dac> ./configure -opteron -verbose -big pgf90
dac> Don't use the -big option, until you have tried everything without this
dac> option. I don't think it really works, or indeed in necessary.
The idea of -big is to create executables that can access more than
2^32 bytes of memory. If you do not need that much address space then
even on a 64 bit machine you do not need -big.
Indeed -big may have a performance penalty.
When we had an opteron at Scripps in Aug 2003 we had troubles building
Charmm and Amber. But ultimately sander was built with
configure -opteron -big pgf90 and configure -opteron pgf90.
I do not have an opteron now, but I do not notice any glaring
problems in the pgf90 section of configure.
On Sun, 11 Jul 2004, Mark Williamson wrote:
> ./configure -opteron -verbose pgf90
>
> [root.e01 root]# pgf90 -V
> pgf90 5.1-6
> pgf90 -c -tp k8-32 -O -v -V -Minfo=all -Minform=inform -Mneginfo -o new2oldparm.o _new2oldparm.f
> pgf90-Error--tp k8-32 not supported in this installation//usr/pgi/linux86/5.1/bin
>
> I'm assuming the error has occurred because I've only got the 64bit
> compiler installed.
pgf90 5.1-3 did not complain about -tp k8-32 when used on a pentium IV.
> Now, why was configure using k8-32?
because the default build size for amber is 32 bit regardless of the
default word size of the architecture.
> Surely this will only produce 32bit code?
hopefully, since that is the intention.
> In fact, there is no reference to k8-64 (AMD K8 Processor 64-bit mode)
> in the configure script:
again that is intentional because according to the pgf90 man page :
The default in the absence of the -tp flag is
to compile for type of CPU on which the compiler is running.
So one only gets k8-64 when one builds on an opteron; this should be changed
if one wants to do cross compilation, but my advice is 'builder beware'.
> No idea why, but replacing every instance of k8-32 with k8-64 in the
> config.h (from ./configure -opteron -verbose pgf90) produced a good build!
This seems reasonable to me, but I have not been following every nuance
of this thread to know if you are implying some contradiction.
Jiten> Another problem I face after the compilation is that xleap seg faults.
xleap and tleap may not be 64 bit clean.
Build leap for 32 bit, ie, without -big.
Scott Brozell
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Mon Jul 12 2004 - 19:53:00 PDT
