-
-
Notifications
You must be signed in to change notification settings - Fork 54
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
Batch mode #552
Comments
Yeah we really need to do this at some point. @avik-pal you setup a batch mode for SimpleGMRES at one time? Is that code still around? |
That is a different kind of batching. It is essentially |
@avik-pal I reviewed SimpleGMRES again, but I’m still not entirely sure what it's doing. It does make sense to combine some operations to leverage BLAS2/BLAS3 instead of BLAS1. I quickly opened PR #554 in case it’s helpful. |
No, that's the batching @avik-pal was talking about. The most normal batching is But it is natural to want the B batching. I think we'd need a |
Hey there,
I am trying to optimise the ConScape.jl package, a library for ecological connectivity analysis. The computational bottleneck relates to these lines, where a large linear problem is solved.
https://github.com/ConScape/ConScape.jl/blob/c6d8e3d82a26f181c4f734f5d34e6dcd1b3dc99b/src/gridrsp.jl#L21-L25
I'd love to be able to try some of LinearSolve.jl solvers for this task, but it requires batch mode, a feature that is supported by
\
:but not by LinearSolve.jl
I have suggested this feature on Slack in July, but I make it official here. @amontoison raised the point that other SciML users could be interested, and had some idea on how to implement it.
It is not in my capacity to investigate this, but maybe one of the developers here could have a look at it?
Cheers!
The text was updated successfully, but these errors were encountered: