Overview

HeteroSolver is a library, used for solving large scale sparse linear equations $Ax=b$ that are derived from numerical simulation programs in a wide range. Here, $A$ is a sparse square matrix of the dimension $N\times N$ , $x$ and $b$ are $N$ element vectors. The library is implemented by using sparse direct methods, which are robust and high performance.

HeteroSolver has the following distinctive characteristics:

Solves effectively linear equations by cooperating with Vector Engine(VE) and Vector Host(VH).
Uses BLAS routines of NEC Numeric Library Collection that are highly optimized for VE.
Supports OpenMP version and Hybrid version (MPI-OpenMP parallel implementation), which can be used from either Fortran or C programs.

Overview of HeteroSolver

HeteroSolver allows users to solve sparse simultaneous linear equations without implementing complex algorithms regarding sparse direct methods, greatly improving the productivity of numerical simulation program development.

This guide explains how to call this library from Fortran programs and execute them. For details, see the following topics:

Sparse Matrix Storage Formats describes data storage formats of a sparse matrix,
Compiling and Linking describes how to compile your program and how to link it with the library,
Remarks on Execution describes what you need before running your program linked with the library, and
Algorithms Used describes algorithms are used in this library.

For details of the user interface, see below:

When solving the equations by using multiple MPI processes, you must use Hybrid version. It enables a sparse matrix $A$ , right-hand side vectors $b$ , and solution vectors $x$ to distribute among MPI processes. For details, refer to the description of the parameter idist of the handle initialization subroutine phs_init_handle.

Version

The API version this manual targets: 1.1
This manual version: 3.0.0-230222

License Notice

HeteroSolver includes the work that is distributed in the Apache License, Version 2.0.