[petsc-dev] error configuring on KNL

[Mark Adams]

Yes, I submitted jobs before I went to bed last night (in the UK) and it
configure and built. So far so good.
Thanks,

On Fri, Jun 30, 2017 at 11:32 PM, Richard Tran Mills <rtmills at anl.gov>
wrote:

>
>
> On Fri, Jun 30, 2017 at 3:27 PM, Balay, Satish <balay at mcs.anl.gov> wrote:
>
>> On Fri, 30 Jun 2017, Richard Tran Mills wrote:
>>
>> > Hmm. I get the same error when adding "--download-hypre=0" to an
>> otherwise
>> > functional configure script for Cori KNL. Hypre's configure doesn't like
>> > something about cross-compiling for KNL. It might be possible to make
>> this
>> > work by building on the compute nodes, though this is, unfortunately,
>> not
>> > officially supported on Cori -- NERSC staff say that builds are not
>> > guaranteed to work on the compute nodes, though I've managed to make it
>> > work sometimes.
>>
>> Here is (a part of a prior email) regarding installing hypre on theta
>> (using compute nodes).
>>
>
> The approach Satish outlines below sounds like the most reasonable course
> of action we ahve. I wonder if it might help a little bit to try to just
> run the standalone Hypre configuration (using the incantation in the
> configure.log) in the batch script, rather than invoking PETSc's
> configure.py.
>
> --Richard
>
>
>> Satish
>>
>> --------------------------------
>>
>> The alternative is to run configure on the KNL nodes. But this is very
>> slow. We could split up the build into 2 parts - only build some
>> packages on KNL node and the rest on front-end.
>>
>> Ok - I tried "--download-hypre=1 --prefix=/home/balay/soft/hypre-xsdk-0.2.0-rc2"
>> on the KNL node.
>> [this way - I can --with-hypre-dir=/home/balay/soft/hypre-xsdk-0.2.0-rc2
>> for subsequent builds]
>>
>> This crashed halfway with 'Intel License Server' errors. The crash
>> happened after hypre configure completed - so I could complete the
>> build on the frontend.
>>
>> Anyhow - here is the process"
>>
>> caveat: PETSC_ARCH/externalpackages should be populated before doing
>> this - as compute nodes don't have external network access.
>>
>> 1. schedule configure run in the batch queue:
>>
>> balay at thetalogin5:~/petsc.test> cat reconfigure-arch-theta-opt-cray.py
>> #!/usr/bin/python
>> if __name__ == '__main__':
>>   import sys
>>   import os
>>   sys.path.insert(0, os.path.abspath('config'))
>>   import configure
>>   configure_options = [
>>     '--download-hypre=1',
>>     '--prefix=/home/balay/soft/hypre-xsdk-0.2.0-rc2',
>>     '--with-cc=cc',
>>     '--with-clib-autodetect=0',
>>     '--with-cxx=CC',
>>     '--with-cxxlib-autodetect=0',
>>     '--with-debugging=0',
>>     '--with-fc=ftn',
>>     '--with-fortranlib-autodetect=0',
>>     '--with-mpiexec=mpirun',
>>     '--with-memalign=64',
>>     'COPTFLAGS=-g -O3 -fp-model fast -xMIC-AVX512',
>>     'CXXOPTFLAGS=-g -O3 -fp-model fast -xMIC-AVX512',
>>     'FOPTFLAGS=-g -O3 -fp-model fast -xMIC-AVX512',
>>     'LIBS=-lstdc++',
>>     'PETSC_ARCH=arch-theta-opt-cray',
>>   ]
>>   configure.petsc_configure(configure_options)
>>
>> balay at thetalogin5:~/petsc.test> cat x.sh
>> #!/bin/sh
>> aprun -n 1 python reconfigure-arch-theta-opt-cray.py
>>
>> balay at thetalogin5:~/petsc.test> qsub -t 45 -n 1 -A EarlyPerf_theta x.sh
>>
>> <<<<<<<<<<<<<<<<<<
>>
>> 2. wait for job to complete - [or atleast hypre configure to complete].
>>
>> 3. if incomplete - complete the build on frontend - and install hypre
>> libraries.
>>
>> 4. subsequently install PETSc using --with-hypre-dir=/home/balay/s
>> oft/hypre-xsdk-0.2.0-rc2
>> [on the frontend]
>>
>> Satish
>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20170701/ad543060/attachment.html>