wrlib/x86_specific.c

   1 /* x86_convert.c - convert RImage to XImage with x86 optimizations
   2  *
   3  * Raster graphics library
   4  *
   5  * Copyright (c) 2000-2003 Alfredo K. Kojima
   6  *
   7  *  This library is free software; you can redistribute it and/or
   8  *  modify it under the terms of the GNU Library General Public
   9  *  License as published by the Free Software Foundation; either
  10  *  version 2 of the License, or (at your option) any later version.
  11  *
  12  *  This library is distributed in the hope that it will be useful,
  13  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  *  Library General Public License for more details.
  16  *
  17  *  You should have received a copy of the GNU Library General Public
  18  *  License along with this library; if not, write to the Free
  19  *  Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21
  22 #include <config.h>
  23
  24 #ifdef ASM_X86
  25
  26
  27 #ifdef ASM_X86_MMX
  28
  29 int
  30 x86_check_mmx()
  31 {
  32     static int result = -1;
  33
  34     if (result >= 0)
  35         return result;
  36
  37     result = 0;
  38
  39     asm volatile
  40         ("pushal                \n\t" // please dont forget this in any asm
  41          "pushfl                \n\t" // check whether cpuid supported
  42          "pop %%eax             \n\t"
  43          "movl %%eax, %%ebx     \n\t"
  44          "xorl $(1<<21), %%eax  \n\t"
  45          "pushl %%eax           \n\t"
  46          "popfl                 \n\t"
  47          "pushfl                \n\t"
  48          "popl %%eax            \n\t"
  49          "xorl %%ebx, %%eax     \n\t"
  50          "andl $(1<<21), %%eax  \n\t"
  51          "jz .NotPentium        \n\t"
  52          "xorl %%eax, %%eax     \n\t" // no eax effect because of the movl below
  53                                       // except reseting flags. is it needed?
  54          "movl $1, %%eax        \n\t"
  55          "cpuid                 \n\t"
  56          "test $(1<<23), %%edx  \n\t"
  57          "jz .NotMMX            \n\t"
  58
  59          "popal                 \n\t" // popal needed because the address of
  60          "movl $1, %0           \n\t" // variable %0 may be kept in a register
  61          "jmp .noPop            \n"
  62
  63 ".NotMMX:                       \n"
  64 ".NotPentium:                   \n\t"
  65          "popal                 \n"
  66 ".noPop:                        \n\t"
  67
  68          : "=m" (result));
  69
  70     return result;
  71 }
  72
  73
  74 /*
  75  * TODO:
  76  *              32/8    24/8    32/16   24/16   32/24   24/24
  77  * PPlain       YES     YES
  78  * MMX                          DONE
  79  *
  80  *
  81  * - try to align stack (local variable space) into quadword boundary
  82  */
  83 void
  84 x86_mmx_TrueColor_32_to_16(unsigned char *image,
  85                            unsigned short *ximage,
  86                            short *err,
  87                            short *nerr,
  88                            short *rtable,
  89                            short *gtable,
  90                            short *btable,
  91                            int dr,
  92                            int dg,
  93                            int db,
  94                            unsigned int roffs,
  95                            unsigned int goffs,
  96                            unsigned int boffs,
  97                            int width,
  98                            int height,
  99                            int line_offset)
 100 {
 101     long long rrggbbaa;
 102     long long pixel;
 103     short *tmp_err;
 104     short *tmp_nerr;
 105     int x;
 106
 107     asm volatile
 108         (
 109          "pushal                        \n\t"
 110
 111          // pack dr, dg and db into mm6
 112          "movl  %7, %%eax               \n\t"
 113          "movl  %8, %%ebx               \n\t"
 114          "movl  %9, %%ecx               \n\t"
 115          "movw  %%ax, %16               \n\t"
 116          "movw  %%bx, %17               \n\t"
 117          "movw  %%cx, %18               \n\t"
 118          "movw  $0,  %19                \n\t"
 119
 120          "movq  %16, %%mm6              \n\t" // dr dg db 0
 121
 122          // pack 4|4|4|4 into mm7, for shifting (/16)
 123          "movl $0x00040004, %16         \n\t"
 124          "movl $0x00040004, %18         \n\t"
 125          "movq %16, %%mm7               \n\t"
 126
 127          // store constant values for using with mmx when dithering
 128          "movl $0x00070007, %16         \n\t"
 129          "movl $0x00070007, %18         \n\t"
 130          "movq %16, %%mm5               \n\t"
 131
 132          "movl $0x00050005, %16         \n\t"
 133          "movl $0x00050005, %18         \n\t"
 134          "movq %16, %%mm4               \n\t"
 135
 136          "movl $0x00030003, %16         \n\t"
 137          "movl $0x00030003, %18         \n\t"
 138          "movq %16, %%mm3               \n\t"
 139
 140          // process 1 pixel / cycle, each component treated as 16bit
 141          "movl %0, %%esi                \n"   // esi = image->data
 142
 143 ".LoopYa:                               \n\t"
 144          "movl %13, %%eax               \n\t"
 145          "movl %%eax, %26               \n\t" // x = width
 146
 147          "movl %14, %%eax               \n\t"
 148          "decl %%eax                    \n\t" // y--
 149          "movl %%eax, %14               \n\t"
 150          "js .Enda                      \n\t" // if y < 0, goto end
 151          "andl $1, %%eax                \n\t"
 152          "jz .LoopY_1a                  \n"   // if (y&1) goto LoopY_1
 153
 154 ".LoopY_0a:                             \n\t"
 155
 156          "movl %2, %%ebx                \n\t" // ebx = err
 157          "movl %%ebx, %25               \n\t" // [-36] = err
 158          "movl %3, %%eax                \n\t" //
 159          "movl %%eax, %24               \n\t" // [-32] = nerr
 160
 161          "jmp .LoopXa                   \n"
 162
 163 ".LoopY_1a:                             \n\t"
 164
 165          "movl %3, %%ebx                \n\t" // ebx = nerr
 166          "movl %%ebx, %25               \n\t" // [-36] = nerr
 167          "movl %2, %%eax                \n\t" //
 168          "movl %%eax, %24               \n\t" // [-32] = eerr
 169
 170          ".align 16                     \n"
 171 ".LoopXa:                               \n\t"
 172
 173          // calculate errors and pixel components
 174
 175          // depend on ebx, esi, mm6
 176          "movq (%%ebx), %%mm1           \n\t" // mm1 = error[0..3]
 177          "punpcklbw (%%esi), %%mm0      \n\t" // mm0 = image->data[0..3]
 178          "psrlw $8, %%mm0               \n\t" // fixup mm0
 179          "paddusb %%mm1, %%mm0          \n\t" // mm0 = mm0 + mm1 (sat. to 255)
 180          "movq %%mm0, %20               \n\t" // save the pixel
 181
 182          "movzwl %20, %%ecx             \n\t" // ecx = pixel.red
 183          "movl %4, %%edi                \n\t" // edi = rtable
 184          //agi
 185          "leal (%%edi, %%ecx, 2), %%eax \n\t" // eax = &rtable[pixel.red]
 186          // agi
 187          "movw (%%eax), %%dx            \n\t" // dx = rtable[pixel.red]
 188          "movw %%dx, %16                \n\t" // save rr
 189
 190          "movzwl %21, %%ecx             \n\t" // ecx = pixel.green
 191          "movl %5, %%edi                \n\t" // edi = gtable
 192          //agi
 193          "leal (%%edi, %%ecx, 2), %%eax \n\t" // eax = &gtable[pixel.green]
 194          //agi
 195          "movw (%%eax), %%dx            \n\t" // dx = gtable[pixel.green]
 196          "movw %%dx, %17                \n\t" // save gg
 197
 198          "movzwl %22, %%ecx             \n\t" // ecx = pixel.blue
 199          "movl %6, %%edi                \n\t" // ebx = btable
 200          //agi
 201          "leal (%%edi, %%ecx, 2), %%eax \n\t" // eax = &btable[pixel.blue]
 202          //agi
 203          "movw (%%eax), %%dx            \n\t" // dx = btable[pixel.blue]
 204          "movw %%dx, %18                \n\t" // save bb
 205
 206          "movw $0, %19                  \n\t" // save dummy aa
 207
 208          "movq %16, %%mm1               \n\t" // load mm1 with rrggbbaa
 209          "pmullw %%mm6, %%mm1           \n\t" // mm1 = rr*dr|...
 210          "psubsw %%mm1, %%mm0           \n\t" // error = pixel - mm1
 211
 212
 213          // distribute the error
 214
 215          // depend on mm0, mm7, mm3, mm4, mm5
 216
 217          "movl %25, %%ebx               \n\t"
 218
 219          "movq %%mm0, %%mm1             \n\t"
 220          "pmullw %%mm5, %%mm1           \n\t" // mm1 = mm1*7
 221          "psrlw %%mm7, %%mm1            \n\t" // mm1 = mm1/16
 222          "paddw 8(%%ebx), %%mm1         \n\t"
 223          "movq %%mm1, 8(%%ebx)          \n\t" // err[x+1,y] = rer*7/16
 224
 225
 226          "movl %24, %%ebx               \n\t"
 227
 228          "movq %%mm0, %%mm1             \n\t"
 229          "pmullw %%mm4, %%mm1           \n\t" // mm1 = mm1*5
 230          "psrlw %%mm7, %%mm1            \n\t" // mm1 = mm1/16
 231          "paddw -8(%%ebx), %%mm1                \n\t"
 232          "movq %%mm1, -8(%%ebx)         \n\t" // err[x-1,y+1] += rer*3/16
 233
 234          "movq %%mm0, %%mm1             \n\t"
 235          "pmullw %%mm3, %%mm1           \n\t" // mm1 = mm1*3
 236          "psrlw %%mm7, %%mm1            \n\t" // mm1 = mm1/16
 237          "paddw 8(%%ebx), %%mm1         \n\t"
 238          "movq %%mm1, (%%ebx)           \n\t" // err[x,y+1] += rer*5/16
 239
 240          "psrlw %%mm7, %%mm0            \n\t" // mm0 = mm0/16
 241          "movq %%mm0, 8(%%ebx)          \n\t" // err[x+1,y+1] = rer/16
 242
 243
 244          // calculate final pixel value and store
 245          "movl %10, %%ecx               \n\t"
 246          "movw %16, %%ax                \n\t"
 247          "shlw %%cl, %%ax               \n\t" //NP* ax = r<<roffs
 248
 249          "movl %11, %%ecx               \n\t"
 250          "movw %17, %%bx                \n\t"
 251          "shlw %%cl, %%bx               \n\t" //NP*
 252          "orw %%bx, %%ax                \n\t"
 253
 254          "movl %12, %%ecx               \n\t"
 255          "movw %18, %%bx                \n\t"
 256          "shlw %%cl, %%bx               \n\t" //NP*
 257          "orw %%bx, %%ax                \n\t"
 258
 259          "movl %1, %%edx                \n\t"
 260          "movw %%ax, (%%edx)            \n\t"
 261          "addl $2, %%edx                \n\t" // increment ximage
 262          "movl %%edx, %1                \n\t"
 263
 264          // prepare for next iteration on X
 265
 266          "addl $8, %24                  \n\t" // nerr += 8
 267
 268          "movl %25, %%ebx               \n\t"
 269          "addl $8, %%ebx                \n\t"
 270          "movl %%ebx, %25               \n\t" // ebx = err += 8
 271
 272
 273          // Note: in the last pixel, this would cause an invalid memory access
 274          // because, punpcklbw is used (which reads 8 bytes) and the last
 275          // pixel is only 4 bytes. This is no problem because the image data
 276          // was allocated with extra 4 bytes when created.
 277          "addl $4, %%esi                        \n\t" // image->data += 4
 278
 279
 280          "decl %26                      \n\t" // x--
 281          "jnz .LoopXa                   \n\t" // if x>0, goto .LoopX
 282
 283
 284          // depend on edx
 285          "addl %15, %%edx               \n\t" // add extra offset to ximage
 286          "movl %%edx, %1                \n\t"
 287
 288
 289          "jmp .LoopYa                   \n"
 290
 291 ".Enda:                                 \n\t" // THE END
 292          "emms                          \n\t"
 293          "popal                         \n\t"
 294          :
 295          :
 296          "m" (image),                      // %0
 297          "m" (ximage),                     // %1
 298          "m" (err),                        // %2
 299          "m" (nerr),                       // %3
 300          "m" (rtable),                     // %4
 301          "m" (gtable),                     // %5
 302          "m" (btable),                     // %6
 303          "m" (dr),                         // %7
 304          "m" (dg),                         // %8
 305          "m" (db),                         // %9
 306          "m" (roffs),                      // %10
 307          "m" (goffs),                      // %11
 308          "m" (boffs),                      // %12
 309          "m" (width),                      // %13
 310          "m" (height),                     // %14
 311          "m" (line_offset),                // %15
 312          "m" (rrggbbaa),                   // %16 (access to rr)
 313          "m" ((*((short*)(&rrggbbaa)+1))), // %17 (access to gg)
 314          "m" ((*((short*)(&rrggbbaa)+2))), // %18 (access to bb)
 315          "m" ((*((short*)(&rrggbbaa)+3))), // %19 (access to aa)
 316          "m" (pixel),                      // %20 (access to pixel.r)
 317          "m" ((*((short*)(&pixel)+1))),    // %21 (access to pixel.g)
 318          "m" ((*((short*)(&pixel)+2))),    // %22 (access to pixel.b)
 319          "m" ((*((short*)(&pixel)+3))),    // %23 (access to pixel.a)
 320          "m" (tmp_err),                    // %24
 321          "m" (tmp_nerr),                   // %25
 322          "m" (x)                           // %26
 323          );
 324 }
 325
 326
 327 void
 328 x86_mmx_TrueColor_24_to_16(unsigned char *image,
 329                            unsigned short *ximage,
 330                            short *err,
 331                            short *nerr,
 332                            short *rtable,
 333                            short *gtable,
 334                            short *btable,
 335                            int dr,
 336                            int dg,
 337                            int db,
 338                            unsigned int roffs,
 339                            unsigned int goffs,
 340                            unsigned int boffs,
 341                            int width,
 342                            int height,
 343                            int line_offset)
 344 {
 345     long long rrggbbaa;
 346     long long pixel;
 347
 348     short *tmp_err;
 349     short *tmp_nerr;
 350
 351     int x;
 352     int w1;
 353     int w2;
 354
 355     asm volatile
 356         (
 357          "pushal                        \n\t"
 358
 359          "movl %13, %%eax               \n\t" // eax = width
 360          "movl %%eax, %%ebx             \n\t"
 361          "shrl $2, %%eax                \n\t"
 362          "movl %%eax, %27               \n\t" // w1 = width / 4
 363          "andl $3, %%ebx                \n\t"
 364          "movl %%ebx, %28               \n"   // w2 = width %% 4
 365
 366
 367 ".LoopYc:                               \n\t"
 368          "movl %13, %%eax               \n\t"
 369          "movl %%eax, %26               \n\t" // x = width
 370
 371          "decl %14                      \n\t" // height--
 372          "js .Endc                      \n\t" // if height < 0 then end
 373
 374          "movl %14, %%eax               \n\t"
 375          "decl %%eax                    \n\t" // y--
 376          "movl %%eax, %14               \n\t"
 377          "js .Endc                      \n\t" // if y < 0, goto end
 378          "andl $1, %%eax                \n\t"
 379          "jz .LoopY_1c                  \n"   // if (y&1) goto LoopY_1
 380
 381 ".LoopY_0c:                             \n\t"
 382
 383          "movl %2, %%ebx                \n\t" // ebx = err
 384          "movl %%ebx, %25               \n\t" // [-36] = err
 385          "movl %3, %%eax                \n\t" //
 386          "movl %%eax, %24               \n\t" // [-32] = nerr
 387
 388          "jmp .LoopX_1c                 \n"
 389
 390 ".LoopY_1c:                             \n\t"
 391
 392          "movl %3, %%ebx                \n\t" // ebx = nerr
 393          "movl %%ebx, %25               \n\t" // [-36] = nerr
 394          "movl %2, %%eax                \n\t" //
 395          "movl %%eax, %24               \n\t" // [-32] = eerr
 396
 397          ".align 16                     \n\t"
 398
 399          "movl %%eax, %26               \n"   // x = w1
 400 ".LoopX_1c:                             \n\t"
 401          "decl %26                      \n\t" // x--
 402          "js .Xend1_c                   \n\t" // if x < 0 then end
 403
 404          // do conversion of 4 pixels
 405          "movq %2, %%mm0                \n\t" // mm0 = err
 406
 407
 408
 409
 410          "jmp .LoopX_1c                 \n"
 411 ".Xend1_c:                              \n\t"
 412
 413          "movl %28, %%eax               \n\t"
 414          "movl %%eax, %26               \n"   // x = w2
 415 ".LoopX_2c:                             \n\t"
 416          "decl %26                      \n\t" // x--
 417          "js .Xend2_c                   \n\t" //
 418          // do conversion
 419          "jmp .LoopX_2c                 \n"
 420 ".Xend2_c:                              \n\t"
 421
 422          "movl %27, %%eax               \n\t"
 423          "jmp .LoopYc                   \n"
 424
 425 ".Endc:                                 \n\t" // THE END
 426          "emms                          \n\t"
 427          "popal                         \n\t"
 428          :
 429          :
 430          "m" (image),                      // %0
 431          "m" (ximage),                     // %1
 432          "m" (err),                        // %2
 433          "m" (nerr),                       // %3
 434          "m" (rtable),                     // %4
 435          "m" (gtable),                     // %5
 436          "m" (btable),                     // %6
 437          "m" (dr),                         // %7
 438          "m" (dg),                         // %8
 439          "m" (db),                         // %9
 440          "m" (roffs),                      // %10
 441          "m" (goffs),                      // %11
 442          "m" (boffs),                      // %12
 443          "m" (width),                      // %13
 444          "m" (height),                     // %14
 445          "m" (line_offset),                // %15
 446          "m" (rrggbbaa),                   // %16 (access to rr)
 447          "m" ((*((short*)(&rrggbbaa)+1))), // %17 (access to gg)
 448          "m" ((*((short*)(&rrggbbaa)+2))), // %18 (access to bb)
 449          "m" ((*((short*)(&rrggbbaa)+3))), // %19 (access to aa)
 450          "m" (pixel),                      // %20 (access to pixel.r)
 451          "m" ((*((short*)(&pixel)+1))),    // %21 (access to pixel.g)
 452          "m" ((*((short*)(&pixel)+2))),    // %22 (access to pixel.b)
 453          "m" ((*((short*)(&pixel)+3))),    // %23 (access to pixel.a)
 454          "m" (tmp_err),                    // %24
 455          "m" (tmp_nerr),                   // %25
 456          "m" (x),                          // %26
 457          "m" (w1),                         // %27
 458          "m" (w2)                          // %28
 459          );
 460 }
 461
 462
 463
 464 #endif /* ASM_X86_MMX */
 465
 466
 467
 468 void
 469 x86_PseudoColor_32_to_8(unsigned char *image,
 470                         unsigned char *ximage,
 471                         char *err,
 472                         char *nerr,
 473                         short *ctable,
