-
Notifications
You must be signed in to change notification settings - Fork 120
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
[HIP] Revert add prefetch for USM hip allocations a6b8fa66b537753415d24076f… #936
Conversation
Has this been done through |
77845ea
to
7de531a
Compare
Yes it was but I think maybe there was some conflict so added the second commit. Have squashed the second commit so just one commit now |
7de531a
to
b3887e6
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
…f atomic builtin (#11876) 1. Fix a race condition in cmpxchg 2. Change `atomic_xor` to use a CAS loop instead of atomic builtin. This is needed to merge this in UR oneapi-src/unified-runtime#936 so that perf regression can be fixed. The long term fix is to use a compiler flag to choose between builtin and safe CAS implementation, but talks upstream may take some time to figure out the details. See llvm/llvm-project#69229
This needs an intel/llvm testing PR with passing checks before we can merge here. |
b3887e6
to
1151231
Compare
1151231
to
841a287
Compare
Run CI for oneapi-src/unified-runtime#936 Depends on #11718 merging first. --------- Co-authored-by: Kenneth Benzie (Benie) <k.benzie@codeplay.com>
Run CI for oneapi-src/unified-runtime#936 Depends on intel#11718 merging first. --------- Co-authored-by: Kenneth Benzie (Benie) <k.benzie@codeplay.com>
intel/llvm#10430 caused a performance regression. The new way to ensure safety when using xor atomics is by changing the xor impl to use a CAS loop. See here intel/llvm#11876.
Tested in intel/llvm#11893.