An efficient MPI/openMP parallelization of the Hartree-Fock method for the second generation of Intel^®Xeon Phi^™processor

Abstract

Modern OpenMP threading techniques are used to convert the MPI-only Hartree-Fock code in the GAMESS program to a hybrid MPI/OpenMP algorithm. Two separate implementations that di er by the sharing or replication of key data structures among threads are considered, density and Fock matrices. All implementations are benchmarked on a super-computer of 3,000 Intel® Xeon Phi TM processors. With 64 cores per processor, scaling numbers are reported on up to 192,000 cores. e hybrid MPI/OpenMP implementation reduces the memory footprint by approximately 200 times compared to the legacy code. e MPI/OpenMP code was shown to run up to six times faster than the original for a range of molecular system sizes.

Optimizations of Unstructured Aerodynamics Computations for Many-core Architectures

“…In addition, KNL-specific optimizations and tuning are proposed in [63] for seismic computations, with detailed comparisons between KNC and Haswell. In [78], the performance of the hybrid MPI+OpenMP programming paradigm is provided on Theta supercomputer based on KNL compute nodes. The many-core scalability of the edgebased graph coloring algorithm performance is provided in [79], which targets both GPU as well as KNC.…”

Multithreaded parallelization of the energy and analytic gradient in the fragment molecular orbital method

“…It also eliminates intranode MPI communications. This hybrid MPI+OpenMP parallelization has been recently implemented in GAMESS …”

“…This matrix (Equation ) is composed of the internal contribution and ESP. The former is the usual Fock matrix as in non‐FMO calculations, and its parallelization is described elsewhere . Therefore, the description below (Sections 2.2.1 and 2.2.2) focuses on ESP, which is FMO specific, and according to Equation consists from one‐electron and two‐electron contributions.…”

“…The scheme in Algorithm 2 is a modification of the Fock matrix calculation . It contains nested loops over shells of fragment X (indices k and l ) and monomer L (indices i and j ).…”

“…This model executes multiple copies of GAMESS program per node, which is not very efficient on CPUs with a very large number of cores (eg, on Intel Xeon Phi processors 128 MPI processes per node are executed). To improve the memory usage and increase the computational efficiency, an OpenMP version of GAMESS has been developed independent of the two older OpenMP codes of GAMESS on the K computer and Cray XT5…”