-
-
Notifications
You must be signed in to change notification settings - Fork 195
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
Performance regression in non-autonomous Linear ODE solvers #2121
Performance regression in non-autonomous Linear ODE solvers #2121
Comments
Profile Trace for dense caseProfile Trace for sparse case
Package Status
|
https://github.com/SciML/ExponentialUtilities.jl/blob/c9fc842f00a090feb79dcc460b426d9fdb11359a/src/exp_generic.jl#L126-L143 this dispatch isn't using the alloc_mem cache in its later pieces, so all of the time is re-allocating its memory. We need to alloc_mem in the cache construction here and just reuse that cache, and we should see massive savings. |
@oscardssmith can you find the time for this one as well? Relatively trivial to fix. |
I'm not sure this was fully ready to close. I think I only fixed the sparse regression. |
There are two performance issues when using state-independent solvers for non-autonomous linear ODEs:
ExponentialUtilities.exponential!
instead ofexp
and addmethod
as an option in solvers that currently useexp
andexpv
#2009.Probably caused by Faster linear_perform_step by reducing deepcopies and using sparse matrix when possible. #2022.
These problems persist in the latest v6.69.0.
Here are the timings (in seconds):
MRE:
Pkg.status() on v6.54.0
Pkg.status() on v6.55.0
Pkg.status() on v6.56.0
Pkg.status() on v6.69.0
versioninfo()
The above results were obtained specifically for
MagnusGauss4()
, but several brief tests indicate that the problem shows up for other solvers as well.I am attaching the profile traces.
The text was updated successfully, but these errors were encountered: