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

   1 /*
   2     Copyright (C) 2001-2002 Michael Niedermayer <michaelni@gmx.at>
   3
   4     This program is free software; you can redistribute it and/or modify
   5     it under the terms of the GNU General Public License as published by
   6     the Free Software Foundation; either version 2 of the License, or
   7     (at your option) any later version.
   8
   9     This program is distributed in the hope that it will be useful,
  10     but WITHOUT ANY WARRANTY; without even the implied warranty of
  11     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12     GNU General Public License for more details.
  13
  14     You should have received a copy of the GNU General Public License
  15     along with this program; if not, write to the Free Software
  16     Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  17 */
  18
  19 /*
  20   supported Input formats: YV12, I420/IYUV, YUY2, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
  21   supported output formats: YV12, I420/IYUV, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
  22   BGR15/16 support dithering
  23
  24   unscaled special converters
  25   YV12/I420/IYUV -> BGR15/BGR16/BGR24/BGR32
  26   YV12/I420/IYUV -> YV12/I420/IYUV
  27   YUY2/BGR15/BGR16/BGR24/BGR32/RGB24/RGB32 -> same format
  28   BGR24 -> BGR32 & RGB24 -> RGB32
  29   BGR32 -> BGR24 & RGB32 -> RGB24
  30   BGR15 -> BGR16
  31 */
  32
  33 /*
  34 tested special converters
  35  YV12/I420 -> BGR16
  36  YV12 -> YV12
  37  BGR15 -> BGR16
  38  BGR16 -> BGR16
  39
  40 untested special converters
  41   YV12/I420 -> BGR15/BGR24/BGR32 (its the yuv2rgb stuff, so it should be ok)
  42   YV12/I420 -> YV12/I420
  43   YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
  44   BGR24 -> BGR32 & RGB24 -> RGB32
  45   BGR32 -> BGR24 & RGB32 -> RGB24
  46   BGR24 -> YV12
  47 */
  48
  49 #include <inttypes.h>
  50 #include <string.h>
  51 #include <math.h>
  52 #include <stdio.h>
  53 #include "../config.h"
  54 #include "../mangle.h"
  55 #include <assert.h>
  56 #ifdef HAVE_MALLOC_H
  57 #include <malloc.h>
  58 #else
  59 #include <stdlib.h>
  60 #endif
  61 #include "swscale.h"
  62 #include "../cpudetect.h"
  63 #include "../bswap.h"
  64 #include "../libvo/img_format.h"
  65 #include "rgb2rgb.h"
  66 #include "../libvo/fastmemcpy.h"
  67 #include "../mp_msg.h"
  68
  69 #define MSG_WARN(args...) mp_msg(MSGT_SWS,MSGL_WARN, ##args )
  70 #define MSG_FATAL(args...) mp_msg(MSGT_SWS,MSGL_FATAL, ##args )
  71 #define MSG_ERR(args...) mp_msg(MSGT_SWS,MSGL_ERR, ##args )
  72 #define MSG_V(args...) mp_msg(MSGT_SWS,MSGL_V, ##args )
  73 #define MSG_DBG2(args...) mp_msg(MSGT_SWS,MSGL_DBG2, ##args )
  74 #define MSG_INFO(args...) mp_msg(MSGT_SWS,MSGL_INFO, ##args )
  75
  76 #undef MOVNTQ
  77 #undef PAVGB
  78
  79 //#undef HAVE_MMX2
  80 //#define HAVE_3DNOW
  81 //#undef HAVE_MMX
  82 //#undef ARCH_X86
  83 //#define WORDS_BIGENDIAN
  84 #define DITHER1XBPP
  85
  86 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
  87
  88 #define RET 0xC3 //near return opcode for X86
  89
  90 #ifdef MP_DEBUG
  91 #define ASSERT(x) assert(x);
  92 #else
  93 #define ASSERT(x) ;
  94 #endif
  95
  96 #ifdef M_PI
  97 #define PI M_PI
  98 #else
  99 #define PI 3.14159265358979323846
 100 #endif
 101
 102 //FIXME replace this with something faster
 103 #define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420 || (x)==IMGFMT_YVU9)
 104 #define isYUV(x)       ((x)==IMGFMT_YUY2 || isPlanarYUV(x))
 105 #define isGray(x)      ((x)==IMGFMT_Y800)
 106 #define isRGB(x)       (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
 107 #define isBGR(x)       (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
 108 #define isSupportedIn(x)  ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420 || (x)==IMGFMT_YUY2 \
 109                         || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
 110                         || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
 111                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
 112 #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420 \
 113                         || isRGB(x) || isBGR(x)\
 114                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
 115 #define isPacked(x)    ((x)==IMGFMT_YUY2 || isRGB(x) || isBGR(x))
 116
 117 #define RGB2YUV_SHIFT 16
 118 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
 119 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
 120 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
 121 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
 122 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
 123 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
 124 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
 125 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
 126 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
 127
 128 extern int verbose; // defined in mplayer.c
 129 /*
 130 NOTES
 131 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
 132
 133 TODO
 134 more intelligent missalignment avoidance for the horizontal scaler
 135 write special vertical cubic upscale version
 136 Optimize C code (yv12 / minmax)
 137 add support for packed pixel yuv input & output
 138 add support for Y8 output
 139 optimize bgr24 & bgr32
 140 add BGR4 output support
 141 write special BGR->BGR scaler
 142 deglobalize yuv2rgb*.c
 143 */
 144
 145 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
 146 #define MIN(a,b) ((a) > (b) ? (b) : (a))
 147 #define MAX(a,b) ((a) < (b) ? (b) : (a))
 148
 149 #ifdef ARCH_X86
 150 #define CAN_COMPILE_X86_ASM
 151 #endif
 152
 153 #ifdef CAN_COMPILE_X86_ASM
 154 static uint64_t __attribute__((aligned(8))) yCoeff=    0x2568256825682568LL;
 155 static uint64_t __attribute__((aligned(8))) vrCoeff=   0x3343334333433343LL;
 156 static uint64_t __attribute__((aligned(8))) ubCoeff=   0x40cf40cf40cf40cfLL;
 157 static uint64_t __attribute__((aligned(8))) vgCoeff=   0xE5E2E5E2E5E2E5E2LL;
 158 static uint64_t __attribute__((aligned(8))) ugCoeff=   0xF36EF36EF36EF36ELL;
 159 static uint64_t __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
 160 static uint64_t __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
 161 static uint64_t __attribute__((aligned(8))) w400=      0x0400040004000400LL;
 162 static uint64_t __attribute__((aligned(8))) w80=       0x0080008000800080LL;
 163 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
 164 static uint64_t __attribute__((aligned(8))) w02=       0x0002000200020002LL;
 165 static uint64_t __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
 166 static uint64_t __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
 167 static uint64_t __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
 168 static uint64_t __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
 169
 170 static volatile uint64_t __attribute__((aligned(8))) b5Dither;
 171 static volatile uint64_t __attribute__((aligned(8))) g5Dither;
 172 static volatile uint64_t __attribute__((aligned(8))) g6Dither;
 173 static volatile uint64_t __attribute__((aligned(8))) r5Dither;
 174
 175 static uint64_t __attribute__((aligned(8))) dither4[2]={
 176         0x0103010301030103LL,
 177         0x0200020002000200LL,};
 178
 179 static uint64_t __attribute__((aligned(8))) dither8[2]={
 180         0x0602060206020602LL,
 181         0x0004000400040004LL,};
 182
 183 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
 184 static uint64_t __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
 185 static uint64_t __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
 186 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
 187 static uint64_t __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
 188 static uint64_t __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
 189
 190 static uint64_t __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
 191 static uint64_t __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
 192 static uint64_t __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
 193
 194 #ifdef FAST_BGR2YV12
 195 static const uint64_t bgr2YCoeff  __attribute__((aligned(8))) = 0x000000210041000DULL;
 196 static const uint64_t bgr2UCoeff  __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
 197 static const uint64_t bgr2VCoeff  __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
 198 #else
 199 static const uint64_t bgr2YCoeff  __attribute__((aligned(8))) = 0x000020E540830C8BULL;
 200 static const uint64_t bgr2UCoeff  __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
 201 static const uint64_t bgr2VCoeff  __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
 202 #endif
 203 static const uint64_t bgr2YOffset __attribute__((aligned(8))) = 0x1010101010101010ULL;
 204 static const uint64_t bgr2UVOffset __attribute__((aligned(8)))= 0x8080808080808080ULL;
 205 static const uint64_t w1111       __attribute__((aligned(8))) = 0x0001000100010001ULL;
 206 #endif
 207
 208 // clipping helper table for C implementations:
 209 static unsigned char clip_table[768];
 210
 211 //global sws_flags from the command line
 212 int sws_flags=2;
 213
 214 //global srcFilter
 215 SwsFilter src_filter= {NULL, NULL, NULL, NULL};
 216
 217 float sws_lum_gblur= 0.0;
 218 float sws_chr_gblur= 0.0;
 219 int sws_chr_vshift= 0;
 220 int sws_chr_hshift= 0;
 221 float sws_chr_sharpen= 0.0;
 222 float sws_lum_sharpen= 0.0;
 223
 224 /* cpuCaps combined from cpudetect and whats actually compiled in
 225    (if there is no support for something compiled in it wont appear here) */
 226 static CpuCaps cpuCaps;
 227
 228 void (*swScale)(SwsContext *context, uint8_t* src[], int srcStride[], int srcSliceY,
 229              int srcSliceH, uint8_t* dst[], int dstStride[])=NULL;
 230
 231 static SwsVector *getConvVec(SwsVector *a, SwsVector *b);
 232 static inline void orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]);
 233 void *yuv2rgb_c_init (unsigned bpp, int mode, void *table_rV[256], void *table_gU[256], int table_gV[256], void *table_bU[256]);
 234
 235 extern const uint8_t dither_2x2_4[2][8];
 236 extern const uint8_t dither_2x2_8[2][8];
 237 extern const uint8_t dither_8x8_32[8][8];
 238 extern const uint8_t dither_8x8_73[8][8];
 239 extern const uint8_t dither_8x8_220[8][8];
 240
 241 #ifdef CAN_COMPILE_X86_ASM
 242 void in_asm_used_var_warning_killer()
 243 {
 244  volatile int i= yCoeff+vrCoeff+ubCoeff+vgCoeff+ugCoeff+bF8+bFC+w400+w80+w10+
 245  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
 246  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
 247  if(i) i=0;
 248 }
 249 #endif
 250
 251 static int testFormat[]={
 252 IMGFMT_YVU9,
 253 IMGFMT_YV12,
 254 //IMGFMT_IYUV,
 255 IMGFMT_I420,
 256 IMGFMT_BGR15,
 257 IMGFMT_BGR16,
 258 IMGFMT_BGR24,
 259 IMGFMT_BGR32,
 260 IMGFMT_RGB24,
 261 IMGFMT_RGB32,
 262 //IMGFMT_Y8,
 263 IMGFMT_Y800,
 264 //IMGFMT_YUY2,
 265 0
 266 };
 267
 268 static uint64_t getSSD(uint8_t *src1, uint8_t *src2, int stride1, int stride2, int w, int h){
 269         int x,y;
 270         uint64_t ssd=0;
 271
 272         for(y=0; y<h; y++){
 273                 for(x=0; x<w; x++){
 274                         int d= src1[x + y*stride1] - src2[x + y*stride2];
 275                         ssd+= d*d;
 276                 }
 277         }
 278         return ssd;
 279 }
 280
 281 // test by ref -> src -> dst -> out & compare out against ref
 282 // ref & out are YV12
 283 static void doTest(uint8_t *ref[3], int refStride[3], int w, int h, int srcFormat, int dstFormat,
 284                    int srcW, int srcH, int dstW, int dstH, int flags){
 285         uint8_t *src[3];
 286         uint8_t *dst[3];
 287         uint8_t *out[3];
 288         int srcStride[3], dstStride[3];
 289         int i;
 290         uint64_t ssdY, ssdU, ssdV;
 291         SwsContext *srcContext, *dstContext, *outContext;
 292
 293         for(i=0; i<3; i++){
 294                 srcStride[i]= srcW*4;
 295                 dstStride[i]= dstW*4;
 296                 src[i]= malloc(srcStride[i]*srcH);
 297                 dst[i]= malloc(dstStride[i]*dstH);
 298                 out[i]= malloc(refStride[i]*h);
 299         }
 300
 301         srcContext= getSwsContext(w, h, IMGFMT_YV12, srcW, srcH, srcFormat, flags, NULL, NULL);
 302         dstContext= getSwsContext(srcW, srcH, srcFormat, dstW, dstH, dstFormat, flags, NULL, NULL);
 303         outContext= getSwsContext(dstW, dstH, dstFormat, w, h, IMGFMT_YV12, flags, NULL, NULL);
 304         if(srcContext==NULL ||dstContext==NULL ||outContext==NULL){
 305                 printf("Failed allocating swsContext\n");
 306                 goto end;
 307         }
 308 //      printf("test %X %X %X -> %X %X %X\n", (int)ref[0], (int)ref[1], (int)ref[2],
 309 //              (int)src[0], (int)src[1], (int)src[2]);
 310
 311         srcContext->swScale(srcContext, ref, refStride, 0, h   , src, srcStride);
 312         dstContext->swScale(dstContext, src, srcStride, 0, srcH, dst, dstStride);
 313         outContext->swScale(outContext, dst, dstStride, 0, dstH, out, refStride);
 314
 315         ssdY= getSSD(ref[0], out[0], refStride[0], refStride[0], w, h);
 316         ssdU= getSSD(ref[1], out[1], refStride[1], refStride[1], (w+1)>>1, (h+1)>>1);
 317         ssdV= getSSD(ref[2], out[2], refStride[2], refStride[2], (w+1)>>1, (h+1)>>1);
 318
 319         if(isGray(srcFormat) || isGray(dstFormat)) ssdU=ssdV=0; //FIXME check that output is really gray
 320
 321         ssdY/= w*h;
 322         ssdU/= w*h/4;
 323         ssdV/= w*h/4;
 324
 325         if(ssdY>100 || ssdU>50 || ssdV>50){
 326                 printf(" %s %dx%d -> %s %4dx%4d flags=%2d SSD=%5lld,%5lld,%5lld\n",
 327                         vo_format_name(srcFormat), srcW, srcH,
 328                         vo_format_name(dstFormat), dstW, dstH,
 329                         flags,
 330                         ssdY, ssdU, ssdV);
 331         }
 332
 333         end:
 334
 335         freeSwsContext(srcContext);
 336         freeSwsContext(dstContext);
 337         freeSwsContext(outContext);
 338
 339         for(i=0; i<3; i++){
 340                 free(src[i]);
 341                 free(dst[i]);
 342                 free(out[i]);
 343         }
 344 }
 345
 346 static void selfTest(uint8_t *src[3], int stride[3], int w, int h){
 347         int srcFormat, dstFormat, srcFormatIndex, dstFormatIndex;
 348         int srcW, srcH, dstW, dstH;
 349         int flags;
 350
 351         for(srcFormatIndex=0; ;srcFormatIndex++){
 352                 srcFormat= testFormat[srcFormatIndex];
