diku-dk/futhark

Health v2 · maintenance only

Adoption not part of the score

2793 stars · 208 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Futhark is a purely functional, statically typed data-parallel array programming language in the ML family, developed at the University of Copenhagen. Its optimizing ahead-of-time compiler generates efficient GPU code (CUDA, HIP, OpenCL) or multi-threaded CPU code, intended for compute-intensive kernels called from other languages.

Use cases

• accelerate array computations on the GPU without writing CUDA or OpenCL by hand
• compile a small compute-intensive kernel to call from Python, Rust, Haskell, or C
• write high-performance parallel numeric code in a functional language
• prototype GPGPU data-parallel algorithms with automatic differentiation
• generate multi-threaded CPU code for parallel workloads

When to choose

• you need high-performance data-parallel array code and want to avoid hand-written GPU kernels
• you want a small, purely functional language with a heavily optimizing compiler
• you need to embed fast compute kernels into applications written in other languages

When to avoid

• you need a general-purpose language for full applications - Futhark is meant for small compute-intensive parts
• you need graphics programming - it targets compute, not rendering
• you rely on features Futhark omits to guarantee high-performance parallelism

Facets

cli-tool · maturity active

compiler programming-language gpu-computing math concurrency programming-languages compilers gpu-computing performance windows cli cross-platform futhark data-parallel functional-programming array-language opencl cuda hip hpc ml-family automatic-differentiation algorithms high-performance-computing linux macos gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="diku-dk/futhark")