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

   1
   2 // Software scaling and colorspace conversion routines for MPlayer
   3
   4 #include <inttypes.h>
   5 #include "../config.h"
   6
   7 #undef HAVE_MMX2 //code is buggy
   8 //#undef HAVE_MMX
   9
  10 #define RET 0xC3 //near return opcode
  11
  12 // temporary storage for 4 yuv lines:
  13 // 16bit for now (mmx likes it more compact)
  14 static uint16_t pix_buf_y[4][2048];
  15 static uint16_t pix_buf_uv[2][2048*2];
  16
  17 // clipping helper table for C implementations:
  18 static unsigned char clip_table[768];
  19
  20 // yuv->rgb conversion tables:
  21 static    int yuvtab_2568[256];
  22 static    int yuvtab_3343[256];
  23 static    int yuvtab_0c92[256];
  24 static    int yuvtab_1a1e[256];
  25 static    int yuvtab_40cf[256];
  26
  27 static uint64_t yCoeff=    0x2568256825682568LL;
  28 static uint64_t ubCoeff=   0x3343334333433343LL;
  29 static uint64_t vrCoeff=   0x40cf40cf40cf40cfLL;
  30 static uint64_t ugCoeff=   0xE5E2E5E2E5E2E5E2LL;
  31 static uint64_t vgCoeff=   0xF36EF36EF36EF36ELL;
  32 static uint64_t w80=       0x0080008000800080LL;
  33 static uint64_t w10=       0x0010001000100010LL;
  34
  35 static uint64_t b16Dither= 0x0004000400040004LL;
  36 static uint64_t b16Dither1=0x0004000400040004LL;
  37 static uint64_t b16Dither2=0x0602060206020602LL;
  38 static uint64_t g16Dither= 0x0002000200020002LL;
  39 static uint64_t g16Dither1=0x0002000200020002LL;
  40 static uint64_t g16Dither2=0x0301030103010301LL;
  41
  42 static uint64_t b16Mask=   0x001F001F001F001FLL;
  43 static uint64_t g16Mask=   0x07E007E007E007E0LL;
  44 static uint64_t r16Mask=   0xF800F800F800F800LL;
  45 static uint64_t temp0;
  46
  47 static uint8_t funnyYCode[10000];
  48 static uint8_t funnyUVCode[10000];
  49
  50
  51
  52 // *** bilinear scaling and yuv->rgb conversion of yv12 slices:
  53 // *** Note: it's called multiple times while decoding a frame, first time y==0
  54 // *** Designed to upscale, but may work for downscale too.
  55 // s_xinc = (src_width << 8) / dst_width
  56 // s_yinc = (src_height << 16) / dst_height
  57 void SwScale_YV12slice_brg24(unsigned char* srcptr[],int stride[], int y, int h,
  58                              unsigned char* dstptr, int dststride, int dstw, int dstbpp,
  59                              unsigned int s_xinc,unsigned int s_yinc){
  60
  61 // scaling factors:
  62 //static int s_yinc=(vo_dga_src_height<<16)/vo_dga_vp_height;
  63 //static int s_xinc=(vo_dga_src_width<<8)/vo_dga_vp_width;
  64
  65 unsigned int s_xinc2=s_xinc>>1;
  66
  67 static int s_srcypos;
  68 static int s_ypos;
  69 static int s_last_ypos;
  70 static int static_dstw;
  71
  72 #ifdef HAVE_MMX2
  73 static int old_dstw= -1;
  74 static int old_s_xinc= -1;
  75 #endif
  76
  77 s_xinc&= -2; //clear last bit or uv and y might be shifted relative to each other
  78
  79   if(y==0){
  80       s_srcypos=-2*s_yinc;
  81       s_ypos=-2;
  82       s_last_ypos=-2;
  83 #ifdef HAVE_MMX2
  84 // cant downscale !!!
  85         if(old_s_xinc != s_xinc || old_dstw!=dstw)
  86         {
  87                 uint8_t *fragment;
  88                 int imm8OfPShufW1;
  89                 int imm8OfPShufW2;
  90                 int fragmentLength;
  91
  92                 int xpos, xx, xalpha, i;
  93
  94                 old_s_xinc= s_xinc;
  95                 old_dstw= dstw;
  96
  97                 static_dstw= dstw;
  98
  99                 // create an optimized horizontal scaling routine
 100
 101                 //code fragment
 102
 103 //              fragmentLength=0;
 104 //              printf("%d, %d\n", fragmentLength,imm8OfPShufW1);
 105
 106                 asm volatile(
 107                         "jmp 9f                         \n\t"
 108                 // Begin
 109                         "0:                             \n\t"
 110                         "movq (%%esi, %%ebx), %%mm0     \n\t" //FIXME Alignment
 111                         "movq %%mm0, %%mm1              \n\t"
 112                         "psrlq $8, %%mm0                \n\t"
 113                         "punpcklbw %%mm7, %%mm1 \n\t"
 114                         "punpcklbw %%mm7, %%mm0 \n\t"
 115                         "pshufw $0xFF, %%mm1, %%mm1     \n\t"
 116                         "1:                             \n\t"
 117                         "pshufw $0xFF, %%mm0, %%mm0     \n\t"
 118                         "2:                             \n\t"
 119                         "psubw %%mm1, %%mm0             \n\t"
 120                         "psraw $1, %%mm0                \n\t"
 121                         "pmullw %%mm2, %%mm0            \n\t"
 122                         "psllw $7, %%mm1                \n\t"
 123                         "paddw %%mm1, %%mm0             \n\t"
 124                         "movq %%mm0, (%%edi, %%eax)     \n\t"
 125                         "paddb %%mm6, %%mm2             \n\t" // 2*alpha += xpos&0xFF
 126
 127                         "addb %%ch, %%cl                \n\t" //2*xalpha += (4*s_xinc)&0xFF
 128                         "adcl %%edx, %%ebx              \n\t" //xx+= (4*s_xinc)>>8 + carry
 129
 130                         "addl $8, %%eax                 \n\t"
 131                 // End
 132                         "9:                             \n\t"
 133 //              "int $3\n\t"
 134                         "leal 0b, %0                    \n\t"
 135                         "leal 1b, %1                    \n\t"
 136                         "leal 2b, %2                    \n\t"
 137                         "decl %1                        \n\t"
 138                         "decl %2                        \n\t"
 139                         "subl %0, %1                    \n\t"
 140                         "subl %0, %2                    \n\t"
 141                         "leal 9b, %3                    \n\t"
 142                         "subl %0, %3                    \n\t"
 143                         :"=r" (fragment), "=r" (imm8OfPShufW1), "=r" (imm8OfPShufW2),
 144                          "=r" (fragmentLength)
 145                 );
 146
 147                 xpos= xx=xalpha= 0;
 148                 //FIXME choose size and or xinc so that they fit exactly
 149                 for(i=0; i<dstw/8; i++)
 150                 {
 151                         int xx=xpos>>8;
 152
 153                         if((i&3) == 0)
 154                         {
 155                                 int a=0;
 156                                 int b=((xpos+s_xinc)>>8) - xx;
 157                                 int c=((xpos+s_xinc*2)>>8) - xx;
 158                                 int d=((xpos+s_xinc*3)>>8) - xx;
 159
 160                                 memcpy(funnyYCode + fragmentLength*i/4, fragment, fragmentLength);
 161
 162                                 funnyYCode[fragmentLength*i/4 + imm8OfPShufW1]=
 163                                 funnyYCode[fragmentLength*i/4 + imm8OfPShufW2]=
 164                                         a | (b<<2) | (c<<4) | (d<<6);
 165
 166                                 funnyYCode[fragmentLength*(i+4)/4]= RET;
 167                         }
 168                         xpos+=s_xinc;
 169                 }
 170
 171                 xpos= xx=xalpha= 0;
 172                 //FIXME choose size and or xinc so that they fit exactly
 173                 for(i=0; i<dstw/8; i++)
 174                 {
 175                         int xx=xpos>>8;
 176
 177                         if((i&3) == 0)
 178                         {
 179                                 int a=0;
 180                                 int b=((xpos+s_xinc2)>>8) - xx;
 181                                 int c=((xpos+s_xinc2*2)>>8) - xx;
 182                                 int d=((xpos+s_xinc2*3)>>8) - xx;
 183
 184                                 memcpy(funnyUVCode + fragmentLength*i/4, fragment, fragmentLength);
 185
 186                                 funnyUVCode[fragmentLength*i/4 + imm8OfPShufW1]=
 187                                 funnyUVCode[fragmentLength*i/4 + imm8OfPShufW2]=
 188                                         a | (b<<2) | (c<<4) | (d<<6);
 189
 190                                 funnyUVCode[fragmentLength*(i+4)/4]= RET;
 191                         }
 192                         xpos+=s_xinc2;
 193                 }
 194 //              funnyCode[0]= RET;
 195
 196
 197         }
 198 #endif
 199   } // reset counters
 200
 201   while(1){
 202     unsigned char *dest=dstptr+dststride*s_ypos;
 203     int y0=2+(s_srcypos>>16);
 204     int y1=1+(s_srcypos>>17);
 205     int yalpha=(s_srcypos&0xFFFF)>>7;
 206     int yalpha1=yalpha^511;
 207     int uvalpha=((s_srcypos>>1)&0xFFFF)>>7;
 208     int uvalpha1=uvalpha^511;
 209     uint16_t *buf0=pix_buf_y[y0&3];
 210     uint16_t *buf1=pix_buf_y[((y0+1)&3)];
 211     uint16_t *uvbuf0=pix_buf_uv[y1&1];
 212     uint16_t *uvbuf1=pix_buf_uv[(y1&1)^1];
 213     int i;
 214
 215     if(y0>=y+h) break;
 216
 217     s_ypos++; s_srcypos+=s_yinc;
 218
 219     if(s_last_ypos!=y0){
 220       unsigned char *src=srcptr[0]+(y0-y)*stride[0];
 221       unsigned int xpos=0;
 222       s_last_ypos=y0;
 223       // *** horizontal scale Y line to temp buffer
 224       // this loop should be rewritten in MMX assembly!!!!
 225 #ifdef HAVE_MMX2
 226         asm volatile(
 227                 "pxor %%mm7, %%mm7              \n\t"
 228                 "pxor %%mm2, %%mm2              \n\t" // 2*xalpha
 229                 "movd %5, %%mm6                 \n\t" // s_xinc&0xFF
 230                 "punpcklwd %%mm6, %%mm6         \n\t"
 231                 "punpcklwd %%mm6, %%mm6         \n\t"
 232                 "movq %%mm6, %%mm2              \n\t"
 233                 "psllq $16, %%mm2               \n\t"
 234                 "paddb %%mm6, %%mm2             \n\t"
 235                 "psllq $16, %%mm2               \n\t"
 236                 "paddb %%mm6, %%mm2             \n\t"
 237                 "psllq $16, %%mm2               \n\t" //0,t,2t,3t               t=s_xinc&0xFF
 238                 "movq %%mm2, temp0              \n\t"
 239                 "movd %4, %%mm6                 \n\t" //(s_xinc*4)&0xFF
 240                 "punpcklwd %%mm6, %%mm6         \n\t"
 241                 "punpcklwd %%mm6, %%mm6         \n\t"
 242                 "xorl %%eax, %%eax              \n\t" // i
 243                 "xorl %%ebx, %%ebx              \n\t" // xx
 244                 "movl %0, %%esi                 \n\t" // src
 245                 "movl %1, %%edi                 \n\t" // buf1
 246                 "movl %3, %%edx                 \n\t" // (s_xinc*4)>>8
 247                 "xorl %%ecx, %%ecx              \n\t"
 248                 "movb %4, %%ch                  \n\t" // (s_xinc*4)&0xFF
 249 //      "int $3\n\t"
 250                 "call funnyYCode                        \n\t"
 251                 "movq temp0, %%mm2              \n\t"
 252                 "xorb %%cl, %%cl                \n\t"
 253                 "call funnyYCode                        \n\t"
 254                 "movq temp0, %%mm2              \n\t"
 255                 "xorb %%cl, %%cl                \n\t"
 256                 "call funnyYCode                        \n\t"
 257                 "movq temp0, %%mm2              \n\t"
 258                 "xorb %%cl, %%cl                \n\t"
 259                 "call funnyYCode                        \n\t"
 260                 "movq temp0, %%mm2              \n\t"
 261                 "xorb %%cl, %%cl                \n\t"
 262                 "call funnyYCode                        \n\t"
 263                 "movq temp0, %%mm2              \n\t"
 264                 "xorb %%cl, %%cl                \n\t"
 265                 "call funnyYCode                        \n\t"
 266                 "movq temp0, %%mm2              \n\t"
 267                 "xorb %%cl, %%cl                \n\t"
 268                 "call funnyYCode                        \n\t"
 269                 "movq temp0, %%mm2              \n\t"
 270                 "xorb %%cl, %%cl                \n\t"
 271                 "call funnyYCode                        \n\t"
 272                 :: "m" (src), "m" (buf1), "m" (dstw), "m" ((s_xinc*4)>>8),
 273                   "m" ((s_xinc*4)&0xFF), "m" (s_xinc&0xFF)
 274                 : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
 275         );
 276
 277 #elif defined (ARCH_X86)
 278         //NO MMX just normal asm ... FIXME try/write funny MMX2 variant
 279         //FIXME add prefetch
 280         asm volatile(
 281                 "xorl %%eax, %%eax              \n\t" // i
 282                 "xorl %%ebx, %%ebx              \n\t" // xx
 283                 "xorl %%ecx, %%ecx              \n\t" // 2*xalpha
 284                 "1:                             \n\t"
 285                 "movzbl  (%0, %%ebx), %%edi     \n\t" //src[xx]
 286                 "movzbl 1(%0, %%ebx), %%esi     \n\t" //src[xx+1]
 287                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
 288                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
 289                 "shll $8, %%edi                 \n\t"
 290                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
 291                 "movl %1, %%edi                 \n\t"
 292                 "shrl $1, %%esi                 \n\t"
 293                 "movw %%si, (%%edi, %%eax, 2)   \n\t"
 294                 "addb %4, %%cl                  \n\t" //2*xalpha += s_xinc&0xFF
 295                 "adcl %3, %%ebx                 \n\t" //xx+= s_xinc>>8 + carry
 296
 297                 "movzbl (%0, %%ebx), %%edi      \n\t" //src[xx]
 298                 "movzbl 1(%0, %%ebx), %%esi     \n\t" //src[xx+1]
 299                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
 300                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
 301                 "shll $8, %%edi                 \n\t"
 302                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
 303                 "movl %1, %%edi                 \n\t"
 304                 "shrl $1, %%esi                 \n\t"
 305                 "movw %%si, 2(%%edi, %%eax, 2)  \n\t"
 306                 "addb %4, %%cl                  \n\t" //2*xalpha += s_xinc&0xFF
 307                 "adcl %3, %%ebx                 \n\t" //xx+= s_xinc>>8 + carry
 308
 309
 310                 "addl $2, %%eax                 \n\t"
 311                 "cmpl %2, %%eax                 \n\t"
 312                 " jb 1b                         \n\t"
 313
 314
 315                 :: "r" (src), "m" (buf1), "m" (dstw), "m" (s_xinc>>8), "m" (s_xinc&0xFF)
 316                 : "%eax", "%ebx", "%ecx", "%edi", "%esi"
 317                 );
 318 #else
 319       for(i=0;i<dstw;i++){
 320         register unsigned int xx=xpos>>8;
 321         register unsigned int xalpha=(xpos&0xFF)>>1;
 322         buf1[i]=(src[xx]*(xalpha^127)+src[xx+1]*xalpha);
 323         xpos+=s_xinc;
 324       }
 325 #endif
 326       // *** horizontal scale U and V lines to temp buffer
 327       if(!(y0&1)){
 328         unsigned char *src1=srcptr[1]+(y1-y/2)*stride[1];
 329         unsigned char *src2=srcptr[2]+(y1-y/2)*stride[2];
 330         xpos=0;
 331         // this loop should be rewritten in MMX assembly!!!!
 332 #ifdef HAVE_MMX2
 333         asm volatile(
 334                 "pxor %%mm7, %%mm7              \n\t"
 335                 "pxor %%mm2, %%mm2              \n\t" // 2*xalpha
 336                 "movd %5, %%mm6                 \n\t" // s_xinc&0xFF
 337                 "punpcklwd %%mm6, %%mm6         \n\t"
 338                 "punpcklwd %%mm6, %%mm6         \n\t"
 339                 "movq %%mm6, %%mm2              \n\t"
 340                 "psllq $16, %%mm2               \n\t"
 341                 "paddb %%mm6, %%mm2             \n\t"
 342                 "psllq $16, %%mm2               \n\t"
 343                 "paddb %%mm6, %%mm2             \n\t"
 344                 "psllq $16, %%mm2               \n\t" //0,t,2t,3t               t=s_xinc&0xFF
 345                 "movq %%mm2, temp0              \n\t"
 346                 "movd %4, %%mm6                 \n\t" //(s_xinc*4)&0xFF
 347                 "punpcklwd %%mm6, %%mm6         \n\t"
 348                 "punpcklwd %%mm6, %%mm6         \n\t"
 349                 "xorl %%eax, %%eax              \n\t" // i
 350                 "xorl %%ebx, %%ebx              \n\t" // xx
 351                 "movl %0, %%esi                 \n\t" // src
 352                 "movl %1, %%edi                 \n\t" // buf1
 353                 "movl %3, %%edx                 \n\t" // (s_xinc*4)>>8
 354                 "xorl %%ecx, %%ecx              \n\t"
 355                 "movb %4, %%ch                  \n\t" // (s_xinc*4)&0xFF
 356 //      "int $3\n\t"
 357                 "call funnyUVCode                       \n\t"
 358                 "movq temp0, %%mm2              \n\t"
 359                 "xorb %%cl, %%cl                \n\t"
 360                 "call funnyUVCode                       \n\t"
 361                 "movq temp0, %%mm2              \n\t"
 362                 "xorb %%cl, %%cl                \n\t"
 363                 "call funnyUVCode                       \n\t"
 364                 "movq temp0, %%mm2              \n\t"
 365                 "xorb %%cl, %%cl                \n\t"
 366                 "call funnyUVCode                       \n\t"
 367                 "movq temp0, %%mm2              \n\t"
 368                 "xorb %%cl, %%cl                \n\t"
 369                 "call funnyUVCode                       \n\t"
 370                 "movq temp0, %%mm2              \n\t"
 371                 "xorb %%cl, %%cl                \n\t"
 372                 "call funnyUVCode                       \n\t"
 373                 "movq temp0, %%mm2              \n\t"
 374                 "xorb %%cl, %%cl                \n\t"
 375                 "call funnyUVCode                       \n\t"
 376                 "movq temp0, %%mm2              \n\t"
 377                 "xorb %%cl, %%cl                \n\t"
 378                 "call funnyUVCode                       \n\t"
 379
 380                 "xorl %%eax, %%eax              \n\t" // i
 381                 "xorl %%ebx, %%ebx              \n\t" // xx
 382                 "movl %6, %%esi                 \n\t" // src
 383                 "movl %1, %%edi                 \n\t" // buf1
 384                 "addl $4096, %%edi              \n\t"
 385
 386                 "call funnyUVCode                       \n\t"
 387                 "movq temp0, %%mm2              \n\t"
 388                 "xorb %%cl, %%cl                \n\t"
 389                 "call funnyUVCode                       \n\t"
 390                 "movq temp0, %%mm2              \n\t"
 391                 "xorb %%cl, %%cl                \n\t"
 392                 "call funnyUVCode                       \n\t"
 393                 "movq temp0, %%mm2              \n\t"
 394                 "xorb %%cl, %%cl                \n\t"
 395                 "call funnyUVCode                       \n\t"
 396                 "movq temp0, %%mm2              \n\t"
 397                 "xorb %%cl, %%cl                \n\t"