 353                 if(!srcFormat) break;
 354                 for(dstFormatIndex=0; ;dstFormatIndex++){
 355                         dstFormat= testFormat[dstFormatIndex];
 356                         if(!dstFormat) break;
 357                         if(!isSupportedOut(dstFormat)) continue;
 358
 359                         srcW= w+w/3;
 360                         srcH= h+h/3;
 361                         for(dstW=w; dstW<w*2; dstW+= dstW/3){
 362                                 for(dstH=h; dstH<h*2; dstH+= dstH/3){
 363                                         for(flags=1; flags<33; flags*=2)
 364                                                 doTest(src, stride, w, h, srcFormat, dstFormat,
 365                                                         srcW, srcH, dstW, dstH, flags);
 366                                 }
 367                         }
 368                 }
 369         }
 370 }
 371
 372 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
 373                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
 374                                     uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
 375 {
 376         //FIXME Optimize (just quickly writen not opti..)
 377         int i;
 378         for(i=0; i<dstW; i++)
 379         {
 380                 int val=0;
 381                 int j;
 382                 for(j=0; j<lumFilterSize; j++)
 383                         val += lumSrc[j][i] * lumFilter[j];
 384
 385                 dest[i]= MIN(MAX(val>>19, 0), 255);
 386         }
 387
 388         if(uDest != NULL)
 389                 for(i=0; i<chrDstW; i++)
 390                 {
 391                         int u=0;
 392                         int v=0;
 393                         int j;
 394                         for(j=0; j<chrFilterSize; j++)
 395                         {
 396                                 u += chrSrc[j][i] * chrFilter[j];
 397                                 v += chrSrc[j][i + 2048] * chrFilter[j];
 398                         }
 399
 400                         uDest[i]= MIN(MAX(u>>19, 0), 255);
 401                         vDest[i]= MIN(MAX(v>>19, 0), 255);
 402                 }
 403 }
 404
 405 #define YSCALE_YUV_2_RGBX_C(type) \
 406                 for(i=0; i<(dstW>>1); i++){\
 407                         int j;\
 408                         int Y1=0;\
 409                         int Y2=0;\
 410                         int U=0;\
 411                         int V=0;\
 412                         type *r, *b, *g;\
 413                         const int i2= 2*i;\
 414                         \
 415                         for(j=0; j<lumFilterSize; j++)\
 416                         {\
 417                                 Y1 += lumSrc[j][i2] * lumFilter[j];\
 418                                 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
 419                         }\
 420                         for(j=0; j<chrFilterSize; j++)\
 421                         {\
 422                                 U += chrSrc[j][i] * chrFilter[j];\
 423                                 V += chrSrc[j][i+2048] * chrFilter[j];\
 424                         }\
 425                         Y1>>=19;\
 426                         Y2>>=19;\
 427                         U >>=19;\
 428                         V >>=19;\
 429                         if((Y1|Y2|U|V)&256)\
 430                         {\
 431                                 if(Y1>255)   Y1=255;\
 432                                 else if(Y1<0)Y1=0;\
 433                                 if(Y2>255)   Y2=255;\
 434                                 else if(Y2<0)Y2=0;\
 435                                 if(U>255)    U=255;\
 436                                 else if(U<0) U=0;\
 437                                 if(V>255)    V=255;\
 438                                 else if(V<0) V=0;\
 439                         }\
 440                         r = c->table_rV[V];\
 441                         g = c->table_gU[U] + c->table_gV[V];\
 442                         b = c->table_bU[U];\
 443
 444 #define YSCALE_YUV_2_RGB2_C(type) \
 445                 for(i=0; i<(dstW>>1); i++){\
 446                         const int i2= 2*i;\
 447                         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
 448                         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
 449                         int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
 450                         int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
 451                         type *r, *b, *g;\
 452                         r = c->table_rV[V];\
 453                         g = c->table_gU[U] + c->table_gV[V];\
 454                         b = c->table_bU[U];\
 455
 456 #define YSCALE_YUV_2_RGB1_C(type) \
 457                 for(i=0; i<(dstW>>1); i++){\
 458                         const int i2= 2*i;\
 459                         int Y1= buf0[i2  ]>>7;\
 460                         int Y2= buf0[i2+1]>>7;\
 461                         int U= (uvbuf1[i     ])>>7;\
 462                         int V= (uvbuf1[i+2048])>>7;\
 463                         type *r, *b, *g;\
 464                         r = c->table_rV[V];\
 465                         g = c->table_gU[U] + c->table_gV[V];\
 466                         b = c->table_bU[U];\
 467
 468 #define YSCALE_YUV_2_RGB1B_C(type) \
 469                 for(i=0; i<(dstW>>1); i++){\
 470                         const int i2= 2*i;\
 471                         int Y1= buf0[i2  ]>>7;\
 472                         int Y2= buf0[i2+1]>>7;\
 473                         int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
 474                         int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
 475                         type *r, *b, *g;\
 476                         r = c->table_rV[V];\
 477                         g = c->table_gU[U] + c->table_gV[V];\
 478                         b = c->table_bU[U];\
 479
 480 #define YSCALE_YUV_2_ANYRGB_C(func)\
 481         switch(c->dstFormat)\
 482         {\
 483         case IMGFMT_BGR32:\
 484         case IMGFMT_RGB32:\
 485                 func(uint32_t)\
 486                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
 487                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
 488                 }               \
 489                 break;\
 490         case IMGFMT_RGB24:\
 491                 func(uint8_t)\
 492                         ((uint8_t*)dest)[0]= r[Y1];\
 493                         ((uint8_t*)dest)[1]= g[Y1];\
 494                         ((uint8_t*)dest)[2]= b[Y1];\
 495                         ((uint8_t*)dest)[3]= r[Y2];\
 496                         ((uint8_t*)dest)[4]= g[Y2];\
 497                         ((uint8_t*)dest)[5]= b[Y2];\
 498                         ((uint8_t*)dest)+=6;\
 499                 }\
 500                 break;\
 501         case IMGFMT_BGR24:\
 502                 func(uint8_t)\
 503                         ((uint8_t*)dest)[0]= b[Y1];\
 504                         ((uint8_t*)dest)[1]= g[Y1];\
 505                         ((uint8_t*)dest)[2]= r[Y1];\
 506                         ((uint8_t*)dest)[3]= b[Y2];\
 507                         ((uint8_t*)dest)[4]= g[Y2];\
 508                         ((uint8_t*)dest)[5]= r[Y2];\
 509                         ((uint8_t*)dest)+=6;\
 510                 }\
 511                 break;\
 512         case IMGFMT_RGB16:\
 513         case IMGFMT_BGR16:\
 514                 {\
 515                         const int dr1= dither_2x2_8[y&1    ][0];\
 516                         const int dg1= dither_2x2_4[y&1    ][0];\
 517                         const int db1= dither_2x2_8[(y&1)^1][0];\
 518                         const int dr2= dither_2x2_8[y&1    ][1];\
 519                         const int dg2= dither_2x2_4[y&1    ][1];\
 520                         const int db2= dither_2x2_8[(y&1)^1][1];\
 521                         func(uint16_t)\
 522                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
 523                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
 524                         }\
 525                 }\
 526                 break;\
 527         case IMGFMT_RGB15:\
 528         case IMGFMT_BGR15:\
 529                 {\
 530                         const int dr1= dither_2x2_8[y&1    ][0];\
 531                         const int dg1= dither_2x2_8[y&1    ][1];\
 532                         const int db1= dither_2x2_8[(y&1)^1][0];\
 533                         const int dr2= dither_2x2_8[y&1    ][1];\
 534                         const int dg2= dither_2x2_8[y&1    ][0];\
 535                         const int db2= dither_2x2_8[(y&1)^1][1];\
 536                         func(uint16_t)\
 537                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
 538                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
 539                         }\
 540                 }\
 541                 break;\
 542         case IMGFMT_RGB8:\
 543         case IMGFMT_BGR8:\
 544                 {\
 545                         const uint8_t * const d64= dither_8x8_73[y&7];\
 546                         const uint8_t * const d32= dither_8x8_32[y&7];\
 547                         func(uint8_t)\
 548                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
 549                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
 550                         }\
 551                 }\
 552                 break;\
 553         case IMGFMT_RGB4:\
 554         case IMGFMT_BGR4:\
 555                 {\
 556                         const uint8_t * const d64= dither_8x8_73 [y&7];\
 557                         const uint8_t * const d128=dither_8x8_220[y&7];\
 558                         func(uint8_t)\
 559                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
 560                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
 561                         }\
 562                 }\
 563                 break;\
 564         case IMGFMT_RGB1:\
 565         case IMGFMT_BGR1:\
 566                 {\
 567                         const uint8_t * const d128=dither_8x8_220[y&7];\
 568                         uint8_t *g= c->table_gU[128] + c->table_gV[128];\
 569                         for(i=0; i<dstW-7; i+=8){\
 570                                 int acc;\
 571                                 acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
 572                                 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
 573                                 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
 574                                 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
 575                                 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
 576                                 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
 577                                 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
 578                                 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
 579                                 ((uint8_t*)dest)[0]= acc;\
 580                                 ((uint8_t*)dest)++;\
 581                         }\
 582 \
 583 /*\
 584 ((uint8_t*)dest)-= dstW>>4;\
 585 {\
 586                         int acc=0;\
 587                         int left=0;\
 588                         static int top[1024];\
 589                         static int last_new[1024][1024];\
 590                         static int last_in3[1024][1024];\
 591                         static int drift[1024][1024];\
 592                         int topLeft=0;\
 593                         int shift=0;\
 594                         int count=0;\
 595                         const uint8_t * const d128=dither_8x8_220[y&7];\
 596                         int error_new=0;\
 597                         int error_in3=0;\
 598                         int f=0;\
 599                         \
 600                         for(i=dstW>>1; i<dstW; i++){\
 601                                 int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
 602                                 int in2 = (76309 * (in - 16) + 32768) >> 16;\
 603                                 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
 604                                 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
 605                                         + (last_new[y][i] - in3)*f/256;\
 606                                 int new= old> 128 ? 255 : 0;\
 607 \
 608                                 error_new+= ABS(last_new[y][i] - new);\
 609                                 error_in3+= ABS(last_in3[y][i] - in3);\
 610                                 f= error_new - error_in3*4;\
 611                                 if(f<0) f=0;\
 612                                 if(f>256) f=256;\
 613 \
 614                                 topLeft= top[i];\
 615                                 left= top[i]= old - new;\
 616                                 last_new[y][i]= new;\
 617                                 last_in3[y][i]= in3;\
 618 \
 619                                 acc+= acc + (new&1);\
 620                                 if((i&7)==6){\
 621                                         ((uint8_t*)dest)[0]= acc;\
 622                                         ((uint8_t*)dest)++;\
 623                                 }\
 624                         }\
 625 }\
 626 */\
 627                 }\
 628                 break;\
 629         }\
 630
 631
 632 static inline void yuv2rgbXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
 633                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
 634                                     uint8_t *dest, int dstW, int y)
 635 {
 636         int i;
 637         switch(c->dstFormat)
 638         {
 639         case IMGFMT_RGB32:
 640         case IMGFMT_BGR32:
 641                 YSCALE_YUV_2_RGBX_C(uint32_t)
 642                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
 643                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
 644                 }
 645                 break;
 646         case IMGFMT_RGB24:
 647                 YSCALE_YUV_2_RGBX_C(uint8_t)
 648                         ((uint8_t*)dest)[0]= r[Y1];
 649                         ((uint8_t*)dest)[1]= g[Y1];
 650                         ((uint8_t*)dest)[2]= b[Y1];
 651                         ((uint8_t*)dest)[3]= r[Y2];
 652                         ((uint8_t*)dest)[4]= g[Y2];
 653                         ((uint8_t*)dest)[5]= b[Y2];
 654                         ((uint8_t*)dest)+=6;
 655                 }
 656                 break;
 657         case IMGFMT_BGR24:
 658                 YSCALE_YUV_2_RGBX_C(uint8_t)
 659                         ((uint8_t*)dest)[0]= b[Y1];
 660                         ((uint8_t*)dest)[1]= g[Y1];
 661                         ((uint8_t*)dest)[2]= r[Y1];
 662                         ((uint8_t*)dest)[3]= b[Y2];
 663                         ((uint8_t*)dest)[4]= g[Y2];
 664                         ((uint8_t*)dest)[5]= r[Y2];
 665                         ((uint8_t*)dest)+=6;
 666                 }
 667                 break;
 668         case IMGFMT_RGB16:
 669         case IMGFMT_BGR16:
 670                 {
 671                         const int dr1= dither_2x2_8[y&1    ][0];
 672                         const int dg1= dither_2x2_4[y&1    ][0];
 673                         const int db1= dither_2x2_8[(y&1)^1][0];
 674                         const int dr2= dither_2x2_8[y&1    ][1];
 675                         const int dg2= dither_2x2_4[y&1    ][1];
 676                         const int db2= dither_2x2_8[(y&1)^1][1];
 677                         YSCALE_YUV_2_RGBX_C(uint16_t)
 678                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
 679                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
 680                         }
 681                 }
 682                 break;
 683         case IMGFMT_RGB15:
 684         case IMGFMT_BGR15:
 685                 {
 686                         const int dr1= dither_2x2_8[y&1    ][0];
 687                         const int dg1= dither_2x2_8[y&1    ][1];
 688                         const int db1= dither_2x2_8[(y&1)^1][0];
 689                         const int dr2= dither_2x2_8[y&1    ][1];
 690                         const int dg2= dither_2x2_8[y&1    ][0];
 691                         const int db2= dither_2x2_8[(y&1)^1][1];
 692                         YSCALE_YUV_2_RGBX_C(uint16_t)
 693                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
 694                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
 695                         }
 696                 }
 697                 break;
 698         case IMGFMT_RGB8:
 699         case IMGFMT_BGR8:
 700                 {
 701                         const uint8_t * const d64= dither_8x8_73[y&7];
 702                         const uint8_t * const d32= dither_8x8_32[y&7];
 703                         YSCALE_YUV_2_RGBX_C(uint8_t)
 704                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
 705                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
 706                         }
 707                 }
 708                 break;
 709         case IMGFMT_RGB4:
 710         case IMGFMT_BGR4:
 711                 {
 712                         const uint8_t * const d64= dither_8x8_73 [y&7];
 713                         const uint8_t * const d128=dither_8x8_220[y&7];
 714                         YSCALE_YUV_2_RGBX_C(uint8_t)
 715                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
 716                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
 717                         }
 718                 }
 719                 break;
 720         case IMGFMT_RGB1:
 721         case IMGFMT_BGR1:
 722                 {
 723                         const uint8_t * const d128=dither_8x8_220[y&7];
 724                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
 725                         int acc=0;
 726                         for(i=0; i<dstW-1; i+=2){
 727                                 int j;
 728                                 int Y1=0;
 729                                 int Y2=0;
 730
 731                                 for(j=0; j<lumFilterSize; j++)
 732                                 {
 733                                         Y1 += lumSrc[j][i] * lumFilter[j];
 734                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
 735                                 }
 736                                 Y1>>=19;
 737                                 Y2>>=19;
 738                                 if((Y1|Y2)&256)
 739                                 {
 740                                         if(Y1>255)   Y1=255;
 741                                         else if(Y1<0)Y1=0;
 742                                         if(Y2>255)   Y2=255;
 743                                         else if(Y2<0)Y2=0;
 744                                 }
 745                                 acc+= acc + g[Y1+d128[(i+0)&7]];
 746                                 acc+= acc + g[Y2+d128[(i+1)&7]];
 747                                 if((i&7)==6){
 748                                         ((uint8_t*)dest)[0]= acc;
 749                                         ((uint8_t*)dest)++;
 750                                 }
 751                         }
 752                 }
 753                 break;
 754         }
 755 }
 756
 757
 758 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
 759 //Plain C versions
 760 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
 761 #define COMPILE_C
 762 #endif
 763
 764 #ifdef CAN_COMPILE_X86_ASM
 765
 766 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
 767 #define COMPILE_MMX
 768 #endif
 769
 770 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
 771 #define COMPILE_MMX2
 772 #endif
 773
 774 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
 775 #define COMPILE_3DNOW
 776 #endif
 777 #endif //CAN_COMPILE_X86_ASM
 778
 779 #undef HAVE_MMX
 780 #undef HAVE_MMX2
 781 #undef HAVE_3DNOW
 782
 783 #ifdef COMPILE_C
 784 #undef HAVE_MMX
 785 #undef HAVE_MMX2
 786 #undef HAVE_3DNOW
 787 #define RENAME(a) a ## _C
 788 #include "swscale_template.c"
 789 #endif
 790
 791 #ifdef CAN_COMPILE_X86_ASM
 792
 793 //X86 versions
 794 /*
 795 #undef RENAME
 796 #undef HAVE_MMX
 797 #undef HAVE_MMX2
 798 #undef HAVE_3DNOW
 799 #define ARCH_X86
 800 #define RENAME(a) a ## _X86
 801 #include "swscale_template.c"
 802 */
 803 //MMX versions
 804 #ifdef COMPILE_MMX
 805 #undef RENAME
 806 #define HAVE_MMX
 807 #undef HAVE_MMX2
 808 #undef HAVE_3DNOW
 809 #define RENAME(a) a ## _MMX
 810 #include "swscale_template.c"
 811 #endif
 812
 813 //MMX2 versions
 814 #ifdef COMPILE_MMX2
 815 #undef RENAME
 816 #define HAVE_MMX
 817 #define HAVE_MMX2
 818 #undef HAVE_3DNOW
 819 #define RENAME(a) a ## _MMX2
 820 #include "swscale_template.c"
 821 #endif
 822
 823 //3DNOW versions
 824 #ifdef COMPILE_3DNOW
 825 #undef RENAME
 826 #define HAVE_MMX
 827 #undef HAVE_MMX2
 828 #define HAVE_3DNOW
 829 #define RENAME(a) a ## _3DNow
 830 #include "swscale_template.c"
 831 #endif
 832
 833 #endif //CAN_COMPILE_X86_ASM
 834
 835 // minor note: the HAVE_xyz is messed up after that line so dont use it
 836
 837
 838 // old global scaler, dont use for new code
 839 // will use sws_flags from the command line
 840 void SwScale_YV12slice(unsigned char* src[], int srcStride[], int srcSliceY ,
 841                              int srcSliceH, uint8_t* dst[], int dstStride, int dstbpp,
 842                              int srcW, int srcH, int dstW, int dstH){
 843
 844         static SwsContext *context=NULL;
 845         int dstFormat;
 846         int dstStride3[3]= {dstStride, dstStride>>1, dstStride>>1};
 847
 848         switch(dstbpp)
 849         {
 850                 case 8 : dstFormat= IMGFMT_Y8;          break;
 851                 case 12: dstFormat= IMGFMT_YV12;        break;
 852                 case 15: dstFormat= IMGFMT_BGR15;       break;
 853                 case 16: dstFormat= IMGFMT_BGR16;       break;
 854                 case 24: dstFormat= IMGFMT_BGR24;       break;
 855                 case 32: dstFormat= IMGFMT_BGR32;       break;
 856                 default: return;
 857         }
 858
 859         if(!context) context=getSwsContextFromCmdLine(srcW, srcH, IMGFMT_YV12, dstW, dstH, dstFormat);
 860
 861         context->swScale(context, src, srcStride, srcSliceY, srcSliceH, dst, dstStride3);
 862 }
 863
 864 void swsGetFlagsAndFilterFromCmdLine(int *flags, SwsFilter **srcFilterParam, SwsFilter **dstFilterParam)
 865 {
 866         static int firstTime=1;
 867         *flags=0;
 868
 869 #ifdef ARCH_X86
 870         if(gCpuCaps.hasMMX)
 871                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
 872 #endif
 873         if(firstTime)
 874         {
 875                 firstTime=0;
 876                 *flags= SWS_PRINT_INFO;
 877         }
 878         else if(verbose>1) *flags= SWS_PRINT_INFO;
 879
 880         if(src_filter.lumH) freeVec(src_filter.lumH);
 881         if(src_filter.lumV) freeVec(src_filter.lumV);
 882         if(src_filter.chrH) freeVec(src_filter.chrH);
 883         if(src_filter.chrV) freeVec(src_filter.chrV);
 884
 885         if(sws_lum_gblur!=0.0){
 886                 src_filter.lumH= getGaussianVec(sws_lum_gblur, 3.0);
 887                 src_filter.lumV= getGaussianVec(sws_lum_gblur, 3.0);
 888         }else{
 889                 src_filter.lumH= getIdentityVec();
 890                 src_filter.lumV= getIdentityVec();
 891         }
 892
 893         if(sws_chr_gblur!=0.0){
 894                 src_filter.chrH= getGaussianVec(sws_chr_gblur, 3.0);
 895                 src_filter.chrV= getGaussianVec(sws_chr_gblur, 3.0);
 896         }else{
 897                 src_filter.chrH= getIdentityVec();
 898                 src_filter.chrV= getIdentityVec();
 899         }
 900
 901         if(sws_chr_sharpen!=0.0){
 902                 SwsVector *g= getConstVec(-1.0, 3);
 903                 SwsVector *id= getConstVec(10.0/sws_chr_sharpen, 1);
 904                 g->coeff[1]=2.0;
 905                 addVec(id, g);
 906                 convVec(src_filter.chrH, id);
 907                 convVec(src_filter.chrV, id);
 908                 freeVec(g);
 909                 freeVec(id);
 910         }
 911
 912         if(sws_lum_sharpen!=0.0){
 913                 SwsVector *g= getConstVec(-1.0, 3);
 914                 SwsVector *id= getConstVec(10.0/sws_lum_sharpen, 1);
 915                 g->coeff[1]=2.0;
 916                 addVec(id, g);
 917                 convVec(src_filter.lumH, id);
 918                 convVec(src_filter.lumV, id);
 919                 freeVec(g);
 920                 freeVec(id);
 921         }
 922
 923         if(sws_chr_hshift)
 924                 shiftVec(src_filter.chrH, sws_chr_hshift);
 925
 926         if(sws_chr_vshift)
 927                 shiftVec(src_filter.chrV, sws_chr_vshift);
 928
 929         normalizeVec(src_filter.chrH, 1.0);
 930         normalizeVec(src_filter.chrV, 1.0);
 931         normalizeVec(src_filter.lumH, 1.0);
 932         normalizeVec(src_filter.lumV, 1.0);
 933
 934         if(verbose > 1) printVec(src_filter.chrH);
 935         if(verbose > 1) printVec(src_filter.lumH);
 936
 937         switch(sws_flags)
 938         {
 939                 case 0: *flags|= SWS_FAST_BILINEAR; break;
 940                 case 1: *flags|= SWS_BILINEAR; break;
 941                 case 2: *flags|= SWS_BICUBIC; break;
 942                 case 3: *flags|= SWS_X; break;
 943                 case 4: *flags|= SWS_POINT; break;
 944                 case 5: *flags|= SWS_AREA; break;
 945                 case 6: *flags|= SWS_BICUBLIN; break;
 946                 default:*flags|= SWS_BILINEAR; break;
 947         }
 948
 949         *srcFilterParam= &src_filter;
 950         *dstFilterParam= NULL;
 951 }
 952
 953 // will use sws_flags & src_filter (from cmd line)
 954 SwsContext *getSwsContextFromCmdLine(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat)
 955 {
 956         int flags;
 957         SwsFilter *dstFilterParam, *srcFilterParam;
 958         swsGetFlagsAndFilterFromCmdLine(&flags, &srcFilterParam, &dstFilterParam);
 959
 960         return getSwsContext(srcW, srcH, srcFormat, dstW, dstH, dstFormat, flags, srcFilterParam, dstFilterParam);
 961 }
 962
 963
 964 static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
 965                               int srcW, int dstW, int filterAlign, int one, int flags,
 966                               SwsVector *srcFilter, SwsVector *dstFilter)
 967 {
 968         int i;
 969         int filterSize;
 970         int filter2Size;
 971         int minFilterSize;
 972         double *filter=NULL;
 973         double *filter2=NULL;
 974 #ifdef ARCH_X86
 975         if(gCpuCaps.hasMMX)
 976                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
 977 #endif
 978
 979         // Note the +1 is for the MMXscaler which reads over the end
 980         *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
 981
 982         if(ABS(xInc - 0x10000) <10) // unscaled
 983         {
 984                 int i;
 985                 filterSize= 1;
 986                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
 987                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
 988
 989                 for(i=0; i<dstW; i++)
 990                 {
 991                         filter[i*filterSize]=1;
 992                         (*filterPos)[i]=i;
 993                 }
 994
 995         }
 996         else if(flags&SWS_POINT) // lame looking point sampling mode
 997         {
 998                 int i;
 999                 int xDstInSrc;
1000                 filterSize= 1;
1001                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
1002
1003                 xDstInSrc= xInc/2 - 0x8000;
1004                 for(i=0; i<dstW; i++)
1005                 {
1006                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1007
1008                         (*filterPos)[i]= xx;
1009                         filter[i]= 1.0;
1010                         xDstInSrc+= xInc;
1011                 }
1012         }
1013         else if(xInc <= (1<<16) || (flags&SWS_FAST_BILINEAR)) // upscale
1014         {
1015                 int i;
1016                 int xDstInSrc;
1017                 if     (flags&SWS_BICUBIC) filterSize= 4;
1018                 else if(flags&SWS_X      ) filterSize= 4;
1019                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA
1020                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
1021
1022                 xDstInSrc= xInc/2 - 0x8000;
1023                 for(i=0; i<dstW; i++)
1024                 {
1025                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1026                         int j;
1027
1028                         (*filterPos)[i]= xx;
1029                         if((flags & SWS_BICUBIC) || (flags & SWS_X))
1030                         {
1031                                 double d= ABS(((xx+1)<<16) - xDstInSrc)/(double)(1<<16);
1032                                 double y1,y2,y3,y4;
1033                                 double A= -0.6;
1034                                 if(flags & SWS_BICUBIC){
1035                                                 // Equation is from VirtualDub
1036                                         y1 = (        +     A*d -       2.0*A*d*d +       A*d*d*d);
1037                                         y2 = (+ 1.0             -     (A+3.0)*d*d + (A+2.0)*d*d*d);
1038                                         y3 = (        -     A*d + (2.0*A+3.0)*d*d - (A+2.0)*d*d*d);
1039                                         y4 = (                  +           A*d*d -       A*d*d*d);
1040                                 }else{
1041                                                 // cubic interpolation (derived it myself)
1042                                         y1 = (    -2.0*d + 3.0*d*d - 1.0*d*d*d)/6.0;
1043                                         y2 = (6.0 -3.0*d - 6.0*d*d + 3.0*d*d*d)/6.0;
1044                                         y3 = (    +6.0*d + 3.0*d*d - 3.0*d*d*d)/6.0;
1045                                         y4 = (    -1.0*d           + 1.0*d*d*d)/6.0;
1046                                 }
1047
1048                                 filter[i*filterSize + 0]= y1;
1049                                 filter[i*filterSize + 1]= y2;
1050                                 filter[i*filterSize + 2]= y3;
1051                                 filter[i*filterSize + 3]= y4;
1052                         }
1053                         else
1054                         {
1055                                 //Bilinear upscale / linear interpolate / Area averaging
1056                                 for(j=0; j<filterSize; j++)
1057                                 {
1058                                         double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
1059                                         double coeff= 1.0 - d;
1060                                         if(coeff<0) coeff=0;
1061                                         filter[i*filterSize + j]= coeff;
1062                                         xx++;
1063                                 }
1064                         }
1065                         xDstInSrc+= xInc;
1066                 }
1067         }
1068         else // downscale
1069         {
1070                 int xDstInSrc;
1071                 ASSERT(dstW <= srcW)
1072
1073                 if(flags&SWS_BICUBIC)   filterSize= (int)ceil(1 + 4.0*srcW / (double)dstW);
1074                 else if(flags&SWS_X)    filterSize= (int)ceil(1 + 4.0*srcW / (double)dstW);
1075                 else if(flags&SWS_AREA) filterSize= (int)ceil(1 + 1.0*srcW / (double)dstW);
1076                 else /* BILINEAR */     filterSize= (int)ceil(1 + 2.0*srcW / (double)dstW);
1077                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
1078
1079                 xDstInSrc= xInc/2 - 0x8000;
1080                 for(i=0; i<dstW; i++)
1081                 {
1082                         int xx= (int)((double)xDstInSrc/(double)(1<<16) - (filterSize-1)*0.5 + 0.5);
1083                         int j;
1084                         (*filterPos)[i]= xx;
1085                         for(j=0; j<filterSize; j++)
1086                         {
1087                                 double d= ABS((xx<<16) - xDstInSrc)/(double)xInc;
1088                                 double coeff;
1089                                 if((flags & SWS_BICUBIC) || (flags & SWS_X))
1090                                 {
1091                                         double A= -0.75;
1092 //                                      d*=2;
1093                                         // Equation is from VirtualDub
1094                                         if(d<1.0)
1095                                                 coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
1096                                         else if(d<2.0)
1097                                                 coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
1098                                         else
1099                                                 coeff=0.0;
1100                                 }
1101                                 else if(flags & SWS_AREA)
1102                                 {
1103                                         double srcPixelSize= (1<<16)/(double)xInc;
1104                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
1105                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1106                                         else coeff=0.0;
1107                                 }
1108                                 else
1109                                 {
1110                                         coeff= 1.0 - d;
1111                                         if(coeff<0) coeff=0;
1112                                 }
1113                                 filter[i*filterSize + j]= coeff;
1114                                 xx++;
1115                         }
1116                         xDstInSrc+= xInc;
1117                 }
1118         }
1119
1120         /* apply src & dst Filter to filter -> filter2
1121            free(filter);
1122         */
1123         ASSERT(filterSize>0)
1124         filter2Size= filterSize;
1125         if(srcFilter) filter2Size+= srcFilter->length - 1;
1126         if(dstFilter) filter2Size+= dstFilter->length - 1;
1127         ASSERT(filter2Size>0)
1128         filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
1129
1130         for(i=0; i<dstW; i++)
1131         {
1132                 int j;
1133                 SwsVector scaleFilter;
1134                 SwsVector *outVec;
1135
1136                 scaleFilter.coeff= filter + i*filterSize;
1137                 scaleFilter.length= filterSize;
1138
1139                 if(srcFilter) outVec= getConvVec(srcFilter, &scaleFilter);
1140                 else          outVec= &scaleFilter;
1141
1142                 ASSERT(outVec->length == filter2Size)
1143                 //FIXME dstFilter
1144
1145                 for(j=0; j<outVec->length; j++)
1146                 {
1147                         filter2[i*filter2Size + j]= outVec->coeff[j];
1148                 }
1149
1150                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1151
1152                 if(outVec != &scaleFilter) freeVec(outVec);
1153         }
1154         free(filter); filter=NULL;
1155
1156         /* try to reduce the filter-size (step1 find size and shift left) */
1157         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1158         minFilterSize= 0;
1159         for(i=dstW-1; i>=0; i--)
1160         {
1161                 int min= filter2Size;
1162                 int j;
1163                 double cutOff=0.0;
1164
1165                 /* get rid off near zero elements on the left by shifting left */
1166                 for(j=0; j<filter2Size; j++)
1167                 {
1168                         int k;
1169                         cutOff += ABS(filter2[i*filter2Size]);
1170
1171                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1172
1173                         /* preserve Monotonicity because the core cant handle the filter otherwise */
1174                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1175
1176                         // Move filter coeffs left
1177                         for(k=1; k<filter2Size; k++)
1178                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1179                         filter2[i*filter2Size + k - 1]= 0.0;
1180                         (*filterPos)[i]++;
1181                 }
1182
1183                 cutOff=0.0;
1184                 /* count near zeros on the right */
1185                 for(j=filter2Size-1; j>0; j--)
1186                 {
1187                         cutOff += ABS(filter2[i*filter2Size + j]);
1188
1189                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1190                         min--;
1191                 }
1192
1193                 if(min>minFilterSize) minFilterSize= min;
1194         }
1195
1196         ASSERT(minFilterSize > 0)
1197         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1198         ASSERT(filterSize > 0)
1199         filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1200         *outFilterSize= filterSize;
1201
1202         if(flags&SWS_PRINT_INFO)
1203                 MSG_INFO("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1204         /* try to reduce the filter-size (step2 reduce it) */
1205         for(i=0; i<dstW; i++)
1206         {
1207                 int j;
1208
1209                 for(j=0; j<filterSize; j++)
1210                 {
1211                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1212                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1213                 }
1214         }
1215         free(filter2); filter2=NULL;
1216
1217
1218         //FIXME try to align filterpos if possible
1219
1220         //fix borders
1221         for(i=0; i<dstW; i++)
1222         {
1223                 int j;
1224                 if((*filterPos)[i] < 0)
1225                 {
1226                         // Move filter coeffs left to compensate for filterPos
1227                         for(j=1; j<filterSize; j++)
1228                         {
1229                                 int left= MAX(j + (*filterPos)[i], 0);
1230                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1231                                 filter[i*filterSize + j]=0;
1232                         }
1233                         (*filterPos)[i]= 0;
1234                 }
1235
1236                 if((*filterPos)[i] + filterSize > srcW)
1237                 {
1238                         int shift= (*filterPos)[i] + filterSize - srcW;
1239                         // Move filter coeffs right to compensate for filterPos
1240                         for(j=filterSize-2; j>=0; j--)
1241                         {
1242                                 int right= MIN(j + shift, filterSize-1);
1243                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1244                                 filter[i*filterSize +j]=0;
1245                         }
1246                         (*filterPos)[i]= srcW - filterSize;
1247                 }
1248         }
1249
1250         // Note the +1 is for the MMXscaler which reads over the end
1251         *outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));
1252         memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1253
1254         /* Normalize & Store in outFilter */
1255         for(i=0; i<dstW; i++)
1256         {
1257                 int j;
1258                 double sum=0;
1259                 double scale= one;
1260                 for(j=0; j<filterSize; j++)
1261                 {
1262                         sum+= filter[i*filterSize + j];
1263                 }
1264                 scale/= sum;
1265                 for(j=0; j<filterSize; j++)
1266                 {
1267                         (*outFilter)[i*(*outFilterSize) + j]= (int)(filter[i*filterSize + j]*scale);
1268                 }
1269         }
1270
1271         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1272         for(i=0; i<*outFilterSize; i++)
1273         {
1274                 int j= dstW*(*outFilterSize);
1275                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1276         }
1277
1278         free(filter);
1279 }
1280
1281 #ifdef ARCH_X86
1282 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1283 {
1284         uint8_t *fragmentA;
1285         int imm8OfPShufW1A;
1286         int imm8OfPShufW2A;
1287         int fragmentLengthA;
1288         uint8_t *fragmentB;
1289         int imm8OfPShufW1B;
1290         int imm8OfPShufW2B;
1291         int fragmentLengthB;
1292         int fragmentPos;
1293
1294         int xpos, i;
1295
1296         // create an optimized horizontal scaling routine
1297
1298         //code fragment
1299
1300         asm volatile(
1301                 "jmp 9f                         \n\t"
1302         // Begin
1303                 "0:                             \n\t"
1304                 "movq (%%edx, %%eax), %%mm3     \n\t"
1305                 "movd (%%ecx, %%esi), %%mm0     \n\t"
1306                 "movd 1(%%ecx, %%esi), %%mm1    \n\t"
1307                 "punpcklbw %%mm7, %%mm1         \n\t"
1308                 "punpcklbw %%mm7, %%mm0         \n\t"
1309                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1310                 "1:                             \n\t"
1311                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1312                 "2:                             \n\t"
1313                 "psubw %%mm1, %%mm0             \n\t"
1314                 "movl 8(%%ebx, %%eax), %%esi    \n\t"
1315                 "pmullw %%mm3, %%mm0            \n\t"
1316                 "psllw $7, %%mm1                \n\t"
1317                 "paddw %%mm1, %%mm0             \n\t"
1318
1319                 "movq %%mm0, (%%edi, %%eax)     \n\t"
1320
1321                 "addl $8, %%eax                 \n\t"
1322         // End
1323                 "9:                             \n\t"
1324 //              "int $3\n\t"
1325                 "leal 0b, %0                    \n\t"
1326                 "leal 1b, %1                    \n\t"
1327                 "leal 2b, %2                    \n\t"
1328                 "decl %1                        \n\t"
1329                 "decl %2                        \n\t"
1330                 "subl %0, %1                    \n\t"
1331                 "subl %0, %2                    \n\t"
1332                 "leal 9b, %3                    \n\t"
1333                 "subl %0, %3                    \n\t"
1334
1335
1336                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1337                 "=r" (fragmentLengthA)
1338         );
1339
1340         asm volatile(
1341                 "jmp 9f                         \n\t"
1342         // Begin
1343                 "0:                             \n\t"
1344                 "movq (%%edx, %%eax), %%mm3     \n\t"
1345                 "movd (%%ecx, %%esi), %%mm0     \n\t"
1346                 "punpcklbw %%mm7, %%mm0         \n\t"
1347                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1348                 "1:                             \n\t"
1349                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1350                 "2:                             \n\t"
1351                 "psubw %%mm1, %%mm0             \n\t"
1352                 "movl 8(%%ebx, %%eax), %%esi    \n\t"
1353                 "pmullw %%mm3, %%mm0            \n\t"
1354                 "psllw $7, %%mm1                \n\t"
1355                 "paddw %%mm1, %%mm0             \n\t"
1356
1357                 "movq %%mm0, (%%edi, %%eax)     \n\t"
1358
1359                 "addl $8, %%eax                 \n\t"
1360         // End
1361                 "9:                             \n\t"
1362 //              "int $3\n\t"
1363                 "leal 0b, %0                    \n\t"
1364                 "leal 1b, %1                    \n\t"
1365                 "leal 2b, %2                    \n\t"
1366                 "decl %1                        \n\t"
1367                 "decl %2                        \n\t"
1368                 "subl %0, %1                    \n\t"
1369                 "subl %0, %2                    \n\t"
1370                 "leal 9b, %3                    \n\t"
1371                 "subl %0, %3                    \n\t"
1372
1373
1374                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1375                 "=r" (fragmentLengthB)
1376         );
1377
1378         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1379         fragmentPos=0;
1380
1381         for(i=0; i<dstW/numSplits; i++)
1382         {
1383                 int xx=xpos>>16;
1384
1385                 if((i&3) == 0)
1386                 {
1387                         int a=0;
1388                         int b=((xpos+xInc)>>16) - xx;
1389                         int c=((xpos+xInc*2)>>16) - xx;
1390                         int d=((xpos+xInc*3)>>16) - xx;
1391
1392                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1393                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1394                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1395                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1396                         filterPos[i/2]= xx;
1397
1398                         if(d+1<4)
1399                         {
1400                                 int maxShift= 3-(d+1);
1401                                 int shift=0;
1402
1403                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1404
1405                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1406                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1407                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1408                                         a | (b<<2) | (c<<4) | (d<<6);
1409
1410                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1411                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1412
1413                                 if(shift && i>=shift)
1414                                 {
1415                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1416                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1417                                         filterPos[i/2]-=shift;
1418                                 }
1419
1420                                 fragmentPos+= fragmentLengthB;
1421                         }
1422                         else
1423                         {
1424                                 int maxShift= 3-d;
1425                                 int shift=0;
1426
1427                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1428
1429                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1430                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1431                                         a | (b<<2) | (c<<4) | (d<<6);
1432
1433                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1434                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1435
1436                                 if(shift && i>=shift)
1437                                 {
1438                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1439                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1440                                         filterPos[i/2]-=shift;
1441                                 }
1442
1443                                 fragmentPos+= fragmentLengthA;
1444                         }
1445
1446                         funnyCode[fragmentPos]= RET;
1447                 }
1448                 xpos+=xInc;
1449         }
1450         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1451 }
1452 #endif // ARCH_X86
1453
1454 //FIXME remove
1455 void SwScale_Init(){
1456 }
1457
1458 static void globalInit(){
1459     // generating tables:
1460     int i;
1461     for(i=0; i<768; i++){
1462         int c= MIN(MAX(i-256, 0), 255);
1463         clip_table[i]=c;
1464     }
1465
1466 cpuCaps= gCpuCaps;
1467
1468 #ifdef RUNTIME_CPUDETECT
1469 #ifdef CAN_COMPILE_X86_ASM
1470         // ordered per speed fasterst first
1471         if(gCpuCaps.hasMMX2)
1472                 swScale= swScale_MMX2;
1473         else if(gCpuCaps.has3DNow)
1474                 swScale= swScale_3DNow;
1475         else if(gCpuCaps.hasMMX)
1476                 swScale= swScale_MMX;
1477         else
1478                 swScale= swScale_C;
1479
1480 #else
1481         swScale= swScale_C;
1482         cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;
1483 #endif
1484 #else //RUNTIME_CPUDETECT
1485 #ifdef HAVE_MMX2
1486         swScale= swScale_MMX2;
1487         cpuCaps.has3DNow = 0;
1488 #elif defined (HAVE_3DNOW)
1489         swScale= swScale_3DNow;
1490         cpuCaps.hasMMX2 = 0;
1491 #elif defined (HAVE_MMX)
1492         swScale= swScale_MMX;
1493         cpuCaps.hasMMX2 = cpuCaps.has3DNow = 0;
1494 #else
1495         swScale= swScale_C;
1496         cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;
1497 #endif
1498 #endif //!RUNTIME_CPUDETECT
1499 }
1500
1501 static void PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1502              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1503         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1504         /* Copy Y plane */
1505         if(dstStride[0]==srcStride[0])
1506                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1507         else
1508         {
1509                 int i;
1510                 uint8_t *srcPtr= src[0];
1511                 uint8_t *dstPtr= dst;
1512                 for(i=0; i<srcSliceH; i++)
1513                 {
1514                         memcpy(dstPtr, srcPtr, srcStride[0]);
1515                         srcPtr+= srcStride[0];
1516                         dstPtr+= dstStride[0];
1517                 }
1518         }
1519         dst = dstParam[1] + dstStride[1]*srcSliceY;
1520         if(c->srcFormat==IMGFMT_YV12)
1521                 interleaveBytes( src[1],src[2],dst,c->srcW,srcSliceH,srcStride[1],srcStride[2],dstStride[0] );
1522         else /* I420 & IYUV */
1523                 interleaveBytes( src[2],src[1],dst,c->srcW,srcSliceH,srcStride[2],srcStride[1],dstStride[0] );
1524 }
1525
1526
1527 /* Warper functions for yuv2bgr */
1528 static void planarYuvToBgr(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1529              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1530         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1531
1532         if(c->srcFormat==IMGFMT_YV12)
1533                 yuv2rgb( dst,src[0],src[1],src[2],c->srcW,srcSliceH,dstStride[0],srcStride[0],srcStride[1] );
1534         else /* I420 & IYUV */
1535                 yuv2rgb( dst,src[0],src[2],src[1],c->srcW,srcSliceH,dstStride[0],srcStride[0],srcStride[1] );
1536 }
1537
1538 static void PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1539              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1540         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1541
1542         if(c->srcFormat==IMGFMT_YV12)
1543                 yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1544         else /* I420 & IYUV */
1545                 yv12toyuy2( src[0],src[2],src[1],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1546 }
1547
1548 static void bgr24to32Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1549              int srcSliceH, uint8_t* dst[], int dstStride[]){
1550
1551         if(dstStride[0]*3==srcStride[0]*4)
1552                 rgb24to32(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1553         else
1554         {
1555                 int i;
1556                 uint8_t *srcPtr= src[0];
1557                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1558
1559                 for(i=0; i<srcSliceH; i++)
1560                 {
1561                         rgb24to32(srcPtr, dstPtr, c->srcW*3);
1562                         srcPtr+= srcStride[0];
1563                         dstPtr+= dstStride[0];
1564                 }
1565         }
1566 }
1567
1568 static void bgr24to16Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1569              int srcSliceH, uint8_t* dst[], int dstStride[]){
1570
1571         if(dstStride[0]*3==srcStride[0]*2)
1572                 rgb24to16(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1573         else
1574         {
1575                 int i;
1576                 uint8_t *srcPtr= src[0];
1577                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1578
1579                 for(i=0; i<srcSliceH; i++)
1580                 {
1581                         rgb24to16(srcPtr, dstPtr, c->srcW*3);
1582                         srcPtr+= srcStride[0];
1583                         dstPtr+= dstStride[0];
1584                 }
1585         }
1586 }
1587
1588 static void bgr24to15Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1589              int srcSliceH, uint8_t* dst[], int dstStride[]){
1590
1591         if(dstStride[0]*3==srcStride[0]*2)
1592                 rgb24to15(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1593         else
1594         {
1595                 int i;
1596                 uint8_t *srcPtr= src[0];
1597                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1598
1599                 for(i=0; i<srcSliceH; i++)
1600                 {
1601                         rgb24to15(srcPtr, dstPtr, c->srcW*3);
1602                         srcPtr+= srcStride[0];
1603                         dstPtr+= dstStride[0];
1604                 }
1605         }
1606 }
1607
1608 static void bgr32to24Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1609              int srcSliceH, uint8_t* dst[], int dstStride[]){
1610
1611         if(dstStride[0]*4==srcStride[0]*3)
1612                 rgb32to24(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1613         else
1614         {
1615                 int i;
1616                 uint8_t *srcPtr= src[0];
1617                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1618
1619                 for(i=0; i<srcSliceH; i++)
1620                 {
1621                         rgb32to24(srcPtr, dstPtr, c->srcW<<2);
1622                         srcPtr+= srcStride[0];
1623                         dstPtr+= dstStride[0];
1624                 }
1625         }
1626 }
1627
1628 static void bgr32to16Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1629              int srcSliceH, uint8_t* dst[], int dstStride[]){
1630
1631         if(dstStride[0]*4==srcStride[0]*2)
1632                 rgb32to16(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1633         else
1634         {
1635                 int i;
1636                 uint8_t *srcPtr= src[0];
1637                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1638
1639                 for(i=0; i<srcSliceH; i++)
1640                 {
1641                         rgb32to16(srcPtr, dstPtr, c->srcW<<2);
1642                         srcPtr+= srcStride[0];
1643                         dstPtr+= dstStride[0];
1644                 }
1645         }
1646 }
1647
1648 static void bgr32to15Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1649              int srcSliceH, uint8_t* dst[], int dstStride[]){
1650
1651         if(dstStride[0]*4==srcStride[0]*2)
1652                 rgb32to15(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1653         else
1654         {
1655                 int i;
1656                 uint8_t *srcPtr= src[0];
1657                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1658
1659                 for(i=0; i<srcSliceH; i++)
1660                 {
1661                         rgb32to15(srcPtr, dstPtr, c->srcW<<2);
1662                         srcPtr+= srcStride[0];
1663                         dstPtr+= dstStride[0];
1664                 }
1665         }
1666 }
1667
1668 static void bgr15to16Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1669              int srcSliceH, uint8_t* dst[], int dstStride[]){
1670
1671         if(dstStride[0]==srcStride[0])
1672                 rgb15to16(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1673         else
1674         {
1675                 int i;
1676                 uint8_t *srcPtr= src[0];
1677                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1678
1679                 for(i=0; i<srcSliceH; i++)
1680                 {
1681                         rgb15to16(srcPtr, dstPtr, c->srcW<<1);
1682                         srcPtr+= srcStride[0];
1683                         dstPtr+= dstStride[0];
1684                 }
1685         }
1686 }
1687
1688 static void bgr15to24Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1689              int srcSliceH, uint8_t* dst[], int dstStride[]){
1690
1691         if(dstStride[0]*2==srcStride[0]*3)
1692                 rgb15to24(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1693         else
1694         {
1695                 int i;
1696                 uint8_t *srcPtr= src[0];
1697                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1698
1699                 for(i=0; i<srcSliceH; i++)
1700                 {
1701                         rgb15to24(srcPtr, dstPtr, c->srcW<<1);
1702                         srcPtr+= srcStride[0];
1703                         dstPtr+= dstStride[0];
1704                 }
1705         }
1706 }
1707
1708 static void bgr15to32Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1709              int srcSliceH, uint8_t* dst[], int dstStride[]){
1710
1711         if(dstStride[0]*2==srcStride[0]*4)
1712                 rgb15to32(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1713         else
