General Purpose Task-Dependence Management Hardware for Task-Based Dataflow Programming Models
Authors:
- Xubin TanXTXubin Tan
- Jaume BoschJBJaume Bosch
- Miquel VidalMVMiquel Vidal
- Carlos AlvarezCACarlos Alvarez
- Daniel Jimenez-GonzalezDJDaniel Jimenez-Gonzalez
- Eduard AyguadeEAEduard Ayguade
- Mateo ValeroMVMateo Valero
- 1 May 2017
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 244-253
Abstract
Task-based programming models such as OpenMP, IntelTBB and OmpSs offer the possibility of expressing dependences among tasks to drive their execution at runtime. Managing these dependences introduces noticeable overheads when targeting fine-grained tasks, diminishing the potential speedups or even introducing performance losses. To overcome this drawback, we present a general purpose hardware accelerator, Picos++, to manage the inter-task dependences efficiently in both time and energy. Our design also includes a novel nested task support. To this end, a new hardware/software co-design is presented to overcome the fact that nested tasks with dependences could result in system deadlocks due to the limited amount of resources in hardware task dependence managers. In this paper we describe a detailed implementation of this design and evaluate a parallel task-based programming model using Picos++ in a Linux embedded system with two ARM Cortex-A9 and a FPGA. The scalability and energy consumption of the real system implemented have been studied and compared against a software runtime. Even in a system limited to 2 threads, using Picos++ results in more than 1.8x speedup and 40% of energy savings in the most demanding parallelizations of real benchmarks. As a matter of fact, a hardware task dependence manager should be able to achieve much higher speedup and provide more energy savings with more threads.Keywords
This publication has 11 references indexed in Scilit:
- Picos: A hardware runtime architecture support for OmpSsFuture Generation Computer Systems, 2015
- Nexus#: A Distributed Hardware Task Manager for Task-Based Programming Models2015
- An Integrated Hardware-Software Approach to Task Graph Management2014
- Microarchitecture of a Coarse-Grain Out-of-Order Superscalar ProcessorIEEE Transactions on Parallel and Distributed Systems, 2012
- A Multithreaded Multicore System for Embedded Media ProcessingLecture Notes in Computer Science, 2011
- Task Superscalar: An Out-of-Order Task Pipeline2010
- Flexible architectural support for fine-grain scheduling2010
- A Hardware Task Scheduler for Embedded Video Processing2009
- Characterizing and improving the performance of Intel Threading Building Blocks2008
- Carbon2007