Rapid prototyping of image processing workflows on massively parallel architectures

Bo Li, János Sallai, Péter Völgyesi, Ákos Lédeczi

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Many-core Graphics Processing Units (GPU) provide a high-performance parallel hardware platform on the desktop at an incredibly low cost. However, the widespread use of this computational capacity is hindered by the fact that programming GPUs is difficult. The state-of-the-art is to develop code utilizing the NVIDIA Compute Unified Device Architecture (CUDA). However, effective use of CUDA requires developers highly skilled in both low-level systems programming and parallel processing. Recognizing this roadblock to widespread adaption of General-Purpose Computing on GPUs (GPGPU), the NVIDIA Performance Primitives (NPP) library was released recently. While greatly easing the burden, utilizing NPP still requires one to learn CUDA. In this paper, we introduce a graphical environment for the design of image processing workflows that automatically generates all the CUDA code including NPP calls necessary to run the application on a GPU. Experimental results show that the generated code is almost as efficient as the equivalent hand written program and 10 times faster than running on the CPU alone in the typical case.

Original languageEnglish (US)
Title of host publicationWISES 2012 - Proceedings, Workshop on Intelligent Solutions in Embedded Systems
Pages15-20
Number of pages6
StatePublished - Oct 16 2012
Externally publishedYes
EventWorkshop on Intelligent Solutions in Embedded Systems, WISES 2012 - Klagenfurt, Austria
Duration: Jul 5 2012Jul 6 2012

Publication series

NameWISES 2012 - Proceedings, Workshop on Intelligent Solutions in Embedded Systems

Other

OtherWorkshop on Intelligent Solutions in Embedded Systems, WISES 2012
CountryAustria
CityKlagenfurt
Period7/5/127/6/12

Keywords

  • CUDA
  • GPU
  • NPP
  • code generation
  • image processing
  • model-based

ASJC Scopus subject areas

  • Artificial Intelligence
  • Control and Systems Engineering

Fingerprint Dive into the research topics of 'Rapid prototyping of image processing workflows on massively parallel architectures'. Together they form a unique fingerprint.

Cite this