Automatic OpenCL Code Generation for Multi-device Heterogeneous Architectures

Li, Pei; Brunet, Élisabeth; Parrot, Christian; Thomas, Gaël; Namyst, Raymond

doi:10.1109/icpp.2015.105

Cited by 12 publications

(7 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We confirm the findings from Reference 4 that the MASON backend performed considerably better than the C with OpenMP backend when the agent number is high enough. Interestingly, the performance of the FPGA, which is restricted by the relatively low operating frequency and slow off-chip global memory access, is able to catch-up with OpenMP at 2 14 agents and even outperform it for the 2 16 agent scenario in both Game-of-Life and Sugarscape ( Figure 7A-C). This is caused by the efficient neighbor search and the pipelining parallelism the FPGA implements.…”

Section: Methodsmentioning

confidence: 99%

“…We demonstrate the benefit of the online dispatcher as well as the co-execution capability using three simulation models: Circle, Crowd, and Traffic. The former two were simulated using 2 16 agents, while the more complex traffic simulation was populated by 2 13 agents. The merge_functions for Crowd and Traffic are defined as: combining the results from the path finding model and the social force model for Crowd and taking the result of the car-following model for Traffic.…”

Section: Online Dispatchermentioning

confidence: 99%

“…To achieve domain‐independent code generation for heterogeneous platforms, methods such as pattern‐matching to detect parallelizable C snippets 15 and the use of code templates 16 have been proposed. Most of these approaches focus on detecting localized sections of the source code that can be parallelized, for example, nested loops with predictable control flows, whereas our work addresses the more irregular control flow of ABSs.…”

Section: Related Work and Backgroundmentioning

confidence: 99%

See 2 more Smart Citations

OpenABLext: An automatic code generation framework for agent‐based simulations on CPU‐GPU‐FPGA heterogeneous platforms

Xiao

Andelfinger

Cai

et al. 2020

Concurrency and Computation

View full text Add to dashboard Cite

The execution of agent-based simulations (ABSs) on hardware accelerator devices such as graphics processing units (GPUs) has been shown to offer great performance potentials. However, in heterogeneous hardware environments, it can become increasingly difficult to find viable partitions of the simulation and provide implementations for different hardware devices. To automate this process, we present OpenABLext, an extension to OpenABL, a model specification language for ABSs. By providing a device-aware OpenCL backend, OpenABLext enables the co-execution of ABS on heterogeneous hardware platforms consisting of central processing units, GPUs, and field programmable gate arrays (FPGAs). We present a novel online dispatching method that efficiently profiles partitions of the simulation during run-time to optimize the hardware assignment while using the profiling results to advance the simulation itself. In addition, OpenABLext features automated conflict resolution based on user-specified rules, supports graph-based simulation spaces, and utilizes an efficient neighbor search algorithm. We show the improved performance of OpenABLext and demonstrate the potential of FPGAs in the context of ABS. We illustrate how co-execution can be used to further lower execution times. OpenABLext can be seen as an enabler to tap the computing power of heterogeneous hardware platforms for ABS.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Online Dispatchermentioning

confidence: 99%

Section: Related Work and Backgroundmentioning

confidence: 99%

See 1 more Smart Citation

OpenABLext: An automatic code generation framework for agent‐based simulations on CPU‐GPU‐FPGA heterogeneous platforms

Xiao

Andelfinger

Cai

et al. 2020

Concurrency and Computation

View full text Add to dashboard Cite

show abstract

“…Meta Functions Li et al [11] as well as Diop et al [4] employ meta functions to determine data splits independently of the problem size. These meta functions have to be supplied by the programmer, describing the memory access patterns of the employed algorithms.…”

Section: Job Designmentioning

confidence: 99%

CloudCL: Single-Paradigm Distributed Heterogeneous Computing for Cloud Infrastructures

