Folks -
Just a heads-up on a potential source of grief. In earlier releases of
pmemd, I included the -static switch in the ifc machine files because it
helped avoid problems with needing to distribute the compiler libraries on
all the machines pmemd was running on. However, with later if not earlier
releases of ifc, there has been a lot of ill-defined grief associated with
the interface to system libraries. Now, in an ideal world, there would be a
clear delineation between the compiler-related libraries and the system
libraries, and you would be able to have the right thing happening by using
static compiler libraries and dynamic system libraries. The dynamic system
libraries would present consistent call interfaces and behaviour, and if
they didn't, there would be provision for versioning. I am not an expert on
this stuff in the linux world, but it seems like things are less than ideal,
and in the case of the intel fortran compilers, I have not had much luck
determining what the loader links in dynamically or statically (it is hidden
in some binaries, and the documentation is just the wrong side of hopeless).
SO for now it seems it is safest to NOT use -static with ifc, because it
seems to not get things right; probably doesn't always get things right
otherwise either. SO if you are building old versions of pmemd, consider
taking the -static switch out of the machine files. Also, in building mpich
for ifc, don't use the -static switch, just in case (I did this for redhat
7.2 and it worked fine; I am just now in the process of building mpich for
later versions of redhat). I bring this up for two reasons 1) Chun Wu in
Prof. Yong Duan's group was recently exchanging email with me regarding
problems building pmemd 3.03; we got the other problems straightened out,
but he also mentioned this grief with the -static switch (part of Chun's
last mail embedded below), and 2) I had been helping Carsten D. get mpich
and amber 8 built, and feared she might have used the -static switch which
is in a script I have given out, but she hadn't, and fortunately it is not
in the pmemd 3.0 release note.
I am hoping we have stable solutions for amber 8 on the later redhat
releases with the intel fortran compiler real soon now...
This is Chun's last mail:
Hi Rob,
Thanks! Now it works!
I used dynamic linking to avoid pthread problem and to avoid ghost process
problem introduced by static linking(the processes can not killed by
SGE_execd).
chun
FYI
CPU--->
Intel(R) Xeon(TM) CPU 2.40GHz
glib--->
glibc-2.3.2-95.6
kernel-->
2.4.21-4.0.1.ELsmp
ifc-->
l_fc_pc_7[1].1.038.tar--->intel-ifc7-7.1-40.i386.rpm
../ifcbin -V
Intel(R) Fortran Compiler for 32-bit applications, Version 7.1 Build
20031225Z
icc-->
l_cc_pc_7.1.037.tar-->
intel-isubh7-7.1-40.i386.rpm
intel-icc7-7.1-40.i386.rpm
mpich--->
/raid12t/packages/mpich-1.2.5
pmemd--->
pmemd3.03
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Fri Apr 02 2004 - 22:53:00 PST
