A coprocessor design for the architectural support of non-numeric operations

Computer Science is concerned with the electronic manipulation of information. Continually increasing amounts of computer time are being expended on information that is not numeric. This is represented in part by modem computing requirements such as the block moves associated with context switching and virtual memory management, peripheral device communication, compilers, editors, word processors, databases, and text retrieval. This dissertation examines the traditional support of non-numeric information from a software, firmware, and hardware perspective and presents a coprocessor design to improve the performance of a set of non-numeric operations. Simple micro-coding of operations can provide a degree of performance improvement through parallel execution of instructions and control store access speeds. New special purpose parallel hardware algorithms can yield complexity improvements. This dissertation presents a parallel hardware regular expression searching algorithm which requires linear time and quadratic space compared to software uniprocessor algorithms which require exponential time and space. A very large scale integration (VLSD implementation of a version of this algorithm was designed, fabricated, and tested. The hardware. searching algorithm is then combined with other special purpose hardware to implement a set of operations. Simulation is then used to quantify the performance improvement of the operations when compared to software solutions. A coprocessor approach allows the optional addition of hardware to accelerate a set of operations. This is appropriate from a complex instruction set computer (CISC) perspective since hardware acceleration is being utilized. It is also appropriate from a reduced instruction set computer (RISC) perspective since the operations are distributed away from the central processing unit (CPU)