[CP2K:115] Re: CP2K compilation errors

Matthias Krack kr... at phys.chem.ethz.ch
Sat Jun 2 09:43:24 UTC 2007


Eric,

does your ScaLAPACK build include the toolslib?
I always use the lib target

lib: toolslib pblaslib redistlib scalapacklib

in the ScaLAPACK makefile for the library build.

Matthias


On Sat, 2 Jun 2007, Eric Shamay wrote:

> 
> The final compilation of all the libraries went through just fine.
> I've managed to get the linking process down in the arch (popt) file
> to the point where I'm only getting reference errors from scalapack
> (they look like the following):
> 
> /home/users/eshamay/tmp/SCALAPACK/libscalapack.a(pdlase2.o)(.text
> +0x1b00): In function `pdlase2_':
> : undefined reference to `dlaset_'
> /home/users/eshamay/tmp/SCALAPACK/libscalapack.a(pzlarfb.o)(.text
> +0x5440): In function `pzlarfb_':
> : undefined reference to `zlacpy_'
> /home/users/eshamay/tmp/SCALAPACK/libscalapack.a(pzlarfb.o)(.text
> +0x5460): In function `pzlarfb_':
> : undefined reference to `zlaset_'
> /home/users/eshamay/tmp/SCALAPACK/libscalapack.a(pzlarft.o)(.text
> +0x3b80): In function `pzlarft_':
> : undefined reference to `zlaset_'
> /home/users/eshamay/tmp/SCALAPACK/libscalapack.a(pzlarft.o)(.text
> +0x3ba0): In function `pzlarft_':
> : undefined reference to `zlacgv_'
> /home/users/eshamay/tmp/SCALAPACK/libscalapack.a(pdlarfb.o)(.text
> +0x5140): In function `pdlarfb_':
> : undefined reference to `dlacpy_'
> /home/users/eshamay/tmp/SCALAPACK/libscalapack.a(pdlarfb.o)(.text
> +0x5160): In function `pdlarfb_':
> : undefined reference to `dlaset_'
> /home/users/eshamay/tmp/SCALAPACK/libscalapack.a(pdlarft.o)(.text
> +0x3600): In function `pdlarft_':
> : undefined reference to `dlaset_'
> /home/users/eshamay/tmp/SCALAPACK/libscalapack.a(pzlarf.o)(.text
> +0x39c0): In function `pzlarf_':
> : undefined reference to `zlaset_'
> /home/users/eshamay/tmp/SCALAPACK/libscalapack.a(pzlarfc.o)(.text
> +0x39c0): In function `pzlarfc_':
> : undefined reference to `zlaset_'
> /home/users/eshamay/tmp/SCALAPACK/libscalapack.a(pdlarf.o)(.text
> +0x3780): In function `pdlarf_':
> : undefined reference to `dlaset_'
> 
> I can't find any info on the dlaset_, zlaset_, etc. functions, or what
> library I need to link to in addition. Also, I want to note for future
> compilers the linking order I came up with, so below is my updated
> popt file:
> 
> ########################################################3
> # using openmpi 1.2
> #       ATLAS 3.6.0
> #       mpiBLACS 1.1 with patch03
> #       scalapack 1.8.0
> #
> PERL     = perl
> CC       = mpicc
> CPP      = cpp
> FC       = mpif90 -FR
> FC_fixed = mpif90 -FI
> LD       = mpif90
> AR       = ar -r
> ###########
> # Here are the locations of the newly built libraries
> ###########
> SCALAPACK   = $(HOME)/tmp/SCALAPACK
> ATLASLIB    = $(HOME)/tmp/ATLAS/lib/Linux_IA64Itan2_16
> BLACSLIB    = $(HOME)/tmp/BLACS/LIB
> ###########
> DFLAGS   = -D__INTEL -D__FFTSG -D__FFTW3 -D__parallel -D__BLACS -
> D__SCALAPACK -I$(FFTW)/include
> CPPFLAGS = -C $(DFLAGS) -P -traditional
> FCFLAGS  = -O2 -Vaxlib -tpp2
> LDFLAGS  = $(FCFLAGS) -L$(FFTW)/lib -L$(MPI)/lib -L$(BLACSLIB) -L$
> (ATLASLIB) -L$(SCALAPACK)
> ###########
> # These use the home-built libraries
> LIBS     = -lscalapack $(BLACSLIB)/blacs_MPI-LINUX-0.a $(BLACSLIB)/
> blacsCinit_MPI-LINUX-0.a $(BLACSLIB)/blacsF77init_MPI-LINUX- 0.a $
> (BLACSLIB)/blacs_MPI-LINUX-0.a -llapack -lcblas -lf77blas -latlas -
> lblas -lg2c -lguide -lmpi -lfftw3
> ###########
> 
> OBJECTS_ARCHITECTURE = machine_intel.o
> 
> 
> Note that the call to blacs_MPI-LINUX-0.a happens before and after the
> other BLACS library files. Also, the order of the calls to lapack and
> blas is important: -llapack -lcblas -lf77blas -latlas -lblas
> 
> Anyhow, any help with the undefined reference errors would be greatly
> appreciated!
> Thanks,
> ~Eric
> 
> 
> 
> 
> 
> On Jun 1, 3:39 pm, akohlmey <akoh... at gmail.com> wrote:
> > On Jun 1, 6:17 pm, Eric Shamay <eric.... at gmail.com> wrote:
> >
> > > The errors do point to BLACS, so I hope this will work. I didn't
> > > realize that I'd have to go back in the recompiling-tree and redo each
> > > component everytime I update a lower component (in this case, mpi).
> >
> > the question is always whether that are changes in the binary
> > interface, i.e. the ABI. also you have to make sure that your
> > configuration is correct.
> >
> > > Thanks for the heads up, I'll take note of what each was compiled
> > > against for records from now on.
> >
> > > Why is it that cp2k is so touchy with regards to minor version changes
> > > in the mpi routines? Do mpi libraries deprecate calls from earlier
> > > versions?
> >
> > the problem is not a problem of cp2k directly (unless you want to
> > blame
> > cp2k for using a larger subset of the functionality of the underlying
> > libraries and language than other packages), but in the fact, that
> > openmpi
> > uses shared libraries and thus you have to always make sure, that you
> > use a set of libraries, that match the definitions of the header/
> > include files.
> >
> > the best strategy is in fact, to always build everything from ground
> > up
> > until you have a consistent set. it helps to script the build process,
> > so that you know what set of flags you were using where.
> >
> > just mix and matching different libraries can work, but you'd never
> > know why something was working (or not).
> >
> > cp2k _is_ a large and complex package and that shows in the build
> > process. however, once you have a working setup, keeping up-to-date
> > is rather simple.
> >
> > cheers,
> >    axel.
> 
> 
> > 
> 

Dr. Matthias Krack
Computational Science
Department of Chemistry and Applied Biosciences
ETH Zurich
USI-Campus, via G. Buffi 13
6900 Lugano
Switzerland

Phone: +41 (0)58 666 48 05 (direct)
       +41 (0)58 666 48 00 (secretary)
Fax:   +41 (0)58 666 48 17

Email: kr... at phys.chem.ethz.ch
URL:   http://www.phys.chem.ethz.ch



More information about the CP2K-user mailing list