Add possibility to configure dynamic shared memory on CUDA backend. #767
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
It can be configured via 'sharedMemBytes' field of type int in the kernel properties. Also example of dynamic shared memory usage is added. Note that transpiler doesn't support this feature, so it is usable with only native cuda kernels for now.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.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.
Dynamic shared memory might be required in some applications, since CUDA allows to allocate more shared memory per block with it.
In this PR It is proposed to add a possibility to configure dynamic shared memory for native CUDA kernels with 'sharedMemBytes' field of type int in the kernel properties.
Also an example of dynamic shared memory usage is added.
Similar thing can be done for HIP backend, but I couldn't get my hands on an AMD gpu right now.