git.sesse.net Git - ffmpeg/blob - postproc/swscale.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 static uint64_t bm00000111=0x0000000000FFFFFFLL;
  35 static uint64_t bm11111000=0xFFFFFFFFFF000000LL;
  36
  37 static uint64_t b16Dither= 0x0004000400040004LL;
  38 static uint64_t b16Dither1=0x0004000400040004LL;
  39 static uint64_t b16Dither2=0x0602060206020602LL;
  40 static uint64_t g16Dither= 0x0002000200020002LL;
  41 static uint64_t g16Dither1=0x0002000200020002LL;
  42 static uint64_t g16Dither2=0x0301030103010301LL;
  43
  44 static uint64_t b16Mask=   0x001F001F001F001FLL;
  45 static uint64_t g16Mask=   0x07E007E007E007E0LL;
  46 static uint64_t r16Mask=   0xF800F800F800F800LL;
  47 static uint64_t temp0;
  48
  49 static uint8_t funnyYCode[10000];
  50 static uint8_t funnyUVCode[10000];
  51
  52
  53
  54 // *** bilinear scaling and yuv->rgb conversion of yv12 slices:
  55 // *** Note: it's called multiple times while decoding a frame, first time y==0
  56 // *** Designed to upscale, but may work for downscale too.
  57 // s_xinc = (src_width << 8) / dst_width
  58 // s_yinc = (src_height << 16) / dst_height
  59 void SwScale_YV12slice_brg24(unsigned char* srcptr[],int stride[], int y, int h,
  60                              unsigned char* dstptr, int dststride, int dstw, int dstbpp,
  61                              unsigned int s_xinc,unsigned int s_yinc){
  62
  63 // scaling factors:
  64 //static int s_yinc=(vo_dga_src_height<<16)/vo_dga_vp_height;
  65 //static int s_xinc=(vo_dga_src_width<<8)/vo_dga_vp_width;
  66
  67 unsigned int s_xinc2=s_xinc>>1;
  68
  69 static int s_srcypos;
  70 static int s_ypos;
  71 static int s_last_ypos;
  72 static int static_dstw;
  73
  74 #ifdef HAVE_MMX2
  75 static int old_dstw= -1;
  76 static int old_s_xinc= -1;
  77 #endif
  78
  79 s_xinc&= -2; //clear last bit or uv and y might be shifted relative to each other
  80
  81   if(y==0){
  82       s_srcypos=-2*s_yinc;
  83       s_ypos=-2;
  84       s_last_ypos=-2;
  85 #ifdef HAVE_MMX2
  86 // cant downscale !!!
  87         if(old_s_xinc != s_xinc || old_dstw!=dstw)
  88         {
  89                 uint8_t *fragment;
  90                 int imm8OfPShufW1;
  91                 int imm8OfPShufW2;
  92                 int fragmentLength;
  93
  94                 int xpos, xx, xalpha, i;
  95
  96                 old_s_xinc= s_xinc;
  97                 old_dstw= dstw;
  98
  99                 static_dstw= dstw;
 100
 101                 // create an optimized horizontal scaling routine
 102
 103                 //code fragment
 104
 105 //              fragmentLength=0;
 106 //              printf("%d, %d\n", fragmentLength,imm8OfPShufW1);
 107
 108                 asm volatile(
 109                         "jmp 9f                         \n\t"
 110                 // Begin
 111                         "0:                             \n\t"
 112                         "movq (%%esi, %%ebx), %%mm0     \n\t" //FIXME Alignment
 113                         "movq %%mm0, %%mm1              \n\t"
 114                         "psrlq $8, %%mm0                \n\t"
 115                         "punpcklbw %%mm7, %%mm1 \n\t"
 116                         "punpcklbw %%mm7, %%mm0 \n\t"
 117                         "pshufw $0xFF, %%mm1, %%mm1     \n\t"
 118                         "1:                             \n\t"
 119                         "pshufw $0xFF, %%mm0, %%mm0     \n\t"
 120                         "2:                             \n\t"
 121                         "psubw %%mm1, %%mm0             \n\t"
 122                         "psraw $1, %%mm0                \n\t"
 123                         "pmullw %%mm2, %%mm0            \n\t"
 124                         "psllw $7, %%mm1                \n\t"
 125                         "paddw %%mm1, %%mm0             \n\t"
 126                         "movq %%mm0, (%%edi, %%eax)     \n\t"
 127                         "paddb %%mm6, %%mm2             \n\t" // 2*alpha += xpos&0xFF
 128
 129                         "addb %%ch, %%cl                \n\t" //2*xalpha += (4*s_xinc)&0xFF
 130                         "adcl %%edx, %%ebx              \n\t" //xx+= (4*s_xinc)>>8 + carry
 131
 132                         "addl $8, %%eax                 \n\t"
 133                 // End
 134                         "9:                             \n\t"
 135 //              "int $3\n\t"
 136                         "leal 0b, %0                    \n\t"
 137                         "leal 1b, %1                    \n\t"
 138                         "leal 2b, %2                    \n\t"
 139                         "decl %1                        \n\t"
 140                         "decl %2                        \n\t"
 141                         "subl %0, %1                    \n\t"
 142                         "subl %0, %2                    \n\t"
 143                         "leal 9b, %3                    \n\t"
 144                         "subl %0, %3                    \n\t"
 145                         :"=r" (fragment), "=r" (imm8OfPShufW1), "=r" (imm8OfPShufW2),
 146                          "=r" (fragmentLength)
 147                 );
 148
 149                 xpos= xx=xalpha= 0;
 150                 //FIXME choose size and or xinc so that they fit exactly
 151                 for(i=0; i<dstw/8; i++)
 152                 {
 153                         int xx=xpos>>8;
 154
 155                         if((i&3) == 0)
 156                         {
 157                                 int a=0;
 158                                 int b=((xpos+s_xinc)>>8) - xx;
 159                                 int c=((xpos+s_xinc*2)>>8) - xx;
 160                                 int d=((xpos+s_xinc*3)>>8) - xx;
 161
 162                                 memcpy(funnyYCode + fragmentLength*i/4, fragment, fragmentLength);
 163
 164                                 funnyYCode[fragmentLength*i/4 + imm8OfPShufW1]=
 165                                 funnyYCode[fragmentLength*i/4 + imm8OfPShufW2]=
 166                                         a | (b<<2) | (c<<4) | (d<<6);
 167
 168                                 funnyYCode[fragmentLength*(i+4)/4]= RET;
 169                         }
 170                         xpos+=s_xinc;
 171                 }
 172
 173                 xpos= xx=xalpha= 0;
 174                 //FIXME choose size and or xinc so that they fit exactly
 175                 for(i=0; i<dstw/8; i++)
 176                 {
 177                         int xx=xpos>>8;
 178
 179                         if((i&3) == 0)
 180                         {
 181                                 int a=0;
 182                                 int b=((xpos+s_xinc2)>>8) - xx;
 183                                 int c=((xpos+s_xinc2*2)>>8) - xx;
 184                                 int d=((xpos+s_xinc2*3)>>8) - xx;
 185
 186                                 memcpy(funnyUVCode + fragmentLength*i/4, fragment, fragmentLength);
 187
 188                                 funnyUVCode[fragmentLength*i/4 + imm8OfPShufW1]=
 189                                 funnyUVCode[fragmentLength*i/4 + imm8OfPShufW2]=
 190                                         a | (b<<2) | (c<<4) | (d<<6);
 191
 192                                 funnyUVCode[fragmentLength*(i+4)/4]= RET;
 193                         }
 194                         xpos+=s_xinc2;
 195                 }
 196 //              funnyCode[0]= RET;
 197
 198
 199         }
 200 #endif
 201   } // reset counters
 202
 203   while(1){
 204     unsigned char *dest=dstptr+dststride*s_ypos;
 205     int y0=2+(s_srcypos>>16);
 206     int y1=1+(s_srcypos>>17);
 207     int yalpha=(s_srcypos&0xFFFF)>>7;
 208     int yalpha1=yalpha^511;
 209     int uvalpha=((s_srcypos>>1)&0xFFFF)>>7;
 210     int uvalpha1=uvalpha^511;
 211     uint16_t *buf0=pix_buf_y[y0&3];
 212     uint16_t *buf1=pix_buf_y[((y0+1)&3)];
 213     uint16_t *uvbuf0=pix_buf_uv[y1&1];
 214     uint16_t *uvbuf1=pix_buf_uv[(y1&1)^1];
 215     int i;
 216
 217     if(y0>=y+h) break;
 218
 219     s_ypos++; s_srcypos+=s_yinc;
 220
 221     if(s_last_ypos!=y0){
 222       unsigned char *src=srcptr[0]+(y0-y)*stride[0];
 223       unsigned int xpos=0;
 224       s_last_ypos=y0;
 225       // *** horizontal scale Y line to temp buffer
 226       // this loop should be rewritten in MMX assembly!!!!
