Profiling of Data-Parallel Processors
Institute of Computer Engineering, Ruprecht-Karls University of Heidelberg
Ruprecht-Karls University of Heidelberg, 2013
Profiling data can help to improve an application with respect to various objectives like execution time, energy consumption or even thermal sensor placement for an upcoming device. This survey reviews state-of-the-art profiling tools for dataparallel processors like Nsight, PAPI and TAU as well as Lynx. Additionally, the attained knowledge is utilized to detect the bottleneck of a reduction kernel for a CUDA-enabled device.
November 3, 2014 by hgpu
