RE: AMBER: Re: Amber_ifort_x86_64 test error

From: Ilyas Yildirim <yildirim.pas.rochester.edu>
Date: Sun, 9 Dec 2007 16:56:09 -0500 (EST)

Ross,

In case it might help you, I have tried to install AMBER 9 to Ubuntu 7.10
with MKL 10.0.1.014. The machine is a Dell Vostro 1400 (Intel Core 2 Duo
CPU T7250). In parallel testing, it passes the test case circ_dna, but
fails at 'cd cnstph; ./Run.cnstph' with an error of

==============================================================
cd cnstph; ./Run.cnstph

MKL func load error: /opt/intel/mkl/10.0.1.014/lib/32/libmkl_vml_p4m.so:
undefined symbol: vmlGetErrorCallBack

MKL func load error: /opt/intel/mkl/10.0.1.014/lib/32/libmkl_vml_p4m.so:
undefined symbol: vmlGetErrorCallBack
rank 1 in job 62 yildirim_34642 caused collective abort of all ranks
  exit status of rank 1: killed by signal 9
rank 0 in job 62 yildirim_34642 caused collective abort of all ranks
  exit status of rank 0: killed by signal 9
  ./Run.cnstph: Program error
make[1]: *** [test.sander.GB] Error 1
make[1]: Leaving directory `/home/programs/amber9/test'
make: *** [test.sander.GB.MPI] Error 2
==============================================================

I have intel c and fortran compilers ver. 10.1.008. I am using Mpich2 ver
1.0.6. I compiled MPICH2 with these intel compilers.

I disabled the MKL option in AMBER9 installation and every test passed.

Best,

On Sun, 9 Dec 2007, Ross Walker wrote:

> Hi Ilya,
>
> How did you compile Amber with MKL V10?
>
> I would suggest just doing:
>
> make clean
> ./configure -static ifort_x86_64
> make
>
> And see what happens. I haven't tried MKL 10 myself yet and know that Intel
> changed the libraries that you link but apparently the old way of doing it
> should still work.
>
> I'll try getting a copy of MKL 10 myself next week and try it.
>
> All the best
> Ross
>
> /\
> \/
> |\oss Walker
>
> | HPC Consultant and Staff Scientist |
> | San Diego Supercomputer Center |
> | Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
> | http://www.rosswalker.co.uk <http://www.rosswalker.co.uk/> | PGP Key
> available on request |
>
> Note: Electronic Mail is not secure, has no guarantee of delivery, may not
> be read every day, and should not be used for urgent or sensitive issues.
>
>
>
> _____
>
> From: owner-amber.scripps.edu [mailto:owner-amber.scripps.edu] On Behalf Of
> Ilya Chorny
> Sent: Sunday, December 09, 2007 10:12
> To: amber.scripps.edu
> Subject: Re: AMBER: Re: Amber_ifort_x86_64 test error
>
>
> I only have one version of MKL( I do have different version for differetn
> architechtures (i.e. 32, 64, and emt64). In the make log and in the error,
> the linking is to the emt64.
>
> Has anyone successfully compiled Amber 9 with MKL 10?
>
> Thanks,
>
> Ilya
>
>
>
> On Dec 9, 2007 10:01 AM, Ross Walker <ross.rosswalker.co.uk> wrote:
>
>
> Hi Ilya,
>
> If this is for running production jobs then yes you should be using ifort
> over gfortran. That makes a 30% or more difference in performance. If it is
> only for local 'debugging / testing' on your local machine then you are fine
> sticking with gfortran.
>
> Note, while Dave says that MKL doesn't make much of a difference it can
> depend on what type of simulation you run, GB sims benefit more than PME
> sims. The QMMM calculations benefit quite a bit from MKL and in AMBER 10
> will benefit a signficantly (read > 50% for systems with more than 90 QM
> atoms) from MKL10 and other math libraries that have optimized diagonalizers
> in them.
>
> What seems weird with the error you are seeing below is that this is a
> runtime error that should have been picked up at compile time. Hence it
> looks to me like you are using a different version of MKL at runtime than
> you linked against at compile time. If this is a cluster you should probably
> check that each of the nodes has identical versions of MKL installed (in the
> same paths) as the login node. You should also check your environment
> variables are the same on the compute nodes as the one you compiled on.
>
> I would also suggest trying to compile statically so you don't run into
> these problems. This sometimes work, sometimes doesn't depending on whether
> statically linkable libraries are available. For MKL this is generally okay
> but I haven't tried it with v10 yet.
>
> Good luck,
>
> Ross
>
>
> /\
> \/
> |\oss Walker
>
> | HPC Consultant and Staff Scientist |
> | San Diego Supercomputer Center |
> | Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
> | http://www.rosswalker.co.uk <http://www.rosswalker.co.uk/> | PGP Key
> available on request |
>
> Note: Electronic Mail is not secure, has no guarantee of delivery, may not
> be read every day, and should not be used for urgent or sensitive issues.
>
>
>
>
> _____
>
> From: owner-amber.scripps.edu [mailto:owner-amber.scripps.edu] On Behalf Of
> Ilya Chorny
> Sent: Sunday, December 09, 2007 09:22
> To: amber.scripps.edu
> Subject: Re: AMBER: Re: Amber_ifort_x86_64 test error
>
>
> should I even be using ifort?
>
>
> On Dec 9, 2007 9:13 AM, David A. Case <case.scripps.edu> wrote:
>
>
> On Sun, Dec 09, 2007, Ilya Chorny wrote:
> >
> >
> > cd circ_dna; ./Run.circdna
> >
> > MKL func load error:
> > /data0/software/intel/mkl/10.0.1.014/lib/em64t/libmkl_vml_mc.so: undefined
>
> > symbol: vmlGetErrorCallBack
> > ./Run.circdna: Program error
> > make: *** [test.sander.GB] Error 1
> >
>
>
> Amber 9 was released before Intel had version 10 of the MKL libraries, and
> the calling sequences have been changed. Someone on the list who has access
> to version 10 might be able to post a bug-fix here.
>
> In the meantime, please realize the MKL has very little impact on overall
> performance, and you should probably just un-define your MKL_HOME variable
> and
> re-configure.
>
> ...dac
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber.scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
>
>
>
>
>
>

-- 
  Ilyas Yildirim
  ---------------------------------------------------------------
  = Department of Chemistry      -                              =
  = University of Rochester      -                              =
  = Rochester, NY 14627-0216     - Ph.:(585) 275 67 66 (Office) =
  = http://www.pas.rochester.edu/~yildirim/                     =
  ---------------------------------------------------------------
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Wed Dec 12 2007 - 06:07:13 PST
Custom Search