  
CONFIGURING
===========

We are using standard GNU configure scripts but since we rely upon
MPI, BLAS, LAPACK and other non-standard libraries most machines
require options passed to configure and build.  More information on
the MADNESS software environment is available on the wiki at

   https://github.com/m-a-d-n-e-s-s/madness


Hardware requirements:

  o A recent x86 cpu (32 or 64 bit) with SSE3 instructions (if you do
    not have SSE3, see below), or

  o IBM BG/P


Software requirements:

  o  MPI-2 with mpicc and mpicxx (tested with mpich2 and openmpi)

  o  BLAS

  o  LAPACK

  o  C and C++ compilers: gcc, g++ (most versions 4.7.x, 4.8.x, 4.9.x
     should work). Your compiler must also support C++11.

  o  If you are building from the respository you will also need
     autoconf and automake.


Optional:

  o  Google performance tools (the malloc library is fast and it is the only
     free multithreaded performance analysis tool)
     
  o  GTest for google unit test


If you checked the source out of subversion:

  o  You will need to run autoreconf (no options necessary) in the trunk directory
     to build the configure scripts


Configure options:

  Run the configure script with option --help to obtain a full list of options.
  The most commonly used are:

  o  --with-libunwind=dir ... directory containing libunwind.a for google profiler

  o  --with-google-perf=dir ... installation directory for google performance tools 
     (that specified with --prefix=dir when configuring the tools)

  o  --with-google-test=dir ... installation directory for google unit test library
     (that specified with --prefix=dir when configuring the library)
  
  o  --enable-debugging[=yes|no|OPTION] ... Enable debugging C and C++ compilers 
     [default=no]
     
  o  --enable-optimization[=yes|no|OPTION] ... Enable optimization for C and C++ 
     [default=yes]
     
  o  --enable-warning[=yes|no|GNU|Pathscale|Portland|Intel|IBM] ... Automatically set 
     warnings for compiler.[default=yes]
     
  o  --enable-optimal[=yes|no|GNU|Pathscale|Portland|Intel|IBM] ... Auto detect 
     optimal CXXFLAGS for compiler, or specify compiler vendor.[default=yes] 

  o  --enable-spinlocks ... forces use of spin locks instead of mutex (enabled by
     default on Cray XT and IBM BG/P)

  o  --enable-never-spin ... completely disables use of spinlocks (necessary for using VMs)


CONFIGURE COMMAND EXAMPLES
==========================

  o  Linux workstation or cluster using GNU compilers
     (some systems might require additional libraries to resolve Fortran
      symbols used by LAPACK/BLAS)

     - 64-bit with Intel MKL installed

       ./configure LDFLAGS="-L/opt/intel/mkl/10.0.1.014/lib/em64t" LIBS="-lmkl -lguide -lpthread -lm" CXXFLAGS="-std=c++0x"

     - 32-bit with Intel MKL installed

       ./configure LDFLAGS="-L/opt/intel/mkl/10.0.1.014/lib/32" LIBS="-lmkl -lguide -lpthread -lm" CXXFLAGS="-std=c++0x"

     - 32-bit with Intel MKL installed and also including MPFR and GSL 

       ./configure LDFLAGS="-I/usr/include/gsl -L/opt/intel/mkl/10.0.1.014/lib/32" LIBS="-lgsl -lgslcblas -lmkl -lguide -lpthread -lm -lmpfr -lgmp" CXXFLAGS="-std=c++0x"

     - 64-bit with MKL, google's fast malloc and profiler library

       ./configure LDFLAGS="-L/opt/intel/mkl/10.0.1.014/lib/em64t" LIBS="-lmkl -lguide -lpthread -lm" -with-google-perf=/usr/local/gperftools CXXFLAGS="-std=c++0x"

      - 64-bit with AMD ACML

      ./configure LDFLAGS="-L/usr/local/acml4.2.0/gfortran64_int64/lib" LIBS="-lacml  -lacml_mv -lgfortran -lstdc++" CXXFLAGS="-std=c++0x"

      - 64-bit with AMD ACML, google's super-fast malloc and google's profiler library

      ./configure LDFLAGS="-L/usr/local/acml4.2.0/gfortran64_int64/lib -L/usr/local/gperftools-20090306/lib -L/usr/local/libunwind-20090306" LIBS="-lacml  -lacml_mv -lprofiler -ltcmalloc -lunwind-x86_64 -lgfortran -lstdc++" CXXFLAGS="-std=c++0x"


  o  Cray-XT (e.g., jaguar@ornl and jaguarpf@ornl, kraken at NICS and franklin at NERSC) 
     [Module commands assume you start from the default which is PGI and scilib]

