RIA: Combining Static and Dynamic Techniques for ExploitingLocality
RIA: Combining Static and Dynamic Techniques for ExploitingLocality
批准号:
9110766
负责人:
Anne Rogers
金额:
$6.25万
依托单位:
依托单位国家:
美国
项目类别:
Standard Grant
财政年份:
1991
资助国家:
美国
项目状态:
已结题
起止时间:
1991-08-01 至 1994-01-31
中文摘要
这项研究的最终目标是为并行机器开发复杂的编译器。内存延迟是我们研究的重点,因为它是当今并行架构中驱动许多设计决策的瓶颈。随着系统变得越来越大,让所有内存靠近所有处理器变得越来越困难,这可能会增加有效的内存延迟。当前减少内存延迟的解决方案可以分为动态和静态两类。动态解决方案包括多缓存和共享虚拟内存。这些方法基于这样的假设:程序表现出高度的局部性,因此,处理缓存缺失或页面错误的成本可以平摊到缓存行或页面的许多引用上。当前的静态解决方案依赖于程序员有效地使用内存。提出了三个相关的研究方向:减少有效内存延迟问题的静态解决方案,这比目前的解决方案对机器的详细知识要求要低得多;减少内存延迟的静态和动态技术的结合;在递归定义的动态数据结构中利用局域性的静态和动态方法。//
英文摘要
The ultimate goal of this research is to develop sophisticated compilers for parallel machines. Memory latency is the focus of our research because it is the bottleneck that drives many design decisions in parallel architectures today. As systems become larger, it is increasingly difficult to have all memory close to all processors which can increase effective memory latency. Current solutions for reducing memory latency can be categorized as either dynamic or static. Dynamic solutions include multi-caches and shared virtual memory. These methods are based on the assumption that programs exhibit a high degree of locality and therefore, the cost of servicing a cache miss or a page fault can be amortized over many references that cache line or page. Current static solutions rely on the programmer to use memory efficiently. Three related research directions are proposed: a static solution to the problem of reducing effective memory latency that requires much less detailed knowledge of the machine by the programmer than present solutions, the combination of static and dynamic techniques for reducing memory latency, and static and dynamic methods for exploiting locality in recursively defined dynamic data structures.//
期刊论文(0)
专著(0)
科研奖励(0)
会议论文
海外基金