[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [condor-users] Interaction of Condor and Threads



De-Wei Yin wrote:
> I am also using Intel compilers (ifc and icc version 7.1) and MKL here
> with Condor and have never been able to condor_compile before with
> Condor 6.4 (on Linux 7.3).
> 
> Which versions of Intel compilers, Condor, and Linux (kernel, glibc) are
> you using, and how do you get condor_compile to work with Intel at all?
> I last tried this quite some time ago and failed, and the Condor team
> told me that they have no plans to support Intel compilers at the moment.

Hi Dewey,

I am using mkl v6.1 on a Red Hat 7.3 box (kernel 2.4.20-28.7 with glibc
2.2.5-42). I think it should also be possible to get things working on a
Red Hat 9 box, but I haven't tried this yet. I'm using condor 6.6.0.
Since 90% of the speedup in my code comes from using the FFT routines form
MKL, not and not from icc, I compile with gcc but link in MKL.

First I did a full install of condor_compile, described at:

http://www.cs.wisc.edu/condor/manual/v6.6/3_10Setting_Up.html#SECTION004103000000000000000

and built my executable with condor_compile gcc linking in libmkl_ia32.a and
libguide.a to get the FFT routines. This failed, as libguide.a needs
libpthread.a, which isn't compatible with condor, so I wrote the dummy library
libqthread.a to fake up the functions that MKL needs (qthread.c was attached
in the last mail I sent).  This seems to work as long as MKL is set to run
without threads, which can be done by setting the environment variable
MKL_SERIAL=TRUE. 

I haven't thoroughly validated the results of this hack yet and I don't know
what the effect on condor is, hence the mail to this list. The results seem
correct and I can checkpoint and resume the code hapily enough. I also have
filed a support request with Intel so see if they can see any possible
problems with MKL using the fake threading functions.

Cheers,
Duncan.

-- 
Duncan Brown                                  University of Wisconsin-Milwaukee
duncan@xxxxxxxxxxxxxxxxxxxx                 Physics Department, 1900 E. Kenwood
http://www.lsc-group.phys.uwm.edu/~duncan              Milwaukee, WI 53211, USA
Condor Support Information:
http://www.cs.wisc.edu/condor/condor-support/
To Unsubscribe, send mail to majordomo@xxxxxxxxxxx with
unsubscribe condor-users <your_email_address>