Coordinating parallel processors

A variety of coordination primitives have been proposed for memorysharing MIMD parallel processors. The most familiar of these primitives are probably test-and-set and swap. In this report we show how a simple modification of the fetch-and-add coordination primitive [I] and [3] recently proposed for the "NYU Ultracomputer" [2] yields a hardware technique that may also be used for test-and-set and swap. enables one to support a wide range of operations on any multiprocessor. implementation This technique network-based