[...] > > > Btw if the MPI implementation is openmpi, LD_LIBRARY_PATH doesn't need > to be changed , as mpirun/mpiexec will set the correct environment > automatically (not to mention implementations like mpich v1, which > link > statically by default)... > > > Well , that was the impression I had earlier, that it can set the > correct environment. But if the system contains both gcc compiled > openmpi ( most common location is /usr/lib/openmpi , iff installed by > the OS !!! ) and intel compiled MPI ( installed by user ) then my > personal experience tells me to better set the LD_LIBRARY_PATH. > openmpi faq's are also
I agree with you that setting the LD_LIBRARY_PATH is a good idea, since if the binary is executed directly instead of using mpirun/mpiexec (which is a common practice, e.g. for debugging) afaik there is no other way to let it work without admin privileges (and it woulnd't hurt in all other cases anyway). And it's a good thing you suggested it in your mail, since I completely forgot about that. As far as how much I liked the suggestion about the PATH variable, that's the only reason why I have put politeness aside in my reply. If you correct somebody by pushing for a solution that's far worse than the original one, being bashed is the physiological reaction and not a personal matter. This is just the way the dialectic process works as far as the hacking community is concerned: no hard feelings (you should lurk a bit in the ##C channel on freenode or sample a couple of mail in the kernel mailing list and you'll see what I mean: man, the things you'll see there... :-| ). Anyway, we are both here to help, and you had a point about LD_LIBRARY_PATH, so thank you for your contribution Best Regards, Riccardo
