Thanks for the info, I was trying to profile something simple for starters so went with intro/ex3.
You were right in that I wasn't compiling PETSc with -pg, but unfortunately that seemed to be only part of the problem. I ended up having to configure libMesh with --disable-shared --enable-static and with these changes the produced perf logs seem to behave as expected - as this satisfies my needs Ill just stick with this configuration for now. Thanks again, Boris On Tue, Jul 10, 2018 at 10:50 PM, Roy Stogner <[email protected]> wrote: > > On Tue, 10 Jul 2018, Boris Boutkov wrote: > > Oh, that's unfortunate. I've uploaded the logs in question to a public >> google drive folder, they should be available for download at : >> >> https://drive.google.com/drive/u/1/folders/1-FqKx9FhtSgRAttG >> VW-4cYNUfzehbpTz >> > > The PerfLog in intro_ex3.log says 97.66% of your time is being spent > in PetscLinearSolver::solve(). What exactly are you trying to profile > here? There's not even any time spent within PerfLog'ed subroutines > called from within solve(), so you can't be doing a multigrid sweep > here, can you? > > If you're not compiling PETSc with -pg that would explain where 97.66% > of your runtime vanished in the gprof output, but I don't see why 99% > of your *call graph* seems to have vanished. That gprof.log looks as > if you linked a METHOD=prof compiled example against a METHOD=opt > compiled libMesh somehow. > --- > Roy > ------------------------------------------------------------------------------ Check out the vibrant tech community on one of the world's most engaging tech sites, Slashdot.org! http://sdm.link/slashdot _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
