Change search
ReferencesLink to record
Permanent link

Direct link
Auto-tunable GPU BLAS
Norwegian University of Science and Technology, Faculty of Information Technology, Mathematics and Electrical Engineering, Department of Computer and Information Science.
2012 (English)MasteroppgaveStudent thesis
Abstract [en]

In this paper, we present our implementation of an Auto tuning system, written in C++, which incorporate the use of OpenCL kernels. We deploy this approach on different GPU architectures, evaluating the performance of the approach. Our main focus is to easily generate tuned code, that would otherwise require a large amount of empirical testing, and then run it on any kind of device. This is achieved through the auto tuning framework, which will create different kernels, compile and run them on the device and output the best performing kernel on the given platform. BLAS is much used in performance critical applications, and is a good candidate for execution on GPUs due to its potential performance increase. Our implementation was benchmarked on various of test environments, with different GPUs, where we achieved comparable results to the ViennaCL library. We also tested against the native vendor specific BLAS libraries from AMD and NVIDIA.

Place, publisher, year, edition, pages
Institutt for datateknikk og informasjonsvitenskap , 2012. , 63 p.
Keyword [no]
ntnudaim:5976, MIT informatikk, Komplekse datasystemer
URN: urn:nbn:no:ntnu:diva-18411Local ID: ntnudaim:5976OAI: diva2:565905
Available from: 2012-11-08 Created: 2012-11-08

Open Access in DiVA

fulltext(1244 kB)352 downloads
File information
File name FULLTEXT01.pdfFile size 1244 kBChecksum SHA-512
Type fulltextMimetype application/pdf
cover(184 kB)25 downloads
File information
File name COVER01.pdfFile size 184 kBChecksum SHA-512
Type coverMimetype application/pdf
attachment(12539 kB)1362 downloads
File information
File name ATTACHMENT01.zipFile size 12539 kBChecksum SHA-512
Type attachmentMimetype application/zip

By organisation
Department of Computer and Information Science

Search outside of DiVA

GoogleGoogle Scholar
Total: 352 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

Total: 28 hits
ReferencesLink to record
Permanent link

Direct link