2021
DOI: 10.1016/j.parco.2021.102829
|View full text |Cite
|
Sign up to set email alerts
|

Achieving performance portability in Gaussian basis set density functional theory on accelerator based architectures in NWChemEx

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1

Citation Types

0
3
0

Year Published

2022
2022
2024
2024

Publication Types

Select...
5
1
1

Relationship

1
6

Authors

Journals

citations
Cited by 11 publications
(3 citation statements)
references
References 53 publications
0
3
0
Order By: Relevance
“…In addition, domain-driven tensor frameworks typically outperform generic tensor frameworks for computational chemistry workloads. As such, many chemistry-community driven libraries have been developed for low-level operations such as analytical (e.g., libint, libcint, and simint) and numerical (e.g., GauXC, libGridXC) operator integrals, functional evaluation (libxc, XCfun, ExchCXX), and tensor algebra (e.g., TiledArray, , TAMM, and the Cyclops Tensor Framework (CTF)) to name a few. The development and optimization of these libraries is challenging; thus, it is of critical interest to the development of sustainable computational chemistry software that these kernels be implemented in a performance portable manner.…”
Section: Programming Models and Software Integrationmentioning
confidence: 99%
“…In addition, domain-driven tensor frameworks typically outperform generic tensor frameworks for computational chemistry workloads. As such, many chemistry-community driven libraries have been developed for low-level operations such as analytical (e.g., libint, libcint, and simint) and numerical (e.g., GauXC, libGridXC) operator integrals, functional evaluation (libxc, XCfun, ExchCXX), and tensor algebra (e.g., TiledArray, , TAMM, and the Cyclops Tensor Framework (CTF)) to name a few. The development and optimization of these libraries is challenging; thus, it is of critical interest to the development of sustainable computational chemistry software that these kernels be implemented in a performance portable manner.…”
Section: Programming Models and Software Integrationmentioning
confidence: 99%
“…As of this review, we have CUDA, HIP and SYCL ports of the density functional theory (DFT)-XC integrator. There has recently been a large structural refactor to allow for porting to new architectures (per the algorithmic design outlined in [12]) which allows the decoupling of high-level algorithmic workflows from the implementation details of performance critical kernels to be optimized for each architecture of interest. However, due to lack of resources, only the CUDA and HIP ports of the integrator have been ported in this new software infrastructure as of this report, the SYCL functionality will be ported over in the coming fiscal year.…”
Section: Dftmentioning
confidence: 99%
“…However, when reasonable large systems (≥1000 atoms) are studied, accurate wavefunction-based methods cannot be employed due to their high computational cost (above all, for excited state properties), although some progress has been recently achieved using graphical processing units [49] and localization procedures [50,51]. Thus, density functional theory (DFT) and time dependent (TD-) DFT, the latter required for excited state quantities, are usually chosen as a good compromise between accuracy and computational costs [10,[52][53][54][55][56][57][58][59][60][61][62].…”
Section: Introductionmentioning
confidence: 99%