Parallel bit stream technology

提出的研究是关于在商用多核处理器上使用并行比特流技术进行高性能文本处理的工具和技术。目前的研究成果已经证实，在现代流水线处理器上，在字符串搜索、字符集转码、词法分析、正则表达式匹配和XML解析等应用中，并行比特流技术可以提供比传统的一次字节字符处理高一个数量级的性能改进。实际上，并行位流技术充分利用了寄存器内的并行性，以便使用128个位寄存器一次处理多达128个字符。然而，这涉及高级工具支持(例如，高级编程语言)尚不可用的新的编程形式。此外，当前商用处理器设计中涉及多核处理器的趋势正在产生需求，需要找到新的并行编程形式来利用芯片内并行。通过开发新的技术来利用寄存器内并行来支持支路内并行，可以预期这些并行文本处理技术相对于顺序一次字节数技术具有进一步的性能优势。通过开发映射到基于并行比特流技术的实现的高级编程工具，程序员可以广泛地获得这些性能优势，而不需要棘手且容易出错的低级编程。

The proposed research is into tools and technology for high-performance text processing using parallel bit stream technology on commodity multicore processors. Current research results have established that parallel bit stream technology can provide order of magnitude performance improvements over traditional byte-at-a-time character processing on modern pipelined processors in applications such as string search, character set transcoding, lexical analysis, regular expression matching and XML parsing. In effect, the parallel bit stream technology makes substantial use of intraregister parallelism in order to process up to 128 characters at a time using 128 bit registers. However, this involves new forms of programming for which high-level tool support (e.g., high-level programming languages) are not yet available. Furthermore, current trends in commodity processor design involving multicore processors are creating demands to find new forms of parallel programming to take advantage of intrachip parallelism. By developing new techniques to leverage intraregister parallelism in support of intrachip parallelism, further performance advantages of these parallel text processing techniques over sequential byte-at-a-time techniques can be expected. By developing high-level programming tools that map to implementations based on parallel bit stream technology, these performance benefits can be made broadly available to programmers without the need for tricky and error-prone low-level programming.