Dear Lee,
I would avoid playing around trying to get things 64 bit until you can get
the standard 32bit serial installation to compile. Note, the only reason you
would typically want 64bit functionality with amber is to be able to write
files greater than 2GB since using 64 bit integers for everything will
typically result in a performance hit. This is especially true on the new
P4's which don't actually do true 64 bit addressing.
Attached is a standard config.h script that I have used with Amber8
(bugfixes 1 to 41). This was created with the configure options:
../configure -p4 -static ifort
The only real difference between this version and yours is that I don't link
against the MKL libraries.
This compiles perfectly on our bluefish cluster at Scripps which is dual
3.4GHz Xeon EM64T's with ifort version fce 8.1 Build 20050203 Package ID:
l_fce_pc_8.1.025.
GNU ld version is 2.15.91.0.2.20040727 (SuSE Linux) - try upgrading your
version of ld if it is older than this.
The operating system is SuSE Linux 9.2 (x86-64).
So the following works fine for me on this architecture:
setenv AMBERHOME ~rcw/amber8
cd $AMBERHOME/src
source /opt/intel_fce_80/bin/ifortvars.csh
../configure -p4 -static ifort
make clean
make
This compiles the entire amber8 suite with no errors (just a warning about
getpwuid requirign dynamic access to glibc at runtime) and all test cases
pass. I also believe that this method should give support for large files
with no other changes necessary but I haven't verified this.
So, I would look to making sure your version of ifort and ld matches mine.
All the best
Ross
/\
\/
|\oss Walker
| Department of Molecular Biology TPC15 |
| The Scripps Research Institute |
| Tel:- +1 858 784 8889 | EMail:- ross.rosswalker.co.uk |
|
http://www.rosswalker.co.uk/ | PGP Key available on request |
Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.
> -----Original Message-----
> From: owner-amber.scripps.edu
> [mailto:owner-amber.scripps.edu] On Behalf Of Shekter, Lee
> Sent: 21 March 2005 16:12
> To: amber.scripps.edu
> Subject: AMBER: Dual Xeon EM64T Intel Fortran Compile Problem
>
> Hello -
> This question was originally posted by another user back in
> September, but the responses did not solve the problem. I
> contacted the user, keiodai.shohoku.ac.jp directly but he
> still was only able to compile AMBER 8 in serial mode, not
> parallel. I have been trying to compile AMBER 8, with all bug
> fixes, on an IBM xSeries 336, with 2 X 3.0 GHz Xeon EM64T
> cpus, 8G RAM, 36G U320 SCSI HDD. The OS is RHEL 3.0 (AMD64
> version) with update 4. The compiler is Intel Fortran, v. 8.1
> along with Intel MKL v. 7.2, and gcc v. 3.2.3. I have tried
> compiling in both 32-bit and 64-bit mode. The 32-bit mode was
> attempted by simply running ./configure -p4 ifort since that
> seemed to be the most similar and then running configure. I
> am attaching the config.h that was created along with the
> output of the compile. The latter is a little long, but I'm
> sure any experts reading this will be able to sort the wheat
> from the chaff pretty quickly. One of the responses to the
> earlier question was that this was a compiler error, but it
> looks like an error in "ld". To wit:
> ld: BFD 2.14.90.0.4 20030523 internal error, aborting at
> ../../bfd/elfcode.h line 226 in bfd_elf64_swap_symbol_in
> I also tried downgrading the Intel compiler to v. 8.0 as per
> an earlier suggestion but that made things worse. I then
> tried to force configure to recognize the "64-bitness" of my
> machine by substituting $arch = 'ia64' with $arch = 'x86_64',
> since uname -m yields x86_64, not ia64. This seemed to create
> a 64-bit config.h, which now recognized the 64-bit MKL
> (though not the EM64T MKL which is included). When I ran this
> configure, compilation ran a little longer but still died,
> reporting: /opt/intel/mkl72/lib/64/libmkl_lapack.a: could not
> read symbols: Bad value.
> I am including the output of that as well, called
> "configure_64.log. The 32-bit error and the 64-bit error have
> both been reported on the net, just not in the context of
> compiling AMBER 8.
>
> In summary, I am attaching 5 files:
> config.h: the original config.h
> make_serial.log: the original output from compiling
> configure_64: the 64 bit configure script
> config.h_64: the resulting config.h
> make_serial_64.log: the output from running the 64-bit configure
> I am not attaching the original configure, since that is
> included in the /src directory.
>
> Yes, I did set "ulimit" to unlimited.
>
> Thanks in advance,
>
> Lee Shekter
>
> Assistant Professor (Adjunct) Medical Science
> Brown University
> Divison of Biology and Medicine
> Box G
> Providence, RI 02912-9107
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Tue Mar 22 2005 - 01:53:00 PST
