AMBER Archive (2008)

Subject: Re: AMBER: AmberTools on OS X Leopard with Intel compilers

From: Yu Chen (chen_at_hhmi.umbc.edu)
Date: Mon Apr 21 2008 - 13:47:06 CDT


>>>
>>> We are tring to compile/test/install AmberTools as the first step in Amber
>>> 10 installation on a xserver with Leopard with Intel compiler (10.1.014).
>>
>> You may be the first to use icc on Leopard ... if someone else on the list
>> has tried that, they may chime in here.
>>>
>>> 1. The "configure_at icc" finished fine except for the g77 part(since we
>>> don't have g77/gfortran installed yet), which says " We will be unable to
>>> compile mopac or resp", since we are not going to use those two for now,
>>> we assume that we can ignore it for now, is it right? Or do we have to
>>> have
>>> it compiled for everything to work?
>>
>> If you don't need mopac or resp, you are fine.
>>
>>> So we assume it compiled successfully. Although one strange thing, I
>>> specified the complier as 'icc' as you can see in first step, but some
>>> progames still used 'gcc' and 'g++' as compilers (like Makefile in
>>> gleap/freelib/readline), is this normal? Do those programs have to be
>>> compiled by 'gcc'?
>>
>> Yes: sleap has only been tested with gcc, so that is hard-wireds.
>>
>>>
>>> 3. After the 'make', we did 'make -f Makefile_at test' in the 'test'
>>> directory, but it gave out error right away with these:
>>> #######################################
>>> ( cd nab; make test )
>>> =====================================================
>>> Running test to make dna duplex:
>>>
>>> /bin/sh: line 1: 43282 Segmentation fault ./duplex < duplex.in
>>
>> Pretty serious. We know icc works on linux (x86_64 and ia64), but is not
>> working for you for some reason. And, as far as I know, we have only
>> tested
>> up to icc version 10.0.023. [You already reported off-line that icc 10.1
>> fails on the netcdf configure step, whereas 10.0.023 has no such problem.]
>>
>> The obvious workaround is to use gcc. It's not clear that icc is actually
>> faster, and it is clearly much less well tested. Unless you end up running
>> NAB for days on end, you would probably never notice the difference between
>> gcc and icc. The other programs in AmberTools are setup and analysis
>> routines, where it doesn't make sense to try to squeeze out the last bit of
>> performance.
>
> Thank you, Dr. Case for detailed explaination. I will try gcc. The reason we
> used icc is that the Apple software tech told us that Intel compiler used on
> Intel architecture is much faster than gcc.

I tried both Apple stock gcc (4.0) and latest gcc (4.3) on AmberTools,
both failed compiling with the following: wondering if anybody can help

=======================
ld: warning in /mfslab/software/amber/v10/lib/libpdb.a, file is not of
required architecture
Undefined symbols:
   "_pdb_read_record", referenced from:
       _loadPdb in trajectory.o
       _checkCoordinates in ptraj.o
       _ptrajInitializeState in ptraj.o
   "_pdb_read_string", referenced from:
       _checkCoordinates in ptraj.o
       _checkCoordinates in ptraj.o
   "_pdb_write_record", referenced from:
       _savePdb in trajectory.o
       _savePdbHigh in trajectory.o
ld: symbol(s) not found
collect2: ld returned 1 exit status
make[1]: *** [rdparm] Error 1
make: *** [install] Error 2

Thanks,

>
>>
>> Of course, it you want, please go ahead and compile a debug version and try
>> to
>> get a traceback on the seg fault. That program is a trivial one, so the
>> error
>> is probably generic somewhere. As a start, specify -static in
>> configure_at,
>> and modify the config.h file to turn of optimization and turn on debugging.
>>
>
> I am not sure how to do this. I am not good at debugging. Could you give a
> little bit more detail on how to do this?
>
> Thanks,
>
> CY
>
>> ....good luck...dac
>>
>> -----------------------------------------------------------------------
>> The AMBER Mail Reflector
>> To post, send mail to amber_at_scripps.edu
>> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>>
>
>
> ===========================================
> Yu Chen
> Howard Hughes Medical Institute
> Chemistry Building, Rm 182
> University of Maryland at Baltimore County
> 1000 Hilltop Circle
> Baltimore, MD 21250
>
> phone: (410)455-1728 (primary)
> (410)455-6347 (secondary)
> fax: (410)455-1174
> email: chen_at_hhmi.umbc.edu
> ===========================================
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>

===========================================
Yu Chen
Howard Hughes Medical Institute
Chemistry Building, Rm 182
University of Maryland at Baltimore County
1000 Hilltop Circle
Baltimore, MD 21250

phone: (410)455-1728 (primary)
         (410)455-6347 (secondary)
fax: (410)455-1174
email: chen_at_hhmi.umbc.edu
===========================================
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu