[Q-e-developers] Compiling error on Cray XE6 (UNCLASSIFIED)

Gabriele Sclauzero gabriele.sclauzero at epfl.ch
Mon Aug 20 14:22:02 CEST 2012


Dear Giovanni, Filippo, and all

    I've tried to compile QE-5.0.1 and it works for PGI, CCE (Cray), and Intel compilers on Rosa (XE6) at CSCS.
However, I've discarded the latter because it was apparently not possible to link against MKL scalapack (which is a bad thing, since you usually want to study big systems on those machines), since they usually come with OpenMPI and Intel MPI flavors only. The user support at CSCS got in touch with Cray staff on that issue, but they weren't able
to provide a solution. Has any of you succeeded in doing this?
My CCE compiled version is a bit faster than the PGI one, but I did not have any previous experience with these compilers. A member of CSCS staff also managed to compile with both PGI and CCE, and his PGI build was slightly faster than my CCE. So, if you are familiar with PGI, you should probably stick to that.

HTH

Gabriele


Il giorno 20/ago/2012, alle ore 10.43, Giovanni Pizzi ha scritto:
> Dear all,
> just to follow up, indeed my experience on the Cray system rosa.cscs.ch was similar to Filippo's one, i.e. Intel compiler working, CRAY compiler not working. The only difference being that I wasn't able to compile with PGI (probably because I wasn't aware of the IOTK workaround debug flag, and then I stopped trying when the intel compilation worked).
> 
> Giovanni
> 
> 
> On 08/16/2012 05:21 PM, Filippo Spiga wrote:
>> Dear Paolo,
>> 
>> On Aug 16, 2012, at 10:12 PM, Paolo Giannozzi <giannozz at democritos.it> wrote:
>>> On Aug 16, 2012, at 11:51 , Filippo Spiga wrote:
>>> 
>>>> I do not see a any reasonable reason to compile QE using Intel compiler on a CRAY, the PGI compiler (usually the default) works. Moreover it is weird that mpif90 is called because CRAY systems the correct FORTRAN compiler is always "ftn" (that is actually a wrapper to whatever compiler is selected through the modules loaded). Anyway, I am currently using the CRAY XE6 installed at CSCS in Switzerland (MonteRosa) and I simply use the PGI compiler and the default CRAY LibSci.
>>> 
>>> before leaving for vacations, I had a different report by Giovanni Pizzi (CC: here) for
>>> exactly the same machine: various problems with the cray compiler, either in the
>>> compilation phase or in execution. I was surprised because the ftn compiler on cray
>>> has always been working very reliably. Did you need ./configure --disable-wrappers
>>> to prevent ./configure from stopping in the C-to-fortran interface check, by the way?
>> 
>> From my experience no, I never had big or blocking problems with the ftn/cc wrappers. However I have to admit that despite CRAY provide an "uniform and fairly easy" environment, every machine is unique. My experience on CRAY systems can be summarized in this way:
>> - PGI compiler (loading pgi-env) + CRAY LibSci + ftn/cc wrappers  --> ok 
>> - CRAY compiler (loading cray-env) + CRAY LibSci + ftn/cc wrappers  --> compilation fails
>> - PGI compiler (loading pgi-env) + CRAY LibSci + ftn/cc wrappers + CUDA (XK6 machines)  --> ok 
>> - Intel compiler (loading intel-env) + MKL + ftn/cc wrappers + CUDA (XK6 machines)  --> ok (but I tried only few times)
>> - Intel compiler (loading intel-env) + ACML + ftn/cc wrappers + CUDA (XK6 machines)  --> ok (but I tried only few times)
>> 
>> On a CRAY testing machine (so, not a common case) I had to unload some modules ("$ module unload atp totalview-support xt-totalview hss-llm"), This is the only "weird" thing I did to compile and run QE.
>> 
>> I did not mention in my previous email the option "--disable-wrapper: because I was curious to hear from Miller if using PGI compiler (instead of Intel) and "ARCH=crayxt" was enough to solve the problem. Of course if something goes wrong the first thing that I probably would suggest is to try --disable-wrapper but also I would check if there is some error/mistake/typo in the modulefile. That's something that the user support people should do, not the user.
>> 
>> @Giovanni, I wrote an email to Emine with some hints related to Monte Rosa. She has more details for that specific machine.
>> 
>> HTH
>> 
>> Ciao!
>> F.
>>  
>> --
>> Mr. Filippo SPIGA, M.Sc., Ph.D. Candidate 
>> CADMOS - Chair of Numerical Algorithms and HPC (ANCHP)
>> École Polytechnique Fédérale de Lausanne (EPFL)
>> http://anchp.epfl.ch ~ http://filippospiga.me ~ skype: filippo.spiga
>> 
>> «Nobody will drive us out of Cantor's paradise.» ~ David Hilbert
>> 
> 
> 
> -- 
> Giovanni Pizzi
> Post-doctoral Research Scientist
> EPFL STI IMX THEOS
> MXC 319 (Bâtiment MXC)
> Station 12
> CH-1015 Lausanne (Switzerland)
> Phone: +41 21 69 31159
> _______________________________________________
> Q-e-developers mailing list
> Q-e-developers at qe-forge.org
> http://qe-forge.org/mailman/listinfo/q-e-developers


§ Gabriele Sclauzero, EPFL SB ITP CSEA
   PH H2 462, Station 3, CH-1015 Lausanne









§ Gabriele Sclauzero, EPFL SB ITP CSEA
   PH H2 462, Station 3, CH-1015 Lausanne







-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/developers/attachments/20120820/f24078be/attachment.html>


More information about the developers mailing list