Skip to content
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

Add size to parameter space of data layout #11

Closed
Tracked by #2632
simonbyrne opened this issue Apr 16, 2021 · 6 comments · Fixed by #1894
Closed
Tracked by #2632

Add size to parameter space of data layout #11

simonbyrne opened this issue Apr 16, 2021 · 6 comments · Fixed by #1894

Comments

@simonbyrne
Copy link
Member

Some experiments by @vchuravy suggest there are some performance benefits, particularly on the GPU, to having the fast indices (i,j,k and field size) known at compile time. We could do this by adding these sizes to the parameters of the data layout objects.

@simonbyrne
Copy link
Member Author

@jakebolewski
Copy link
Contributor

jakebolewski commented Jan 13, 2022

We might need to add the 5 tuple size (Ni, Nj, Nk, Nv, Nh) to be able to better optimize field / getproperty access for all cases consistently.

@jakebolewski
Copy link
Contributor

jakebolewski commented Jan 13, 2022

A benefit of that would be that the packing data / array wouldn't have to actually be an nd array giving a bit more flexibility there

@charleskawczynski
Copy link
Member

Just to note: Nf is probably not needed in the type space, as it can be statically determined from S and eltype(parent(data::AbstractData)), using DataLayouts.typesize

@charleskawczynski
Copy link
Member

Closed by #1757

@charleskawczynski
Copy link
Member

Going to reopen this, as we do still have some type parameters that are only runtime known. This turns out to be important, based on recent benchmarks (#1893).

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 a pull request may close this issue.

3 participants