 227 #ifdef HAVE_MMX2
 228         asm volatile(
 229                 "pxor %%mm7, %%mm7              \n\t"
 230                 "pxor %%mm2, %%mm2              \n\t" // 2*xalpha
 231                 "movd %5, %%mm6                 \n\t" // s_xinc&0xFF
 232                 "punpcklwd %%mm6, %%mm6         \n\t"
 233                 "punpcklwd %%mm6, %%mm6         \n\t"
 234                 "movq %%mm6, %%mm2              \n\t"
 235                 "psllq $16, %%mm2               \n\t"
 236                 "paddb %%mm6, %%mm2             \n\t"
 237                 "psllq $16, %%mm2               \n\t"
 238                 "paddb %%mm6, %%mm2             \n\t"
 239                 "psllq $16, %%mm2               \n\t" //0,t,2t,3t               t=s_xinc&0xFF
 240                 "movq %%mm2, temp0              \n\t"
 241                 "movd %4, %%mm6                 \n\t" //(s_xinc*4)&0xFF
 242                 "punpcklwd %%mm6, %%mm6         \n\t"
 243                 "punpcklwd %%mm6, %%mm6         \n\t"
 244                 "xorl %%eax, %%eax              \n\t" // i
 245                 "xorl %%ebx, %%ebx              \n\t" // xx
 246                 "movl %0, %%esi                 \n\t" // src
 247                 "movl %1, %%edi                 \n\t" // buf1
 248                 "movl %3, %%edx                 \n\t" // (s_xinc*4)>>8
 249                 "xorl %%ecx, %%ecx              \n\t"
 250                 "movb %4, %%ch                  \n\t" // (s_xinc*4)&0xFF
 251 //      "int $3\n\t"
 252                 "call funnyYCode                        \n\t"
 253                 "movq temp0, %%mm2              \n\t"
 254                 "xorb %%cl, %%cl                \n\t"
 255                 "call funnyYCode                        \n\t"
 256                 "movq temp0, %%mm2              \n\t"
 257                 "xorb %%cl, %%cl                \n\t"
 258                 "call funnyYCode                        \n\t"
 259                 "movq temp0, %%mm2              \n\t"
 260                 "xorb %%cl, %%cl                \n\t"
 261                 "call funnyYCode                        \n\t"
 262                 "movq temp0, %%mm2              \n\t"
 263                 "xorb %%cl, %%cl                \n\t"
 264                 "call funnyYCode                        \n\t"
 265                 "movq temp0, %%mm2              \n\t"
 266                 "xorb %%cl, %%cl                \n\t"
 267                 "call funnyYCode                        \n\t"
 268                 "movq temp0, %%mm2              \n\t"
 269                 "xorb %%cl, %%cl                \n\t"
 270                 "call funnyYCode                        \n\t"
 271                 "movq temp0, %%mm2              \n\t"
 272                 "xorb %%cl, %%cl                \n\t"
 273                 "call funnyYCode                        \n\t"
 274                 :: "m" (src), "m" (buf1), "m" (dstw), "m" ((s_xinc*4)>>8),
 275                   "m" ((s_xinc*4)&0xFF), "m" (s_xinc&0xFF)
 276                 : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
 277         );
 278
 279 #elif defined (ARCH_X86)
 280         //NO MMX just normal asm ... FIXME try/write funny MMX2 variant
 281         //FIXME add prefetch
 282         asm volatile(
 283                 "xorl %%eax, %%eax              \n\t" // i
 284                 "xorl %%ebx, %%ebx              \n\t" // xx
 285                 "xorl %%ecx, %%ecx              \n\t" // 2*xalpha
 286                 "1:                             \n\t"
 287                 "movzbl  (%0, %%ebx), %%edi     \n\t" //src[xx]
 288                 "movzbl 1(%0, %%ebx), %%esi     \n\t" //src[xx+1]
 289                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
 290                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
 291                 "shll $8, %%edi                 \n\t"
 292                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
 293                 "movl %1, %%edi                 \n\t"
 294                 "shrl $1, %%esi                 \n\t"
 295                 "movw %%si, (%%edi, %%eax, 2)   \n\t"
 296                 "addb %4, %%cl                  \n\t" //2*xalpha += s_xinc&0xFF
 297                 "adcl %3, %%ebx                 \n\t" //xx+= s_xinc>>8 + carry
 298
 299                 "movzbl (%0, %%ebx), %%edi      \n\t" //src[xx]
 300                 "movzbl 1(%0, %%ebx), %%esi     \n\t" //src[xx+1]
 301                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
 302                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
 303                 "shll $8, %%edi                 \n\t"
 304                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
 305                 "movl %1, %%edi                 \n\t"
 306                 "shrl $1, %%esi                 \n\t"
 307                 "movw %%si, 2(%%edi, %%eax, 2)  \n\t"
 308                 "addb %4, %%cl                  \n\t" //2*xalpha += s_xinc&0xFF
 309                 "adcl %3, %%ebx                 \n\t" //xx+= s_xinc>>8 + carry
 310
 311
 312                 "addl $2, %%eax                 \n\t"
 313                 "cmpl %2, %%eax                 \n\t"
 314                 " jb 1b                         \n\t"
 315
 316
 317                 :: "r" (src), "m" (buf1), "m" (dstw), "m" (s_xinc>>8), "m" (s_xinc&0xFF)
 318                 : "%eax", "%ebx", "%ecx", "%edi", "%esi"
 319                 );
 320 #else
 321       for(i=0;i<dstw;i++){
 322         register unsigned int xx=xpos>>8;
 323         register unsigned int xalpha=(xpos&0xFF)>>1;
 324         buf1[i]=(src[xx]*(xalpha^127)+src[xx+1]*xalpha);
 325         xpos+=s_xinc;
 326       }
 327 #endif
 328       // *** horizontal scale U and V lines to temp buffer
 329       if(!(y0&1)){
 330         unsigned char *src1=srcptr[1]+(y1-y/2)*stride[1];
 331         unsigned char *src2=srcptr[2]+(y1-y/2)*stride[2];
 332         xpos=0;
 333         // this loop should be rewritten in MMX assembly!!!!
 334 #ifdef HAVE_MMX2
 335         asm volatile(
 336                 "pxor %%mm7, %%mm7              \n\t"
 337                 "pxor %%mm2, %%mm2              \n\t" // 2*xalpha
 338                 "movd %5, %%mm6                 \n\t" // s_xinc&0xFF
 339                 "punpcklwd %%mm6, %%mm6         \n\t"
 340                 "punpcklwd %%mm6, %%mm6         \n\t"
 341                 "movq %%mm6, %%mm2              \n\t"
 342                 "psllq $16, %%mm2               \n\t"
 343                 "paddb %%mm6, %%mm2             \n\t"
 344                 "psllq $16, %%mm2               \n\t"
 345                 "paddb %%mm6, %%mm2             \n\t"
 346                 "psllq $16, %%mm2               \n\t" //0,t,2t,3t               t=s_xinc&0xFF
 347                 "movq %%mm2, temp0              \n\t"
 348                 "movd %4, %%mm6                 \n\t" //(s_xinc*4)&0xFF
 349                 "punpcklwd %%mm6, %%mm6         \n\t"
 350                 "punpcklwd %%mm6, %%mm6         \n\t"
 351                 "xorl %%eax, %%eax              \n\t" // i
 352                 "xorl %%ebx, %%ebx              \n\t" // xx
 353                 "movl %0, %%esi                 \n\t" // src
 354                 "movl %1, %%edi                 \n\t" // buf1
 355                 "movl %3, %%edx                 \n\t" // (s_xinc*4)>>8
 356                 "xorl %%ecx, %%ecx              \n\t"
 357                 "movb %4, %%ch                  \n\t" // (s_xinc*4)&0xFF
 358 //      "int $3\n\t"
 359                 "call funnyUVCode                       \n\t"
 360                 "movq temp0, %%mm2              \n\t"
 361                 "xorb %%cl, %%cl                \n\t"
 362                 "call funnyUVCode                       \n\t"
 363                 "movq temp0, %%mm2              \n\t"
 364                 "xorb %%cl, %%cl                \n\t"
 365                 "call funnyUVCode                       \n\t"
 366                 "movq temp0, %%mm2              \n\t"
 367                 "xorb %%cl, %%cl                \n\t"
 368                 "call funnyUVCode                       \n\t"
 369                 "movq temp0, %%mm2              \n\t"
 370                 "xorb %%cl, %%cl                \n\t"
 371                 "call funnyUVCode                       \n\t"
 372                 "movq temp0, %%mm2              \n\t"
 373                 "xorb %%cl, %%cl                \n\t"
 374                 "call funnyUVCode                       \n\t"
 375                 "movq temp0, %%mm2              \n\t"
 376                 "xorb %%cl, %%cl                \n\t"
 377                 "call funnyUVCode                       \n\t"
 378                 "movq temp0, %%mm2              \n\t"
 379                 "xorb %%cl, %%cl                \n\t"
 380                 "call funnyUVCode                       \n\t"
 381
 382                 "xorl %%eax, %%eax              \n\t" // i
 383                 "xorl %%ebx, %%ebx              \n\t" // xx
