Assembly directly to HYPREMatrix. #5
Conversation
Codecov ReportBase: 74.39% // Head: 75.97% // Increases project coverage by
Additional details and impacted files@@ Coverage Diff @@
## master #5 +/- ##
==========================================
+ Coverage 74.39% 75.97% +1.58%
==========================================
Files 4 4
Lines 984 1053 +69
==========================================
+ Hits 732 800 +68
- Misses 252 253 +1
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here. ☔ View full report at Codecov. |
|
With this you can either assemble every element contribution (might be too slow due to MPI communication, but require less memory): A = HYPREMatrix(comm, proc_row_first, proc_row_last)
hypre_assembler = HYPRE.start_assemble!(A)
for cell in cell_for_proc
ke = ...
global_dofs_for_cell = ...
HYPRE.assemble!(hypre_assembler, global_dofs_for_cell, ke)
end
HYPRE.finish_assemble!(hypre_assembler)or assemble the processor-matrix at once: A = HYPREMatrix(comm, proc_row_first, proc_row_last)
hypre_assembler = HYPRE.start_assemble!(A)
Alocal = process_local_sparsity_pattern(...)
ferrite_assembler = start_assemble(Alocal)
for cell in cell_for_proc
ke = ...
local_dofs_for_cell = ...
Ferrite.assemble!(ferrite_assembler, local_dofs_for_cell, ke)
end
global_dofs_for_proc = ...
HYPRE.assemble!(hypre_assembler, global_dofs_for_proc, Alocal)
HYPRE.finish_assemble!(hypre_assembler) |
|
Nice. But would the first example really be that slow? HYPRE_IJMatrixAddToValues does not do any MPI communication right, it is only at assemble_matrix(A.A) where mpi communication is done? I think I like the first approach better since it more resembles the current assembler in Ferrite and does not require local dof ids |
|
Yea you are right, it is actually documented to be
|
This patch adds an assembler interface such that it is possible to
directly assemble HYPRE(Vector|Matrix) without going through a sparse
matrix datastructure in Julia. This is done as follows:
1. Create a new (empty) matrix/vector using the constructor.
2. Create an assembler and initialize the assembly using
HYPRE.start_assemble!.
3. Assemble all contributions using HYPRE.assemble!.
4. Finalize the assembly using HYPRE.finish_assemble!.
HYPRE.start_assemble!
The assembler caches some buffers that are (re)used by every call to
HYPRE.assemble! so this should be efficient. All MPI communication
should happen in the finalization step.
fredrikekre
force-pushed
the
fe/assembly
branch
from
f0366fb to
9a91f0b
Compare
fredrikekre
changed the title
No description provided.