The CMAX translator converts applications written in scalable Fortran 77 to parallel Connection Machine Fortran. The most obvious part of the translation problem, and one addressed by a number of previous translators as well as CMAX, is loop vectorization: the substitution of array syntax and intrinsics for Fortran 77 DO loops. A less obvious (but equally important) part of the translation problem is the conversion of constructs arising from Fortran 77's linear memory model into code which does not rely on storage or sequence association. Most such constructs can only be detected and repaired through interprocedural analysis and transformation; CMAX is the first translator to perform such repairs. This paper describes the CMAX translator and some of the more important transforms that we have distilled from our users' porting experiences.

This work presents the first MPP execution of the a Cray Y-MP.Finally, we achieved at least 2 times the price/performance of a Cray Y-MP.

The performance of a program on a distributed-memory parallel computer depends on the algorithms employed, the structure and speed of the machine's communication network, and the ways in which data are distributed to the processors. This paper addresses the last of these concerns, the problem of data mapping.