AMBER Archive (2002)

Subject: RE: 64-bit Compilation SUN's forte compiler

From: Fernando Martin (fmartin_at_plexxikon.com)
Date: Mon Oct 28 2002 - 18:24:21 CST

Hi Scott,

>Generally, a 64 bit addressable executable is only recommended for
>applications that require the extra addressing space because a
>32 bit addressable executable will have superior performance by a
>couple of percent.
>For reference, I do not believe that the large benchmark rt_polymerase
>requires 64 bits, but I haven't verifyed this recently.

Well, if it was up to me I would stick to 32-bit, but 'nmode' will poop for
moderate to large systems (~3000 atoms). If you want to use this tool to
assess these system's entropy you have no choice but to re-dimension
'nmode'-dependent routines, produce 64-bit executables and cross your
fingers for the ride.

>I have not built Amber on a Sun platform, but have built lots of
>software on Suns. The default address size is platform dependent
>and -fast will very probably give the best choice.
>Regarding performance, the default optimization level included by the
>-fast option is O4; hence, one might alter Machine.sparc like this
>setenv L3 "f90 -c -fast -O3 "
>->
>setenv L3 "f90 -c -fast "

Yes, unfortunately these compilation flags will produce 32-bit applications.
Thanks though,

Fernando