Acceptance of an object-oriented approach to scientific computing is being hindered by the performance differential between programs written in traditional languages and those written in object-oriented languages. These performance issues stem from the limitations on static dependence analysis and compiler optimization posed by the presence of objects. An approach to increasing the degree of optimization in object-oriented environments is incorporating information about the object semantics and performing runtime optimizations. As proved true for compiler optimization using static dependence information,informal semantics and intuition are inadequate. A semantic foundation for optimization which includes support for dynamic dependence and for the richer data types exploited in object-oriented languages provides an important foundation for advancing compiler research. The semantic model studied in this research is intended to unify reasoning both in the compiler (static) environment and the runtime (dynamic)environment. Support for dynamic object creation and even dynamic class definition can be considered in a setting exploiting both static and dynamic dependence information. The resulting model, utilizing tools such as denotational semantics and equational calculi, will allow the creation of automated tools to assist in the development of advanced compiler optimizations with the additional formal assurances that these optimizations do not alter the meaning of the program. The computational demands posed by computational biology provide another realistic setting within which to study the complexities of object-oriented parallel and distributed program development and the concomitant optimization needs for language systems. The educational goals of this research are to both alter the programming languages and compiler curricula to address, at all levels, issues of semantics, optimization, and the degree to which each affect the other. The computational biology work provides a promising framework within which to study these issues. Thus, the computational biology curricula is being developed in parallel with the languages work. This research is expected to contribute to an understanding of the nature of program dependences; how different language constructs and program usage contribute to different patterns of dependence; and how these patterns can be exploited in language design or in language translation. Work in this area has already uncovered unsuspected connections between lazy evaluation and imperative languages. Further study should provide insight into relationships among languages in the context of their interactions at runtime. A goal of this research is to support the mixing of languages without sacrificing the performance of the program, allowing the programmer to select the most appropriate language for the individual parts of a problem. The fundamental notion of program dependence is a natural basis on which to construct a system supporting this style of programming. ***
