A multiprocessor computer system for integrated numeric and symbolic computation

A class of problems, called signal understanding (SU) systems, that involves intensive numeric, as well as symbolic testing, is considered. A shared-memory multiprocessor computer system designed specifically to facilitate the development and evaluation of parallel SU algorithms, which would be a basic tool for developing real-time SU systems, is described. This machine, called the MX-1, runs Common LISP, which has been augmented with extensions for user-directed parallelism. It is, in essence, a parallel LISP machine with digital signal processors (DSPs) integrated into the architecture to provide powerful numeric computational capability. The MX-1 has 16 crossbar interconnected processing nodes, each with a Motorola 68020 CPU operating at 16 MHz, 8 Mbytes of dynamic random-access memory and its own Weitek DSP coprocessor that can do 10 million multiply-accumulate operations per second. High-speed I/O ports are provided for direct entry of sensor data into the DSPs. Two 4-node systems and a 16-node system have been fabricated, and several applications have been run on the systems