Supervisor: Barbara Chapman (Stony Brook University, Brookhaven National Laboratory)
Abstract: More researchers and developers desire to port their applications to GPU-based clusters, due to their abundant parallelism and energy efficiency. Unfortunately porting or writing an application for accelerators, such as GPUs, requires extensive knowledge of the underlying architectures, the application/algorithm and the interfacing programming model (e.g. OpenMP). Often applications spend a significant portion of their execution time on data transfer. Exploiting data reuse opportunities in an application can reduce its overall execution time. In this research, we present an approach to automatically recognize data reuse opportunities in an application which uses OpenMP for exploiting GPU parallelism, and consequently insert pertinent code to take advantage of data reuse on GPU. Using our approach we were able to retain reused data on the GPU and reduce the overall execution time of multiple benchmark application.
ACM-SRC Semi-Finalist: yes
Poster Summary: PDF
Back to Poster Archive Listing