next up previous
Next: Future Work Up: Automatic Inline Allocation of Previous: Related Work

Conclusions

 

We presented object inlining, an optimization that automatically inline allocates objects, allowing us obtain similar efficiency to a system with explicit inline allocation while preserving the programmability benefits of a uniform object model. To do this requires a novel analysis and transformation mechanism that builds upon previous work on analysis and specialization of object-oriented programs. Our evaluation shows that our analysis can handle not only situations where inline allocation is done in C++, but also situations where such manual allocation would be conceptually disruptive. We showed several object-oriented benchmarks running up three times as fast due to object inlining as opposed to without it, and matching the performance of code with inlining specified by hand.



Julian Dolby
dolby@cs.uiuc.edu