Question on getting NNZ of the preconditioner with Euclid ILU

[RickGY]

Hi,

I'm trying to run some experiments with Euclid's ILU preconditioner. It would be very helpful if there's some way to access and report the nnz of the L and U matrix built as the preconditioner, for profiling purpose. In particular, I'm wondering if it's possible to integrate that into the workflow similar to the one in src/test/ij.c:

if (eu_level > -1) { HYPRE_EuclidSetLevel(pcg_precond, eu_level); }
if (eu_ilut) { HYPRE_EuclidSetILUT(pcg_precond, eu_ilut); }
if (eu_sparse_A) { HYPRE_EuclidSetSparseA(pcg_precond, eu_sparse_A); }
if (eu_row_scale) { HYPRE_EuclidSetRowScale(pcg_precond, eu_row_scale); }
if (eu_bj) { HYPRE_EuclidSetBJ(pcg_precond, eu_bj); }
HYPRE_EuclidSetStats(pcg_precond, eu_stats);
HYPRE_EuclidSetMem(pcg_precond, eu_mem);

/*HYPRE_EuclidSetParams(pcg_precond, argc, argv);*/

HYPRE_PCGSetPrecond(pcg_solver,
                   (HYPRE_PtrToSolverFcn) HYPRE_EuclidSolve,
                   (HYPRE_PtrToSolverFcn) HYPRE_EuclidSetup,
                   pcg_precond);
HYPRE_PCGSetup(pcg_solver, (HYPRE_Matrix) parcsr_M,
                     (HYPRE_Vector) b, (HYPRE_Vector) x);
HYPRE_PCGSolve(pcg_solver, (HYPRE_Matrix)parcsr_A,
                     (HYPRE_Vector)b, (HYPRE_Vector)x);

[victorapm]

Hi @RickGY,

Support for new features in Euclid has been discontinued. However, if you’d like to contribute a patch, we welcome PRs for improvements.

That said, we generally recommend using the ILU preconditioner instead, which is actively maintained and may better suit your needs. You can find more details on ILU here. Particularly, print_level=1 in HYPRE_ILUSetPrintLevel gives you nnz(L) + nnz(U) / nnz(A)

Let us know if you need further guidance!