PDA

View Full Version : strange Display list problem



zed
04-10-2006, 11:44 PM
ive had this problem thats been bothering me for a while, so i decided to work until i solved it today, anyways heres what i came up with + im completely stumped

version A+B are used somewhere in my code and should have no influence with the rendering
version A/
while [ count++ < 5 ]
func[]; // only cpu stuff ie no graphics calls

version B/
while [ count++ < 1 ]
func[]; // only cpu stuff ie no graphics calls


the following rendering funcs are elsewhere in my code ie the above funcs shouldnt interfer at all

~0.3ms rendering with displaylists version B // expected
~5ms rendering with displaylists version A // NOT expected

~1.1ms rendering with vertex arrays version B // expected
~1.1ms rendering with vertex arrays version A // expected

so whats happening? somehow a slightly more taxing cpu version is somehow making the version with DLs fall apart, has anyone seen something similar to this before?

zed

sqrt[-1]
04-11-2006, 05:30 AM
Humm, are you 100% sure there is no difference in the rendering calls between the two versions?

Perhaps do a XML/text frame grab with GLIntercept on both versions then do a directory "diff"? (ensure to compare the dumped display lists)

jide
04-11-2006, 03:41 PM
I don't understand a word zed.

zed
04-11-2006, 05:07 PM
ok ive narrowed down the problem.
i tried GLIntercept (sweet looks like u only log DLs onscreen, ie not all current ones), also perhaps having the option to log a certain frame number would be an idea instead of triggering it with a key press.
anyways there was no differences between the two.

now im thinking (my app uses quite a bit of memory + must be getting close to the cards 128mb limit, in fact i had to downsize a few textures to get it to work).
so i decided to free up a bit of card memory by disabling antialiasing, now both DL versions run at the same speed (as they should)

ie heres the results

antialiasing enabled
~0.3ms rendering with displaylists version B // expected
~5ms rendering with displaylists version A // NOT expected

antialiasing disabled
~0.3ms rendering with displaylists version B // expected
~0.3ms rendering with displaylists version A // expected

i wouldnt say this was a driver bug, but its certainly a driver issue that ppl should be aware of (ive never read about anything similar).

ild be happy to email a test case to the nvidia guys if necessary.

jide - i thought this was one of my more legitable posts? then again i am insane so what do i know, cocopops.

jide
04-12-2006, 03:28 AM
Don't be insane :) I just wanted to understand what you mean: it's not so easy for me.

I guess now I understand more what you wrote.