1714         {
1715                 int i;
1716                 uint8_t *srcPtr= src[0];
1717                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1718
1719                 for(i=0; i<srcSliceH; i++)
1720                 {
1721                         rgb15to32(srcPtr, dstPtr, c->srcW<<1);
1722                         srcPtr+= srcStride[0];
1723                         dstPtr+= dstStride[0];
1724                 }
1725         }
1726 }
1727
1728 static void bgr16to24Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1729              int srcSliceH, uint8_t* dst[], int dstStride[]){
1730
1731         if(dstStride[0]*2==srcStride[0]*3)
1732                 rgb16to24(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1733         else
1734         {
1735                 int i;
1736                 uint8_t *srcPtr= src[0];
1737                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1738
1739                 for(i=0; i<srcSliceH; i++)
1740                 {
1741                         rgb16to24(srcPtr, dstPtr, c->srcW<<1);
1742                         srcPtr+= srcStride[0];
1743                         dstPtr+= dstStride[0];
1744                 }
1745         }
1746 }
1747
1748 static void bgr16to32Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1749              int srcSliceH, uint8_t* dst[], int dstStride[]){
1750
1751         if(dstStride[0]*2==srcStride[0]*4)
1752                 rgb16to32(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1753         else
1754         {
1755                 int i;
1756                 uint8_t *srcPtr= src[0];
1757                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1758
1759                 for(i=0; i<srcSliceH; i++)
1760                 {
1761                         rgb16to32(srcPtr, dstPtr, c->srcW<<1);
1762                         srcPtr+= srcStride[0];
1763                         dstPtr+= dstStride[0];
1764                 }
1765         }
1766 }
1767
1768 static void bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1769              int srcSliceH, uint8_t* dst[], int dstStride[]){
1770
1771         rgb24toyv12(
1772                 src[0],
1773                 dst[0]+ srcSliceY    *dstStride[0],
1774                 dst[1]+(srcSliceY>>1)*dstStride[1],
1775                 dst[2]+(srcSliceY>>1)*dstStride[2],
1776                 c->srcW, srcSliceH,
1777                 dstStride[0], dstStride[1], srcStride[0]);
1778 }
1779
1780 static void yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1781              int srcSliceH, uint8_t* dst[], int dstStride[]){
1782         int i;
1783
1784         /* copy Y */
1785         if(srcStride[0]==dstStride[0])
1786                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1787         else{
1788                 uint8_t *srcPtr= src[0];
1789                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1790
1791                 for(i=0; i<srcSliceH; i++)
1792                 {
1793                         memcpy(dstPtr, srcPtr, c->srcW);
1794                         srcPtr+= srcStride[0];
1795                         dstPtr+= dstStride[0];
1796                 }
1797         }
1798
1799         if(c->dstFormat==IMGFMT_YV12){
1800                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1801                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1802         }else{
1803                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1804                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1805         }
1806 }
1807
1808 /**
1809  * bring pointers in YUV order instead of YVU
1810  */
1811 static inline void orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1812         if(format == IMGFMT_YV12 || format == IMGFMT_YVU9){
1813                 sortedP[0]= p[0];
1814                 sortedP[1]= p[1];
1815                 sortedP[2]= p[2];
1816                 sortedStride[0]= stride[0];
1817                 sortedStride[1]= stride[1];
1818                 sortedStride[2]= stride[2];
1819         }
1820         else if(isPacked(format) || isGray(format))
1821         {
1822                 sortedP[0]= p[0];
1823                 sortedP[1]=
1824                 sortedP[2]= NULL;
1825                 sortedStride[0]= stride[0];
1826                 sortedStride[1]=
1827                 sortedStride[2]= 0;
1828         }
1829         else /* I420 */
1830         {
1831                 sortedP[0]= p[0];
1832                 sortedP[1]= p[2];
1833                 sortedP[2]= p[1];
1834                 sortedStride[0]= stride[0];
1835                 sortedStride[1]= stride[2];
1836                 sortedStride[2]= stride[1];
1837         }
1838 }
1839
1840 /* unscaled copy like stuff (assumes nearly identical formats) */
1841 static void simpleCopy(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
1842              int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
1843
1844         int srcStride[3];
1845         int dstStride[3];
1846         uint8_t *src[3];
1847         uint8_t *dst[3];
1848
1849         orderYUV(c->srcFormat, src, srcStride, srcParam, srcStrideParam);
1850         orderYUV(c->dstFormat, dst, dstStride, dstParam, dstStrideParam);
1851
1852         if(isPacked(c->srcFormat))
1853         {
1854                 if(dstStride[0]==srcStride[0])
1855                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1856                 else
1857                 {
1858                         int i;
1859                         uint8_t *srcPtr= src[0];
1860                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1861                         int length=0;
1862
1863                         /* universal length finder */
1864                         while(length+c->srcW <= ABS(dstStride[0])
1865                            && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
1866                         ASSERT(length!=0);
1867
1868                         for(i=0; i<srcSliceH; i++)
1869                         {
1870                                 memcpy(dstPtr, srcPtr, length);
1871                                 srcPtr+= srcStride[0];
1872                                 dstPtr+= dstStride[0];
1873                         }
1874                 }
1875         }
1876         else
1877         { /* Planar YUV or gray */
1878                 int plane;
1879                 for(plane=0; plane<3; plane++)
1880                 {
1881                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1882                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1883                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1884
1885                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1886                         {
1887                                 if(!isGray(c->dstFormat))
1888                                         memset(dst[plane], 128, dstStride[plane]*height);
1889                         }
1890                         else
1891                         {
1892                                 if(dstStride[plane]==srcStride[plane])
1893                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1894                                 else
1895                                 {
1896                                         int i;
1897                                         uint8_t *srcPtr= src[plane];
1898                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1899                                         for(i=0; i<height; i++)
1900                                         {
1901                                                 memcpy(dstPtr, srcPtr, length);
1902                                                 srcPtr+= srcStride[plane];
1903                                                 dstPtr+= dstStride[plane];
1904                                         }
1905                                 }
1906                         }
1907                 }
1908         }
1909 }
1910
1911 static int remove_dup_fourcc(int fourcc)
1912 {
1913         switch(fourcc)
1914         {
1915             case IMGFMT_IYUV: return IMGFMT_I420;
1916             case IMGFMT_Y8  : return IMGFMT_Y800;
1917             case IMGFMT_IF09: return IMGFMT_YVU9;
1918             default: return fourcc;
1919         }
1920 }
1921
1922 static void getSubSampleFactors(int *h, int *v, int format){
1923         switch(format){
1924         case IMGFMT_YUY2:
1925                 *h=1;
1926                 *v=0;
1927                 break;
1928         case IMGFMT_YV12:
1929         case IMGFMT_I420:
1930         case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
1931                 *h=1;
1932                 *v=1;
1933                 break;
1934         case IMGFMT_YVU9:
1935                 *h=2;
1936                 *v=2;
1937                 break;
1938         default:
1939                 *h=0;
1940                 *v=0;
1941                 break;
1942         }
1943 }
1944
1945 SwsContext *getSwsContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
1946                          SwsFilter *srcFilter, SwsFilter *dstFilter){
1947
1948         SwsContext *c;
1949         int i;
1950         int usesFilter;
1951         int unscaled;
1952         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1953 #ifdef ARCH_X86
1954         if(gCpuCaps.hasMMX)
1955                 asm volatile("emms\n\t"::: "memory");
1956 #endif
1957         if(swScale==NULL) globalInit();
1958 //srcFormat= IMGFMT_Y800;
1959 //srcFormat= IMGFMT_YVU9;
1960         /* avoid dupplicate Formats, so we dont need to check to much */
1961         srcFormat = remove_dup_fourcc(srcFormat);
1962         dstFormat = remove_dup_fourcc(dstFormat);
1963
1964         unscaled = (srcW == dstW && srcH == dstH);
1965
1966         if(!isSupportedIn(srcFormat))
1967         {
1968                 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1969                 return NULL;
1970         }
1971         if(!isSupportedOut(dstFormat))
1972         {
1973                 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1974                 return NULL;
1975         }
1976
1977         /* sanity check */
1978         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
1979         {
1980                  MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
1981                         srcW, srcH, dstW, dstH);
1982                 return NULL;
1983         }
1984
1985         if(!dstFilter) dstFilter= &dummyFilter;
1986         if(!srcFilter) srcFilter= &dummyFilter;
1987
1988         c= memalign(64, sizeof(SwsContext));
1989         memset(c, 0, sizeof(SwsContext));
1990
1991         c->srcW= srcW;
1992         c->srcH= srcH;
1993         c->dstW= dstW;
1994         c->dstH= dstH;
1995         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1996         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1997         c->flags= flags;
1998         c->dstFormat= dstFormat;
1999         c->srcFormat= srcFormat;
2000
2001         usesFilter=0;
2002         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesFilter=1;
2003         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesFilter=1;
2004         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesFilter=1;
2005         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesFilter=1;
2006         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesFilter=1;
2007         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesFilter=1;
2008         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesFilter=1;
2009         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesFilter=1;
2010
2011         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2012         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2013
2014         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2015         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2016
2017         // drop some chroma lines if the user wants it
2018         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2019         c->chrSrcVSubSample+= c->vChrDrop;
2020
2021         // drop every 2. pixel for chroma calculation unless user wants full chroma
2022         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
2023                 c->chrSrcHSubSample=1;
2024
2025         c->chrIntHSubSample= c->chrDstHSubSample;
2026         c->chrIntVSubSample= c->chrSrcVSubSample;
2027
2028         // note the -((-x)>>y) is so that we allways round toward +inf
2029         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2030         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2031         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2032         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2033
2034         if(isBGR(dstFormat))
2035                 c->yuvTable= yuv2rgb_c_init(dstFormat & 0xFF, MODE_RGB, c->table_rV, c->table_gU, c->table_gV, c->table_bU);
2036         if(isRGB(dstFormat))
2037                 c->yuvTable= yuv2rgb_c_init(dstFormat & 0xFF, MODE_BGR, c->table_rV, c->table_gU, c->table_gV, c->table_bU);
2038
2039         /* unscaled special Cases */
2040         if(unscaled && !usesFilter)
2041         {
2042                 /* yv12_to_nv12 */
2043                 if((srcFormat == IMGFMT_YV12||srcFormat==IMGFMT_I420)&&dstFormat == IMGFMT_NV12)
2044                 {
2045                         c->swScale= PlanarToNV12Wrapper;
2046
2047                         if(flags&SWS_PRINT_INFO)
2048                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2049                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2050                         return c;
2051                 }
2052                 /* yv12_to_yuy2 */
2053                 if((srcFormat == IMGFMT_YV12||srcFormat==IMGFMT_I420)&&dstFormat == IMGFMT_YUY2)
2054                 {
2055                         c->swScale= PlanarToYuy2Wrapper;
2056
2057                         if(flags&SWS_PRINT_INFO)
2058                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2059                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2060                         return c;
2061                 }
2062                 /* yuv2bgr */
2063                 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_I420) && isBGR(dstFormat))
2064                 {
2065                         // FIXME multiple yuv2rgb converters wont work that way cuz that thing is full of globals&statics
2066                         //FIXME rgb vs. bgr ?
