Cray Research Inc, Minneapolis, Minnesota, has added a new automatic parallel processing capability to its CFT77 Fortran compiling system. Called Autotasking, the software is said to restructure programs automatically for processing in parallel, removing the need for users to write complex code specifically for parallel environments. Cray claims that on an eight processor system, Autotasking performs program execution up to 7.8 times faster than a single processor system. It is available with version 3.0 of the CFT77 and can be used on the Cray Y-MP, X-MP EA, and X-MP multiprocessor systems running under Cray’s Unicos version of Unix. Parallel processing performance varies according to the number of processors applied and the amount of parallelism a program contains, which has traditionally meant a high level of programming expertise is required. This restriction has tended to limit the spread of parallel systems. But Autotasking is only one of a number of recently announced systems that claim to decompose programs for users automatically. The Cray system follows hard on the heels of Artificial Intelligence’s Strand88, a language which performs the same task, allowing the transcription of sequential code written on concurrent machines such as workstations to the parallel environment, or being used as a writing tool in its own right. Strand88 can embed to Fortran, C, or Cobol code for transcription, whereas Autotasking is available only on Cray’s Fortran compiler at present – the company intends to migrate the system for other languages in the future. And Japan’s ICOT Institute for Computers Of Tomorrow is also set to release a similar fifth generation language for parallel processing called KL1, Kernel Language 1, although there are no details as yet on when it will become commercially available.
This article is from the CBROnline archive: some formatting and images may not be present.
CBR Online legacy content.