From cTuning.org

Jump to: navigation, search
Navigation: cTuning.org > CTools > MilepostGCC

Here we try to answer some frequently asked questions:

Does MILEPOST GCC solve all optimization problems?

Answer: NO!

It's just the first practical attempt to move previously mostly theoretical research techniques to production compilers and show that it is possible to automate program, compiler and architecture design and optimization using feedback directed iterative compilation, statistical collective optimization and machine learning in real production environments. However, many things are still unsolved or unclear and we hope that using such common open-source production infrastructure opens up many more interested collaborative R&D opportunities in this area. cTuning community continues parametrization of all parts of computing systems (compiler fine-grain optimizations, run-time adaptation, program data/code partitioning, parallelization, virtualization, OS control, etc) or understanding how to better characterize programs and architectures (features, behavior) or how to collect and reuse huge amount of optimization information.

Is MILEPOST GCC similar to ACOVEA tool?

Answer: MILEPOST GCC is orthogonal to ACOVEA.

ACOVEA is a useful tool to find good optimization flags for a given program, dataset and architecture using genetic algorithms. MILEPOST GCC learns optimizations ACROSS programs and architectures by correlating program features and program optimizations so that it can PREDICT profitable optimizations for an UNSEEN program. ACOVEA has some similarities with CCC framework and can be used to build an optimization training set for MILEPOST GCC. In fact, we have been discussing a possibility to implement ACOVEA plugin inside CCC framework one day ...

Locations of visitors to this page

Tweet