2067 #ifdef WORDS_BIGENDIAN
2068                         if(dstFormat==IMGFMT_BGR32)
2069                                 yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_BGR);
2070                         else
2071                                 yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_RGB);
2072 #else
2073                         yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_RGB);
2074 #endif
2075                         c->swScale= planarYuvToBgr;
2076
2077                         if(flags&SWS_PRINT_INFO)
2078                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2079                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2080                         return c;
2081                 }
2082
2083                 /* simple copy */
2084                 if(   srcFormat == dstFormat
2085                    || (srcFormat==IMGFMT_YV12 && dstFormat==IMGFMT_I420)
2086                    || (srcFormat==IMGFMT_I420 && dstFormat==IMGFMT_YV12)
2087                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2088                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2089                   )
2090                 {
2091                         c->swScale= simpleCopy;
2092
2093                         if(flags&SWS_PRINT_INFO)
2094                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2095                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2096                         return c;
2097                 }
2098
2099                 if( srcFormat==IMGFMT_YVU9 && (dstFormat==IMGFMT_YV12 || dstFormat==IMGFMT_I420) )
2100                 {
2101                         c->swScale= yvu9toyv12Wrapper;
2102
2103                         if(flags&SWS_PRINT_INFO)
2104                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2105                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2106                         return c;
2107                 }
2108
2109                 /* bgr32to24 & rgb32to24*/
2110                 if((srcFormat==IMGFMT_BGR32 && dstFormat==IMGFMT_BGR24)
2111                  ||(srcFormat==IMGFMT_RGB32 && dstFormat==IMGFMT_RGB24))
2112                 {
2113                         c->swScale= bgr32to24Wrapper;
2114
2115                         if(flags&SWS_PRINT_INFO)
2116                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2117                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2118                         return c;
2119                 }
2120
2121                 /* bgr32to16 & rgb32to16*/
2122                 if((srcFormat==IMGFMT_BGR32 && dstFormat==IMGFMT_BGR16)
2123                  ||(srcFormat==IMGFMT_RGB32 && dstFormat==IMGFMT_RGB16))
2124                 {
2125                         c->swScale= bgr32to16Wrapper;
2126
2127                         if(flags&SWS_PRINT_INFO)
2128                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2129                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2130                         return c;
2131                 }
2132
2133                 /* bgr32to15 & rgb32to15*/
2134                 if((srcFormat==IMGFMT_BGR32 && dstFormat==IMGFMT_BGR15)
2135                  ||(srcFormat==IMGFMT_RGB32 && dstFormat==IMGFMT_RGB15))
2136                 {
2137                         c->swScale= bgr32to15Wrapper;
2138
2139                         if(flags&SWS_PRINT_INFO)
2140                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2141                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2142                         return c;
2143                 }
2144
2145                 /* bgr24to32 & rgb24to32*/
2146                 if((srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_BGR32)
2147                  ||(srcFormat==IMGFMT_RGB24 && dstFormat==IMGFMT_RGB32))
2148                 {
2149                         c->swScale= bgr24to32Wrapper;
2150
2151                         if(flags&SWS_PRINT_INFO)
2152                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2153                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2154                         return c;
2155                 }
2156
2157                 /* bgr24to16 & rgb24to16*/
2158                 if((srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_BGR16)
2159                  ||(srcFormat==IMGFMT_RGB24 && dstFormat==IMGFMT_RGB16))
2160                 {
2161                         c->swScale= bgr24to16Wrapper;
2162
2163                         if(flags&SWS_PRINT_INFO)
2164                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2165                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2166                         return c;
2167                 }
2168
2169                 /* bgr24to15 & rgb24to15*/
2170                 if((srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_BGR15)
2171                  ||(srcFormat==IMGFMT_RGB24 && dstFormat==IMGFMT_RGB15))
2172                 {
2173                         c->swScale= bgr24to15Wrapper;
2174
2175                         if(flags&SWS_PRINT_INFO)
2176                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2177                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2178                         return c;
2179                 }
2180
2181                 /* bgr15to16 */
2182                 if(srcFormat==IMGFMT_BGR15 && dstFormat==IMGFMT_BGR16)
2183                 {
2184                         c->swScale= bgr15to16Wrapper;
2185
2186                         if(flags&SWS_PRINT_INFO)
2187                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2188                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2189                         return c;
2190                 }
2191
2192                 /* bgr15to24 */
2193                 if((srcFormat==IMGFMT_BGR15 && dstFormat==IMGFMT_BGR24)
2194                  ||(srcFormat==IMGFMT_RGB15 && dstFormat==IMGFMT_RGB24))
2195                 {
2196                         c->swScale= bgr15to24Wrapper;
2197
2198                         if(flags&SWS_PRINT_INFO)
2199                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2200                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2201                         return c;
2202                 }
2203
2204 #if 0 //segfaults
2205                 /* bgr15to32 */
2206                 if((srcFormat==IMGFMT_BGR15 && dstFormat==IMGFMT_BGR32)
2207                  ||(srcFormat==IMGFMT_RGB15 && dstFormat==IMGFMT_RGB32))
2208                 {
2209                         c->swScale= bgr15to32Wrapper;
2210
2211                         if(flags&SWS_PRINT_INFO)
2212                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2213                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2214                         return c;
2215                 }
2216 #endif
2217                 /* bgr16to24 */
2218                 if((srcFormat==IMGFMT_BGR16 && dstFormat==IMGFMT_BGR24)
2219                  ||(srcFormat==IMGFMT_RGB16 && dstFormat==IMGFMT_RGB24))
2220                 {
2221                         c->swScale= bgr16to24Wrapper;
2222
2223                         if(flags&SWS_PRINT_INFO)
2224                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2225                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2226                         return c;
2227                 }
2228
2229 #if 0 //segfaults
2230                 /* bgr16to32 */
2231                 if((srcFormat==IMGFMT_BGR16 && dstFormat==IMGFMT_BGR32)
2232                  ||(srcFormat==IMGFMT_RGB16 && dstFormat==IMGFMT_RGB32))
2233                 {
2234                         c->swScale= bgr16to32Wrapper;
2235
2236                         if(flags&SWS_PRINT_INFO)
2237                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2238                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2239                         return c;
2240                 }
2241 #endif
2242                 /* bgr24toYV12 */
2243                 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
2244                 {
2245                         c->swScale= bgr24toyv12Wrapper;
2246
2247                         if(flags&SWS_PRINT_INFO)
2248                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2249                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2250                         return c;
2251                 }
2252         }
2253
2254         if(cpuCaps.hasMMX2)
2255         {
2256                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2257                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2258                 {
2259                         if(flags&SWS_PRINT_INFO)
2260                                 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
2261                 }
2262         }
2263         else
2264                 c->canMMX2BeUsed=0;
2265
2266         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2267         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2268
2269         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2270         // but only for the FAST_BILINEAR mode otherwise do correct scaling
2271         // n-2 is the last chrominance sample available
2272         // this is not perfect, but noone shuld notice the difference, the more correct variant
2273         // would be like the vertical one, but that would require some special code for the
2274         // first and last pixel
2275         if(flags&SWS_FAST_BILINEAR)
2276         {
2277                 if(c->canMMX2BeUsed)
2278                 {
2279                         c->lumXInc+= 20;
2280                         c->chrXInc+= 20;
2281                 }
2282                 //we dont use the x86asm scaler if mmx is available
2283                 else if(cpuCaps.hasMMX)
2284                 {
2285                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2286                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2287                 }
2288         }
2289
2290         /* precalculate horizontal scaler filter coefficients */
2291         {
2292                 const int filterAlign= cpuCaps.hasMMX ? 4 : 1;
2293
2294                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2295                                  srcW      ,       dstW, filterAlign, 1<<14,
2296                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2297                                  srcFilter->lumH, dstFilter->lumH);
2298                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2299                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2300                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2301                                  srcFilter->chrH, dstFilter->chrH);
2302
2303 #ifdef ARCH_X86
2304 // cant downscale !!!
2305                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2306                 {
2307                         c->lumMmx2Filter   = (int16_t*)memalign(8, (dstW        /8+8)*sizeof(int16_t));
2308                         c->chrMmx2Filter   = (int16_t*)memalign(8, (c->chrDstW  /4+8)*sizeof(int16_t));
2309                         c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW      /2/8+8)*sizeof(int32_t));
2310                         c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
2311
2312                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2313                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2314                 }
2315 #endif
2316         } // Init Horizontal stuff
2317
2318
2319
2320         /* precalculate vertical scaler filter coefficients */
2321         initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2322                         srcH      ,        dstH, 1, (1<<12)-4,
2323                         (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2324                         srcFilter->lumV, dstFilter->lumV);
2325         initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2326                         c->chrSrcH, c->chrDstH, 1, (1<<12)-4,
2327                         (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2328                         srcFilter->chrV, dstFilter->chrV);
2329
2330         // Calculate Buffer Sizes so that they wont run out while handling these damn slices
2331         c->vLumBufSize= c->vLumFilterSize;
2332         c->vChrBufSize= c->vChrFilterSize;
2333         for(i=0; i<dstH; i++)
2334         {
2335                 int chrI= i*c->chrDstH / dstH;
2336                 int nextSlice= MAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
2337                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2338                 nextSlice&= ~3; // Slices start at boundaries which are divisable through 4
2339                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
2340                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
2341                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2342                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2343         }
2344
2345         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2346         c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
2347         c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
2348         //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)
2349         for(i=0; i<c->vLumBufSize; i++)
2350                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
2351         for(i=0; i<c->vChrBufSize; i++)
2352                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
2353
2354         //try to avoid drawing green stuff between the right end and the stride end
2355         for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2356         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2357
2358         ASSERT(c->chrDstH <= dstH)
2359
2360         // pack filter data for mmx code
2361         if(cpuCaps.hasMMX)
2362         {
2363                 c->lumMmxFilter= (int16_t*)memalign(8, c->vLumFilterSize*      dstH*4*sizeof(int16_t));
2364                 c->chrMmxFilter= (int16_t*)memalign(8, c->vChrFilterSize*c->chrDstH*4*sizeof(int16_t));
2365                 for(i=0; i<c->vLumFilterSize*dstH; i++)
2366                         c->lumMmxFilter[4*i]=c->lumMmxFilter[4*i+1]=c->lumMmxFilter[4*i+2]=c->lumMmxFilter[4*i+3]=
2367                                 c->vLumFilter[i];
2368                 for(i=0; i<c->vChrFilterSize*c->chrDstH; i++)
2369                         c->chrMmxFilter[4*i]=c->chrMmxFilter[4*i+1]=c->chrMmxFilter[4*i+2]=c->chrMmxFilter[4*i+3]=
2370                                 c->vChrFilter[i];
2371         }
2372
2373         if(flags&SWS_PRINT_INFO)
2374         {
2375 #ifdef DITHER1XBPP
2376                 char *dither= " dithered";
2377 #else
2378                 char *dither= "";
2379 #endif
2380                 if(flags&SWS_FAST_BILINEAR)
2381                         MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
2382                 else if(flags&SWS_BILINEAR)
2383                         MSG_INFO("\nSwScaler: BILINEAR scaler, ");
2384                 else if(flags&SWS_BICUBIC)
2385                         MSG_INFO("\nSwScaler: BICUBIC scaler, ");
2386                 else if(flags&SWS_X)
2387                         MSG_INFO("\nSwScaler: Experimental scaler, ");
2388                 else if(flags&SWS_POINT)
2389                         MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
2390                 else if(flags&SWS_AREA)
2391                         MSG_INFO("\nSwScaler: Area Averageing scaler, ");
2392                 else if(flags&SWS_BICUBLIN)
2393                         MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR, ");
2394                 else
2395                         MSG_INFO("\nSwScaler: ehh flags invalid?! ");
2396
2397                 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2398                         MSG_INFO("from %s to%s %s ",
2399                                 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
2400                 else
2401                         MSG_INFO("from %s to %s ",
2402                                 vo_format_name(srcFormat), vo_format_name(dstFormat));
2403
2404                 if(cpuCaps.hasMMX2)
2405                         MSG_INFO("using MMX2\n");
2406                 else if(cpuCaps.has3DNow)
2407                         MSG_INFO("using 3DNOW\n");
2408                 else if(cpuCaps.hasMMX)
2409                         MSG_INFO("using MMX\n");
2410                 else
2411                         MSG_INFO("using C\n");
2412         }
2413
2414         if((flags & SWS_PRINT_INFO) && verbose)
2415         {
2416                 if(cpuCaps.hasMMX)
2417                 {
2418                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2419                                 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2420                         else
2421                         {
2422                                 if(c->hLumFilterSize==4)
2423                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2424                                 else if(c->hLumFilterSize==8)
2425                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2426                                 else
2427                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2428
2429                                 if(c->hChrFilterSize==4)
2430                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2431                                 else if(c->hChrFilterSize==8)
2432                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2433                                 else
2434                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2435                         }
2436                 }
2437                 else
2438                 {
2439 #ifdef ARCH_X86
2440                         MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2441 #else
2442                         if(flags & SWS_FAST_BILINEAR)
2443                                 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2444                         else
2445                                 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2446 #endif
2447                 }
2448                 if(isPlanarYUV(dstFormat))
2449                 {
2450                         if(c->vLumFilterSize==1)
2451                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");
2452                         else
2453                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");
2454                 }
2455                 else
2456                 {
2457                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2458                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2459                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",cpuCaps.hasMMX ? "MMX" : "C");
2460                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2461                                 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");
2462                         else
2463                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");
2464                 }
2465
2466                 if(dstFormat==IMGFMT_BGR24)
2467                         MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2468                                 cpuCaps.hasMMX2 ? "MMX2" : (cpuCaps.hasMMX ? "MMX" : "C"));
2469                 else if(dstFormat==IMGFMT_BGR32)
2470                         MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
2471                 else if(dstFormat==IMGFMT_BGR16)
2472                         MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
2473                 else if(dstFormat==IMGFMT_BGR15)
2474                         MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
2475
2476                 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2477         }
2478         if((flags & SWS_PRINT_INFO) && verbose>1)
2479         {
2480                 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2481                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2482                 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2483                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2484         }
2485
2486         c->swScale= swScale;
2487         return c;
2488 }
2489
2490 /**
2491  * returns a normalized gaussian curve used to filter stuff
2492  * quality=3 is high quality, lowwer is lowwer quality
2493  */
2494
2495 SwsVector *getGaussianVec(double variance, double quality){
2496         const int length= (int)(variance*quality + 0.5) | 1;
2497         int i;
2498         double *coeff= memalign(sizeof(double), length*sizeof(double));
2499         double middle= (length-1)*0.5;
2500         SwsVector *vec= malloc(sizeof(SwsVector));
2501
2502         vec->coeff= coeff;
2503         vec->length= length;
2504
2505         for(i=0; i<length; i++)
2506         {
2507                 double dist= i-middle;
2508                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2509         }
2510
2511         normalizeVec(vec, 1.0);
2512
2513         return vec;
2514 }
2515
2516 SwsVector *getConstVec(double c, int length){
2517         int i;
2518         double *coeff= memalign(sizeof(double), length*sizeof(double));
2519         SwsVector *vec= malloc(sizeof(SwsVector));
2520
2521         vec->coeff= coeff;
2522         vec->length= length;
2523
2524         for(i=0; i<length; i++)
2525                 coeff[i]= c;
2526
2527         return vec;
2528 }
2529
2530
2531 SwsVector *getIdentityVec(void){
2532         double *coeff= memalign(sizeof(double), sizeof(double));
2533         SwsVector *vec= malloc(sizeof(SwsVector));
2534         coeff[0]= 1.0;
2535
2536         vec->coeff= coeff;
2537         vec->length= 1;
2538
2539         return vec;
2540 }
2541
2542 void normalizeVec(SwsVector *a, double height){
2543         int i;
2544         double sum=0;
2545         double inv;
2546
2547         for(i=0; i<a->length; i++)
2548                 sum+= a->coeff[i];
2549
2550         inv= height/sum;
2551
2552         for(i=0; i<a->length; i++)
2553                 a->coeff[i]*= height;
2554 }
2555
2556 void scaleVec(SwsVector *a, double scalar){
2557         int i;
2558
2559         for(i=0; i<a->length; i++)
2560                 a->coeff[i]*= scalar;
2561 }
2562
2563 static SwsVector *getConvVec(SwsVector *a, SwsVector *b){
2564         int length= a->length + b->length - 1;
2565         double *coeff= memalign(sizeof(double), length*sizeof(double));
2566         int i, j;
2567         SwsVector *vec= malloc(sizeof(SwsVector));
2568
2569         vec->coeff= coeff;
2570         vec->length= length;
2571
2572         for(i=0; i<length; i++) coeff[i]= 0.0;
2573
2574         for(i=0; i<a->length; i++)
2575         {
2576                 for(j=0; j<b->length; j++)
2577                 {
2578                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2579                 }
2580         }
2581
2582         return vec;
2583 }
2584
2585 static SwsVector *sumVec(SwsVector *a, SwsVector *b){
2586         int length= MAX(a->length, b->length);
2587         double *coeff= memalign(sizeof(double), length*sizeof(double));
2588         int i;
2589         SwsVector *vec= malloc(sizeof(SwsVector));
2590
2591         vec->coeff= coeff;
2592         vec->length= length;
2593
2594         for(i=0; i<length; i++) coeff[i]= 0.0;
2595
2596         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2597         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2598
2599         return vec;
2600 }
2601
2602 static SwsVector *diffVec(SwsVector *a, SwsVector *b){
2603         int length= MAX(a->length, b->length);
2604         double *coeff= memalign(sizeof(double), length*sizeof(double));
2605         int i;
2606         SwsVector *vec= malloc(sizeof(SwsVector));
2607
2608         vec->coeff= coeff;
2609         vec->length= length;
2610
2611         for(i=0; i<length; i++) coeff[i]= 0.0;
2612
2613         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2614         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2615
2616         return vec;
2617 }
2618
2619 /* shift left / or right if "shift" is negative */
2620 static SwsVector *getShiftedVec(SwsVector *a, int shift){
2621         int length= a->length + ABS(shift)*2;
2622         double *coeff= memalign(sizeof(double), length*sizeof(double));
2623         int i;
2624         SwsVector *vec= malloc(sizeof(SwsVector));
2625
2626         vec->coeff= coeff;
2627         vec->length= length;
2628
2629         for(i=0; i<length; i++) coeff[i]= 0.0;
2630
2631         for(i=0; i<a->length; i++)
2632         {
2633                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2634         }
2635
2636         return vec;
2637 }
2638
2639 void shiftVec(SwsVector *a, int shift){
2640         SwsVector *shifted= getShiftedVec(a, shift);
2641         free(a->coeff);
2642         a->coeff= shifted->coeff;
2643         a->length= shifted->length;
2644         free(shifted);
2645 }
2646
2647 void addVec(SwsVector *a, SwsVector *b){
2648         SwsVector *sum= sumVec(a, b);
2649         free(a->coeff);
2650         a->coeff= sum->coeff;
2651         a->length= sum->length;
2652         free(sum);
2653 }
2654
2655 void subVec(SwsVector *a, SwsVector *b){
2656         SwsVector *diff= diffVec(a, b);
2657         free(a->coeff);
2658         a->coeff= diff->coeff;
2659         a->length= diff->length;
2660         free(diff);
2661 }
2662
2663 void convVec(SwsVector *a, SwsVector *b){
2664         SwsVector *conv= getConvVec(a, b);
2665         free(a->coeff);
2666         a->coeff= conv->coeff;
2667         a->length= conv->length;
2668         free(conv);
2669 }
2670
2671 SwsVector *cloneVec(SwsVector *a){
2672         double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2673         int i;
2674         SwsVector *vec= malloc(sizeof(SwsVector));
2675
2676         vec->coeff= coeff;
2677         vec->length= a->length;
2678
2679         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2680
2681         return vec;
2682 }
2683
2684 void printVec(SwsVector *a){
2685         int i;
2686         double max=0;
2687         double min=0;
2688         double range;
2689
2690         for(i=0; i<a->length; i++)
2691                 if(a->coeff[i]>max) max= a->coeff[i];
2692
2693         for(i=0; i<a->length; i++)
2694                 if(a->coeff[i]<min) min= a->coeff[i];
2695
2696         range= max - min;
2697
2698         for(i=0; i<a->length; i++)
2699         {
2700                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2701                 MSG_DBG2("%1.3f ", a->coeff[i]);
2702                 for(;x>0; x--) MSG_DBG2(" ");
2703                 MSG_DBG2("|\n");
2704         }
2705 }
2706
2707 void freeVec(SwsVector *a){
2708         if(!a) return;
2709         if(a->coeff) free(a->coeff);
2710         a->coeff=NULL;
2711         a->length=0;
2712         free(a);
2713 }
2714
2715 void freeSwsContext(SwsContext *c){
2716         int i;
2717         if(!c) return;
2718
2719         if(c->lumPixBuf)
2720         {
2721                 for(i=0; i<c->vLumBufSize; i++)
2722                 {
2723                         if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2724                         c->lumPixBuf[i]=NULL;
2725                 }
2726                 free(c->lumPixBuf);
2727                 c->lumPixBuf=NULL;
2728         }
2729
2730         if(c->chrPixBuf)
2731         {
2732                 for(i=0; i<c->vChrBufSize; i++)
2733                 {
2734                         if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2735                         c->chrPixBuf[i]=NULL;
2736                 }
2737                 free(c->chrPixBuf);
2738                 c->chrPixBuf=NULL;
2739         }
2740
2741         if(c->vLumFilter) free(c->vLumFilter);
2742         c->vLumFilter = NULL;
2743         if(c->vChrFilter) free(c->vChrFilter);
2744         c->vChrFilter = NULL;
2745         if(c->hLumFilter) free(c->hLumFilter);
2746         c->hLumFilter = NULL;
2747         if(c->hChrFilter) free(c->hChrFilter);
2748         c->hChrFilter = NULL;
2749
2750         if(c->vLumFilterPos) free(c->vLumFilterPos);
2751         c->vLumFilterPos = NULL;
2752         if(c->vChrFilterPos) free(c->vChrFilterPos);
2753         c->vChrFilterPos = NULL;
2754         if(c->hLumFilterPos) free(c->hLumFilterPos);
2755         c->hLumFilterPos = NULL;
2756         if(c->hChrFilterPos) free(c->hChrFilterPos);
2757         c->hChrFilterPos = NULL;
2758
2759         if(c->lumMmxFilter) free(c->lumMmxFilter);
2760         c->lumMmxFilter = NULL;
2761         if(c->chrMmxFilter) free(c->chrMmxFilter);
2762         c->chrMmxFilter = NULL;
2763
2764         if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2765         c->lumMmx2Filter=NULL;
2766         if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2767         c->chrMmx2Filter=NULL;
2768         if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2769         c->lumMmx2FilterPos=NULL;
2770         if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2771         c->chrMmx2FilterPos=NULL;
2772         if(c->yuvTable) free(c->yuvTable);
2773         c->yuvTable=NULL;
2774
2775         free(c);
2776 }
2777
2778