Programming with a high degree of parallelism in fortran

✍ Scribed by C.R. Jesshope

• Publisher: Elsevier Science
• Year: 1982
• Tongue: English
• Weight: 864 KB
• Volume: 26
• Category: Article
• ISSN: 0010-4655
• DOI: 10.1016/0010-4655(82)90111-4

✦ Synopsis

Many parallel extensions to FORTRAN have been proposed by 'supercomputer' manufacturers. The major differences between these language extensions is reviewed briefly. The Principle of Conservation of Parallelism is also introduced, which is argued to be a desirable foundation on which to base the development of code for parallel computers. Simply stated it requires that the degree of parallelism should not increase during the translation of an algorithm from a concept to a high level language (FORTRAN say) and finally into the machine code of the target computer. Cray FORTRAN and other vectorising compilers do not adhere to this principle, as the parallelism increases from I to some greater degree during the compilation process. A simple example will be used to illustrate the implications of this principle, which shows that it will reduce operations at the expense of storage locations. Vectorising compilers may reduce this storage requirement but will increase the number of operations. Two further examples of highly parallel and practical codes are also presented. These illustrate the compactness of code and the close relationship between the mathematical description of the problem and the FORTRAN implementation. The examples show the matrix multiplication and fast Fourier transform algorithms.