Hello,
I have a RX 5700 XT. My application is calling `imageAtomicAdd(uimage3D, coords, data)` in a fragment shader which works fine on drivers released before 22.7.1 and on an other nvidia gpu. Since 22.7.1 this function has no effect. Not only that but it also causes all other images writes in the same shader to also have no effect. Regular imageStore works but thats not sufficient.
It's a dsa immutable 3D texture with GL_RGBA8 as it's internalformat. I bind the image by calling glBindImageTexture(0, texture, 0, true, 0, GL_READ_WRITE, GL_R32UI).
Also one thing I noticed is that when using the imageAtomicAdd the program runs unbelievably fast (measured with timer queries).
Solved! Go to Solution.
Update:
The OpenGL team has implemented a fix for this issue.
Thanks.
Hi @BoyBaykiller ,
Thank you for reporting it.
Could you please provide the source code that reproduces the issue? I will forward the issue to the OpenGL team.
Thanks.
I've uploaded the shader program code.
Vertex: https://gist.github.com/BoyBaykiller/d0e8217146df7f9975d783f6c4ae06ed
Geometry: https://gist.github.com/BoyBaykiller/9f2bbd6eed6145747fbdd57d1f251c43
Fragment: https://gist.github.com/BoyBaykiller/88ca63f7d8f119ab5e3b90289dd258c9
The important part is in the fragment shader. I've added some comments to explain my findings. I hope this is enough - let me know if it's not and I see what I can do.
Thanks for providing the reproducible code. I have reported the issue to the OpenGL team.
A bug ticket has been created to investigate the issue. I will let you know if I get any update on this.
Thanks.
Update:
The OpenGL team has implemented a fix for this issue.
Thanks.
I can confirm this has been fixed in 23.7.1. Thanks!
Thanks for the confirmation.