Build CP2K using Intel Fortran compiler Professional Edition

Introduction : 

CP2K (http://cp2k.berlios.de/) is a freely available (GPL) program, written in Fortran, to perform atomistic and molecular simulations of solid state, liquid, molecular and biological systems. It provides a general framework for different methods such as e.g. density functional theory (DFT) using a mixed Gaussian and plane waves approach (GPW), and classical pair and many-body potentials.

Version : 

1.08


Obtaining Source Code :

CP2K stable version
- the cp2k-2_1-branch: ftp://ftp.berlios.de/pub/cp2k/cp2k-2_1-branch.tar.gz
- the cp2k-2_2-branch: ftp://ftp.berlios.de/pub/cp2k/cp2k-2_2-branch.tar.gz

CP2K development version
- (nightly snapshot): ftp://ftp.berlios.de/pub/cp2k/cp2k.tar.gz


Prerequisites :

CP2K requires that you already have installed the Fortran compiler, math library with FFTW 2.x/3.x support.  We suggest using both Intel Fortran compiler and Intel Math Kernel Library for better performance.


Configuration Set Up :


n/a


Source Code Changes :

n/a


Building the Application :

Here we show the detail on how to build the CP2K source with Intel Fortran Compiler for Intel 64 platform.

Part I. Intel Fortran Composer XE 2011


1) Download and install Intel Fortran Composer XE 2011 (first release version 2011.0.084).
Make sure you select to install both Intel Fortran compiler and Intel Math Kernel Library (MKL).  The default installation directory is /opt/intel/composerxe-2011/ .

2) Build FFTW version 3.x wrappers library for Intel MKL Library
From Intel MKL version 10.2 and later, FFTW3 interface for Intel Fortran compiler is pre-built and ntegrated in Intel MKL core libraries.  See /en-us/articles/intel-mkl-main-libraries-contain-fftw3-interfaces

Source code is still available to create wrappers for use with other compilers.  Here is example to build Fortran version by gnu compiler.

] cd /opt/intel/composerxe-2011/mkl/interfaces/fftw3xf

] make libintel64 compiler=gnu

3) Untar the CP2K.tar.gz, and modify configuration file “Linux-x86-64-intel.sopt” in cp2k/arch directory, which is targeting for Intel 64 platform.

Here is an example of Linux-x86-64-intel.sopt for reference:

INTEL_MKL = /opt/intel/composerxe-2011/mkl
INTEL_INC = $(INTEL_MKL)/include/fftw
INTEL_LIB = $(INTEL_MKL)/lib/intel64

LIBINT_PATH = /usr/lib/libint-1.1.4
LIBINT_LIB  = $(LIBINT_PATH)/lib/libderiv.a $(LIBINT_PATH)/lib/libint.a -lstdc++

FC = ifort
LD = ifort
AR = xiar -r

DFLAGS = -D__INTEL -D__FFTSG -D__FFTW3 -D__FFTMKL -D__LIBINT
CPPFLAGS = -C -traditional $(DFLAGS) -I$(INTEL_INC)
FCFLAGS = $(DFLAGS) -I$(INTEL_INC) -O2 -xHost -heap-arrays 64 -fpp -free
LDFLAGS = $(FCFLAGS)

#LIBS = -mkl
LIBS = -L$(INTEL_LIB) -mkl  $(LIBINT_LIB)

OBJECTS_ARCHITECTURE = machine_intel.o
Notice: The options (-O2 -xHost) are available for both Intel® and non-Intel microprocessors but it may result in more optimizations for Intel microprocessors than for non-Intel microprocessors.


4) Setup the environment of Intel Fortran compiler, then build CP2K source:

] source /opt/intel/composerxe-2011/bin/compilervars.sh intel64

] cd cp2k/makefiles/

] make ARCH=Linux-x86-64-intel VERSION=sopt


Part II. Intel Fortran Compiler Professional Edition (version 11.0 / 11.1)

Please follow the steps listing below:.

1) Download and install Intel Fortran compiler Professional Edition version.
Make sure you select to install both Intel Fortran compiler and Intel Math Kernel Library (MKL). For example, the default installation directory is under /opt/intel/Compiler/11.0/083 or /opt/intel/Compiler/11.1/072.

2) Build FFTW version 3.x wrappers library for Intel MKL Library

Note: Please skip this step if you are using Intel Fortran compiler version 11.1, or Intel MKL libraries version 10.2 and later.

The source code for the wrappers and makefiles with the wrapper list files for Fortran are located in the /interfaces/fftw3xf sub-directories in the Intel MKL directory. Please use Intel C++ compiler (or GCC) to generate the wrappers library (libffw3xf_intel.a, or libfftw3xf_gnu.a):

] cd /opt/intel/Compiler/11.0/083/mkl/interfaces/fftw3xf

] make libem64t compiler=intel (or make libem64t compiler=gnu)

3) Untar the CP2K.tar.gz, and modify configuration file “Linux-x86-64-intel.sopt” in cp2k/arch directory, which is targeting for Intel 64 platform.

Here is an example of Linux-x86-64-intel.sopt for reference:

INTEL_MKL = /opt/intel/Compiler/11.0/083/mkl
INTEL_INC = $(INTEL_MKL)/include/fftw
INTEL_LIB = $(INTEL_MKL)/lib/em64t

FC = ifort
LD = ifort
AR = xiar -r

DFLAGS = -D__INTEL -D__FFTSG -D__FFTW3 -D__FFTMKL
CPPFLAGS = -C -traditional $(DFLAGS) -I$(INTEL_INC)
FCFLAGS = $(DFLAGS) -I$(INTEL_INC) -O2 -xHost -heap-arrays 64 -fpp -free
LDFLAGS = $(FCFLAGS)

# If you are using ifort verison 11.1 with MKL version 10.2 and later, Intel MKL libraries and FFTW interfaces will be linked automatically with -mkl option
# LIBS = -L$(INTEL_LIB) -mkl

LIBS = -L$(INTEL_LIB) -lfftw3xf_intel -lmkl_intel_lp64 -lmkl_sequential -lmkl_core

OBJECTS_ARCHITECTURE = machine_intel.o
Notice: The options (-O2 -xHost) are available for both Intel® and non-Intel microprocessors but it may result in more optimizations for Intel microprocessors than for non-Intel microprocessors.


4) Setup the environment of Intel Fortran compiler, then build CP2K source:

] source /opt/intel/Compiler/11.0/083/bin/ifortvars.sh intel64

] cd cp2k/makefiles/

] make ARCH=Linux-x86-64-intel VERSION=sopt


Running the Application :

CP2K provide several test cases, which can be used for stability testing and verify the results to see if the compilation is correct.

] cd cp2k/tests/QS

] ./../../exe/Linux-x86-64-intel/cp2k.sopt H2O-32.inp

] ./../../exe/Linux-x86-64-intel/cp2k.sopt H2O-3x3.inp


Known Issues or Limitations :

1) For CP2K version 2.1 branch

We recommended to use Intel Fortran Compiler version 11.1 update 6 (version 11.1.072) or later version.

If you want to use new Intel Fortran Composer XE 2011 (aka version 12.0), we suggest that one source file "qs_vxc_atom.F" should be compiled with -O1 option only.  You could add the following lines into Linux-x86-64-intel.sopt to avoid this problem:

FCFLAGS2 = -I$(INTEL_INC) -O1 -xHost -heap-arrays 64 -fpp -free $(DFLAGS)

qs_vxc_atom.o: qs_vxc_atom.F
$(FC) -c $(FCFLAGS2) $<

2) For CP2K version 2.2 branch

Suggested to use the Intel Fortran compiler 11.1 update 6 and later version. 

There are some known issues with new Composer XE 2011 (ifort version 12.x.xxx).  Please use the XE 2011 update 8 and future version,  also apply the following patches in configuration file "Linux-x86-64-intel.sopt":

FCFLAGS2 = -I$(INTEL_INC) -O1 -xHost -heap-arrays 64 -fpp -free $(DFLAGS)

et_coupling.o: et_coupling.F
$(FC) -c $(FCFLAGS2) $<

qs_vxc_atom.o: qs_vxc_atom.F
$(FC) -c $(FCFLAGS2) $<

<end>

- DPD200161616
- DPD200172602



Optimization Notice in English

For more complete information about compiler optimizations, see our Optimization Notice.

23 comments

Top
Mathieu Dubois's picture

I am having trouble trying to run CP2K 2.3 revision version 12219 in a mixed MPI / OpenMP way. I am having either precisions erros (NaN) or segmentation fault errors.

Does someone has a best practice guide for CP2K compilation ?

I am using Intel Compilers 12.1.3 and Intel MPI 4.0.3

THanks

Hao Jiang (Intel)'s picture

The second issue reported in Sept. 9 (source file 'xc_pot_saop.F') has been fixed in Composer XE 2011 update 8. We still need work around for the first issue.

Hao Jiang (Intel)'s picture

Sorry, the correct configuration to use libint should looks like this:

LIBINT_PATH= <your install path>/libint-1.1.4
LIBINT_LIB = $(LIBINT_PATH)/lib/libderiv.a $(LIBINT_PATH)/lib/libint.a -lstdc++

Hao Jiang (Intel)'s picture

There are some known issues found on Fortran Composer XE 2011 update 6 (or version 12.1.233) with latest CP2K 2.2 branch (using CP2K version 2.2.347 in my testing)

In summary :
1) Two source files 'et_coupling.F' and 'qs_vxc_atom.F' need to compiled with -O1 option. Please apply FCFLAGS2 patch to them.

