View Full Version : imageAtomicCompSwap not working on AMD 12.6?
07-04-2012, 05:49 AM
it seems to me the issue with imageAtomicCompSwap as detailed in this thread
has returned. All of my tests indicate that imageAtomicCompSwap does not write to the image (the return value seems to be ok though). The simplest way to reproduce the issue is to run the simple demo program provided in the thread above (https://github.com/jdupuy/imageAtomicCompSwap/zipball/master).
Additonal Infos: I'm running the latest Win 7, 64Bit, and my GPU is a ATI Radeon HD 5970
08-04-2012, 11:52 AM
Same issue here on Gentoo (amd64) with fglrx 12.6 and HD 5970 GPU.
Unfortunately, imageAtomicExchange doesn't appear to work either. :(
A friend who is using a HD 7950 GPU (also Gentoo/fglrx) confirmed that imageAtomicCompSwap is broken for him too, but imageAtomicExchange works (at least until he resized the window).
08-12-2012, 04:57 PM
I did some tests with the example program from Github. All results are from my HD 5970 with driver version 12.6:
08-17-2012, 09:14 AM
Still broken for driver version 12.8.
Tested on ATI Radeon HD 7970.
11-06-2012, 08:22 AM
With the latest 12.11 beta driver ( http://www2.ati.com/drivers/beta/amd-driver-installer-catalyst-12.11-beta-x86.x86_64.zip ) it still doesn't work - all while the driver still claims GL 4.2 and GL_ARB_shader_atomic_counters with my 7950. Worse even, all drivers after 12.8 have been crashy as fuck - literally no bugs fixed while new ones are introduced.
AMD/ATI really needs to get their shit together, especially with Nvidia putting more resources into their driver development with the whole Valve deal recently.
11-16-2012, 09:14 AM
I've tried running the linked example program. There are two issues:
1) Under core and compatibility profiles, with both imageAtomicCompSwap and imageAtomicExchange, the output of the program is "buffer content : 0 1 2 3 4 5 6 7", which I assume is correct.
2) The application uses image operations in a vertex shader. Our drivers currently report GL_MAX_VERTEX_IMAGE_UNIFORMS as 0 - i.e., not supported. The application is at fault - our bug is that we don't report an error.
If someone has a correctly written, legal application that reproduces an issue, we'd be more than happy to take a look at it. Feel free to PM me here if you wish.
Powered by vBulletin® Version 4.2.2 Copyright © 2014 vBulletin Solutions, Inc. All rights reserved.