nV News Forums

 
 

nV News Forums (http://www.nvnews.net/vbulletin/index.php)
-   NVIDIA Linux (http://www.nvnews.net/vbulletin/forumdisplay.php?f=14)
-   -   1.0-9625 GLSL rendering bug (http://www.nvnews.net/vbulletin/showthread.php?t=77720)

tachyon_john 10-05-06 03:26 PM

1.0-9625 GLSL rendering bug
 
3 Attachment(s)
Hi,
I've begun testing the latest 1.0-9625 beta Linux driver release (as have various users of our molecular graphics software). The good news first: the new version has fixed long standing problems I previously reported with the GLSL compilers emitting internal compiler errors and failing. I was happy to see this, and this is an important step forward from the way things have been for a number of months with the older driver series. With the new drivers, a number of correctly-written GLSL shaders that previously failed to compile now compile without errors or warnings. The new drivers do have a problem however, in that one of the GLSL shaders used by VMD (the sphere shader) fails to execute correctly, resulting in spheres not being displayed at runtime. I've attached the sphere shader and the nvidia-bug-report.log file for your perusal. I have previously posted complete test scripts along with sample output images and instructions here, and these existing scripts illustrate the problem clearly (all of the atom spheres disappear when the rendering mode is changed from 'Normal' to 'GLSL'...)
http://www.ks.uiuc.edu/Research/vmd/alpha/test/

Let me know if I can be of assistance in debugging this problem.

Cheers,
John Stone
vmd@ks.uiuc.edu

netllama 10-05-06 03:42 PM

Re: 1.0-9625 GLSL rendering bug
 
John,
I'm not really clear on how to reproduce the problem you're reporting. The tarball that you referenced seems to contain the same content that was used in the previous bug (which you're now reporting as resolved in 1.0-9625).

Can you please provide detailed steps describing how to reproduce this problem?

Thanks,
Lonni

tachyon_john 10-05-06 04:46 PM

Re: 1.0-9625 GLSL rendering bug
 
Hi,
The tarball is indeed the same one used to reproduce the previously reported bugs,
and the steps are the same as before. Only the results are different, the procedure for running the tests is the same. For 1.0-9625 the result this time is that all of the spheres in the test scene are missing when GLSL is enabled. (previous driver versions had compiler errors etc, 9625 has fixed the shader compiler problems, but it executes the sphere shader incorrectly..) Does that help clarify? The test case I have in that tarball is pretty generic, and can be used to test any driver version, of any video card, on any of the operating systems supported by VMD...

Cheers,
John

tachyon_john 10-05-06 05:12 PM

Re: 1.0-9625 GLSL rendering bug
 
Hi,
I have just traced down the problem with the 9625 driver...
It appears that the problem is related to the execution of code that sets the OpenGL fragment depth:
// Output the ray-sphere intersection point as the fragment depth
// rather than the depth of the bounding box polygons.
// The eye coordinate Z value must be transformed to normalized device
// coordinates before being assigned as the final fragment depth.
if (vmdprojectionmode == 1) {
// perspective projection = 0.5 + (hfpn + (f * n / pnt.z)) / diff
gl_FragDepth = 0.5 + (vmdprojparms[2] + (vmdprojparms[1] * vmdprojparms[0] /
pnt.z)) / vmdprojparms[3];
} else {
// orthographic projection = 0.5 + (-hfpn - pnt.z) / diff
gl_FragDepth = 0.5 + (-vmdprojparms[2] - pnt.z) / vmdprojparms[3];
}

When I #ifdef out the fragment depth code, the spheres render on the screen as one would expect. This tells me that the fragment depth portion of the shader is not being executed correctly for some reason. The fragment depth equations are standard OpenGL and are documented in various books.

Cheers,
John

netllama 10-05-06 05:25 PM

Re: 1.0-9625 GLSL rendering bug
 
Thanks for clarifying. I've reproduced this problem, and opened bug 258748.

Thanks,
Lonni

tachyon_john 10-05-06 05:30 PM

Re: 1.0-9625 GLSL rendering bug
 
Excellent, that's great to hear. If I manage to come up with any additional info about this I'll let you know.

netllama 10-06-06 09:43 AM

Re: 1.0-9625 GLSL rendering bug
 
John,
The developers looking into this bug are asking in which driver version this last worked as expected? Also, have you tested this with any other vendor's graphics cards?

Thanks,
Lonni

tachyon_john 10-06-06 05:11 PM

Re: 1.0-9625 GLSL rendering bug
 
Lonni,
Linux driver 1.0-7676 worked fine with the fragment depth and lighting portions of the shaders. (though it had a different bug in another area as I recall)
I believe the most recent nVidia driver that I tested that worked 100% perfectly was
the nvidia Windows 8421 driver. The nvidia Windows 9131 driver had working fragment depth, but broken 3-D texgen handling.

I have also tested these shaders extensively on the 3DLabs Wildcat Realizm cards and they are 100% bug-free with the Realizm 4.05-0854 drivers which are the most recent ones I've tested with.

Let me know if I can provide other helpful information.

Cheers,
John Stone

tachyon_john 10-10-06 01:22 PM

Re: 1.0-9625 GLSL rendering bug
 
Hi,
Any word from the developers on whether or not the driver versions I listed were helpful?

John

netllama 10-10-06 01:24 PM

Re: 1.0-9625 GLSL rendering bug
 
They're still investigating.

Thanks,
Lonni

bram 11-16-07 03:53 PM

Re: 1.0-9625 GLSL rendering bug
 
Is this issue fixed in the current drivers?

AaronP 11-16-07 04:22 PM

Re: 1.0-9625 GLSL rendering bug
 
Yes, that bug should have been fixed in 96.43.01.


All times are GMT -5. The time now is 10:24 PM.

Powered by vBulletin® Version 3.7.1
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Copyright 1998 - 2014, nV News.