Abstract: "This paper presents and evaluates the main architectural features of a parallel inference machine, PIM/p, which is now being developed for the high-speed execution of AI programs written in the concurrent logic programming language KL1, PIM/p has a hierarchical structure where eight...

Ausführliche Beschreibung

Bibliographische Detailangaben

Format:	Buch
Sprache:	English
Veröffentlicht:	Tokyo, Japan 1989
Schriftenreihe:	Shin-Sedai-Konpyūta-Gijutsu-Kaihatsu-Kikō <Tōkyō>: ICOT technical report 514
Schlagworte:	Inference Logic programming Parallel programming (Computer science)
Zusammenfassung:	Abstract: "This paper presents and evaluates the main architectural features of a parallel inference machine, PIM/p, which is now being developed for the high-speed execution of AI programs written in the concurrent logic programming language KL1, PIM/p has a hierarchical structure where eight processing elements are combined with shared memory to form a cluster, and more than sixteen clusters are connected by a hypercube network. To decrease the numver of code fetches on the common bus of a cluster, we have developed a macro instruction call mechanism that allows high-level instructions to be executed in the otherwise RISC-like instruction stream. Some software controlled cache memory assist instructions have also been introduced to delete unneeded operand data transfers on the common bus. Simulation has shown their effectiveness. We also evaluated the application of the two main deadlock-free routing methods to the intercluster network, and selected the E-cube method because of its relatively good performance and simple hardware."
Beschreibung:	6 S. graph. Darst.