Auto-tuning the Matrix Powers Kernel with SEJITS
Jeffrey Morlan
EECS Department
University of California, Berkeley
Technical Report No. UCB/EECS-2012-95
May 11, 2012
http://www.eecs.berkeley.edu/Pubs/TechRpts/2012/EECS-2012-95.pdf
The matrix powers kernel, used in communication-avoiding Krylov subspace methods, requires runtime auto-tuning for best performance. We demonstrate how the SEJITS (Selective Embedded Just-In-Time Specialization) approach can be used to deliver a high-performance and performance-portable implementation of the matrix powers kernel to application authors, while separating their high-level concerns from those of auto-tuner implementers involving low-level optimizations. The benefits of delivering this kernel in the form of a specializer, rather than a traditional library, are discussed. Performance of the matrix powers kernel specializer is evaluated in the context of a communication-avoiding conjugate gradient (CA-CG) solver, which compares favorably to traditional CG.
Advisor: Armando Fox
BibTeX citation:
@mastersthesis{Morlan:EECS-2012-95,
Author = {Morlan, Jeffrey},
Title = {Auto-tuning the Matrix Powers Kernel with SEJITS},
School = {EECS Department, University of California, Berkeley},
Year = {2012},
Month = {May},
URL = {http://www.eecs.berkeley.edu/Pubs/TechRpts/2012/EECS-2012-95.html},
Number = {UCB/EECS-2012-95},
Abstract = {The matrix powers kernel, used in communication-avoiding Krylov subspace methods, requires runtime auto-tuning for best performance. We demonstrate how the SEJITS (Selective Embedded Just-In-Time Specialization) approach can be used to deliver a high-performance and performance-portable implementation of the matrix powers kernel to application authors, while separating their high-level concerns from those of auto-tuner implementers involving low-level optimizations. The benefits of delivering this kernel in the form of a specializer, rather than a traditional library, are discussed. Performance of the matrix powers kernel specializer is evaluated in the context of a communication-avoiding conjugate gradient (CA-CG) solver, which compares favorably to traditional CG.}
}
EndNote citation:
%0 Thesis %A Morlan, Jeffrey %T Auto-tuning the Matrix Powers Kernel with SEJITS %I EECS Department, University of California, Berkeley %D 2012 %8 May 11 %@ UCB/EECS-2012-95 %U http://www.eecs.berkeley.edu/Pubs/TechRpts/2012/EECS-2012-95.html %F Morlan:EECS-2012-95