-
Notifications
You must be signed in to change notification settings - Fork 117
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.
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
Consider interface to expose P2P capabilities #68
Comments
Add requirements for each adapter to a table on this issue to kick off the design process of a combined multi-device context and P2P proposal. |
related: this is what L0 has https://spec.oneapi.io/level-zero/latest/core/PROG.html#peer-to-peer-access-and-queries |
FYI I made a draft impl (https://github.com/intel/llvm/pull/8303) of Intel's sycl extension proposal for P2P USM: intel/llvm#6104. This is fully functional now for the cuda backend only, but should support HIP and L0 in the future using the same pi functions. I'm pretty sure that these three functions will be required, although the naming is subject to change, particularly "piextCanAccessPeer". |
@kbenzie : what is the next step on this issue? are there any points to clarify? |
I asked @JackAKirk to bring these changes over to UR here. |
Added in #631 |
Some devices have peer to peer capabilities for memory transfers and/or USM buffers, we may need to consider a way to expose this in the unified runtime.
There's currently an extension proposal in DPC++ to handle this, with some discussion on it:
It's currently only a SYCL level extension but I suspect implementing it would require some changes in PI or UR, so we should consider it.
The text was updated successfully, but these errors were encountered: