Title :
Performance analysis between aparapi (a parallel API) and JAVA by implementing sobel edge detection Algorithm
Author :
Gupta, Krishan Gopal ; Agrawal, Nidhi ; Maity, Suman Kalyan
Author_Institution :
HPC Frontier Technol., Exploration Group, C-DAC, Pune Univ., Pune, India
Abstract :
This paper presents performance comparison between aparapi (a parallel API for GPU) and java by implementing sobel edge detection Algorithm in java (run on CPU) and aparapi (run on GPU). Our GPU implementation using Aparapi shows speedup of 6x against CPU implementation using java (serial implementation) and speedup of 2x using java prallel implementation (less than 8 threads). Experiments indicate that java threaded version shows speedup up to 4X against Aparapi implementation (more than 8 threads). This comparison study also include implementation of sobel edge detection algorithm on CPU (sequential, threaded version) and aparapi version for enabled on GPU. This article also discusses how to implement Aparapi kernels for data-parallel operations of Typical Edge detection algorithms based on Sobel operator within Java applications The results for performance gains that can be achieved using with and without Aparapi framework.
Keywords :
Java; application program interfaces; edge detection; parallel processing; Aparapi; Aparapi kernels; CPU implementation; GPU implementation; Java prallel implementation; Java threaded version; Sobel edge detection algorithm; Sobel operator; a parallel API; data-parallel operations; performance analysis; serial implementation; Approximation methods; Graphics processing units; Hardware; Image edge detection; Image resolution; Java; Kernel; Aparapi; Edge Detection; Gpu; java;
Conference_Titel :
Parallel Computing Technologies (PARCOMPTECH), 2013 National Conference on
Conference_Location :
Bangalore
Print_ISBN :
978-1-4799-1589-7
DOI :
10.1109/ParCompTech.2013.6621408