Add computeDivProcBoundImpl and computeLaplacianProcBoundImpl for pro… by greole · Pull Request #512 · exasim-project/NeoN

greole · 2026-05-20T09:40:08Z

Implements matrix-coefficient assembly at processor-boundary faces for both the Gauss-Green divergence and Laplacian operators and tests it against non distributed counterparts. This is also taken from #414 and currently misses the distributed solver test.

github-actions · 2026-05-20T09:42:04Z

Thank you for your PR, here are some useful tips:

If this is your first PR, please add yourself to the AUTHORS.md file
Use labels to change the workflow behavior. Set auto-fix to automatically apply fixes, full-ci for a comprehensive set of tests and sanitizer runs, ready-for-review to show that you're done and need a review.
I deployed a test documentation to https://exasim-project.com/NeoN/Build_PR_512
For full-ci builds benchmark results are available at https://github.com/exasim-project/NeoFOAM-BenchmarkData/512
Also check if the GPU pipelines on https://gitlab-ce.lrz.de/greole/neon/-/jobs are successful.

…cessor boundaries Implements matrix-coefficient assembly at processor-boundary faces for both the Gauss-Green divergence and Laplacian operators. Includes the required LinearSystem::nonLocalMatrix() member and FaceToMatrixAddress extensions that track local/nonLocal/boundary sparsity patterns and row counts.

Copilot

Pull request overview

This PR extends NeoN’s finite-volume implicit operator assembly to properly handle processor-boundary (inter-rank) faces for Gauss-Green divergence and Laplacian, storing cross-rank coefficients separately and adding MPI unit tests to validate equivalence with the non-distributed assembly.

Changes:

Add implicit proc-boundary coefficient assembly for div and laplacian, storing non-local contributions in a new LinearSystem::offDiagonalMatrix().
Extend geometry handling to compute/exchange proc-boundary distances used in non-orthogonal delta coefficients.
Add a new MPI unit test (test/distributed/operator.cpp) and wire it into the distributed test CMake.

Reviewed changes

Copilot reviewed 21 out of 21 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
test/linearAlgebra/utilities.cpp	Updates test setup to provide a dedicated boundary RHS vector.
test/linearAlgebra/linearSystem.cpp	Refactors test variable naming/usage around `LinearSystem` construction and host copying.
test/linearAlgebra/faceToMatrixAddress.cpp	Minor rename/cleanup in sparsity pattern test.
test/distributed/operator.cpp	New MPI test validating distributed operator assembly (main + off-diagonal parts) against global assembly.
test/distributed/CMakeLists.txt	Registers the new distributed operator test when MPI is enabled.
src/mesh/unstructured/unstructuredMesh.cpp	Minor cleanup in uniform mesh creation implementation.
src/linearAlgebra/ginkgo.cpp	Silences unused-parameter warnings for an unsupported COO(Vec3) overload.
src/linearAlgebra/faceToMatrixAddress.cpp	Moves `FaceToMatrixAddress` methods out-of-header; adds proc-boundary sparsity helper (currently TODO).
src/finiteVolume/cellCentred/stencil/cellToFaceStencil.cpp	Minor cleanup removing unused view extraction.
src/finiteVolume/cellCentred/stencil/basicGeometryScheme.cpp	Adds MPI exchange to compute proc-boundary neighbour distances for non-orth delta coeffs.
src/finiteVolume/cellCentred/operators/gaussGreenLaplacian.cpp	Adds `computeLaplacianProcBoundImpl` and integrates it into implicit Laplacian assembly.
src/finiteVolume/cellCentred/operators/gaussGreenDiv.cpp	Adds `computeDivProcBoundImpl` and integrates it into implicit divergence assembly.
src/finiteVolume/cellCentred/interpolation/upwind.cpp	Extends upwind weights to proc-boundary faces.
include/NeoN/mesh/unstructured/unstructuredMesh.hpp	Reorders/clarifies `nProcBoundaryFaces()` API with documentation.
include/NeoN/linearAlgebra/matrix.hpp	Adds a placeholder `Matrix::reset()` API (currently no-op).
include/NeoN/linearAlgebra/linearSystem.hpp	Adds `offDiagonalMatrix_` storage and constructors; resets it in `LinearSystem::reset()`.
include/NeoN/linearAlgebra/faceToMatrixAddress.hpp	Moves inline implementations to `.cpp`.
include/NeoN/finiteVolume/cellCentred/operators/gaussGreenLaplacian.hpp	Header cleanup (removes declarations/extern templates).
include/NeoN/finiteVolume/cellCentred/operators/gaussGreenDiv.hpp	Header cleanup (removes extern templates).
include/NeoN/finiteVolume/cellCentred/faceNormalGradient/faceNormalGradient.hpp	Silences unused-parameter warnings in default `implicitCorrection`.
include/NeoN/distributed/partitioning.hpp	Populates proc-boundary boundary values when partitioning 1D fields.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

    [[nodiscard]] LinearSystem<ValueType, SystemMatrixType, BoundaryMatrixType>
    copyToExecutor(Executor exec) const override
    {
        return {
            matrix_.copyToExecutor(exec),
            rhs_.copyToExecutor(exec),
            boundaryMatrix_.copyToExecutor(exec),
-            boundaryRhs_.copyToExecutor(exec)
+            boundaryRhs_.copyToExecutor(exec),
+            offDiagonalMatrix_.copyToExecutor(exec)
        };


+    float epsilon = 1e-32;
+


+    auto solverStats = solver.solve(ls, x);
+    auto solverStatsDist = solver.solve(lsDst, xPart);
+
+    auto [numIterDist, initResNormDist, finalResNormDist, solveTimeDist] =
+        solverStatsDist.entries[0];
+    auto [numIter, initResNorm, finalResNorm, solveTime] = solverStats.entries[0];


lupeterm

LGTM

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

greole force-pushed the enh/procOperatorImpl branch 3 times, most recently from 501303f to e5df305 Compare May 22, 2026 07:39