Major changes in processor architecture over the last decade have created a demand for new compiler optimization technologies. Optimizing compilers have risen to this challenge by steadily increasing the uniprocessor performance gap between optimized compiled and unoptimized compiled code to a level that already exceeds the performance gap between two successive generations of processor hardware. These traditional optimizations [2] have been developed in the context of sequential programs -- the assumption of sequential control flow is intrinsic to the definition of dasic optimization data structures such as the control flow graph (CFG), and pervades all the optimization algorithms.
展开▼
