
 281                         int u=1<<18;
 282                         int v=1<<18;
 283                         int j;
 284                         for(j=0; j<chrFilterSize; j++)
 285                         {
 286                                 u += chrSrc[j][i] * chrFilter[j];
 287                                 v += chrSrc[j][i + 2048] * chrFilter[j];
 288                         }
 289
 290                         uDest[2*i]= MIN(MAX(u>>19, 0), 255);
 291                         uDest[2*i+1]= MIN(MAX(v>>19, 0), 255);
 292                 }
 293         else
 294                 for(i=0; i<chrDstW; i++)
 295                 {
 296                         int u=1<<18;
 297                         int v=1<<18;
 298                         int j;
 299                         for(j=0; j<chrFilterSize; j++)
 300                         {
 301                                 u += chrSrc[j][i] * chrFilter[j];
 302                                 v += chrSrc[j][i + 2048] * chrFilter[j];
 303                         }
 304
 305                         uDest[2*i]= MIN(MAX(v>>19, 0), 255);
 306                         uDest[2*i+1]= MIN(MAX(u>>19, 0), 255);
 307                 }
 308 }
 309
 310 #define YSCALE_YUV_2_PACKEDX_C(type) \
 311                 for(i=0; i<(dstW>>1); i++){\
 312                         int j;\
 313                         int Y1=1<<18;\
 314                         int Y2=1<<18;\
 315                         int U=1<<18;\
 316                         int V=1<<18;\
 317                         type *r, *b, *g;\
 318                         const int i2= 2*i;\
 319                         \
 320                         for(j=0; j<lumFilterSize; j++)\
 321                         {\
 322                                 Y1 += lumSrc[j][i2] * lumFilter[j];\
 323                                 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
 324                         }\
 325                         for(j=0; j<chrFilterSize; j++)\
 326                         {\
 327                                 U += chrSrc[j][i] * chrFilter[j];\
 328                                 V += chrSrc[j][i+2048] * chrFilter[j];\
 329                         }\
 330                         Y1>>=19;\
 331                         Y2>>=19;\
 332                         U >>=19;\
 333                         V >>=19;\
 334                         if((Y1|Y2|U|V)&256)\
 335                         {\
 336                                 if(Y1>255)   Y1=255;\
 337                                 else if(Y1<0)Y1=0;\
 338                                 if(Y2>255)   Y2=255;\
 339                                 else if(Y2<0)Y2=0;\
 340                                 if(U>255)    U=255;\
 341                                 else if(U<0) U=0;\
 342                                 if(V>255)    V=255;\
 343                                 else if(V<0) V=0;\
 344                         }
 345
 346 #define YSCALE_YUV_2_RGBX_C(type) \
 347                         YSCALE_YUV_2_PACKEDX_C(type)\
 348                         r = c->table_rV[V];\
 349                         g = c->table_gU[U] + c->table_gV[V];\
 350                         b = c->table_bU[U];\
 351
 352 #define YSCALE_YUV_2_PACKED2_C \
 353                 for(i=0; i<(dstW>>1); i++){\
 354                         const int i2= 2*i;\
 355                         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
 356                         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
 357                         int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
 358                         int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
 359
 360 #define YSCALE_YUV_2_RGB2_C(type) \
 361                         YSCALE_YUV_2_PACKED2_C\
 362                         type *r, *b, *g;\
 363                         r = c->table_rV[V];\
 364                         g = c->table_gU[U] + c->table_gV[V];\
 365                         b = c->table_bU[U];\
 366
 367 #define YSCALE_YUV_2_PACKED1_C \
 368                 for(i=0; i<(dstW>>1); i++){\
 369                         const int i2= 2*i;\
 370                         int Y1= buf0[i2  ]>>7;\
 371                         int Y2= buf0[i2+1]>>7;\
 372                         int U= (uvbuf1[i     ])>>7;\
 373                         int V= (uvbuf1[i+2048])>>7;\
 374
 375 #define YSCALE_YUV_2_RGB1_C(type) \
 376                         YSCALE_YUV_2_PACKED1_C\
 377                         type *r, *b, *g;\
 378                         r = c->table_rV[V];\
 379                         g = c->table_gU[U] + c->table_gV[V];\
 380                         b = c->table_bU[U];\
 381
 382 #define YSCALE_YUV_2_PACKED1B_C \
 383                 for(i=0; i<(dstW>>1); i++){\
 384                         const int i2= 2*i;\
 385                         int Y1= buf0[i2  ]>>7;\
 386                         int Y2= buf0[i2+1]>>7;\
 387                         int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
 388                         int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
 389
 390 #define YSCALE_YUV_2_RGB1B_C(type) \
 391                         YSCALE_YUV_2_PACKED1B_C\
 392                         type *r, *b, *g;\
 393                         r = c->table_rV[V];\
 394                         g = c->table_gU[U] + c->table_gV[V];\
 395                         b = c->table_bU[U];\
 396
 397 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
 398         switch(c->dstFormat)\
 399         {\
 400         case IMGFMT_BGR32:\
 401         case IMGFMT_RGB32:\
 402                 func(uint32_t)\
 403                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
 404                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
 405                 }               \
 406                 break;\
 407         case IMGFMT_RGB24:\
 408                 func(uint8_t)\
 409                         ((uint8_t*)dest)[0]= r[Y1];\
 410                         ((uint8_t*)dest)[1]= g[Y1];\
 411                         ((uint8_t*)dest)[2]= b[Y1];\
 412                         ((uint8_t*)dest)[3]= r[Y2];\
 413                         ((uint8_t*)dest)[4]= g[Y2];\
 414                         ((uint8_t*)dest)[5]= b[Y2];\
 415                         dest+=6;\
 416                 }\
 417                 break;\
 418         case IMGFMT_BGR24:\
 419                 func(uint8_t)\
 420                         ((uint8_t*)dest)[0]= b[Y1];\
 421                         ((uint8_t*)dest)[1]= g[Y1];\
 422                         ((uint8_t*)dest)[2]= r[Y1];\
 423                         ((uint8_t*)dest)[3]= b[Y2];\
 424                         ((uint8_t*)dest)[4]= g[Y2];\
 425                         ((uint8_t*)dest)[5]= r[Y2];\
 426                         dest+=6;\
 427                 }\
 428                 break;\
 429         case IMGFMT_RGB16:\
 430         case IMGFMT_BGR16:\
 431                 {\
 432                         const int dr1= dither_2x2_8[y&1    ][0];\
 433                         const int dg1= dither_2x2_4[y&1    ][0];\
 434                         const int db1= dither_2x2_8[(y&1)^1][0];\
 435                         const int dr2= dither_2x2_8[y&1    ][1];\
 436                         const int dg2= dither_2x2_4[y&1    ][1];\
 437                         const int db2= dither_2x2_8[(y&1)^1][1];\
 438                         func(uint16_t)\
 439                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
 440                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
 441                         }\
 442                 }\
 443                 break;\
 444         case IMGFMT_RGB15:\
 445         case IMGFMT_BGR15:\
 446                 {\
 447                         const int dr1= dither_2x2_8[y&1    ][0];\
 448                         const int dg1= dither_2x2_8[y&1    ][1];\
 449                         const int db1= dither_2x2_8[(y&1)^1][0];\
 450                         const int dr2= dither_2x2_8[y&1    ][1];\
 451                         const int dg2= dither_2x2_8[y&1    ][0];\
 452                         const int db2= dither_2x2_8[(y&1)^1][1];\
 453                         func(uint16_t)\
 454                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
 455                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
 456                         }\
 457                 }\
 458                 break;\
 459         case IMGFMT_RGB8:\
 460         case IMGFMT_BGR8:\
 461                 {\
 462                         const uint8_t * const d64= dither_8x8_73[y&7];\
 463                         const uint8_t * const d32= dither_8x8_32[y&7];\
 464                         func(uint8_t)\
 465                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
 466                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
 467                         }\
 468                 }\
 469                 break;\
 470         case IMGFMT_RGB4:\
 471         case IMGFMT_BGR4:\
 472                 {\
 473                         const uint8_t * const d64= dither_8x8_73 [y&7];\
 474                         const uint8_t * const d128=dither_8x8_220[y&7];\
 475                         func(uint8_t)\
 476                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
 477                                                  + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
 478                         }\
 479                 }\
 480                 break;\
 481         case IMGFMT_RG4B:\
 482         case IMGFMT_BG4B:\
 483                 {\
 484                         const uint8_t * const d64= dither_8x8_73 [y&7];\
 485                         const uint8_t * const d128=dither_8x8_220[y&7];\
 486                         func(uint8_t)\
 487                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
 488                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
 489                         }\
 490                 }\
 491                 break;\
 492         case IMGFMT_RGB1:\
 493         case IMGFMT_BGR1:\
 494                 {\
 495                         const uint8_t * const d128=dither_8x8_220[y&7];\
 496                         uint8_t *g= c->table_gU[128] + c->table_gV[128];\
 497                         for(i=0; i<dstW-7; i+=8){\
 498                                 int acc;\
 499                                 acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
 500                                 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
 501                                 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
 502                                 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
 503                                 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
 504                                 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
 505                                 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
 506                                 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
 507                                 ((uint8_t*)dest)[0]= acc;\
 508                                 dest++;\
 509                         }\
 510 \
 511 /*\
 512 ((uint8_t*)dest)-= dstW>>4;\
 513 {\
 514                         int acc=0;\
 515                         int left=0;\
 516                         static int top[1024];\
 517                         static int last_new[1024][1024];\
 518                         static int last_in3[1024][1024];\
 519                         static int drift[1024][1024];\
 520                         int topLeft=0;\
 521                         int shift=0;\
 522                         int count=0;\
 523                         const uint8_t * const d128=dither_8x8_220[y&7];\
 524                         int error_new=0;\
 525                         int error_in3=0;\
 526                         int f=0;\
 527                         \
 528                         for(i=dstW>>1; i<dstW; i++){\
 529                                 int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
 530                                 int in2 = (76309 * (in - 16) + 32768) >> 16;\
 531                                 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
 532                                 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
 533                                         + (last_new[y][i] - in3)*f/256;\
 534                                 int new= old> 128 ? 255 : 0;\
 535 \
 536                                 error_new+= ABS(last_new[y][i] - new);\
 537                                 error_in3+= ABS(last_in3[y][i] - in3);\
 538                                 f= error_new - error_in3*4;\
 539                                 if(f<0) f=0;\
 540                                 if(f>256) f=256;\
 541 \
 542                                 topLeft= top[i];\
 543                                 left= top[i]= old - new;\
 544                                 last_new[y][i]= new;\
 545                                 last_in3[y][i]= in3;\
 546 \
 547                                 acc+= acc + (new&1);\
 548                                 if((i&7)==6){\
 549                                         ((uint8_t*)dest)[0]= acc;\
 550                                         ((uint8_t*)dest)++;\
 551                                 }\
 552                         }\
 553 }\
 554 */\
 555                 }\
 556                 break;\
 557         case IMGFMT_YUY2:\
 558                 func2\
 559                         ((uint8_t*)dest)[2*i2+0]= Y1;\
 560                         ((uint8_t*)dest)[2*i2+1]= U;\
 561                         ((uint8_t*)dest)[2*i2+2]= Y2;\
 562                         ((uint8_t*)dest)[2*i2+3]= V;\
 563                 }               \
 564                 break;\
 565         case IMGFMT_UYVY:\
 566                 func2\
 567                         ((uint8_t*)dest)[2*i2+0]= U;\
 568                         ((uint8_t*)dest)[2*i2+1]= Y1;\
 569                         ((uint8_t*)dest)[2*i2+2]= V;\
 570                         ((uint8_t*)dest)[2*i2+3]= Y2;\
 571                 }               \
 572                 break;\
 573         }\
 574
 575
 576 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
 577                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
 578                                     uint8_t *dest, int dstW, int y)
 579 {
 580         int i;
 581         switch(c->dstFormat)
 582         {
 583         case IMGFMT_RGB32:
 584         case IMGFMT_BGR32:
 585                 YSCALE_YUV_2_RGBX_C(uint32_t)
 586                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
 587                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
 588                 }
 589                 break;
 590         case IMGFMT_RGB24:
 591                 YSCALE_YUV_2_RGBX_C(uint8_t)
 592                         ((uint8_t*)dest)[0]= r[Y1];
 593                         ((uint8_t*)dest)[1]= g[Y1];
 594                         ((uint8_t*)dest)[2]= b[Y1];
 595                         ((uint8_t*)dest)[3]= r[Y2];
 596                         ((uint8_t*)dest)[4]= g[Y2];
 597                         ((uint8_t*)dest)[5]= b[Y2];
 598                         dest+=6;
 599                 }
 600                 break;
 601         case IMGFMT_BGR24:
 602                 YSCALE_YUV_2_RGBX_C(uint8_t)
 603                         ((uint8_t*)dest)[0]= b[Y1];
 604                         ((uint8_t*)dest)[1]= g[Y1];
 605                         ((uint8_t*)dest)[2]= r[Y1];
 606                         ((uint8_t*)dest)[3]= b[Y2];
 607                         ((uint8_t*)dest)[4]= g[Y2];
 608                         ((uint8_t*)dest)[5]= r[Y2];
 609                         dest+=6;
 610                 }
 611                 break;
 612         case IMGFMT_RGB16:
 613         case IMGFMT_BGR16:
 614                 {
 615                         const int dr1= dither_2x2_8[y&1    ][0];
 616                         const int dg1= dither_2x2_4[y&1    ][0];
 617                         const int db1= dither_2x2_8[(y&1)^1][0];
 618                         const int dr2= dither_2x2_8[y&1    ][1];
 619                         const int dg2= dither_2x2_4[y&1    ][1];
 620                         const int db2= dither_2x2_8[(y&1)^1][1];
 621                         YSCALE_YUV_2_RGBX_C(uint16_t)
 622                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
 623                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
 624                         }
 625                 }
 626                 break;
 627         case IMGFMT_RGB15:
 628         case IMGFMT_BGR15:
 629                 {
 630                         const int dr1= dither_2x2_8[y&1    ][0];
 631                         const int dg1= dither_2x2_8[y&1    ][1];
 632                         const int db1= dither_2x2_8[(y&1)^1][0];
 633                         const int dr2= dither_2x2_8[y&1    ][1];
 634                         const int dg2= dither_2x2_8[y&1    ][0];
 635                         const int db2= dither_2x2_8[(y&1)^1][1];
 636                         YSCALE_YUV_2_RGBX_C(uint16_t)
 637                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
 638                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
 639                         }
 640                 }
 641                 break;
 642         case IMGFMT_RGB8:
 643         case IMGFMT_BGR8:
 644                 {
 645                         const uint8_t * const d64= dither_8x8_73[y&7];
 646                         const uint8_t * const d32= dither_8x8_32[y&7];
 647                         YSCALE_YUV_2_RGBX_C(uint8_t)
 648                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
 649                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
 650                         }
 651                 }
 652                 break;
 653         case IMGFMT_RGB4:
 654         case IMGFMT_BGR4:
 655                 {
 656                         const uint8_t * const d64= dither_8x8_73 [y&7];
 657                         const uint8_t * const d128=dither_8x8_220[y&7];
 658                         YSCALE_YUV_2_RGBX_C(uint8_t)
 659                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
 660                                                   +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
 661                         }
 662                 }
 663                 break;
 664         case IMGFMT_RG4B:
 665         case IMGFMT_BG4B:
 666                 {
 667                         const uint8_t * const d64= dither_8x8_73 [y&7];
 668                         const uint8_t * const d128=dither_8x8_220[y&7];
 669                         YSCALE_YUV_2_RGBX_C(uint8_t)
 670                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
 671                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
 672                         }
 673                 }
 674                 break;
 675         case IMGFMT_RGB1:
 676         case IMGFMT_BGR1:
 677                 {
 678                         const uint8_t * const d128=dither_8x8_220[y&7];
 679                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
 680                         int acc=0;
 681                         for(i=0; i<dstW-1; i+=2){
 682                                 int j;
 683                                 int Y1=1<<18;
 684                                 int Y2=1<<18;
 685
 686                                 for(j=0; j<lumFilterSize; j++)
 687                                 {
 688                                         Y1 += lumSrc[j][i] * lumFilter[j];
 689                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
 690                                 }
 691                                 Y1>>=19;
 692                                 Y2>>=19;
 693                                 if((Y1|Y2)&256)
 694                                 {
 695                                         if(Y1>255)   Y1=255;
 696                                         else if(Y1<0)Y1=0;
 697                                         if(Y2>255)   Y2=255;
 698                                         else if(Y2<0)Y2=0;
 699                                 }
 700                                 acc+= acc + g[Y1+d128[(i+0)&7]];
 701                                 acc+= acc + g[Y2+d128[(i+1)&7]];
 702                                 if((i&7)==6){
 703                                         ((uint8_t*)dest)[0]= acc;
 704                                         dest++;
 705                                 }
 706                         }
 707                 }
 708                 break;
 709         case IMGFMT_YUY2:
 710                 YSCALE_YUV_2_PACKEDX_C(void)
 711                         ((uint8_t*)dest)[2*i2+0]= Y1;
 712                         ((uint8_t*)dest)[2*i2+1]= U;
 713                         ((uint8_t*)dest)[2*i2+2]= Y2;
 714                         ((uint8_t*)dest)[2*i2+3]= V;
 715                 }
 716                 break;
 717         case IMGFMT_UYVY:
 718                 YSCALE_YUV_2_PACKEDX_C(void)
 719                         ((uint8_t*)dest)[2*i2+0]= U;
 720                         ((uint8_t*)dest)[2*i2+1]= Y1;
 721                         ((uint8_t*)dest)[2*i2+2]= V;
 722                         ((uint8_t*)dest)[2*i2+3]= Y2;
 723                 }
 724                 break;
 725         }
 726 }
 727
 728
 729 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
 730 //Plain C versions
 731 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
 732 #define COMPILE_C
 733 #endif
 734
 735 #ifdef ARCH_POWERPC
 736 #if defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)
 737 #define COMPILE_ALTIVEC
 738 #endif //HAVE_ALTIVEC
 739 #endif //ARCH_POWERPC
 740
 741 #if defined(ARCH_X86) || defined(ARCH_X86_64)
 742
 743 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
 744 #define COMPILE_MMX
 745 #endif
 746
 747 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
 748 #define COMPILE_MMX2
 749 #endif
 750
 751 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
 752 #define COMPILE_3DNOW
 753 #endif
 754 #endif //ARCH_X86 || ARCH_X86_64
 755
 756 #undef HAVE_MMX
 757 #undef HAVE_MMX2
 758 #undef HAVE_3DNOW
 759
 760 #ifdef COMPILE_C
 761 #undef HAVE_MMX
 762 #undef HAVE_MMX2
 763 #undef HAVE_3DNOW
 764 #undef HAVE_ALTIVEC
 765 #define RENAME(a) a ## _C
 766 #include "swscale_template.c"
 767 #endif
 768
 769 #ifdef ARCH_POWERPC
 770 #ifdef COMPILE_ALTIVEC
 771 #undef RENAME
 772 #define HAVE_ALTIVEC
 773 #define RENAME(a) a ## _altivec
 774 #include "swscale_template.c"
 775 #endif
 776 #endif //ARCH_POWERPC
 777
 778 #if defined(ARCH_X86) || defined(ARCH_X86_64)
 779
 780 //X86 versions
 781 /*
 782 #undef RENAME
 783 #undef HAVE_MMX
 784 #undef HAVE_MMX2
 785 #undef HAVE_3DNOW
 786 #define ARCH_X86
 787 #define RENAME(a) a ## _X86
 788 #include "swscale_template.c"
 789 */
 790 //MMX versions
 791 #ifdef COMPILE_MMX
 792 #undef RENAME
 793 #define HAVE_MMX
 794 #undef HAVE_MMX2
 795 #undef HAVE_3DNOW
 796 #define RENAME(a) a ## _MMX
 797 #include "swscale_template.c"
 798 #endif
 799
 800 //MMX2 versions
 801 #ifdef COMPILE_MMX2
 802 #undef RENAME
 803 #define HAVE_MMX
 804 #define HAVE_MMX2
 805 #undef HAVE_3DNOW
 806 #define RENAME(a) a ## _MMX2
 807 #include "swscale_template.c"
 808 #endif
 809
 810 //3DNOW versions
 811 #ifdef COMPILE_3DNOW
 812 #undef RENAME
 813 #define HAVE_MMX
 814 #undef HAVE_MMX2
 815 #define HAVE_3DNOW
 816 #define RENAME(a) a ## _3DNow
 817 #include "swscale_template.c"
 818 #endif
 819
 820 #endif //ARCH_X86 || ARCH_X86_64
 821
 822 // minor note: the HAVE_xyz is messed up after that line so don't use it
 823
 824 static double getSplineCoeff(double a, double b, double c, double d, double dist)
 825 {
 826 //      printf("%f %f %f %f %f\n", a,b,c,d,dist);
 827         if(dist<=1.0)   return ((d*dist + c)*dist + b)*dist +a;
 828         else            return getSplineCoeff(  0.0,
 829                                                  b+ 2.0*c + 3.0*d,
 830                                                         c + 3.0*d,
 831                                                 -b- 3.0*c - 6.0*d,
 832                                                 dist-1.0);
 833 }
 834
 835 static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
 836                               int srcW, int dstW, int filterAlign, int one, int flags,
 837                               SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
 838 {
 839         int i;
 840         int filterSize;
 841         int filter2Size;
 842         int minFilterSize;
 843         double *filter=NULL;
 844         double *filter2=NULL;
 845 #if defined(ARCH_X86) || defined(ARCH_X86_64)
 846         if(flags & SWS_CPU_CAPS_MMX)
 847                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
 848 #endif
 849
 850         // Note the +1 is for the MMXscaler which reads over the end
 851         *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
 852
 853         if(ABS(xInc - 0x10000) <10) // unscaled
 854         {
 855                 int i;
 856                 filterSize= 1;
 857                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
 858                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
 859
 860                 for(i=0; i<dstW; i++)
 861                 {
 862                         filter[i*filterSize]=1;
 863                         (*filterPos)[i]=i;
 864                 }
 865
 866         }
 867         else if(flags&SWS_POINT) // lame looking point sampling mode
 868         {
 869                 int i;
 870                 int xDstInSrc;
 871                 filterSize= 1;
 872                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
 873
 874                 xDstInSrc= xInc/2 - 0x8000;
 875                 for(i=0; i<dstW; i++)
 876                 {
 877                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
 878
 879                         (*filterPos)[i]= xx;
 880                         filter[i]= 1.0;
 881                         xDstInSrc+= xInc;
 882                 }
 883         }
 884         else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
 885         {
 886                 int i;
 887                 int xDstInSrc;
 888                 if     (flags&SWS_BICUBIC) filterSize= 4;
 889                 else if(flags&SWS_X      ) filterSize= 4;
 890                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA
 891                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
 892
 893                 xDstInSrc= xInc/2 - 0x8000;
 894                 for(i=0; i<dstW; i++)
 895                 {
 896                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
 897                         int j;
 898
 899                         (*filterPos)[i]= xx;
 900                                 //Bilinear upscale / linear interpolate / Area averaging
 901                                 for(j=0; j<filterSize; j++)
 902                                 {
 903                                         double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
 904                                         double coeff= 1.0 - d;
 905                                         if(coeff<0) coeff=0;
 906                                         filter[i*filterSize + j]= coeff;
 907                                         xx++;
 908                                 }
 909                         xDstInSrc+= xInc;
 910                 }
 911         }
 912         else
 913         {
 914                 double xDstInSrc;
 915                 double sizeFactor, filterSizeInSrc;
 916                 const double xInc1= (double)xInc / (double)(1<<16);
 917
 918                 if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
 919                 else if(flags&SWS_X)            sizeFactor= 8.0;
 920                 else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
 921                 else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
 922                 else if(flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
 923                 else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
 924                 else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
 925                 else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
 926                 else {
 927                         sizeFactor= 0.0; //GCC warning killer
 928                         ASSERT(0)
 929                 }
 930
 931                 if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
 932                 else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
 933
 934                 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
 935                 if(filterSize > srcW-2) filterSize=srcW-2;
 936
 937                 filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
 938
 939                 xDstInSrc= xInc1 / 2.0 - 0.5;
 940                 for(i=0; i<dstW; i++)
 941                 {
 942                         int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
 943                         int j;
 944                         (*filterPos)[i]= xx;
 945                         for(j=0; j<filterSize; j++)
 946                         {
 947                                 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
 948                                 double coeff;
 949                                 if(flags & SWS_BICUBIC)
 950                                 {
 951                                         double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
 952                                         double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
 953
 954                                         if(d<1.0)
 955                                                 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
 956                                         else if(d<2.0)
 957                                                 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
 958                                         else
 959                                                 coeff=0.0;
 960                                 }
 961 /*                              else if(flags & SWS_X)
 962                                 {
 963                                         double p= param ? param*0.01 : 0.3;
 964                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
 965                                         coeff*= pow(2.0, - p*d*d);
 966                                 }*/
 967                                 else if(flags & SWS_X)
 968                                 {
 969                                         double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
 970
 971                                         if(d<1.0)
 972                                                 coeff = cos(d*PI);
 973                                         else
 974                                                 coeff=-1.0;
 975                                         if(coeff<0.0)   coeff= -pow(-coeff, A);
 976                                         else            coeff=  pow( coeff, A);
 977                                         coeff= coeff*0.5 + 0.5;
 978                                 }
 979                                 else if(flags & SWS_AREA)
 980                                 {
 981                                         double srcPixelSize= 1.0/xInc1;
 982                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
 983                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
 984                                         else coeff=0.0;
 985                                 }
 986                                 else if(flags & SWS_GAUSS)
 987                                 {
 988                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
 989                                         coeff = pow(2.0, - p*d*d);
 990                                 }
 991                                 else if(flags & SWS_SINC)
 992                                 {
 993                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
 994                                 }
 995                                 else if(flags & SWS_LANCZOS)
 996                                 {
 997                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
 998                                         coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
 999                                         if(d>p) coeff=0;
1000                                 }
1001                                 else if(flags & SWS_BILINEAR)
1002                                 {
1003                                         coeff= 1.0 - d;
1004                                         if(coeff<0) coeff=0;
1005                                 }
1006                                 else if(flags & SWS_SPLINE)
1007                                 {
1008                                         double p=-2.196152422706632;
1009                                         coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1010                                 }
1011                                 else {
1012                                         coeff= 0.0; //GCC warning killer
1013                                         ASSERT(0)
1014                                 }
1015
1016                                 filter[i*filterSize + j]= coeff;
1017                                 xx++;
1018                         }
1019                         xDstInSrc+= xInc1;
1020                 }
1021         }
1022
1023         /* apply src & dst Filter to filter -> filter2
1024            free(filter);
1025         */
1026         ASSERT(filterSize>0)
1027         filter2Size= filterSize;
1028         if(srcFilter) filter2Size+= srcFilter->length - 1;
1029         if(dstFilter) filter2Size+= dstFilter->length - 1;
1030         ASSERT(filter2Size>0)
1031         filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
1032
1033         for(i=0; i<dstW; i++)
1034         {
1035                 int j;
1036                 SwsVector scaleFilter;
1037                 SwsVector *outVec;
1038
1039                 scaleFilter.coeff= filter + i*filterSize;
1040                 scaleFilter.length= filterSize;
1041
1042                 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1043                 else          outVec= &scaleFilter;
1044
1045                 ASSERT(outVec->length == filter2Size)
1046                 //FIXME dstFilter
1047
1048                 for(j=0; j<outVec->length; j++)
1049                 {
1050                         filter2[i*filter2Size + j]= outVec->coeff[j];
1051                 }
1052
1053                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1054
1055                 if(outVec != &scaleFilter) sws_freeVec(outVec);
1056         }
1057         free(filter); filter=NULL;
1058
1059         /* try to reduce the filter-size (step1 find size and shift left) */
1060         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1061         minFilterSize= 0;
1062         for(i=dstW-1; i>=0; i--)
1063         {
1064                 int min= filter2Size;
1065                 int j;
1066                 double cutOff=0.0;
1067
1068                 /* get rid off near zero elements on the left by shifting left */
1069                 for(j=0; j<filter2Size; j++)
1070                 {
1071                         int k;
1072                         cutOff += ABS(filter2[i*filter2Size]);
1073
1074                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1075
1076                         /* preserve Monotonicity because the core can't handle the filter otherwise */
1077                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1078
1079                         // Move filter coeffs left
1080                         for(k=1; k<filter2Size; k++)
1081                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1082                         filter2[i*filter2Size + k - 1]= 0.0;
1083                         (*filterPos)[i]++;
1084                 }
1085
1086                 cutOff=0.0;
1087                 /* count near zeros on the right */
1088                 for(j=filter2Size-1; j>0; j--)
1089                 {
1090                         cutOff += ABS(filter2[i*filter2Size + j]);
1091
1092                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1093                         min--;
1094                 }
1095
1096                 if(min>minFilterSize) minFilterSize= min;
1097         }
1098
1099         if (flags & SWS_CPU_CAPS_ALTIVEC) {
1100           // we can handle the special case 4,
1101           // so we don't want to go to the full 8
1102           if (minFilterSize < 5)
1103             filterAlign = 4;
1104
1105           // we really don't want to waste our time
1106           // doing useless computation, so fall-back on
1107           // the scalar C code for very small filter.
1108           // vectorizing is worth it only if you have
1109           // decent-sized vector.
1110           if (minFilterSize < 3)
1111             filterAlign = 1;
1112         }
1113
1114         ASSERT(minFilterSize > 0)
1115         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1116         ASSERT(filterSize > 0)
1117         filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1118         *outFilterSize= filterSize;
1119
1120         if(flags&SWS_PRINT_INFO)
1121                 MSG_V("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1122         /* try to reduce the filter-size (step2 reduce it) */
1123         for(i=0; i<dstW; i++)
1124         {
1125                 int j;
1126
1127                 for(j=0; j<filterSize; j++)
1128                 {
1129                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1130                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1131                 }
1132         }
1133         free(filter2); filter2=NULL;
1134
1135
1136         //FIXME try to align filterpos if possible
1137
1138         //fix borders
1139         for(i=0; i<dstW; i++)
1140         {
1141                 int j;
1142                 if((*filterPos)[i] < 0)
1143                 {
1144                         // Move filter coeffs left to compensate for filterPos
1145                         for(j=1; j<filterSize; j++)
1146                         {
1147                                 int left= MAX(j + (*filterPos)[i], 0);
1148                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1149                                 filter[i*filterSize + j]=0;
1150                         }
1151                         (*filterPos)[i]= 0;
1152                 }
1153
1154                 if((*filterPos)[i] + filterSize > srcW)
1155                 {
1156                         int shift= (*filterPos)[i] + filterSize - srcW;
1157                         // Move filter coeffs right to compensate for filterPos
1158                         for(j=filterSize-2; j>=0; j--)
1159                         {
1160                                 int right= MIN(j + shift, filterSize-1);
1161                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1162                                 filter[i*filterSize +j]=0;
1163                         }
1164                         (*filterPos)[i]= srcW - filterSize;
1165                 }
1166         }
1167
1168         // Note the +1 is for the MMXscaler which reads over the end
1169         /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1170         *outFilter= (int16_t*)memalign(16, *outFilterSize*(dstW+1)*sizeof(int16_t));
1171         memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1172
1173         /* Normalize & Store in outFilter */
1174         for(i=0; i<dstW; i++)
1175         {
1176                 int j;
1177                 double error=0;
1178                 double sum=0;
1179                 double scale= one;
1180
1181                 for(j=0; j<filterSize; j++)
1182                 {
1183                         sum+= filter[i*filterSize + j];
1184                 }
1185                 scale/= sum;
1186                 for(j=0; j<*outFilterSize; j++)
1187                 {
1188                         double v= filter[i*filterSize + j]*scale + error;
1189                         int intV= floor(v + 0.5);
1190                         (*outFilter)[i*(*outFilterSize) + j]= intV;
1191                         error = v - intV;
1192                 }
1193         }
1194
1195         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1196         for(i=0; i<*outFilterSize; i++)
1197         {
1198                 int j= dstW*(*outFilterSize);
1199                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1200         }
1201
1202         free(filter);
1203 }
1204
1205 #if defined(ARCH_X86) || defined(ARCH_X86_64)
1206 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1207 {
1208         uint8_t *fragmentA;
1209         long imm8OfPShufW1A;
1210         long imm8OfPShufW2A;
1211         long fragmentLengthA;
1212         uint8_t *fragmentB;
1213         long imm8OfPShufW1B;
1214         long imm8OfPShufW2B;
1215         long fragmentLengthB;
1216         int fragmentPos;
1217
1218         int xpos, i;
1219
1220         // create an optimized horizontal scaling routine
1221
1222         //code fragment
1223
1224         asm volatile(
1225                 "jmp 9f                         \n\t"
1226         // Begin
1227                 "0:                             \n\t"
1228                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t"
1229                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t"
1230                 "movd 1(%%"REG_c", %%"REG_S"), %%mm1\n\t"
1231                 "punpcklbw %%mm7, %%mm1         \n\t"
1232                 "punpcklbw %%mm7, %%mm0         \n\t"
1233                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1234                 "1:                             \n\t"
1235                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1236                 "2:                             \n\t"
1237                 "psubw %%mm1, %%mm0             \n\t"
1238                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1239                 "pmullw %%mm3, %%mm0            \n\t"
1240                 "psllw $7, %%mm1                \n\t"
1241                 "paddw %%mm1, %%mm0             \n\t"
1242
1243                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1244
1245                 "add $8, %%"REG_a"              \n\t"
1246         // End
1247                 "9:                             \n\t"
1248 //              "int $3\n\t"
1249                 "lea 0b, %0                     \n\t"
1250                 "lea 1b, %1                     \n\t"
1251                 "lea 2b, %2                     \n\t"
1252                 "dec %1                         \n\t"
1253                 "dec %2                         \n\t"
1254                 "sub %0, %1                     \n\t"
1255                 "sub %0, %2                     \n\t"
1256                 "lea 9b, %3                     \n\t"
1257                 "sub %0, %3                     \n\t"
1258
1259
1260                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1261                 "=r" (fragmentLengthA)
1262         );
1263
1264         asm volatile(
1265                 "jmp 9f                         \n\t"
1266         // Begin
1267                 "0:                             \n\t"
1268                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t"
1269                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t"
1270                 "punpcklbw %%mm7, %%mm0         \n\t"
1271                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1272                 "1:                             \n\t"
1273                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1274                 "2:                             \n\t"
1275                 "psubw %%mm1, %%mm0             \n\t"
1276                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1277                 "pmullw %%mm3, %%mm0            \n\t"
1278                 "psllw $7, %%mm1                \n\t"
1279                 "paddw %%mm1, %%mm0             \n\t"
1280
1281                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1282
1283                 "add $8, %%"REG_a"              \n\t"
1284         // End
1285                 "9:                             \n\t"
1286 //              "int $3\n\t"
1287                 "lea 0b, %0                     \n\t"
1288                 "lea 1b, %1                     \n\t"
1289                 "lea 2b, %2                     \n\t"
1290                 "dec %1                         \n\t"
1291                 "dec %2                         \n\t"
1292                 "sub %0, %1                     \n\t"
1293                 "sub %0, %2                     \n\t"
1294                 "lea 9b, %3                     \n\t"
1295                 "sub %0, %3                     \n\t"
1296
1297
1298                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1299                 "=r" (fragmentLengthB)
1300         );
1301
1302         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1303         fragmentPos=0;
1304
1305         for(i=0; i<dstW/numSplits; i++)
1306         {
1307                 int xx=xpos>>16;
1308
1309                 if((i&3) == 0)
1310                 {
1311                         int a=0;
1312                         int b=((xpos+xInc)>>16) - xx;
1313                         int c=((xpos+xInc*2)>>16) - xx;
1314                         int d=((xpos+xInc*3)>>16) - xx;
1315
1316                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1317                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1318                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1319                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1320                         filterPos[i/2]= xx;
1321
1322                         if(d+1<4)
1323                         {
1324                                 int maxShift= 3-(d+1);
1325                                 int shift=0;
1326
1327                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1328
1329                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1330                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1331                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1332                                         a | (b<<2) | (c<<4) | (d<<6);
1333
1334                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1335                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1336
1337                                 if(shift && i>=shift)
1338                                 {
1339                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1340                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1341                                         filterPos[i/2]-=shift;
1342                                 }
1343
1344                                 fragmentPos+= fragmentLengthB;
1345                         }
1346                         else
1347                         {
1348                                 int maxShift= 3-d;
1349                                 int shift=0;
1350
1351                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1352
1353                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1354                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1355                                         a | (b<<2) | (c<<4) | (d<<6);
1356
1357                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1358                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1359
1360                                 if(shift && i>=shift)
1361                                 {
1362                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1363                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1364                                         filterPos[i/2]-=shift;
1365                                 }
1366
1367                                 fragmentPos+= fragmentLengthA;
1368                         }
1369
1370                         funnyCode[fragmentPos]= RET;
1371                 }
1372                 xpos+=xInc;
1373         }
1374         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1375 }
1376 #endif // ARCH_X86 || ARCH_X86_64
1377
1378 static void globalInit(void){
1379     // generating tables:
1380     int i;
1381     for(i=0; i<768; i++){
1382         int c= MIN(MAX(i-256, 0), 255);
1383         clip_table[i]=c;
1384     }
1385 }
1386
1387 static SwsFunc getSwsFunc(int flags){
1388
1389 #ifdef RUNTIME_CPUDETECT
1390 #if defined(ARCH_X86) || defined(ARCH_X86_64)
1391         // ordered per speed fasterst first
1392         if(flags & SWS_CPU_CAPS_MMX2)
1393                 return swScale_MMX2;
1394         else if(flags & SWS_CPU_CAPS_3DNOW)
1395                 return swScale_3DNow;
1396         else if(flags & SWS_CPU_CAPS_MMX)
1397                 return swScale_MMX;
1398         else
1399                 return swScale_C;
1400
1401 #else
1402 #ifdef ARCH_POWERPC
1403         if(flags & SWS_CPU_CAPS_ALTIVEC)
1404           return swScale_altivec;
1405         else
1406           return swScale_C;
1407 #endif
1408         return swScale_C;
1409 #endif
1410 #else //RUNTIME_CPUDETECT
1411 #ifdef HAVE_MMX2
1412         return swScale_MMX2;
1413 #elif defined (HAVE_3DNOW)
1414         return swScale_3DNow;
1415 #elif defined (HAVE_MMX)
1416         return swScale_MMX;
1417 #elif defined (HAVE_ALTIVEC)
1418         return swScale_altivec;
1419 #else
1420         return swScale_C;
1421 #endif
1422 #endif //!RUNTIME_CPUDETECT
1423 }
1424
1425 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1426              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1427         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1428         /* Copy Y plane */
1429         if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1430                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1431         else
1432         {
1433                 int i;
1434                 uint8_t *srcPtr= src[0];
1435                 uint8_t *dstPtr= dst;
1436                 for(i=0; i<srcSliceH; i++)
1437                 {
1438                         memcpy(dstPtr, srcPtr, c->srcW);
1439                         srcPtr+= srcStride[0];
1440                         dstPtr+= dstStride[0];
1441                 }
1442         }
1443         dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1444         if (c->dstFormat == IMGFMT_NV12)
1445                 interleaveBytes( src[1],src[2],dst,c->srcW/2,srcSliceH/2,srcStride[1],srcStride[2],dstStride[0] );
1446         else
1447                 interleaveBytes( src[2],src[1],dst,c->srcW/2,srcSliceH/2,srcStride[2],srcStride[1],dstStride[0] );
1448
1449         return srcSliceH;
1450 }
1451
1452 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1453              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1454         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1455
1456         yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1457
1458         return srcSliceH;
1459 }
1460
1461 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1462              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1463         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1464
1465         yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1466
1467         return srcSliceH;
1468 }
1469
1470 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1471 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1472                            int srcSliceH, uint8_t* dst[], int dstStride[]){
1473         const int srcFormat= c->srcFormat;
1474         const int dstFormat= c->dstFormat;
1475         const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1476         const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1477         const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8
1478         const int dstId= (dstFormat&0xFF)>>2;
1479         void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1480
1481         /* BGR -> BGR */
1482         if(   (isBGR(srcFormat) && isBGR(dstFormat))
1483            || (isRGB(srcFormat) && isRGB(dstFormat))){
1484                 switch(srcId | (dstId<<4)){
1485                 case 0x34: conv= rgb16to15; break;
1486                 case 0x36: conv= rgb24to15; break;
1487                 case 0x38: conv= rgb32to15; break;
1488                 case 0x43: conv= rgb15to16; break;
1489                 case 0x46: conv= rgb24to16; break;
1490                 case 0x48: conv= rgb32to16; break;
1491                 case 0x63: conv= rgb15to24; break;
1492                 case 0x64: conv= rgb16to24; break;
1493                 case 0x68: conv= rgb32to24; break;
1494                 case 0x83: conv= rgb15to32; break;
1495                 case 0x84: conv= rgb16to32; break;
1496                 case 0x86: conv= rgb24to32; break;
1497                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1498                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1499                 }
1500         }else if(   (isBGR(srcFormat) && isRGB(dstFormat))
1501                  || (isRGB(srcFormat) && isBGR(dstFormat))){
1502                 switch(srcId | (dstId<<4)){
1503                 case 0x33: conv= rgb15tobgr15; break;
1504                 case 0x34: conv= rgb16tobgr15; break;
1505                 case 0x36: conv= rgb24tobgr15; break;
1506                 case 0x38: conv= rgb32tobgr15; break;
1507                 case 0x43: conv= rgb15tobgr16; break;
1508                 case 0x44: conv= rgb16tobgr16; break;
1509                 case 0x46: conv= rgb24tobgr16; break;
1510                 case 0x48: conv= rgb32tobgr16; break;
1511                 case 0x63: conv= rgb15tobgr24; break;
1512                 case 0x64: conv= rgb16tobgr24; break;
1513                 case 0x66: conv= rgb24tobgr24; break;
1514                 case 0x68: conv= rgb32tobgr24; break;
1515                 case 0x83: conv= rgb15tobgr32; break;
1516                 case 0x84: conv= rgb16tobgr32; break;
1517                 case 0x86: conv= rgb24tobgr32; break;
1518                 case 0x88: conv= rgb32tobgr32; break;
1519                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1520                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1521                 }
1522         }else{
1523                 MSG_ERR("swScaler: internal error %s -> %s converter\n",
1524                          vo_format_name(srcFormat), vo_format_name(dstFormat));
1525         }
1526
1527         if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1528                 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1529         else
1530         {
1531                 int i;
1532                 uint8_t *srcPtr= src[0];
1533                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1534
1535                 for(i=0; i<srcSliceH; i++)
1536                 {
1537                         conv(srcPtr, dstPtr, c->srcW*srcBpp);
1538                         srcPtr+= srcStride[0];
1539                         dstPtr+= dstStride[0];
1540                 }
1541         }
1542         return srcSliceH;
1543 }
1544
1545 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1546              int srcSliceH, uint8_t* dst[], int dstStride[]){
1547
1548         rgb24toyv12(
1549                 src[0],
1550                 dst[0]+ srcSliceY    *dstStride[0],
1551                 dst[1]+(srcSliceY>>1)*dstStride[1],
1552                 dst[2]+(srcSliceY>>1)*dstStride[2],
1553                 c->srcW, srcSliceH,
1554                 dstStride[0], dstStride[1], srcStride[0]);
1555         return srcSliceH;
1556 }
1557
1558 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1559              int srcSliceH, uint8_t* dst[], int dstStride[]){
1560         int i;
1561
1562         /* copy Y */
1563         if(srcStride[0]==dstStride[0] && srcStride[0] > 0)
1564                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1565         else{
1566                 uint8_t *srcPtr= src[0];
1567                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1568
1569                 for(i=0; i<srcSliceH; i++)
1570                 {
1571                         memcpy(dstPtr, srcPtr, c->srcW);
1572                         srcPtr+= srcStride[0];
1573                         dstPtr+= dstStride[0];
1574                 }
1575         }
1576
1577         if(c->dstFormat==IMGFMT_YV12){
1578                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1579                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1580         }else{
1581                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1582                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1583         }
1584         return srcSliceH;
1585 }
1586
1587 /**
1588  * bring pointers in YUV order instead of YVU
1589  */
1590 static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1591         if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
1592            || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
1593                 sortedP[0]= p[0];
1594                 sortedP[1]= p[2];
1595                 sortedP[2]= p[1];
1596                 sortedStride[0]= stride[0];
1597                 sortedStride[1]= stride[2];
1598                 sortedStride[2]= stride[1];
1599         }
1600         else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8)
1601         {
1602                 sortedP[0]= p[0];
1603                 sortedP[1]=
1604                 sortedP[2]= NULL;
1605                 sortedStride[0]= stride[0];
1606                 sortedStride[1]=
1607                 sortedStride[2]= 0;
1608         }
1609         else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)
1610         {
1611                 sortedP[0]= p[0];
1612                 sortedP[1]= p[1];
1613                 sortedP[2]= p[2];
1614                 sortedStride[0]= stride[0];
1615                 sortedStride[1]= stride[1];
1616                 sortedStride[2]= stride[2];
1617         }
1618         else if(format == IMGFMT_NV12 || format == IMGFMT_NV21)
1619         {
1620                 sortedP[0]= p[0];
1621                 sortedP[1]= p[1];
1622                 sortedP[2]= NULL;
1623                 sortedStride[0]= stride[0];
1624                 sortedStride[1]= stride[1];
1625                 sortedStride[2]= 0;
1626         }else{
1627                 MSG_ERR("internal error in orderYUV\n");
1628         }
1629 }
1630
1631 /* unscaled copy like stuff (assumes nearly identical formats) */
1632 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1633              int srcSliceH, uint8_t* dst[], int dstStride[]){
1634
1635         if(isPacked(c->srcFormat))
1636         {
1637                 if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1638                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1639                 else
1640                 {
1641                         int i;
1642                         uint8_t *srcPtr= src[0];
1643                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1644                         int length=0;
1645
1646                         /* universal length finder */
1647                         while(length+c->srcW <= ABS(dstStride[0])
1648                            && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
1649                         ASSERT(length!=0);
1650
1651                         for(i=0; i<srcSliceH; i++)
1652                         {
1653                                 memcpy(dstPtr, srcPtr, length);
1654                                 srcPtr+= srcStride[0];
1655                                 dstPtr+= dstStride[0];
1656                         }
1657                 }
1658         }
1659         else
1660         { /* Planar YUV or gray */
1661                 int plane;
1662                 for(plane=0; plane<3; plane++)
1663                 {
1664                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1665                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1666                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1667
1668                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1669                         {
1670                                 if(!isGray(c->dstFormat))
1671                                         memset(dst[plane], 128, dstStride[plane]*height);
1672                         }
1673                         else
1674                         {
1675                                 if(dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1676                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1677                                 else
1678                                 {
1679                                         int i;
1680                                         uint8_t *srcPtr= src[plane];
1681                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1682                                         for(i=0; i<height; i++)
1683                                         {
1684                                                 memcpy(dstPtr, srcPtr, length);
1685                                                 srcPtr+= srcStride[plane];
1686                                                 dstPtr+= dstStride[plane];
1687                                         }
1688                                 }
1689                         }
1690                 }
1691         }
1692         return srcSliceH;
1693 }
1694
1695 static int remove_dup_fourcc(int fourcc)
1696 {
1697         switch(fourcc)
1698         {
1699             case IMGFMT_I420:
1700             case IMGFMT_IYUV: return IMGFMT_YV12;
1701             case IMGFMT_Y8  : return IMGFMT_Y800;
1702             case IMGFMT_IF09: return IMGFMT_YVU9;
1703             default: return fourcc;
1704         }
1705 }
1706
1707 static void getSubSampleFactors(int *h, int *v, int format){
1708         switch(format){
1709         case IMGFMT_UYVY:
1710         case IMGFMT_YUY2:
1711                 *h=1;
1712                 *v=0;
1713                 break;
1714         case IMGFMT_YV12:
1715         case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
1716         case IMGFMT_NV12:
1717         case IMGFMT_NV21:
1718                 *h=1;
1719                 *v=1;
1720                 break;
1721         case IMGFMT_YVU9:
1722                 *h=2;
1723                 *v=2;
1724                 break;
1725         case IMGFMT_444P:
1726                 *h=0;
1727                 *v=0;
1728                 break;
1729         case IMGFMT_422P:
1730                 *h=1;
1731                 *v=0;
1732                 break;
1733         case IMGFMT_411P:
1734                 *h=2;
1735                 *v=0;
1736                 break;
1737         default:
1738                 *h=0;
1739                 *v=0;
1740                 break;
1741         }
1742 }
1743
1744 static uint16_t roundToInt16(int64_t f){
1745         int r= (f + (1<<15))>>16;
1746              if(r<-0x7FFF) return 0x8000;
1747         else if(r> 0x7FFF) return 0x7FFF;
1748         else               return r;
1749 }
1750
1751 /**
1752  * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1753  * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1754  * @return -1 if not supported
1755  */
1756 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1757         int64_t crv =  inv_table[0];
1758         int64_t cbu =  inv_table[1];
1759         int64_t cgu = -inv_table[2];
1760         int64_t cgv = -inv_table[3];
1761         int64_t cy  = 1<<16;
1762         int64_t oy  = 0;
1763
1764         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1765         memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1766         memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
1767
1768         c->brightness= brightness;
1769         c->contrast  = contrast;
1770         c->saturation= saturation;
1771         c->srcRange  = srcRange;
1772         c->dstRange  = dstRange;
1773
1774         c->uOffset=   0x0400040004000400LL;
1775         c->vOffset=   0x0400040004000400LL;
1776
1777         if(!srcRange){
1778                 cy= (cy*255) / 219;
1779                 oy= 16<<16;
1780         }
1781
1782         cy = (cy *contrast             )>>16;
1783         crv= (crv*contrast * saturation)>>32;
1784         cbu= (cbu*contrast * saturation)>>32;
1785         cgu= (cgu*contrast * saturation)>>32;
1786         cgv= (cgv*contrast * saturation)>>32;
1787
1788         oy -= 256*brightness;
1789
1790         c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
1791         c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
1792         c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1793         c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1794         c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1795         c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
1796
1797         yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1798         //FIXME factorize
1799
1800 #ifdef COMPILE_ALTIVEC
1801         if (c->flags & SWS_CPU_CAPS_ALTIVEC)
1802             yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
1803 #endif
1804         return 0;
1805 }
1806
1807 /**
1808  * @return -1 if not supported
1809  */
1810 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1811         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1812
1813         *inv_table = c->srcColorspaceTable;
1814         *table     = c->dstColorspaceTable;
1815         *srcRange  = c->srcRange;
1816         *dstRange  = c->dstRange;
1817         *brightness= c->brightness;
1818         *contrast  = c->contrast;
1819         *saturation= c->saturation;
1820
1821         return 0;
1822 }
1823
1824 SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,
1825                          SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
1826
1827         SwsContext *c;
1828         int i;
1829         int usesVFilter, usesHFilter;
1830         int unscaled, needsDither;
1831         int srcFormat, dstFormat;
1832         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1833 #if defined(ARCH_X86) || defined(ARCH_X86_64)
1834         if(flags & SWS_CPU_CAPS_MMX)
1835                 asm volatile("emms\n\t"::: "memory");
1836 #endif
1837
1838 #ifndef RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
1839         flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1840 #ifdef HAVE_MMX2
1841         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1842 #elif defined (HAVE_3DNOW)
1843         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1844 #elif defined (HAVE_MMX)
1845         flags |= SWS_CPU_CAPS_MMX;
1846 #elif defined (HAVE_ALTIVEC)
1847         flags |= SWS_CPU_CAPS_ALTIVEC;
1848 #endif
1849 #endif
1850         if(clip_table[512] != 255) globalInit();
1851         if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1852
1853         /* avoid duplicate Formats, so we don't need to check to much */
1854         srcFormat = remove_dup_fourcc(origSrcFormat);
1855         dstFormat = remove_dup_fourcc(origDstFormat);
1856
1857         unscaled = (srcW == dstW && srcH == dstH);
1858         needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
1859                      && (dstFormat&0xFF)<24
1860                      && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1861
1862         if(!isSupportedIn(srcFormat))
1863         {
1864                 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1865                 return NULL;
1866         }
1867         if(!isSupportedOut(dstFormat))
1868         {
1869                 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1870                 return NULL;
1871         }
1872
1873         /* sanity check */
1874         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
1875         {
1876                  MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
1877                         srcW, srcH, dstW, dstH);
1878                 return NULL;
1879         }
1880
1881         if(!dstFilter) dstFilter= &dummyFilter;
1882         if(!srcFilter) srcFilter= &dummyFilter;
1883
1884         c= memalign(64, sizeof(SwsContext));
1885         memset(c, 0, sizeof(SwsContext));
1886
1887         c->srcW= srcW;
1888         c->srcH= srcH;
1889         c->dstW= dstW;
1890         c->dstH= dstH;
1891         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1892         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1893         c->flags= flags;
1894         c->dstFormat= dstFormat;
1895         c->srcFormat= srcFormat;
1896         c->origDstFormat= origDstFormat;
1897         c->origSrcFormat= origSrcFormat;
1898         c->vRounder= 4* 0x0001000100010001ULL;
1899
1900         usesHFilter= usesVFilter= 0;
1901         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
1902         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
1903         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
1904         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
1905         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
1906         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
1907         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
1908         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
1909
1910         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1911         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1912
1913         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1914         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1915
1916         // drop some chroma lines if the user wants it
1917         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1918         c->chrSrcVSubSample+= c->vChrDrop;
1919
1920         // drop every 2. pixel for chroma calculation unless user wants full chroma
1921         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
1922                 c->chrSrcHSubSample=1;
1923
1924         if(param){
1925                 c->param[0] = param[0];
1926                 c->param[1] = param[1];
1927         }else{
1928                 c->param[0] =
1929                 c->param[1] = SWS_PARAM_DEFAULT;
1930         }
1931
1932         c->chrIntHSubSample= c->chrDstHSubSample;
1933         c->chrIntVSubSample= c->chrSrcVSubSample;
1934
1935         // note the -((-x)>>y) is so that we allways round toward +inf
1936         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1937         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
1938         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
1939         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
1940
1941         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);
1942
1943         /* unscaled special Cases */
1944         if(unscaled && !usesHFilter && !usesVFilter)
1945         {
1946                 /* yv12_to_nv12 */
1947                 if(srcFormat == IMGFMT_YV12 && (dstFormat == IMGFMT_NV12 || dstFormat == IMGFMT_NV21))
1948                 {
1949                         c->swScale= PlanarToNV12Wrapper;
1950                 }
1951                 /* yuv2bgr */
1952                 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
1953                 {
1954                         c->swScale= yuv2rgb_get_func_ptr(c);
1955                 }
1956
1957                 if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 )
1958                 {
1959                         c->swScale= yvu9toyv12Wrapper;
1960                 }
1961
1962                 /* bgr24toYV12 */
1963                 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
1964                         c->swScale= bgr24toyv12Wrapper;
1965
1966                 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
1967                 if(   (isBGR(srcFormat) || isRGB(srcFormat))
1968                    && (isBGR(dstFormat) || isRGB(dstFormat))
1969                    && !needsDither)
1970                         c->swScale= rgb2rgbWrapper;
1971
1972                 /* LQ converters if -sws 0 or -sws 4*/
1973                 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
1974                         /* rgb/bgr -> rgb/bgr (dither needed forms) */
1975                         if(  (isBGR(srcFormat) || isRGB(srcFormat))
1976                           && (isBGR(dstFormat) || isRGB(dstFormat))
1977                           && needsDither)
1978                                 c->swScale= rgb2rgbWrapper;
1979
1980                         /* yv12_to_yuy2 */
1981                         if(srcFormat == IMGFMT_YV12 &&
1982                             (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY))
1983                         {
1984                                 if (dstFormat == IMGFMT_YUY2)
1985                                     c->swScale= PlanarToYuy2Wrapper;
1986                                 else
1987                                     c->swScale= PlanarToUyvyWrapper;
1988                         }
1989                 }
1990
1991 #ifdef COMPILE_ALTIVEC
1992                 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
1993                     ((srcFormat == IMGFMT_YV12 &&
1994                       (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY)))) {
1995                   // unscaled YV12 -> packed YUV, we want speed
1996                   if (dstFormat == IMGFMT_YUY2)
1997                     c->swScale= yv12toyuy2_unscaled_altivec;
1998                   else
1999                     c->swScale= yv12touyvy_unscaled_altivec;
2000                 }
2001 #endif
2002
2003                 /* simple copy */
2004                 if(   srcFormat == dstFormat
2005                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2006                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2007                   )
2008                 {
2009                         c->swScale= simpleCopy;
2010                 }
2011
2012                 if(c->swScale){
2013                         if(flags&SWS_PRINT_INFO)
2014                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2015                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2016                         return c;
2017                 }
2018         }
2019
2020         if(flags & SWS_CPU_CAPS_MMX2)
2021         {
2022                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2023                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2024                 {
2025                         if(flags&SWS_PRINT_INFO)
2026                                 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
2027                 }
2028                 if(usesHFilter) c->canMMX2BeUsed=0;
2029         }
2030         else
2031                 c->canMMX2BeUsed=0;
2032
2033         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2034         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2035
2036         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2037         // but only for the FAST_BILINEAR mode otherwise do correct scaling
2038         // n-2 is the last chrominance sample available
2039         // this is not perfect, but noone shuld notice the difference, the more correct variant
2040         // would be like the vertical one, but that would require some special code for the
2041         // first and last pixel
2042         if(flags&SWS_FAST_BILINEAR)
2043         {
2044                 if(c->canMMX2BeUsed)
2045                 {
2046                         c->lumXInc+= 20;
2047                         c->chrXInc+= 20;
2048                 }
2049                 //we don't use the x86asm scaler if mmx is available
2050                 else if(flags & SWS_CPU_CAPS_MMX)
2051                 {
2052                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2053                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2054                 }
2055         }
2056
2057         /* precalculate horizontal scaler filter coefficients */
2058         {
2059                 const int filterAlign=
2060                   (flags & SWS_CPU_CAPS_MMX) ? 4 :
2061                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2062                   1;
2063
2064                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2065                                  srcW      ,       dstW, filterAlign, 1<<14,
2066                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2067                                  srcFilter->lumH, dstFilter->lumH, c->param);
2068                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2069                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2070                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2071                                  srcFilter->chrH, dstFilter->chrH, c->param);
2072
2073 #if defined(ARCH_X86) || defined(ARCH_X86_64)
2074 // can't downscale !!!
2075                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2076                 {
2077 #define MAX_FUNNY_CODE_SIZE 10000
2078 #ifdef MAP_ANONYMOUS
2079                         c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2080                         c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2081 #else
2082                         c->funnyYCode = (uint8_t*)memalign(32, MAX_FUNNY_CODE_SIZE);
2083                         c->funnyUVCode = (uint8_t*)memalign(32, MAX_FUNNY_CODE_SIZE);
2084 #endif
2085
2086                         c->lumMmx2Filter   = (int16_t*)memalign(8, (dstW        /8+8)*sizeof(int16_t));
2087                         c->chrMmx2Filter   = (int16_t*)memalign(8, (c->chrDstW  /4+8)*sizeof(int16_t));
2088                         c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW      /2/8+8)*sizeof(int32_t));
2089                         c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
2090
2091                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2092                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2093                 }
2094 #endif
2095         } // Init Horizontal stuff
2096
2097
2098
2099         /* precalculate vertical scaler filter coefficients */
2100         {
2101                 const int filterAlign=
2102                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2103                   1;
2104
2105                 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2106                                 srcH      ,        dstH, filterAlign, (1<<12)-4,
2107                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2108                                 srcFilter->lumV, dstFilter->lumV, c->param);
2109                 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2110                                 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2111                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2112                                 srcFilter->chrV, dstFilter->chrV, c->param);
2113
2114 #ifdef HAVE_ALTIVEC
2115                 c->vYCoeffsBank = memalign (16, sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2116                 c->vCCoeffsBank = memalign (16, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2117
2118                 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2119                   int j;
2120                   short *p = (short *)&c->vYCoeffsBank[i];
2121                   for (j=0;j<8;j++)
2122                     p[j] = c->vLumFilter[i];
2123                 }
2124
2125                 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2126                   int j;
2127                   short *p = (short *)&c->vCCoeffsBank[i];
2128                   for (j=0;j<8;j++)
2129                     p[j] = c->vChrFilter[i];
2130                 }
2131 #endif
2132         }
2133
2134         // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2135         c->vLumBufSize= c->vLumFilterSize;
2136         c->vChrBufSize= c->vChrFilterSize;
2137         for(i=0; i<dstH; i++)
2138         {
2139                 int chrI= i*c->chrDstH / dstH;
2140                 int nextSlice= MAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
2141                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2142
2143                 nextSlice>>= c->chrSrcVSubSample;
2144                 nextSlice<<= c->chrSrcVSubSample;
2145                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
2146                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
2147                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2148                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2149         }
2150
2151         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2152         c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
2153         c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
2154         //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)
2155         /* align at 16 bytes for AltiVec */
2156         for(i=0; i<c->vLumBufSize; i++)
2157                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(16, 4000);
2158         for(i=0; i<c->vChrBufSize; i++)
2159                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(16, 8000);
2160
2161         //try to avoid drawing green stuff between the right end and the stride end
2162         for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2163         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2164
2165         ASSERT(c->chrDstH <= dstH)
2166
2167         if(flags&SWS_PRINT_INFO)
2168         {
2169 #ifdef DITHER1XBPP
2170                 char *dither= " dithered";
2171 #else
2172                 char *dither= "";
2173 #endif
2174                 if(flags&SWS_FAST_BILINEAR)
2175                         MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
2176                 else if(flags&SWS_BILINEAR)
2177                         MSG_INFO("\nSwScaler: BILINEAR scaler, ");
2178                 else if(flags&SWS_BICUBIC)
2179                         MSG_INFO("\nSwScaler: BICUBIC scaler, ");
2180                 else if(flags&SWS_X)
2181                         MSG_INFO("\nSwScaler: Experimental scaler, ");
2182                 else if(flags&SWS_POINT)
2183                         MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
2184                 else if(flags&SWS_AREA)
2185                         MSG_INFO("\nSwScaler: Area Averageing scaler, ");
2186                 else if(flags&SWS_BICUBLIN)
2187                         MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2188                 else if(flags&SWS_GAUSS)
2189                         MSG_INFO("\nSwScaler: Gaussian scaler, ");
2190                 else if(flags&SWS_SINC)
2191                         MSG_INFO("\nSwScaler: Sinc scaler, ");
2192                 else if(flags&SWS_LANCZOS)
2193                         MSG_INFO("\nSwScaler: Lanczos scaler, ");
2194                 else if(flags&SWS_SPLINE)
2195                         MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
2196                 else
2197                         MSG_INFO("\nSwScaler: ehh flags invalid?! ");
2198
2199                 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2200                         MSG_INFO("from %s to%s %s ",
2201                                 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
2202                 else
2203                         MSG_INFO("from %s to %s ",
2204                                 vo_format_name(srcFormat), vo_format_name(dstFormat));
2205
2206                 if(flags & SWS_CPU_CAPS_MMX2)
2207                         MSG_INFO("using MMX2\n");
2208                 else if(flags & SWS_CPU_CAPS_3DNOW)
2209                         MSG_INFO("using 3DNOW\n");
2210                 else if(flags & SWS_CPU_CAPS_MMX)
2211                         MSG_INFO("using MMX\n");
2212                 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2213                         MSG_INFO("using AltiVec\n");
2214                 else
2215                         MSG_INFO("using C\n");
2216         }
2217
2218         if(flags & SWS_PRINT_INFO)
2219         {
2220                 if(flags & SWS_CPU_CAPS_MMX)
2221                 {
2222                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2223                                 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2224                         else
2225                         {
2226                                 if(c->hLumFilterSize==4)
2227                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2228                                 else if(c->hLumFilterSize==8)
2229                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2230                                 else
2231                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2232
2233                                 if(c->hChrFilterSize==4)
2234                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2235                                 else if(c->hChrFilterSize==8)
2236                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2237                                 else
2238                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2239                         }
2240                 }
2241                 else
2242                 {
2243 #if defined(ARCH_X86) || defined(ARCH_X86_64)
2244                         MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2245 #else
2246                         if(flags & SWS_FAST_BILINEAR)
2247                                 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2248                         else
2249                                 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2250 #endif
2251                 }
2252                 if(isPlanarYUV(dstFormat))
2253                 {
2254                         if(c->vLumFilterSize==1)
2255                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2256                         else
2257                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2258                 }
2259                 else
2260                 {
2261                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2262                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2263                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2264                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2265                                 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2266                         else
2267                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2268                 }
2269
2270                 if(dstFormat==IMGFMT_BGR24)
2271                         MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2272                                 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2273                 else if(dstFormat==IMGFMT_BGR32)
2274                         MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2275                 else if(dstFormat==IMGFMT_BGR16)
2276                         MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2277                 else if(dstFormat==IMGFMT_BGR15)
2278                         MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2279
2280                 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2281         }
2282         if(flags & SWS_PRINT_INFO)
2283         {
2284                 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2285                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2286                 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2287                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2288         }
2289
2290         c->swScale= getSwsFunc(flags);
2291         return c;
2292 }
2293
2294 /**
2295  * swscale warper, so we don't need to export the SwsContext.
2296  * assumes planar YUV to be in YUV order instead of YVU
2297  */
2298 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2299                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2300         if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2301             MSG_ERR("swScaler: slices start in the middle!\n");
2302             return 0;
2303         }
2304         if (c->sliceDir == 0) {
2305             if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2306         }
2307
2308         // copy strides, so they can safely be modified
2309         if (c->sliceDir == 1) {
2310             // slices go from top to bottom
2311             int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
2312             int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
2313             return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2314         } else {
2315             // slices go from bottom to top => we flip the image internally
2316             uint8_t* src2[3]= {src[0] + (srcSliceH-1)*srcStride[0],
2317                                src[1] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1],
2318                                src[2] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2]
2319             };
2320             uint8_t* dst2[3]= {dst[0] + (c->dstH-1)*dstStride[0],
2321                                dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2322                                dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2323             int srcStride2[3]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2324             int dstStride2[3]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2325
2326             return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2327         }
2328 }
2329
2330 /**
2331  * swscale warper, so we don't need to export the SwsContext
2332  */
2333 int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
2334                            int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
2335         int srcStride[3];
2336         int dstStride[3];
2337         uint8_t *src[3];
2338         uint8_t *dst[3];
2339         sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam);
2340         sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam);
2341 //printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
2342
2343         return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2344 }
2345
2346 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2347                                 float lumaSharpen, float chromaSharpen,
2348                                 float chromaHShift, float chromaVShift,
2349                                 int verbose)
2350 {
2351         SwsFilter *filter= malloc(sizeof(SwsFilter));
2352
2353         if(lumaGBlur!=0.0){
2354                 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2355                 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2356         }else{
2357                 filter->lumH= sws_getIdentityVec();
2358                 filter->lumV= sws_getIdentityVec();
2359         }
2360
2361         if(chromaGBlur!=0.0){
2362                 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2363                 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2364         }else{
2365                 filter->chrH= sws_getIdentityVec();
2366                 filter->chrV= sws_getIdentityVec();
2367         }
2368
2369         if(chromaSharpen!=0.0){
2370                 SwsVector *id= sws_getIdentityVec();
2371                 sws_scaleVec(filter->chrH, -chromaSharpen);
2372                 sws_scaleVec(filter->chrV, -chromaSharpen);
2373                 sws_addVec(filter->chrH, id);
2374                 sws_addVec(filter->chrV, id);
2375                 sws_freeVec(id);
2376         }
2377
2378         if(lumaSharpen!=0.0){
2379                 SwsVector *id= sws_getIdentityVec();
2380                 sws_scaleVec(filter->lumH, -lumaSharpen);
2381                 sws_scaleVec(filter->lumV, -lumaSharpen);
2382                 sws_addVec(filter->lumH, id);
2383                 sws_addVec(filter->lumV, id);
2384                 sws_freeVec(id);
2385         }
2386
2387         if(chromaHShift != 0.0)
2388                 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2389
2390         if(chromaVShift != 0.0)
2391                 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2392
2393         sws_normalizeVec(filter->chrH, 1.0);
2394         sws_normalizeVec(filter->chrV, 1.0);
2395         sws_normalizeVec(filter->lumH, 1.0);
2396         sws_normalizeVec(filter->lumV, 1.0);
2397
2398         if(verbose) sws_printVec(filter->chrH);
2399         if(verbose) sws_printVec(filter->lumH);
2400
2401         return filter;
2402 }
2403
2404 /**
2405  * returns a normalized gaussian curve used to filter stuff
2406  * quality=3 is high quality, lowwer is lowwer quality
2407  */
2408 SwsVector *sws_getGaussianVec(double variance, double quality){
2409         const int length= (int)(variance*quality + 0.5) | 1;
2410         int i;
2411         double *coeff= memalign(sizeof(double), length*sizeof(double));
2412         double middle= (length-1)*0.5;
2413         SwsVector *vec= malloc(sizeof(SwsVector));
2414
2415         vec->coeff= coeff;
2416         vec->length= length;
2417
2418         for(i=0; i<length; i++)
2419         {
2420                 double dist= i-middle;
2421                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2422         }
2423
2424         sws_normalizeVec(vec, 1.0);
2425
2426         return vec;
2427 }
2428
2429 SwsVector *sws_getConstVec(double c, int length){
2430         int i;
2431         double *coeff= memalign(sizeof(double), length*sizeof(double));
2432         SwsVector *vec= malloc(sizeof(SwsVector));
2433
2434         vec->coeff= coeff;
2435         vec->length= length;
2436
2437         for(i=0; i<length; i++)
2438                 coeff[i]= c;
2439
2440         return vec;
2441 }
2442
2443
2444 SwsVector *sws_getIdentityVec(void){
2445         return sws_getConstVec(1.0, 1);
2446 }
2447
2448 double sws_dcVec(SwsVector *a){
2449         int i;
2450         double sum=0;
2451
2452         for(i=0; i<a->length; i++)
2453                 sum+= a->coeff[i];
2454
2455         return sum;
2456 }
2457
2458 void sws_scaleVec(SwsVector *a, double scalar){
2459         int i;
2460
2461         for(i=0; i<a->length; i++)
2462                 a->coeff[i]*= scalar;
2463 }
2464
2465 void sws_normalizeVec(SwsVector *a, double height){
2466         sws_scaleVec(a, height/sws_dcVec(a));
2467 }
2468
2469 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2470         int length= a->length + b->length - 1;
2471         double *coeff= memalign(sizeof(double), length*sizeof(double));
2472         int i, j;
2473         SwsVector *vec= malloc(sizeof(SwsVector));
2474
2475         vec->coeff= coeff;
2476         vec->length= length;
2477
2478         for(i=0; i<length; i++) coeff[i]= 0.0;
2479
2480         for(i=0; i<a->length; i++)
2481         {
2482                 for(j=0; j<b->length; j++)
2483                 {
2484                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2485                 }
2486         }
2487
2488         return vec;
2489 }
2490
2491 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2492         int length= MAX(a->length, b->length);
2493         double *coeff= memalign(sizeof(double), length*sizeof(double));
2494         int i;
2495         SwsVector *vec= malloc(sizeof(SwsVector));
2496
2497         vec->coeff= coeff;
2498         vec->length= length;
2499
2500         for(i=0; i<length; i++) coeff[i]= 0.0;
2501
2502         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2503         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2504
2505         return vec;
2506 }
2507
2508 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2509         int length= MAX(a->length, b->length);
2510         double *coeff= memalign(sizeof(double), length*sizeof(double));
2511         int i;
2512         SwsVector *vec= malloc(sizeof(SwsVector));
2513
2514         vec->coeff= coeff;
2515         vec->length= length;
2516
2517         for(i=0; i<length; i++) coeff[i]= 0.0;
2518
2519         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2520         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2521
2522         return vec;
2523 }
2524
2525 /* shift left / or right if "shift" is negative */
2526 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2527         int length= a->length + ABS(shift)*2;
2528         double *coeff= memalign(sizeof(double), length*sizeof(double));
2529         int i;
2530         SwsVector *vec= malloc(sizeof(SwsVector));
2531
2532         vec->coeff= coeff;
2533         vec->length= length;
2534
2535         for(i=0; i<length; i++) coeff[i]= 0.0;
2536
2537         for(i=0; i<a->length; i++)
2538         {
2539                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2540         }
2541
2542         return vec;
2543 }
2544
2545 void sws_shiftVec(SwsVector *a, int shift){
2546         SwsVector *shifted= sws_getShiftedVec(a, shift);
2547         free(a->coeff);
2548         a->coeff= shifted->coeff;
2549         a->length= shifted->length;
2550         free(shifted);
2551 }
2552
2553 void sws_addVec(SwsVector *a, SwsVector *b){
2554         SwsVector *sum= sws_sumVec(a, b);
2555         free(a->coeff);
2556         a->coeff= sum->coeff;
2557         a->length= sum->length;
2558         free(sum);
2559 }
2560
2561 void sws_subVec(SwsVector *a, SwsVector *b){
2562         SwsVector *diff= sws_diffVec(a, b);
2563         free(a->coeff);
2564         a->coeff= diff->coeff;
2565         a->length= diff->length;
2566         free(diff);
2567 }
2568
2569 void sws_convVec(SwsVector *a, SwsVector *b){
2570         SwsVector *conv= sws_getConvVec(a, b);
2571         free(a->coeff);
2572         a->coeff= conv->coeff;
2573         a->length= conv->length;
2574         free(conv);
2575 }
2576
2577 SwsVector *sws_cloneVec(SwsVector *a){
2578         double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2579         int i;
2580         SwsVector *vec= malloc(sizeof(SwsVector));
2581
2582         vec->coeff= coeff;
2583         vec->length= a->length;
2584
2585         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2586
2587         return vec;
2588 }
2589
2590 void sws_printVec(SwsVector *a){
2591         int i;
2592         double max=0;
2593         double min=0;
2594         double range;
2595
2596         for(i=0; i<a->length; i++)
2597                 if(a->coeff[i]>max) max= a->coeff[i];
2598
2599         for(i=0; i<a->length; i++)
2600                 if(a->coeff[i]<min) min= a->coeff[i];
2601
2602         range= max - min;
2603
2604         for(i=0; i<a->length; i++)
2605         {
2606                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2607                 MSG_DBG2("%1.3f ", a->coeff[i]);
2608                 for(;x>0; x--) MSG_DBG2(" ");
2609                 MSG_DBG2("|\n");
2610         }
2611 }
2612
2613 void sws_freeVec(SwsVector *a){
2614         if(!a) return;
2615         if(a->coeff) free(a->coeff);
2616         a->coeff=NULL;
2617         a->length=0;
2618         free(a);
2619 }
2620
2621 void sws_freeFilter(SwsFilter *filter){
2622         if(!filter) return;
2623
2624         if(filter->lumH) sws_freeVec(filter->lumH);
2625         if(filter->lumV) sws_freeVec(filter->lumV);
2626         if(filter->chrH) sws_freeVec(filter->chrH);
2627         if(filter->chrV) sws_freeVec(filter->chrV);
2628         free(filter);
2629 }
2630
2631
2632 void sws_freeContext(SwsContext *c){
2633         int i;
2634         if(!c) return;
2635
2636         if(c->lumPixBuf)
2637         {
2638                 for(i=0; i<c->vLumBufSize; i++)
2639                 {
2640                         if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2641                         c->lumPixBuf[i]=NULL;
2642                 }
2643                 free(c->lumPixBuf);
2644                 c->lumPixBuf=NULL;
2645         }
2646
2647         if(c->chrPixBuf)
2648         {
2649                 for(i=0; i<c->vChrBufSize; i++)
2650                 {
2651                         if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2652                         c->chrPixBuf[i]=NULL;
2653                 }
2654                 free(c->chrPixBuf);
2655                 c->chrPixBuf=NULL;
2656         }
2657
2658         if(c->vLumFilter) free(c->vLumFilter);
2659         c->vLumFilter = NULL;
2660         if(c->vChrFilter) free(c->vChrFilter);
2661         c->vChrFilter = NULL;
2662         if(c->hLumFilter) free(c->hLumFilter);
2663         c->hLumFilter = NULL;
2664         if(c->hChrFilter) free(c->hChrFilter);
2665         c->hChrFilter = NULL;
2666 #ifdef HAVE_ALTIVEC
2667         if(c->vYCoeffsBank) free(c->vYCoeffsBank);
2668         c->vYCoeffsBank = NULL;
2669         if(c->vCCoeffsBank) free(c->vCCoeffsBank);
2670         c->vCCoeffsBank = NULL;
2671 #endif
2672
2673         if(c->vLumFilterPos) free(c->vLumFilterPos);
2674         c->vLumFilterPos = NULL;
2675         if(c->vChrFilterPos) free(c->vChrFilterPos);
2676         c->vChrFilterPos = NULL;
2677         if(c->hLumFilterPos) free(c->hLumFilterPos);
2678         c->hLumFilterPos = NULL;
2679         if(c->hChrFilterPos) free(c->hChrFilterPos);
2680         c->hChrFilterPos = NULL;
2681
2682 #if defined(ARCH_X86) || defined(ARCH_X86_64)
2683 #ifdef MAP_ANONYMOUS
2684         if(c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
2685         if(c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
2686 #else
2687         if(c->funnyYCode) free(c->funnyYCode);
2688         if(c->funnyUVCode) free(c->funnyUVCode);
2689 #endif
2690         c->funnyYCode=NULL;
2691         c->funnyUVCode=NULL;
2692 #endif
2693
2694         if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2695         c->lumMmx2Filter=NULL;
2696         if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2697         c->chrMmx2Filter=NULL;
2698         if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2699         c->lumMmx2FilterPos=NULL;
2700         if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2701         c->chrMmx2FilterPos=NULL;
2702         if(c->yuvTable) free(c->yuvTable);
2703         c->yuvTable=NULL;
2704
2705         free(c);
2706 }
2707