Compatibility with multithreading

[mjyshin]

I wanted to look at a range of parameters on the same system, so I tried using Threads.@threads on the for loop, and I got the following error:

    nested task error: Allocations in assembly loop: cells: 208, bfaces: 0
    See the documentation of `check_allocs!` for more information

It runs just fine if I run it in series, but I thought I would try to speed up the process any way I could. Is this something VoronoiFVM cannot handle (yet), or did I fumble somewhere? Thank you!

[j-fu]

Hi, thanks for asking. Parallelization is in the plans, but this will need some work:

• grid partitioning + partition graph coloring
• multithreaded sparse matrix assembly
• parallel iterative solvers The later point possibly can be handeled relatively independently, there is a master thesis running on this and we will see if this functionality could be added first.

In the moment the multithreaded loop would end up in write conflicts.

[mjyshin]

Thank you for your reply, that sounds great, and I will patiently await the updates!

Two unrelated questions I had were 1) if there was a way to do unstructured/random Delaunay triangulation, even on a domain like (0,1) x (0,1), and 2) if you had any plans for an adaptive (i.e., time dependent) mesh at any point in the future?

[j-fu]

Hi,

as for 1) Have a look at SimplexGridFactory.jl - this package has been made for this purpose. For 2): this has lesser priority than other things (and would need some input from math regarding error estimators etc.)

[j-fu]

Multithreaded assembly has now bee implemented starting with v1.22, but not yet well documented. In the moment, see Examples 201 and 206 for clues. Please note that for being efficient, it needs to be combined with multihreaded linear solvers like Pardiso.jl or AMGCLWrap.jl. More options in this direction are WIP...