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                            unsigned short *rtable,
  89                            unsigned short *gtable,
  90                            unsigned 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     union {
 102         long long rrggbbaa;
 103         struct {
 104             short int rr, gg, bb, aa;
 105         } words;
 106     } rrggbbaa;
 107
 108     union {
 109         long long pixel;
 110         struct {
 111             short int rr, gg, bb, aa;
 112         } words;
 113     } pixel;
 114
 115     short *tmp_err;
 116     short *tmp_nerr;
 117     int x;
 118
 119     asm volatile
 120         (
 121          "pushl %%ebx                   \n\t"
 122
 123          // pack dr, dg and db into mm6
 124          "movl  %7, %%eax               \n\t"
 125          "movl  %8, %%ebx               \n\t"
 126          "movl  %9, %%ecx               \n\t"
 127          "movw  %%ax, %16               \n\t"
 128          "movw  %%bx, %17               \n\t"
 129          "movw  %%cx, %18               \n\t"
 130          "movw  $0,  %19                \n\t"
 131
 132          "movq  %16, %%mm6              \n\t" // dr dg db 0
 133
 134          // pack 4|4|4|4 into mm7, for shifting (/16)
 135          "movl $0x00040004, %16         \n\t"
 136          "movl $0x00040004, %18         \n\t"
 137          "movq %16, %%mm7               \n\t"
 138
 139          // store constant values for using with mmx when dithering
 140          "movl $0x00070007, %16         \n\t"
 141          "movl $0x00070007, %18         \n\t"
 142          "movq %16, %%mm5               \n\t"
 143
 144          "movl $0x00050005, %16         \n\t"
 145          "movl $0x00050005, %18         \n\t"
 146          "movq %16, %%mm4               \n\t"
 147
 148          "movl $0x00030003, %16         \n\t"
 149          "movl $0x00030003, %18         \n\t"
 150          "movq %16, %%mm3               \n\t"
 151
 152          // process 1 pixel / cycle, each component treated as 16bit
 153          "movl %0, %%esi                \n"   // esi = image->data
 154
 155 ".LoopYa:                               \n\t"
 156          "movl %13, %%eax               \n\t"
 157          "movl %%eax, %26               \n\t" // x = width
 158
 159          "movl %14, %%eax               \n\t"
 160          "decl %%eax                    \n\t" // y--
 161          "movl %%eax, %14               \n\t"
 162          "js .Enda                      \n\t" // if y < 0, goto end
 163          "andl $1, %%eax                \n\t"
 164          "jz .LoopY_1a                  \n"   // if (y&1) goto LoopY_1
 165
 166 ".LoopY_0a:                             \n\t"
 167
 168          "movl %2, %%ebx                \n\t" // ebx = err
 169          "movl %%ebx, %25               \n\t" // [-36] = err
 170          "movl %3, %%eax                \n\t" //
 171          "movl %%eax, %24               \n\t" // [-32] = nerr
 172
 173          "jmp .LoopXa                   \n"
 174
 175 ".LoopY_1a:                             \n\t"
 176
 177          "movl %3, %%ebx                \n\t" // ebx = nerr
 178          "movl %%ebx, %25               \n\t" // [-36] = nerr
 179          "movl %2, %%eax                \n\t" //
 180          "movl %%eax, %24               \n\t" // [-32] = eerr
 181
 182          ".align 16                     \n"
 183 ".LoopXa:                               \n\t"
 184
 185          // calculate errors and pixel components
 186
 187          // depend on ebx, esi, mm6
 188          "movq (%%ebx), %%mm1           \n\t" // mm1 = error[0..3]
 189          "punpcklbw (%%esi), %%mm0      \n\t" // mm0 = image->data[0..3]
 190          "psrlw $8, %%mm0               \n\t" // fixup mm0
 191          "paddusb %%mm1, %%mm0          \n\t" // mm0 = mm0 + mm1 (sat. to 255)
 192          "movq %%mm0, %20               \n\t" // save the pixel
 193
 194          "movzwl %20, %%ecx             \n\t" // ecx = pixel.red
 195          "movl %4, %%edi                \n\t" // edi = rtable
 196          //agi
 197          "leal (%%edi, %%ecx, 2), %%eax \n\t" // eax = &rtable[pixel.red]
 198          // agi
 199          "movw (%%eax), %%dx            \n\t" // dx = rtable[pixel.red]
 200          "movw %%dx, %16                \n\t" // save rr
 201
 202          "movzwl %21, %%ecx             \n\t" // ecx = pixel.green
 203          "movl %5, %%edi                \n\t" // edi = gtable
 204          //agi
 205          "leal (%%edi, %%ecx, 2), %%eax \n\t" // eax = &gtable[pixel.green]
 206          //agi
 207          "movw (%%eax), %%dx            \n\t" // dx = gtable[pixel.green]
 208          "movw %%dx, %17                \n\t" // save gg
 209
 210          "movzwl %22, %%ecx             \n\t" // ecx = pixel.blue
 211          "movl %6, %%edi                \n\t" // ebx = btable
 212          //agi
 213          "leal (%%edi, %%ecx, 2), %%eax \n\t" // eax = &btable[pixel.blue]
 214          //agi
 215          "movw (%%eax), %%dx            \n\t" // dx = btable[pixel.blue]
 216          "movw %%dx, %18                \n\t" // save bb
 217
 218          "movw $0, %19                  \n\t" // save dummy aa
 219
 220          "movq %16, %%mm1               \n\t" // load mm1 with rrggbbaa
 221          "pmullw %%mm6, %%mm1           \n\t" // mm1 = rr*dr|...
 222          "psubsw %%mm1, %%mm0           \n\t" // error = pixel - mm1
 223
 224
 225          // distribute the error
 226
 227          // depend on mm0, mm7, mm3, mm4, mm5
 228
 229          "movl %25, %%ebx               \n\t"
 230
 231          "movq %%mm0, %%mm1             \n\t"
 232          "pmullw %%mm5, %%mm1           \n\t" // mm1 = mm1*7
 233          "psrlw %%mm7, %%mm1            \n\t" // mm1 = mm1/16
 234          "paddw 8(%%ebx), %%mm1         \n\t"
 235          "movq %%mm1, 8(%%ebx)          \n\t" // err[x+1,y] = rer*7/16
 236
 237
 238          "movl %24, %%ebx               \n\t"
 239
 240          "movq %%mm0, %%mm1             \n\t"
 241          "pmullw %%mm4, %%mm1           \n\t" // mm1 = mm1*5
 242          "psrlw %%mm7, %%mm1            \n\t" // mm1 = mm1/16
 243          "paddw -8(%%ebx), %%mm1        \n\t"
 244          "movq %%mm1, -8(%%ebx)         \n\t" // err[x-1,y+1] += rer*3/16
 245
 246          "movq %%mm0, %%mm1             \n\t"
 247          "pmullw %%mm3, %%mm1           \n\t" // mm1 = mm1*3
 248          "psrlw %%mm7, %%mm1            \n\t" // mm1 = mm1/16
 249          "paddw 8(%%ebx), %%mm1         \n\t"
 250          "movq %%mm1, (%%ebx)           \n\t" // err[x,y+1] += rer*5/16
 251
 252          "psrlw %%mm7, %%mm0            \n\t" // mm0 = mm0/16
 253          "movq %%mm0, 8(%%ebx)          \n\t" // err[x+1,y+1] = rer/16
 254
 255
 256          // calculate final pixel value and store
 257          "movl %10, %%ecx               \n\t"
 258          "movw %16, %%ax                \n\t"
 259          "shlw %%cl, %%ax               \n\t" //NP* ax = r<<roffs
 260
 261          "movl %11, %%ecx               \n\t"
 262          "movw %17, %%bx                \n\t"
 263          "shlw %%cl, %%bx               \n\t" //NP*
 264          "orw %%bx, %%ax                \n\t"
 265
 266          "movl %12, %%ecx               \n\t"
 267          "movw %18, %%bx                \n\t"
 268          "shlw %%cl, %%bx               \n\t" //NP*
