Profile-guided optimization across process boundaries

Abstract

We describe a profile-driven compiler optimization technique for inter-process optimization, which dynamically inlines the effects of sending messages. Profiling is used to find optimization opportunities, and to dynamically trigger recompilation and optimization at run-time. We apply the optimization technique on the concurrent programming language ERLANG, letting recompilation take place in a separate ERLANG process, and taking advantage of the facilities provided by ERLANG to dynamically replace code at run-time. We have implemented a prototype inter-process profiler and optimizer, that can handle small programs. Measurements on synthetic benchmarks show encouraging speedups of up to 1.8 times.