Software Technology

Programming Models and Runtimes

xGA: Global Arrays on Extreme Scale Architectures

Principal Investigators: Jeff Daily, lead, Pacific Northwest National Laboratory (PNNL); Bruce Palmer, PNNL

This project is endeavoring to improve the performance, scalability, and user productivity of the Global Arrays (GA) library for exascale systems. This is essential to the ECP applications that already depend on GA such as NWChemEx, GAMESS and GridPACK (used as part of the Stochastic Grid Dynamics project).  In addition, GA is being considered for use in the ECP application QMCPACK.  GA supports a shared memory-like programming model on distributed memory platforms. This allows users to create distributed multidimensional arrays that can be accessed from any processor using simple one-sided put/get/accumulate communication primitives. The GA library is being extended to take advantage of exascale architecture features including deep memory hierarchies and accelerators, while continuing to tune and improve the performance of the GA runtime.