2010Unpublished venueRequires access

A cost estimation based speculative path determination method for speculative thread partitioning

Yuancheng Li, Yinliang Zhao, Meirong Li, Yonggang Zhao

Open publisher page 1 citations

Abstract

As an effective mechanism for parallelizing irregular programs which are hard by conventional approaches, speculative multithreading (SpMT) technology has been proposed and being rapidly developed. While speculative parallelization can potentially deliver significant speedup for programs, the accuracy of the control- and data- speculation would limit these speedups in practice. In fact, different thread partitioning results by different speculative paths would lead to different overheads; and these overheads would seriously affect the parallel performance. Therefore, it is very important to understand the trade-offs between different speculative paths. The largest problem is that these existing prediction and speculation methods do not integrate the data dependence into the control speculation in order to obtain the maximum benefit. In this paper, by discussing the trade-offs between different speculative paths, we propose a novel cost estimation based speculative path determination method which comprehensively takes account of control- and data-dependence. By this method, we attempt to seek a speculative path which has the minimum cost overhead. The experimental results show that there are interesting trade-offs between different speculative paths and we can indeed get better performance. On average, we achieve 5.37% performance improvement.

About this research paper

What this paper is about

As an effective mechanism for parallelizing irregular programs which are hard by conventional approaches, speculative multithreading (SpMT) technology has been proposed and being rapidly developed. While speculative parallelization can potentially deliver significant speedup for programs, the accuracy of the control- and data- speculation would limit these speedups in practice. In fact, different thread partitioning results by different speculative paths would lead to different overheads; and these overheads would seriously affect the parallel performance. Therefore, it is very important to understand the trade-offs between different speculative paths. The largest problem is that these existing prediction and speculation methods do not integrate the data dependence into the control speculation in order to obtain the maximum benefit. In this paper, by discussing the trade-offs between different speculative paths, we propose a novel cost estimation based speculative path determination method which comprehensively takes account of control- and data-dependence. By this method, we attempt to seek a speculative path which has the minimum cost overhead. The experimental results show that there are interesting trade-offs between different speculative paths and we can indeed get better performance. On average, we achieve 5.37% performance improvement.

Why it matters

OpenAlex reports 1 citations for this work. Citation counts describe recorded attention and do not establish research quality.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

As an effective mechanism for parallelizing irregular programs which are hard by conventional approaches, speculative multithreading (SpMT) technology has been proposed and being rapidly developed. While speculative parallelization can potentially deliver significant speedup for programs, the accuracy of the control- and data- speculation would limit these speedups in practice. In fact, different thread partitioning results by different speculative paths would lead to different overheads; and these overheads would seriously affect the parallel performance. Therefore, it is very important to understand the trade-offs between different speculative paths. The largest problem is that these existing prediction and speculation methods do not integrate the data dependence into the control speculation in order to obtain the maximum benefit. In this paper, by discussing the trade-offs between different speculative paths, we propose a novel cost estimation based speculative path determination method which comprehensively takes account of control- and data-dependence. By this method, we attempt to seek a speculative path which has the minimum cost overhead. The experimental results show that there are interesting trade-offs between different speculative paths and we can indeed get better performance. On average, we achieve 5.37% performance improvement.

Key concepts: Speculative multithreading, Speculative execution, Speculation, Computer science, Multithreading, Thread (computing), Parallel computing, Speedup

Related papers

Back to paper searchBrowse research topicsOriginal source
A cost estimation based speculative path determination method for speculative thread partitioning — Research Paper | ScholarLens