A Study on the Viability of a Production-Quality Metaobject Protocol-Based Statically Parallelizing Compiler

Abstract


A compiler that automatically chooses a program's parallelization is often unable to choose either the best one or the particular one that a programmer has in mind. This has led to systems that provide the programmer with explicit control over a program's parallelization, for example, via compiler pragmas. The pragma approach is like the metaobject protocol (MOP) approach in that pragmas provide control over what would otherwise be hidden aspect's of an implementation. However, it differs because the set of pragmas is fixed, thereby limiting the amount of control provided. We investigated whether it was possible to increase the amount of control using the full MOP approach. We were in fact, successful, but the resulting MOP differs from previous ones in that it is present at compile-time rather than at run-time. In this paper, we compare the MOP approach with other approaches, and discuss what is needed in order to produce a production-quality MOP-based statically parallelizing compiler.


Go to to Selected Papers page.