06:45 Venemo: karolherbst: it's a good idea to fix the VVL errors first. then, if the issue is still reproducible let us know. if allbos fixes it that could mean that your submission contained a wrong bo
08:01 karolherbst: Venemo: I wonder if this one could mess things up in this area? https://github.com/KhronosGroup/OpenCL-CTS/issues/2489
08:02 Venemo: karolherbst: I'm afraid I need more context to judge that
08:03 karolherbst: I mean, is it plausible that violating this VVL _could_ lead to wrong bos within the command submission
08:03 karolherbst: not trying to get a "this error causes this", just if it's worth digging into this one first
08:05 Venemo: I see
08:06 Venemo: My best guess is that it could. It sounds like there are come commands in progress which access some memory that you are trying to free. So it is plausible that the BO is not included in the submission but it should be.
08:06 karolherbst: there is another one that complains about importing a semaphore into an object currently in used by a queue, but I doubt it's causing this and the others look harmless
08:07 karolherbst: well there is another one about BeginCommandBuffer used on an active command buffer, but...
08:08 karolherbst: to me it sounds like the freememory is the most likely, followed by the command buffer one.. anyway, will probably try to look into those two first
08:08 Venemo: Hmm
08:08 Venemo: What is this? radv+zink+rusticl?
08:09 karolherbst: rusticl+radeonsi and radv on the vulkan side. This is related to a test that exports objects from vulkan and imports them into CL
08:10 karolherbst: so this is just OpenCL CTS code doing Vulkan calls and that is throwing some errors with the VVL
08:10 Venemo: aha
08:11 karolherbst: my code works on iris + anv perfectly fine, but you never know
08:11 Venemo: regardless of whether it works or not, the CTS shouldn't have validation errors
08:11 karolherbst: yeah...
08:12 karolherbst: and I'd rather trust radv here than the CTS :D
08:12 Venemo: haha, thank you
08:13 karolherbst: I'm also hitting weird race conditions with rusticl+zink, but that's just use-after-frees and pipe_resources getting deallocated too early, but that's a zink issue
09:18 Venemo: I empathize. It must be super annoying to deal with this sort of issue