Lazy task creation: a technique for increasing the granularity of parallel programs
LFP '90 Proceedings of the 1990 ACM conference on LISP and functional programming
Introduction to the cell multiprocessor
IBM Journal of Research and Development - POWER5 and packaging
CellSs: a programming model for the cell BE architecture
Proceedings of the 2006 ACM/IEEE conference on Supercomputing
Carbon: architectural support for fine-grained parallelism on chip multiprocessors
Proceedings of the 34th annual international symposium on Computer architecture
Introduction to the cell broadband engine architecture
IBM Journal of Research and Development
CellSs: making it easier to program the cell broadband engine processor
IBM Journal of Research and Development
An adaptive cut-off for task parallelism
Proceedings of the 2008 ACM/IEEE conference on Supercomputing
Available task-level parallelism on the Cell BE
Scientific Programming - High Performance Computing with the Cell Broadband Engine
Numerical Simulation in Molecular Dynamics: Numerics, Algorithms, Parallelization, Applications
Numerical Simulation in Molecular Dynamics: Numerics, Algorithms, Parallelization, Applications
Dynamic Task Scheduling and Load Balancing on Cell Processors
PDP '10 Proceedings of the 2010 18th Euromicro Conference on Parallel, Distributed and Network-based Processing
| Hi-index | 0.00 |
Driven by increasing specialization, multicore integration will soon enable large-scale chip multiprocessors (CMPs) with many processing cores. In order to take advantage of increasingly parallel hardware, independent tasks must be expressed at a fine level of granularity to maximize the available parallelism and thus potential speedup. However, the efficiency of this approach depends on the runtime system, which is responsible for managing and distributing the tasks. In this paper, we present a hierarchically distributed task pool for task parallel programming on Cell processors. By storing subsets of the task pool in the local memories of the Synergistic Processing Elements (SPEs), access latency and thus overheads are greatly reduced. Our experiments show that only a worker-centric runtime system that utilizes the SPEs for both task creation and execution is suitable for exploiting fine-grained parallelism.