Mail Archives: djgpp/1997/10/23/01:31:26
Alistair Bain writes:
>I'm programming modex and have the routine running at 60fps no problem.
>As soon as I initialise the sound I start getting jerks in the motion,
>if I cut the number of sprites down to, say, 1 then I still get the same
>problem, what's going on?!!
Being in mode-X, you are probably using some kind of page flipping
system, right? I can explain what is causing the problem, but I'm afraid
I don't have any easy answer about how to fix it :-)
Probably the worst thing about the PC hardware is that there is no
interrupt for the video retrace, so you have to poll a status bit to
determine when it is taking place (actually a few cards do support an
interrupt, but not in a sufficiently standard or widespread way for it
to be useful). So any time that your program syncs with the retrace, or
does a page flip (which requires a retrace sync to know when the flip
has taken place), it has to sit there in a little loop polling the
retrace bit in the VGA card until it changes from a zero to a one :-)
The trouble is, if a soundcard interrupt comes along at the wrong time,
it is possible for the soundcard handler to get control just as the
retrace begins, and release control just after it is finished, so this
little poll loop never notices that the retrace has occured. Result: it
waits until the next retrace, and your animation skips a frame.
How to avoid this is a big problem, and there are no easy answers...
Fix #1: don't use Win95. This problem seems to be far less frequent in a
clean DOS environment, I've no idea why.
Fix #2: don't sync with the retrace at all :-) That is just ducking the
issue, but it does work! Maybe you could provide both a mode-X page
flipping system, and a simple RAM->VRAM double buffer without any
retrace sync, and let people choose whichever works best. Some machines
(like mine) never suffer from the problem, I suppose because the
soundcard processes quickly enough, or the retrace interval lasts longer
than normal, while on other systems it seems to happen all the time...
Fix #3: do clever things with the timer chip to detect when the retrace
should be. Allegro can do this in a clean DOS environment (use the
timer_simulate_retrace() function), but unfortunately that doesn't work
under Win95 because Microsoft don't provide good enough emulation of the
PIT hardware. Try running the Allegro demo.exe in a mode-X resolution,
though, and notice the difference between when you select the retrace
synced vs. normal page flipping modes...
Fix #4: (hypothetical, not yet any use at all :-) Wait a year or so
until you can use the VBE 3.0 or VBE/AF 2.0 interfaces. Neither of these
has yet been ratified, let alone actually implemented, and I've yet to
see a copy of the specs, but according to Kendall Bennett (from SciTech)
they both provide a triple buffering mechanism. This will allow you to
request a page flip, carry on doing other stuff in your mainline code,
and then at some later point poll the video card to see if the flip has
yet occured. This will be a kind of "dirty" flag rather than a "retrace
in progress" bit, so it won't matter if you are a little bit late in
detecting it, and all our problems will be solved! I don't think this is
possible with the standard VGA register set, though, so it will probably
only work on SVGA boards and will need a driver like UniVBE...
--
Shawn Hargreaves - shawn AT talula DOT demon DOT co DOT uk - http://www.talula.demon.co.uk/
Beauty is a French phonetic corruption of a short cloth neck ornament.
- Raw text -