Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[BUG][onetrace][IMME CmdList] Tool got less kernel calls in report than actual submitted #43

Open
xunsongh opened this issue Aug 24, 2023 · 2 comments
Open

[BUG][onetrace][IMME CmdList] Tool got less kernel calls in report than actual submitted #43

xunsongh opened this issue Aug 24, 2023 · 2 comments

Comments

@xunsongh
Copy link

I made a simple case - submitting and executing 101 kernels (1 M2D and 100 add_kernel) with enabling immediate command list on PVC . I use onetrace and pass the flag -s. I found that in the report, only 1 M2D and 78 add_kernels were captured. And the Append(ns) always be 0.
image
I guess this might be a bug. So I report it to you and look forward to an solution in a quick fix.
Thank you.
@xunsongh xunsongh changed the title [onetrace][PVC][IMME CmdList] Tool got less kernel submissions in report than actual [BUG][onetrace][IMME CmdList] Tool got less kernel calls in report than actual submitted Aug 24, 2023
@xunsongh
Copy link
Author

[Additional]
I tried to add a queue.wait() at the end to make sure every kernels run to an end. But the number of calls is still wrong.
image

@jfedorov
Copy link
Contributor

jfedorov commented Dec 5, 2023

@xunsongh sorry for responding in such a delay. yes - this issue is reproduced (on simplest dpc_gemm sample) and likely caused in runtime or level-zero behavior re- Immediate command lists change. We are exploring the best solution and should update (fix) on this soon.

jfedorov pushed a commit that referenced this issue Dec 15, 2023
@mschilling0
Fix Warnings After Switching to oneAPI 2024.0 release (#43)
166eeba 
* Fix warnings in SYCL code due to icpx/libsycl upgrade.
  * Most of these are due to SYCL 2020 spec deprecations (e.g. `get_pointer` -> `get_multi_ptr`).
* Prevent dlworkload from crashing but provide notice it is not working
  as expected.

Signed-off-by: Schilling, Matthew <matthew.schilling@intel.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@xunsongh @jfedorov