Computing Architectures and Programming Methodologies for Heterogeneous High-Performance Computing and Embedded Applications

现代的编译器和编程工具使计算机能够被许多领域的专家使用，以便他们能够开发自己的应用程序。 然而，今天需要使用并行性来获得性能，这增加了普通程序员不容易理解的重大编程挑战。此外，随着我们转向GPGPU（通用图形处理单元）和FPGA（现场可编程门阵列）等专用架构以寻求更高的性能和更高的能效，编程变得更加困难。 大多数领域专家不具备有效利用这种计算技术的专业知识。 本提案中的计划是与几位领域专家合作，他们必须具有重大挑战性质的重要计算需求-具有重大影响的大问题。 大脑建模就是一个例子。 拟议工作的关键要素是，计算机工程（CE）研究生将“嵌入”需要帮助开发应用程序的协作组。 这将使CE学生获得有关领域专家的要求，同时帮助他们与他们的应用程序的深入了解。 期望这种合作将激发新的计算方法，从而导致计算机架构和系统的新变化。 大多数计算机架构师处理系统的非常具体的底层方面，例如内存、互连和处理器的组织。 该提案承认工作应用程序是特定领域用户的工具。 在真实的系统上构建工作应用程序将更好地告知计算机架构师下一代计算系统的需求。 与此同时，这种研究方法为特定领域的用户提供了显着的好处，帮助他们实现最有效和最高性能的应用程序，他们可以用来推进自己的科学研究。 这有可能比计算机体系结构研究产生更直接的影响！

Modern compilers and programming tools have made computers accessible to the experts in many domains so that they can develop their own applications. However, the need to use parallelism to get performance today adds significant programming challenges that are not easily understood by the average programmer. Moreover, programming is becoming even more difficult as we move to specialized architectures such as GPGPUs (General-Purpose Graphics Processing Units) and FPGAs (Field-Programmable Gate Arrays) to find more performance and better energy efficiency. Most domain experts do not have the expertise to make effective use of such computing technologies. The plan in this proposal is to work with several domain experts who must have significant computational requirements of a grand challenge nature -- large problems with significant impact. Modeling the brain is one example. The key element of the proposed work is that a computer engineering (CE) graduate student will be "embedded" with the collaborating group that needs help developing an application. This will allow the CE student to gain intimate knowledge about the requirements of the domain experts while helping them with their application. The expectation is that this collaboration will inspire new approaches to computing that will result in novel changes to computer architectures and systems. Most computer architects address very specific low-level aspects of systems, such as the memory, the interconnect and the organization of the processors. This proposal recognizes that working applications are the tools of the domain-specific users. Building working applications on real systems will better inform the computer architect of the requirements for next-generation computing systems. At the same time, this approach to research provides a significant benefit to the domain-specific users by helping them to implement the most efficient and highest-performing applications that they can use to advance their own scientific research. This has the potential to create more immediate impact than the computer architecture study!