configure.in

#
# Process this file with autoconf to produce a configure script
#
AC_PREREQ(2.59)
AC_INIT(tmLQCD, 5.1.6, curbach@gmx.de)
AC_CONFIG_HEADER(config.h)
AC_CONFIG_SRCDIR([hmc_tm.c])
AC_CANONICAL_HOST()
AC_PREFIX_DEFAULT($HOME)
AC_ARG_PROGRAM

if test "$host_vendor" = "cray"; then
  ac_cv_c_bigendian=yes
fi

AC_PROG_CC
AC_PROG_CC_C99
dnl AC_PROG_CC_STDC
AC_C_CONST
AC_C_INLINE
AC_C_RESTRICT
AC_F77_LIBRARY_LDFLAGS
AC_CHECK_TOOL(AR, ar, [ar])
LIBS="$LIBS $FLIBS -lm"

AC_PROG_LEX
dnl AC_PROG_LEX sets $LEX to ":" if neither lex nor flex are found! 
if test "$LEX" = ":"; then
  AC_MSG_ERROR([(F)LEX is required for building read_input.c. Please install it and run configure again.])
fi

AC_PROG_MAKE_SET
AC_PROG_RANLIB
AC_CHECK_PROG(CCDEP, gcc, "gcc", "$CC")
#(endian="", AC_DEFINE(LITTLE_ENDIAN,1,The endian of the architechture))

# AC_PROG_FC([ifort gfortran])
# AC_FC_FUNC(testfunc, )

LDFLAGS="$LDFLAGS -L\${HOME}/lib -L\${top_builddir}/lib"
CCLD=${CC}

USESUBDIRS="buffers cu io solver linalg"

AC_CHECK_HEADERS([stdint.h],
[ dnl for inttypes.h and stdint.h for uint_xxx types
  dnl if successful check for the actual types too
  AC_CHECK_TYPES([uint16_t, uint32_t, uint64_t],
                 [],
                 [AC_MSG_ERROR([stdint.h found but either uint16_t, uint32_t or uint64_t not found]) ]
                )
],
[
  dnl no inttypes.h or stdint.h found check common unsigned types
  dnl for sizes and make appropriate decisions in the lime_fixed_types.h file
  AC_CHECK_SIZEOF(unsigned char)
  AC_CHECK_SIZEOF(unsigned short)
  AC_CHECK_SIZEOF(unsigned int)
  AC_CHECK_SIZEOF(unsigned long)
  AC_CHECK_SIZEOF(unsigned long long)
]
)

AC_MSG_CHECKING(where to find lime)
AC_ARG_WITH(limedir,
  AS_HELP_STRING([--with-limedir[=dir]], [search lime in dir [default=./lime]]),
  lime_dir=$withval, lime_dir="./lime")
AC_MSG_RESULT($lime_dir)
LDFLAGS="$LDFLAGS -L${lime_dir}/lib/"
AC_CHECK_LIB([lime], [limeReaderNextRecord],[],
              [AC_MSG_ERROR([library liblime is missing or needed function is not available])])

#LIBS="$LIBS $FLIBS -lm"

AC_MSG_CHECKING(whether we want to use lemon)
AC_ARG_WITH(lemondir,
            AS_HELP_STRING([--with-lemondir[=dir]], [use lemon, to be found in dir]),
             [echo yes
              LEMON_AVAILABLE=1
              lemon_dir=$withval
              LDFLAGS="$LDFLAGS -L${lemon_dir}/lib"
              AC_CHECK_LIB([lemon],
                           [lemonReaderNextRecord],
                           [],
                           [AC_MSG_ERROR([library liblemon was not found])])],
             [echo no
              LEMON_AVAILABLE=0])

AC_MSG_CHECKING(whether we use the general geometry)
AC_ARG_ENABLE(indexindepgeom,
  AS_HELP_STRING([--enable-indexindepgeom], [enable Index independent addressing [default=no]]),
  enable_iig=$enableval, enable_iig=no)
if test $enable_iig = yes; then
  AC_MSG_RESULT(yes)
  AC_DEFINE(_INDEX_INDEP_GEOM,1,Index independent addressing)
else
  AC_MSG_RESULT(no)
fi

AC_MSG_CHECKING(whether we want to use MPI)
AC_ARG_ENABLE(mpi,
  AS_HELP_STRING([--enable-mpi], [enable use of mpi [default=yes]]),
  enable_mpi=$enableval, enable_mpi=yes)
if test $enable_mpi = yes; then
  AC_MSG_RESULT(yes)
  AC_DEFINE(MPI,1,Compile with MPI support)
else
  AC_MSG_RESULT(no)
fi

AC_MSG_CHECKING(whether to use QPX intrinsics)
AC_ARG_ENABLE(qpx,
  AS_HELP_STRING([--enable-qpx], [enable use of qpx intrinsics [default=no]]),
  enable_qpx=$enableval, enable_qpx=no)
if test $enable_qpx = yes; then
  AC_MSG_RESULT(yes)
  AC_DEFINE(BGQ,1,Compile with QPX intrinsics)
  AC_MSG_NOTICE([Compiling with QPX intrinsics on BGQ, enabling compiler optimizations for XLC.])
  OPTARGS="-O3 -qstrict=all -qtune=qp -qarch=qp -qmaxmem=-1"
  SOPTARGS="$OPTARGS" 
else
  AC_MSG_RESULT(no)
fi

AC_MSG_CHECKING(whether to use IBM BG/Q SPI for communications)
AC_ARG_ENABLE(spi,
  AS_HELP_STRING([--enable-spi], [enable use of SPI [default=no]]),
  enable_spi=$enableval, enable_spi=no)
if test $enable_spi = yes; then
  AC_MSG_RESULT(yes)
  AC_DEFINE(SPI,1,Compile with SPI for communications)
  SPI_FILES="DirectPut"
else
  AC_MSG_RESULT(no)
  SPI_FILES=""
fi


AC_MSG_CHECKING(whether we want to use OpenMP)
AC_ARG_ENABLE(omp,
  AS_HELP_STRING([--enable-omp], [enable use of OpenMP [default=yes]]),
  enable_omp=$enableval, enable_omp=yes)
if test $enable_omp = yes; then
  AC_MSG_RESULT(yes)
  AC_DEFINE(OMP,1,Compile with OpenMP support)
  AC_CHECK_HEADERS([omp.h],,[AC_MSG_ERROR([Cannot find OpenMP headers!])])
  AC_OPENMP
# -- AC_OPENMP provides a compiler-dependent OPENMP_CFLAGS so we can set it here
# on the BG/Q with XLC we force a special set of options for OpenMP support
  if test $enable_qpx = yes; then
    AC_MSG_NOTICE([Using OpenMP with XLC on BG/Q. Compiling with "-qsmp=omp:noauto:schedule=static".])
    CFLAGS="$CFLAGS -qsmp=omp:noauto:schedule=static"
    CPPFLAGS="$CPPFLAGS -qsmp=omp:noauto:schedule=static"
    LDFLAGS="$LDFLAGS -qsmp=omp:noauto:schedule=static" 
  else
    CFLAGS="$CFLAGS $OPENMP_CFLAGS"
    CPPFLAGS="$CPPFLAGS $OPENMP_CFLAGS"
    LDFLAGS="$LDFLAGS $OPENMP_CFLAGS"
  fi
else
  AC_MSG_RESULT(no)
fi

fftw_lib=/usr
AC_MSG_CHECKING(whether we want to use FFTW)
AC_ARG_ENABLE(fftw,
  AS_HELP_STRING([--enable-fftw], [enable use of fftw [default=no]]),
  enable_fftw=$enableval, enable_fftw=no)
if test $enable_fftw = yes; then
  AC_MSG_RESULT(yes)
  AC_DEFINE(HAVE_FFTW,1,Compile with FFTW support)
  LIBS="-lfftw3 ${LIBS}"
elif test $enable_fftw = no; then
  AC_MSG_RESULT(no)
else
  AC_MSG_RESULT(yes)
  AC_DEFINE(HAVE_FFTW,1,Compile with FFTW support)
  fftw_lib=${enable_fftw}
  LDFLAGS="$LDFLAGS -L${fftw_lib}/lib64"
  LIBS="-lfftw3 ${LIBS}"
  INCLUDES="-I${fftw_lib}/include ${INCLUDES}"
fi