 384                 "movl %6, %%esi                 \n\t" // src
 385                 "movl %1, %%edi                 \n\t" // buf1
 386                 "addl $4096, %%edi              \n\t"
 387
 388                 "call funnyUVCode                       \n\t"
 389                 "movq temp0, %%mm2              \n\t"
 390                 "xorb %%cl, %%cl                \n\t"
 391                 "call funnyUVCode                       \n\t"
 392                 "movq temp0, %%mm2              \n\t"
 393                 "xorb %%cl, %%cl                \n\t"
 394                 "call funnyUVCode                       \n\t"
 395                 "movq temp0, %%mm2              \n\t"
 396                 "xorb %%cl, %%cl                \n\t"
 397                 "call funnyUVCode                       \n\t"
 398                 "movq temp0, %%mm2              \n\t"
 399                 "xorb %%cl, %%cl                \n\t"
 400                 "call funnyUVCode                       \n\t"
 401                 "movq temp0, %%mm2              \n\t"
 402                 "xorb %%cl, %%cl                \n\t"
 403                 "call funnyUVCode                       \n\t"
 404                 "movq temp0, %%mm2              \n\t"
 405                 "xorb %%cl, %%cl                \n\t"
 406                 "call funnyUVCode                       \n\t"
 407                 "movq temp0, %%mm2              \n\t"
 408                 "xorb %%cl, %%cl                \n\t"
 409                 "call funnyUVCode                       \n\t"
 410
 411                 :: "m" (src1), "m" (uvbuf1), "m" (dstw), "m" ((s_xinc2*4)>>8),
 412                   "m" ((s_xinc2*4)&0xFF), "m" (s_xinc2&0xFF), "m" (src2)
 413                 : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
 414         );
 415
 416 #elif defined (ARCH_X86)
 417         asm volatile(
 418                 "xorl %%eax, %%eax              \n\t" // i
 419                 "xorl %%ebx, %%ebx              \n\t" // xx
 420                 "xorl %%ecx, %%ecx              \n\t" // 2*xalpha
 421                 "1:                             \n\t"
 422                 "movl %0, %%esi                 \n\t"
 423                 "movzbl  (%%esi, %%ebx), %%edi  \n\t" //src[xx]
 424                 "movzbl 1(%%esi, %%ebx), %%esi  \n\t" //src[xx+1]
 425                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
 426                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
 427                 "shll $8, %%edi                 \n\t"
 428                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
 429                 "movl %1, %%edi                 \n\t"
 430                 "shrl $1, %%esi                 \n\t"
 431                 "movw %%si, (%%edi, %%eax, 2)   \n\t"
 432
 433                 "movzbl  (%5, %%ebx), %%edi     \n\t" //src[xx]
 434                 "movzbl 1(%5, %%ebx), %%esi     \n\t" //src[xx+1]
 435                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
 436                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
 437                 "shll $8, %%edi                 \n\t"
 438                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
 439                 "movl %1, %%edi                 \n\t"
 440                 "shrl $1, %%esi                 \n\t"
 441                 "movw %%si, 4096(%%edi, %%eax, 2)\n\t"
 442
 443                 "addb %4, %%cl                  \n\t" //2*xalpha += s_xinc&0xFF
 444                 "adcl %3, %%ebx                 \n\t" //xx+= s_xinc>>8 + carry
 445                 "addl $1, %%eax                 \n\t"
 446                 "cmpl %2, %%eax                 \n\t"
 447                 " jb 1b                         \n\t"
 448
 449
 450                 :: "m" (src1), "m" (uvbuf1), "m" (dstw), "m" (s_xinc2>>8), "m" (s_xinc2&0xFF),
 451                 "r" (src2)
 452                 : "%eax", "%ebx", "%ecx", "%edi", "%esi"
 453                 );
 454 #else
 455         for(i=0;i<dstw;i++){
 456           register unsigned int xx=xpos>>8;
 457           register unsigned int xalpha=(xpos&0xFF)>>1;
 458           uvbuf1[i]=(src1[xx]*(xalpha^127)+src1[xx+1]*xalpha);
 459           uvbuf1[i+2048]=(src2[xx]*(xalpha^127)+src2[xx+1]*xalpha);
 460           xpos+=s_xinc2;
 461         }
 462 #endif
 463       }
 464       if(!y0) continue;
 465     }
 466
 467     // Note1: this code can be resticted to n*8 (or n*16) width lines to simplify optimization...
