Hypre solver eliminate boundaries (New API) - #3460
Open
bendudson wants to merge 12 commits into
Open
Conversation
This reverts commit 81befc2.
The vector should be assembled once after importing values. Assembling twice with elimBE over-corrects the RHS.
row_indexes is free'd using HypreFree, so should be allocated using HypreMalloc.
The row elimination is now an implementation detail that doesn't require changes to the HypreMatrix and HypreVector interface. The elimination is applied to temporary working vectors when assembling the HypreVector, so the input values are not mutated. New tests check the boundary elimination.
Fixes the integrated Laplace-Hypre3D tests. GPT says: The boundary-elimination logic in [src/sys/hypre_interface.cxx (line 48)](/Users/dudson2/code/BOUT-next/src/sys/hypre_interface.cxx:48) was only removing each boundary variable from one coupled row; in the hypre3d stencil, the same Y-boundary value can appear in multiple nearby rows, especially in the SOL case. I changed BoundaryElimination to collect all representable row couplings for each boundary equation, store them with per-boundary offsets, and apply them during matrix reduction, RHS reduction, and matvec reconstruction. The interface metadata was updated in [include/bout/hypre_interface.hxx (line 103)](/Users/dudson2/code/BOUT-next/include/bout/hypre_interface.hxx:103), and I added multi-coupling unit coverage in [tests/unit/include/bout/test_hypre_interface.cxx (line 618)](/Users/dudson2/code/BOUT-next/tests/unit/include/bout/test_hypre_interface.cxx:618).
ZedThree
requested changes
Aug 26, 2026
ZedThree
left a comment
ZedThree
left a comment
Member
There was a problem hiding this comment.
I can't judge the accuracy of the numerics, but I have a few comments on the code.
Comment on lines
+140
to
+151
| HYPRE_Int nb; | ||
| HYPRE_Int* binum_array; | ||
| HYPRE_Int* bjnum_array; | ||
| HYPRE_Int* bdep_array; | ||
| HYPRE_Complex* bii_array; | ||
| HYPRE_Complex* bij_array; | ||
| HYPRE_Int na; | ||
| HYPRE_Int* aoffset_array; | ||
| HYPRE_Int* aknum_array; | ||
| HYPRE_Complex* aki_array; | ||
| std::vector<HYPRE_Int> reduction_order; | ||
| std::vector<HYPRE_Int> expansion_order; |
Member
There was a problem hiding this comment.
What's the reason some of these are raw pointers, and some are std::vector? Is this a device/host distinction?
Contributor
Author
There was a problem hiding this comment.
Mostly yes. std::vector can't be used on host and arrays that are passed into Hypre should use HypreMalloc.
Comment on lines
+58
to
+64
| std::vector<std::vector<BoundaryOccurrence>> occurrences_by_boundary(nb); | ||
| for (HYPRE_Int rownum = 0; rownum < nrows; rownum++) { | ||
| const HYPRE_Int row_start = row_indexes[rownum]; | ||
| for (HYPRE_Int m = 0; m < ncols[rownum]; m++) { | ||
| const auto boundary_position = boundary_number_for_row.find(cols[row_start + m]); | ||
| if (boundary_position != boundary_number_for_row.end()) { | ||
| occurrences_by_boundary[boundary_position->second].push_back({rownum, m}); |
Member
There was a problem hiding this comment.
I'm not sure what this is supposed to be doing, but should occurrences_by_boundary just be std::set?
I think this block could use a comment explaining what it's doing
| HYPRE_BigInt* cols, HYPRE_Complex* values, | ||
| HYPRE_Int nb, HYPRE_Int* bi_array) | ||
| : nb(nb) { | ||
| HYPRE_Int* row_indexes; |
Member
There was a problem hiding this comment.
I think this constructor should be executed on the host, so we can use vector here?
Suggested change
| HYPRE_Int* row_indexes; | |
| std::vector<HYPRE_Int> row_indexes; | |
| row_indexes.reserve(nrows); |
Contributor
Author
There was a problem hiding this comment.
The lifetime of this is somewhat complicated: row_indexes is allocated but then assigned to the "out" parameter row_indexes_ptr. Very C-like behavior. This could perhaps be tidied up in future.
| BoundaryElimination::BoundaryElimination(HYPRE_Int nrows, HYPRE_Int* ncols, | ||
| HYPRE_BigInt* rows, HYPRE_Int** row_indexes_ptr, | ||
| HYPRE_BigInt* cols, HYPRE_Complex* values, | ||
| HYPRE_Int nb, HYPRE_Int* bi_array) |
Member
There was a problem hiding this comment.
I think most of these should at least be const*, but I don't think there's any need for pointers here? Most of them seem to be copied elementwise?
Corner equations are likely not handled correctly, but it probably doesn't matter.
Calculation of boundary rows can just use the size of the container.
More descriptive flag. The API member function `setElimBE` is also renamed to `setUseBoundaryElimination`
Many small things. Replaced some raw pointers with shared_ptr. Added `modernize-use-ranges` to `.clangd`
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
Replaces #3439, using a new API that doesn't require sharing boundary information between matrices, rhs and solution vectors. Avoids hidden mutations that could be hard to reason about.
Includes additional tests and some changes to the boundary row elimination code.
Written almost entirely by Codex/gpt-5.4 with review and iterations.