Regression testing CP2K 2.6.0

Matthias Krack matthia... at
Sat Feb 21 17:38:18 UTC 2015

Dear Justin,

the precompiled CP2K binaries 
<> are compiled 
using the corresponding arch files Linux-x86-64-gfortran-static.sopt 
and Linux-x86-64-gfortran-static.ssmp 
(GNU compiler 4.9 and MPICH 3.1.2). The libs have been downloaded as source 
code and compiled as follows:

   - blas/lapack 3.5.0: gcc/gfortram -O1 -fno-omit-frame-pointer -g 
   -mtune=generic (change  -O1 to -O0 for NOOPT)
   - blacs/scalapack 2.0.2: mpicc/mpif90 -O1 -fno-omit-frame-pointer -g 
   -mtune=generic (change -O1 to -O0 for NOOPT)
   - fftw 3.3: ./configure FFLAGS="-O1 -fno-omit-frame-pointer -g 
   -mtune=generic" CFLAGS="-O1 -fno-omit-frame-pointer -g -mtune=generic" 
   --prefix=$PREFIX --libdir=$PREFIX/lib64 --enable-shared --enable-static 
   --enable-threads --enable-mpi
   - libint 1.1.4: ./configure --prefix=$PREFIX --libdir=$PREFIX/lib64 
   FFLAGS="-O1 -fno-omit-frame-pointer -g -mtune=generic" --with-cc="$CC" 
   --with-cc-optflags="-O1 -fno-omit-frame-pointer -g -mtune=generic" 
   --with-cxx="$CXX" --with-cxx-optflags="-O1 -fno-omit-frame-pointer -g 
   - libxc 2.2.0: ./configure FCCPP=cpp --prefix=${PREFIX} 
   --libdir=${PREFIX}/lib64 CFLAGS="-O1 -fno-omit-frame-pointer -g 
   -mtune=generic" FCFLAGS="-O1 -fno-omit-frame-pointer -g -mtune=generic



On Saturday, 21 February 2015 08:00:16 UTC+1, Justin Oelgoetz wrote:
> All,
> I hadn't heard anything so I thought I'd restate and bump this request 
> again.
> I'm in the process of trying to produce a reasonably optimized cp2k binary 
> for  STAMPEDE which will hopefully make some use the Intel Xeon Phi 
> hardware on the machine.  As part of this I need a good baseline set of 
> compilation options I can compare my changes to.  
> My initial binary, compiled with the intel compiler & mkl fails roughly 70 
> regression tests when compared to the Linux binary 
> (cp2k-2.6.0-Linux-x86_64.sopt) which is distributed on the sourceforge 
> site.  I then created a binary with the gnu compiler and mol which also 
> failed the same tests.  I've played with compiler options and have gotten 
> that number down to 15.  Some of these 15 are quite large (some as large as 
> 1E-2).
> To resolve the last 15, I really need to know how the binary being 
> distributed is compiled.  Is it compiled with the 
> "Linux-x86-64-gfortran-regtest.ssmp" arch files in the distribution?  If 
> so, can anyone shed any light on how the libraries cp2k depends on (fftw3, 
> lapack, blas, libxc, libint) are compiled or where they are drawn from?  If 
> it is another set of compilations options, knowing those would be very 
> useful.  Can anyone out there help me?  
> Justin Oelgoetz
> Justin Oelgoetz
> joe... at <javascript:>
> On Sat, Jan 17, 2015 at 9:52 PM, Justin Oelgoetz <joe... at 
> <javascript:>> wrote:
>> All,
>> I'm in the process of testing an installation using the regression tests 
>> provided as part of the source distribution.  I have a handful which are 
>> giving wrong results (versus the precompiled binary, 
>> cp2k-2.6.0-Linux-x86_64.sopt).  I'd like to trace what compilation 
>> flags/libraries cause the larger differences.  Would any of you developed 
>> be able to tell me what compiler options were used to compile cp2k 2.6.0 
>> and information about its dependent libraries (compilation flags, and 
>> versions for libint, libxc, etc).
>> Thanks,
>> Justin
>> --
>> Justin Oelgoetz, Ph.D.
>> Associate Professor & Interim Chair
>> Department of Physics and Astronomy
>> XSEDE Campus Champion
>> Austin Peay State University
>> PO Box 4608
>> Clarksville, TN 37044
>> Office: SSC: B321
>> Phone: (931)221-6158
>> Fax:   (931)221-6129
>> oel... at <javascript:>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the CP2K-user mailing list