Embracing a new era of highly efficient and productive quantum Monte Carlo simulations

Abstract

QMCPACK has enabled cutting-edge materials research on supercomputers for over a decade. It scales nearly ideally but has low single-node e ciency due to the physics-based abstractions using array-of-structures objects, causing ine cient vectorization. We present a systematic approach to transform QMCPACK to better exploit the new hardware features of modern CPUs in portable and maintainable ways. We develop miniapps for fast prototyping and optimizations. We implement new containers in structure-of-arrays data layout to facilitate vectorizations by the compilers. Further speedup and smaller memory-footprints are obtained by computing data on the y with the vectorized routines and expanding single-precision use. All these are seamlessly incorporated in production QMCPACK. We demonstrate upto 4.5x speedups on recent Intel ® processors and IBM Blue Gene/Q for representative workloads. Energy consumption is reduced signi cantly commensurate to the speedup factor. Memory-footprints are reduced by up-to 3.8x, opening the possibility to solve much larger problems of future.

QMCPACK: an open sourceab initioquantum Monte Carlo package for the electronic structure of atoms, molecules and solids

“…This also reduces the memory requirements of the application. We have recently completed extensive analysis and reimplementation of the core compute kernels of the application, more than doubling the speed of many calculations on modern multicore processors [37]. The performance obtained for several key kernels is shown in figure 2.…”

Raising the Performance of the Tinker-HP Molecular Modeling Package [Article v1.0]

“…As the two functions can give slightly different numerical results, care must be taken to be sure to always have the same numerical results than the Rel. 4. Limit the use of Fortran intrinsics The use of Fortran intrinsics (SUM, PACK, etc...) is a common approach for vectorization in Molecular Dynamics package.…”

“…In a recent study, Watanabe and Nakagawa [25] have obtained a boost factor between 1. 4 the data setup, for the vectorization of the Lennard-Jones potential on AVX2 and AVX-512 architectures. So, achieving comparable, and often superior, boost factors on all the vectorized routines of Tinker-HP seems quite satisfactory.…”

“…So, achieving comparable, and often superior, boost factors on all the vectorized routines of Tinker-HP seems quite satisfactory. One interesting way to see what could be achieved is also to look at what has been done in another community code developed at Argonne and devoted to Quantum Monte-Carlo [4]. By definition, Monte-Carlo is the opposite of MD, as it is a highly parallel approach with few communications.…”

“…Second, it is organized to present in a pedagogical fashion our code optimization efforts, outlining an optimal strategy for vectorization of Tinker-HP. Although there have been theoretical discussions of AVX-512 vectorization [3], and a description of the effective vectorization of a Quantum Monte-Carlo production code [4], such practical examples are rarely documented and we think our experience could prove useful to other software developers.…”