Install kokkos with Homebrew

Project links

• Homepage
• Repository

Commands and aliases

• hpcbind
• kokkos_launch_compiler
• nvcc_wrapper

Project history

Kokkos originated in the U.S. Department of Energy national-lab HPC world, with Sandia National Laboratories as the original developer. Its early evolution moved from node abstractions around 2008-2010 into a fuller programming model around 2010-2014, combining execution abstractions with data abstractions.

The 2014 Kokkos paper described the core idea: applications and domain libraries need performance portability across manycore architectures, and memory access patterns are as important as parallel execution. Sandia describes the Kokkos ecosystem as released in 2017 to provide a vendor-independent performance-portable programming system for scientific, engineering, and mathematical C++ applications.

The project later became part of the Linux Foundation's High Performance Software Foundation ecosystem. Kokkos documentation describes Core as the programming model, with related projects for math kernels, profiling/debugging tools, Python bindings, remote spaces, and resilience/checkpointing.

Adoption history

Kokkos adoption tracks the HPC shift from single-architecture tuning toward heterogeneous systems with NVIDIA, AMD, Intel, and CPU backends. Its support for CUDA, HIP, SYCL, HPX, OpenMP, and C++ threads made it a common portability layer for DOE and scientific codes preparing for exascale machines.

Package-manager adoption matters because Kokkos is often both a library dependency and a compiler-wrapper/tooling dependency. Making it available through Homebrew, Linux distribution packages, MacPorts, Nix, and openSUSE helps developers test and build performance-portable code outside facility-specific software stacks.

How it is used

Users build C++ applications against Kokkos Core to express parallel execution policies and memory spaces without hard-coding one backend. The package's CLI-visible tools support compilation flows, especially CUDA-oriented builds through wrappers such as nvcc_wrapper.

In practice, Kokkos often appears inside larger scientific applications and libraries rather than as an end-user command. Its package entry is still valuable because the build system has to find headers, libraries, compiler wrappers, enabled backends, and compatible C++ compiler settings.

Why package nerds care

Kokkos is a stress test for package systems because it is C++, HPC, backend-sensitive, and toolchain-sensitive. A good package has to expose enough knobs for OpenMP, CUDA, HIP, SYCL, and compiler integration without making every user rebuild a bespoke supercomputer stack.

It is also a marker of HPC software moving into mainstream package indexes. Homebrew carrying Kokkos means macOS and workstation developers can inspect, compile, and prototype with an exascale-era programming model before moving to clusters.

Timeline

• 2008-2010: Kokkos history materials describe early node abstractions and sparse-kernel work.
• 2010-2014: Kokkos evolved into a complete programming model with execution and data abstractions.
• 2014: The original Kokkos paper appeared in Journal of Parallel and Distributed Computing.
• 2017: Sandia describes the Kokkos ecosystem as released for vendor-independent performance portability.
• 2020-02-28: GitHub release metadata records Kokkos 3.0.00.
• 2023-03-03: GitHub release metadata records Kokkos 4.0.00.

Related projects

• The Kokkos ecosystem includes kokkos-kernels, kokkos-tools, pykokkos, kokkos-remote-spaces, and kokkos-resilience. It is used alongside HPC projects and libraries such as Trilinos and LAMMPS, and conceptually relates to RAJA, OpenMP, CUDA, HIP, SYCL, and C++ standardization efforts such as mdspan.

Sources

• https://api.github.com/repos/kokkos/kokkos/releases
• https://kokkos.org/about/overview/
• https://kokkos.org/kokkos-core-wiki/
• https://kokkos.org/kokkos-core-wiki/citation.html
• https://www.osti.gov/servlets/purl/1765805
• https://www.sandia.gov/news/publications/hpc-annual-reports/article/kokkos/