v0.5.0

Lots of new goodies! Now completely memory leak-free, and valgrind-clean!

New Features 🔥

• New comm_t object to encapsulate all MPI
• Error handling via std::exceptions
• New memory<T>, deviceMemory<T>, and pinnedMemory<T> types
  • Memory management objects, with shallow copy semantics and automatic garbage collection
  • memory<T> is essentially a thin wrapper of std::shared_ptr<T[]>
  • deviceMemory<T> and pinnedMemory<T> are thin wrappers of occa::memory
• New device-aware gather scatter functionality in ogs library, replacing gslib
• New ParAdogs library for distributed parallel mesh partitioning
• Device/Host Timers

v0.4.0

New features:

• Smoothed Aggregation AMG
• Improved initial guesses for successive solutions of linear systems, including
  • History projection method with restarts
  • History projection method with rolling QR update
  • History extrapolation
  • Sparse history extrapolation
• Preconditioned GMRES linearSolver
• Preconditioned MINRES linearSolver

Introducing platform for GPU memory management & removed BLAS+LAPACK

Minor version bump.

New features:

1. platform to mediate DEVICE memory allocation.
2. removed inclusion of LAPACK and BLAS from netlib. User must now supply precompiled binaries.

v0.3.1

Minor version bump. Added scripted testing framework written with Python

Testing and CI

This release includes various bug fixes, improved testing, and continuous integration.

Switching staging to master

Noel Chalmers has completed his staging branch and switched it to be the master branch.

Main innovations include:

improved make system.
improved C++ hierarchy.
reorg into core and libraries.
overall streamlining of code.
See readme for new details.

Refactored parAlmond & gather-scatter

Major changes:

1. The parAlmond multigrid code has been refactored and relocated to the libs/parAlmond sub-directory.

2. The OCCA gather-scatter code has been consolidated and relocated to the libs/gatherScatter sub-directory.

Removed experimental solvers

Deleted some deprecated solver code and old scripts.

Initial copy from holmes

This is the first version of libParanumal developed by the Parallel Numerical Algorithms Group at Virginia Tech.