2) One source file 'xc_pot_saop.F' need to compile with -xSSE2 option (instead of -xHost). Suggest to apply another FCFLAGS3 patch looks like:

FCFLAGS3 = -I$(INTEL_INC) -O2 -xSSE2 -heap-arrays 64 -fpp -free $(DFLAGS)

xc_pot_saop.o: xc_pot_saop.F
$(FC) -c $(FCFLAGS3) $<

Please report any other issues you have found. Thanks

Hao Jiang (Intel)'s picture

If want to use libint, please add/modify the following lines in “Linux-x86-64-intel.sopt”:

LIBINT_PATH= <your install path>/libint-1.1.4
LIBINT_LIB = $(LIBINT_LIB)/lib/libderiv.a $(LIBINT_LIB)/lib/libint.a -lstdc++

DFLAGS = -D__INTEL -D__FFTSG -D__FFTW3 -D__FFTMKL -D__LIBINT

LIBS = -L$(INTEL_LIB) -mkl $(LIBINT_LIB)

It works with Composer XE 2011 (aka 12.0) and later version.

anonymous's picture

It seems that ifort (e.g.11.1.072) doesn't support ISO_C_BINDINGS. Using the following setup in Linux-x86-64-intel.popt, I successfully compiled cp2k with libint for the hybrid functional calculation without segmental fault.

CC = cc
CPP =
FC = mpif90
LD = mpif90
AR = ar -r
DFLAGS = -D__INTEL -D__FFTSG -D__parallel -D__BLACS -D__SCALAPACK -D__FFTW3 -D__LIBINT -D__HAS_NO_ISO_C_BINDING
CPPFLAGS =

INTEL_INC = /usr/apps/isv/intel/11.1.072/include
LIBINT_INC = $(HOME)/soft/libint-1.1.4/include
FFTW_INC = /usr/apps/free/fftw/3.2.2_i/include
FCFLAGS = $(DFLAGS) -I$(LIBINT_INC) -I$(INTEL_INC) -I$(FFTW_INC) -O2 -xW -heap-arrays 64 -funroll-loops -fpp -free
FCFLAGS2 = $(DFLAGS) -I$(LIBINT_INC) -I$(INTEL_INC) -O1 -xW -heap-arrays 64 -fpp -free
LDFLAGS = $(FCFLAGS) -I$(INTEL_INC)

FFTW_LIB = /usr/apps/free/fftw/3.2.2_i
INTEL_LIB = /usr/apps/isv/intel/11.1.072/mkl/lib/em64t
LIBINT = $(HOME)/soft/libint-1.1.4
LIBS =-L$(INTEL_LIB)/ -lmkl_scalapack_lp64 -lmkl_blacs_openmpi_lp64 -lmkl_intel_lp64 -lmkl_sequential -lmkl_core
$(HOME)/soft/cp2k-2.1/tools/hfx_tools/libint_tools/libint_cpp_wrapper.o
$(LIBINT)/lib/libderiv.a
$(LIBINT)/lib/libint.a
-lstdc++
$(FFTW_LIB)/lib/libfftw3.a

OBJECTS_ARCHITECTURE = machine_intel.o

graphcon.o: graphcon.F
$(FC) -c $(FCFLAGS2) $<

qs_vxc_atom.o: qs_vxc_atom.F
$(FC) -c $(FCFLAGS2) $<

-------
The gcc and ifort are used for compiling the libint-1.1.4.

anonymous's picture

I was able to compile cp2k with libint and calculating with hybrid functionals without getting seg faults. One simply has to remove -xHost from FCFLAGS. I'll try to find out which source file actually is responsible for the problem, therefore, it should be possible to optimize the rest of the program with -xHost put and apply FCFLAGS2 to the affected routine, module, source file.

anonymous's picture

Can you indicate me how to compile cp2k with libint? I always get segmentation errors at the stage where HFX_MEM_INFO would be printed out:
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
cp2k.popt 0000000001321BCE Unknown Unknown Unknown
...

Usually, it does not happen with gfortran. Does ifort really not support ISO_C_BINDINGS?
Thanks!

anonymous's picture

I got error in ifort v 12.0.2.137

/home/joonho/cp2k/makefiles/../lib/Linux-x86-64-intel/sopt/libcp2k_fft_lib.a(fftacml_lib.o): In function `fftacml3d_':
/home/joonho/cp2k/makefiles/../src/lib/fftacml_lib.F:(.text+0xbe0): undefined reference to `zfft3dx_'
/home/joonho/cp2k/makefiles/../src/lib/fftacml_lib.F:(.text+0xc50): undefined reference to `zfft3dx_'

Hao Jiang (Intel)'s picture

Thanks for your update, mecej4. Welocome to report any issue you find and share with us.

Pages

Add a Comment

Have a technical question? Visit our forums. Have site or software product issues? Contact support.