Child pages
  • cp2k on sisu
Skip to end of metadata
Go to start of metadata

arch file for mpi-only parallel version (popt):

# Program environments:
# - module switch PrgEnv-cray PrgEnv-gnu/5.0.15
# - module load fftw/3.3.0.1

 

HWTOPO   = no
CC       = cc
CPP      =
FC       = ftn
LD       = ftn
AR       = ar -r
CPPFLAGS =
DFLAGS   = -D__GFORTRAN -D__FFTSG -D__FFTW3 -D__HAS_NO_GETLOG -D__parallel -D__SCALAPACK -D__BLACS  
CFLAGS   = $(DFLAGS)
FCFLAGS  = -O3 -march=native -ffree-form -ftree-vectorize $(DFLAGS) -ffree-line-length-512
LDFLAGS  = $(FCFLAGS)
LIBS     =

OBJECTS_ARCHITECTURE = machine_gfortran.o
  • No labels

2 Comments

  1. Actually, Intel is a lot faster:

    # Program environments:
    # - module switch PrgEnv-cray/5.0.15 PrgEnv-intel/5.0.15
    # - module load fftw/3.3.0.1
    # export LD_LIBRARY_PATH=/opt/intel/composer_xe_2013.1.117/mkl/lib/intel64:$LD_LIBRARY_PATH
    # export MKLROOT=/opt/intel/composer_xe_2013.1.117/mkl

     

    HWTOPO   = no
    CC       = cc
    CPP      =
    FC       = ftn
    LD       = ftn
    AR       = xiar -r
    CPPFLAGS =
    DFLAGS   = -D__INTEL -D__FFTSG -D__FFTW3 -D__HAS_NO_GETLOG -D__parallel -D__SCALAPACK -D__BLACS  
    CFLAGS   = $(DFLAGS)
    FCFLAGS  = $(DFLAGS)  -O3 -opt-prefetch -unroll-aggressive -no-prec-div -fp-model fast=2  -fpp -free
    FCFLAGS2 = $(DFLAGS)  -O1 -heap-arrays 64 -fpp -free
    LDFLAGS  = $(FCFLAGS)
    LIBS     = $(MKLROOT)/lib/intel64/libmkl_blas95_lp64.a \
               $(MKLROOT)/lib/intel64/libmkl_lapack95_lp64.a \
               $(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
     
    # the upper stuff is from mkl link advisor

     

    OBJECTS_ARCHITECTURE = machine_intel.o

     

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

      

  2. I learned today from the Cray specialists that cp2k can't at the moment be compiled with the intel compiler using mixed parallelization. Intel is aware of this and is working on it.

     

    Has anyone built the gnu-version using mixed parallelization? Any performance data?