Lock constant memory in Cuda/HIP kernel launch with a mutex #4525
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The constant memory used for kernel launches is shared between all
Cuda
/HIP
execution space instances. We already guard the access to make sure that the last kernel using the constant memory has finished before copying the next one. This doesn't prevent (different) kernels submitted from independent threads to be copied to the constant memory in a conflict way. Note that we already have a mutex for the case that the same kernel is submitted from different threads.This pull request adds another mutex that guards access to the constant memory globally.