PDA

View Full Version : Problem with sampler objects on Radeon 4800s



lazyeval
12-27-2010, 07:29 AM
Hi, I believe ATI Radeon 4800s drivers have problems with sampler objects.

For example if you have a texture bound to unit 2, and also a sampler object bound to unit 2, its states will not be used for the texture. Also if at the same time you have a sampler object bound to unit 0, it will use their states instead.

I've attached the source code to a test I made, it requires the glew library (http://glew.sourceforge.net/), I used the latest version 1.5.7. Also it's setup to run in windows only. The program doesn't require any resources (e.g. texture is generated using a framebuffer object, and shaders are stored in strings).

You can see that the sampler states are not being used if the wrap_s and wrap_t aren't using the clamp_to_edge (aka a single image at the bottom left) and instead are using the default repeat (aka the image tiled across the screen).

I've tested the program with a ATI Radeon 4850 (using latest drivers) and 4890, which both failed the test, also I had someone else try with a nVidia 8600 M, which passed the test.

I wasn't sure where to post this problem, but this seemed like the best place.

Thanks.

Here's another link (http://files.myopera.com/andrewhh/files/Test.cpp) to the source code, I can't seem to be able to access the attachment.

Alfonse Reinheart
12-27-2010, 12:27 PM
Verified on the HD 3300, driver version 10.8. I haven't installed more recent drivers to test this.

frank li
12-27-2010, 10:55 PM
It seems a bug for the driver. We are looking at it.
Thanks for your feedback.

Chris Lux
12-29-2010, 02:39 AM
I reported this to AMD devrel and Pierre Boudier months ago, do these things not get logged in a bug data base? It is frustrating to read this, as you never saw this problem before....

frank li
12-29-2010, 05:48 AM
I think it's a misunderstanding. Pierre is our fellow and he will assign the bug to others. But it doesn't mean all developers know there exists such a bug. For the person who manages the bug data base, he knows the two bugs are actually the same.

I will ask for the status of the issue after the holiday.

Groovounet
01-07-2011, 05:47 AM
Frank, I am sorry but I agree with Chris on the regard that it seems something is wrong at AMD with the bug report management. I report bugs on sampler object and uniform buffer months ago (3 / 6 months) and I am bit surprised to receive now a mail to confirm a bug that I already confirm remain.

I am happy with AMD concerns for making more stable drivers but I guess something could be better regarding the bug reports management.

Alfonse Reinheart
02-08-2011, 02:34 PM
So, where are we on this bug? Frank said, "I will ask for the status of the issue after the holiday," but is there movement on it?

Groovounet
02-09-2011, 10:36 AM
Not so far :p

frank li
02-11-2011, 07:46 PM
Yes, it's still being investigated.

Chris Lux
02-14-2011, 01:12 AM
wow... just WOW. this should tell us something about the OpenGL priorities at AMD.

Groovounet
02-15-2011, 11:31 AM
Nothing better in 11.2 :p

Alfonse Reinheart
05-01-2011, 06:27 PM
So, the Cat 11.4's are out. Has anyone tried Sampler Objects yet to see if they work?

kozmo
05-10-2011, 01:58 PM
With 11.5, it seems working, at least at first glance

DarkGKnight
05-10-2011, 02:55 PM
Strange ... I get a blank screen whilst using the latest hotfix (Cat 11.5a)

kozmo
05-10-2011, 03:15 PM
I tested this with someone's quick test. At first I also got black screen, but the texture had only 1st mipmap, and the sampler parameters for texture filtering wasn't never set. I can confirm it now reads the sampler parameters from the correct texture unit -checked with few simple demos.

Source (http://pastebin.com/zpGKKSR8) - Credits goes to whoever wrote the demo.

Groovounet
05-11-2011, 06:38 AM
I confirmed that the sampler object is now perfectly working with Catalyst 11.5a.

DarkGKnight
05-11-2011, 01:39 PM
I confirmed that the sampler object is now perfectly working with Catalyst 11.5a.
We now have to wait for the OpenGL 3 Scissor Tests in your sample pack to work.

Groovounet
05-12-2011, 12:23 PM
It almost works actually! Only when rendering tout SRGB colorbuffer fails.

Groovounet
05-12-2011, 12:30 PM
+ ARB_debug_output supports which works fine, apparently exactly the way is works on nVidia which means under debug context only and without asynchronous support but the specification doesn't really state about when the messages are retrieved (as far as I know) so that's fine.

Good release for AMD!

Alfonse Reinheart
05-12-2011, 06:08 PM
They also seem to have fixed a bug regarding arrays in uniform blocks. I was able to remove a hack from one of my tutorials to get around it.

Groovounet
05-13-2011, 04:57 AM
They also seem to have fixed a bug regarding arrays in uniform blocks. I was able to remove a hack from one of my tutorials to get around it.

Doesn't work for me... :p
Can you point me where you fixed it in your tutorial?

Alfonse Reinheart
05-13-2011, 12:50 PM
Can you point me where you fixed it in your tutorial?

Tutorial 13, Geometry Impostors. Specifically, the "GeomImpostor.frag" file. I had to use a vec4 for the "specularShininess" in the MaterialEntry array instead of a float. This was because I couldn't get the array entry padding to work in accord with std140 layout. But with 11.5a, it works.