DirectSolverPlanConfig#

class nvmath.sparse.advanced.DirectSolverPlanConfig(solver)[source]#

An interface to configure nvmath.sparse.advanced.DirectSolver.plan(). The current configuration can also be queried.

Attributes

host_nthreads#: Query or set the number of host threads. See the cuDSS documentation for more information.

matching_algorithm#: Query or set the matching algorithm used. See nvmath.bindings.cudss.AlgType and the cuDSS documentation for more information.

max_nnz#: Query or set the maximum limit for non-zero factors in the LU decomposition. See the cuDSS documentation for more information.

nd_min_levels#: Query or set the minimum number of levels for the nested dissection reordering. See the cuDSS documentation for more information.

pivot_threshold#: Query or set the pivot threshold. See the cuDSS documentation for more information.

pivot_type#: Query or set the type of pivoting. See nvmath.bindings.cudss.PivotType and the cuDSS documentation for more information.

reordering_algorithm#: Query or set the reordering algorithm used. See nvmath.bindings.cudss.AlgType and the cuDSS documentation for more information.

use_matching#: Query or set the option to enable or disable matching. See the cuDSS documentation for more information.

Note

Matching is recommended for general sparse matrices as it typically reduces pivot perturbations and improves solution accuracy with minimal performance impact. For symmetric positive definite (SPD) matrices, matching is unnecessary. For symmetric indefinite matrices, the benefits of matching is unknown.

use_superpanels#: Query or set the option to enable or disable superpanel optimization. See the cuDSS documentation for more information.

Methods