Re: gfortran and a compilation
On Tuesday 27 June 2006 16:44, Jo Shields wrote:
> Francesco Pietra wrote:
> >>> However, I do not understand whether dft, like nevpt is only
> >>> distributed in cvs version. In that case I can do nothing.
> >
> > I have been informed that dft allow parallelization, so that it is
> > interesting to go on.
> >
> > Hi Jo: I have also been indicated - by people who probably do not know
> > debian but are masters of Dalton - to compile mpich, whereby all
> > libraries are compiled at 64bit on my system. Do the above packages do
> > the job, with no need to compile mpich? Having examined the packages I
> > expect Yes.
>
> That depends. MPICH is fairly generic, and at compile-time, you need to
> specify which parallelization model to use - for example, a work, we use
> MPICH-GM, which is for use on Myrinet high-speed proprietary
> interconnects. If you want to use shared memory (i.e. MPI on a local
> machine, not over a network), then the prepackaged *mpich-shmem*
> packages are fine.
mpich-shmem is the one for me
> If you want to use gigabit ethernet over TCP/IP on a
> cluster, you want the vanilla mpich packages, or *mpich-mpd* packages if
> your cluster provides an mpich execution daemon.
>
> I would *never* compile something if it can be avoided - why repeat work
> someone else has already done?
In fact, I asked above
> The advice you received about 64-bitness
> is false, as almost all packages on AMD64 debian are already 64-bit
> (rare exceptions are 32-bit compatibility libraries
I purged from them
> , and things like GRUB)
That I have
Incidentally, migjt it be that the problems I have encounterd with mpqc
(failure to reach convergence with complex molecules) were due to not having
installed mpich? Actually, however, mpqc detected the four nodes, and on less
complex molecules, where it reached convergenge and led to
experimental-identical results, it was tremendously faster that on serial
computation.
Thanks again
francesco pietra
Reply to: