We develop a simple model of hierarchical memories and we use it to determine an optimal strategy for blocking matrices. This model predicts the form of current, state-of-the-art L1 kernels. Additionally, it shows that current L1 kernels can continue to produce their high performance on operand matrices that are as large as the L2 cache. For a hierarchical memory with L memory levels (main memory and L-1 caches), our model reduces the number of potential matrix multiply algorithms from 6L to four. We use the shape of the matrix input operands to select one of our four algorithms. Because of space limitations, we do not include performance results.
By: John A. Gunnels; Fred G. Gustavson; Greg M. Henry; Robert A. van de Geijn
Published in: RC23443 in 2004
LIMITED DISTRIBUTION NOTICE:
This Research Report is available. This report has been submitted for publication outside of IBM and will probably be copyrighted if accepted for publication. It has been issued as a Research Report for early dissemination of its contents. In view of the transfer of copyright to the outside publisher, its distribution outside of IBM prior to publication should be limited to peer communications and specific requests. After outside publication, requests should be filled only by reprints or legally obtained copies of the article (e.g., payment of royalties). I have read and understand this notice and am a member of the scientific community outside or inside of IBM seeking a single copy only.
Questions about this service can be mailed to reports@us.ibm.com .