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

   1
   2 // Software scaling and colorspace conversion routines for MPlayer
   3
   4 // Orginal C implementation by ?
   5 // current version mostly by Michael Niedermayer (michaelni@gmx.at)
   6
   7 #include <inttypes.h>
   8 #include "../config.h"
   9
  10 //#undef HAVE_MMX2
  11 //#undef HAVE_MMX
  12 //#undef ARCH_X86
  13 #define DITHER16BPP
  14 #define ALT_ERROR
  15
  16 #define RET 0xC3 //near return opcode
  17 /*
  18 NOTES
  19
  20 known BUGS with known cause (no bugreports please!)
  21 line at the right (c,asm and mmx2)
  22 code reads 1 sample too much (might cause a sig11)
  23
  24 TODO
  25 check alignment off everything
  26 */
  27
  28 static uint64_t yCoeff=    0x2568256825682568LL;
  29 static uint64_t ubCoeff=   0x3343334333433343LL;
  30 static uint64_t vrCoeff=   0x40cf40cf40cf40cfLL;
  31 static uint64_t ugCoeff=   0xE5E2E5E2E5E2E5E2LL;
  32 static uint64_t vgCoeff=   0xF36EF36EF36EF36ELL;
  33 static uint64_t w80=       0x0080008000800080LL;
  34 static uint64_t w10=       0x0010001000100010LL;
  35 static uint64_t bm00000111=0x0000000000FFFFFFLL;
  36 static uint64_t bm11111000=0xFFFFFFFFFF000000LL;
  37
  38 static uint64_t b16Dither= 0x0004000400040004LL;
  39 static uint64_t b16Dither1=0x0004000400040004LL;
  40 static uint64_t b16Dither2=0x0602060206020602LL;
  41 static uint64_t g16Dither= 0x0002000200020002LL;
  42 static uint64_t g16Dither1=0x0002000200020002LL;
  43 static uint64_t g16Dither2=0x0301030103010301LL;
  44
  45 static uint64_t b16Mask=   0x001F001F001F001FLL;
  46 static uint64_t g16Mask=   0x07E007E007E007E0LL;
  47 static uint64_t r16Mask=   0xF800F800F800F800LL;
  48 static uint64_t temp0;
  49
  50
  51 // temporary storage for 4 yuv lines:
  52 // 16bit for now (mmx likes it more compact)
  53 static uint16_t pix_buf_y[4][2048];
  54 static uint16_t pix_buf_uv[2][2048*2];
  55
  56 // clipping helper table for C implementations:
  57 static unsigned char clip_table[768];
  58
  59 // yuv->rgb conversion tables:
  60 static    int yuvtab_2568[256];
  61 static    int yuvtab_3343[256];
  62 static    int yuvtab_0c92[256];
  63 static    int yuvtab_1a1e[256];
  64 static    int yuvtab_40cf[256];
  65
  66
  67 static uint8_t funnyYCode[10000];
  68 static uint8_t funnyUVCode[10000];
  69
  70
  71 // *** bilinear scaling and yuv->rgb conversion of yv12 slices:
  72 // *** Note: it's called multiple times while decoding a frame, first time y==0
  73 // *** Designed to upscale, but may work for downscale too.
  74 // s_xinc = (src_width << 8) / dst_width
  75 // s_yinc = (src_height << 16) / dst_height
  76 void SwScale_YV12slice_brg24(unsigned char* srcptr[],int stride[], int y, int h,
  77                              unsigned char* dstptr, int dststride, int dstw, int dstbpp,
  78                              unsigned int s_xinc,unsigned int s_yinc){
  79
  80 // scaling factors:
  81 //static int s_yinc=(vo_dga_src_height<<16)/vo_dga_vp_height;
  82 //static int s_xinc=(vo_dga_src_width<<8)/vo_dga_vp_width;
  83
  84 unsigned int s_xinc2;
  85
  86 static int s_srcypos; // points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
  87 static int s_ypos;
  88
  89 // last horzontally interpolated lines, used to avoid unnecessary calculations
  90 static int s_last_ypos;
  91 static int s_last_y1pos;
  92
  93 static int static_dstw;
  94
  95 #ifdef HAVE_MMX2
  96 // used to detect a horizontal size change
  97 static int old_dstw= -1;
  98 static int old_s_xinc= -1;
  99
 100 // difference between the requested xinc and the required one for the mmx2 routine
 101 static int s_xinc_diff=0;
 102 static int s_xinc2_diff=0;
 103 #endif
 104 int canMMX2BeUsed;
 105
 106 // we need that precission at least for the mmx2 code
 107 s_xinc*= 256;
 108 s_xinc2=s_xinc>>1;
 109 canMMX2BeUsed= (s_xinc <= 0x10000 && (dstw&31)==0) ? 1 : 0;
 110
 111   if(y==0){
 112       s_srcypos= s_yinc/2 - 0x8000;
 113       s_ypos=0;
 114
 115       // force calculation of the horizontal interpolation of the first line
 116       s_last_ypos=-99;
 117       s_last_y1pos=-99;
 118 #ifdef HAVE_MMX2
 119 // cant downscale !!!
