INCREMENTAL GLOBAL REOPTIMIZATION OF PROGRAMS

Although optimizing compilers have been quite successful in producing excellent code, two factors that limit their usefulness are the accomp anying long compilation times and the lack of good symbolic debuggers for optimized code. One approach to attaining faster recompilations is to reduce the redundant analysis that is performed for optimization i n response to edits, and in particular, small maintenance changes, wit hout affecting the quality of the generated code. Although modular pro gramming with separate compilation aids in eliminating unnecessary rec ompilation and reoptimization, recent studies have discovered that mor e efficient code can be generated by collapsing a modular program thro ugh procedure inlining. To avoid having to reoptimize the resultant la rge procedures, this paper presents techniques for incrementally incor porating changes into globally optimized code. An algorithm is given f or determining which optimizations are no longer safe after a program change, and for discovering which new optimizations can be performed i n order to maintain a high level of optimization. An intermediate repr esentation is incrementally updated to reflect the current optimizatio ns in the program. Analysis is performed in response to changes rather than in preparation for possible changes, so analysis is not wasted i f an edit has no far-reaching effects. The techniques developed in thi s paper have also been exploited to improve on the current techniques for symbolic debugging of optimized code.