Compiling Fortran 8x array features for the connection machine computer system

Abstract

The Connection Machine@ computer system supports a data parallel programmin g style, making it a natural target architecture for Fortran 8x array constructs. The Connection Machine Fortran compiler generates VAX code that performs scalar operations and directs the Connection Machine to perform array operations.The Connection Machine virtual processor mechanism supports elemental operations on very large arrays. Most array operators and intrinsic functions map into single instructions or short instruction sequences. Noncontiguous array sections, array-valued subscripts, and parallel con@ructs such as WHERE and PORALL are also readily accommodated on the Connection Machine. In addition to such customary optimizations as common subexpression elimination, the CM Fortran compiler minimizes data motion for aligning array operations, minimizes transfers between the Connection Machine and the VAX and minimizes context switching for masked computations.

Automatic parallelization of APL-style programs

“…We note that data-parallelism can be exploited by an interpreter; but an interpreter cannot exploit functional parallelism since it requires dependency analysis of code. And data parallelism is also the primary source of parallelism so far effectively exploited on the Connection Machine ( [ 2 ] ) by declaring some variable to be parallel. But we believe that if only data parallelism can be effectively exploited, then the kind of applications which can benefit from parallel execution will be severely limited.…”

A framework for efficient execution of array-based languages on SIMD computers

“…The implementation of Fortran-90 for the CM is discussed in [Alb88]. The approach is the dual of ours with respect to data motion: the layout of an array in [Alb88] is a function of its shape, hence conformable arrays are always aligned.…”

“…The approach is the dual of ours with respect to data motion: the layout of an array in [Alb88] is a function of its shape, hence conformable arrays are always aligned. Thus motion is generated for selector operations but not for transformer operations.…”

“…This discussion is cast primarily in terms of discovering opportuntities for improvement of the default strategy from [Alb88] by analysis of the source and hence addresses the problem of automatic domain decomposition. What is less clear about that work is what notion of layout is used and how it supports calculation of the amount and cost of motion.…”

User-Defined Data Distributions in High-Level Programming Languages

“…Kali [21] was the first language to introduce distribution declarations in the context of distributed-memory architectures, together with a simple mechanism for userdefined distributions. Fortran D [14], Vienna Fortran [5], and Connection Machine Fortran [1], the major predecessors of the High Performance Fortran [15] effort, all offered facilities for combining multi-dimensional array declarations with the specification of a data distribution or alignment. Several other academic as well as commercial projects also contributed to the understanding necessary for the development of data parallel languages and the required compilation technology.…”