A generic parallel pattern interface for stream and data processing

摘要

Current parallel programming frameworks aid developers to a great extent in implementingrnapplications that exploit parallel hardware resources. Nevertheless, developers require additionalrnexpertise to properly use and tune them to operate efficiently on specific parallel platforms.rnOn the other hand, porting applications between different parallel programming modelsrnand platforms is not straightforward and demands considerable efforts and specific knowledge.rnApart from that, the lack of high-level parallel pattern abstractions, in those frameworks, furtherrnincreases the complexity in developing parallel applications. To pave the way in this direction, thisrnpaper proposes GRPPI, a generic and reusable parallel pattern interface for both stream processingrnand data-intensive C++ applications. GRPPI accommodates a layer between developers andrnexisting parallel programming frameworks targetingmulti-core processors, such as C++ threads,rnOpenMP and Intel TBB, and accelerators, as CUDA Thrust. Furthermore, thanks to its high-levelrnC++ application programming interface and pattern composability features, GRPPI allows usersrnto easily expose parallelism via standalone patterns or patterns compositions matching in sequentialrnapplications.We evaluate this interface using an image processing use case and demonstraternits benefits from the usability, flexibility, and performance points of view. Furthermore, we analyzernthe impact of using stream anddata pattern compositions on CPUs,GPUsand heterogeneousrnconfigurations.