 120         if((old_s_xinc != s_xinc || old_dstw!=dstw) && canMMX2BeUsed)
 121         {
 122                 uint8_t *fragment;
 123                 int imm8OfPShufW1;
 124                 int imm8OfPShufW2;
 125                 int fragmentLength;
 126
 127                 int xpos, xx, xalpha, i;
 128
 129                 old_s_xinc= s_xinc;
 130                 old_dstw= dstw;
 131
 132                 static_dstw= dstw;
 133
 134                 // create an optimized horizontal scaling routine
 135
 136                 //code fragment
 137
 138                 asm volatile(
 139                         "jmp 9f                         \n\t"
 140                 // Begin
 141                         "0:                             \n\t"
 142                         "movq (%%esi), %%mm0            \n\t" //FIXME Alignment
 143                         "movq %%mm0, %%mm1              \n\t"
 144                         "psrlq $8, %%mm0                \n\t"
 145                         "punpcklbw %%mm7, %%mm1 \n\t"
 146                         "movq %%mm2, %%mm3              \n\t"
 147                         "punpcklbw %%mm7, %%mm0 \n\t"
 148                         "addw %%bx, %%cx                \n\t" //2*xalpha += (4*s_xinc)&0xFFFF
 149                         "pshufw $0xFF, %%mm1, %%mm1     \n\t"
 150                         "1:                             \n\t"
 151                         "adcl %%edx, %%esi              \n\t" //xx+= (4*s_xinc)>>16 + carry
 152                         "pshufw $0xFF, %%mm0, %%mm0     \n\t"
 153                         "2:                             \n\t"
 154                         "psrlw $9, %%mm3                \n\t"
 155                         "psubw %%mm1, %%mm0             \n\t"
 156                         "pmullw %%mm3, %%mm0            \n\t"
 157                         "paddw %%mm6, %%mm2             \n\t" // 2*alpha += xpos&0xFFFF
 158                         "psllw $7, %%mm1                \n\t"
 159                         "paddw %%mm1, %%mm0             \n\t"
 160
 161                         "movq %%mm0, (%%edi, %%eax)     \n\t"
 162
 163                         "addl $8, %%eax                 \n\t"
 164                 // End
 165                         "9:                             \n\t"
 166 //              "int $3\n\t"
 167                         "leal 0b, %0                    \n\t"
 168                         "leal 1b, %1                    \n\t"
 169                         "leal 2b, %2                    \n\t"
 170                         "decl %1                        \n\t"
 171                         "decl %2                        \n\t"
 172                         "subl %0, %1                    \n\t"
 173                         "subl %0, %2                    \n\t"
 174                         "leal 9b, %3                    \n\t"
 175                         "subl %0, %3                    \n\t"
 176                         :"=r" (fragment), "=r" (imm8OfPShufW1), "=r" (imm8OfPShufW2),
 177                          "=r" (fragmentLength)
 178                 );
 179
 180                 xpos= xx=xalpha= 0;
 181
 182                 /* choose xinc so that all 8 parts fit exactly
 183                    Note: we cannot use just 1 part because it would not fit in the code cache */
 184                 s_xinc2_diff= -((((s_xinc2*(dstw/8))&0xFFFF))/(dstw/8))+10;
 185 //              s_xinc_diff= -((((s_xinc*(dstw/8))&0xFFFF))/(dstw/8));
 186 #ifdef ALT_ERROR
 187                 s_xinc2_diff+= ((0x10000/(dstw/8)));
 188 #endif
 189                 s_xinc_diff= s_xinc2_diff*2;
 190
 191                 s_xinc2+= s_xinc2_diff;
 192                 s_xinc+= s_xinc_diff;
 193                 for(i=0; i<dstw/8; i++)
 194                 {
 195                         int xx=xpos>>16;
 196
 197                         if((i&3) == 0)
 198                         {
 199                                 int a=0;
 200                                 int b=((xpos+s_xinc)>>16) - xx;
 201                                 int c=((xpos+s_xinc*2)>>16) - xx;
 202                                 int d=((xpos+s_xinc*3)>>16) - xx;
 203
 204                                 memcpy(funnyYCode + fragmentLength*i/4, fragment, fragmentLength);
 205
 206                                 funnyYCode[fragmentLength*i/4 + imm8OfPShufW1]=
 207                                 funnyYCode[fragmentLength*i/4 + imm8OfPShufW2]=
 208                                         a | (b<<2) | (c<<4) | (d<<6);
 209
 210                                 funnyYCode[fragmentLength*(i+4)/4]= RET;
 211                         }
 212                         xpos+=s_xinc;
 213                 }
 214
 215                 xpos= xx=xalpha= 0;
 216                 //FIXME choose size and or xinc so that they fit exactly
 217                 for(i=0; i<dstw/8; i++)
 218                 {
 219                         int xx=xpos>>16;
 220
 221                         if((i&3) == 0)
 222                         {
 223                                 int a=0;
 224                                 int b=((xpos+s_xinc2)>>16) - xx;
 225                                 int c=((xpos+s_xinc2*2)>>16) - xx;
 226                                 int d=((xpos+s_xinc2*3)>>16) - xx;
 227
 228                                 memcpy(funnyUVCode + fragmentLength*i/4, fragment, fragmentLength);
 229
 230                                 funnyUVCode[fragmentLength*i/4 + imm8OfPShufW1]=
 231                                 funnyUVCode[fragmentLength*i/4 + imm8OfPShufW2]=
 232                                         a | (b<<2) | (c<<4) | (d<<6);
 233
 234                                 funnyUVCode[fragmentLength*(i+4)/4]= RET;
 235                         }
 236                         xpos+=s_xinc2;
 237                 }
 238 //              funnyCode[0]= RET;
 239
 240
 241         }
 242
 243         if(canMMX2BeUsed)
 244         {
 245                 s_xinc+= s_xinc_diff;
 246                 s_xinc2+= s_xinc2_diff;
 247         }
 248 #endif // HAVE_MMX2
 249   } // reset counters
 250
 251   while(1){
 252     unsigned char *dest=dstptr+dststride*s_ypos;
 253     int y0=(s_srcypos + 0xFFFF)>>16;  // first luminance source line number below the dst line
 254         // points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
 255     int srcuvpos= s_srcypos + s_yinc/2 - 0x8000;
 256     int y1=(srcuvpos + 0x1FFFF)>>17; // first chrominance source line number below the dst line
 257     int yalpha=((s_srcypos-1)&0xFFFF)>>7;
 258     int yalpha1=yalpha^511;
 259     int uvalpha=((srcuvpos-1)&0x1FFFF)>>8;
 260     int uvalpha1=uvalpha^511;
 261     uint16_t *buf0=pix_buf_y[y0&1];             // top line of the interpolated slice
 262     uint16_t *buf1=pix_buf_y[((y0+1)&1)];       // bottom line of the interpolated slice
 263     uint16_t *uvbuf0=pix_buf_uv[y1&1];          // top line of the interpolated slice
 264     uint16_t *uvbuf1=pix_buf_uv[(y1+1)&1];      // bottom line of the interpolated slice
 265     int i;
 266
 267     // if this is before the first line than use only the first src line
 268     if(y0==0) buf0= buf1;
 269     if(y1==0) uvbuf0= uvbuf1; // yes we do have to check this, its not the same as y0==0
 270
 271     if(y0>=y+h) break; // FIXME wrong, skips last lines, but they are dupliactes anyway
 272
 273     // if this is after the last line than use only the last src line
 274     if(y0>=y+h)
 275     {
 276         buf1= buf0;
 277         s_last_ypos=y0;
 278     }
 279     if(y1>=(y+h)/2)
 280     {
 281         uvbuf1= uvbuf0;
 282         s_last_y1pos=y1;
 283     }
 284
 285
 286     s_ypos++; s_srcypos+=s_yinc;
 287
 288     //only interpolate the src line horizontally if we didnt do it allready
 289     if(s_last_ypos!=y0){
 290       unsigned char *src=srcptr[0]+(y0-y)*stride[0];
 291       unsigned int xpos=0;
 292       s_last_ypos=y0;
 293       // *** horizontal scale Y line to temp buffer
 294 #ifdef ARCH_X86
 295
 296 #ifdef HAVE_MMX2
 297         if(canMMX2BeUsed)
 298         {
 299                 asm volatile(
 300                         "pxor %%mm7, %%mm7              \n\t"
 301                         "pxor %%mm2, %%mm2              \n\t" // 2*xalpha
 302                         "movd %5, %%mm6                 \n\t" // s_xinc&0xFFFF
 303                         "punpcklwd %%mm6, %%mm6         \n\t"
 304                         "punpcklwd %%mm6, %%mm6         \n\t"
 305                         "movq %%mm6, %%mm2              \n\t"
 306                         "psllq $16, %%mm2               \n\t"
 307                         "paddw %%mm6, %%mm2             \n\t"
 308                         "psllq $16, %%mm2               \n\t"
 309                         "paddw %%mm6, %%mm2             \n\t"
 310                         "psllq $16, %%mm2               \n\t" //0,t,2t,3t               t=s_xinc&0xFF
 311                         "movq %%mm2, temp0              \n\t"
 312                         "movd %4, %%mm6                 \n\t" //(s_xinc*4)&0xFFFF
 313                         "punpcklwd %%mm6, %%mm6         \n\t"
 314                         "punpcklwd %%mm6, %%mm6         \n\t"
 315                         "xorl %%eax, %%eax              \n\t" // i
 316                         "movl %0, %%esi                 \n\t" // src
 317                         "movl %1, %%edi                 \n\t" // buf1
 318                         "movl %3, %%edx                 \n\t" // (s_xinc*4)>>16
 319                         "xorl %%ecx, %%ecx              \n\t"
 320                         "xorl %%ebx, %%ebx              \n\t"
 321                         "movw %4, %%bx                  \n\t" // (s_xinc*4)&0xFFFF
 322         //      "int $3\n\t"
 323                         "call funnyYCode                        \n\t"
 324                         "movq temp0, %%mm2              \n\t"
 325                         "xorl %%ecx, %%ecx              \n\t"
 326                         "call funnyYCode                        \n\t"
 327                         "movq temp0, %%mm2              \n\t"
 328                         "xorl %%ecx, %%ecx              \n\t"
 329                         "call funnyYCode                        \n\t"
 330                         "movq temp0, %%mm2              \n\t"
 331                         "xorl %%ecx, %%ecx              \n\t"
 332                         "call funnyYCode                        \n\t"
 333                         "movq temp0, %%mm2              \n\t"
 334                         "xorl %%ecx, %%ecx              \n\t"
 335                         "call funnyYCode                        \n\t"
 336                         "movq temp0, %%mm2              \n\t"
 337                         "xorl %%ecx, %%ecx              \n\t"
 338                         "call funnyYCode                        \n\t"
 339                         "movq temp0, %%mm2              \n\t"
 340                         "xorl %%ecx, %%ecx              \n\t"
 341                         "call funnyYCode                        \n\t"
 342                         "movq temp0, %%mm2              \n\t"
 343                         "xorl %%ecx, %%ecx              \n\t"
 344                         "call funnyYCode                        \n\t"
 345                         :: "m" (src), "m" (buf1), "m" (dstw), "m" ((s_xinc*4)>>16),
 346                         "m" ((s_xinc*4)&0xFFFF), "m" (s_xinc&0xFFFF)
 347                         : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
 348                 );
 349         }
 350         else
 351         {
 352 #endif
 353         //NO MMX just normal asm ... FIXME try/write funny MMX2 variant
 354         //FIXME add prefetch
 355         asm volatile(
 356                 "xorl %%eax, %%eax              \n\t" // i
 357                 "xorl %%ebx, %%ebx              \n\t" // xx
 358                 "xorl %%ecx, %%ecx              \n\t" // 2*xalpha
 359                 "1:                             \n\t"
 360                 "movzbl  (%0, %%ebx), %%edi     \n\t" //src[xx]
 361                 "movzbl 1(%0, %%ebx), %%esi     \n\t" //src[xx+1]
 362                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
 363                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
 364                 "shll $16, %%edi                \n\t"
 365                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
 366                 "movl %1, %%edi                 \n\t"
 367                 "shrl $9, %%esi                 \n\t"
 368                 "movw %%si, (%%edi, %%eax, 2)   \n\t"
 369                 "addw %4, %%cx                  \n\t" //2*xalpha += s_xinc&0xFF
 370                 "adcl %3, %%ebx                 \n\t" //xx+= s_xinc>>8 + carry
 371
 372                 "movzbl (%0, %%ebx), %%edi      \n\t" //src[xx]
 373                 "movzbl 1(%0, %%ebx), %%esi     \n\t" //src[xx+1]
 374                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
 375                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
 376                 "shll $16, %%edi                \n\t"
 377                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
 378                 "movl %1, %%edi                 \n\t"
 379                 "shrl $9, %%esi                 \n\t"
 380                 "movw %%si, 2(%%edi, %%eax, 2)  \n\t"
 381                 "addw %4, %%cx                  \n\t" //2*xalpha += s_xinc&0xFF
 382                 "adcl %3, %%ebx                 \n\t" //xx+= s_xinc>>8 + carry
 383
 384
 385                 "addl $2, %%eax                 \n\t"
 386                 "cmpl %2, %%eax                 \n\t"
 387                 " jb 1b                         \n\t"
 388
 389
 390                 :: "r" (src), "m" (buf1), "m" (dstw), "m" (s_xinc>>16), "m" (s_xinc&0xFFFF)
 391                 : "%eax", "%ebx", "%ecx", "%edi", "%esi"
 392                 );
 393 #ifdef HAVE_MMX2
 394         } //if MMX2 cant be used
 395 #endif
 396 #else
 397       for(i=0;i<dstw;i++){
 398         register unsigned int xx=xpos>>16;
 399         register unsigned int xalpha=(xpos&0xFFFF)>>9;
 400         buf1[i]=(src[xx]*(xalpha^127)+src[xx+1]*xalpha);
 401         xpos+=s_xinc;
 402       }
 403 #endif
 404     }
 405       // *** horizontal scale U and V lines to temp buffer
 406     if(s_last_y1pos!=y1){
 407         unsigned char *src1=srcptr[1]+(y1-y/2)*stride[1];
 408         unsigned char *src2=srcptr[2]+(y1-y/2)*stride[2];
 409         int xpos=0;
 410         s_last_y1pos= y1;
 411 #ifdef ARCH_X86
 412 #ifdef HAVE_MMX2
 413         if(canMMX2BeUsed)
 414         {
 415                 asm volatile(
 416                 "pxor %%mm7, %%mm7              \n\t"
 417                 "pxor %%mm2, %%mm2              \n\t" // 2*xalpha
 418                 "movd %5, %%mm6                 \n\t" // s_xinc&0xFFFF
 419                 "punpcklwd %%mm6, %%mm6         \n\t"
 420                 "punpcklwd %%mm6, %%mm6         \n\t"
 421                 "movq %%mm6, %%mm2              \n\t"
 422                 "psllq $16, %%mm2               \n\t"
 423                 "paddw %%mm6, %%mm2             \n\t"
 424                 "psllq $16, %%mm2               \n\t"
 425                 "paddw %%mm6, %%mm2             \n\t"
 426                 "psllq $16, %%mm2               \n\t" //0,t,2t,3t               t=s_xinc&0xFFFF
 427                 "movq %%mm2, temp0              \n\t"
 428                 "movd %4, %%mm6                 \n\t" //(s_xinc*4)&0xFFFF
 429                 "punpcklwd %%mm6, %%mm6         \n\t"
 430                 "punpcklwd %%mm6, %%mm6         \n\t"
 431                 "xorl %%eax, %%eax              \n\t" // i
 432                 "movl %0, %%esi                 \n\t" // src
 433                 "movl %1, %%edi                 \n\t" // buf1
 434                 "movl %3, %%edx                 \n\t" // (s_xinc*4)>>16
 435                 "xorl %%ecx, %%ecx              \n\t"
 436                 "xorl %%ebx, %%ebx              \n\t"
 437                 "movw %4, %%bx                  \n\t" // (s_xinc*4)&0xFFFF
 438
 439 //      "int $3\n\t"
 440 #define FUNNYUVCODE \
 441                 "call funnyUVCode               \n\t"\
 442                 "movq temp0, %%mm2              \n\t"\
 443                 "xorl %%ecx, %%ecx              \n\t"
 444
 445 FUNNYUVCODE
 446 FUNNYUVCODE
 447 FUNNYUVCODE
 448 FUNNYUVCODE
 449
 450 FUNNYUVCODE
 451 FUNNYUVCODE
 452 FUNNYUVCODE
 453 FUNNYUVCODE
 454
 455
 456
 457                 "xorl %%eax, %%eax              \n\t" // i
 458                 "movl %6, %%esi                 \n\t" // src
 459                 "movl %1, %%edi                 \n\t" // buf1
 460                 "addl $4096, %%edi              \n\t"
 461
 462 FUNNYUVCODE
 463 FUNNYUVCODE
 464 FUNNYUVCODE
 465 FUNNYUVCODE
 466
 467 FUNNYUVCODE
 468 FUNNYUVCODE
 469 FUNNYUVCODE
 470 FUNNYUVCODE
 471
 472                 :: "m" (src1), "m" (uvbuf1), "m" (dstw), "m" ((s_xinc2*4)>>16),
 473                   "m" ((s_xinc2*4)&0xFFFF), "m" (s_xinc2&0xFFFF), "m" (src2)
 474                 : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
 475         );
 476         }
 477         else
 478         {
 479 #endif
 480         asm volatile(
 481                 "xorl %%eax, %%eax              \n\t" // i
 482                 "xorl %%ebx, %%ebx              \n\t" // xx
 483                 "xorl %%ecx, %%ecx              \n\t" // 2*xalpha
 484                 "1:                             \n\t"
 485                 "movl %0, %%esi                 \n\t"
 486                 "movzbl  (%%esi, %%ebx), %%edi  \n\t" //src[xx]
 487                 "movzbl 1(%%esi, %%ebx), %%esi  \n\t" //src[xx+1]
 488                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
 489                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
 490                 "shll $16, %%edi                \n\t"
 491                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
 492                 "movl %1, %%edi                 \n\t"
 493                 "shrl $9, %%esi                 \n\t"
 494                 "movw %%si, (%%edi, %%eax, 2)   \n\t"
 495
 496                 "movzbl  (%5, %%ebx), %%edi     \n\t" //src[xx]
 497                 "movzbl 1(%5, %%ebx), %%esi     \n\t" //src[xx+1]
 498                 "subl %%edi, %%esi              \n\t" //src[xx+1] - src[xx]
 499                 "imull %%ecx, %%esi             \n\t" //(src[xx+1] - src[xx])*2*xalpha
 500                 "shll $16, %%edi                \n\t"
 501                 "addl %%edi, %%esi              \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
 502                 "movl %1, %%edi                 \n\t"
 503                 "shrl $9, %%esi                 \n\t"
 504                 "movw %%si, 4096(%%edi, %%eax, 2)\n\t"
 505
 506                 "addw %4, %%cx                  \n\t" //2*xalpha += s_xinc&0xFF
 507                 "adcl %3, %%ebx                 \n\t" //xx+= s_xinc>>8 + carry
 508                 "addl $1, %%eax                 \n\t"
 509                 "cmpl %2, %%eax                 \n\t"
 510                 " jb 1b                         \n\t"
 511
 512
 513                 :: "m" (src1), "m" (uvbuf1), "m" (dstw), "m" (s_xinc2>>16), "m" (s_xinc2&0xFFFF),
 514                 "r" (src2)
 515                 : "%eax", "%ebx", "%ecx", "%edi", "%esi"
 516                 );
 517 #ifdef HAVE_MMX2
 518         } //if MMX2 cant be used
 519 #endif
 520 #else
 521       for(i=0;i<dstw;i++){
 522           register unsigned int xx=xpos>>16;
 523           register unsigned int xalpha=(xpos&0xFFFF)>>9;
 524           uvbuf1[i]=(src1[xx]*(xalpha^127)+src1[xx+1]*xalpha);
 525           uvbuf1[i+2048]=(src2[xx]*(xalpha^127)+src2[xx+1]*xalpha);
 526           xpos+=s_xinc2;
 527       }
 528 #endif
 529     }
 530
 531
 532     // Note1: this code can be resticted to n*8 (or n*16) width lines to simplify optimization...
 533     // Re: Note1: ok n*4 for now
 534     // Note2: instead of using lookup tabs, mmx version could do the multiply...
 535     // Re: Note2: yep
 536     // Note3: maybe we should make separated 15/16, 24 and 32bpp version of this:
 537     // Re: done (32 & 16) and 16 has dithering :) but 16 is untested
 538 #ifdef HAVE_MMX
 539         //FIXME write lq version with less uv ...
 540         //FIXME reorder / optimize
 541         if(dstbpp == 32)
 542         {
 543                 asm volatile(
 544
 545 #define YSCALEYUV2RGB \
 546                 "pxor %%mm7, %%mm7              \n\t"\
 547                 "movd %6, %%mm6                 \n\t" /*yalpha1*/\
 548                 "punpcklwd %%mm6, %%mm6         \n\t"\
 549                 "punpcklwd %%mm6, %%mm6         \n\t"\
 550                 "movd %7, %%mm5                 \n\t" /*uvalpha1*/\
 551                 "punpcklwd %%mm5, %%mm5         \n\t"\
 552                 "punpcklwd %%mm5, %%mm5         \n\t"\
 553                 "xorl %%eax, %%eax              \n\t"\
 554                 "1:                             \n\t"\
 555                 "movq (%0, %%eax, 2), %%mm0     \n\t" /*buf0[eax]*/\
 556                 "movq (%1, %%eax, 2), %%mm1     \n\t" /*buf1[eax]*/\
 557                 "movq (%2, %%eax,2), %%mm2      \n\t" /* uvbuf0[eax]*/\
 558                 "movq (%3, %%eax,2), %%mm3      \n\t" /* uvbuf1[eax]*/\
 559                 "psubw %%mm1, %%mm0             \n\t" /* buf0[eax] - buf1[eax]*/\
 560                 "psubw %%mm3, %%mm2             \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
 561                 "pmulhw %%mm6, %%mm0            \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
 562                 "pmulhw %%mm5, %%mm2            \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
 563                 "psraw $7, %%mm1                \n\t" /* buf0[eax] - buf1[eax] >>7*/\
 564                 "movq 4096(%2, %%eax,2), %%mm4  \n\t" /* uvbuf0[eax+2048]*/\
 565                 "psraw $7, %%mm3                \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>7*/\
 566                 "paddw %%mm0, %%mm1             \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
 567                 "movq 4096(%3, %%eax,2), %%mm0  \n\t" /* uvbuf1[eax+2048]*/\
 568                 "paddw %%mm2, %%mm3             \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
 569                 "psubw %%mm0, %%mm4             \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
 570                 "psubw w10, %%mm1               \n\t" /* Y-16*/\
 571                 "psubw w80, %%mm3               \n\t" /* (U-128)*/\
 572                 "psllw $3, %%mm1                \n\t" /* (y-16)*8*/\
 573                 "psllw $3, %%mm3                \n\t" /*(U-128)8*/\
 574                 "pmulhw yCoeff, %%mm1           \n\t"\
 575 \
 576 \
 577                 "pmulhw %%mm5, %%mm4            \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
 578                 "movq %%mm3, %%mm2              \n\t" /* (U-128)8*/\
 579                 "pmulhw ubCoeff, %%mm3          \n\t"\
 580                 "psraw $7, %%mm0                \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>7*/\
 581                 "pmulhw ugCoeff, %%mm2          \n\t"\
 582                 "paddw %%mm4, %%mm0             \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
 583                 "psubw w80, %%mm0               \n\t" /* (V-128)*/\
 584                 "psllw $3, %%mm0                \n\t" /* (V-128)8*/\
 585 \
 586 \
 587                 "movq %%mm0, %%mm4              \n\t" /* (V-128)8*/\
 588                 "pmulhw vrCoeff, %%mm0          \n\t"\
 589                 "pmulhw vgCoeff, %%mm4          \n\t"\
 590                 "paddw %%mm1, %%mm3             \n\t" /* B*/\
 591                 "paddw %%mm1, %%mm0             \n\t" /* R*/\
 592                 "packuswb %%mm3, %%mm3          \n\t"\
 593 \
 594                 "packuswb %%mm0, %%mm0          \n\t"\
 595                 "paddw %%mm4, %%mm2             \n\t"\
 596                 "paddw %%mm2, %%mm1             \n\t" /* G*/\
 597 \
 598                 "packuswb %%mm1, %%mm1          \n\t"
 599
 600 YSCALEYUV2RGB
 601                 "punpcklbw %%mm1, %%mm3         \n\t" // BGBGBGBG
 602                 "punpcklbw %%mm7, %%mm0         \n\t" // R0R0R0R0
 603
 604                 "movq %%mm3, %%mm1              \n\t"
 605                 "punpcklwd %%mm0, %%mm3         \n\t" // BGR0BGR0
 606                 "punpckhwd %%mm0, %%mm1         \n\t" // BGR0BGR0
 607 #ifdef HAVE_MMX2
 608                 "movntq %%mm3, (%4, %%eax, 4)   \n\t"
 609                 "movntq %%mm1, 8(%4, %%eax, 4)  \n\t"
 610 #else
 611                 "movq %%mm3, (%4, %%eax, 4)     \n\t"
 612                 "movq %%mm1, 8(%4, %%eax, 4)    \n\t"
 613 #endif
 614                 "addl $4, %%eax                 \n\t"
 615                 "cmpl %5, %%eax                 \n\t"
 616                 " jb 1b                         \n\t"
 617
 618
 619                 :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 620                 "m" (yalpha1), "m" (uvalpha1)
 621                 : "%eax"
 622                 );
 623         }
 624         else if(dstbpp==24)
 625         {
 626                 asm volatile(
 627
 628 YSCALEYUV2RGB
 629
 630                                                         // lsb ... msb
 631                 "punpcklbw %%mm1, %%mm3         \n\t" // BGBGBGBG
 632                 "punpcklbw %%mm7, %%mm0         \n\t" // R0R0R0R0
 633
 634                 "movq %%mm3, %%mm1              \n\t"
 635                 "punpcklwd %%mm0, %%mm3         \n\t" // BGR0BGR0
 636                 "punpckhwd %%mm0, %%mm1         \n\t" // BGR0BGR0
 637
 638                 "movq %%mm3, %%mm2              \n\t" // BGR0BGR0
 639                 "psrlq $8, %%mm3                \n\t" // GR0BGR00
 640                 "pand bm00000111, %%mm2         \n\t" // BGR00000
 641                 "pand bm11111000, %%mm3         \n\t" // 000BGR00
 642                 "por %%mm2, %%mm3               \n\t" // BGRBGR00
 643                 "movq %%mm1, %%mm2              \n\t"
 644                 "psllq $48, %%mm1               \n\t" // 000000BG
 645                 "por %%mm1, %%mm3               \n\t" // BGRBGRBG
 646
 647                 "movq %%mm2, %%mm1              \n\t" // BGR0BGR0
 648                 "psrld $16, %%mm2               \n\t" // R000R000
 649                 "psrlq $24, %%mm1               \n\t" // 0BGR0000
 650                 "por %%mm2, %%mm1               \n\t" // RBGRR000
 651
 652                 "movl %4, %%ebx                 \n\t"
 653                 "addl %%eax, %%ebx              \n\t"
 654 #ifdef HAVE_MMX2
 655                 //FIXME Alignment
 656                 "movntq %%mm3, (%%ebx, %%eax, 2)\n\t"
 657                 "movntq %%mm1, 8(%%ebx, %%eax, 2)\n\t"
 658 #else
 659                 "movd %%mm3, (%%ebx, %%eax, 2)  \n\t"
 660                 "psrlq $32, %%mm3               \n\t"
 661                 "movd %%mm3, 4(%%ebx, %%eax, 2) \n\t"
 662                 "movd %%mm1, 8(%%ebx, %%eax, 2) \n\t"
 663 #endif
 664                 "addl $4, %%eax                 \n\t"
 665                 "cmpl %5, %%eax                 \n\t"
 666                 " jb 1b                         \n\t"
 667
 668                 :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
 669                 "m" (yalpha1), "m" (uvalpha1)
 670                 : "%eax", "%ebx"
 671                 );
 672         }
 673         else if(dstbpp==16)
 674         {
 675                 asm volatile(
 676
 677 YSCALEYUV2RGB
 678 #ifdef DITHER16BPP
 679                 "paddusb g16Dither, %%mm1       \n\t"
 680                 "paddusb b16Dither, %%mm0       \n\t"
 681                 "paddusb b16Dither, %%mm3       \n\t"
