Year: 2011
Type: report
Abstract: The speed of software algorithms can be greatly improved by using a co-processor to offload computations from the main processor. Multiple co-processors can further increase the speed of a given algorithm. Based on this idea, three versions of an alpha blending algorithm were implemented on a NIOS II/f. The first implementation was entirely software based. This software based solution was then used as a baseline against which to test a single co-processor hardware solution and a multiple co-processor hardware solution. We showed that a single co-processor implementation achieved a speedup of 13.2 times, whereas the 2 co-processor solution achieved a speedup of 14.5 times with respect to this baseline. As further co-processors were added, the system became memory-bound as the algorithmic bottleneck moved from processing power to memory throughput.
Authors: Daniel Shapiro, Jonathan Parri, John-Marc Desmarais, Abdullah Kouri, Jean-Philippe Bergeron, Miodrag Bolić
Institution: University of Ottawa
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2123431586
