On Fri, 7 Apr 2006, David Leuenberger wrote:
I was not able to compile the parallel version of hdf5-1.6.5 only the
serial one. However the manual mentions that there is a hack in the
parallel version of meep that makes it work with the serial hdf5.
(Yes, although the hack does not scale well to large numbers of
processors.)
I was able to compile successfully configure and compile meep. However
when I run ./make check, the following error occurs. Din anyone have a
similar experience? Does anybody has an idea how I could localize the
error? It seems to occur in some mathematical operation...
From your output, it looks like you are seeing two separate errors. A
segfault in bench.cpp:
bench:, Periodic 12x12, 1.38281, 0.40198
p1_13870: p4_error: interrupt SIGSEGV: 11
rm_l_1_13877: (11.261719) net_send: could not write to fd=5, errno = 32
and also a NaN in bragg_transmission.cpp:
Using MPI version 1.2, 2 processes
meep: large error inf at freq = 0.1: T = nan instead of 0.704209
[0] MPI Abort by user Aborting program !
[0] Aborting program!
p0_22184: p4_error: : 1
make[2]: *** [bragg_transmission.done] Error 1
It would be helpful if you could narrow down the problem.
First, it would be good to check whether this is a problem specific to
MPI. Try compiling the serial version of Meep and see if that passes
'make check'.
Second, it would be nice if the possibility of a compiler bug could be
eliminated. We've encountered compiler bugs with gcc 3.3.1 on x86,
although I think 3.3.5 was okay, but I haven't tested on Itanium. Two
things to try would be: (a) try configuring with --enable-debug to turn
off optimization and turn on debugging flags (usually compiler bugs occur
in the optimization phase), and (b) try a newer version of gcc if you have
one, or perhaps the Intel or HP compilers (./configure
CXX=<compilername> MPICXX=<MPIcompilername>).
If the problems still occur with --enable-debug, then you could give me a
stack track by running under gdb and typing "where" when it crashes.
(--enable-debug will change the floating-point mode to crash when a NaN is
encountered, so the bragg_transmission.cpp test should crash as well.)
Steven
_______________________________________________
meep-discuss mailing list
[email protected]
http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
