Section: New Software and Platforms

Tensor Comprehensions

Keywords: Machine learning - Matrix calculation - Polyhedral compilation - GPU - CUDA

Functional Description: Tensor Comprehensions (TC) is a notation based on generalized Einstein notation for computing on multi-dimensional arrays. TC greatly simplifies ML framework implementations by providing a concise and powerful syntax which can be efficiently translated to high-performance computation kernels, automatically.

Release Functional Description: Integration of the loop tactics matching framework for identifying linear algebra operations and optimizing them.

• Partner: Eindhoven University of Technology

• Contact: Albert Cohen