Using ArrayOL to Identify Potentially Shareable Data in Thread Work-Groups of GPUs

International audienceOver recent years, using Graphics Processing Units (GPUs) has become as an effective method for increasing the performance of many applications. However, these performance benefits from GPUs come at a price. First, extensive programming expertise and intimate knowledge of the underlying hardware are essential for gaining good speedups. Second, the expressibility of GPU-based programs are not powerful enough to retain the high-level abstractions of the solutions. Although the programming experience has been significantly improved by existing frameworks like CUDA and OpenCL, it is still a challenge to effectively utilise these devices while still retaining the programming abstractions. To this end, performing a model-to-source transformation, whereby a highlevel language is mapped to CUDA or OpenCL, is an attractive option. In particular, it enables to harness the power of GPUs without any expertise on the GPGPU programming. In this work, we purpose an approach based on MDE and ArrayOL to detect shareable data zone. The tilers from ArrayOL, which allow express the data parallelism from repetitive tasks, are analyzed in time compilation to create areas of shared data. The identification of these areas is crucial to allow us loading data on shared areas of memory that have high throughput. Consequently, programs automatically generated shall have performances comparable to manually well written programs