dataflow programming languages are used in a variety of settings, and defects in their programs can have serious consequences. However, prior work in automated program repair (APR) emphasizes control flow over dataflo...
详细信息
ISBN:
(纸本)9781665444668
dataflow programming languages are used in a variety of settings, and defects in their programs can have serious consequences. However, prior work in automated program repair (APR) emphasizes control flow over dataflowlanguages. We identify three impediments to the use of APR in dataflowprogramming-parallelism, state, and evaluation-and highlight opportunities for overcoming them.
The RVC-CAL language is used for implementing dataflow process networks (DPNs), i.e., distributed systems of actors. The behavior of an actor is defined by a set of actions which can consume input tokens and produce o...
详细信息
ISBN:
(纸本)9781728128610
The RVC-CAL language is used for implementing dataflow process networks (DPNs), i.e., distributed systems of actors. The behavior of an actor is defined by a set of actions which can consume input tokens and produce output tokens. RVC-CAL DPNs can offer parallelism both at the level of actors and at the level of actions. To efficiently execute these models on a target hardware, it is important to generate parallel code based on the entire parallelism provided by these two levels. In this paper, we discuss criteria for the generation of parallel software from RVC-CAL models based on the potential parallelism of modeled behaviors. The approach considers both the coarse-grained (task-parallel) execution of actors using multi-threading and the fine-grained (data-parallel) execution of their actions using the open computing language (OpenCL) or even a higher-level layer of OpenCL, namely SYCL. The methodology is validated by benchmarks on OpenCL abstracted hardware platforms. Based on the experimental results, the methodology is evaluated for efficiency (performance) in comparison with a pure multithreaded C++ approach and a well-known reference framework.
暂无评论