 269          "orw %%bx, %%ax                \n\t"
 270
 271          "movl %1, %%edx                \n\t"
 272          "movw %%ax, (%%edx)            \n\t"
 273          "addl $2, %%edx                \n\t" // increment ximage
 274          "movl %%edx, %1                \n\t"
 275
 276          // prepare for next iteration on X
 277
 278          "addl $8, %24                  \n\t" // nerr += 8
 279
 280          "movl %25, %%ebx               \n\t"
 281          "addl $8, %%ebx                \n\t"
 282          "movl %%ebx, %25               \n\t" // ebx = err += 8
 283
 284
 285          // Note: in the last pixel, this would cause an invalid memory access
 286          // because, punpcklbw is used (which reads 8 bytes) and the last
 287          // pixel is only 4 bytes. This is no problem because the image data
 288          // was allocated with extra 4 bytes when created.
 289          "addl $4, %%esi                \n\t" // image->data += 4
 290
 291
 292          "decl %26                      \n\t" // x--
 293          "jnz .LoopXa                   \n\t" // if x>0, goto .LoopX
 294
 295
 296          // depend on edx
 297          "addl %15, %%edx               \n\t" // add extra offset to ximage
 298          "movl %%edx, %1                \n\t"
 299
 300
 301          "jmp .LoopYa                   \n"
 302
 303 ".Enda:                                 \n\t" // THE END
 304          "emms                          \n\t"
 305          "popl %%ebx                    \n\t"
 306          :
 307          :
 308          "m" (image),                      // %0
 309          "m" (ximage),                     // %1
 310          "m" (err),                        // %2
 311          "m" (nerr),                       // %3
 312          "m" (rtable),                     // %4
 313          "m" (gtable),                     // %5
 314          "m" (btable),                     // %6
 315          "m" (dr),                         // %7
 316          "m" (dg),                         // %8
 317          "m" (db),                         // %9
 318          "m" (roffs),                      // %10
 319          "m" (goffs),                      // %11
 320          "m" (boffs),                      // %12
 321          "m" (width),                      // %13
 322          "m" (height),                     // %14
 323          "m" (line_offset),                // %15
 324          "m" (rrggbbaa.words.rr),          // %16 (access to rr)
 325          "m" (rrggbbaa.words.gg),          // %17 (access to gg)
 326          "m" (rrggbbaa.words.bb),          // %18 (access to bb)
 327          "m" (rrggbbaa.words.aa),          // %19 (access to aa)
 328          "m" (pixel.words.rr),             // %20 (access to pixel.r)
 329          "m" (pixel.words.gg),             // %21 (access to pixel.g)
 330          "m" (pixel.words.bb),             // %22 (access to pixel.b)
 331          "m" (pixel.words.aa),             // %23 (access to pixel.a)
 332          "m" (tmp_err),                    // %24
 333          "m" (tmp_nerr),                   // %25
 334          "m" (x)                           // %26
 335          : "eax", "ecx", "edx", "esi", "edi"
 336         );
 337 }
 338
 339
 340 void
 341 x86_mmx_TrueColor_24_to_16(unsigned char *image,
 342                            unsigned short *ximage,
 343                            short *err,
 344                            short *nerr,
 345                            short *rtable,
 346                            short *gtable,
 347                            short *btable,
 348                            int dr,
 349                            int dg,
 350                            int db,
 351                            unsigned int roffs,
 352                            unsigned int goffs,
 353                            unsigned int boffs,
 354                            int width,
 355                            int height,
 356                            int line_offset)
 357 {
 358     union {
 359         long long rrggbbaa;
 360         struct {
 361             short int rr, gg, bb, aa;
 362         } words;
 363     } rrggbbaa;
 364
 365     union {
 366         long long pixel;
 367         struct {
 368             short int rr, gg, bb, aa;
 369         } words;
 370     } pixel;
 371
 372     short *tmp_err;
 373     short *tmp_nerr;
 374
 375     int x;
 376     int w1;
 377     int w2;
 378
 379     asm volatile
 380         (
 381          "pushl %%ebx                   \n\t"
 382
 383          "movl %13, %%eax               \n\t" // eax = width
 384          "movl %%eax, %%ebx             \n\t"
 385          "shrl $2, %%eax                \n\t"
 386          "movl %%eax, %27               \n\t" // w1 = width / 4
 387          "andl $3, %%ebx                \n\t"
 388          "movl %%ebx, %28               \n"   // w2 = width %% 4
 389
 390
 391 ".LoopYc:                               \n\t"
 392          "movl %13, %%eax               \n\t"
 393          "movl %%eax, %26               \n\t" // x = width
 394
 395          "decl %14                      \n\t" // height--
 396          "js .Endc                      \n\t" // if height < 0 then end
 397
 398          "movl %14, %%eax               \n\t"
 399          "decl %%eax                    \n\t" // y--
 400          "movl %%eax, %14               \n\t"
 401          "js .Endc                      \n\t" // if y < 0, goto end
 402          "andl $1, %%eax                \n\t"
 403          "jz .LoopY_1c                  \n"   // if (y&1) goto LoopY_1
 404
 405 ".LoopY_0c:                             \n\t"
 406
 407          "movl %2, %%ebx                \n\t" // ebx = err
 408          "movl %%ebx, %25               \n\t" // [-36] = err
 409          "movl %3, %%eax                \n\t" //
 410          "movl %%eax, %24               \n\t" // [-32] = nerr
 411
 412          "jmp .LoopX_1c                 \n"
 413
 414 ".LoopY_1c:                             \n\t"
 415
 416          "movl %3, %%ebx                \n\t" // ebx = nerr
 417          "movl %%ebx, %25               \n\t" // [-36] = nerr
 418          "movl %2, %%eax                \n\t" //
 419          "movl %%eax, %24               \n\t" // [-32] = eerr
 420
 421          ".align 16                     \n\t"
 422
 423          "movl %%eax, %26               \n"   // x = w1
 424 ".LoopX_1c:                             \n\t"
 425          "decl %26                      \n\t" // x--
 426          "js .Xend1_c                   \n\t" // if x < 0 then end
 427
 428          // do conversion of 4 pixels
 429          "movq %2, %%mm0                \n\t" // mm0 = err
 430
 431
 432
 433
 434          "jmp .LoopX_1c                 \n"
 435 ".Xend1_c:                              \n\t"
 436
 437          "movl %28, %%eax               \n\t"
 438          "movl %%eax, %26               \n"   // x = w2
 439 ".LoopX_2c:                             \n\t"
 440          "decl %26                      \n\t" // x--
 441          "js .Xend2_c                   \n\t" //
 442          // do conversion
 443          "jmp .LoopX_2c                 \n"
 444 ".Xend2_c:                              \n\t"
 445
 446          "movl %27, %%eax               \n\t"
 447          "jmp .LoopYc                   \n"
 448
 449 ".Endc:                                 \n\t" // THE END
