- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hey,
Has anyone experienced in slowdown of execution on the host due to offloading using LEO or COI? I know there is an entry of a known issue in the MPSS changelog where remote process creation causes a slowness in the host and I'm not sure if that implies the effect I'm seeing.
I'm creating two threads at the start of a function, one of the threads uses TBB to run a plain data parallel computation (BlackScholes), while the other is executing code on a Xeon Phi. If I don't execute code on the MIC then the code on the host completes 5x faster. This remains even if I reduce the number of worker threads in the TBB thread pool. I've reduced the work on the MIC thread too such that it finishes before the CPU without effect.
Has anyone experienced something similar? Potential avenues to debug or compiler/runtime settings to use?
Best,
Jacques
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Jacques,
I am curious to know how you are timing you application. Are waiting for the code to complete running on the coprocessor?
It would be very helpful if you could share a reproducer with us.
Thanks,
Sumedh
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page