 398                 "call funnyUVCode                       \n\t"
 399                 "movq temp0, %%mm2              \n\t"
 400                 "xorb %%cl, %%cl                \n\t"
 401                 "call funnyUVCode                       \n\t"
 402                 "movq temp0, %%mm2              \n\t"
 403                 "xorb %%cl, %%cl                \n\t"
 404                 "call funnyUVCode                       \n\t"
 405                 "movq temp0, %%mm2              \n\t"
 406                 "xorb %%cl, %%cl                \n\t"
 407                 "call funnyUVCode                       \n\t"
 408
 409                 :: "m" (src1), "m" (uvbuf1), "m" (dstw), "m" ((s_xinc2*4)>>8),
 410                   "m" ((s_xinc2*4)&0xFF), "m" (s_xinc2&0xFF), "m" (src2)
 411                 : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
 412         );
 413
 414 #elif defined (ARCH_X86)
 415         //NO MMX just normal asm ... FIXME try/write funny MMX2 variant
 416         asm volatile(
 417                 "xorl %%eax, %%eax              \n\t" // i
 418                 "xorl %%ebx, %%ebx              \n\t" // xx
 419                 "xorl %%ecx, %%ecx              \n\t" // 2*xalpha
 420                 "1:                             \n\t"
 421                 "movl %0, %%esi                 \n\t"
 422                 "movzbl  (%%esi, %%ebx), %%edi  \n\t" //src[xx]
 423                 "movzbl 1(%%esi, %%ebx), %%esi  \n\t" //src[xx+1]
 424                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
 425                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
 426                 "shll $8, %%edi                 \n\t"
 427                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
 428                 "movl %1, %%edi                 \n\t"
 429                 "shrl $1, %%esi                 \n\t"
 430                 "movw %%si, (%%edi, %%eax, 2)   \n\t"
 431
 432                 "movzbl  (%5, %%ebx), %%edi     \n\t" //src[xx]
 433                 "movzbl 1(%5, %%ebx), %%esi     \n\t" //src[xx+1]
 434                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
 435                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
 436                 "shll $8, %%edi                 \n\t"
 437                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
 438                 "movl %1, %%edi                 \n\t"
 439                 "shrl $1, %%esi                 \n\t"
 440                 "movw %%si, 4096(%%edi, %%eax, 2)\n\t"
 441
 442                 "addb %4, %%cl                  \n\t" //2*xalpha += s_xinc&0xFF
 443                 "adcl %3, %%ebx                 \n\t" //xx+= s_xinc>>8 + carry
 444                 "addl $1, %%eax                 \n\t"
 445                 "cmpl %2, %%eax                 \n\t"
 446                 " jb 1b                         \n\t"
 447
 448
 449                 :: "m" (src1), "m" (uvbuf1), "m" (dstw), "m" (s_xinc2>>8), "m" (s_xinc2&0xFF),
 450                 "r" (src2)
 451                 : "%eax", "%ebx", "%ecx", "%edi", "%esi"
 452                 );
 453 #else
 454         for(i=0;i<dstw;i++){
 455           register unsigned int xx=xpos>>8;
 456           register unsigned int xalpha=(xpos&0xFF)>>1;
 457           uvbuf1[i]=(src1[xx]*(xalpha^127)+src1[xx+1]*xalpha);
 458           uvbuf1[i+2048]=(src2[xx]*(xalpha^127)+src2[xx+1]*xalpha);
 459           xpos+=s_xinc2;
 460         }
 461 #endif
 462       }
 463       if(!y0) continue;
 464     }
 465
 466     // Note1: this code can be resticted to n*8 (or n*16) width lines to simplify optimization...
 467     // Re: Note1: ok n*4 for now
 468     // Note2: instead of using lookup tabs, mmx version could do the multiply...
 469     // Re: Note2: yep
 470     // Note3: maybe we should make separated 15/16, 24 and 32bpp version of this:
 471     // Re: done (32 & 16) and 16 has dithering :) but 16 is untested
 472 #ifdef HAVE_MMX
 473         //FIXME write lq version with less uv ...
 474         //FIXME reorder / optimize
 475         if(dstbpp == 4)
 476         {
 477                 asm volatile(
 478
 479 #define YSCALEYUV2RGB \
 480                 "pxor %%mm7, %%mm7              \n\t"\
 481                 "movd %6, %%mm6                 \n\t" /*yalpha1*/\
 482                 "punpcklwd %%mm6, %%mm6         \n\t"\
 483                 "punpcklwd %%mm6, %%mm6         \n\t"\
 484                 "movd %7, %%mm5                 \n\t" /*uvalpha1*/\
 485                 "punpcklwd %%mm5, %%mm5         \n\t"\
 486                 "punpcklwd %%mm5, %%mm5         \n\t"\
 487                 "xorl %%eax, %%eax              \n\t"\
 488                 "1:                             \n\t"\
 489                 "movq (%0, %%eax, 2), %%mm0     \n\t" /*buf0[eax]*/\
 490                 "movq (%1, %%eax, 2), %%mm1     \n\t" /*buf1[eax]*/\
 491                 "psubw %%mm1, %%mm0             \n\t" /* buf0[eax] - buf1[eax]*/\
 492                 "pmulhw %%mm6, %%mm0            \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
 493                 "psraw $7, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>7*/\
 494                 "paddw %%mm0, %%mm1             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
 495                 "psubw w10, %%mm1               \n\t" /* Y-16*/\
 496                 "psllw $3, %%mm1                \n\t" /* (y-16)*8*/\
 497                 "pmulhw yCoeff, %%mm1           \n\t"\
 498 \
 499                 "movq (%2, %%eax,2), %%mm2      \n\t" /* uvbuf0[eax]*/\
 500                 "movq (%3, %%eax,2), %%mm3      \n\t" /* uvbuf1[eax]*/\
 501                 "psubw %%mm3, %%mm2             \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
 502                 "pmulhw %%mm5, %%mm2            \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
 503                 "psraw $7, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>7*/\
 504                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
 505                 "psubw w80, %%mm3               \n\t" /* (U-128)*/\
 506                 "psllw $3, %%mm3                \n\t" /*(U-128)8*/\
 507 \
 508                 "movq 4096(%2, %%eax,2), %%mm4  \n\t" /* uvbuf0[eax+2048]*/\
 509                 "movq 4096(%3, %%eax,2), %%mm0  \n\t" /* uvbuf1[eax+2048]*/\
 510                 "psubw %%mm0, %%mm4             \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
 511                 "pmulhw %%mm5, %%mm4            \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
 512                 "psraw $7, %%mm0                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>7*/\
 513                 "paddw %%mm4, %%mm0             \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
 514                 "psubw w80, %%mm0               \n\t" /* (V-128)*/\
 515                 "psllw $3, %%mm0                \n\t" /* (V-128)8*/\
 516 \
 517                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
 518                 "pmulhw ubCoeff, %%mm3          \n\t"\
 519                 "paddw %%mm1, %%mm3             \n\t" /* B*/\
 520 \
 521                 "movq %%mm0, %%mm4              \n\t" /* (V-128)8*/\
 522                 "pmulhw vrCoeff, %%mm0          \n\t"\
 523                 "paddw %%mm1, %%mm0             \n\t" /* R*/\
 524 \
 525                 "pmulhw ugCoeff, %%mm2          \n\t"\
 526                 "pmulhw vgCoeff, %%mm4          \n\t"\
 527                 "paddw %%mm4, %%mm2             \n\t"\
 528                 "paddw %%mm2, %%mm1             \n\t" /* G*/\
 529 \
 530                 "packuswb %%mm3, %%mm3          \n\t"\
 531                 "packuswb %%mm0, %%mm0          \n\t"\
 532                 "packuswb %%mm1, %%mm1          \n\t"
 533
 534 YSCALEYUV2RGB
 535                 "punpcklbw %%mm1, %%mm3         \n\t" // BGBGBGBG
 536                 "punpcklbw %%mm7, %%mm0         \n\t" // R0R0R0R0
 537
 538                 "movq %%mm3, %%mm1              \n\t"
 539                 "punpcklwd %%mm0, %%mm3         \n\t" // BGR0BGR0
 540                 "punpckhwd %%mm0, %%mm1         \n\t" // BGR0BGR0
 541 #ifdef HAVE_MMX2
 542                 "movntq %%mm3, (%4, %%eax, 4)   \n\t"
 543                 "movntq %%mm1, 8(%4, %%eax, 4)  \n\t"
 544 #else
 545                 "movq %%mm3, (%4, %%eax, 4)     \n\t"
 546                 "movq %%mm1, 8(%4, %%eax, 4)    \n\t"
 547 #endif
 548                 "addl $4, %%eax                 \n\t"
 549                 "cmpl %5, %%eax                 \n\t"
 550                 " jb 1b                         \n\t"
 551
 552
 553                 :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 554                 "m" (yalpha1), "m" (uvalpha1)
 555                 : "%eax"
 556                 );
 557         }
 558         else if(dstbpp==2)
 559         {
 560                 asm volatile(
 561
 562 YSCALEYUV2RGB
 563                 "paddusb g16Dither, %%mm1       \n\t"
 564                 "paddusb b16Dither, %%mm0       \n\t"
 565                 "paddusb b16Dither, %%mm3       \n\t"
 566                 "punpcklbw %%mm7, %%mm1         \n\t" // 0G0G0G0G
 567                 "punpcklbw %%mm7, %%mm3         \n\t" // 0B0B0B0B
 568                 "punpcklbw %%mm7, %%mm0         \n\t" // 0R0R0R0R
 569
 570                 "psrlw $3, %%mm3                \n\t"
 571                 "psllw $3, %%mm1                \n\t"
 572                 "psllw $8, %%mm0                \n\t"
 573                 "pand g16Mask, %%mm1            \n\t"
 574                 "pand r16Mask, %%mm0            \n\t"
 575
 576                 "por %%mm3, %%mm1               \n\t"
 577                 "por %%mm1, %%mm0               \n\t"
 578 #ifdef HAVE_MMX2
 579                 "movntq %%mm0, (%4, %%eax, 2)   \n\t"
 580 #else
 581                 "movq %%mm0, (%4, %%eax, 2)     \n\t"
 582 #endif
 583                 "addl $4, %%eax                 \n\t"
 584                 "cmpl %5, %%eax                 \n\t"
 585                 " jb 1b                         \n\t"
 586
 587                 :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 588                 "m" (yalpha1), "m" (uvalpha1)
 589                 : "%eax"
 590                 );
 591         }
 592 #else
 593         if(dstbpp==4 || dstbpp==3)
 594         {
 595                 for(i=0;i<dstw;i++){
 596                         // vertical linear interpolation && yuv2rgb in a single step:
 597                         int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
 598                         int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
 599                         int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
 600                         dest[0]=clip_table[((Y + yuvtab_3343[U]) >>13)];
 601                         dest[1]=clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)];
 602                         dest[2]=clip_table[((Y + yuvtab_40cf[V]) >>13)];
 603                         dest+=dstbpp;
 604                 }
 605         }
 606         else if(dstbpp==2) //16bit
 607         {
 608                 for(i=0;i<dstw;i++){
 609                         // vertical linear interpolation && yuv2rgb in a single step:
 610                         int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
 611                         int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
 612                         int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
 613
 614                         ((uint16_t*)dest)[0] =
 615                                 (clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
 616                                 (clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<3)&0x07E0 |
 617                                 (clip_table[((Y + yuvtab_40cf[V]) >>13)]<<8)&0xF800;
 618                         dest+=dstbpp;
 619                 }
 620         }
 621         else if(dstbpp==2) //15bit FIXME how do i figure out if its 15 or 16?
 622         {
 623                 for(i=0;i<dstw;i++){
 624                         // vertical linear interpolation && yuv2rgb in a single step:
 625                         int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
 626                         int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
 627                         int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
 628
 629                         ((uint16_t*)dest)[0] =
 630                                 (clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
 631                                 (clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<2)&0x03E0 |
 632                                 (clip_table[((Y + yuvtab_40cf[V]) >>13)]<<7)&0x7C00;
 633                         dest+=dstbpp;
 634                 }
 635         }
 636 #endif
 637
 638         b16Dither= b16Dither1;
 639         b16Dither1= b16Dither2;
 640         b16Dither2= b16Dither;
 641
 642         g16Dither= g16Dither1;
 643         g16Dither1= g16Dither2;
 644         g16Dither2= g16Dither;
 645   }
 646
 647 }
 648
 649
 650 void SwScale_Init(){
 651     // generating tables:
 652     int i;
 653     for(i=0;i<256;i++){
 654         clip_table[i]=0;
 655         clip_table[i+256]=i;
 656         clip_table[i+512]=255;
 657         yuvtab_2568[i]=(0x2568*(i-16))+(256<<13);
 658         yuvtab_3343[i]=0x3343*(i-128);
 659         yuvtab_0c92[i]=-0x0c92*(i-128);
 660         yuvtab_1a1e[i]=-0x1a1e*(i-128);
 661         yuvtab_40cf[i]=0x40cf*(i-128);
 662     }
 663
 664 }