Section: New Results

Adaptive Partitioning for Iterated Sequences of Irregular OpenCL Kernels

OpenCL defines a common parallel programming language for CPU and GPU devices, although writing tasks adapted to the devices, managing communication and load-balancing issues are left to the programmer. We propose [10] a static/dynamic approach for the execution of an iterated sequence of data-dependent kernels on a multi-device heterogeneous architecture. The method allows to automatically distribute irregular kernels onto multiple devices and tackles, without training, both load balancing and data transfers issues coming from hardware heterogeneity, load imbalance within the application itself and load variations between repeated executions of the sequence. Our evaluation on some benchmarks and a complex N-body application, SOTL, simulating the electromagnetic Coulomb force applied on particles, show the interest of our approach.