Section: New Results

The XFOR programming structure

We have proposed a new programming control structure called “xfor” or “multifor”, providing users a way to schedule explicitly the statements of a loop nest, and take advantage of optimization and parallelization opportunities that are not easily attainable using the standard programming structures. This work is the PhD work of Imen Fassi, who started her work in 2013 and who is co-advised by Yosr Slama, Assistant Professor at the University El Manar in Tunis, Tunisia, and Philippe Clauss.

Data locality optimization is a well-known goal when handling programs that must run as fast as possible or use a minimum amount of energy. However, usual techniques never address the significant impact of numerous stalled processor cycles that may occur when consecutive load and store instructions are accessing the same memory location. In [15] , we show that two versions of the same program may exhibit similar memory performance, while performing very differently regarding their execution times because of the stalled processor cycles generated by many pipeline hazards. The xfor structure enables the explicit control of the way data locality is optimized in a program and thus, to control the amount of stalled processor cycles. In [15] , we also show the benefits of xfor regarding execution time and energy saving.

While many advanced and fully automatic program analysis and optimization techniques have been developed thanks to the accuracy and expressiveness of the polyhedral model, these techniques may fail in producing efficient codes in some circumstances. The xfor structure eases the manual application of optimizing transformations on loop nests for expert programmers and allows to generate executable codes that may be significantly faster than those generated automatically using well-established polyhedral strategies. we highlight five main gaps regarding these strategies and discuss some ideas on how to bridge them in [14] .