 450          "emms                          \n\t"
 451          "popl %%ebx                    \n\t"
 452          :
 453          :
 454          "m" (image),                      // %0
 455          "m" (ximage),                     // %1
 456          "m" (err),                        // %2
 457          "m" (nerr),                       // %3
 458          "m" (rtable),                     // %4
 459          "m" (gtable),                     // %5
 460          "m" (btable),                     // %6
 461          "m" (dr),                         // %7
 462          "m" (dg),                         // %8
 463          "m" (db),                         // %9
 464          "m" (roffs),                      // %10
 465          "m" (goffs),                      // %11
 466          "m" (boffs),                      // %12
 467          "m" (width),                      // %13
 468          "m" (height),                     // %14
 469          "m" (line_offset),                // %15
 470          "m" (rrggbbaa.words.rr),          // %16 (access to rr)
 471          "m" (rrggbbaa.words.gg),          // %17 (access to gg)
 472          "m" (rrggbbaa.words.bb),          // %18 (access to bb)
 473          "m" (rrggbbaa.words.aa),          // %19 (access to aa)
 474          "m" (pixel.words.rr),             // %20 (access to pixel.r)
 475          "m" (pixel.words.gg),             // %21 (access to pixel.g)
 476          "m" (pixel.words.bb),             // %22 (access to pixel.b)
 477          "m" (pixel.words.aa),             // %23 (access to pixel.a)
 478          "m" (tmp_err),                    // %24
 479          "m" (tmp_nerr),                   // %25
 480          "m" (x),                          // %26
 481          "m" (w1),                         // %27
 482          "m" (w2)                          // %28
 483          : "eax", "ecx", "edx", "esi", "edi"
 484         );
 485 }
 486
 487
 488
 489 #endif /* ASM_X86_MMX */
 490
 491
 492
 493 void
 494 x86_PseudoColor_32_to_8(unsigned char *image,
 495                         unsigned char *ximage,
 496                         char *err,
 497                         char *nerr,
 498                         short *ctable,
 499                         int dr,
 500                         int dg,
 501                         int db,
 502                         unsigned long *pixels,
 503                         int cpc,
 504                         int width,
 505                         int height,
 506                         int bytesPerPixel,
 507                         int line_offset)
 508 {
 509     int x;
 510     int cpcpc;
 511
 512     int rr;
 513     int gg;
 514     int bb;
 515
 516     char *tmp_err;
 517     char *tmp_nerr;
 518
 519     char ndr; // aparently not used
 520     char ndg; // aparently not used
 521     char ndb; // aparently not used
 522
 523     asm volatile
 524         (
 525          "pushal                        \n\t"
 526
 527          "movl %9, %%eax                \n\t"
 528          "mulb %9                       \n\t"
 529          "movl %%eax, %15               \n\t" // cpcpc = cpc*cpc
 530
 531          // eax will always be <= 0xffff
 532
 533          // process 1 pixel / cycle, each component treated as 16bit
 534          "movl %0, %%esi                \n"   // esi = image->data
 535
 536 ".LoopYb:                               \n\t"
 537          "movl %10, %%ecx               \n\t"
 538          "movl %%ecx, %14               \n\t" // x = width
 539
 540          "movl %11, %%ecx               \n\t"
 541          "decl %%ecx                    \n\t" // y--
 542          "movl %%ecx, %11               \n\t"
 543          "js .Endb                      \n\t" // if y < 0, goto end
 544          "andl $1, %%ecx                \n\t"
 545          "jz .LoopY_1b                  \n"   // if (y&1) goto LoopY_1
 546
 547 ".LoopY_0b:                             \n\t"
 548
 549          "movl %2, %%ebx                \n\t" // ebx = err
 550 //useless "movl %%ebx, %20              \n\t" // [-36] = err
 551          "movl %3, %%ecx                \n\t" //
 552          "movl %%ecx, %19               \n\t" // [-32] = nerr
 553
 554          "movl $0, (%%ecx)              \n\t" // init error of nerr[0] to 0
 555
 556          "jmp .LoopXb                   \n"
 557
 558 ".LoopY_1b:                             \n\t"
 559
 560          "movl %3, %%ebx                \n\t" // ebx = nerr
 561 //useless "movl %%ebx, %20              \n\t" // [-36] = nerr
 562          "movl %2, %%ecx                \n\t" //
 563          "movl %%ecx, %19               \n\t" // [-32] = err
 564
 565          "movl $0, (%%ecx)              \n\t" // init error of nerr[0] to 0
 566
 567
 568          ".align 16                     \n"
 569 ".LoopXb:                               \n\t"
 570
 571
 572          "movl %4, %%edi                \n\t" // edi = ctable
 573          "xorl %%edx, %%edx             \n\t" // zero the upper word on edx
 574
 575          // RED
 576
 577          // depends on ebx==err, esi==image->data, edi
 578          "movzbw (%%esi), %%dx          \n\t" // dx = image->data[0]
 579          "movsbw (%%ebx), %%ax          \n\t" // ax = error[0]
 580          "addw %%ax, %%dx               \n\t" // pixel.red = data[0] + error[0]
 581
 582          "testb %%dh, %%dh              \n\t" // test if pixel.red < 0 or > 255
 583          "jz .OKRb                      \n\t" // 0 <= pixel.red <= 255
 584          "js .NEGRb                     \n\t" // pixel.red < 0
 585          "movw $0xff, %%dx              \n\t" // pixel.red > 255
 586          "jmp .OKRb                     \n"
 587 ".NEGRb:                                \n\t"
 588          "xorw %%dx, %%dx               \n"
 589 ".OKRb:                                 \n\t"
 590          //partial reg
 591          "leal (%%edi, %%edx, 2), %%ecx \n\t" // ecx = &ctable[pixel.red]
 592          //agi
 593          "movl (%%ecx), %%eax           \n\t" // ax = ctable[pixel.red]
 594          "movw %%ax, %16                \n\t" // save rr
 595
 596          "mulb %5                       \n\t" // ax = rr*dr
 597          "subw %%ax, %%dx               \n\t" // rer = dx = dx - rr*dr
 598
 599          "movswl %%dx, %%eax            \n\t" // save rer
 600
 601          // distribute error
 602          "leal (, %%eax, 8), %%ecx      \n\t"
 603          "subw %%dx, %%cx               \n\t" // cx = rer * 7
 604          "sarw $4, %%cx                 \n\t" // cx = rer * 7 / 16
 605          "addb %%cl, 4(%%ebx)           \n\t" // err[x+1] += rer * 7 / 16
 606
 607          "movl %19, %%ecx               \n\t" // ecx = nerr
 608
 609          "leaw (%%eax, %%eax, 4), %%dx  \n\t" // dx = rer * 5
 610          "sarw $4, %%dx                 \n\t" // dx = rer * 5 / 16
 611          "addb %%dl, (%%ecx)            \n\t" // nerr[x] += rer * 5 / 16
 612
 613          "leaw (%%eax, %%eax, 2), %%dx  \n\t" // dx = rer * 3
 614          "sarw $4, %%dx                 \n\t" // dx = rer * 3 / 16
 615          "addb %%dl, -4(%%ecx)          \n\t" // nerr[x-1] += rer * 3 / 16
 616
 617          "sarw $4, %%ax                 \n\t" // ax = rer / 16
 618          "movb %%al, 4(%%ecx)           \n\t" // nerr[x+1] = rer / 16
 619
 620
 621          // GREEN
 622
 623          // depends on ebx, esi, edi
 624          "movzbw 1(%%esi), %%dx         \n\t" // dx = image->data[1]
 625          "movsbw 1(%%ebx), %%ax         \n\t" // ax = error[1]
 626          "addw %%ax, %%dx               \n\t" // pixel.grn = data[1] + error[1]
 627
 628          "testb %%dh, %%dh              \n\t" // test if pixel.grn < 0 or > 255
 629          "jz .OKGb                      \n\t" // 0 <= pixel.grn <= 255
 630          "js .NEGGb                     \n\t" // pixel.grn < 0
 631          "movw $0xff, %%dx              \n\t" // pixel.grn > 255
 632          "jmp .OKGb                     \n"
 633 ".NEGGb:                                \n\t"
 634          "xorw %%dx, %%dx               \n"
 635 ".OKGb:                                 \n\t"
 636          // partial reg
 637          "leal (%%edi, %%edx, 2), %%ecx \n\t" // ecx = &ctable[pixel.grn]
 638          //agi
 639          "movw (%%ecx), %%ax            \n\t" // ax = ctable[pixel.grn]
 640          "movw %%ax, %17                \n\t" // save gg
 641
 642          "mulb %6                       \n\t" // ax = gg*dg
 643          "subw %%ax, %%dx               \n\t" // ger = dx = dx - gg*dg
 644
 645          "movswl %%dx, %%eax            \n\t" // save ger
 646
 647          // distribute error
 648
 649          "leal (, %%eax, 8), %%ecx      \n\t"
 650          "subw %%dx, %%cx               \n\t" // cx = ger * 7
 651          "sarw $4, %%cx                 \n\t" // cx = ger * 7 / 16
 652          "addb %%cl, 5(%%ebx)           \n\t" // err[x+1] += ger * 7 / 16
 653
 654          "movl %19, %%ecx               \n\t" // ecx = nerr
 655
 656          "leaw (%%eax, %%eax, 4), %%dx  \n\t" // dx = ger * 5
 657          "sarw $4, %%dx                 \n\t" // dx = ger * 5 / 16
 658          "addb %%dl, 1(%%ecx)           \n\t" // nerr[x] += ger * 5 / 16
 659
 660          "leaw (%%eax, %%eax, 2), %%dx  \n\t" // dx = ger * 3
 661          "sarw $4, %%dx                 \n\t" // dx = ger * 3 / 16
 662          "addb %%dl, -3(%%ecx)          \n\t" // nerr[x-1] += ger * 3 / 16
 663
 664          "sarw $4, %%ax                 \n\t" // ax = ger / 16
 665          "movb %%al, 5(%%ecx)           \n\t" // nerr[x+1] = ger / 16
 666
 667
 668          // BLUE
 669
 670          // depends on ebx, esi
 671          "movzbw 2(%%esi), %%dx         \n\t" // dx = image->data[2]
 672          "movsbw 2(%%ebx), %%ax         \n\t" // ax = error[2]
 673          "addw %%ax, %%dx               \n\t" // pixel.grn = data[2] + error[2]
 674
 675          "testb %%dh, %%dh              \n\t" // test if pixel.blu < 0 or > 255
 676          "jz .OKBb                      \n\t" // 0 <= pixel.blu <= 255
 677          "js .NEGBb                     \n\t" // pixel.blu < 0
 678          "movw $0xff, %%dx              \n\t" // pixel.blu > 255
 679          "jmp .OKBb                     \n"
 680 ".NEGBb:                                \n\t"
 681          "xorw %%dx, %%dx               \n"
 682 ".OKBb:                                 \n\t"
 683          //partial reg
 684          "leal (%%edi, %%edx, 2), %%ecx \n\t" // ecx = &ctable[pixel.blu]
 685          //agi
 686          "movw (%%ecx), %%ax            \n\t" // ax = ctable[pixel.blu]
 687          "movw %%ax, %18                \n\t" // save bb
 688
 689          "mulb %7                       \n\t" // ax = bb*db
 690          "subw %%ax, %%dx               \n\t" // ber = dx = dx - bb*db
 691          "movswl %%dx, %%eax            \n\t" // save ber
 692
 693          // distribute error
 694          "leal (, %%eax, 8), %%ecx      \n\t"
 695          "subw %%dx, %%cx               \n\t" // cx = ber * 7
 696          "sarw $4, %%cx                 \n\t" // cx = ber * 7 / 16
 697          "addb %%cl, 6(%%ebx)           \n\t" // err[x+1] += ber * 7 / 16
 698
 699          "movl %19, %%ecx               \n\t" // ecx = nerr
 700
 701          "leaw (%%eax, %%eax, 4), %%dx  \n\t" // dx = ber * 5
 702          "sarw $4, %%dx                 \n\t" // dx = ber * 5 / 16
 703          "addb %%dl, 2(%%ecx)           \n\t" // nerr[x] += ber * 5 / 16
 704
 705          "leaw (%%eax, %%eax, 2), %%dx  \n\t" // dx = ber * 3
 706          "sarw $4, %%dx                 \n\t" // dx = ber * 3 / 16
 707          "addb %%dl, -4(%%ecx)          \n\t" // nerr[x-1] += ber * 3 / 16
 708
 709          "sarw $4, %%ax                 \n\t" // ax = ber / 16
 710          "movb %%al, 6(%%ecx)           \n\t" // nerr[x+1] = ber / 16
 711
 712          "andl $0xffff, %%eax           \n\t"
 713          // depends on eax & 0xffff0000 == 0
 714          // calculate the index of the value of the pixel
 715          "movw %16, %%ax                \n\t" // ax = rr
 716          "mulb %15                      \n\t" // ax = cpcpc*rr
 717          "movw %%ax, %%cx               \n\t"
 718          "movw %17, %%ax                \n\t" // ax = gg
 719          "mulb %9                       \n\t" // ax = cpc*gg
 720          "addw %%cx, %%ax               \n\t" // ax = cpc*gg + cpcpc*rr
 721          "addw %18, %%ax                \n\t" // ax = cpcpc*rr + cpc*gg + bb
 722
 723          "movl %8, %%ecx                \n\t"
 724          //agi
 725          "leal (%%ecx, %%eax, 4), %%edx \n\t"
 726          //agi
 727          "movb (%%edx), %%cl            \n\t" // cl = pixels[ax]
 728
 729          // store the pixel
 730          "movl %1, %%eax                \n\t"
 731          "movb %%cl, (%%eax)            \n\t" // *ximage = cl
 732          "incl %1                       \n\t" // ximage++
 733
 734          // prepare for next iteration on X
 735
 736          "addl $4, %19                  \n\t" // nerr += 4
 737          "addl $4, %%ebx                \n\t" // err += 4
 738
 739          "addl %12, %%esi               \n\t" // image->data += bpp
 740
 741          "decl %14                      \n\t" // x--
 742          "jnz .LoopXb                   \n\t" // if x>0, goto .LoopX
 743
 744
 745          "movl %13, %%eax               \n\t"
 746          "addl %%eax, %1                \n\t" // add extra offset to ximage
 747
 748          "jmp .LoopYb                   \n"
 749
 750 ".Endb:                                 \n\t"
 751          "emms                          \n\t"
 752          "popal                         \n\t"
 753          :
 754          :
 755          "m" (image),         // %0
 756          "m" (ximage),        // %1
 757          "m" (err),           // %2
 758          "m" (nerr),          // %3
 759          "m" (ctable),        // %4
 760          "m" (dr),            // %5
 761          "m" (dg),            // %6
 762          "m" (db),            // %7
 763          "m" (pixels),        // %8
 764          "m" (cpc),           // %9
 765          "m" (width),         // %10
 766          "m" (height),        // %11
 767          "m" (bytesPerPixel), // %12
 768          "m" (line_offset),   // %13
 769          "m" (x),             // %14
 770          "m" (cpcpc),         // %15
 771          "m" (rr),            // %16
 772          "m" (gg),            // %17
 773          "m" (bb),            // %18
 774          "m" (tmp_err),       // %19
 775          "m" (tmp_nerr),      // %20
 776          "m" (ndr),           // %21
 777          "m" (ndg),           // %22
 778          "m" (ndb)            // %23
 779         );
 780 }
 781
 782 #endif /* ASM_X86 */
 783