AMBER Archive (2007)

Subject: Re: AMBER: AMBER8 compiler with intel 9.1 compiler

From: Peter Z. Qin (pzq_at_usc.edu)
Date: Thu May 17 2007 - 18:17:29 CDT

----- Original Message -----
From: "David A. Case" <case_at_scripps.edu>
To: <amber_at_scripps.edu>
Sent: Wednesday, May 16, 2007 2:01 PM
Subject: Re: AMBER: AMBER8 compiler with intel 9.1 compiler

> On Wed, May 16, 2007, Peter Z. Qin wrote:
>>
>> [pzq_at_hpc-login1 src]$ ifort -V
>> Intel(R) Fortran Compiler for 32-bit applications, Version 9.1 Build
>> 20060323Z Package ID: l_fc_p_9.1.032
>> Copyright (C) 1985-2006 Intel Corporation. All rights reserved.
>> FOR NON-COMMERCIAL USE ONLY
>>
>>
>> 2. The following tests failed in the intel 9.1 compilation but passed
>> for
>> the intel 8.1 compilation. Could this give us a clue
>> cytosine, dhfr, dhfr.noshake,mtrx, trx.cpln, alp.
>>
>
> Be sure to apply bugfix.64 from the Amber web site. This fixes a problem
> with
> newer Intel compilers. I had thought that it only applied to ia64
> compilers,
> but it might well affect ia32 as well. And, as Scott suggested, if bugfix
> 64
> doesn't help, try compilations without optimizing (make FOPTFLAGS in
> config.h
> the same as FFLAGS).
>
> Another reasonable option is to upgrade to Amber 9. Basically, the
> problem is
> that ifort version 9.1 didn't exist when Amber 8 was released. When
> problems are found with newer compilers, fixes or work-arounds may get
> applied
> to the current release, but often will not be back-ported to earlier
> releases.
> Or, the new code may be sufficiently different than the old that it never
> triggers the problem, so we would never see it.
>

Compiled without optimizing fixed the problems for tests of cytosine, dhfr,
dhfr.noshake,mtrx, trx.cpln, alp. There are still (potential) problems in
all tests in the antechambe directory. A typical difference file is:

[pzq_at_hpc-login1 ash]$ more prmtop.dif
17,19c17,19
< -1.1E+1 -1.0E+1 6.1 9.4E-1 1.8
< 1.1E+1 -1.1E+1 -1.0E+1 6.1 1.4
< 1.3 7.4E-1 4.9E-1 -2.8 2.0 

---
>  -1.1E+1 -1.0E+1  6.1  9.2E-1  1.9
>   1.1E+1 -1.1E+1 -1.0E+1  6.1  1.3
>   1.3  7.0E-1  5.0E-1 -2.8  1.9
[pzq_at_hpc-login1 ash]$

Looks like rounding errors to me.  My questions are:

1. Are these difference (1.9 vs. 1.8, 1.4 vs. 13, etc.) significant?  If so, 
what is the fix?

2. Does this affect other modules, such as sander.  I mostly study 
macromolecules, and probably will not use antechamber?

-peter 

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