Add MPI support for conforming P4estMeshes in 2D
#977
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
If the p4est library does not support MPI, it is not allowed to pass an MPI communicator when creating a new p4est object. If it does support MPI, an MPI communicator has to be passed even if only one process is used.
This enables adding methods specialized on the parallel type for MPI runs.
Adds a new container for the interfaces at the MPI domain boundaries and methods for initializing such container and adds specialized versions for some initialization methods for the existing containers for parallel runs (e.g. to distinguish between regular and mpi interfaces).
Codecov Report
@@ Coverage Diff @@
## main #977 +/- ##
==========================================
- Coverage 93.68% 92.18% -1.50%
==========================================
Files 294 295 +1
Lines 22606 22995 +389
==========================================
+ Hits 21177 21196 +19
- Misses 1429 1799 +370
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
sloede
requested changes
Jan 27, 2022
This was referenced Jan 28, 2022
This avoids a few unnecessary "where {IsParallel}" and makes the code
more readable overall.
- New P4est.jl version supports MPI, hence enable `ParallelP4estMesh` tests - `p4est_has_mpi()` function no longer required as it's not built into P4est.jl
sloede
approved these changes
Feb 4, 2022
lchristm
changed the title
P4estMeshes in 2DP4estMeshes in 2D
|
From my side this is ready once all checks passed. |
ranocha
approved these changes
Feb 4, 2022
|
Nice work, @lchristm 👍 Out of curiosity: Did you run some benchmarks of MPI vs. multithreading? |
efaulhaber
approved these changes
Feb 4, 2022
|
@ranocha, @efaulhaber I didn't do any extensive benchmarking but I did save some numbers when I checked the correctness of my implementation. The following numbers were obtained by running a modified conforming version of Multithreading with 6 threads: MPI with 6 processes: |
|
Thanks! Did you use multi-threaded or serial RK methods? |
I forgot - does this only time the MPI operation or also the unpacking of buffers? Especially the |
Serial. I think that's where difference in runtime is coming from.
It includes unpacking the buffers. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds containers and solver functions for running simulations on a conforming 2D
P4estMeshin parallel. TheP4estMeshhas been slightly adapted so that parallel and serialP4estMeshes can be distinguished which allows for elixirs using theP4estMeshto "just work" when executed in parallel, similar to theTreeMeshimplementation.There is a new container for interfaces shared by two MPI domains (MPI interfaces). Furthermore, there are new functions for initializing the parallel and serial containers for parallel runs (see
containers_parallel.jlandcontainers_parallel_2d.jl). Similar to theTreeMeshimplementation, there is also a new cache for holding data structures related to MPI communication (seedg_parallel.jl).Todo: