Dírojed CGA by CRTC

For Sundown 2014 I had planned on doing an impressive 8088 4.77MHz PC speaker
related project but I was rapidly running out of time and couldn't concentrate
on it properly with all the noise in the hall. However, I still wanted to
submit *something*.

I got my XT set up and it's nice to have some little animation or picture
running on the screen for passers-by to look at. I decided to knock up some
kind of quick little display hack to keep the machine busy, but what? Where
could I find a little animation program, ideally x86 16-bit real mode DOS?
Inspiration hit: I searched pouet for the best rated demos under 64 bytes
(extreme sizecoding is usually 16-bit real mode DOS because doing anything with
32-bits, protected mode or Windows costs a lot of bytes).

Most of the effects (I discovered later) are too slow to run at 4.77MHz
(extreme sizecoded effects are often "de-optimized" to save bytes at the
cost of requiring a modern machine to run at a decent speed) or use FPU
instructions but the top hit, D¡rojed by Rrrola
(www.pouet.net/prod.php?which=28913 - itself a port of the 64-byte demo
Lochfrass by T$) turned out to be perfect for what I wanted. It's a 32-byter,
and requires VGA (mode 13h, as usual for extreme sizecoding). It also uses one
286+ instruction (imul rw,imm).

After fixing those two issues I had a nice "attract screen" in 42 bytes (the
"press esc to exit" functionality was the first to go, since I was running
it from the keyboard port anyway (using the hack I described at
http://www.reenigne.org/blog/i-bought-an-xt) and always do a hard reset to load
a new program when I'm doing that. I also changed a "add cl,[bx-321]" to "add
cl,[bx-81]" to reflect the reduced stride, which saved another byte.

I was chatting to another CRTC member (doz) about it over lunch on Saturday
and he suggested that I should enter it as a prod (he has been trying to get
me to enter something the past 3 Sundowns). I told him I would, but only if
I could get it back under 32 bytes. The challenge was on!

One problem with using the keyboard port to run extreme sizecoded programs
is that the initial register values aren't as useful as the ones DOS
provides for a .com program. Of course, I wrote the bootloader so I could
put whatever initial register values I wanted in there. Doz (correctly)
pointed out that that would be cheating. He then suggested I actually make
it a proper .com file so that other people can run it too. Of course! How
did I miss this obvious idea? So I did exactly that. That got it to 34
bytes. Two to go!

The first one was related to my changing "imul bx,0xe5" to an 8088-compatible
  mov ax,0xe5
  imul bx
  mov bx,ax
I realized I could replace the third instruction with "xchg ax,bx".

A neat trick that sizecoders have is to put:
  lds bx,[bx]
at the start of the program. The initial value of bx is 0 and the bytes at the
start of the program segment are:
  int 0x20
  dw 0x9fff  ; top segment of conventional RAM
So that two byte instruction puts 0x20cd in BX and 0x9fff in DS which is almost
perfect for writing to video memory. My first attempt to "CGA-ise" this code
replaced the byte in the PSP before the lds instruction:
  mov [bx+3],0xb7
The final byte was saved by realizing that the intial value of BX doesn't
matter, so I could replace the mov/lds sequence with the more mundane but
  mov bx,0xb800
  mov ds,bx

At that point it was about 5 minutes before the deadline so I started the
submission process, but unfortunately I had a last second bug! Since the 8088's
imul instruction stomps ah, the "increment [bx] only if carry set" instruction:
 adc [bx],ah
became an "always add something to [bx]" instruction, which ruined the effect.

I found and fixed the bug (by using ch there instead of ah) pretty quickly but
my process for testing .com files on the XT is cumbersome: I use DOSBox to
create/modify a disk image to contain the program I want to test. Then I send
the "write_image.bin" program and the disk image over the serial port.
write_image writes the image to disk and boots to it. This all takes a few

I was also trying to test it on DOSBox but failing - my usual go-to DOSBox is
the one with the debugger, but that emits a debug message whenver a program
writes to ROM-mapped space (which this demo does a lot, unfortunately) so it's
too slow and for some reason was not going into graphics mode. Going back to
the standard DOSBox fixed both of those problems and I got the fixed version
submitted in the nick of time. We ended up using DOSBox to display it on the
big screen rather than trying to move the XT to the projector or stringing up a
long composite connection - we also used the default 3000 cycles rather than XT
speed (which is a bit on the slow side for a compo presentation) but since I
described it as "an 8088/CGA port of D¡rojed" rather than "a 4.77MHz 8088/CGA
port" I think that's allowable.

The effect looks a little different on the XT than on DOSBox because of memory
layout differences - on DOSBox the pattern just goes up from the bottom, on the
XT it comes down from the top as well. It looks different again on DOSBox with
machine=vgaonly (comes down from the top corners).

https://www.youtube.com/watch?v=OCH8A4tn5nE is the live video from the party.
https://www.youtube.com/watch?v=Ys6koiu1lF0 is a capture from DOSBox with
machine=vgaonly which keeps going until the image becomes stable.

D¡rojed CGA came 3rd in the Oldskool demo competition at Sundown 2014.

Greetz to: pup, megmeg, doz, rc55, psonice, Deltafire, tunk, Alegend45, joey,
Topy44, stavs, kabuto, m0d, Trixter, Scali, VileRancour, gasman, danbee, ne7,
rrrola, T$ and anyone I forgot.