Next Generation Software: An Integrated Framework for Performance Engineering and Resource-Aware Compilation

EIA-9975019 Polychronopoulos，ConstantineUniversity of Illinois-Urbana下一代软件：性能工程和资源感知编译的集成框架该提案将进行研究，以开发系统和协同方法，用于开发集成的性能工程框架和资源感知编译和运行时系统，他们建议在测试的背景下进行这项研究，在重要应用中验证和演示已开发的这些技术，其中视觉应用是本提案的重点。 关于完整的分布式和并行计算系统的性能工程框架的努力包括建模语言的硬件和软件组件开发异构组件模型，结合这些模型来表示硬件和软件系统和该系统的视图的手段。 为了创建这样的功能，他们将开发组合技术，使模型能够相互交互，共享状态，事件或结果，以及可扩展的建模能力。编译器和运行时支持系统将设计用于程序转换的资源感知算法，多语言资源感知优化编译器，以及由编译器生成并嵌入用户代码的低开销自适应运行时支持。 通过使用有关应用程序内部的信息和目标体系结构的知识，将开发出可动态扩展并响应系统资源类型和数量的重新配置的可执行代码的技术。上述建模、编译和运行时技术将协同组合，以获得集成的编译器/运行时/性能工程框架，他们将展示这样一个框架的能力与应用程序，如面部跟踪，视频过滤和索引。

EIA-9975019Polychronopoulos, ConstantineUniversity of Illinois-UrbanaNext Generation Software: An Integrated Framework for Performance Engineering and Resource-Aware CompilationThis proposal will perform research to develop systematic and synergistic approaches for developing an integrated performance engineering framework and resource aware compilation and run-time system, and they propose to perform this research in the context of testing, validation and demonstration of the developed these technologies on important applications, a vision application being a focal one in this proposal. With respect to the performance engineering framework for complete distributed and parallel computing systems the efforts include modeling languages of hardware and software components for developing heterogeneous component models, means of combining these models to represent the hardware and software system and views of that system. To create such capabilities they will develop composition techniques that enable models to interact with each other, share state, events or results, and ability for scalable modeling.The compiler and runtime support system will design resource-aware algorithms for program transformations, a multilingual resource-aware optimizing compiler, and low-overhead, adaptive run-time support to be generated by the compiler and embedded in user code. By using information about application internals and knowledge of the target architectures will develop techniques that will produce executable code that scales dynamically and is responsive to reconfigurations in the type and number of system resources.The modeling, compilation and runtime techniques above will be synergistically combined to obtain an integrated compiler/runtime/performance engineering framework, and they will demonstrate the abilities of such a framework with applications like face-tracking, video filtering and indexing.