Port MPI to Taal

[ranocha]

As discussed, I've tried to use a minimally invasive approach to port MPI to Taal. In most cases, I dispatch on the mesh type. In the future, I would like to create an MPI array type for u etc., which handles gather operations etc. to enable error-based timestep control in DiffEq. However, that's left for future work and can be done once we've settled on our approach to MPI.

Closes #287

[codecov]

Codecov Report

Merging #292 (4c716ea) into dev (2a9d8c7) will increase coverage by 0.13%.
The diff coverage is 89.23%.

@@            Coverage Diff             @@
##              dev     #292      +/-   ##
==========================================
+ Coverage   88.44%   88.57%   +0.13%     
==========================================
  Files          87       90       +3     
  Lines       13460    13765     +305     
==========================================
+ Hits        11905    12193     +288     
- Misses       1555     1572      +17     

Flag	Coverage Δ
unittests	88.57% <89.23%> (+0.13%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
src/io/io.jl	86.07% <ø> (-2.05%)	⬇️
src/io/parallel.jl	84.69% <ø> (-1.26%)	⬇️
src/mesh/parallel.jl	96.10% <0.00%> (ø)
src/semidiscretization.jl	76.33% <ø> (ø)
src/solvers/dg/2d/dg.jl	91.90% <ø> (ø)
src/solvers/dg/2d/parallel.jl	93.77% <60.00%> (ø)
src/callbacks/stepsize_dg2d.jl	86.20% <75.00%> (-4.27%)	⬇️
src/mesh/mesh.jl	80.70% <76.47%> (+0.32%)	⬆️
src/callbacks/analysis.jl	83.49% <78.68%> (+0.32%)	⬆️
src/callbacks/save_restart_dg.jl	85.10% <86.44%> (+2.60%)	⬆️
... and 25 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2a9d8c7...4c716ea. Read the comment docs.

[sloede]

So, first round of reviews done. Great job in porting all that MPI stuff to Taal! There are still a few issues to iron out, but nothing I see is unsolvable.

If it helps, we should consider setting up a call to talk about some of the larger issues instead of long a back-and-forth via comments. Just let me know...

[sloede]

LGTM and can be merged when all tests pass. Great work!

By the way, have you ever seen one of the MPI tests fail? IIRC, I've never experienced that before, so I cannot say for certain that the overall testing would fail in that case.

[sloede]

By the way, have you ever seen one of the MPI tests fail? IIRC, I've never experienced that before, so I cannot say for certain that the overall testing would fail in that case.

Or put differently: Should the tests currently pass, we have a problem with MPI testing (caused by me, since I set it up in the first place), since currently a file is tested that does not even exist:

Trixi.jl/test/test_examples_2d_parallel.jl

Line 32 in f774c9d

@testset "elixir_hyp_diff_llf.jl" begin

[sloede]

As expected. The MPI tests should have failed but they don't.

[ranocha]

As expected. The MPI tests should have failed but they don't.

Why should they have failed? We get

Test Summary: | Pass  Test Summary: | Pass  Total
Total
Examples 2D   |   82     82
Examples 2D   |   82     82
799.767451 seconds (1.93 M allocations: 96.366 MiB, 0.00% gc time)
Test Summary:  | Pass  Total
Trixi.jl tests |    1      1
800.634312 seconds (3.64 M allocations: 183.784 MiB, 0.02% gc time)
    Testing Trixi tests passed 

in https://github.com/trixi-framework/Trixi.jl/pull/292/checks?check_run_id=1390251395.

[sloede]

Ah, never mind. Didn't see that you fixed the elixir names.