On Mon, 2009-07-27 at 17:53 +0200, Paolo Giannozzi wrote: > Eduardo Ariel Menendez Proupin wrote: > > > Testing the speed, version 4.1 is a bit slower than 4.0.4 (about 9% > > more time in this benchmark: 39.5 vs 36 minutes, using 32 cpus). > > I don't see any good reason why this should happen. Not that I trust
memory alignment of data structures and arrays. particularly with intel fortran, the compiler will automatically try to use SSE instructions instead of regular FP math, as those can process two double precision numbers at the same time. however, SSE requires 16-byte aligned data or you take a performance hit, but the standard malloc usually only returns 8-byte aligned data. > timings. Not since I saw one job running 15% slower after removal of > a variable THAT WASN'T EVEN USED. depends on where the variable was placed. if it was in a place, where the compiler was not allowed to optimize it away, than you can run into exactly the alignment issue that i was mentioning. cheers, axel. > > P. -- ======================================================================= Axel Kohlmeyer akohlmey at cmm.chem.upenn.edu http://www.cmm.upenn.edu Center for Molecular Modeling -- University of Pennsylvania Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323 tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425 ======================================================================= If you make something idiot-proof, the universe creates a better idiot.
