Sparse Linear Algebra#

Overview#

The sparse linear algebra modulenvmath.sparse in nvmath-python leverages variousNVIDIA math libraries to support sparse[1] linear algebra computations. As of the currentBeta release, we offer the specialized sparse direct solver API based on thecuDSS library.

API Reference#

Generic Linear Algebra APIs (nvmath.sparse)#

Generic APIs will be available in a later release.

Specialized Linear Algebra APIs (nvmath.sparse.advanced)#

direct_solver(a, b, /, *[, options, ...])

Solve\(a @ x = b\) for\(x\).

DirectSolver(a, b, *[, options, execution, ...])

Create a stateful object that encapsulates the specified sparse direct solver computations and required resources.

DirectSolverFactorizationConfig(solver)

An interface to configurenvmath.sparse.advanced.DirectSolver.factorize().

DirectSolverFactorizationInfo(solver)

An interface to query information returned bynvmath.sparse.advanced.DirectSolver.factorize().

DirectSolverPlanConfig(solver)

An interface to configurenvmath.sparse.advanced.DirectSolver.plan().

DirectSolverPlanInfo(solver)

An interface to query information returned bynvmath.sparse.advanced.DirectSolver.plan().

DirectSolverSolutionConfig(solver)

An interface to configurenvmath.sparse.advanced.DirectSolver.solve().

memory_estimates_dtype

NumPy dtype object that encapsulates the memory estimates in sparse.advanced.

DirectSolverAlgType

alias ofAlgType

DirectSolverMatrixType

alias ofMatrixType

DirectSolverMatrixViewType

alias ofMatrixViewType

DirectSolverOptions([sparse_system_type, ...])

A data class for providing options to theDirectSolver object and the wrapper functiondirect_solver().

ExecutionCUDA(device_id, ...)

A data class for providing GPU execution options to theDirectSolver object and the wrapper functiondirect_solver().

ExecutionHybrid([device_id, num_threads])

A data class for providing hybrid (CPU-GPU) execution options to theDirectSolver object and the wrapper functiondirect_solver().

HybridMemoryModeOptions([...])

A data class for providing options related to the use of hybrid (CPU-GPU) memory to those execution spaces that support it.

Footnotes

[1]

SeeLinear Algebra for dense operations.