Shared-memory parallel programming in C++

Abstract

A set of portable parallel-programming constructs for C, implemented in M4 macros called Parmacs, developed by researchers at Argonne National Laboratory, is considered. The Parmacs macros make it possible to write parallel C programs for shared-memory, distributed-memory, and mixed-memory systems; are portable, highly functional, and efficient; and provide sufficient functions to build a variety of parallel programs. However, they have several problems. The M4 macros are awkward to use and hard to read, so it is easy to make errors when using them. They are a mechanism outside the C language; thus, someone familiar only with C cannot readily read the code. They are not cleanly extensible, and their use is not type-checked. It is shown that using C++, rather than C, addresses these problems adequately. The Parmacs macros are described, the C++ Parmacs classes and runtime model are presented, and examples that show the use of the C++ Parmacs constructs are given. The work described considers only shared-memory constructs. Directions for future work are indicated.