Post-Doctoral Experimental Research for the Evaluation of Multithreaded Architectures

9805216 Kavi，Krishna M.阿拉巴马大学亨茨维尔分校Cise多线程体系结构评估博士后实验研究最近的技术趋势导致内存和处理器之间的性能差距扩大。多线程被吹捧为一种解决方案，可以通过同时执行几个指令流来最大限度地减少由于内存延迟造成的CPU周期损失。在过去的几年中，PI一直在研究基于数据流范例和混合数据流控制流范例的多线程体系结构。这项研究导致了用于数据流体系结构的高速缓存存储器、连贯的I结构高速缓存存储器以及用于多线程体系结构的DOACROSS循环的交错分配。博士后研究人员将扩展跟踪生成工具的工作，为数据流和混合体系结构生成足够大的指令跟踪。只要有可能，传统体系结构的跟踪将被转换为数据流和多线程系统的跟踪。这将大大有助于PI的研究，以评估数据流/混合模型和新的体系结构创新(如计划的数据流)的适用性，并推测性地为更广泛的编程模型和应用程序预加载线程上下文。这项研究的成功完成将对计算机体系结构研究产生更广泛的影响，因为它有助于产生创新体系结构的大量痕迹，而不需要编译器和运行时系统来对新体系结构进行试验性评估，同时有助于扩大博士后助理。

9805216 Kavi, Krishna M. University of Alabama in Huntsville CISE Postdoctoral Experimental Research for the Evaluation of Multithreaded Architectures Recent trends in technology have resulted in a widening of the performance gap between memory and processors. Multithreading has been touted as the solution to minimize the loss of CPU cycles due to memory latency by executing several instruction streams simultaneously. Over the past several years, the PI has been investigating multithreaded architectures, both based on dataflow paradigm and hybrid dataflow-control flow paradigms. This research led to the introduction of cache memories for dataflow architectures, coherent I-structure cache memories, and staggered allocation of DOACROSS loops for multithreaded architectures. The post-doctoral researcher will extend the work on trace generation tools to generate sufficiently large instruction traces for dataflow and hybrid architectures. Whenever possible, traces for conventional architectures will be converted into traces for dataflow and multithreaded systems. This will aid the PI's research significantly in evaluating the applicability of the dataflow/hybrid models and new architectural innovations such as scheduled dataflow, and speculatively preloading thread contexts for a wider range of programming models and applications. The successful completion of the research will have a broader impact on computer architecture research by facilitating the generation of large traces for innovative architectures without the need for compilers and runtime systems for experimentally evaluating the new architectures, and at the same time help to broaden the Postdoctoral Associate.