PDA

View Full Version : What’s up with the point_sprites?



Leif
08-25-2004, 10:32 AM
Before the summer I added stars from the Hipparcos catalog to my skydome using gl_points for the faint stars and GL_ARB_point_sprite for the bright ones and everything was just fine and dandy in my own little world.

But now I updated to the latest ATI-drivers and all of a sudden there are no bright stars anymore. Does anyone else get the point_sprites to work with the latest ati-drivers or should I forget the problem and wait for the next release?

Leif…

dorbie
08-25-2004, 01:44 PM
A while back I did the same thing with this catalog (actually a merged variation used by Open Universe), I'm interested in your color calculation from that data to RGB.

However I didn't use point sprites, I just used points and adjusted the brightness using the alpha of the per point color. You could I suppose have simply used color intensity with a suitable blendfunc.

Point sprites seem like way overkill for this kind of thing, did you have a particular reason for using them?

Even point parameters on its own would probably be a better choice.

If this just stopped working then you've probably found a bug on a driver path less trodden, OTOH maybe they fixed something. You should submit a test case that's also broken with the same point parameters and blend operations etc. Maybe run it against Mesa first.

Humus
08-25-2004, 04:53 PM
I recall I verified a point sprites bug a couple of months ago. I don't remember the specifics of it though. It could perhaps be the same bug. If you got a small test app, send it to me at epersson 'at' ati.com and I'll take a look at it.

Leif
08-25-2004, 10:29 PM
@ dorbie, I do blend my points and slightly adjust their size according to their brightness. I’m not using any colours though, hm maybe I should.

Anyhow I couldn’t get the bright stars to stand out as much as they do in reality and thus it was hard to recognise the constellations (the few I knew anyway). So I tried point_sprites for the brightest stars and I think it worked out quite well, perhaps a bit cartoonish but I like it. This is how it looked before the new drivers:
http://w1.510.telia.com/~u51008935/VIIC/stars1.jpg
http://w1.510.telia.com/~u51008935/VIIC/stars2.jpg

@Humus, thanks for the offer, I don’t have a small application to send though so I will dig a little deeper before I give up. Perhaps I’ve missed something.

Leif…

Eric Lengyel
08-26-2004, 06:29 PM
ATI has had some very bad point sprite bugs for several months now. They claimed to have fixed them at one point, but it turned out that they hadn't. We currently disable point sprites altogether when we detect that we're running on ATI hardware.

I can't check whether the latest drivers work because running our engine, or actually anything other than Doom3, crashes my machine so badly that the monitor turns off and I have to reboot. When the drivers are so unstable that they have to include a special VPU Recover mechanism (not that it does a lot of good), you should probably take it as a sign that the software engineering sucks. I'll be sticking with Nvidia, thank you.

zeckensack
08-27-2004, 05:12 AM
Eric,
this may be a little off topic, but anyway: I don't think "VPU Recover" is necessarily an admittal of bad software engineering. I've seen many people bash it, but the point is that it can cover up for hardware instability, mainly overclocking various parts of your system too much and running too aggressive timings, or just flat out misconfigurations (keyword: "PSU"). I'm not claiming that's what you're doing, but very often people just "shoot the messenger". The system would have crashed anyway. VPU Recover just makes it a tiny bit more graceful.