Skip to content

Conversation

odiazib
Copy link
Contributor

@odiazib odiazib commented May 10, 2025

Using a regular range policy instead of MDRangePolicy, as it has been reported that the latter has performance issues. We can revert this commit once the Kokkos team addresses these concerns.

@bartgol
Copy link
Contributor

bartgol commented May 12, 2025

Did you measure the perf hit? If you did, and there is a sizeable hit, then go ahead. Otherwise, I think MD policies are much more self-expressive and compact (and less bug prone) than manual index arithmetic...

…as been reported that the latter has performance issues. We can revert this commit once the Kokkos team addresses these concerns.
@odiazib odiazib closed this May 13, 2025
@odiazib odiazib force-pushed the eamxx/remove_MDRangePolicy branch from d90282e to 811f283 Compare May 13, 2025 14:45
@odiazib
Copy link
Contributor Author

odiazib commented May 13, 2025

There is no performance hit for MDRangePolicy in the mam4xx interfaces. See the next figure.
pr
For this reason, I will proceed to close this pull request.

@bartgol
Copy link
Contributor

bartgol commented May 13, 2025

To be clear, compared to master, MD policy was only added in a subset of the aerosols section, right?

@odiazib
Copy link
Contributor Author

odiazib commented May 13, 2025

To be clear, compared to master, MD policy was only added in a subset of the aerosols section, right?

Yes, we have only replaced MDRange policies in two of the MAM4xx interfaces: microphysics and optics.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants