Productivity and Performance of Global-View Programming with XcalableMP PGAS Language

Nakao, Masahiro; Lee, Jinpil; Boku, Taisuke; Sato, Mitsuhisa

doi:10.1109/ccgrid.2012.118

Cited by 25 publications

(12 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As these hardware environments are becoming more and more complex, the cost of developing efficient parallel programs that utilize these environments is also increasing. Under these circumstances, increasing the productivity of parallel programming is an important research topic, and several programming languages have been developed to achieve this aim such as X10 [4], Chapel [1], and XcalableMP [16]. Functional programming is often said to be highly productive due to the strong modularity of functional languages [8].…”

Section: Introductionmentioning

confidence: 99%

Evaluation of Libraries for Parallel Computing in Haskell — A Case Study with a Super-resolution Application

Matsumoto

Matsuzaki

2017

Journal of Information Processing

View full text Add to dashboard Cite

Abstract:Haskell is a functional language featuring lazy evaluation and referential transparency. On one hand, Referential transparency is useful for parallel computing because the results do not depend on the evaluation order, but on the other hand, parallel computing requires an evaluation order that is different from that of lazy evaluation. There are some parallel programming libraries for Haskell, such as Repa (regular parallel arrays) and Accelerate. However, little research has been conducted on evaluation with real applications, and the usefulness of these libraries remains unclear. In this study, we evaluated the usefulness of parallel programming libraries for Haskell with an application that applies a super-resolution technique to fMRI images. We developed a CPU-based parallel program with Repa and GPU-based parallel program with Accelerate and compared their performance. We obtained reasonable speedups for the program with Repa, but not for the program with Accelerate. We also investigated Accelerate's performance issues with an implementation in C and CUDA and the log from the Accelerate program. In this paper, we report our findings through a case study, focusing on the advantages and difficulties in parallel program development with Haskell.

show abstract

Section: Introductionmentioning

confidence: 99%

Evaluation of Libraries for Parallel Computing in Haskell — A Case Study with a Super-resolution Application

Matsumoto

Matsuzaki

2017

Journal of Information Processing

View full text Add to dashboard Cite

show abstract

“…The first "int-expr" specifies the lower bound region, while the second one specifies the upper bound in the dimension. If values of both widths are the same, only "int-expr" needs [7] to be specified. The "*" means that the entire region of the distributed array is allocated on other nodes as a halo region.…”

Section: Definition and Synchronization Of Halo Regionmentioning

confidence: 99%

“…With the aim of replacing MPI, we have been engaged in the design and development of XcalableMP (XMP) [6][7][8], which is a directive-based language extension based on the Partitioned Global Address Space (PGAS) [9] model. XMP supports both global-view and local-view memory models of PGAS.…”

Section: Introductionmentioning

confidence: 99%

XcalableACC: Extension of XcalableMP PGAS Language Using OpenACC for Accelerator Clusters

Nakao

Murai

Shimosaka

et al. 2014

2014 First Workshop on Accelerator Programming Using Directives

Self Cite

View full text Add to dashboard Cite

The present paper introduces the XcalableACC (XACC) programming model, which is a hybrid model of the XcalableMP (XMP) Partitioned Global Address Space (PGAS) language and OpenACC. XACC defines directives that enable programmers to mix XMP and OpenACC directives in order to develop applications that can use accelerator clusters with ease. Moreover, in order to improve the performance of stencil applications, the Omni XACC compiler provides functions that can transfer a halo region on accelerator memory via Tightly Coupled Accelerators (TCA), which is a proprietary network for transferring data directly among accelerators. In the present paper, we evaluate the productivity and the performance of XACC through implementations of the HIMENO Benchmark. The results show that thanks to the productivity improvements, XACC requires less than half the source lines of code compare to a combination of Message Passing Interface (MPI) and OpenACC, which is commonly used together as a typical programming model. As a result of these performance improvements, XACC using TCA achieved up to 2.7 times faster performance than could be obtained via the combination of OpenACC and MPI programming model using GPUDirect RDMA over InfiniBand.

show abstract

“…While Chapel does provide a means to take more explicit control of the parallelism when required, many programs can be written in such a way that they appear no different than serial programs. Other PGAS languages, or language extensions such as XcalableMP [55], provide "global view" objects where the array objects appear to be indexed conventionally, but ranges of indices reside on physically distinct distributed memories. For example, a "global view" matrix is represented as a two-dimensional array with non-overlapping portions residing in physically distinct memories.…”

Section: Coarray Fortranmentioning

confidence: 99%

Application of Modern Fortran to Spacecraft Trajectory Design and Optimization

Williams

Falck

Beekman

2018

2018 Space Flight Mechanics Meeting

View full text Add to dashboard Cite

Productivity and Performance of Global-View Programming with XcalableMP PGAS Language

Cited by 25 publications

References 8 publications

Evaluation of Libraries for Parallel Computing in Haskell — A Case Study with a Super-resolution Application

Evaluation of Libraries for Parallel Computing in Haskell — A Case Study with a Super-resolution Application

XcalableACC: Extension of XcalableMP PGAS Language Using OpenACC for Accelerator Clusters

Application of Modern Fortran to Spacecraft Trajectory Design and Optimization

Contact Info

Product

Resources

About