Team Compsys

Overall Objectives
Scientific Foundations
New Results
Contracts and Grants with Industry
Other Grants and Activities

Section: New Results

Loop Transformations for High Level Synthesis

Participants : Alain Darte, Alexandru Plesco, Clément Quinson, Tanguy Risset.

We have started a study on the use of a loop transformation front-end to high level synthesis HLS tools. The study is based on the Wrapit loop transformation tools developed by the Alchemy team project and integrated into the orc open-source compiler. This tool allows the user to identify part of C programs which are loops with static control, to indicate many loop transformations (fusion, code motion, strip mining, etc.) and to generate back a C program where the loops have been transformed. The Wrapit tool has been applied to C code, synthesized by the Spark HLS tool, showing important performance improvements of the resulting design (in terms of silicon area and communication optimization). This work was done by Alexandru Plesco during his Master, it is currently under publication.

The topic of using compiling techniques as front-end optimizations to HLS tools is currently very hot. In parallel to this study, Clément Quinson has spent several months at stm icroelectronics (synthesis team in Crolles) to analyze what program transformations are required on source codes so that they can be accepted as input of HLS tools and achieve a good enough Quality of Result , i.e., good results both in terms of area and performance when compared to hand-written HDL. He worked with an industrial HLS tool on several applications from stm icroelectronics, that were so far not accepted or that required transformations to get acceptable results. These – still preliminary – experiments give quantitative and qualitative results from a leading edge commercial tool on real-life applications, which represents a valuable feedback for our research. These manually-applied transformations could in the near future be automated or, at least, semi-automated, guided by the user. The ``painful'' part that can be automated would then be done by a compiler or, more precisely, a high-level source-to-source transformer.


Logo Inria