Implement DLA in legacy CUDA #2887
Merged
Conversation
This file contains 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Proposed changes
Implement the DLA method in the legacy CUDA code. The NLRatio calls and the plumbing to them are updated following the easy to follow recipe used in the CPU code by @ye-luo.
We don't have established DLA tests for any implementation. I will create some in a later PR.
I verified on a water molecule that the energies were appropriately close to the CPU implementation. In the legacy CUDA code the timings for J1NL* and J2NL* operations not in the output when DLA is enabled indicating that the code is working.
What type(s) of changes does this code introduce?
Does this introduce a breaking change?
What systems has this change been tested on?
Nitrogen ( AMD Rome + NVIDIA V100 ).
Checklist