|
|
|
Foundation ::
Parallel Computing ::
MPICH
|
MPICH Support
Fortran Compilers
Back to top
Some users have had difficulty when mixing C and
Fortran compiler vendors. For example, when using GNU gcc and the Portland Group
pgf77, C programs would fail to link. This is because some of the code used to
provide definitions of some Fortran types is computed at MPI_Init time in MPICH
1.2.3 by using a small Fortran routine, and in some cases, these routines refer
to symbols that are defined only in the Fortran run time libraries. To fix this,
you need to add the Fortran libraries when building C programs. You can do this
with the -lib switch in configure:
configure -cc=gcc -fc=pgf77 -lib="-L/usr/local/pgi/linux86/lib
-lpgftnrtl -lpgc" ... The next release of MPICH will attempt to
avoid the need for these libraries when possible and will determine them for you
otherwise.
If you are using a Fortran 90 compiler as your Fortran compiler (e.g.,
-fc=pgf90), you'll need a longer list of libaries. Look in the library directory
for your Fortran compiler for the libraries. For example,
configure -cc=gcc -fc=pgf90 -lib="-L/usr/local/pgi/linux86/lib
-lpgf90 -lpgf90rtl -lpgftnrtl -lpgc" ...
Redhat 7.0 Compilation Problems
Back to top
Redhat 7.0 ships with a version of gcc that is a development version and should not have been used by Redhat. See http://gcc.gnu.org/gcc-2.96.html for GNU's discussion of this.
LINUX TCP Performance
Back to top
There is a bug in the implementation of TCP in
some versions of Linux. This has been documented by Josip Loncaric. His description of the problem and
fix are available on the Web. He reports that as of Linux 2.4.3, this TCP
patch is no longer necessary.
Fortran 77 and Fortran 90/95
Back to top
MPICH currently supports Fortran 77. There is some support for using Fortran
90/95, but this is still imperfect. There is a table
that shows some of the combinations of configure options that can be used to
make MPICH work with Fortran 90 and Fortran 95 compilers.
Failures in System Includes on Solaris
Back to top
Some users have had problems with the make failing when compiling the routines
in mpid/ch_p4 on Solaris. A typical output is
...
In file included from /usr/include/rpc/rpc.h:38,
from p4/include/p4.h:16,
from chdef.h:9,
from ../ch2/packets.h:350,
from mpiddev.h:15,
from adi2recv.c:9:
/usr/include/rpc/auth_des.h:58: field `adv_ctime' has incomplete type
...
This appears to be caused by the MPICH configure deciding to reject cc
as the C compiler (because it does not support function prototypes) and picking
gcc instead. Unfortunately, configure is still using cc
for the C preprocessor, and gets confused (this is actually a bug in autoconf;
it should not be using a C preprocessor to decide what the C compiler, whose
search paths might depend on various options, will find).
To fix this, add -cc=gcc to the configure line, reconfigure and
remake.
Unsolved Problems
Back to top
errno=110 or connection timed out in Linux
This indicates that Linux has closed a TCP connection. This can happen if, just when MPICH is trying to communicate a message, the network interconnect becomes very busy. Other operating systems have less fragile TCP connections. We are working on a fix for this, but a better one would be for Linux to provide more robust TCP connections.
Shared Memory in LINUX
Shared memory in LINUX (ch_shmem and we expect ch_p4 with -comm=shared) may
fail. In some Linux systems, the system call 'semget()' fails to create a
new semaphore set. In particular, MPICH will return the following error
message when attempting to run an executable.
semget failed for setnum = 0
This seems to be a problem with LINUX and not with MPICH. While attempting
to uncover the problem, it was found that most LINUX machines worked, however,
there were some that did not. Further investigation led us to two C programs
from the "LINUX Programmer's Guide",
semtool.c - command
line tool for tinkering with SysV style Semaphore Sets
semstat.c -
command line tool for the semtool package which displays the current value of
all semaphores in the set created by semtool (not necessary).
These tools are useful in determining if your LINUX machine will run MPICH in
shared memory. If you are able to create a semaphore as the result of
running semtool, it is highly probable that MPICH will run on your LINUX
machine with shared memory.
At the moment, we are attempting to get to the root of the problem and would
appreciate any suggestions or comments from the LINUX world.
One user found that the include file /usr/include/linux/sem.h (used by the kernel) and /usr/include/sys/sem.h and /usr/include/sys/ipc.h were not consistent, causing programs using semaphores to fail. If you have this problem, you need to correct your installation of Linux.
Problems with MPI-IO and NFS
Back to top
The network file system (NFS) must be configured extremely carefully for
MPI-IO (and many other programs) to work correctly. Unfortunately, few systems
are so configured, and doing so can adversely impact performance. As a result,
programs using files on an NFS system may hang or produce incorrect results.
Note that this is, officially, a design feature of NFS; unless the
NFS system is configured with no attribute caching, any two
processes, accessing the same file, may produce incorrect results.
You can use the -file_system=ufs option of configure to build an MPICH that
supports only UFS (Unix File System); MPI-IO works correctly with UFS, XFS,
PIOFS, HFS, SFS, etc. (more precisely, those file system that
correctly implement basic Unix I/O system calls; something that NFS does not
do).
Detailed instructions on setting up NFS can be found in the installation manual or online.
Problems not in MPICH
Back to top
Some problems are caused by compiler problems. Some of the problems are
- Solaris C
- Version SC3.0.1 will fail with
cg: assertion failed in file ../src/regman/regman_reporter.h at line 36
cg: add_edges_to_new_node -- unexpected edges
cg: 42 warnings, 1 errors
cc: cg failed for ad_read_coll.c
Removing -O from all of the Makefiles (particularly in ROMIO) may fix this
- GNU gcc
- Version 2.8.x does not handle the command line argument -I./ correctly.
Compaq Fortran
Version 5.3-915 of f95 has a bug that causes it to fail with files that
have the extension .F (e.g., to be processed with the preprocessor). In this
case, specify the Fortran 90 compiler to be f90 instead of f95.
An update to Compaq Fortran 5.3 exists in Compaq's FTP
repository at ftp://ftp.compaq.com/pub/products/fortran/Tru64/.
Please see the file readme.txt in this directory.
You may also want to check on patches for the system that you are running
on.
- HP patch database:
- Europe or
US
Back to top
|
|
|
|
|
Open Channel Software runs entirely on Open Source Software. We return
value to the Software community in the form of
services and original software. Most of our
content is currently available as source code, with the
copyright owned by the original author, All
Rights Reserved. Everything else is Copyright
©2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010 Open Channel Software.
View our privacy statement.
Contact webmaster at openchannelsoftware dot org with questions.
|
|
|
