Added threading to ABM algorithms using @.. #2556
Conversation
1. in *_perform_step.jl
in functions using @..
in unpacking of the "cache" object
add a "thread" variable
2. in *_caches.jl
for the cache types affected
(determined in 1.)
- add parametric type Thread
- and a field thread::Thread
- in the alg_cache function,
add 'alg.thread' as argument to the cache type constructor
3. in *_algorithms.jl,
for the alg type affected
(determined in 2.)
- add parametric type Thread
- add field thread::Thread
- add outer constructor taking a thread argument
with default value False(),
that builds the respective algorithm object
forwarding the thread parameter to it.
|
This is set as a draft, but it seems like it's correct? What's the incomplete part here? |
Still need to make sure all the boxes can be ticked! |
|
If you can add some regression tests for the threading choices then I think that's sufficient. |
Run the reformatting command suggested in CONTRIBUTING.md, but keeping only the changes on the ABM directory (which my current PR is affecting)
Where should those tests go? Should they be included in the |
|
https://github.com/mmesiti/OrdinaryDiffEq.jl/blob/multithreaded-ABM/lib/OrdinaryDiffEqAdamsBashforthMoulton/test/abm_convergence_tests.jl extend the convergence tests to have a few of the multithreaded choices. |
Checklist
contributor guidelines, in particular the SciML Style Guide and
COLPRAC.
Additional context
Add any other context about the problem here.
Referring to this discussion: https://discourse.julialang.org/t/differentialequations-jl-mpi-jl-pencilarrays-jl-lack-of-scaling-observed/122998/18?u=mmesiti