AMBER Archive (2004)

Subject: Re: AMBER: Intel Fortran Compiler 8.0 / pmemd 3.1 warning

From: David E. Konerding (dekonerding_at_lbl.gov)
Date: Wed Jan 07 2004 - 10:16:49 CST

Robert Duke wrote:

>1) ifc 7.x and ifc 8 work on the 2.4.x linux kernel, which is the common one
>out there (at least since redhat 7.2, and possibly earlier).
>2) ifc 8 requires either glibc 2.2.5, 2.2.93, or 2.3.2. This means it won't
>work with my trusty redhat 7.2, which has glibc 2.2.4, but should work with
>later releases. Unfortunately ifc 8 has a internal compiler error in it
>right now that keeps it from compiling pmemd 3.1, so we need a patch from
>intel to make it work.
>3) ifc 7.1, latest ia32 release (7.1.037) claims in the release notes to
>require glibc 2.2.4 or 2.2.5. Missing is 2.3.2 which is what the latest
>redhat releases use. I have not confirmed that this is a problem, but would
>not be at all surprised.
>4) pmemd 3.1, built with static libraries on a redhat 7.2 system with glibc
>2.2.4 and ifc 7.1 seems to work fine on redhat enterprise 3.0 with glibc
>2.3.2.
>
>My intention in posting the warning was to help folks avoid a systems
>upgrade trap, and provide a possible workaround if they have already
>upgraded. What a mess!
>
>
Robert,

Thanks for the collection of sets of compiler, glibc, kernel sets that
you've found operational (this sort of accumulated knowledge
is quite useful in the (sigh) ever-changing world of Linux distributions).

A few points:

One thing that complexicates the compatibility table above is that Red
Hat modifies the glibc and kernel enough to make it non-compatible
with other vendors. The best example is in the threading libraries, as
RH back-ported the Linux 2.6 threading library, into the Red Hat Linux
kernel 2.4 and the glibc 2.3.2 that they ship. This causes binary
incompatibility when trying to run apps compiled against the older threading
libraries (or on another contemporaneous distibution without a kernel
2.4 NPTL implementation). This can usually be fixed by setting the env
var LD_ASSUME_KERNEL=2.2.4 which causes the older threading mechanisms
to be used.

The only problem I had with the most up-to-date Intel Fortran 7.1
release on a Red Hat 9 system was in the final linking, a missing
__ctype_b_loc symbol which I believe is the one other people have been
seeing. If you're brave you can read
http://softwareforums.intel.com/ids/board/message?board.id=11&message.id=71
and see if anythign there helps you.

>The software is available for the ifc 7.1 releases under the
>premium.intel.com webpage. You do have to log in to your account, but a
>free account is available even when you get the "noncommercial unsupported
>version" of the compilers - ie., they actually do provide some free low
>priority support. I am skeptical that the latest ifc 7.1 will help, based
>on the glibc requirements listed in the release notes, but I have not
>
>

You should definitely hang on the ifc 7.1 stuff for a while, especially
if you're going to have Red Hat 7.3 systems around for a long time.
The one thing this taught me is that paying for and getting the premiere
support will actually help us, since they seem relatively
responsive. I got an update this morning that they have reproduced the
problem in-house and will be making a small test case they send to the
compiler group, so I think we'll see a fix in the next week or two ;-)

Dave

>actually installed it to check it out.
>

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu