Interior Point Methods involve the solution of many large sparse symmetric linear systems. The main bottleneck involved in the solution of these systems is memory bandwidth. Through the use of single precision arithmetic we can halve the amount of floating point data we move through memory, resulting in a significant gain in speed of a sparse factorization. If double precision accuracy is desired it can normally be recovered through the use of iterative methods preconditioned by the sparse factorization.
We report our investigations into the practical success of these methods and describe a new code HSL_MA79 which implements them. Numerical results for a wide range of problems including those drawn from Optimisation are presented.
Current 2016 2015 2014 2013 2012 2011 2010 2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 1998 1997 1996