 468     // Re: Note1: ok n*4 for now
 469     // Note2: instead of using lookup tabs, mmx version could do the multiply...
 470     // Re: Note2: yep
 471     // Note3: maybe we should make separated 15/16, 24 and 32bpp version of this:
 472     // Re: done (32 & 16) and 16 has dithering :) but 16 is untested
 473 #ifdef HAVE_MMX
 474         //FIXME write lq version with less uv ...
 475         //FIXME reorder / optimize
 476         if(dstbpp == 32)
 477         {
 478                 asm volatile(
 479
 480 #define YSCALEYUV2RGB \
 481                 "pxor %%mm7, %%mm7              \n\t"\
 482                 "movd %6, %%mm6                 \n\t" /*yalpha1*/\
 483                 "punpcklwd %%mm6, %%mm6         \n\t"\
 484                 "punpcklwd %%mm6, %%mm6         \n\t"\
 485                 "movd %7, %%mm5                 \n\t" /*uvalpha1*/\
 486                 "punpcklwd %%mm5, %%mm5         \n\t"\
 487                 "punpcklwd %%mm5, %%mm5         \n\t"\
 488                 "xorl %%eax, %%eax              \n\t"\
 489                 "1:                             \n\t"\
 490                 "movq (%0, %%eax, 2), %%mm0     \n\t" /*buf0[eax]*/\
 491                 "movq (%1, %%eax, 2), %%mm1     \n\t" /*buf1[eax]*/\
 492                 "psubw %%mm1, %%mm0             \n\t" /* buf0[eax] - buf1[eax]*/\
 493                 "pmulhw %%mm6, %%mm0            \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
 494                 "psraw $7, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>7*/\
 495                 "paddw %%mm0, %%mm1             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
 496                 "psubw w10, %%mm1               \n\t" /* Y-16*/\
 497                 "psllw $3, %%mm1                \n\t" /* (y-16)*8*/\
 498                 "pmulhw yCoeff, %%mm1           \n\t"\
 499 \
 500                 "movq (%2, %%eax,2), %%mm2      \n\t" /* uvbuf0[eax]*/\
 501                 "movq (%3, %%eax,2), %%mm3      \n\t" /* uvbuf1[eax]*/\
 502                 "psubw %%mm3, %%mm2             \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
 503                 "pmulhw %%mm5, %%mm2            \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
 504                 "psraw $7, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>7*/\
 505                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
 506                 "psubw w80, %%mm3               \n\t" /* (U-128)*/\
 507                 "psllw $3, %%mm3                \n\t" /*(U-128)8*/\
 508 \
 509                 "movq 4096(%2, %%eax,2), %%mm4  \n\t" /* uvbuf0[eax+2048]*/\
 510                 "movq 4096(%3, %%eax,2), %%mm0  \n\t" /* uvbuf1[eax+2048]*/\
 511                 "psubw %%mm0, %%mm4             \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
 512                 "pmulhw %%mm5, %%mm4            \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
 513                 "psraw $7, %%mm0                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>7*/\
 514                 "paddw %%mm4, %%mm0             \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
 515                 "psubw w80, %%mm0               \n\t" /* (V-128)*/\
 516                 "psllw $3, %%mm0                \n\t" /* (V-128)8*/\
 517 \
 518                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
 519                 "pmulhw ubCoeff, %%mm3          \n\t"\
 520                 "paddw %%mm1, %%mm3             \n\t" /* B*/\
 521 \
 522                 "movq %%mm0, %%mm4              \n\t" /* (V-128)8*/\
 523                 "pmulhw vrCoeff, %%mm0          \n\t"\
 524                 "paddw %%mm1, %%mm0             \n\t" /* R*/\
 525 \
 526                 "pmulhw ugCoeff, %%mm2          \n\t"\
 527                 "pmulhw vgCoeff, %%mm4          \n\t"\
 528                 "paddw %%mm4, %%mm2             \n\t"\
 529                 "paddw %%mm2, %%mm1             \n\t" /* G*/\
 530 \
 531                 "packuswb %%mm3, %%mm3          \n\t"\
 532                 "packuswb %%mm0, %%mm0          \n\t"\
 533                 "packuswb %%mm1, %%mm1          \n\t"
 534
 535 YSCALEYUV2RGB
 536                 "punpcklbw %%mm1, %%mm3         \n\t" // BGBGBGBG
 537                 "punpcklbw %%mm7, %%mm0         \n\t" // R0R0R0R0
 538
 539                 "movq %%mm3, %%mm1              \n\t"
 540                 "punpcklwd %%mm0, %%mm3         \n\t" // BGR0BGR0
 541                 "punpckhwd %%mm0, %%mm1         \n\t" // BGR0BGR0
 542 #ifdef HAVE_MMX2
 543                 "movntq %%mm3, (%4, %%eax, 4)   \n\t"
 544                 "movntq %%mm1, 8(%4, %%eax, 4)  \n\t"
 545 #else
 546                 "movq %%mm3, (%4, %%eax, 4)     \n\t"
 547                 "movq %%mm1, 8(%4, %%eax, 4)    \n\t"
 548 #endif
 549                 "addl $4, %%eax                 \n\t"
 550                 "cmpl %5, %%eax                 \n\t"
 551                 " jb 1b                         \n\t"
 552
 553
 554                 :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 555                 "m" (yalpha1), "m" (uvalpha1)
 556                 : "%eax"
 557                 );
 558         }
 559         else if(dstbpp==24)
 560         {
 561                 asm volatile(
 562
 563 YSCALEYUV2RGB
 564
 565                                                         // lsb ... msb
 566                 "punpcklbw %%mm1, %%mm3         \n\t" // BGBGBGBG
 567                 "punpcklbw %%mm7, %%mm0         \n\t" // R0R0R0R0
 568
 569                 "movq %%mm3, %%mm1              \n\t"
 570                 "punpcklwd %%mm0, %%mm3         \n\t" // BGR0BGR0
 571                 "punpckhwd %%mm0, %%mm1         \n\t" // BGR0BGR0
 572
 573                 "movq %%mm3, %%mm2              \n\t" // BGR0BGR0
 574                 "psrlq $8, %%mm3                \n\t" // GR0BGR00
 575                 "pand bm00000111, %%mm2         \n\t" // BGR00000
 576                 "pand bm11111000, %%mm3         \n\t" // 000BGR00
 577                 "por %%mm2, %%mm3               \n\t" // BGRBGR00
 578                 "movq %%mm1, %%mm2              \n\t"
 579                 "psllq $48, %%mm1               \n\t" // 000000BG
 580                 "por %%mm1, %%mm3               \n\t" // BGRBGRBG
 581
 582                 "movq %%mm2, %%mm1              \n\t" // BGR0BGR0
 583                 "psrld $16, %%mm2               \n\t" // R000R000
 584                 "psrlq $24, %%mm1               \n\t" // 0BGR0000
 585                 "por %%mm2, %%mm1               \n\t" // RBGRR000
 586
 587                 "movl %4, %%ebx                 \n\t"
 588                 "addl %%eax, %%ebx              \n\t"
 589 #ifdef HAVE_MMX2
 590                 //FIXME Alignment
 591                 "movntq %%mm3, (%%ebx, %%eax, 2)\n\t"
 592                 "movntq %%mm1, 8(%%ebx, %%eax, 2)\n\t"
 593 #else
 594                 "movd %%mm3, (%%ebx, %%eax, 2)  \n\t"
 595                 "psrlq $32, %%mm3               \n\t"
 596                 "movd %%mm3, 4(%%ebx, %%eax, 2) \n\t"
 597                 "movd %%mm1, 8(%%ebx, %%eax, 2) \n\t"
 598 #endif
 599                 "addl $4, %%eax                 \n\t"
 600                 "cmpl %5, %%eax                 \n\t"
 601                 " jb 1b                         \n\t"
 602
 603                 :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
 604                 "m" (yalpha1), "m" (uvalpha1)
 605                 : "%eax", "%ebx"
 606                 );
 607         }
 608         else if(dstbpp==16)
 609         {
 610                 asm volatile(
 611
 612 YSCALEYUV2RGB
 613                 "paddusb g16Dither, %%mm1       \n\t"
 614                 "paddusb b16Dither, %%mm0       \n\t"
 615                 "paddusb b16Dither, %%mm3       \n\t"
 616                 "punpcklbw %%mm7, %%mm1         \n\t" // 0G0G0G0G
 617                 "punpcklbw %%mm7, %%mm3         \n\t" // 0B0B0B0B
 618                 "punpcklbw %%mm7, %%mm0         \n\t" // 0R0R0R0R
 619
 620                 "psrlw $3, %%mm3                \n\t"
 621                 "psllw $3, %%mm1                \n\t"
 622                 "psllw $8, %%mm0                \n\t"
 623                 "pand g16Mask, %%mm1            \n\t"
 624                 "pand r16Mask, %%mm0            \n\t"
 625
 626                 "por %%mm3, %%mm1               \n\t"
 627                 "por %%mm1, %%mm0               \n\t"
 628 #ifdef HAVE_MMX2
 629                 "movntq %%mm0, (%4, %%eax, 2)   \n\t"
 630 #else
 631                 "movq %%mm0, (%4, %%eax, 2)     \n\t"
 632 #endif
 633                 "addl $4, %%eax                 \n\t"
 634                 "cmpl %5, %%eax                 \n\t"
 635                 " jb 1b                         \n\t"
 636
 637                 :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 638                 "m" (yalpha1), "m" (uvalpha1)
 639                 : "%eax"
 640                 );
 641         }
 642 #else
 643         if(dstbpp==32 || dstbpp==24)
 644         {
 645                 for(i=0;i<dstw;i++){
 646                         // vertical linear interpolation && yuv2rgb in a single step:
 647                         int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
 648                         int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
 649                         int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
 650                         dest[0]=clip_table[((Y + yuvtab_3343[U]) >>13)];
 651                         dest[1]=clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)];
 652                         dest[2]=clip_table[((Y + yuvtab_40cf[V]) >>13)];
 653                         dest+=dstbpp>>3;
 654                 }
 655         }
 656         else if(dstbpp==16)
 657         {
 658                 for(i=0;i<dstw;i++){
 659                         // vertical linear interpolation && yuv2rgb in a single step:
 660                         int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
 661                         int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
 662                         int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
 663
 664                         ((uint16_t*)dest)[0] =
 665                                 (clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
 666                                 (clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<3)&0x07E0 |
 667                                 (clip_table[((Y + yuvtab_40cf[V]) >>13)]<<8)&0xF800;
 668                         dest+=2;
 669                 }
 670         }
 671         else if(dstbpp==15) //15bit FIXME how do i figure out if its 15 or 16?
 672         {
 673                 for(i=0;i<dstw;i++){
 674                         // vertical linear interpolation && yuv2rgb in a single step:
 675                         int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
 676                         int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
 677                         int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
 678
 679                         ((uint16_t*)dest)[0] =
 680                                 (clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
 681                                 (clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<2)&0x03E0 |
 682                                 (clip_table[((Y + yuvtab_40cf[V]) >>13)]<<7)&0x7C00;
 683                         dest+=2;
 684                 }
 685         }
 686 #endif
 687
 688         b16Dither= b16Dither1;
 689         b16Dither1= b16Dither2;
 690         b16Dither2= b16Dither;
 691
 692         g16Dither= g16Dither1;
 693         g16Dither1= g16Dither2;
 694         g16Dither2= g16Dither;
 695   }
 696
 697 #ifdef HAVE_3DNOW
 698         asm volatile("femms");
 699 #elif defined (HAVE_MMX)
 700         asm volatile("emms");
 701 #endif
 702
 703
 704 }
 705
 706
 707 void SwScale_Init(){
 708     // generating tables:
 709     int i;
 710     for(i=0;i<256;i++){
 711         clip_table[i]=0;
 712         clip_table[i+256]=i;
 713         clip_table[i+512]=255;
 714         yuvtab_2568[i]=(0x2568*(i-16))+(256<<13);
 715         yuvtab_3343[i]=0x3343*(i-128);
 716         yuvtab_0c92[i]=-0x0c92*(i-128);
 717         yuvtab_1a1e[i]=-0x1a1e*(i-128);
 718         yuvtab_40cf[i]=0x40cf*(i-128);
 719     }
 720
 721 }