A Framework for Efficient Execution of Data Parallel Irregular Applications on Heterogeneous Systems

Abstract

Exploiting the computing power of the diversity of resources available on heterogeneous systems is mandatory but a very challenging task. The diversity of architectures, execution models and programming tools, together with disjoint address spaces and different computing capabilities, raise a number of challenges that severely impact on application performance and programming productivity. This problem is further compounded in the presence of data parallel irregular applications. This paper presents a framework that addresses development and execution of data parallel irregular applications in heterogeneous systems. A unified task-based programming and execution model is proposed, together with inter and intra-device scheduling, which, coupled with a data management system, aim to achieve performance scalability across multiple devices, while maintaining high programming productivity. Intra-device scheduling on wide SIMD/SIMT architectures resorts to consumer-producer kernels, which, by allowing dynamic generation and rescheduling of new work units, enable balancing irregular workloads and increase resource utilization. Results show that regular and irregular applications scale well with the number of devices, while requiring minimal programming effort. Consumer-producer kernels are able to sustain significant performance gains as long as the workload per basic work unit is enough to compensate overheads associated with intra-device scheduling. This not being the case, consumer kernels can still be used for the irregular application. Comparisons with an alternative framework, StarPU, which targets regular workloads, consistently demonstrate significant speedups. This is, to the best of our knowledge, the first published integrated approach that successfully handles irregular workloads over heterogeneous systems.

LOOM: Interweaving tightly coupled visualization and numeric simulation framework

“…Barbosa et al [5,18] proposed allowing the developer to define the partition method for the data and rely on a performance model and scheduler to dice the tasks to be decomposed into smaller ones when needed. This mechanism can be leveraged to enable the interweaving of simulation and visualization tasks, primarily because they operate in different time intervals, t and t − 1, respectively.…”

“…The solution gives rise to a granularity of τ v , i.e., if the workload from τ v (W (τ v )) is too large, it will delay the execution of subsequently scheduled τ s , on the other hand, if W (τ v ) is too small it will increase the scheduling overhead. To address this problem, we follow the approach proposed by Barbosa et al [5,18] referred to as dicing. The dicing strategy enables the developer to define a generic workload applied to a partition created by the scheduler using a developer-defined partitioning method called dice.…”

Exploring heterogeneous computing with advanced path tracing algorithms

“…Furthermore, different architectures usually exhibit different execution and programming models and are deployed with different programming languages and development tools, severely impacting on both code and performance portability. Additionally, the application's workload has to be distributed and balanced among the multiple devices, and, within each device, among its multiple computing units; this leads to multilevel scheduling, which must be effectively handled in order to achieve acceptable performance levels [1]. To efficiently use the available resources in these and future systems, algorithms and software packages have to be revisited and re-evaluated to assess their adequateness to these environments.…”