On Tue, Dec 13, 2016 at 3:52 PM, James Ostrander <jlost.umich.edu> wrote:
> Hi, I am trying to use AmberTools on an IBM POWER8 system (ppc64le). IBM's
> compilers produce the fastest machine code for this system, so I always try
> to use them first.
>
> AmberTools seems to have no direct support for the IBM XL compilers
> although it does offer out-of-the-box options for Intel, Clang, etc. I also
> had a quick look at the build system and it seems non-trivial to manually
> configure or add your own (unless I'm just missing something!).
>
> So from there, I have a few questions:
> - I didn't see any options in configure --full-help that suggested a direct
> compiler override. Am I missing something?
> - If not, are there any plans to support IBM's compilers?
> - If not, and someone offered you a patch that added support, would you be
> willing to accept it into the official codebase?
> - If so, is there any specific guidance you can offer that would get
> someone started on adding the support?
>
​You can always try to configure Amber (I would suggest using the
--with-netcdf flag and use a pre-built NetCDF library if possible) and
specify the gnu compilers. Then replace the compiler binaries in config.h
with the IBM compiler names (and replace compiler flags if necessary). If
the IBM compilers have flags compatible with another supported compiler,
then you may be able to set the CC, FC, and CXX environment variables to
point at the IBM compilers while specifying the flag-compatible compiler
suite to configure.
Note that these compilers are not supported because developers do not have
access to these compilers. And adding support is only a relatively small
part of the job -- the much bigger part is maintaining that support through
multiple versions of the various libraries, codes, and compiler versions.
It can sometimes be better to not have a feature at all than have some
untested code that *might* provide support.
That's not to say there's no value, though. If there are small changes
necessary to improve IBM compiler support​
that also improves the configure script as a whole (and does not introduce
maintenance headaches), I imagine such a patch would be welcome.
HTH,
Jason
--
Jason M. Swails
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Dec 13 2016 - 22:30:05 PST
