Browse Prior Art Database

Accumulator Based Vector Instructions

IP.com Disclosure Number: IPCOM000053641D
Original Publication Date: 1981-Oct-01
Included in the Prior Art Database: 2005-Feb-12

Publishing Venue

IBM

Related People

Authors:
Agerwala, TKM [+details]

Abstract

A class of triadic vector accumulator-based instructions is set forth. This type of machine approach has definite advantages over a register-to-register machine with multiple pipelines or a memory to memory machine. A uni-pipeline machine operates off a central storage with limited parallelism. A multiprocessor approach provides increased parallelism, but if operated off a central storage, a switching problem arises. If each processor has local storage and no central storage exists, a data movement problem is created. A compromise approach is to have multiple pipelines, the bulk of the storage being central with some local storage. An approach along these lines is described here. The utility of the approach is illustrated using two specific triadic accumulator instructions.