UNIVERSITY OF HERTFORDSHIRE
		COMPUTER SCIENCE RESEARCH COLLOQUIUM

			     presents


                 "You have a high-performance computer too --- 
        On the utilisation of graphics cards for parallel computing"

		       Dr. Jeyan Thiyagalingam 
	    (University of Hertfordshire & Oxford University)


		      1 April 2009 (Wednesday)

			 Lecture Theatre B404 
		   Hatfield, College Lane Campus
			      3 - 4 pm

	    Coffee/tea and biscuits will be available.

		   Everyone is Welcome to Attend



Abstract:

The newest generations of graphics processing unit (GPU) architecture, 
such as Nvidia's Compute Unified Device Architecture (CUDA), provide 
massive multi-threading capabilities for high performance computing.  
While being affordable, they feature multiple cores, high-bandwidth bus 
and  simplified  parallel programming model. Applications  written for 
CUDA devices are  already achieving several folds of speedups compared to 
desktop systems.

However, to achieve a notable performance gain,  programs have to be 
optimised paying attention to the details of the underlying CUDA 
architecture. Since these architectures evolve rather quickly, 
learning and thus mastering these architectural aspects are a real burden
on developers.  We argue that this process should be automated  
to minimise these demands on developers and to simplify the programming.   

In this talk, we present our  efforts in automating this process using 
a compiler framework. In particular, we focus on  mapping  one of the 
array-programming languages, Single Assignment C (SaC), to CUDA.  We 
present the overview of the architectural aspects of CUDA, our 
methodology for automating the process and our initial results so far.

---------------------------------------------------
Hertfordshire Computer Science Research Colloquium
http://homepages.feis.herts.ac.uk/~nehaniv/colloq