Lazy Skeletons is a C++ library for building high-performance, data-parallel applications.
Based on the concept of algorithmic skeletons, it provides a simple and expressive API for composing complex computations from a small set of high-level parallel patterns.
Download or clone the repository from GitHub:
git clone https://github.com/rafa-picao/LazySkeletons.gitInclude the lskel.cuh header file in your C++ project and use the lskel namespace to access the library's functionalities.
See real-world code examples (like SAXPY) and detailed performance benchmarks demonstrating the benefits of kernel fusion against NVIDIA Thrust.
Our API delivers optimized performance by automatically fusing multiple operations into a single kernel, dramatically reducing GPU overhead and memory transfers.
Chain multiple Map and Reduce operations together, and the framework compiles them into a single, highly efficient kernel.
Use expressive parallel patterns (skeletons) to build complex algorithms without writing explicit CUDA or memory boilerplate.
Lazy evaluation ensures computation only happens when the result is needed, minimizing unnecessary execution and host-device synchronization.
Have questions about the framework, potential contributions, or collaboration opportunities? Reach out via email or connect on social media.