Hi Kim,
Looking at the Makefile, I see that test.sander.BASIC.MPI does have
some MPI specific tests. GB.MPI, LES.MPI, QMMM.MPI, and MPPIMD.MPI
just call the serial tests.
In my case, most of the tests do run, but all fail in the check if
using more than 1 processor. Trying using MPI but with only 1
processor the tests seem to be working OK (PASSED).
Since this now is a different problem, I'll open a new thread about
that in the list.
Gustavo.
On 10/12/06, Kim F. Wong <kfwong.hec.utah.edu> wrote:
> Gustavo,
>
> Do you mind running the parallel tests and see if you are getting
> similar failures as the following:
>
> (1) test.sander.BASIC.MPI
> (2) test.sander.GB.MPI
> (3) test.sander.LES.MPI
> (4) test.sander.QMMM.MPI
> (5) test.sander.PIMD.MPI
>
> All failures gave stdout error similar to the following:
>
> rm_l_1_1573: (0.332031) net_send: could not write to fd=5, errno = 32
> p0_1536: (2.523438) net_send: could not write to fd=5, errno = 32
>
> I think the above tests are fundamentally serial jobs but running under
> an MPI environment. I really have not done much digging within the test
> scripts, though ... so I am guessing. Thanks for your help.
>
> -Kim
>
>
> Gustavo Seabra wrote:
> > On 10/11/06, David A. Case <case.scripps.edu> wrote:
> >> On Wed, Oct 11, 2006, Gustavo Seabra wrote:
> >>
> >> > I just checked out a clean copy of amber, and compilatin fails due to
> >> > F95 compatibility problems. I attach here the compilation output and
> >> > my config.h file. Although this is a mpi compilation, I get the same
> >> > errors with serial.
> >> >
> >>
> >> Seems like these errors are pretty easy to correct -- please just do so.
> >>
> >
> > I needed sander running asap, so I went ahead and followed Dave's
> > suggestion. Apart from the sander/schlegel_dg.f file which was
> > corrected by Kim Wong, there were similar errors in a number of other
> > files inside sander and dcqtp, which were stopping compilation. I
> > corrected those old '$' suppress-newline formats in those files. (I
> > hope the official maintainers won't mind...). The new files are on the
> > tree now, and compile well (at least for me.)
> >
> > For those interested, the changes I did were, in ALL cases, to replace
> > the '$' in a write statement by a ADVANCE='NO' :
> >
> > write (6,'(<<FORMAT>>,$)') ==> write (6,'(<<FORMAT>>)', ADVANCE='NO')
> >
> > Judging by the 'warnings' the compiler gives, there are a number of
> > syntaxes that have disappeared in F95 that are still present. I have
> > no idea why the compiler chose just this one as fatal.
> >
> > Gustavo.
> > P.S.: The reason I went to get a new tree was that, before, I was only
> > 'updating' my local copy, and at some point some calculations started
> > to segfault with no apparent reason. It turned out to be one of those
> > files being compiled with the old config.h: It compiled, but segfaults
> > unpredictably. (My point here is that, even if the compiler didn't
> > pick, we may get strange segfaults if we've been just updating for a
> > while, as I was.)
>
>
--
---------------------------------------------------------------------------------------------
Gustavo Seabra Postdoctoral Associate
Quantum Theory Project University of Florida
Registered Linux user number 381680
Say NO! to software patents: http://www.nosoftwarepatents.com
---------------------------------------------------------------------------------------------
If at first you don't succeed...
...skydiving is not for you.
Received on Thu Oct 12 2006 - 20:36:13 PDT
