Projects in the topic performance

Moses is a statistical machine translation system that allows you to automatically train translation models for any language pair. Intention of the project is to tune up existing software, where a glimpse shows that majority of time is consumed by memory allocation, dynamic casting and other calculation non-related stuff. I would like to inspect many techniques (like perf profiling, GCC LTO, GCC profile-guided optimization, code refactoring, OpenTuner, etc.) which may bring really significant performance gain. Moreover, it would be really beneficial to come up with a cookbook that can be used by folk in general. If possible, I would like to create a step-by-step performance improvement graphs.