In the project I'm working on, I need to execute a compute shader. It's working, but after I did some modifications to the code calling `vkCreateComputePipelines` started to return the `VK_ERROR_INITIALIZATION_FAILED` error code.
Here is a reduced case of the source code of the shader that triggers the problem: bug.glsl · GitHub
Removing either lines 54-55 or lines 70-73 fixes the problem (and the compute shader then does work). This leads me to think that it's a low-level problem in either glslangValidator or the driver, and not a problem in my code or API usage. Someone on the ##vulkan IRC channel didn't spot anything wrong in the SPIR-V code and thinks it's probably caused by the driver.
The SPIR-V output has been attached. I'm using the latest commit of the glslang repo (7e3e486).
I can send an executable in private if needed.
Windows 7 64bits
Radeon HD 7950
Driver version 16.15.2111.1001 Beta1
Solved! Go to Solution.
We just released a new Vulkan BETA (16.15.2401.1001) which should resolve this issue, along with others:
http://support.amd.com/en-us/kb-articles/Pages/Radeon-Vulkan-Beta.aspx
Please let me know if it does!
This problem is really starting to block me 😕
Any non-trivial modification to the code posted above triggers an INITIALIZATION_FAILED.
Please try updating to the latest AMD Driver 16.4.1 (16.15.2211.1001):
AMD Radeon Software Crimson Edition 16.4.1
If you're still having an issue afterwards, please reply back.
The new version didn't fix the problem.
We just released a new Vulkan BETA (16.15.2401.1001) which should resolve this issue, along with others:
http://support.amd.com/en-us/kb-articles/Pages/Radeon-Vulkan-Beta.aspx
Please let me know if it does!
Thank you! The particular code I reported now works.
Unfortunately when I tried restoring the real code, another INITIALIZATION_FAILED error happened (gist here).
Should I open another topic, or can we continue on this one?
Does this problem continue to occur with the latest driver version?