Hi all,
Given the recent traffic on this topic, I apologise if I seem to be
flagellating deceased equines here. But I've spent a good deal of time
mucking around with this and thought I'd share my discoveries in case
they're helpful.
Detection of architecture
-------------------------
As has been pointed out by others, the "-m" compiler flag picks the
machine type up. "-m64" is used, *if* the computer is booted with the
"6" and "4" keys held down (I kid you not; this forces booting into the
so-called 64-bit kernel).
The Intel compilers
-------------------
I sent a message earlier today regarding an "internal error" which
causes the compiler to choke for AmberTools. Amber itself, on the other
hand, compiles with no significant problems (compiler warnings
notwithstanding).
The GNU compilers
-----------------
These can be used to build AmberTools OK. On the other hand, attempting
to build sander generates a number of errors along these lines:
gfortran -c -O3 -ffree-form -o pb_force.o _pb_force.f
_pb_force.f: In function 'phintp.2255':
_pb_force.f:1286:0: error: type mismatch in component reference
real(kind=8)[0:D.2764] * restrict
real(kind=8)[0:D.7772] *
# VUSE <.MEM_173(D)>
D.5358_39 = CHAIN.410_38(D)->phi;
_pb_force.f:1286:0: error: type mismatch in component reference
real(kind=8)[0:D.2764] * restrict
and so on and so forth.
Apparently this is not so much a sander bug, as a compiler bug related
to gfortran 4.5.0 (although I'd be interested to know whether the sander
code is doing something questionable to bring it about). Nevertheless,
it can be avoided by reducing the Fortran optimisation level (FOPTFLAGS
in config.h) from O3 to O2.
I've cross-posted this at the main Amber list for the benefit of those
users who may be stuck on this point in earlier versions of Amber.
Tests
-----
Both Intel and GNU compilers produce a number of errors and failures of
varying degrees of severity throughout the testing process. What I hope
to do in the near future is come up with a list of what those are, so
they can be marked for fixing (or not, as appropriate).
Cheers,
Ben
--
Benjamin P. Roberts
Postdoctoral Research Associate
Quantum Theory Project
University of Florida
2301 New Physics Building #92
PO Box 118435
Gainesville FL 32611-8435
USA
Phone: +1 352 392 6712
Cell: +1 352 222 3677
Member of the Royal Australian Chemical Institute
and of the American Chemical Society
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Sep 29 2009 - 15:00:03 PDT