 474                         int dr,
 475                         int dg,
 476                         int db,
 477                         unsigned long *pixels,
 478                         int cpc,
 479                         int width,
 480                         int height,
 481                         int bytesPerPixel,
 482                         int line_offset)
 483 {
 484     int x;
 485     int cpcpc;
 486
 487     int rr;
 488     int gg;
 489     int bb;
 490
 491     char *tmp_err;
 492     char *tmp_nerr;
 493
 494     char ndr; // aparently not used
 495     char ndg; // aparently not used
 496     char ndb; // aparently not used
 497
 498     asm volatile
 499         (
 500          "pushal                        \n\t"
 501
 502          "movl %9, %%eax                \n\t"
 503          "mulb %9                       \n\t"
 504          "movl %%eax, %15               \n\t" // cpcpc = cpc*cpc
 505
 506          // eax will always be <= 0xffff
 507
 508          // process 1 pixel / cycle, each component treated as 16bit
 509          "movl %0, %%esi                \n"   // esi = image->data
 510
 511 ".LoopYb:                               \n\t"
 512          "movl %10, %%ecx               \n\t"
 513          "movl %%ecx, %14               \n\t" // x = width
 514
 515          "movl %11, %%ecx               \n\t"
 516          "decl %%ecx                    \n\t" // y--
 517          "movl %%ecx, %11               \n\t"
 518          "js .Endb                      \n\t" // if y < 0, goto end
 519          "andl $1, %%ecx                \n\t"
 520          "jz .LoopY_1b                  \n"   // if (y&1) goto LoopY_1
 521
 522 ".LoopY_0b:                             \n\t"
 523
 524          "movl %2, %%ebx                \n\t" // ebx = err
 525 //useless "movl %%ebx, %20              \n\t" // [-36] = err
 526          "movl %3, %%ecx                \n\t" //
 527          "movl %%ecx, %19               \n\t" // [-32] = nerr
 528
 529          "movl $0, (%%ecx)              \n\t" // init error of nerr[0] to 0
 530
 531          "jmp .LoopXb                   \n"
 532
 533 ".LoopY_1b:                             \n\t"
 534
 535          "movl %3, %%ebx                \n\t" // ebx = nerr
 536 //useless "movl %%ebx, %20              \n\t" // [-36] = nerr
 537          "movl %2, %%ecx                \n\t" //
 538          "movl %%ecx, %19               \n\t" // [-32] = err
 539
 540          "movl $0, (%%ecx)              \n\t" // init error of nerr[0] to 0
 541
 542
 543          ".align 16                     \n"
 544 ".LoopXb:                               \n\t"
 545
 546
 547          "movl %4, %%edi                \n\t" // edi = ctable
 548          "xorl %%edx, %%edx             \n\t" // zero the upper word on edx
 549
 550          // RED
 551
 552          // depends on ebx==err, esi==image->data, edi
 553          "movzbw (%%esi), %%dx          \n\t" // dx = image->data[0]
 554          "movsbw (%%ebx), %%ax          \n\t" // ax = error[0]
 555          "addw %%ax, %%dx               \n\t" // pixel.red = data[0] + error[0]
 556
 557          "testb %%dh, %%dh              \n\t" // test if pixel.red < 0 or > 255
 558          "jz .OKRb                      \n\t" // 0 <= pixel.red <= 255
 559          "js .NEGRb                     \n\t" // pixel.red < 0
 560          "movw $0xff, %%dx              \n\t" // pixel.red > 255
 561          "jmp .OKRb                     \n"
 562 ".NEGRb:                                \n\t"
 563          "xorw %%dx, %%dx                       \n"
 564 ".OKRb:                                 \n\t"
 565          //partial reg
 566          "leal (%%edi, %%edx, 2), %%ecx \n\t" // ecx = &ctable[pixel.red]
 567          //agi
 568          "movl (%%ecx), %%eax           \n\t" // ax = ctable[pixel.red]
 569          "movw %%ax, %16                \n\t" // save rr
 570
 571          "mulb %5                       \n\t" // ax = rr*dr
 572          "subw %%ax, %%dx               \n\t" // rer = dx = dx - rr*dr
 573
 574          "movswl %%dx, %%eax            \n\t" // save rer
 575
 576          // distribute error
 577          "leal (, %%eax, 8), %%ecx      \n\t"
 578          "subw %%dx, %%cx               \n\t" // cx = rer * 7
 579          "sarw $4, %%cx                 \n\t" // cx = rer * 7 / 16
 580          "addb %%cl, 4(%%ebx)           \n\t" // err[x+1] += rer * 7 / 16
 581
 582          "movl %19, %%ecx               \n\t" // ecx = nerr
 583
 584          "leaw (%%eax, %%eax, 4), %%dx  \n\t" // dx = rer * 5
 585          "sarw $4, %%dx                 \n\t" // dx = rer * 5 / 16
 586          "addb %%dl, (%%ecx)            \n\t" // nerr[x] += rer * 5 / 16
 587
 588          "leaw (%%eax, %%eax, 2), %%dx  \n\t" // dx = rer * 3
 589          "sarw $4, %%dx                 \n\t" // dx = rer * 3 / 16
 590          "addb %%dl, -4(%%ecx)          \n\t" // nerr[x-1] += rer * 3 / 16
 591
 592          "sarw $4, %%ax                 \n\t" // ax = rer / 16
 593          "movb %%al, 4(%%ecx)           \n\t" // nerr[x+1] = rer / 16
 594
 595
 596          // GREEN
 597
 598          // depends on ebx, esi, edi
 599          "movzbw 1(%%esi), %%dx         \n\t" // dx = image->data[1]
 600          "movsbw 1(%%ebx), %%ax         \n\t" // ax = error[1]
 601          "addw %%ax, %%dx               \n\t" // pixel.grn = data[1] + error[1]
 602
 603          "testb %%dh, %%dh              \n\t" // test if pixel.grn < 0 or > 255
 604          "jz .OKGb                      \n\t" // 0 <= pixel.grn <= 255
 605          "js .NEGGb                     \n\t" // pixel.grn < 0
 606          "movw $0xff, %%dx              \n\t" // pixel.grn > 255
 607          "jmp .OKGb                     \n"
 608 ".NEGGb:                                \n\t"
 609          "xorw %%dx, %%dx               \n"
 610 ".OKGb:                                 \n\t"
 611          // partial reg
 612          "leal (%%edi, %%edx, 2), %%ecx \n\t" // ecx = &ctable[pixel.grn]
 613          //agi
 614          "movw (%%ecx), %%ax            \n\t" // ax = ctable[pixel.grn]
 615          "movw %%ax, %17                \n\t" // save gg
 616
 617          "mulb %6                       \n\t" // ax = gg*dg
 618          "subw %%ax, %%dx               \n\t" // ger = dx = dx - gg*dg
 619
 620          "movswl %%dx, %%eax            \n\t" // save ger
 621
 622          // distribute error
 623
 624          "leal (, %%eax, 8), %%ecx      \n\t"
 625          "subw %%dx, %%cx               \n\t" // cx = ger * 7
 626          "sarw $4, %%cx                 \n\t" // cx = ger * 7 / 16
 627          "addb %%cl, 5(%%ebx)           \n\t" // err[x+1] += ger * 7 / 16
 628
 629          "movl %19, %%ecx               \n\t" // ecx = nerr
 630
 631          "leaw (%%eax, %%eax, 4), %%dx  \n\t" // dx = ger * 5
 632          "sarw $4, %%dx                 \n\t" // dx = ger * 5 / 16
 633          "addb %%dl, 1(%%ecx)           \n\t" // nerr[x] += ger * 5 / 16
 634
 635          "leaw (%%eax, %%eax, 2), %%dx  \n\t" // dx = ger * 3
 636          "sarw $4, %%dx                 \n\t" // dx = ger * 3 / 16
 637          "addb %%dl, -3(%%ecx)          \n\t" // nerr[x-1] += ger * 3 / 16
 638
 639          "sarw $4, %%ax                 \n\t" // ax = ger / 16
 640          "movb %%al, 5(%%ecx)           \n\t" // nerr[x+1] = ger / 16
 641
 642
 643          // BLUE
 644
 645          // depends on ebx, esi
 646          "movzbw 2(%%esi), %%dx         \n\t" // dx = image->data[2]
 647          "movsbw 2(%%ebx), %%ax         \n\t" // ax = error[2]
 648          "addw %%ax, %%dx               \n\t" // pixel.grn = data[2] + error[2]
 649
 650          "testb %%dh, %%dh              \n\t" // test if pixel.blu < 0 or > 255
 651          "jz .OKBb                      \n\t" // 0 <= pixel.blu <= 255
 652          "js .NEGBb                     \n\t" // pixel.blu < 0
 653          "movw $0xff, %%dx              \n\t" // pixel.blu > 255
 654          "jmp .OKBb                     \n"
 655 ".NEGBb:                                \n\t"
 656          "xorw %%dx, %%dx                       \n"
 657 ".OKBb:                                 \n\t"
 658          //partial reg
 659          "leal (%%edi, %%edx, 2), %%ecx \n\t" // ecx = &ctable[pixel.blu]
 660          //agi
 661          "movw (%%ecx), %%ax            \n\t" // ax = ctable[pixel.blu]
 662          "movw %%ax, %18                \n\t" // save bb
 663
 664          "mulb %7                       \n\t" // ax = bb*db
 665          "subw %%ax, %%dx               \n\t" // ber = dx = dx - bb*db
 666          "movswl %%dx, %%eax            \n\t" // save ber
 667
 668          // distribute error
 669          "leal (, %%eax, 8), %%ecx      \n\t"
 670          "subw %%dx, %%cx               \n\t" // cx = ber * 7
 671          "sarw $4, %%cx                 \n\t" // cx = ber * 7 / 16
 672          "addb %%cl, 6(%%ebx)           \n\t" // err[x+1] += ber * 7 / 16
 673
 674          "movl %19, %%ecx               \n\t" // ecx = nerr
 675
 676          "leaw (%%eax, %%eax, 4), %%dx  \n\t" // dx = ber * 5
 677          "sarw $4, %%dx                 \n\t" // dx = ber * 5 / 16
 678          "addb %%dl, 2(%%ecx)           \n\t" // nerr[x] += ber * 5 / 16
 679
 680          "leaw (%%eax, %%eax, 2), %%dx  \n\t" // dx = ber * 3
 681          "sarw $4, %%dx                 \n\t" // dx = ber * 3 / 16
 682          "addb %%dl, -4(%%ecx)          \n\t" // nerr[x-1] += ber * 3 / 16
 683
 684          "sarw $4, %%ax                 \n\t" // ax = ber / 16
 685          "movb %%al, 6(%%ecx)           \n\t" // nerr[x+1] = ber / 16
 686
 687          "andl $0xffff, %%eax           \n\t"
 688          // depends on eax & 0xffff0000 == 0
 689          // calculate the index of the value of the pixel
 690          "movw %16, %%ax                \n\t" // ax = rr
 691          "mulb %15                      \n\t" // ax = cpcpc*rr
 692          "movw %%ax, %%cx               \n\t"
 693          "movw %17, %%ax                \n\t" // ax = gg
 694          "mulb %9                       \n\t" // ax = cpc*gg
 695          "addw %%cx, %%ax               \n\t" // ax = cpc*gg + cpcpc*rr
 696          "addw %18, %%ax                \n\t" // ax = cpcpc*rr + cpc*gg + bb
 697
 698          "movl %8, %%ecx                \n\t"
 699          //agi
 700          "leal (%%ecx, %%eax, 4), %%edx \n\t"
 701          //agi
 702          "movb (%%edx), %%cl            \n\t" // cl = pixels[ax]
 703
 704          // store the pixel
 705          "movl %1, %%eax                \n\t"
 706          "movb %%cl, (%%eax)            \n\t" // *ximage = cl
 707          "incl %1                       \n\t" // ximage++
 708
 709          // prepare for next iteration on X
 710
 711          "addl $4, %19                  \n\t" // nerr += 4
 712          "addl $4, %%ebx                \n\t" // err += 4
 713
 714          "addl %12, %%esi               \n\t" // image->data += bpp
 715
 716          "decl %14                      \n\t" // x--
 717          "jnz .LoopXb                   \n\t" // if x>0, goto .LoopX
 718
 719
 720          "movl %13, %%eax               \n\t"
 721          "addl %%eax, %1                \n\t" // add extra offset to ximage
 722
 723          "jmp .LoopYb                   \n"
 724
 725 ".Endb:                                 \n\t"
 726          "emms                          \n\t"
 727          "popal                         \n\t"
 728          :
 729          :
 730          "m" (image),         // %0
 731          "m" (ximage),        // %1
 732          "m" (err),           // %2
 733          "m" (nerr),          // %3
 734          "m" (ctable),        // %4
 735          "m" (dr),            // %5
 736          "m" (dg),            // %6
 737          "m" (db),            // %7
 738          "m" (pixels),        // %8
 739          "m" (cpc),           // %9
 740          "m" (width),         // %10
 741          "m" (height),        // %11
 742          "m" (bytesPerPixel), // %12
 743          "m" (line_offset),   // %13
 744          "m" (x),             // %14
 745          "m" (cpcpc),         // %15
 746          "m" (rr),            // %16
 747          "m" (gg),            // %17
 748          "m" (bb),            // %18
 749          "m" (tmp_err),       // %19
 750          "m" (tmp_nerr),      // %20
 751          "m" (ndr),           // %21
 752          "m" (ndg),           // %22
 753          "m" (ndb)            // %23
 754          );
 755 }
 756
 757 #endif /* ASM_X86 */
 758