Plauth

Rösler

Polze

2018

IJNC

View full text Add to dashboard Cite

The ever-growing demand for compute resources has reached a wide range of application domains, and with that has created a larger audience for compute-intensive tasks. In this paper, we present the CloudCL framework, which empowers users to run compute-intensive tasks without having to face the total cost of ownership of operating an extensive high-performance compute infrastructure. CloudCL enables developers to tap the ubiquitous availability of cloudbased heterogeneous resources using a single-paradigm compute framework, without having to consider dynamic resource management and inter-node communication. In an extensive performance evaluation, we demonstrate the feasibility of the framework, yielding close-to-linear scale-out capabilities for certain workloads. International Journal of Networking and ComputingHere, we present the CloudCL 1 framework, which empowers users to run compute-intensive tasks on demand without having to face the total cost of ownership of operating an extensive highperformance compute infrastructure. The presented framework hides the complexity of distributed programming for dynamic cluster configurations, enabling developers to tap the ubiquitous availability of cloud-based heterogeneous resources on demand using a single-paradigm compute framework. At the same time, the single paradigm approach helps in improving productivity, as developers and domain experts may focus their implementation efforts on application logic without having to consider inter-node communication and dynamic resource management. As illustrated in Figure 1, CloudCL combines the capabilities of dOpenCL [7] and Aparapi [10] and augments these enabling technologies with a job scheduling system that can handle the dynamic properties of cloud-based resource-provisioning by supporting dynamic addition and removal of resources at runtime. In an extensive performance evaluation, we demonstrate that the framework provides close-to-linear scaleout capabilities both in an on-premise hosting environment, using Amazon EC2-based public cloud resources as well as hybrid configurations.This paper is structured as follows: Section 2 provides background about MPI, Opencl, dOpenCL and Aparapi for two purposes. On one hand, the goal is to provide a quick walk-through of the enabling technologies behind CloudCL. On the other hand, re-iterating over the basics of both distributed as well as heterogeneous computing should remind readers about the many layers of complexity that are concealed by the CloudCL framework. Subsequently, Section 3 reviews related OpenCL API forwarding approaches and compares them with dOpenCL -the library of our choice. Section 4 proceeds with elucidating the major components making up CloudCL. Afterwards, Section 5 provides an extensive performance evaluation of the scale-out behavior of CloudCL, before a conclusion is reached in Section 6.

show abstract

“…The technique we propose does not require training runs. Li et al [9] present STEPOCL, a tool which takes as input kernels along with a configuration file and generates automatically an OpenCL multi-devices application. The configuration file describes how to split data, the control flow of the program, and allow to have specialized kernels for different architectures.…”

Section: Related Workmentioning

confidence: 99%

Automatic OpenCL Task Adaptation for Heterogeneous Architectures

Huchant

Counilh

Barthou

2016

Euro-Par 2016: Parallel Processing

View full text Add to dashboard Cite

OpenCL defines a common parallel programming language for all devices, although writing tasks adapted to the devices, managing communication and load-balancing issues are left to the programmer. In this work, we propose a novel automatic compiler and runtime technique to execute single OpenCL kernels on heterogeneous multi-device architectures. The technique proposed is completely transparent to the user, does not require off-line training or a performance model. It handles communications and load-balancing issues, resulting from hardware heterogeneity, load imbalance within the kernel itself and load variations between repeated executions of the kernel, in an iterative computation. We present our results on benchmarks and on an N-body application over two platforms, a 12-core CPU with two different GPUs and a 16core CPU with three homogeneous GPUs.

show abstract

Automatic OpenCL Code Generation for Multi-device Heterogeneous Architectures

Cited by 12 publications

References 10 publications

OpenABLext: An automatic code generation framework for agent‐based simulations on CPU‐GPU‐FPGA heterogeneous platforms

OpenABLext: An automatic code generation framework for agent‐based simulations on CPU‐GPU‐FPGA heterogeneous platforms

CloudCL: Single-Paradigm Distributed Heterogeneous Computing for Cloud Infrastructures

Automatic OpenCL Task Adaptation for Heterogeneous Architectures

Contact Info

Product

Resources

About