Computations with tensors arise in many compute-intensive quantum chemistry codes. Unlike linear algebra, high-performance libraries are not readily available for tensor computations. The most common approach is to use layout transformations to convert higher dimensional tensors into 2D matrices and then use highly tuned linear algebra kernels from libraries. However, this approach is not satisfactory in many contexts and it is therefore of interest to develop optimizing compiler technology to automatically generate high-performance code for multiple target platforms from a convenient high-level specification. This talk will discuss current research developments towards the goal of performance-portable code generation for tensor computations.

Towards performance-portable generation of tensor kernels for computational chemistry

248th National Meeting (2014)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.