neither PGI nor Intel compiler working?

[NUCP2K]

Dear cp2k community,

I just downloaded the latest CVS snapshot and tried to compile the
codes using my once working ARCH files for PGI and Intel compilers on
Linux-x86-64 MPI machines. Unfortunately, none of the two compilers is
working though both can get the codes compiled successfully. It seems
to me that the "cp_fm_cholesky_decompose" is the cause of the runtime
trouble since my testing job always exit abnormally when the
subroutine is called.

Anyone has the experience on successfully compiling and running the
latest cp2k code on Linux PC MPI clusters? The sharing of your ARCH
files is highly appreciated.

Thanks.

Hanning

[Ondrej Marsalek]

Hi,

On Mon, Oct 26, 2009 at 16:46, NUCP2K <chenh...@gmail.com> wrote:
>
> Dear cp2k community,
>
>  I just downloaded the latest CVS snapshot and tried to compile the
> codes using my once working ARCH files for PGI and Intel compilers on
> Linux-x86-64 MPI machines. Unfortunately, none of the two compilers is
> working though both can get the codes compiled successfully. It seems
> to me that the "cp_fm_cholesky_decompose" is the cause of the runtime
> trouble since my testing job always exit abnormally when the
> subroutine is called.

Please note that this can also mean (and in my experience usually
does) that the data were corrupted sometime before this point, but the
problem is only exposed because the matrix entering Cholesky
decomposition is not well formed.

>
>  Anyone has the experience on successfully compiling and running the
> latest cp2k code on Linux PC MPI clusters? The sharing of your ARCH
> files is highly appreciated.

I had trouble with the Intel compiler as well, see the Google group
archive. Other people have reported the same, I don't think anyone has
claimed to have a working build of current CP2K with the Intel
compiler, any version. I have a working (well, passed the tests) build
with gfortran 4.3.3 and Intel MKL, I attach the arch file. It seems
that at least gfortran 4.3.3 is needed if you want to build with
libint, I am not sure what is the limitation without libint (if anyone
does know, please tell me). I use current OpenMPI (now 1.3.3).

Good luck,
Ondrej


>  Thanks.
>
> Hanning
> >
>

[Jörg Saßmannshausen]

Hi all,

unless I am mistaken, libint is c++ code, so the fortran compiler does not
matter. Besides, I have build (albeit an older version) of cp2k with gfortran
4.3.2.
Just wanted to mention that. I have not tried the latest version of cp2k, so
things may have changed here.
Personally, I would not recommend PGI for various reasons. If there are
problems with the build, try and use the g95 compiler (for cp2k only) to
check all the other libraries/stuff and see if the problem persists.

All the best

Jörg

--
*************************************************************
Jörg Saßmannshausen
Research Fellow
University of Strathclyde
Department of Pure and Applied Chemistry
295 Cathedral St.
Glasgow
G1 1XL

email: jorg.sassm...@strath.ac.uk
web: http://sassy.formativ.net

Please avoid sending me Word or PowerPoint attachments.
See http://www.gnu.org/philosophy/no-word-attachments.html

[Ondrej Marsalek]

Hi,

On Mon, Oct 26, 2009 at 17:13, Jörg Saßmannshausen
<jorg.sassm...@strath.ac.uk> wrote:
>
> Hi all,
>
> unless I am mistaken, libint is c++ code, so the fortran compiler does not
> matter.

It is true that it does not matter for the compilation of libint itself.
However, you need to link this code to Fortran code and then the Fortran
compiler does matter. You either need full ISO_C_BINDING support, or an
additional wrapper. For details see the file

cp2k/tools/hfx_tools/libint_tools/README_LIBINT

> Besides, I have build (albeit an older version) of cp2k with gfortran
> 4.3.2.

Was that with libint? I certainly got complaints about support for
function pointers with gfortran 4.3.2.

Best,
Ondrej

[Teodoro Laino]

> It is true that it does not matter for the compilation of libint itself.
> However, you need to link this code to Fortran code and then the Fortran
> compiler does matter. You either need full ISO_C_BINDING support, or an
> additional wrapper. For details see the file
>
> cp2k/tools/hfx_tools/libint_tools/README_LIBINT
>
>
>> Besides, I have build (albeit an older version) of cp2k with gfortran
>> 4.3.2.
>>
>
> Was that with libint? I certainly got complaints about support for
> function pointers with gfortran 4.3.2.
>

that's because you tried the ISO_C_BINDING with gfortran (which was
probably broken or not fully working with that version). Nonetheless, if
the rest of cp2k was properly compiled, there is no reason why with
gfortran you had to be stuck with ISO_C_BINDING. You may have as well
tried the additional cpp wrapper, which does not need any ISO_C_BINDING
capability of the fortran compiler. See:

cp2k/tools/hfx_tools/libint_tools/README_LIBINT

so quite probably you may have been able as well to compile cp2k with
4.3.2, even with libint (of course relying on the absence of other
compiler bugs, which may have potentially misinterpreted other parts of
cp2k: the function pointer is only required in the ISO_C_BINDING
instructions)

Teo

[Ondrej Marsalek]

Teo,

I thought I did use the wrapper, but perhaps there was a mistake in my
setup. I will recheck and retry.

Thanks,
Ondrej

[NUCP2K]

Hi guys,

With your generous helps, I just got the latest CVS snapshot on
running on Macintosh OSX using gfortran compiler. I am hoping the code
work for Intel or PGI compilers soon.....

Thanks.

Hanning

ARCH file for my Intel Mac:

CC = cc
CPP =
FC = gfortran
LD = gfortran
AR = ar -r
CPPFLAGS = -D__GFORTRAN -D__FFTSG
FCFLAGS = -cpp -O3 -fopenmp -ffree-form $(CPPFLAGS)
FCFLAGS2 = -cpp -O3 -fopenmp -ffree-form $(CPPFLAGS)
LDFLAGS = $(FCFLAGS)
LIBS = -llapack -lblas

OBJECTS_ARCHITECTURE = machine_gfortran.o




On Oct 26, 11:36 am, Ondrej Marsalek <ondrej.marsa...@gmail.com>
wrote:

[Jörg Saßmannshausen]

Hi Ondrej,

yes, that cp2k code I compiled with the mentioned gfortran compiler was with
libint (version libint-1.1.4).
I know you need the wrappter for some compilers, but I took that as granted.

All the best

Jörg

[Urban]

There was a known issue that resulted in an error in
cp_fm_cholesky_decompose when run with certain processor numbers. That
issue, along with the Intel compiler issues, should have been fixed
with the commits of Oct. 27.

[Hanning Chen]

Urban,

  Awesome ! I just got the latest CVS working with gfortran/mvapich2 on x86-64/Linux, and will test it with Intel compilers.

  Thanks.

Hanning

[Ondrej Marsalek]

Hi,

this sounds promising. I would like to ask which version of the Intel
compiler to use. The readme file for libint use lists 11.1.056 as
having full ISO_C_BINDING support, so presumably it works. On the
other hand, this page:

http://software.intel.com/en-us/articles/build-cp2k-using-intel-fortran-compiler-professional-edition/

says that 11.1 should not be used. I do realize that that information
is from July, but it is still there.

Any suggestions appreciated,
Ondrej

[Ondrej Marsalek]

A quick update. Using the libint wrapper, I have compiled with
11.0.084 and a quick test says that at least one of the previously
failing tests now does work.

The question that now remains - is there any point in trying to get
11.1.x working? Any experience?

Ondrej

[Ilya]

Hello Ondrej.

I have ifort 11.0, but I'm still getting Nan's in the output.

Could you post your arch file used for 11.0.084, please.

Thanks.

Ilya

On Oct 30, 12:33 am, Ondrej Marsalek <ondrej.marsa...@gmail.com>
wrote:

> A quick update. Using the libint wrapper, I have compiled with
> 11.0.084 and a quick test says that at least one of the previously
> failing tests now does work.
>
> The question that now remains - is there any point in trying to get
> 11.1.x working? Any experience?
>
> Ondrej
>
> On Thu, Oct 29, 2009 at 19:38, Ondrej Marsalek
>

> <ondrej.marsa...@gmail.com> wrote:
> > Hi,
>
> > this sounds promising. I would like to ask which version of the Intel
> > compiler to use. The readme file for libint use lists 11.1.056 as
> > having full ISO_C_BINDING support, so presumably it works. On the
> > other hand, this page:
>

> >http://software.intel.com/en-us/articles/build-cp2k-using-intel-fortr...

>
> > says that 11.1 should not be used. I do realize that that information
> > is from July, but it is still there.
>
> > Any suggestions appreciated,
> > Ondrej
>

[Ondrej Marsalek]

Hi Ilya,

running the full test suite, I got some problem in RT propagation
(that printed some warning indicating that they are expected to fail)
and a few problems in HFX, which I have not investigated yet.
Otherwise, it looks fine.

The arch file is attached. If you have any suggestions for further
tuning, please let me know.

Best,
Ondrej

[flo]

Hi Ondrej,

could you sent me the output files from RT propagation. I will have a
look what's going wrong. Just in case these are only the EMD related
tests, could you do me a favor and try to compile the
rt_propagation_utils.F file with -O0 and see if the problems persist.

Thanks
Flo

On Nov 3, 12:07 pm, Ondrej Marsalek <ondrej.marsa...@gmail.com> wrote:
> Hi Ilya,
>
> running the full test suite, I got some problem in RT propagation
> (that printed some warning indicating that they are expected to fail)
> and a few problems in HFX, which I have not investigated yet.
> Otherwise, it looks fine.
>
> The arch file is attached. If you have any suggestions for further
> tuning, please let me know.
>
> Best,
> Ondrej
>

>  Linux-x86-64-intel.popt
> 1KViewDownload

[Ondrej Marsalek]

Hi Flo,

please see the attached file, it was all only in EMD. There are some
others for HFX and EMD combined, but as I said, I'll try looking at
HFX later. I'll recompile without opti and let you know.

Ondrej

[Ondrej Marsalek]

Hi again,

disabling the optimizations did help. Now I get these failures:

$ grep FAIL regtest-03.log
QS/regtest-hfx/H2O-hfx-emd.inp RUNTIME FAIL
QS/regtest-hfx/H2O-hfx-emd-restart.inp RUNTIME FAIL
QS/regtest-epr-1/NO2-EPR-3.inp RUNTIME FAIL
QS/regtest-epr-2/NO2-EPR-1.inp RUNTIME FAIL
FARMING/regtest-1/farming-4.inp RUNTIME FAIL
QS/regtest-hfx-periodic/H2O-trunc-auto.inp RUNTIME FAIL
QS/regtest-hfx-periodic/H2O-trunc-auto-md.inp RUNTIME FAIL
QS/regtest-hfx-periodic/CH3-trunc-auto.inp RUNTIME FAIL
QS/regtest-hfx-periodic/CH3-trunc-auto-md.inp RUNTIME FAIL
number of FAILED tests 9

I run the tests in parallel with 2 CPUs. Are there any more of these
that are expected to fail (except for farming, that's clear)? I'll try
to find some time to have a look at whatever should work and does not.

Best,
Ondrej

On Tue, Nov 3, 2009 at 12:12, flo <fsch...@pci.uzh.ch> wrote:
>

[Ondrej Marsalek]

Dear Naga,

I attach my Intel arch file, but note that I do not claim this
produces a useful binary. See the whole of this thread for details. I
do not have any newer experience than what was already posted. I use
gfortran instead, as it seems to be a safer choice.

Regards,
Ondrej

PS: Please keep the mail in the list, for future reference.

On Fri, Feb 19, 2010 at 12:39, nagaraj <k.nag...@gmail.com> wrote:
> Dear Onddrej,
> Hi,
>
> I used Intel compiler (11.1 trial version)  & intel MPI library for
> building CP2K as a parallel version. The compilation was successfull.
> But while running the CP2K application hangs at some point. The
> details are,
> max_taylor : 4
>
> mixed_precision : F
>
> ----------------------------------- OT
> ---------------------------------------
>
> Step Update method Time Convergence Total energy Change
> -----------------------------------------------------------------
> -------------
>
>
> Alternatively, if I build with Open MPI & Intel compilers using
> similar configuration script, the application fails as follows:
>
>
> *
> *** 04:35:45 ERRORL2 in cp_fm_cholesky:cp_fm_cholesky_decompose
> processor ***
> *** 0 err=-300 condition FAILED at line 92 ***
> *
>
> ===== Routine Calling Stack =====
>
> 7 cp_fm_cholesky_decompose
> 6 make_basis_sm
> 5 calculate_first_density_matrix
> 4 scf_env_initial_rho_setup
> 3 init_scf_run
> 2 qs_energies_scf
> 1 CP2K
> CP2K| condition FAILED at line 92
> CP2K| Abnormal program termination, stopped by process number 0
>
> Are you tested with Intel compilers? If test is success, please send
> me the architecture files.
>
>
> Kindly help us to proceed further.
>
> Thanks in advance.
> Naga
>
> On Nov 3 2009, 4:07 pm, Ondrej Marsalek <ondrej.marsa...@gmail.com>

>>  Linux-x86-64-intel.popt
>> 1KViewDownload- Hide quoted text -
>>
>> - Show quoted text -

[Ondrej Marsalek]

And... here goes the arch file. Sorry...