     - with GNU compilers and ACML - RECOMMENDED

       This is recommended since the GNU compiler is about 10x faster at
       compiling and has comparable execution speed to PGI.  ACML is
       also faster than the Goto BLAS in scilib for the small matrices
       primarily used by MADNESS and does not have a huge piece of
       static memory reserved.
     
       module swap PrgEnv-pgi PrgEnv-gnu
       module swap gcc/4.3.2 gcc/4.4.2
       module load acml/4.3.0

       ./configure CXXFLAGS=-std=c++11

     - with GNU compilers and scilib

       module swap PrgEnv-pgi PrgEnv-gnu
       module swap gcc/4.3.2 gcc/4.4.2
       ./configure CXXFLAGS=-std=c++11


  o  IBM BlueGene-Q @ ANL

     For other BGQ boxes you will need to find these compilers and libraries.  
     They seem to be site dependent.

#!/bin/bash
export LIBS=""
export LIBS="${LIBS} -L/home/robert/install/lib -llapack_bgp"
export LIBS="${LIBS} -L/soft/apps/ESSL-4.4.1-0/lib -lesslbg"
export LIBS="${LIBS} -L/soft/apps/ibmcmp-aug2009/xlf/bg/11.1/bglib -lxlf90_r -lxlfmath"
export LIBS="${LIBS} -L/soft/apps/ibmcmp-aug2009/xlsmp/bg/1.7/bglib -lxlsmp -lxl"
export LIBS="${LIBS} -L/bgsys/drivers/ppcfloor/gnu-linux/powerpc-bgp-linux/lib -lrt"
export LIBS="${LIBS} -lpthread"

export CPPFLAGS="-I/bgsys/drivers/ppcfloor/arch/include"

./configure \
  --build=powerpc-bgp-linux-gnu \
  --with-google-test=/home/robert/install \
  LIBS="${LIBS}" \
  CPPFLAGS="${CPPFLAGS}" 


  o  Macintosh Darwin 10.6, snow leopard
     Tested with Apple GCC 4.2.1, Apple clang 2.0, Intel 11.1 & 12.0, GNU GCC 4.4.x & 4.5.x, 
     and LLVM clang 2.8 
     
     - Apple GCC 4.2.1 using vecLib
       
       Set the following system variables
       export OMPI_CC=/usr/bin/gcc-4.2
       export OMPI_CXX=/usr/bin/g++-4.2

       ./configure  MPICC=/usr/bin/mpicc MPICXX=/usr/bin/mpicxx \
       LIBS="/System/Library/Frameworks/vecLib.framework/vecLib" \
       CPPFLAGS="-I/System/Library/Frameworks/vecLib.framework/Headers"
       
     - Apple clang 2.0 using vecLib
       
       Set the following system variables
       export OMPI_CC=/usr/bin/clang
       export OMPI_CXX=/usr/bin/clang++

       ./configure  MPICC=/usr/bin/mpicc MPICXX=/usr/bin/mpicxx \
       LIBS="/System/Library/Frameworks/vecLib.framework/vecLib" \
       CPPFLAGS="-I/System/Library/Frameworks/vecLib.framework/Headers" \
       CXXFLAGS="-std=c++0x"
     
     - GNU GCC 4.4.x or later using vecLib
     
       Set the following system variables
       export OMPI_CC=/path/to/gnu_gcc/gcc
       export OMPI_CXX=/path/to/gun_gcc/g++

       ./configure  MPICC=/usr/bin/mpicc MPICXX=/usr/bin/mpicxx \
       LIBS="/System/Library/Frameworks/vecLib.framework/vecLib" \
       CPPFLAGS="-I/System/Library/Frameworks/vecLib.framework/Headers" \
       CXXFLAGS="-std=c++0x"
       
     - Intel 11.x and 2011 (12.x) with Apple GCC
     
       Set the following system variables
       export OMPI_CC=icc
       export OMPI_CXX=icpc

       ./configure MPICC=/usr/bin/mpicc MPICXX=/usr/bin/mpicxx

  o On x86 without SSE3 -- configure with the option CPPFLAGS="-DDISABLE_SSE3", requires autoconf-2.65+, automake-1.11+
    
