GitHub

HPCSparseArrays.jl 0.1.0

Pure Julia distributed linear algebra with MPI.

Overview

HPCSparseArrays.jl provides distributed matrix and vector types for parallel computing with MPI. It offers a pure Julia implementation of distributed linear algebra, with MUMPS for sparse direct solves.

Key Features

  • Distributed Types: HPCVector, HPCMatrix, and HPCSparseMatrix for row-partitioned distributed storage
  • MUMPS Solver: Direct solves using MUMPS for sparse linear systems
  • Row-wise Operations: map_rows for efficient distributed row operations
  • Seamless Integration: Works with standard Julia linear algebra operations
  • Plan Caching: Efficient repeated operations through memoized communication plans

Quick Example

using MPI
MPI.Init()
using HPCSparseArrays
using SparseArrays

# Use the default MPI backend
backend = BACKEND_CPU_MPI

# Create distributed sparse matrix
A = HPCSparseMatrix(sprandn(100, 100, 0.1) + 10I, backend)
b = HPCVector(randn(100), backend)

# Solve linear system
x = A \ b

# Row-wise operations
norms = map_rows(row -> norm(row), HPCMatrix(randn(50, 10), backend))

println(io0(), "Solution computed!")

Run with MPI:

mpiexec -n 4 julia --project example.jl

Documentation Contents

Requirements

  • Julia 1.11 or later
  • MPI installation (OpenMPI, MPICH, or Intel MPI)
  • MUMPS for sparse direct solves (bundled via MUMPS.jl)

License

This package is licensed under the MIT License.