Automatic storage optimization

Abstract

RetrospectiveThis paper, based on the author's 1979 thesis with the same title, addresses the problem of storage optimization. As the paper points out optimizing compilers focus on improving execution times while ignoring or even worsening the problem of storage use. The primary motivation was to automatically organize data so as to "optimize" the use of active storage during execution. Designers of optimizing compilers for procedural languages assumed that compilers would optimize execution times while users, language protocols, and run-time systems determined the storage layouts. (This view of the compiler's role in forming the execution stream and organizing storage persists today.)The Fabri paper proposes a storage-optimizing compiler and develops the algorithms and techniques needed to accomplish storage optimization. (Her thesis has the details.) Some of the algorithms are built on the state-of-the-art program optimization algorithms of the time but many were new or surprising variations of what was known. For example the paper develops an extended graph coloring algorithm to do automatic storage overlays and a framework for storage optimizing code generation. A renaming transformation is developed. Empirical results validate the theory. This is a wonderful paper, beautifully written, clear and insightful. It's nine, easy to read pages provide a nice snapshot of the state of compiler design in the late 1970's and a bold thesis for a new attack on an old problem being ignored by the compiler community. Janet Fabri's untimely death in an auto accident was a great loss in many ways. The field lost the founder of an interesting new research path in optimizing compilers.It is now 2002, and storage optimizing compiler techniques are needed more than ever. The problem isn't quite the same as it was in 1979. The motivation now is less about making good use of active storage than about overcoming the memory wall, especially in the context of concurrency, caches, and distribution. Automatic Storage Optimization is worth revisiting. 20