HYPRE.jl solvers and preconditioners #254
Conversation
Codecov Report
@@ Coverage Diff @@
## main #254 +/- ##
==========================================
+ Coverage 65.58% 68.73% +3.15%
==========================================
Files 12 14 +2
Lines 738 838 +100
==========================================
+ Hits 484 576 +92
- Misses 254 262 +8
📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
fredrikekre
force-pushed
the
fe/hypre
branch
3 times, most recently
from
083c5f5
to
ece4472
Compare
| hcache = cache.cacheval | ||
| if hcache.isfresh_A || hcache.A === nothing | ||
| hcache.A = cache.A isa HYPREMatrix ? cache.A : HYPREMatrix(cache.A) | ||
| hcache.isfresh_A = false | ||
| end | ||
| if hcache.isfresh_b || hcache.b === nothing | ||
| hcache.b = cache.b isa HYPREVector ? cache.b : HYPREVector(cache.b) | ||
| hcache.isfresh_b = false | ||
| end | ||
| if hcache.isfresh_u || hcache.u === nothing | ||
| hcache.u = cache.u isa HYPREVector ? cache.u : HYPREVector(cache.u) | ||
| hcache.isfresh_u = false | ||
| end |
There was a problem hiding this comment.
Does this play nicely with Distributed? So for example, if someone used a DistributedArray A and b, will this nicely use the memory? Copy it again? Distribute the same way? I'm curious because if we could have a default solver choice on DistributedArray values then that will fix a few downstream issues.
There was a problem hiding this comment.
Right now it will only work if you have either set up your matrix and vector with HYPRE to begin with or if you use a SparseMatrixCSC or SparseMatrixCSR. For the latter the matrix will be duplicated inside the hypre library, and some buffers need to be allocated in order to send the data to HYPRE.
I don't think it would work with Distributed or DistributedArray since those are not MPI processes, right?
There was a problem hiding this comment.
Makes sense. With Distributed you could make it use the MPI cluster manager, but that's not too common.
|
I'll merge and format |
|
Thanks! |
This patch adds support to LinearSolve.jl for HYPRE.jl solvers and preconditioners. This is implemented as a package extension and thus requires Julia version 1.9 or higher.
Fixes #106, fixes #167.