权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

I-Corps: Compiler Technology for Modern Many-core Architectures

I-Corps：现代多核架构的编译器技术

基本信息

批准号：
1342156
负责人：
Aviral Shrivastava
金额：
$ 5万
依托单位：
Arizona State University
依托单位国家：
美国
项目类别：
Standard Grant
财政年份：
2013
资助国家：
美国
起止时间：
2013-07-01 至 2014-06-30
项目状态：
已结题

来源：
https://www.nsf.gov/awardsearch/showAward?AWD_ID=1342156&HistoricalAwards=false
关键词：
Corps Compiler Technology Modern Many

项目摘要

This project focuses on the development of a complier that automatically performs correct and efficient memory management. The compiler automatically manages the data of the application on a multi-core architecture that does not provide any memory management in hardware. Such architectures are scalable (i.e., processors with hundreds and thousands of cores can be designed), and are more power-efficient.This compiler does this through application analysis. It analyzes each kind of code/data (i.e., code, stack data, heap data and global variables) and inserts data management instructions in the application, so that the application can execute correctly and efficiently on many-core processors that will not have any support for memory management in hardware. This compiler may be needed by system development companies that are developing relatively high-performance computing systems, for example, 3D scanner, high capacity 3D printers, TVs, audio video processing in cars etc. All these companies may want to use the processor with best performance and lowest power. However, top of the line processors will soon not have support for memory management in hardware, and the existing applications will not compile them. This compiler may have the ability to enable them to utilize the power and performance of the latest many-core processor for their systems. To the end user, they receive a power-efficient product with higher performance. Hand-held devices thus developed, while providing improved performance, will be lighter and will require fewer recharges.

这个项目的重点是开发一个编译器，自动执行正确和有效的内存管理。编译器在多核架构上自动管理应用程序的数据，该架构不提供硬件中的任何内存管理。这样的体系结构是可扩展的（即，可以设计具有成百上千个核的处理器），并且更节能。该编译器通过应用程序分析来实现这一点。它分析每种代码/数据（即，代码、堆栈数据、堆数据和全局变量），并在应用程序中插入数据管理指令，使得应用程序可以在不支持硬件中的存储器管理的众核处理器上正确且有效地执行。该编译器可能需要由系统开发公司正在开发相对高性能的计算系统，例如，3D扫描仪，高容量3D打印机，电视，音频视频处理在汽车等所有这些公司可能希望使用具有最佳性能和最低功耗的处理器。然而，顶级处理器很快将不支持硬件中的内存管理，现有的应用程序将无法编译它们。这个编译器可能有能力使他们能够利用最新的众核处理器的能力和性能。对于最终用户来说，他们收到的是具有更高性能的高能效产品。这样开发的手持设备，在提供改进的性能的同时，将更轻并且将需要更少的再充电。