Parallel processing in LIPS

Abstract

Major problems with parallel languages are concerned with the complexity and lack of expressive power which makes it difficult to produce efficient code and the ease of mapping onto parallel hardware. Also, a parallel language must be independent of specific hardware features as well as the underlying run-time support for it to be manned efficiently on a range of hardware. This paper presents a parallel language (LIPS) which allows the programs to be mapped onto a range of parallel hardware in an efficient manner. LIPS encourages a pipelining effect to occur onboard each processor in a parallel system, with data becoming available to each node once the pipe is full. LIPS uses an asynchronous communication subsystem so that synchronisation delays can be avoided. The communication dynamics being of a dataflow nature, where variables are passed from one process to another as the values become available. LIPS offers facilities for representation of data, flow of control and the manipulation of the network of processes in a practical form suitable for constructing programs for MIMD computing.