Techniques for compiler-directed cache coherence

The concepts underlying compiler-directed cache coherence are explained, and various approaches to this strategy are surveyed. Compiler-directed coherence emphasizes the interaction between architectural features and compiler optimization, which affects a variety of system and compiler designers. Several techniques for compiler-directed cache coherence, such as CMU C.mmp and two-phase invalidation (TPI), were implemented on existing microprocessors without hardware support. Overall, the performance of the `off-chip' implementation is comparable to both `on-chip' TPI and full-map hardware directory (HW), suggesting that off-chip TPI using off-the-shelf microprocessors is a practical solution.