Low performance with vasp 4.6.28 and intel mkl 11.0.074

[gorkab]

Dear all,

I'm compiling vasp 4.6.28 in our new SuSE machines (2 Intel(R) Xeon(R) CPU E5430 of 4 cores, 16GB and Gygabit ethernet) with intel 11.0.074 and mpich/1.2.7 (compiled with the same intel compiler).

I tried to follow the instructions related in http://software.intel.com/en-us/article ... l-in-vasp/

but when we try to execute some jobs and compare its results with other machines ( 2 Intel(R) Xeon(R) CPU E5345 of 4 cores, 16GB and Gygabit ethernet) who had another compilation of the same version of vasp we obtain a very low performance (it takes double time or more to solve the job!!!). The jobs are running in one node with 4 or 8 procs.

This is my makefile_linux.ifc.P4 (sorry but I couldn't find the makefile of the "performed vasp"):

.SUFFIXES: .inc .f .f90 .F
#-----------------------------------------------------------------------
# Makefile for Intel Fortran compiler for P4 systems
#
# The makefile was tested only under Linux on Intel platforms
# (Suse 5.3- Suse 9.0)
# the followin compiler versions have been tested
# 5.0, 6.0, 7.0 and 7.1 (some 8.0 versions seem to fail compiling the code)
# presently we recommend version 7.1 or 7.0, since these
# releases have been used to compile the present code versions
#
# it might be required to change some of library pathes, since
# LINUX installation vary a lot
# Hence check ***ALL**** options in this makefile very carefully
#-----------------------------------------------------------------------
#
# BLAS must be installed on the machine
# there are several options:
# 1) very slow but works:
# retrieve the lapackage from ftp.netlib.org
# and compile the blas routines (BLAS/SRC directory)
# please use g77 or f77 for the compilation. When I tried to
# use pgf77 or pgf90 for BLAS, VASP hang up when calling
# ZHEEV (however this was with lapack 1.1 now I use lapack 2.0)
# 2) most desirable: get an optimized BLAS
#
# the two most reliable packages around are presently:
# 3a) Intels own optimised BLAS (PIII, P4, Itanium)
# http://developer.intel.com/software/products/mkl/
# this is really excellent when you use Intel CPU's
#
# 3b) or obtain the atlas based BLAS routines
# http://math-atlas.sourceforge.net/
# you certainly need atlas on the Athlon, since the mkl
# routines are not optimal on the Athlon.
# If you want to use atlas based BLAS, check the lines around LIB=
#
# 3c) mindblowing fast SSE2 (4 GFlops on P4, 2.53 GHz)
# Kazushige Goto's BLAS
# http://www.cs.utexas.edu/users/kgoto/signup_first.html
#
#-----------------------------------------------------------------------

# all CPP processed fortran files have the extension .f90
SUFFIX=.f90

#-----------------------------------------------------------------------
# fortran compiler and linker
#-----------------------------------------------------------------------
FC=ifort
# fortran linker
FCL=$(FC)


#-----------------------------------------------------------------------
# whereis CPP ?? (I need CPP, can't use gcc with proper options)
# that's the location of gcc for SUSE 5.3
#
# CPP_ = /usr/lib/gcc-lib/i486-linux/2.7.2/cpp -P -C
#
# that's probably the right line for some Red Hat distribution:
#
# CPP_ = /usr/lib/gcc-lib/i386-redhat-linux/2.7.2.3/cpp -P -C
#
# SUSE X.X, maybe some Red Hat distributions:

CPP_ = ./preprocess <$*.F | /usr/bin/cpp -P -C -traditional >$*$(SUFFIX)

#-----------------------------------------------------------------------
# possible options for CPP:
# NGXhalf charge density reduced in X direction
# wNGXhalf gamma point only reduced in X direction
# avoidalloc avoid ALLOCATE if possible
# IFC work around some IFC bugs
# CACHE_SIZE 1000 for PII,PIII, 5000 for Athlon, 8000-12000 P4
# RPROMU_DGEMV use DGEMV instead of DGEMM in RPRO (depends on used BLAS)
# RACCMU_DGEMV use DGEMV instead of DGEMM in RACC (depends on used BLAS)
#-----------------------------------------------------------------------

CPP = $(CPP_) -DHOST=\"LinuxIFC\" \
-Dkind8 -DNGXhalf -DCACHE_SIZE=12000 -DPGF90 -Davoidalloc \
-DRPROMU_DGEMV -DRACCMU_DGEMV

#-----------------------------------------------------------------------
# general fortran flags (there must a trailing blank on this line)
#-----------------------------------------------------------------------

FFLAGS =-I/opt/intel/Compiler/11.0/074/mkl/include/fftw -FR -lowercase

#FFLAGS = -quiet -FR -lowercase -assume byterecl

#-----------------------------------------------------------------------
# optimization
# we have tested whether higher optimisation improves performance
# -axK SSE1 optimization, but also generate code executable on all mach.
# xK improves performance somewhat on XP, and a is required in order
# to run the code on older Athlons as well
# -xW SSE2 optimization
# -axW SSE2 optimization, but also generate code executable on all mach.
# -tpp6 P3 optimization
# -tpp7 P4 optimization
#-----------------------------------------------------------------------

OFLAG=-O3 -xW -tpp7

OFLAG_HIGH = $(OFLAG)
OBJ_HIGH =

OBJ_NOOPT =
DEBUG = -FR -O0
INLINE = $(OFLAG)


#-----------------------------------------------------------------------
# the following lines specify the position of BLAS and LAPACK
# on P4, VASP works fastest with the libgoto library
# so that's what I recommend
#-----------------------------------------------------------------------

# Atlas based libraries
#ATLASHOME= $(HOME)/archives/BLAS_OPT/ATLAS/lib/Linux_P4SSE2/
#BLAS= -L$(ATLASHOME) -lf77blas -latlas

# use specific libraries (default library path might point to other libraries)
#BLAS= $(ATLASHOME)/libf77blas.a $(ATLASHOME)/libatlas.a

# use the mkl Intel libraries for p4 (www.intel.com)
# mkl.5.1
# set -DRPROMU_DGEMV -DRACCMU_DGEMV in the CPP lines
#BLAS=-L/opt/intel/mkl/lib/32 -lmkl_p4 -lpthread

# mkl.5.2 requires also to -lguide library
# set -DRPROMU_DGEMV -DRACCMU_DGEMV in the CPP lines
#BLAS=-L/opt/intel/mkl/lib/32 -lmkl_p4 -lguide -lpthread

# even faster Kazushige Goto's BLAS
# http://www.cs.utexas.edu/users/kgoto/signup_first.html
#BLAS= /opt/libs/libgoto/libgoto_p4_512-r0.6.so

# LAPACK, simplest use vasp.4.lib/lapack_double
#LAPACK= ../vasp.4.lib/lapack_double.o

# use atlas optimized part of lapack
#LAPACK= ../vasp.4.lib/lapack_atlas.o -llapack -lcblas

# use the mkl Intel lapack
#LAPACK= -lmkl_lapack

BLAS=-L/opt/intel/Compiler/11.0/074/mkl/lib/em64t -lmkl_intel_lp64 -lmkl_blacs_lp64 -lmkl_intel_thread -lmkl_core -liomp5 -lpthread

#LAPACK=-L/opt/intel/Compiler/11.0/074/mkl/lib/em64t -lmkl_intel_lp64 -lmkl_blacs_lp64 -lmkl_intel_thread -lmkl_core -liomp5 -lpthread

LAPACK=-L/opt/intel/Compiler/11.0/074/mkl/lib/em64t/libmkl_lapack.a

#-----------------------------------------------------------------------

LIB = -L../vasp.4.lib -ldmy \
../vasp.4.lib/linpack_double.o $(LAPACK) \
$(BLAS)

# options for linking (for compiler version 6.X, 7.1) nothing is required
LINK =
# compiler version 7.0 generates some vector statments which are located
# in the svml library, add the LIBPATH and the library (just in case)
#LINK = -L/opt/intel/compiler70/ia32/lib/ -lsvml

#-----------------------------------------------------------------------
# fft libraries:
# VASP.4.6 can use fftw.3.0.X (http://www.fftw.org)
# since this version is faster on P4 machines, we recommend to use it
#-----------------------------------------------------------------------

#FFT3D = fft3dfurth.o fft3dlib.o
#FFT3D = fftw3d.o fft3dlib.o /opt/libs/fftw-3.0.1/lib/libfftw3.a

FFT3D= fftmpi.o fftmpi_map.o fft3dlib.o /opt/intel/Compiler/11.0/074/mkl/lib/em64t/libfftw3xf_intel.a

#=======================================================================
# MPI section, uncomment the following lines
#
# one comment for users of mpich or lam:
# You must *not* compile mpi with g77/f77, because f77/g77
# appends *two* underscores to symbols that contain already an
# underscore (i.e. MPI_SEND becomes mpi_send__). The pgf90/ifc
# compilers however append only one underscore.
# Precompiled mpi version will also not work !!!
#
# We found that mpich.1.2.1 and lam-6.5.X to lam-7.0.4 are stable
# mpich.1.2.1 was configured with
# ./configure -prefix=/usr/local/mpich_nodvdbg -fc="pgf77 -Mx,119,0x200000" \
# -f90="pgf90 " \
# --without-romio --without-mpe -opt=-O \
#
# lam was configured with the line
# ./configure -prefix /opt/libs/lam-7.0.4 --with-cflags=-O -with-fc=ifc \
# --with-f77flags=-O --without-romio
#
# please note that you might be able to use a lam or mpich version
# compiled with f77/g77, but then you need to add the following
# options: -Msecond_underscore (compilation) and -g77libs (linking)
#
# !!! Please do not send me any queries on how to install MPI, I will
# certainly not answer them !!!!
#=======================================================================
#-----------------------------------------------------------------------
# fortran linker for mpi: if you use LAM and compiled it with the options
# suggested above, you can use the following line
#-----------------------------------------------------------------------

FC=/opt/mpich-1.2.7p1/bin/mpif77
FCL=$(FC)

#-----------------------------------------------------------------------
# additional options for CPP in parallel version (see also above):
# NGZhalf charge density reduced in Z direction
# wNGZhalf gamma point only reduced in Z direction
# scaLAPACK use scaLAPACK (usually slower on 100 Mbit Net)
#-----------------------------------------------------------------------

CPP = $(CPP_) -DMPI -DHOST=\"LinuxIFC\" -DIFC \
-Dkind8 -DNGZhalf -DCACHE_SIZE=4000 -DPGF90 -Davoidalloc \
-DMPI_BLOCK=500 \
-DRPROMU_DGEMV -DRACCMU_DGEMV

#-----------------------------------------------------------------------
# location of SCALAPACK
# if you do not use SCALAPACK simply uncomment the line SCA
#-----------------------------------------------------------------------

BLACS=$(HOME)/archives/SCALAPACK/BLACS/
SCA_=$(HOME)/archives/SCALAPACK/SCALAPACK

SCA= $(SCA_)/libscalapack.a \
$(BLACS)/LIB/blacsF77init_MPI-LINUX-0.a $(BLACS)/LIB/blacs_MPI-LINUX-0.a $(BLACS)/LIB/blacsF77init_MPI-LINUX-0.a

SCA=

#-----------------------------------------------------------------------
# libraries for mpi
#-----------------------------------------------------------------------

LIB = -L../vasp.4.lib -ldmy \
../vasp.4.lib/linpack_double.o $(LAPACK) \
$(SCA) $(BLAS)

# FFT: fftmpi.o with fft3dlib of Juergen Furthmueller
#FFT3D = fftmpi.o fftmpi_map.o fft3dlib.o

# fftw.3.0.1 is slighly faster and should be used if available
#FFT3D = fftmpiw.o fftmpi_map.o fft3dlib.o /opt/libs/fftw-3.0.1/lib/libfftw3.a

#-----------------------------------------------------------------------
# general rules and compile lines
#-----------------------------------------------------------------------
BASIC= symmetry.o symlib.o lattlib.o random.o

SOURCE= base.o mpi.o smart_allocate.o xml.o \
constant.o jacobi.o main_mpi.o scala.o \
asa.o lattice.o poscar.o ini.o setex.o radial.o \
pseudo.o mgrid.o mkpoints.o wave.o wave_mpi.o $(BASIC) \
nonl.o nonlr.o dfast.o choleski2.o \
mix.o charge.o xcgrad.o xcspin.o potex1.o potex2.o \
metagga.o constrmag.o pot.o cl_shift.o force.o dos.o elf.o \
tet.o hamil.o steep.o \
chain.o dyna.o relativistic.o LDApU.o sphpro.o paw.o us.o \
ebs.o wavpre.o wavpre_noio.o broyden.o \
dynbr.o rmm-diis.o reader.o writer.o tutor.o xml_writer.o \
brent.o stufak.o fileio.o opergrid.o stepver.o \
dipol.o xclib.o chgloc.o subrot.o optreal.o davidson.o \
edtest.o electron.o shm.o pardens.o paircorrection.o \
optics.o constr_cell_relax.o stm.o finite_diff.o \
elpol.o setlocalpp.o

INC=

vasp: $(SOURCE) $(FFT3D) $(INC) main.o
rm -f vasp
$(FCL) -o vasp $(LINK) main.o $(SOURCE) $(FFT3D) $(LIB)
makeparam: $(SOURCE) $(FFT3D) makeparam.o main.F $(INC)
$(FCL) -o makeparam $(LINK) makeparam.o $(SOURCE) $(FFT3D) $(LIB)
zgemmtest: zgemmtest.o base.o random.o $(INC)
$(FCL) -o zgemmtest $(LINK) zgemmtest.o random.o base.o $(LIB)
dgemmtest: dgemmtest.o base.o random.o $(INC)
$(FCL) -o dgemmtest $(LINK) dgemmtest.o random.o base.o $(LIB)
ffttest: base.o smart_allocate.o mpi.o mgrid.o random.o ffttest.o $(FFT3D) $(INC)
$(FCL) -o ffttest $(LINK) ffttest.o mpi.o mgrid.o random.o smart_allocate.o base.o $(FFT3D) $(LIB)
kpoints: $(SOURCE) $(FFT3D) makekpoints.o main.F $(INC)
$(FCL) -o kpoints $(LINK) makekpoints.o $(SOURCE) $(FFT3D) $(LIB)

clean:
-rm -f *.g *.f *.o *.L *.mod ; touch *.F

main.o: main$(SUFFIX)
$(FC) $(FFLAGS)$(DEBUG) $(INCS) -c main$(SUFFIX)
xcgrad.o: xcgrad$(SUFFIX)
$(FC) $(FFLAGS) $(INLINE) $(INCS) -c xcgrad$(SUFFIX)
xcspin.o: xcspin$(SUFFIX)
$(FC) $(FFLAGS) $(INLINE) $(INCS) -c xcspin$(SUFFIX)

makeparam.o: makeparam$(SUFFIX)
$(FC) $(FFLAGS)$(DEBUG) $(INCS) -c makeparam$(SUFFIX)

makeparam$(SUFFIX): makeparam.F main.F
#
# MIND: I do not have a full dependency list for the include
# and MODULES: here are only the minimal basic dependencies
# if one strucuture is changed then touch_dep must be called
# with the corresponding name of the structure
#
base.o: base.inc base.F
mgrid.o: mgrid.inc mgrid.F
constant.o: constant.inc constant.F
lattice.o: lattice.inc lattice.F
setex.o: setexm.inc setex.F
pseudo.o: pseudo.inc pseudo.F
poscar.o: poscar.inc poscar.F
mkpoints.o: mkpoints.inc mkpoints.F
wave.o: wave.inc wave.F
nonl.o: nonl.inc nonl.F
nonlr.o: nonlr.inc nonlr.F

$(OBJ_HIGH):
$(CPP)
$(FC) $(FFLAGS) $(OFLAG_HIGH) $(INCS) -c $*$(SUFFIX)
$(OBJ_NOOPT):
$(CPP)
$(FC) $(FFLAGS) $(INCS) -c $*$(SUFFIX)

fft3dlib_f77.o: fft3dlib_f77.F
$(CPP)
$(F77) $(FFLAGS_F77) -c $*$(SUFFIX)

.F.o:
$(CPP)
$(FC) $(FFLAGS) $(OFLAG) $(INCS) -c $*$(SUFFIX)
.F$(SUFFIX):
$(CPP)
$(SUFFIX).o:
$(FC) $(FFLAGS) $(OFLAG) $(INCS) -c $*$(SUFFIX)

# special rules
#-----------------------------------------------------------------------
# these special rules are cummulative (that is once failed
# in one compiler version, stays in the list forever)
# -tpp5|6|7 P, PII-PIII, PIV
# -xW use SIMD (does not pay of on PII, since fft3d uses double prec)
# all other options do no affect the code performance since -O1 is used
#-----------------------------------------------------------------------

fft3dlib.o : fft3dlib.F
$(CPP)
$(FC) -FR -lowercase -O1 -tpp7 -xW -prefetch- -unroll0 -vec_report3 -c $*$(SUFFIX)
fft3dfurth.o : fft3dfurth.F
$(CPP)
$(FC) -FR -lowercase -O1 -c $*$(SUFFIX)

radial.o : radial.F
$(CPP)
$(FC) -FR -lowercase -O1 -c $*$(SUFFIX)

symlib.o : symlib.F
$(CPP)
$(FC) -FR -lowercase -O1 -c $*$(SUFFIX)

symmetry.o : symmetry.F
$(CPP)
$(FC) -FR -lowercase -O1 -c $*$(SUFFIX)

dynbr.o : dynbr.F
$(CPP)
$(FC) -FR -lowercase -O1 -c $*$(SUFFIX)

broyden.o : broyden.F
$(CPP)
$(FC) -FR -lowercase -O2 -c $*$(SUFFIX)

us.o : us.F
$(CPP)
$(FC) -FR -lowercase -O1 -c $*$(SUFFIX)

wave.o : wave.F
$(CPP)
$(FC) -FR -lowercase -O0 -c $*$(SUFFIX)

LDApU.o : LDApU.F
$(CPP)
$(FC) -FR -lowercase -O2 -c $*$(SUFFIX)


It seems all correct acording to the intel pages.

Does anyone has encountered the same problem? Can you give me some
advice? Maybe the problem is with the new intel version?


Thanks for any advise,

best regards

jordi ingles

ps: I also tried to follow these instructions with the same result: http://cms.mpi.univie.ac.at/vasp-forum/ ... php?2.5010

[gorkab]

Any idea where could be the problem? Doy you think that the problem is with the old version of vasp (4.6.28) compiled with the newest intel compiler version? Maybe I should recompile with and older version or with libgoto libraries?

I can't imagine where I made the mistake.

thanks in advance

[mbsullivan]

If you want help, you'll need to tell us the error that you are getting when you compile. I suspect that the paths to your libraries aren't set quite right.

[gorkab]

Hi Mbsullivan,

thanks for your response.

Vasp compiles fine and libraries seems ok with "ldd vasp".

The problem is when I execute VASP, same test takes double time that vasp in other (and older) nodes with intel and mkl libraries. The problem is that I haven't access to the makefile of the vasp of this old nodes. So I haven't got any idea where could be the problem with this low performance.

thanks for any advise

[job]

Make sure you have the environment variable

OMP_NUM_THREADS=1

otherwise MKL might use threaded blas routines, e.g. you run 4 MPI processes on 4 processors, you get a total of 16 threads fighting for those procs. Another thing in your Makefile is that you're not using FFTW, for that you need to link in fft_mpiw.o instead of fft_mpi.o.

[gorkab]

Hi!

Thanks for your response job!

If I try to compile vasp with fft_mpiw.o and then execute it I obtain the following error:

FFT: planning ... 1
p1_18452: p4_error: interrupt SIGBUS: 7
p2_18482: p4_error: interrupt SIGx: 13
p0_18446: p4_error: interrupt SIGx: 13
p3_18512: p4_error: interrupt SIGx: 13
Killed by signal 2.
Killed by signal 2.
Killed by signal 2.
p0_18446: (8.247296) net_send: could not write to fd=4, errno = 32

But If I use fft_mpi and set the OMP_NUM_THREADS it works fine!

[pafell]

Hi,

I'd suggest changing optimization params for ifort. Since you have core2 cpu, you should change
OFLAG=-O3 -xW -tpp7
to
OFLAG=-O3 -xSSE4.1

Both tpp7 and xW switches are missing (depraceted in previous ifort) in ifort 11.