Core Features and Functionality

PETSc is a general parallel linear and non-linear solver framework, which provides these general classes of functionality:

Supported Systems

• HPC

• Linux

• MacOS

• Microsoft Windows

General Features

• Parallel vectors

• Vector code for communicating ghost points

• Parallel matrices

• Several sparse and dense matrix storage formats (see MatType):

  • AIJ/CSR (Yale sparse matrix format)

  • LMVM (Limited Memory Variable Metric)

  • Block

  • Nested

  • Dense

• Easy, efficient matrix assembly and interface

• Parallel unstructured mesh support

• Powerful object introspection tools

• Support for a variety of IO options (see PetscViewerType)

• Comprehensive performance testing

Solver Features

• Parallel Krylov subspace methods

• Parallel nonlinear solvers

• Scalable parallel linear and nonlinear preconditioners

• Parallel timestepping (ODE) solvers

• Local and global error estimators

• Forward and adjoint sensitivity capabilities

• Robust optimization through Tao

See also

For full feature list and prerequisites see:

• Linear solver table

• Nonlinear solver table

• Tao solver table

Accelerator/GPU Features

• Matrix/Vector CUDA support

• Kokkos support

• Matrix/Vector OpenCL/ViennaCL support

• Matrix/Vector HIP support

Note

PETSc GPU support is under heavy development! See GPU support roadmap for more information on current support.

Support Features

• Complete documentation

• Comprehensive profiling of floating point and memory usage

• Consistent user interface

• Intensive error checking

• Over one thousand examples

• PETSc is supported and will be actively enhanced for many years

previous

PETSc in a nutshell

next

GPU Support Roadmap