Executing Dynamic Data Rate Actor Networks on OpenCL Platforms

Abstract

Heterogeneous computing platforms consisting of general purpose processors (GPPs) and graphics processing units (GPUs) have become commonplace in personal mobile devices and embedded systems. For years, programming of these platforms was very tedious and simultaneous use of all available GPP and GPU resources required low-level programming to ensure efficient synchronization and data transfer between processors. However, in the last few years several high-level programming frameworks have emerged, which enable programmers to describe applications by means of abstractions such as dataflow or Kahn process networks and leave parallel execution, data transfer and synchronization to be handled by the framework.Unfortunately, even the most advanced high-level programming frameworks have had shortcomings that limit their applicability to certain classes of applications. This paper presents a new, dataflow-flavored programming framework targeting heterogeneous platforms, and differs from previous approaches by allowing GPUmapped actors to have data dependent consumption of inputs / production of outputs. Such flexibility is essential for configurable and adaptive applications that are becoming increasingly common in signal processing. In our experiments it is shown that this feature allows up to 5× increase in application throughput.The proposed framework is validated by application examples from the video processing and wireless communications domains. In the experiments the framework is compared to a well-known reference framework and it is shown that the proposed framework enables both a higher degree of flexibility and better throughput.

PRUNE: Dynamic and Decidable Dataflow for Signal Processing on Heterogeneous Platforms

“…The frame size used was 1920×1080, which resulted in the token size becoming 1.98 megabytes. Due to the large token size, the token rate was kept at 1 (in our previous publication [6] that uses a preliminary version of PRUNE, resolution was 320x240 with a token rate of 4). GPU acceleration was applied to Motion Detection by mapping the Gauss, Thres and Med actors to the GPU.…”

“…Finally, compilation with the target-specific C compiler requires the PRUNE run-time library, which contains application independent actor wrappers, FIFO implementations and OpenCL support. The following detailed description of the PRUNE runtime framework contains some extension compared to our preliminary work [6], where it was first presented. For example, Equation 2 has been generalized to support token delays > 1.…”

“…The MoC presented in this article has been published [6] recently, and in this article the MoC is complemented with design rules that enable decidability analysis. Based on the MoC, the article describes a novel Linux-arXiv:1802.06625v1 [cs.DC] 19 Feb 2018…”

“…PRUNE 1) implements application consistency analysis, 2) provides an efficient runtime memory-and concurrency management framework for heterogeneous platforms, 3) presents a compile-time translator that allows importing programs from previous similar run-time frameworks. Out of these, items 1) and 3) are novel compared to [6].…”

Digital Predistortion for 5G Small Cell: GPU Implementation and RF Measurements

“…The parallel processing offered by the multicore CPU allows executing DPD coefficient learning concurrently with filtering. The software implementation is based on a dataflow programming environment [6] that takes care of data transfer and synchronization between the CPU cores and the GPU. The GPU code is written in OpenCL for cross-platform portability, whereas the DPD functionalities executed on the CPU cores are written in C.…”