Reparallelization and Migration of OpenMP Programs

Klemm, Michael; Bezold, Matthias; Gabriel, Stefan; Veldema, Ronald; Philippsen, Michæl

doi:10.1109/ccgrid.2007.96

Cited by 6 publications

(5 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Another approach to automating the generation of parallel code was developed within the recent research on OpenMP programs and reparallelizing them for the Grid [8]. This work also covers Java programs and the use of distributed shared memory (DSM) for data exchange among tasks, but still requires from programmers dependence-free input and the explicit declaration of parallel loops via OpenMP directives.…”

Section: Resultsmentioning

confidence: 99%

LooPo-HOC: A Grid Component with Embedded Loop Parallelization

et al. 2008

View full text Add to dashboard Cite

This work integrates two distinct research areas of parallel and distributed computing, (1) automatic loop parallelization, and (2) component-based Grid programming. The latter includes technologies developed within CoreGRID for simplifying Grid programming: the Grid Component Model (GCM) and Higher-Order Components (HOCs). Components support developing applications on the Grid without taking all the technical details of the particular platform type into account (network communication, heterogeneity, etc.). The GCM enables a hierarchical composition of program pieces and HOCs enable the reuse of component code in the development of new applications by specifying application-specific operations in a program via code parameters. When a programmer is provided, e. g. , with a compute farm HOC, only the independent worker tasks must be described. But, once an application exhibits data or control dependences, the trivial farm is no longer sufficient. Here, the power of loop parallelization tools, like LooPo, comes into play: by embedding LooPo into a HOC, we show that these two technologies in combination facilitate the automatic transformation of a sequential loop nest with complex dependences (supplied by the user as a HOC parameter) into an ordered task graph, which can be processed on the Grid in parallel. This technique can significantly simplify GCM-based systems which combine multiple HOCs and other components. We use an equation system solver based on the successive overrelaxation method (SOR) as our motivating application example and for performance experiments.

show abstract

Section: Resultsmentioning

confidence: 99%

LooPo-HOC: A Grid Component with Embedded Loop Parallelization

et al. 2008

View full text Add to dashboard Cite

show abstract

“…Some OpenMP extensions have tried to relax the requirement of keeping constant the number of threads participating in the parallel region based on the idea of reparallelization Klemm et al (2007). Such an approach requires specifying safe points along the OpenMP code where the runtime may proceed with any work repartitioning.…”

Section: Related Workmentioning

confidence: 99%

Role-shifting threads: Increasing OpenMP malleability to address load imbalance at MPI and OpenMP

Criado,

Lopez,

Vinyals-Ylla-Catala

et al. 2023

The International Journal of High Performance Computing Applica

View full text Add to dashboard Cite

This paper presents the evolution of the free agent threads for OpenMP to the new role-shifting threads model and their integration with the Dynamic Load Balancing (DLB) library. We demonstrate how free agent threads can improve resource utilization in OpenMP applications with load imbalance in their nested parallel regions. We also demonstrate how DLB efficiently manages the malleability exposed by the role-shifting threads to address load imbalance issues. We use three real-world scientific applications, one of them to demonstrate that free agents alone can improve the OpenMP model without external tools, and two other MPI+OpenMP applications, one of them with a coupling case, to illustrate the potential of the free agent threads’ malleability with an external resource manager to increase the efficiency of the system. In addition, we demonstrate that the new implementation is more usable than the former one, letting the runtime system automatically make decisions that were made by the programmer previously. All software is released open-source.

show abstract

“…However, because N -the number of momentum p[i] is very large, it is quite time-consuming to complete the loop. In this work, we use OpenMP to paralyze our code, following the standard method in the article [10]. The code of a normal "for" loop is shown in the left panel of figure 1, where function A(p[i]) and function 𝐵(p[i]) are subroutines computing the right hand of the equation (13).…”

Section: Automatic Parallelization With Openmpmentioning

confidence: 99%

“…OpenMP has been very successful in exploiting structured parallelism in applications [8][9]. Particularly article [10] introduced the fundamental design of the OpenMP specification v2.5 in GCC. The implementation supports all the programming languages (C, C++, and Fortran), and it is generally available on any platform that supports Portable Operating System Interface (POSIX) threads.…”

Section: Introductionmentioning

confidence: 99%

Fast Numerical Solution of a Kind of Nonlinear Integral Equations—Dyson-Schwinger Equations for Quark Propagator in Hadron Physics

Huang

Chen

et al. 2023

J. Phys.: Conf. Ser.

View full text Add to dashboard Cite

The nonlinear integral equation has been widely studied and has become the heart of the matter in many scientific and engineering fields, such as seismology, optical fiber evolution, radio astronomy, and hadron physics with Quantum Chromodynamics. The Dyson-Schwinger Equations (DSEs) approach provides an essential nonperturbative approach to investigating the properties of hadrons and hot/dense quark matter. Mathematically, the Dyson-Schwinger Equations are a group of coupled nonlinear integral equations of quark propagators, gluon propagators, ghost propagators, and various vertices. On account of the non-linearity and singularity of the coupled equations, it is almost impossible to solve the DSEs analytically. One has to resort to the numerical solution of the equations, in which efficient fast algorithms are key points in practice. In this work, two improvements for numerically solving the nonlinear and singular integral equation for quark propagator in a vacuum are proposed. One is a modified interpolation method for unknown functions in the integral with high degrees of freedom. The other is the parallelization on CPUs with OpenMP in GCC Comparing the CPU times with different algorithms, our results indicate that our proposed methods can greatly improve the efficiency and reduce the computation time of the CPU.

show abstract

Reparallelization and Migration of OpenMP Programs

Cited by 6 publications

References 21 publications

LooPo-HOC: A Grid Component with Embedded Loop Parallelization

LooPo-HOC: A Grid Component with Embedded Loop Parallelization

Role-shifting threads: Increasing OpenMP malleability to address load imbalance at MPI and OpenMP

Fast Numerical Solution of a Kind of Nonlinear Integral Equations—Dyson-Schwinger Equations for Quark Propagator in Hadron Physics

Contact Info

Product

Resources

About