kokkos/kokkos

Kokkos C++ Performance Portability Programming Ecosystem: The Programming Model - Parallel Execution and Memory Abstraction

What it solves

Kokkos solves the problem of performance portability in High Performance Computing (HPC). It allows developers to write C++ applications that can run efficiently across different hardware platforms without needing to rewrite the code for each specific architecture.

How it works

It provides a C++ programming model that abstracts both parallel execution and data management. This allows the software to target complex node architectures with multiple types of execution resources and N-level memory hierarchies. It supports various backend programming models, including CUDA, HIP, SYCL, HPX, OpenMP, and C++ threads.

Who it’s for

It is designed for developers of performance-critical applications targeting major HPC platforms.

Highlights

  • Support for multiple backends including CUDA, HIP, SYCL, and OpenMP.
  • Abstractions for parallel execution and data management.
  • Designed for complex node architectures with multi-level memory hierarchies.
  • Part of the Linux Foundation and the Kokkos C++ Performance Portability Programming Ecosystem.

Related

  • Project
    kokkos/kokkos-kernels

    A collection of performance-portable C++ math kernels for linear algebra and graph operations built on the Kokkos parallel programming model.

  • Project
    AdaptiveCpp/AdaptiveCpp

    A community-driven platform for C++ heterogeneous programming that enables a single binary to run across CPUs and GPUs from all major vendors.

  • Project
    ROCm/composable_kernel

    A library providing a programming model for writing performance-critical machine learning kernels that are portable across different GPU and CPU architectures.

  • Project
    alpaka-group/alpaka

    A header-only C++20 abstraction library that enables performance portability for parallel kernels across CPUs and GPUs from NVIDIA, AMD, and Intel.

  • Project
    pytorch/cpuinfo

    A cross-platform C/C++ library for detecting detailed CPU hardware information, such as instruction sets and cache topology, to enable performance optimizations.