Getting Started

  • package manager

  • mention API docs for full details

Backends

Ginkgo supports running on the following architectures:

  • multi-core x86 CPUs

  • NVIDIA GPUs

  • AMD GPUs

  • Intel GPUs

We denote the architecture specific implementations in Ginkgo as backends. The following backends are provided in Ginkgo:

Reference

A CPU backend that runs on a single core. This is not optimized and should only be used for verification.

OMP

A CPU backend that runs on multiple cores using OpenMP.

CUDA

A GPU backend that runs on a single NVIDIA GPU.

HIP

A GPU backend that runs on a single AMD GPU.

SYCL

A GPU backend that runs on a single Intel GPU.

Multiple backends can be used at the same time. Which backends can be used only depends on the system, see System Requirements for more details.

The particular backend Ginkgo code runs on is chosen at runtime through an implementation of the Executor interface.

Warning

add links, whatever to the executor reference

  • support different backends

  • general overview

  • high level example