On Thu, Mar 3, 2011 at 5:04 PM, Michael Hoffman <[email protected]> wrote: > The real trouble with OpenMP and threading ends up being less about the > threads and more about whether or not anything is gained. > > From experience I've learned to keep a sequentially optimized version and > always test against that. Even if all cores are being used in the openMP > version there is a chance that bottlenecks slow it down enough to make the > sequential no slower. >
fortunately, compiling an openMP code sequentially is trivial (just remove the compile flags) The gain also higly depends on the number of cores that are used. Again it's all in the choice of what areas are to be parallelized. Calculation intensive areas like AI or loading WML are the most likely targets > -- > Michael Hoffman > ROSE Compiler Group > Lawrence Livermore National Laboratories > Department of Energy > > _______________________________________________ > Wesnoth-dev mailing list > [email protected] > https://mail.gna.org/listinfo/wesnoth-dev > > _______________________________________________ Wesnoth-dev mailing list [email protected] https://mail.gna.org/listinfo/wesnoth-dev
