From: Paul H. Hargrove (PHHargrove_at_lbl_dot_gov)
Date: Tue Nov 01 2005 - 11:48:05 PST
Adolfo, I have no idea why your results would change, other than your theory on random numbers. The cr_run script will LD_PRELOAD two libraries, the blcr library and the pthreads library. I don't know of any reason why either one would have any direct effect on your simulation. It is remotely possible that a different random number generator is being linked from the pthread library. You could test this by modifying cr_run to LD_PRELOAD only the pthreads lib (and not libcr.so) to see if you still see modified results. -Paul Adolfo J. Banchio wrote: >Before I try to isolate the problem. >I'd like to ask you if it is possible to understand why >a code (a simulation) lunched using cr_run could >give differen results than when lunched directly. > >If I run > >./simul.x > >I get different results than if I run > >cr_run ./simul.x > > >Actually I noticed this since the difference seems to >be quite dramatic (I get particle overlap when using cr_run). >I'm trying to identify where the difference arise and isolate >the problem. > >But I'm totally surprised that something like this could >happen. May random numbers generation be affected? >(the simulation are compiled with Intel F90). > >thanks for any comment, > >adolfo > > > -- Paul H. Hargrove PHHargrove_at_lbl_dot_gov Future Technologies Group HPC Research Department Tel: +1-510-495-2352 Lawrence Berkeley National Laboratory Fax: +1-510-486-6900