This paper introduces the support for developing efficient parallel programs in plain Java in the Gaspar framework. The framework supports a complete set of data locality optimisations that are introduced with Java annotations, keeping the original Java code platform independent and postponing performance tuning tasks. Performance results show that data layout improvements can provide a 2.5-fold speedup, and, when combined with the exploitation of parallelism, can deliver a 50-fold speedup when compared with an unoptimised sequential base Java application running on a 24-core machine.

中文翻译：

---

使用普通 Java 进行高效的高级编程

本文介绍了在 Gaspar 框架中对使用纯 Java 开发高效并行程序的支持。该框架支持通过 Java 注释引入的一整套数据局部性优化，保持原始 Java 代码平台独立并推迟性能调优任务。性能结果表明，与在 24 核机器上运行的未优化的顺序基础 Java 应用程序相比，数据布局改进可以提供 2.5 倍的加速，并且与并行性的利用相结合时，可以提供 50 倍的加速。