An optimized and scalable eigensolver for sequences of eigenvalue problems

Berljafa, Mario; Wortmann, Daniel; Napoli, Edoardo Di

doi:10.1002/cpe.3394

Cited by 9 publications

(13 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The greater the K max , the larger is the used prefactor -typically a K max = 4.0 implies a prefactor equal to 80. At first glance, the plots in Figure 5 Ivy Bridge (20 hint that higher values of K max favor the use of HSDLA over FLEUR in terms of execution time. Such a hint is confirmed by the data in Table 2, where speedup values are shown to get larger for increasing values of K max .…”

Section: Performance Resultsmentioning

confidence: 99%

See 1 more Smart Citation

High-performance generation of the Hamiltonian and Overlap matrices in FLAPW methods

Napoli

Peise

Hrywniak

et al. 2017

Computer Physics Communications

Self Cite

View full text Add to dashboard Cite

One of the greatest efforts of computational scientists is to translate the mathematical model describing a class of physical phenomena into large and complex codes. Many of these codes face the difficulty of implementing the mathematical operations in the model in terms of low level optimized kernels offering both performance and portability. Legacy codes suffer from the additional curse of rigid design choices based on outdated performance metrics (e.g. minimization of memory footprint). Using a representative code from the Materials Science community, we propose a methodology to restructure the most expensive operations in terms of an optimized combination of dense linear algebra (BLAS3) kernels. The resulting algorithm guarantees an increased performance and an extended life span of this code, enabling larger scale simulations.

show abstract

Section: Performance Resultsmentioning

confidence: 99%

“…( 11) and ( 12) within the FLEUR software, and illustrate how the traditional implementation can be re-engineered and optimized to take advantage of Basic Linear Algebra Subroutines (BLAS). For the reader interested in improving the computational aspects of the eigenproblem solution in FLAPW, we refer to [19,20,21].…”

Section: [ ]mentioning

confidence: 99%

High-performance generation of the Hamiltonian and Overlap matrices in FLAPW methods

Napoli

Peise

Hrywniak

et al. 2017

Computer Physics Communications

Self Cite

View full text Add to dashboard Cite

show abstract

“…It has been proposed for use in DFT in refs. [28,27], and has recently been adopted by several groups [4,2].…”

Section: Introductionmentioning

confidence: 99%

Parallel eigensolvers in plane-wave Density Functional Theory

Levitt

Torrent

2015

Computer Physics Communications

View full text Add to dashboard Cite

We consider the problem of parallelizing electronic structure computations in plane-wave Density Functional Theory. Because of the limited scalability of Fourier transforms, parallelism has to be found at the eigensolver level. We show how a recently proposed algorithm based on Chebyshev polynomials can scale into the tens of thousands of processors, outperforming block conjugate gradient algorithms for large computations.

show abstract

“…They range from general-purpose and digital signal multicore processors to GPUs. This analysis is especially timely in the decade where the power wall has arisen as a major obstacle to build faster processors.An alternative approach to the solution of a sequence of correlated eigenproblems is proposed in paper [2]. The resulting eigensolver is optimized regarding the number of matrix-vector multiplications and parallelized for distributed memory architectures using the Elemental library framework.…”

mentioning

confidence: 99%

“…An alternative approach to the solution of a sequence of correlated eigenproblems is proposed in paper . The resulting eigensolver is optimized regarding the number of matrix–vector multiplications and parallelized for distributed memory architectures using the Elemental library framework.…”

mentioning

confidence: 99%

10th international conference on Parallel Processing and Applied Mathematics, PPAM 2013

Wyrzykowski

Tudruj

2014

Concurrency and Computation

View full text Add to dashboard Cite

in Warsaw, Poland.PPAM is a biennial series of international conferences dedicated to exchanging ideas between researchers involved in parallel and distributed computing, including theory and applications, as well as applied and computational mathematics. The focus of PPAM 2013 was on models, algorithms and software tools that facilitate efficient and convenient use of modern parallel and distributed computing systems, as well as on large-scale applications.PPAM 2013, the jubilee PPAM conference, was organized by the Department of Computer and Information Science of the Czestochowa University of Technology, under the patronage of the Committee of Informatics of the Polish Academy of Sciences, in cooperation with the Polish-Japanese Institute of Information Technology.This meeting gathered the largest number of participants in the history of PPAM conferences-more than 230 participants from 32 countries. A strict reviewing process, with each submission reviewed at least three times, resulted in acceptance of 143 contributed papers with the acceptance rate of 56%. The accepted papers were presented at the regular tracks of the PPAM 2013 conference, as well as during the workshops, which were important and integral components of PPAM meetings.Based on the results of the reviews, selected papers were recommended for a special journal issue. Besides quality, another important goal which influenced the paper selection was a maximum possible thematic consistency of the issue. The authors were contacted after the conference and invited to submit revised and extended versions of their papers. These new versions were reviewed independently by three reviewers. Finally, ten contributions were accepted for publication. They are summarized below.Paper [1] analyzes interaction occurring in the triangle: performance-power-energy for the execution of a pivotal numerical algorithm, the iterative Conjugate Gradient (CG) method, on a diverse collection of parallel multithreaded architectures. They range from general-purpose and digital signal multicore processors to GPUs. This analysis is especially timely in the decade where the power wall has arisen as a major obstacle to build faster processors.An alternative approach to the solution of a sequence of correlated eigenproblems is proposed in paper [2]. The resulting eigensolver is optimized regarding the number of matrix-vector multiplications and parallelized for distributed memory architectures using the Elemental library framework. Numerical results show that the proposed solver achieves excellent scalability and is competitive with current dense linear algebra parallel eigensolvers.Paper [3] focuses on an efficient implementation of the multidimensional Monte-Carlo integration on various distributed-memory parallel computers and clusters of multi-/manycore nodes. In particular, it addresses the issue how to use multiple cores of CPUs together with multiple GPU accelerators within a single node, achieving a reasonable load balancing of available computational resources.The adapt...

show abstract

An optimized and scalable eigensolver for sequences of eigenvalue problems

Cited by 9 publications

References 30 publications

High-performance generation of the Hamiltonian and Overlap matrices in FLAPW methods

High-performance generation of the Hamiltonian and Overlap matrices in FLAPW methods

Parallel eigensolvers in plane-wave Density Functional Theory

10th international conference on Parallel Processing and Applied Mathematics, PPAM 2013

Contact Info

Product

Resources

About