Skip to content

Compiling BerkeleyGW

Prerequisites

We have tested BerkeleyGW extensively with various configurations, and support the following compilers and libraries:

  • Operating systems: Linux, AIX, MacOS
  • Fortran compilers (required, and must support Fortran 2003): pgf90, ifort, gfortran, g95, openf90, sunf90, pathf90, crayftn, af90 (Absoft), nagfor, xlf90
  • C compilers (required): pgcc, icc, gcc, opencc, pathcc, craycc, clang
  • C++ compilers (required): pgCC, icpc, g++, openCC, pathCC, crayCC, clang++
  • MPI implementation (optional): OpenMPI, MPICH1, MPICH2, MVAPICH2, Intel MPI
  • OpenMP threading (optional)
  • LAPACK/BLAS implementation (required): NetLib, ATLAS, Intel MKL, ACML, Cray LibSci
  • ScaLAPACK/BLACS implementation (required if MPI is used): NetLib, Cray LibSci, Intel MKL, AMD
  • FFTW (required): versions 3.3.x (preferred) or 2.x.

Compilation instructions

The instructions below represent a typical case for compiling BerkeleyGW:

  1. Architecture-specific Makefile-include files appropriate for various supercomputers as well as for using standard Ubuntu or Macports packages are provided in the config/ directory. Copy the configuration file that most resembles your architecture and name it arch.mk. Example:

    cp config/stampede.tacc.utexas.edu_threaded_hdf5.mk arch.mk
    

  2. Edit the arch.mk file to fit your needs. Refer to the compilation flags documentation for details.

  3. Type make to see the available compilation targets. If the arch.mk is properly configured, you only need to type:

    make all-flavors -j 16
    
    where 16 can be substituted by the number of concurrent build processes.

  4. If there is any error in the make processes, it may be beneficial to first build the real version of the code. You can repeatedly type make real and fix any issues in the arch.mk. Once the code builds successfully, clean the code and make the complex version of the code:

    make clean-flavored
    make cplx -j 16
    

Note that the complex version may always be used. Real may be used for systems that have both inversion (about the origin) and time-reversal symmetry, and will be faster and use less memory.

  1. Test your build. See testsuite/README for more info.