Fullscreen programming on the Atari ST

From AtariForumWiki
Jump to navigation Jump to search
            "Fullscreen"-Programming on the Atari ST
                     by Flix of Delta Force

Welcome  to the first part of a series of articles in  which  the 
Delta  Force will explain how to code  "demos",  featuring  fully 
working source-codes on the Maggie-disk.  This article will  show 
you  how  to  display screens with a resolution  of  416  by  276 
pixels, using 16 colours per line.
The  reader  of  this  course  should  have  basic  knowledge  of 
assembler   and  the  ST's  hardware.   The  first  topics   are: 
Annihilation of the screen borders (by Flix),  Sync-Scrolling (by 
New Mode) and Music programming using the YM-2149 (by Big  Alec). 

But now let's talk about the borders:
Allthough nobody expected it to be possible a long time ago,  the 
ST's  borders  can  be removed  by  pure  software.  Demos,  that 
displayed  graphics  in the borders using nothing  but  software, 
were  the  foundation  for  the  development  of  the  well-known 
"Overscan"-hardware.

At first,  I will write a bit about the history of the  so-called 
"border-programming".
The  first human ever to display any graphics in a screen  border 
was  Sven alias "Alyssa" from Mannheim/Germany.  Back in 1987  he 
wrote  an intro,  that showed graphics in the  lower  border.  As 
Sven  vanished  very fast from the  ST-scene,  it  is  completely 
unsettled,  how Sven had the idea to switch the screen  frequency 
to  60 Hz for a short period of time within the last line of  the 
screen.  Without this fundamental idea,  no fullscreens or  sync-
scrolling  would be possible on the ST.  Unfortunately nobody  in 
the  ST-scene  knows  what "Alyssa"  is  doing  today.  TEX  (The 
Exceptions)  firstly  used graphics in the lower  border  in  the 
famous  "B.I.G.  Demo".  To  open the lower  border,  the  screen 
frequency  has  to be switched to 60 Hz at the end of  the  199th 
line.  The  ST  then  reads nearly 50 more lines  of  the  screen 
memory!  The Exceptions revealed this secret in the  B.I.G.  Demo 
at the end of the longest scrolltext (42 Kybtes of  text).  After 
The Exceptions took up Alyssa's trick,  the upper border was  the 
next  one  to  be  removed.  This is  achieved  by  toggling  the 
frequency  13  or  29 lines above the "usual"  beginning  of  the 
screen.  You  may wonder,  why the upper border can be 13  or  29 
lines long.  Unfortunately there are two different  MMU-versions, 
that have a difference of 16 lines.  Due to the place to open the 
upper  border  being  above the screen memory  and  the  Timer  B 
starting  to  count at the beginning of the  screen  memory,  The 
Exceptions  used a quite complicated method.  They set a Timer  B 
Interrupt  in  the last line and waited  until  the  electron-ray 
reached the right position in the upper border.  Doing this, they 
had  to waste a lot of processor-time.  In the "Musical Wonder  - 
1991"  I  used a routine that waited a few  scanlines  after  the 
VBL-Interrupt occured and then toggled the screen-frequency.  But 
there's  still a better method.  Timer B is counting just  "real" 
screen-lines  (200),  but there is another  interrupt,  the  HBL, 
which  counts all scanlines (313).  At the beginning of  the  VBL 
the  HBL  starts counting until the right position  to  open  the 
upper border is reached.  If you now open both, the upper and the 
lower,  border,  you  have  277  lines of  graphics  in  the  low 
resolution.

Soon  the next border was declared to vanish:  The right  border. 
Unfortunately it is not possible to open this border with Timer B 
or HBL interrupts.  The ST displays graphics in the right border, 
if  you  switch the frequency to 60 Hz at a certain  position  in 
every line,  in which the right border is supposed to be  opened. 
This  "certain" position requires a completely new  technique  of 
programming. If you switch colours in an ordinary Timer B, VBL or 
HBL interrupt, you can see, that these interrupts do not occur at 
exactly the same position. The colours shake from the left to the 
right.  In  order to make some commands,  like the colour or  our 
frequency  switchings,  occur at exactly the same  position,  you 
have  to  become synchronized with  the  raster-electron-ray.  An 
ingenious method to achieve this effect is the following one:

WAIT:     MOVE.B    $FF8209.W,D0   ; Low-Byte
          BEQ.S     WAIT           ; mustn't be 0
          NOT.B     D0             ; negate D0
          LSL.B     D0,D0          ; Synchronisation

If  you execute this routine every VBL,  all  following  commands 
will be executed at the same position every VBL,  that means that 
the colour- or frequency-switches are stable.  But what does this 
little routine do?  At first,  the low-byte of the screen-address 
is  loaded  into D0.  This byte exactly determines  the  position 
within  the line.  It is negated and the LSL-command is  executed 
(LSR,  ASL  or  ASR  work as well).  As you  can  read  in  every 
processor-book  the LSL-command takes  8+2*n  clock-cycles.  That 
means  that  the command needs more clock-cycles the  bigger  the 
value in D0 is. That is exactly the shifting that we need! I hope 
that you understood this part,  because all fullscreens and sync-
scrolling-routines  are based upon this effect.  You should  know 
that  one  VBL  (50  Hz) consists  of  160000  clock-cycles  (one 
scanline  consists of 512 clock-cycles).  Now you have to  switch 
the  frequency  at a certain position and the  border  opens.  Of 
course  this takes a lot more processor time than the opening  of 
the  upper  or lower border,  because you've to open  the  border 
every  line.  Your  ST now displays 204 bytes per  line!  A  line 
consists  of 25.5 instead of 20 words without the  right  border, 
but should use only 23 of these 25.5 words,  because the  picture 
is  distorted on some STs (We made this mistake in  the  "Musical 
Wonder  - 1991") if you use too much  words.  Anyway,  there  are 
hardly  any monitors that have such a huge visible right  border. 
A  demo-screen  without  the  lower  and  the  right  border  was 
included in the "Amiga-Demo" by TEX.
For  a  long time it was considered impossible to open  the  left 
border,  because the trick with the 60-hertz-switch did not  work 
in the left border (I was sure about that as well). But than came 
the  "Death  Of The Left Border"-Demo by  the  TNT-Crew.  Finally 
there  was a screen in the legendary Union-Demo that  firstly  in 
ST-history  displayed  no borders at  all:  The  first  so-called 
"Fullscreen"!  The demo-coders made the impossible possible.  But 
how  does  this  trick  work?  It's  pretty  simple:  Instead  of 
switching  to  60  Hz you switch to 71 Hz!  If  you  disable  the 
interrupt  that  resets  at 71 Hz,  you can switch to  71  Hz  in 
coulour mode.  You should do this only for short periods of  time 
(a few clock-cycles).  Being in 71 Hz for a long time can  damage 
your  monitor (allthough I've never seen such damages).  It  goes 
without  saying  that neither Maggie nor the "Delta  Force"  take 
responsibility for damages caused by this technique.  If you  now 
open the right border additionally,  the ST displays 230  instead 
of  160  bytes  per line!  In order to make the  opening  of  the 
"side-borders"  run  on  all STs,  you've  to  install  so-called 
"Stabilisation"-switches  at  the  end  of  each  line  (This  is 
sometimes  called "Closing of the right border" which it  isn't). 
You  can  switch  to  71 Hz (like in the  source)  or  to  medium 
resolution (like ULM (Unlimited Matricks) does it).  I prefer the 
71  Hz  method,  because  I heard that the  other  method  causes 
problems  on few STs.  The "Stabilisation" is needed because  the 
shifter  waits for the last word at the end of the  right  border 
(115  words are displayed in fullscreen) to fill the last  plane. 
This  word  never  arrives and to avoid total  confusion  of  the 
shifter the 71-Hz-switch does something like a shifter-reset.
If you now open all borders, you've a screen memory that consists 
of  160+230*276  (=63640) bytes.  The first line  is  needed  for 
synchronisation, therefore it has 160 bytes. The screen memory is 
almost twice as big as in the "usual" low resolution!
Two registers of the shifter are needed to switch the frequency:

     $FFFF820A:     0: 60 Hertz         2: 50 Hertz     
     $FFFF8260:     0: Low resolution   2: High Res. (71 Hertz)

There  are various methods to cause the  frequency-switches.  ULM 
uses the "Immediate"-Adressation, that means a 60-Hz-switch would 
look like this:

          MOVE.B    #0,$FF820A.W   ; 60 Hz (16 clock-cycles)
          MOVE.B    #2,$FF820A.W   ; 50 Hz (16 clock-cycles)

Besides  that this method takes a lot of processor  time,  it  is 
said  that it does not run on all STs.  Another method is to  use 
registers.  This  is faster,  but takes some Data-  and  Address-
registers.  A  60-Hz-switch  would look like  that  (D0=0,  D1=2, 
A0=$FF820A.W):

          MOVE.B    D0,(A0)        ; 60 Hz (8 clock-cycles)
          MOVE.B    D1,(A0)        ; 50 Hz (8 clock-cycles)

In the source-code on the this Maggie-disk only D0, A0 and A1 are 
used for all frequency switches.  How is this possible? Usually a 
data-register is needed for synchronisation,  but you can use the 
fact that D0 should always contain zero:

WAIT:     MOVE.B    $FF8209.W,D0   ; Low-Byte
          BEQ.S     WAIT           ; mustn't be 0 
          NOT.B     D0             ; negate D0
          LSR.B     D0,D0          ; Synchronisation

After  the LSR-command is executed,  D0 contains zero  for  sure. 
This  zero  can than be used for the  border-routine.  The  other 
data-register  D1 must contain 2.  This register is  superfluous, 
because   you   can  get  the  2  from  somewhere   else   (D0=0, 
A0=$FF820A.W):
        
          MOVE.B    D0,(A0)        ; 60 Hz
          MOVE.W    A0,(A0)        ; 50 Hz

This  step  is not very easy to understand.  The  second  command 
writes the word $820A to $FF820A.  The second byte of the word is 
meaningless,  because  the address $FF820B of the shifter is  not 
used.  The  shifter  only reads the lower two bits of  the  first 
byte ($82), that means that the byte is handled like $02!
The  source-code  on the disk includes a  subroutine  that  tests 
which MMU is present,  that means how many lines the MMU displays 
with an opened upper border (229 or 213).  The program should run 
on every ST,  STe,  Mega ST and Mega STe (not on TTs). It took me 
more  than a month to test the routine on all  possible  ST-chip-
configurations (Thanx to Alien/ST-CNX).  There are more than  two 
dozens  ST-versions.  Atari  changed the Glue,  the MMU  and  the 
Shifter quite often.
It  is  possible to open two more lines of the  lower  border  by 
switching  to 60 Hz for a second time at the end of the  "normal" 
lower border.  Unfortunately this does not work on the new  Atari 
colour  monitors  (Thanx to Aeon/Aura for telling  me!).  For  my 
"Punish Your Machine"-screen I had the idea to code a  fullscreen 
in medium resolution.  It worked fine on my ST very soon,  but on 
the ICC #2 I discovered that the screen only worked on my ST! New 
Mode  made  it work on all STs two months  later,  allthough  all 
"border-experts" on the ICC #2 said that it's impossible to  code 
a fullscreen in medium resolution.
The most difficult thing writing a fullscreen is that the  entire 
code  has  to  fit  synchronous  in  the  waiting  gaps  of   the 
fullscreen-routine.
A  few  days  after  the ICC #2 I  developed  a  new  fullscreen-
routine.  With this routine the main program does not have to  be 
synchronous  with  the electron-ray.  The  disadvantage  of  this 
method  is that the border annihilation takes up to 50  per  cent 
of the processor time.  Therefore you can use  keyboard-requests, 
multiplications,   divisions  and  interrupts.  The  routine  is, 
allthough  it's awfully old,  top secret and cannot be  published 
in Maggie (yet).
Ok,  I  hope  you  understood  how you make  your  ST  display  a 
resolution  of  416*276 pixels.  In the next part New  Mode  will 
explain  how you can scroll the whole screen in 4 planes using  7 
scanlines of processor-time.

Bye...
Flix/DF (The Union)