ranocha / JustRelax.jl

Pseudo-transient accelerated iterative solvers

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

JustRelax.jl

CI

Need to solve a very large multi-physics problem on a GPU cluster? Just Relax!

Pseudo-transient accelerated iterative solvers, ready for extreme-scale, multi-GPU computation.

JustRelax.jl is a collection of pseudo-transient relaxation solvers for multi-physics problems on regular, staggered, parallel grids, using MPI and multiple CPU or GPU backends. It relies on ImplicitGlobalGrid.jl and ParallelStencil.jl. It's part of the PTSolvers project and the GPU4GEO project.

The package serves several purposes:

  • It reduces code duplication between several applications, e.g. PseudoTransientStokes.jl and PseudoTransientAdjoint.jl

  • It provides a collection of solvers to be used in quickly developing new applications

  • It provides some standardization so that application codes can

    • more easily "add more physics"
    • more easily switch between a psuedo-transient solver and another solver (e.g. a direct solve or a multigrid method)
  • It provides a natural place to describe performance benchmarks for the solver routines

  • It defines useful solvers to be encapsulated and used from non-Julia applications

  • It provides a natural location for contributions of new solvers for use by the larger community

We include several miniapps, each designed to solve a well-specified benchmark problem, in order to provide

  • examples of high-performance usage,
  • bases on which to build more full-featured application codes
  • cases for reference and performance tests
  • JustRelax.jl's entries in "bake offs"
  • tests and examples of interfaces with other packages applications might use, in particular
    • CompGrids.jl
    • PETSc.jl

JustRelax.jl is used in the following applications:

  • TODO link to all applications using the package here (crucial for early development)

About

Pseudo-transient accelerated iterative solvers

License:MIT License


Languages

Language:Julia 100.0%