Everyone,
I have a cluster that has a mix of processors in it which has required me
to specify a variety of SSE_TYPES when compiling. This is not working for
me in AMBER14, so I am working on modifying the configure script.
What I found is that the current configure script specifies both -xHost and
-axSSE4.2,SSE3 . Apparently this does not work (at least not for the
processor combination I have). However, if I specify -axSSE4.2,SSE3
without xHost, things are fine.
So, I'm making the appropriate changes, but the way this is handled for
older Intel compilers looks strange to me. I could leave it unchanged
which makes the configure script a little clumsy (internally). Or I could
clean things up a bit, but this will change the behavior silghtly. Since I
have no old Intel with which to test I thought I would consult the list
here before making a change.
Currently with old Intel (10 and earlier) configure will generate coptflags
and foptflags that include (among other things) both xHost and axSTPW. My
understanding is that xHost specifies the highest extension that will run
on the current processer (e.g. xS or xT or xP or xW). So, if one is
requesting the compiler to make all of the various SSE extensions then
xHost isn't doing anything. (Or as I found with Intel 11, xHost is
dominant and the multiple SSE request is being ignored). So, I think this
should be changed to have just axSTPW without the xHost. If I make this
change then things will look analogous in all Intel versions, we will just
be switching from the STPW nomenclature in the older version to the SSE
nomenclature in the newer version.
Does this seem OK to people who know?
Thanks,
Brent
--
_______________________________________________
Brent P. Krueger.....................phone: 616 395 7629
Associate Professor................fax: 616 395 7118
Hope College..........................Schaap Hall 2120
Department of Chemistry
Holland, MI 49423
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Thu May 08 2014 - 08:00:03 PDT
