Heterogeneous Scheduling and Dynamic Parallelism Support in Accelerator Directives

Speaker: Guray Ozen (BSC)
Date: Thursday, 3 December 2015, 10:30-12:00
Session: Programming Models II
Talk type: Short talk (15 min)

Abstract: In this talk, we present our extensions to directive based programming models and their implementation onto MACC compiler with OmpSs runtime. Current directive–based paradigms, such as OpenMP or OpenACC, support both accelerators and multicore-based hosts, they do not provide an effective and efficient way to concurrently use them, usually resulting in accelerated programs in which the potential performance of the host is not exploited. We propose an extension to these directive-based programming models to support multiple-target task sharing onto heterogeneous systems, i.e. the possibility of sharing the execution (of multiple instances) of a task on different devices. Also, we propose how to extend the current OpenMP accelerator model to make the use of DP easy and effective. Because, developing applications able to exploit their dazzling performance is not a trivial task, which becomes even harder when they have irregular data access patterns or control flows. Dynamic Parallelism (DP) has been introduced in the most recent GPU architecture as a mechanism to improve applicability of GPU computing in these situations, resource utilization and execution performance.