-
Notifications
You must be signed in to change notification settings - Fork 124
[DevASAN] Do allocation with USM pool to reduce memory overhead #2394
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
Conversation
|
It might be worth considering that this will raise the bar to entry for supporting dev asan on different platforms in the future, the native_cpu and opencl adapters don't currently have usm pool support and I don't know of any plans to implement it in the near future. If that is a concern I suggest a fallback path if pooling isn't supported, if it isn't feel free to ignore. |
Thanks for your suggestion. But original goal of this PR is to reduce memory overhead on L0 GPU device, opencl adapter doesn't support usm pool is ok for us now. |
|
Hi, I have a patch on syclos which needs this PR to merge, so I update my UR tag to this one. LLVM: intel/llvm#16279 |
|
Hi @oneapi-src/unified-runtime-maintain, could you please help review this PR? We need this PR in v0.11.x, thanks. |
Release mapped physical memory according to its dependency may cause some problems. So, we decide to use USM pool to do allocation to reduce memory overhead.
8151ca1 to
b209eba
Compare
UR Part: oneapi-src/unified-runtime#2394 --------- Co-authored-by: Kenneth Benzie (Benie) <[email protected]>
UR Part: oneapi-src/unified-runtime#2394 --------- Co-authored-by: Kenneth Benzie (Benie) <[email protected]>
Release mapped physical memory according to its dependency may cause
some problems. So, we decide to use USM pool to do allocation to
reduce memory overhead.
intel/llvm: intel/llvm#16280