Glow: An LLVM Optimizing Compiler For Neural Networks

The latest interesting use of the LLVM compiler infrastructure stack is for Glow, a machine learning / neural network optimizing compiler.

Glow is intended to be used by high-level machine learning frameworks and it in turn -- via leveraging LLVM -- will generate optimized code for different hardware targets.

Those interested in machine learning / neural networks, the code and a lot more information on Glow is available via the PyTorch project on GitHub. There is also a whitepaper (PDF) on this optimizing compiler. The developers behind Glow announced the project this weekend on the LLVM mailing list.