You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I have an issue with zeMemAllocHost-allocated memory incombination with a memset-like kernel. If I allocate the host memory, memset it on the device, copy it back using zeCommandListAppendMemoryCopy, everything is OK. If however instead of enqueuing a memory copy, I synchronize the command stream and use a regular CPU-based memcpy, I get the wrong results. Isn't host memory supposed to be readable from arbitrary CPU code? If it is, maybe the driver isn't fully synchronizing the command stream when calling zeCommandQueueSynchronize?
I have an issue with
zeMemAllocHost
-allocated memory incombination with amemset
-like kernel. If I allocate the host memory,memset
it on the device, copy it back usingzeCommandListAppendMemoryCopy
, everything is OK. If however instead of enqueuing a memory copy, I synchronize the command stream and use a regular CPU-basedmemcpy
, I get the wrong results. Isn't host memory supposed to be readable from arbitrary CPU code? If it is, maybe the driver isn't fully synchronizing the command stream when callingzeCommandQueueSynchronize
?Here's a reproducer:
This reads
kernel.spv
, attached here: kernel.zipOriginally reported in: JuliaGPU/oneAPI.jl#458 (comment)
The text was updated successfully, but these errors were encountered: