Tensor Contraction and Operation Minimization for Extreme Scale Computational Chemistry

Computational chemistry codes such as NWChem provide vital simulation capabilities to a wide rangeof scientists. These codes require significant compute resources, and are among the key applications being targeted for upcoming Exascale system at the DOE. Optimizing Tensor operations is a key aspect to reducing the computational requirement of these codes. The Tensor Contraction tools will include Tensor Operation Minimization (OpMin) techniques that include specializations and usability support for computational chemistry codes (e.g., auxiliary indices, projected atomic orbitals, density fitting bases, etc.), and optimized code generation for efficient heterogeneous computing with small-sized tensors. The proposed effort will develop functionality and user support features that will improve computational chemistry software developer productivity in implementing new quantum chemistry models for efficient execution on emerging Extreme Scale architectures. To demonstrate the feasibility of the proposed approach, the technical objectives of the Phase I project will include Tensor operation minimization support will be developed for auxiliary indices. Performance prediction will be developed for the GPU implementation of TTGT. A batched tensor transpose will be implemented for fixed sized tensors on GPUs. A Batched Tensor Contraction will be implemented for fixed size tensors on GPUs. The Tensor Contraction tools will improve simulation performance for computational chemistry scientists in industry and academia, application areas include drug discovery, specialty chemicals, fuel/petroleum products, organic materials, polymers, and consumer products at companies such as Bayer AG, Dow Chemical, Lubrizol, Unilever, RIKEN, DuPont, BP, and BASF. Example chemistry codes that would benefit from the inclusion of the Tensor Contraction tools include QChem, Gaussian, GAMESS, MOLPRO, PWmat, TeraChem, ADF, and NWChem.