Determining the difficulty of accelerating problems on a GPU : research article
Dale Tristram, Karen Bradshaw · South African Computer Journal · 2014
General-purpose computation on graphics processing units (GPGPU) has great potential to accelerate many scientic models and algorithms. However, since some problems are considerably more dicult to accelerate than others, ascertaining the eort required to accelerate a particular problem is challenging. Through the acceleration of three typical scientic problems, seven problem attributes have been identied to assist in the evaluation of the diculty of accelerating a problem on a GPU. These attributes are inherent parallelism, branch divergence, problem size, required computational parallelism, memory access pattern regularity, data transfer overhead, and thread cooperation. Using these attributes as diculty indicators, an initial problem diculty classication framework has been created that aids in evaluating GPU acceleration diculty. The diculty estimates obtained by applying the classication framework to the three case studies correlate well with the actual eort expended in accelerating each problem.