...
首页> 外文期刊>IEEE Transactions on Software Engineering >A transformation approach to derive efficient parallelimplementations
【24h】

A transformation approach to derive efficient parallelimplementations

机译:一种导出有效并行实现的转换方法

获取原文
获取原文并翻译 | 示例
           

摘要

The construction of efficient parallel programs usually requiresnexpert knowledge in the application area and a deep insight into thenarchitecture of a specific parallel machine. Often, the resultingnperformance is not portable, i.e., a program that is efficient on onenmachine is not necessarily efficient on another machine with a differentnarchitecture. Transformation systems provide a more flexible solution.nThey start with a specification of the application problem and allow thengeneration of efficient programs for different parallel machines. Thenprogrammer has to give an exact specification of the algorithmnexpressing the inherent degree of parallelism and is released from thenlow-level details of the architecture. We propose such a transformationnsystem with an emphasis on the exploitation of the data parallelismncombined with a hierarchically organized structure of task parallelism.nStarting with a specification of the maximum degree of task and datanparallelism, the transformations generate a specification of a parallelnprogram for a specific parallel machine. The transformations are basednon a cost model and are applied in a predefined order, fixing the mostnimportant design decisions like the scheduling of independent multitasknactivations, data distributions, pipelining of tasks, and assignment ofnprocessors to task activations. We demonstrate the usefulness of thenapproach with examples from scientific computing
机译:有效的并行程序的构建通常需要应用领域的专业知识,并且需要深入了解特定并行机的体系结构。通常,结果性能不是可移植的,即,在计算机上有效的程序不一定在具有不同体系结构的另一台计算机上有效。转换系统提供了更灵活的解决方案。它们从应用问题的规范开始,然后允许为不同的并行机生成高效的程序。然后,程序员必须给出算法的确切规格,以压制固有的并行度,然后从体系结构的低层详细信息中释放出来。我们提出了一个这样的转换系统,重点是对数据并行性的利用-与任务并行性的分层组织结构相结合。n从最大程度的任务和数据并行性的规范开始,这些转换生成特定并行机的并行程序的规范。转换不是基于成本模型,而是以预定义的顺序应用,固定最重要的设计决策,例如安排独立的多任务激活,数据分布,任务流水线以及将处理器分配给任务激活。我们用科学计算中的例子证明了这种方法的有效性

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号