 682 #endif
 683                 "punpcklbw %%mm7, %%mm1         \n\t" // 0G0G0G0G
 684                 "punpcklbw %%mm7, %%mm3         \n\t" // 0B0B0B0B
 685                 "punpcklbw %%mm7, %%mm0         \n\t" // 0R0R0R0R
 686
 687                 "psrlw $3, %%mm3                \n\t"
 688                 "psllw $3, %%mm1                \n\t"
 689                 "psllw $8, %%mm0                \n\t"
 690                 "pand g16Mask, %%mm1            \n\t"
 691                 "pand r16Mask, %%mm0            \n\t"
 692
 693                 "por %%mm3, %%mm1               \n\t"
 694                 "por %%mm1, %%mm0               \n\t"
 695 #ifdef HAVE_MMX2
 696                 "movntq %%mm0, (%4, %%eax, 2)   \n\t"
 697 #else
 698                 "movq %%mm0, (%4, %%eax, 2)     \n\t"
 699 #endif
 700                 "addl $4, %%eax                 \n\t"
 701                 "cmpl %5, %%eax                 \n\t"
 702                 " jb 1b                         \n\t"
 703
 704                 :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
 705                 "m" (yalpha1), "m" (uvalpha1)
 706                 : "%eax"
 707                 );
 708         }
 709 #else
 710         if(dstbpp==32 || dstbpp==24)
 711         {
 712                 for(i=0;i<dstw;i++){
 713                         // vertical linear interpolation && yuv2rgb in a single step:
 714                         int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
 715                         int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
 716                         int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
 717                         dest[0]=clip_table[((Y + yuvtab_3343[U]) >>13)];
 718                         dest[1]=clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)];
 719                         dest[2]=clip_table[((Y + yuvtab_40cf[V]) >>13)];
 720                         dest+=dstbpp>>3;
 721                 }
 722         }
 723         else if(dstbpp==16)
 724         {
 725                 for(i=0;i<dstw;i++){
 726                         // vertical linear interpolation && yuv2rgb in a single step:
 727                         int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
 728                         int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
 729                         int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
 730
 731                         ((uint16_t*)dest)[0] =
 732                                 (clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
 733                                 (clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<3)&0x07E0 |
 734                                 (clip_table[((Y + yuvtab_40cf[V]) >>13)]<<8)&0xF800;
 735                         dest+=2;
 736                 }
 737         }
 738         else if(dstbpp==15) //15bit FIXME how do i figure out if its 15 or 16?
 739         {
 740                 for(i=0;i<dstw;i++){
 741                         // vertical linear interpolation && yuv2rgb in a single step:
 742                         int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
 743                         int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
 744                         int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
 745
 746                         ((uint16_t*)dest)[0] =
 747                                 (clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
 748                                 (clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<2)&0x03E0 |
 749                                 (clip_table[((Y + yuvtab_40cf[V]) >>13)]<<7)&0x7C00;
 750                         dest+=2;
 751                 }
 752         }
 753 #endif
 754
 755         b16Dither= b16Dither1;
 756         b16Dither1= b16Dither2;
 757         b16Dither2= b16Dither;
 758
 759         g16Dither= g16Dither1;
 760         g16Dither1= g16Dither2;
 761         g16Dither2= g16Dither;
 762   }
 763
 764 #ifdef HAVE_3DNOW
 765         asm volatile("femms");
 766 #elif defined (HAVE_MMX)
 767         asm volatile("emms");
 768 #endif
 769 }
 770
 771
 772 void SwScale_Init(){
 773     // generating tables:
 774     int i;
 775     for(i=0;i<256;i++){
 776         clip_table[i]=0;
 777         clip_table[i+256]=i;
 778         clip_table[i+512]=255;
 779         yuvtab_2568[i]=(0x2568*(i-16))+(256<<13);
 780         yuvtab_3343[i]=0x3343*(i-128);
 781         yuvtab_0c92[i]=-0x0c92*(i-128);
 782         yuvtab_1a1e[i]=-0x1a1e*(i-128);
 783         yuvtab_40cf[i]=0x40cf*(i-128);
 784     }
 785
 786 }