Design and application of the RPA II

In this thesis a highly parallel SIMD machine, the original RPA, is evaluated for numerical processing applications. The results of this evaluation highlighted several deficiencies in this architecture. The most important of these were the progammability aspects,input/output performance, floating-point performance and the fact that the system architecture was not balanced with respect to performance. The experience gained in the original RPA project as well an investigation of hardware and software models was used to develop the RPA II architecture. In this architecture all the deficiencies identified in the original RPA architecture have been corrected. The RPA II architecture is based on an 8 bit wide Processing Element (PE) and supports inter-PE packet switching and a novel inactivity control mechanism. The architecture of the individual processing element also supports several concurrent operations. This is achieved though the provision of a communication co-processor in addition to the normal arithmetic processor. Another feature of the RPA II architecture is the support for floating point operations through a 32 bit normalization shift register and hardware multiplication circuitry. Two patent applications, one for the inter-PE communications scheme and the other for the inactivity control mechanism, are also described. The application of the RPA II architecture to a less regular problem is then described. This application is a parallel implementation of the production system language, OPS 5. The estimated performance of OPS 5 is extremely encouraging and compares favourably with special purpose architectures of a similar complexity.