Re: amber-developers: qmmm2/neb_gb;compile problem with evb_input.f (ifort on linux)

From: Robert Duke <>
Date: Mon, 27 Mar 2006 22:42:54 -0700

Dave -
I saw that Dave Matthews already tried the blank elimination trial, so I
not attempt this. In general, I think fortran is incredibly forgiving
whitespace whereever people want to put it ("forgiving" being the kind
word - I personally like rigid syntax). Anyway, just looking at this
it parses fine in my head at the locus of the problem but perhaps either
there is a problem a bit upstream, or 2) more likely, it is just more
complexity than the expression evaluator can handle (ie., it is not
perfect - gee, compilers are written by people too). There has to be
evaluation of both the parens and the quotes, and they are intermixed in a

way that was perhaps unanticipated or untested. SO I would look at
simplifying the freaking code. Not like it is fun to read when it is this


Regarding ifort 9, the problem I had was when combining it with MKL 8
(latest release), and the problem is that MKL 8 and ifort 9 both ship with

incompatible libguides. So if you pick up "the wrong one" (and either one

could be wrong I would expect, depending on circumstances) you get link
failures. Which one you pick up depends on the order of setup script
execution mkl then ifort vs ifort then mkl; this creates different
LD_LIBRARY_PATH values. For pmemd using MKL, I found I could just omit
libguide and things work fine, despite mkl doc that says you should
it. I did report all this a month or so ago when I was working on GB. I
also encountered link problems when using ifort 9 with mkl 7.21 on an
box at unc when trying to build sander. In this instance it was linear
algebra routines that were causing the grief, not vector math libraries.
ended up deciding that the library-compiler combination was just broken,
no use fighting it after probably an hour of poking (couldn't find the
missing stuff defined anywhere, and couldn't link without dragging in
references to the missing stuff). So for ifort 9 I have encountered two
cases where older versions of mkl don't mix particularly well, and as of a

month ago, there was nothing past mkl 8 (I have not looked, but presume I
would have gotten notified since I have a license).
Regards - Bob

----- Original Message -----
From: "David A. Case" <>
To: <>
Sent: Monday, March 27, 2006 7:32 PM
Subject: Re: amber-developers: qmmm2/neb_gb;compile problem with
(ifort on linux)

> On Mon, Mar 27, 2006, Robert Duke wrote:
>> Yes, I hit this too. Sorry I didn't report it - figured it was more
>> widespread, though on quick exam I did not see what the compiler did
>> like (ie., may well just be a *.#! compiler bug). I was on RHEL 32
>> ifort 8.1.025, which in general I have liked better than 9, but partly
>> because of better compatibility with current MKL releases.
> Dave, Bob:
> If you have a chance, try replacing "% " with "%", and see if that makes
> difference. Other than that, it is not clear what the compiler is
> confused
> about, although this is clearly a pretty awful line of code for a human
> to read....
> ...dac
> p.s. to Bob: what are the problems with MKL and ifort 9? Any particular
> release of MKL that causes the problem? What are the symptoms?
Received on Wed Apr 05 2006 - 23:49:34 PDT
Custom Search