rocALUTION Documentation¶
rocALUTION is a sparse linear algebra library with focus on exploring fine-grained parallelism on top of AMD’s Radeon Open Compute ROCm runtime and toolchains, targeting modern CPU and GPU platforms. Based on C++ and HIP, it provides a portable, generic and flexible design that allows seamless integration with other scientific software packages.
In the following, three separate chapters are available:
User Manual: This is the manual of rocALUTION. It can be seen as a starting guide for new users but also a reference book for more experienced users.
Design Documentation: The Design Document is targeted to advanced users / developers that want to understand, modify or extend the functionality of the rocALUTION library. To embed rocALUTION into your project, it is not required to read the Design Document.
API: This is a list of API functions provided by rocALUTION.
Contents:
- User Manual
- Introduction
- Building and Installing
- Basics
- Single-node Computation
- Multi-node Computation
- Solvers
- Code Structure
- Iterative Linear Solvers
- Building and Solving Phase
- Clear Function and Destructor
- Numerical Update
- Fixed-Point Iteration
- Krylov Subspace Solvers
- Chebyshev Iteration Scheme
- Mixed-Precision Defect Correction Scheme
- MultiGrid Solvers
- Unsmoothed Aggregation AMG
- Smoothed Aggregation AMG
- Ruge-Stueben AMG
- Pairwise AMG
- Direct Linear Solvers
- Preconditioners
- Code Structure
- Jacobi Method
- (Symmetric) Gauss-Seidel / (S)SOR Method
- Incomplete Factorizations
- AI Chebyshev
- FSAI
- SPAI
- TNS
- MultiColored Preconditioners
- Multi-Elimination Incomplete LU
- Diagonal Preconditioner for Saddle-Point Problems
- (Restricted) Additive Schwarz Preconditioner
- Block-Jacobi (MPI) Preconditioner
- Block Preconditioner
- Variable Preconditioner
- Backends
- Remarks
- Supported Targets
- Design Documentation
- API