Upgrade PartitionedArrays from 0.3.x to 0.5.x

[fredrikekre]

No description provided.

[fredrikekre]

Also wanted to ask if using .row_partition and .col_partition like in

HYPRE.jl/src/HYPRE.jl

Line 516 in 5bc2746

map(local_values(B), B.row_partition, B.col_partition) do Bv, Br, Bc

is the correct way?

[fredrikekre]

@fverdugo is using A.block.own_own the correct way or is there some function to extract this part? I saw https://github.com/fverdugo/PartitionedArrays.jl/blob/b211949cc5842f3d6d99f8452b9f0233aef963a5/src/p_sparse_matrix.jl#L670-L672 but not sure what to pass as the second and third argument (although they are unused so doesn't matter...)

[fredrikekre]

Also wanted to ask if using .row_partition and .col_partition like in

https://github.com/fverdugo/PartitionedArrays.jl/blob/b211949cc5842f3d6d99f8452b9f0233aef963a5/src/p_sparse_matrix.jl#L993 looks close. partition(axes(A, 1)) perhaps? Would it make sense to have something like local_axes(A)?

[codecov]

Codecov Report

Attention: Patch coverage is 82.19178% with 13 lines in your changes missing coverage. Please review.

Project coverage is 76.78%. Comparing base (9d35b75) to head (01e91b1).
Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
src/HYPRE.jl	82.19%	13 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #18      +/-   ##
==========================================
- Coverage   77.69%   76.78%   -0.92%     
==========================================
  Files           4        4              
  Lines        1139     1163      +24     
==========================================
+ Hits          885      893       +8     
- Misses        254      270      +16     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[fverdugo]

Nice to see this merged so quickly!

Also wanted to ask if using .row_partition and .col_partition like in

These fields are documented (thus public), but I always use partition(axes(A,x)) outside PAs.

NB. AFAIK, Petsc (and provably Hypre) assumes that the global row ids are partitioned using a "block" partition (consecutive rows go to the same rank) and blocks are mapped to ranks in ascending rank id. Moreover ghost columns are sorted according to global id. None of these restrictions apply to PAs, which makes challenging to convert a PSparseMatrix to a Petsc matrix since you need to reformat data to match Petsc assumptions.

Take a look to this function to check how I am building a Petsc matrix from a PSparseMatrix:

https://github.com/fverdugo/PetscCall.jl/blob/3602086520882797b0210df22fabdb5af9a46a5a/src/ksp.jl#L246

You provably need to do something similar (unless you are already doing it) so that HYPRE.jl is able to work with any PSparseMatrix.

[fredrikekre]

Yea, right now I am just throwing an error instead (

HYPRE.jl/ext/HYPREPartitionedArrays.jl

Line 36 in 10bdaec

@assert iupper - ilower + 1 == length(own_to_global_row)

).