[CP2K-user] [CP2K:10763] Re: cp2k build intel compilers

[贾建峰]

I also have the same error with intel 2018u3. With the newest intel 2019, it works well. 






At 2018-09-20 14:20:46, "Stanislav Šimko" <stanisl... at gmail.com> wrote:

I tried gcc 5.2 (newest available), but compilation didn't succeed. I may look more into it.

On Wednesday, September 19, 2018 at 10:51:32 PM UTC+2, Alfio Lazzaro wrote:
Probably all is too much...
You can use 


-check bounds -check pointers


(see ifort --help for more options).


At this point, I don't have any other suggestion. As a last comment, is there any reason why you want to use the intel compiler? CP2K shows good performance with GCC too (linking with MKL)...

Il giorno mercoledì 19 settembre 2018 20:46:29 UTC+2, Stanislav Šimko ha scritto:
I tried the "-check all -g" flag and I got huge error log that I can not interpret. Do you have any idea what could be going on?

On Wednesday, September 19, 2018 at 8:01:08 PM UTC+2, Alfio Lazzaro wrote:
Sorry, just to make sure, are you using the 17.0.4?


From your stacktrace:


cp2k.popt          000000000D7A6BDD  Unknown               Unknown  Unknown
libpthread-2.17.s  00002AD8E6CFB680  Unknown               Unknown  Unknown
cp2k.popt          0000000007411331  dbcsr_api_mp_dbcs         795  dbcsr_api.F
cp2k.popt          000000000567834C  cp_dbcsr_operatio         944  cp_dbcsr_operations.F


It sees to me that this is a bug with some version of the Intel compiler, namely with Optional Pointer dummy arguments...
Second question, have you tried to run the CP2K regtests?
One more suggestion: you can use the flags "-check all -g", maybe you can get a better error message...


Il giorno mercoledì 19 settembre 2018 18:05:46 UTC+2, Stanislav Šimko ha scritto:
Thank you Alfio for your suggestions. Unfortunately, it is still not working - I compiled only support for fftw3, mkl, mpi (and even without mpi) and I still get the same error. I tried contacting the system admins and I shall see whether they have more luck compiling it. Nevertheless, I'm still open to suggestions.


Thank you.


On Sunday, September 16, 2018 at 9:32:13 PM UTC+2, Alfio Lazzaro wrote:
Hello  Stanislav,
I would suggest starting with a minimal CP2K installation, i.e. without any major library, lust MKL (with FFTW) and IntelMPI.
An arch file would be:


CC       = cc
CPP      = 
FC       = mpiifort 
LD       = mpiifort 
AR       = ar -r


CPPFLAGS =

DFLAGS   = -D__MKL -D__FFTW3 -D__parallel \
   -D__SCALAPACK
CFLAGS   = $(DFLAGS) 
MKLROOT = <path to MKL>
FCFLAGS  = $(DFLAGS) -O2 -g -traceback -fpp -free \
           -I$(MKLROOT)/include -I$(MKLROOT)/include/fftw
FCFLAGS2  = $(DFLAGS) -O0 -g -traceback -fpp -free \
           -I$(MKLROOT)/include -I$(MKLROOT)/include/fftw
LDFLAGS  = $(FCFLAGS) -static-intel 
LDFLAGS_C = $(FCFLAGS) -static-intel -nofor_main
LIBS     = $(MKLROOT)/lib/intel64/libmkl_scalapack_lp64.a \
-Wl,--start-group  $(MKLROOT)/lib/intel64/libmkl_intel_lp64.a \
$(MKLROOT)/lib/intel64/libmkl_sequential.a \
$(MKLROOT)/lib/intel64/libmkl_core.a \
$(MKLROOT)/lib/intel64/libmkl_blacs_intelmpi_lp64.a -Wl,--end-group \
-lpthread -lm


# Required due to memory leak that occurs if high optimisations are used
mp2_optimize_ri_basis.o: mp2_optimize_ri_basis.F
$(FC) -c $(FCFLAGS2) $<


Concerning the compiler, please take a look at this page


https://www.cp2k.org/dev:compiler_support



Intel 17.0.2 is not tested by CP2K people, so I'm not sure if it works. It would be better to use 17.0.4 or 18.0.3. They are both well tested...
The best would be to run the CP2K regression test.


Assuming that everything works fine, then I would move to include other libraries (libxsmm, ELPA, LIBXC, Libint).


Cheers,


Alfio








Il giorno domenica 16 settembre 2018 01:33:30 UTC+2, Stanislav Šimko ha scritto:
Well, it's unfortunately getting a lot messier. I got cp2k compiled on 2 machines in the end, but I get bad results. On both machines, I use intel 17.0.2 compilers. Results on both machines are different (and presumably wrong) from each other. I also compared results of the input file to results obtained by an older versions (4.1 compiled with gcc on one machine, 3.0 compiled with some intel compilers on the second machine; these two installations gave same results). All energies except the self energy is different:
"coorect"
  Overlap energy of the core charge distribution:               0.00000310201993
  Self energy of the core charge distribution:              -1007.31897988265303
  Core Hamiltonian energy:                                    295.26625363987614
  Hartree energy:                                             418.52562360415197
  Exchange-correlation energy:                               -101.48179826891138


  Total energy:                                              -395.00889780551631


  outer SCF iter =    1 RMS gradient =   0.00E+00 energy =       -395.0088978055
  outer SCF loop converged in   1 iterations or   15 steps




 ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.):             -395.008897805517336


wrong output files are attached. I'm now trying to find out how to test LIBINT, LIBXC, LIBXSMM and ELPA. Haven't had much luck yet.


Thank you for any suggestions.

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns... at googlegroups.com.
To post to this group, send email to cp... at googlegroups.com.
Visit this group at https://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20180922/d215368c/attachment.htm>