The recent abundance of interactive programming environments is significantly aiding the programmer of large software systems. Response times are enhanced by decreasing unnecessary recompilation, usually at the expense of sophisticated program optimization. This project addresses the issue of providing an effective interactive environment for developing highly optimized programs. Two principal issues are to be investigated. (1) development of a recompilation system that, in general, provides substantially better interactivity than existing environments for highly optimized code but supports the same level of optimization; (2) assessment of the most efficient recompilation strategies under various scenarios of edits, program structure, and code optimization. This project focuses on adapting the Rn Fortran programming environment at Rice University to provide fine-grained partial recompilation without sacrificing the ambitious (global and interprocedural) program optimization inherent of this environment. The approach consists of applying the incremental, intraprocedural optimization techniques recently developed by the PI to the large-grained partial recompilation system of Rn . The performance of the fine-grained, partial recompilation system will be compared with the existing larger-grained compiler of Rn under different condition. Ideally, the recompilation system would be able to use this information to automatically determine when it is more efficient to perform fine-grained recompilation analysis or a total recompilation. A foreseen contribution of this research is a usable programming environment which offers precise recompilation analyses coupled with the support of highly optimized programs.
