Current State of the Cray MPT Software Stacks on the Cray XC Series Supercomputers

HPC applications heavily rely on Message Passing Interface (MPI) and SHMEM programming models to develop distributed memory parallel applications. This paper describes a set of new features and optimizations that have been introduced in Cray MPT software libraries to optimize the performance of scientific parallel applications on modern Cray XC series supercomputers. For Cray XC systems based on the Intel KNL processor, Cray MPT libraries have been optimized to improve communication performance, memory utilization, while also facilitating better utilization of the MCDRAM technology. Cray MPI continues to improve the performance of hybrid MPI/OpenMP applications that perform communication operations in threaded regions. In addition, Cray MPI is being enhanced to support the MPI Dynamic Process Management Interface and Cray PMI now offers optimized process placement strategies to improve the communication performance of applications on Cray XC systems. Cray SHMEM offers API extensions to create and maintain SHMEM Teams. Finally, this paper describes efforts involved in optimizing real-world applications such as WOMBAT and SNAP on the latest Cray XC supercomputers