Project Home (pi-qmc)

Table Of Contents

Previous topic

Appendices

Next topic

Frequently Asked Questions

This Page

Building pi-qmc

Quick start

The easiest way to build is to use:

./configure make

For a parallel build

./configure --enable-mpi MPICXX=mpic++ MPICC=mpicc MPIF77=mpif77

where you should use the names of your MPI enabled compilers.

You can also build for different numbers of physical dimensions (default is NDIM=3)

./configure --with-ndim=2

Required libraries

We use the following libraries in the pi code:

Advanced build using multiple directories

In research, we often want different versions of the executables, for example, versions with and without MPI, or versions compiled for two-dimensional systems. To accomplish this, we make a pibuilds directory beside our svn checkout directory (pi or pi-qmc). We then make empty subdirectories for each build, for example ndim2mpi for a two dimensional MPI version. A typical directory structure is:

codes/
  pi-qmc/
    configure
    src/
    lib/
  pibuilds/
    ndim1/
    ndim2/
    ndim3/
    ndim1mpi/
    ndim2mpi/
    ndim3mpi/
    debug/

To build, go into the empty build directory,

cd ~/codes/pibuilds/ndim2mpi

Then run the configure script with the desired options

../../configure --with-ndim=2 --enable-mpi

You will probably want more configure options; see the platform specific instructions below for some examples.

Then, make the code in that directory,

make -j2

For conveniance, you can make a soft link to the executable

ln -sf ~/codes/pibuilds/ndim3mpi ~/bin/pi2Dmpi

Platform specific instructions

Mac OS X

All the dependencies are available through [http://www.macports.org/ macports]. It is also handy to install the latest gcc compilers (with gfortran), openmpi, and python utilities for data analysis and plotting.

$ port installed
  libxml2 @2.7.3_0 (active)
  blitz @0.9_0 (active)
  hdf5-18 @1.8.3_0 (active)
  gsl @1.12_0 (active)
  fftw-3 @3.2.2_0 (active)

  gcc44 @4.4.0_0 (active)

  python26 @2.6.2_3 (active)
  py26-numpy @1.3.0_0 (active)
  py26-ipython @0.9.1_0+scientific (active)
  py26-scipy @0.7.0_0+gcc44 (active)
  py26-tables @2.1_0 (active)

A bash script to download the pi-qmc source from git hub, compile on Mac OS X, and run all tests is included in doc/deploy/macosx/start.sh:

#!/bin/bash

git clone git@github.com:phys-tools/pi-qmc.git

mkdir pibuild
cd pibuild
export CC=gcc-mp-4.7
export CXX=g++-mp-4.7
export CXXFLAGS="-O3 -g -Wall -ffast-math -march=native"
export CXXFLAGS+=" -ftree-vectorize -fomit-frame-pointer -pipe"
cmake ../pi-qmc
make -j2

echo; echo "  Running unit tests..."; echo
make -j2 unittest
bin/unittest_pi

cd test/system
echo; echo "Running system integration tests..."; echo
nosetests-2.7 -v --rednose --with-xunit
cd ../..

The following configure works well on an intel mac:

../../pi/configure CXX=g++-mp-4.7 CC=gcc-mp-4.7 \
  CXXFLAGS="-O3 -g -Wall -ffast-math -ftree-vectorize \
  -march=native -fomit-frame-pointer -pipe" \
  F77=gfortran-mp-4.7

or, for an MPI enabled build,

../../pi/configure --enable-mpi CXX=g++-mp-4.4 CC=gcc-mp-4.4 F77=gfortran-mp-4.4 \
  MPICC=openmpicc MPICXX=openmpicxx MPIF77=openmpif77 \
  CXXFLAGS="-O3 -g -Wall -ffast-math -ftree-vectorize \
  -march=native -fomit-frame-pointer -pipe"

On a G5 mac, try:

../../pi/configure --with-ndim=3  F77=gfortran-mp-4.4 CC=gcc-mp-4.4 CXX=g++-mp-4.4\
  CXXFLAGS="-g -O3 -ffast-math -ftree-vectorize -maltivec -mpowerpc-gpopt \
  -mpowerpc64 falign-functions=32 -falign-labels=32 -falign-loops=32 -falign-jumps=32 -funroll-loops"

or, for an MPI enabled build,

../../pi/configure --with-ndim=3 --enable-mpi \
  CXXFLAGS="-g -O3 -ffast-math -ftree-vectorize -maltivec -mpowerpc-gpopt \
  -mpowerpc64 falign-functions=32 -falign-labels=32 -falign-loops=32 -falign-jumps=32 -funroll-loops" \
  F77=gfortran-mp-4.4 CC=gcc-mp-4.4 CXX=g++-mp-4.4  MPICC=openmpicc MPICXX=openmpicxx MPIF77=openmpif77

Linux (CentOS 5.3)

You can download dependencies using yum. First, you may need to add access to the fedora [http://fedoraproject.org/wiki/EPEL Extra Packages for Enterprise Linux (EPEL)].

sudo rpm -Uvh http://download.fedora.redhat.com/pub/epel/5/i386/epel-release-5-3.noarch.rpm

Then install the required packages for _pi_. (You probably want to compile atlas yourself to get automatic performance tuning for your hardware, but the yum install will work if you are impatient.) Note: replace x86_64 with i386 if you are on a 32 bit machine.

sudo yum install libxml2-devel-versionXXX.x86_64 (here I don't know the correct version)
sudo yum install blitz-devel.x86_64
sudo yum install fftw3-devel.x86_64
sudo yum install hdf5-devel.x86_64
sudo yum install atlas-sse3-devel.x86_64
sudo yum install lapack-devel.x86_64
sudo yum install gsl-devel.x86_64

It is useful to install the gcc 4.3 compilers.

sudo yum install gcc43.x86_64
sudo yum install gcc43-c++.x86_64
sudo yum install gcc43-gfortran.x86_64

Also, you will want an MPI implementation if you want to run in parallel,

sudo yum install openmpi-devel.x86_64

The openmpi package will require that you run mpi-selector and open a new terminal to get the executables. Use the mpi-selector --list option to see what is available, then set a system-wide default.

sudo mpi-selector –system –set openmpi-1.2.7-gcc-x86_64

When you configure pi, you will probably need to specify the location of your BLAS and LAPACK routines,

../../pi/configure CXX=g++43 CC=gcc43 F77=gfortran43 CXXFLAGS=\
"-g -O3 -ffast-math -ftree-vectorize -march=native -fomit-frame-pointer -pipe"\
 --with-blas="-L/usr/lib64/atlas -llapack -lf77blas"

For mpi, just add –enable-mpi.

For the python analysis utilities, you’ll want to install ipython and matplotlib.

sudo yum install python-matplotlib
sudo yum install ipython
sudo yum install scipy

The python package pytables for reading HDF5 files is also required for the analysis scripts, but it is not available through yum, so you’ll have to download it and install it yourself.

HPC Centers

ASU Fulton: saguaro

For a serial build in two dimensions,

../../pi/configure –with-ndim=2 –enable-sprng CXX=icpc CC=icc CXXFLAGS=”-O3 -xP -ipo” –with-blas=”-L$MKL_LIB -lmkl_lapack -lmkl_intel_lp64 -lmkl_sequential -lmkl_core” F77=ifort AR=”xild -lib”

or for a parallel version,

../../pi/configure –with-ndim=2 –enable-sprng –enable-mpi MPICC=mpicc MPICXX=mpicxx CXX=icpc CC=icc F77=ifort CXXFLAGS=”-O3 -xP -ipo” AR=”xild -lib” –with-blas=”-L$MKL_LIB -lmkl_lapack -lmkl_intel_lp64 -lmkl_sequential -lmkl_core”

Omit the –enable-sprng option if you do not have the SPRNG library.

LONI-LSU: queenbee

You need to add some lines to your .soft file to include some required libraries,

#My additions (CPATH mimics -I include directories).
CPATH += /usr/local/packages/hdf5-1.8.1-intel10.1/include
+gsl-1.9-intel10.1
+sprng4-mvapich-1.1-intel-10.1
+fftw-3.1.2-intel10.1
CPATH += :/usr/local/packages/fftw-3.1.2-intel10.1/include
+intel-mkl
CPPFLAGS += -DMPICH_IGNORE_CXX_SEEK

For an MPI build, use,

../../pi/configure --with-ndim=3 --enable-mpi MPICC=mpicc MPICXX=mpicxx \
  CXX=icpc CC=icc F77=ifort AR="xild -lib" CXXFLAGS="-O3 -xP -ipo" \
  --with-blas="-lmkl_lapack -lmkl_intel_lp64 -lmkl_sequential -lmkl_core"

NCSA: abe

You need to add some lines to your .soft file to include some required libraries,

#My additions (CPATH mimics -I include directories).
+libxml2-2.6.29
+libxml2
+intel-mkl
+gsl-intel
+hdf5-1.8.2
CPATH += :/usr/apps/hdf/hdf5/v182/include
LD_LIBRARY_PATH += /usr/apps/hdf/szip/lib
+fftw-3.1-intel
LD_LIBRARY_PATH += /usr/apps/math/fftw/fftw-3.1.2/intel10/lib
CPATH += :/usr/apps/math/fftw/fftw-3.1.2/intel10/include
+intel-mkl
CPPFLAGS = "${CPPFLAGS} -DMPICH_IGNORE_CXX_SEEK"
Also have blitz installed locally with --prefix=(your dir choice)

For an MPI build, use,

../../pi/configure –with-ndim=3 –enable-mpi MPICC=mpicc MPICXX=mpicxx CXX=icpc CC=icc CXXFLAGS=”-O3 -xP -ipo” LDFLAGS=”-lsz” –with-blas=”-lmkl_lapack -lmkl_intel_lp64 -lmkl_sequential -lmkl_core” F77=ifort AR=”xild -lib”

TACC: Ranger

Cornell CNF: nanolab

The svn client wasn’t working for me, so I built one in my ~/packages/bin directory. You need to specify the most recent C++ and Fortran compilers by including the following in your .bash_profile,

# Version 10 compilers
source /opt/intel/cc/10.1.017/bin/iccvars.sh
source /opt/intel/fc/10.1.017/bin/ifortvars.shsource /opt/intel/idb/10.1.017/bin/idbvars.sh
source /opt/intel/mkl/10.0.4.023/tools/environment/mklvars32.sh

Also, make sure that /usr/lam-7.4.1_intelv10/bin is in your path to get the correct MPI compilers.

You need to build blitz (again, in my ~/packages directory). For a serial pi build,

../../pi/configure --with-ndim=3 CXX=icpc CC=icc CXXFLAGS="-O3 -ipo" \
--with-blas="-Wl,-rpath,$MKLROOT/lib/32 -L/opt$MKLROOT/lib/32 -lmkl_intel \
-lmkl_sequential -lmkl_core -lpthread -lm" F77=ifort AR="xild -lib"
../../pi/configure --with-ndim=3 CXX=icpc CC=icc CXXFLAGS="-O3 -ipo" \
--with-blas="-Wl,-rpath,$MKLROOT/lib/32 -L$MKLROOT/lib/32 -lmkl_intel \
-lmkl_sequential -lmkl_core -lpthread -lm" F77=ifort AR="xild -lib" \
--enable-mpi MPICXX=mpic++ MPICC=mpicc MPIF77=mpif77