Abstract

Quantum ESPRESSO (QE) is an integrated suite of Open-Source computer codes for electronic-structure calculations and materials modeling at the nanoscale, based on density-functional theory. Filippo Spiga is a long-time contributor to the project and led the effort to port a version of QE to GPU architectures. The first GPU porting was done in CUDA C back in 2012, and a new version based on CUDA Fortran and CUF kernel was developed and released at SC17 . The talk will also review how the porting is done, various tricks to integrate in the same code base both the CPU and the GPU code path, integration with libraries and the use of peer-to-peer GPU -GPU communication. Filippo will also present how QE is developed, built and tested and focus on how QE has been architected for high performance on a variety of HPC architectures. Results running on CSD3 will be presented, best practices are widely applicable to many GPU accelerated codes in many different domains.