basic_iterative_solvers

This is a set of (preconditioned) iterative solvers for Ax = b, where A is typically assumed to be a large sparse matrix.

Usage Examples

./basic_iterative_solvers HPCG-128.mtx -cg

./basic_iterative_solvers Anderson,Lx=100,Ly=100,Lz=50,ranpot=5.0 -gm -p gs

Building basic_iterative_solvers

git clone git@github.com:DanecLacey/basic_iterative_solvers.git
cd basic_iterative_solvers
mkdir build && cd build
cmake .. -DCMAKE_CXX_COMPILER=<CXX> -DCMAKE_PREFIX_PATH=install_path (if using TPLs)
make

Features

• Stacked timers around key code regions
• Optional third party libraries:
  • Instrumentation with likwid markers for collecting hardware performance counters already built-in around native kernels
  • Matrices can be read from .mtx files, or generated with the SCAMAC library
  • Integration with SmaxKernels library for optimized sparse kernel implementations

Supported Solvers

• Jacobi -j
• (Forward/Symmetric) Gauss-Seidel -gs / -sgs
• (Preconditioned) Conjugate-Gradient -cg
• (Preconditioned) (Restarted) GMRES -gm
• (Preconditioned) BiCGSTAB -bi

Supported Preconditioners

• Jacobi -p j
• (Forward/Backward/Symmetric) Gauss-Seidel -p (gs/bgs/sgs)
• (Thresholded) Incomplete LU -p ilu(0/t)

Notes

• We strive to follow the C++ Core Guidelines
• The sparse matrix storage format of A is CRS.
• Only left-preconditioning is implemented.
• This code is mainly for performance investigations.