13 Mar
2024
13 Mar
'24
10:50 a.m.
On Wed Mar 13 10:49:09 2024 +0000, Giovanni Mascellani wrote:
The failures were most likely caused by mishandling of barriers in the
virtualised control flow, resulting in the UAV being written too early, but this is hard to prove. Yeah, that totally makes sense. And indeed I could reproduce the bug on my system using Mesa 22.3.6, but using the same build and cherry picking !711 the bug is solved. Thinking about it, I even wonder how that works on recent Mesa, the old structurizer messes up with control flow completely. The new one should be much better.
Maybe it makes sense to delay this a little bit to have it go in after !711, so you don't even have to add `bug_if()` support in the shader runner.