Architectures for Higher Speed List Processing

点击翻译按钮获取中文摘要

With the recent interest in Artificial Intelligence, nonmumeric (specially list) processing has been brought to the forefront. This has led to a demand for computer architectures that are suited to nonnumeric processing tasks. In this research proposal, the researchers identify four major area in the list processing paradigm namely: 1) function calls, 2) environment maintenance, 3) data structure access and 4) garbage collection. They propose to investigate methods that a) allow fast and efficient implementations of function call and environment maintenance mechanisms, b) allow streamed access to data structures and c) reduce the burden on the garbage collection process. For the efficient implementation of function call and environment maintenance mechanisms, they will investigate new cache management strategies for the stacks that are used to implement the mechanisms. Since streamed access to data is crucial to a pipelined implementation of list-processing architecture, they will investigate representations of list structures that allow streamed access. Current list-processing architectures are constrained to accessing list structures serially. They also will investigate strategies that reduce the overhead due to garbage collection by reclaiming some garbage cells without passing them on to garbage collection process.