Benoit Meister of Reservoir Labs presents a paper, Polyhedral Tensor Schedulers, during The 17th International Conference on High Performance Computing (HPCS 2019), July 15 – 19, in Dublin, Ireland.
This new research allows us to automatically find and generate the complex parallel code needed for computing with tensors. The technology utilizes the polyhedral model and applies it in unique and scalable ways. The technology has uses in optimizing AI applications, from deep learning kernels to modern complex processors and accelerators.
This paper is part of The Second Special Session on Compiler Architecture, Design and Optimization (CADO 2019) and is co-authored by Eric Pappenhausen of Akai Kaeru and Benoit Pradelle of Silexica.