format binary as 'gif'

Width   = 120
Height  = 120

IterLim = 256
Prec    = 29; don't set above 29 nor to low, else image may be scrambled

virtual at 0
  canvas::

    PrecScale = 1 shl (Prec-1)

    StepR = (5 shl (Prec-1) +Width       shr 1) / Width; = 2.5/Width
    StepI = (5 shl (Prec-1) +Height shr 1) / Height; = 3/Height

    PosI = -5 shl (Prec-2); = -1.25
    repeat Height
     PosR = -2 shl Prec; = -2
     repeat Width
      R = PosR
      I = PosI
      RR = (PosR*PosR) shr Prec
      II = (PosI*PosI) shr Prec

      Color = IterLim - 1; Color is also the current iteration number
      repeat IterLim
       if (RR + II) > 4 shl Prec
        break
       end if
       Tmp = (RR - II) + PosR
       I = (R*I) / PrecScale + PosI
       R = Tmp
       RR = (Tmp*Tmp) shr Prec
       II = (I*I) shr Prec
       Color = Color - 1; discounting Color
      end repeat
      if Color < 0
       Color = 0
      end if
      db Color

      PosR = PosR + StepR
     end repeat
     PosI = PosI + StepI
    end repeat

  .size = $
end virtual


signature   db 'GIF87a'
width       dw Width
height      dw Height
bits        db 11110111b
background  db 0
reserved    db 0

; palette: 256 * 3 bytes (RGB)
repeat 256
 db % - 1
 db (%+0A9h) and 0FFh
 db (%+54h) and 0FFh
end repeat

db ','
dw 0,0
dw Width,Height
db 0

db 8
__stream__start = $
db 0

macro __stream__write databyte {
  if $-__stream__start<256
    store byte $-__stream__start at __stream__start
  else
    __stream__start = $
    db 0
  end if
  db databyte
}

__LZW__MAX_ENTRIES = 4096 

virtual at 0 
  __LZW__table:: 
  dd __LZW__MAX_ENTRIES dup (?,?) 
end virtual 

__LZW__count = 102h 

__LZW__last_byte = 0 
__LZW__bit_offset = 0 

macro __LZW__write code {
  local c,d,p 
  d = __LZW__count-1 
  c = 0 
  while d 
    d = d shr 1 
    c = c + 1 
  end while 
  d = code
  while c > 0 
    if __LZW__bit_offset + c >= 8 
      p = 8 - __LZW__bit_offset
      __stream__write __LZW__last_byte or (d and (1 shl p - 1)) shl __LZW__bit_offset
      __LZW__last_byte = 0
      __LZW__bit_offset = 0 
      c = c - p 
      d = d shr p 
    else 
      __LZW__last_byte = __LZW__last_byte or d shl __LZW__bit_offset 
      __LZW__bit_offset = __LZW__bit_offset + c 
      c = 0 
    end if 
  end while 
} 

dictionary_code = 0 
string_offset = 0 
string_length = 0 
scan_from = 102h 

while string_offset + string_length < canvas.size

  string_length = string_length + 1 

  if string_length = 1 

    load dictionary_code byte from canvas:string_offset

  else 

    in_dictionary = 0 

    repeat __LZW__count - scan_from 
       load entry_offset dword from __LZW__table:(scan_from+%-1)*8 
       load entry_length dword from __LZW__table:(scan_from+%-1)*8+4 
       if string_length = entry_length 
         equal = 1 
         repeat string_length 
           load a byte from canvas:string_offset+%-1
           load b byte from canvas:entry_offset+%-1
           if a <> b 
             equal = 0 
             break 
           end if 
         end repeat 
         if equal 
           in_dictionary = 1 
           dictionary_code = scan_from+%-1 
           scan_from = scan_from+% 
           break 
         end if 
       end if 
    end repeat 

    if ~ in_dictionary 

      __LZW__write dictionary_code 

      store dword string_offset at __LZW__table:__LZW__count*8 
      store dword string_length at __LZW__table:__LZW__count*8+4 
      __LZW__count = __LZW__count+1 

      if __LZW__count>=__LZW__MAX_ENTRIES-1 
        __LZW__write 100h 
        __LZW__count = 102h 
      end if 

      string_offset = string_offset+string_length-1 
      string_length = 0 
      scan_from = 102h 
    end if 

  end if 

end while 

if string_length > 0 
  __LZW__write dictionary_code 
  __LZW__count = __LZW__count+1 
  __LZW__write 101h 
end if 

if __LZW__bit_offset > 0 
  __stream__write __LZW__last_byte
end if

db 0

db ';'    

now generates GIF instead of TGA

Fast Assembler, version 1.50.50 (16384 kilobytes memory)
1 passes, 2.1 seconds, 15186 bytes.

Hy again here are my test results:

; Some basic image properties. Feel free to modify (786 bytes header)
; tests in real time, Celeron II @ 933Mhz (600), 103MHz Bus, FSB, 16+16K L1, 256k L2:
; 20x20 : 0.2s 2000 pixels per sec 1186 bytes
; 40x40 : 0.9s 1777.77 pixels per sec 2386 bytes
; 80x80 : 3.6s 1777.77 pixels per sec 7186 bytes
; default 120x120 : 8.2s 1756.09 pixels per sec 15186 bytes
; 160x160 : 14.7s 1741.49 pixels per sec 26386 bytes
; 200x200 : 23.1s 1731.60 pixels per sec 40786 bytes
; 280x280 : 44.9s 1746.10 pixels per sec 79186 bytes
; 800x800 : 366.7s 1745.30 pixels per sec 640786 bytes
; 1200x1200 : 831.3s 1732.23 pixels per sec 1440786 bytes

MATRIX