[petsc-users] problem configuring 3.8.3 with intel mpi

[Smith, Barry F.]

> On Mar 22, 2018, at 10:22 AM, Klaij, Christiaan <C.Klaij at marin.nl> wrote:
> 
> Well, that's a small burden.
> 
> By the way, the option --with-cc=mpicc together with export
> I_MPI_CC=icc doesn't work together with --download-hdf5, while it
> works --with-cc=mpiicc and no export. Guess the export doesn't
> make it into the hdf5 configure.
> 
> Any plans for adding --download-cgns and have it work with HDF5
> and fortran enabled? That would ease my burden.

   Hmm, there is a  --download-cgns that works with HDF5. Perhaps support Fortran could be added to it.

> 
> Chris
> 
> 
> dr. ir. Christiaan Klaij  | Senior Researcher | Research & Development
> MARIN | T +31 317 49 33 44 | mailto:C.Klaij at marin.nl | http://www.marin.nl
> 
> MARIN news: http://www.marin.nl/web/News/News-items/Frequency-of-spill-model-for-area-risk-assessment-of-shipsource-oil-spills-in-Canadian-waters-1.htm
> 
> ________________________________________
> From: Satish Balay <balay at mcs.anl.gov>
> Sent: Thursday, March 22, 2018 4:03 PM
> To: Klaij, Christiaan
> Cc: petsc-users
> Subject: Re: [petsc-users] problem configuring 3.8.3 with intel mpi
> 
> Should mention:
> 
> one nice thing about --with-mpi-dir option is - it attempts to pick up
> related/compatible compilers - i.e MPI_DIR/bin/{mpicc,mpicxx,mpif90}
> 
> When one uses --with-cc=mpicc --with-fc=mpif90 --with-cxx=mpicxx - the
> burden on providing related/compatible compilers is on the user - i.e
> more error cases with configure using unrelated/incompatible mpi
> compilers..
> 
> Will try to add support for
> --with-mpi-dir=/opt/intel/compilers_and_libraries_2017.1.132/linux/mpi
> - and then prefer mpiicc over mpicc [in both with-mpi-dir check and
> PATH check?]
> 
> 
> Satish
> 
> On Thu, 22 Mar 2018, Klaij, Christiaan wrote:
> 
>> Guess I shouldn't have read the intel manual :-)
>> 
>> Lesson learned, no more guessing. And I sure appreciate all the
>> effort you are putting into this, especially the option to
>> configure and build other packages in a consistent way with the
>> --download-package option.
>> 
>> Chris
>> 
>> 
>> dr. ir. Christiaan Klaij  | Senior Researcher | Research & Development
>> MARIN | T +31 317 49 33 44 | mailto:C.Klaij at marin.nl | http://www.marin.nl
>> 
>> MARIN news: http://www.marin.nl/web/News/News-items/SSSRIMARIN-seminar-May-18-2018-Shanghai.htm
>> 
>> ________________________________________
>> From: Satish Balay <balay at mcs.anl.gov>
>> Sent: Thursday, March 22, 2018 3:14 PM
>> To: Klaij, Christiaan
>> Cc: petsc-users
>> Subject: Re: [petsc-users] problem configuring 3.8.3 with intel mpi
>> 
>> Sure - they have so many ways of tweaking things - so one can't things
>> will be the same across multiple users.
>> 
>> Most users dont' set these env variables [thats extra work - and not
>> the default behavior anyway] - so they rely on the default behavior -
>> i.e use mpicc for gnu and mpiicc for intel compilers.
>> 
>> [i.e the guesses picked up by configure might not always be what the
>> user wants - so its best to tell configure not to guess - and tell it
>> exactly what one wants it to use. And there might be scope to improve
>> guessing in favor of most common use cases.]
>> 
>> Satish
>> 
>> On Thu, 22 Mar 2018, Klaij, Christiaan wrote:
>> 
>>> Fair enough.
>>> 
>>> As a side note: if you want intel compilers, there's no need to
>>> specify mpiicc, intel says to export I_MPI_CC=icc, which gives
>>> 
>>> $ which mpicc
>>> /opt/intel/compilers_and_libraries_2017.1.132/linux/mpi/intel64/bin/mpicc
>>> $ mpicc -v
>>> mpiicc for the Intel(R) MPI Library 2017 Update 1 for Linux*
>>> Copyright(C) 2003-2016, Intel Corporation.  All rights reserved.
>>> icc version 17.0.1 (gcc version 4.8.5 compatibility)
>>> 
>>> (otherwise they wrap the gnu compilers by default, baffling)
>>> 
>>> Chris
>>> 
>>> 
>>> dr. ir. Christiaan Klaij  | Senior Researcher | Research & Development
>>> MARIN | T +31 317 49 33 44 | mailto:C.Klaij at marin.nl | http://www.marin.nl
>>> 
>>> MARIN news: http://www.marin.nl/web/News/News-items/Frequency-of-spill-model-for-area-risk-assessment-of-shipsource-oil-spills-in-Canadian-waters-1.htm
>>> 
>>> ________________________________________
>>> From: Satish Balay <balay at mcs.anl.gov>
>>> Sent: Thursday, March 22, 2018 2:54 PM
>>> To: Klaij, Christiaan
>>> Cc: petsc-users
>>> Subject: Re: [petsc-users] problem configuring 3.8.3 with intel mpi
>>> 
>>> On Thu, 22 Mar 2018, Klaij, Christiaan wrote:
>>> 
>>>> OK, configure works with the option --with-cc=/path/to/mpicc,
>>>> thanks!
>>>> 
>>>> I'm not sure you are right about checking --with-mpi-dir/bin,
>>> 
>>> There are 2 types of options to configure:
>>> - tell configure what to use - i.e do not guess [like --with-cc=mpicc']
>>> - tell configure to *guess* with partial info [like --with-package-dir=/path/to/foo/bar]
>>> 
>>> When configure has to guess - it will never be perfect.
>>> 
>>> Sure - perhaps the guessing of --with-mpi-dir can be improved - but it will never be perfect.
>>> 
>>>> because, at least with this install, the binaries are located in
>>>> --with-mpi-dir/intel64/bin:
>>>> 
>>>> $ which mpicc
>>>> /opt/intel/compilers_and_libraries_2017.1.132/linux/mpi/intel64/bin/mpicc
>>> 
>>> So the more appropriate thing here [wrt petsc configure] is
>>> --with-mpi-dir=/opt/intel/compilers_and_libraries_2017.1.132/linux/mpi/intel64
>>> 
>>> And then - what if you want to use intel compilers and not gnu
>>> compilers [with intel mpi]? - you have to specify mpiicc - and not
>>> mpicc - so its best to prefer 'no-guess' options like --with-cc.
>>> [with-mpi-dir is usually convinent - but not always suitable]
>>> 
>>> Satish
>>> 
>>>> 
>>>> Perhaps something to do with intel supporting both 32 and
>>>> 64bit. Anyway, our sysadmin didn't change any of these locations,
>>>> they just run the intel installer and this is what we got.
>>> 
>> 
>> 
>