中文摘要：

针对应用在移植到异构多核高性能计算机系统中所面临的可移植性差以及性能优化难度大的问题,文中提出一种面向异构多核架构的自适应编译框架.通过源到源编译解决传统并行编程模型应用向异构多核架构的映射问题;同时利用动态剖分信息,自适应地调整插桩并配置优化策略,形成迭代式的自动优化过程.文中自适应编译框架将软硬件映射机制与优化策略结合,有效地解决了同构并行应用向异构多核架构的移植问题并提高了应用的整体性能.实验结果表明,文中基于Cell架构实现的原型系统,很好地解决了异构多核架构下应用移植性等问题,同时应用性能有所提高.

英文摘要：

To improve the application performance and portability on the novel hardware, this paper proposes a self-adaptive compilation framework for heterogeneous multi-core architecture. This framework uses source-to-source compiling technique to address the transformation prob- lems of the application from the traditional parallel programming model to the heterogeneous multi-core architecture, and analyzes dynamic profiling information to self-adaptively adjust instrument and configure the optimization strategy. The framework uses an iterative optimization method to combine the mapping mechanisms with the performance optimization strategy. The iterative automatic optimization method is formed to ensure the efficiency of parallel application migration with fully exploiting the ability of the heterogeneous multi-core architecture. The framework has been prototyped on the Cell architecture and tested with a set of examples and the experimental results are promising.