SPECO: Software Platform to Evaluate the impact of Compilation Optimizations
Participants : Maha Idrissi Aouad, Olivier Zendra.
This prototype platform aimed at automatically running a large number of benchmarks and thus get precise concrete results over the actual impact of any compilation optimization we design in our research. The first set of benchmarks, to experiment with the prototype platform, had been chosen among the most relevant according to the bibliography, and mostly comprise multimedia and embedded systems related programs. The platform worked as follows. It extracted static metrics to provide information about the structure and static complexity of the benchmarks. It compiled all the benchmarks, and ran them to extract dynamic metrics helping characterize the behavior of programs at run-time. Those dynamic metrics comprised time, for raw performance and space (memory) information to better understand the memory behavior of the programs. More recently, some work was realized to integrate the energy usage metric to SPECO. This metric is indeed an important part for our current research work, for example in the ANR MORE project, where SPECO used to be intended to be used to characterize programs and obtain some of the metrics needed in MORE for its iterative compilation framework. However, since the development of SPECO was not progressing as forecast, SPECO was put on hold this year to leave place to another, more thoroughly thought of and more mature development that could benefit from this experience (see ANR MORE platform).