if test $enable_mpi = yes; then
  AC_MSG_CHECKING(which parallelisation to use for MPI)
  AC_ARG_WITH(mpidimension,
    AS_HELP_STRING([--with-mpidimension[=n]], [use n dimensional parallelisation [default=1]]),
    withmpidimension=$withval, withmpidimension=1)
  if test $withmpidimension = 1; then
    AC_MSG_RESULT(n=1 [t])
    AC_DEFINE(PARALLELT,1,One dimensional parallelisation)
  elif test $withmpidimension = 2; then
    AC_MSG_RESULT(n=2 [xt])
    AC_DEFINE(PARALLELXT,1,Two dimensional parallelisation)
  elif test $withmpidimension = 3; then
    AC_MSG_RESULT(n=3 [xyt])
    AC_DEFINE(PARALLELXYT,1,Three dimensional parallelisation)
  elif test $withmpidimension = 4; then
    AC_MSG_RESULT(n=4 [xyzt])
    AC_DEFINE(PARALLELXYZT,1,Four dimensional parallelisation)
  elif test $withmpidimension = X; then
    AC_MSG_RESULT(n=1 [x])
    AC_DEFINE(PARALLELX,1, X parallelisation)
  elif test $withmpidimension = XY; then
    AC_MSG_RESULT(n=2 [xy])
    AC_DEFINE(PARALLELXY,1, XY parallelisation)
  elif test $withmpidimension = XYZ; then
    AC_MSG_RESULT(n=3 [xyz])
    AC_DEFINE(PARALLELXYZ,1, XYZ parallelisation)
  elif test $withmpidimension = T; then
    AC_MSG_RESULT(n=1 [t])
    AC_DEFINE(PARALLELT,1, T parallelisation)
  elif test $withmpidimension = XT; then
    AC_MSG_RESULT(n=2 [xt])
    AC_DEFINE(PARALLELXT,1, XT parallelisation)
  elif test $withmpidimension = XYT; then
    AC_MSG_RESULT(n=3 [xyt])
    AC_DEFINE(PARALLELXYT,1, XYT parallelisation)
  elif test $withmpidimension = XYZT; then
    AC_MSG_RESULT(n=4 [xyzt])
    AC_DEFINE(PARALLELXYZT,1, XYZT parallelisation)
  else
    AC_MSG_RESULT(unknown)
    AC_MSG_ERROR([Only t, xt, xyt, xyzt, x, xy, xyz parallelisation available])
  fi

  AC_MSG_CHECKING(whether we shall use persistent MPI calls for halfspinor)
  AC_ARG_WITH([persistentmpi],
    AS_HELP_STRING([--with-persistentmpi], [use persistent MPI calls for halfspinor [default=no]]),
    withpersistent=$withval, withpersistent=no)
  if test $withpersistent = yes; then
    AC_MSG_RESULT(yes)
    AC_DEFINE(_PERSISTENT,1,use persistent MPI calls for halfspinor)
  else
    AC_MSG_RESULT(no)
  fi

  AC_MSG_CHECKING(whether we shall use non-blocking MPI calls)
  AC_ARG_WITH([nonblockingmpi],
    AS_HELP_STRING([--with-nonblockingmpi], [use non-blocking MPI calls for spinor and gauge [default=yes]]),
    withnonblock=$withval, withnonblock=yes)
  if test $withnonblock = yes; then
    AC_MSG_RESULT(yes)
    AC_DEFINE(_NON_BLOCKING,1,use non-blocking MPI calls for spinor ang gauge)
  else
    AC_MSG_RESULT(no)
  fi
fi

AC_MSG_CHECKING([whether we want to fix volume at compiletime])
AC_ARG_WITH([fixedvolume],
  AS_HELP_STRING([--with-fixedvolume], [fix volume at compiletime [default=no]]),
  with_fixvol=$withval, with_fixvol=no)
if test $with_fixvol = yes; then
  AC_MSG_RESULT(yes)
  AC_DEFINE(FIXEDVOLUME,1,Fixed volume at compiletime)
  AC_CONFIG_FILES([fixed_volume.h])
else
  AC_MSG_RESULT(no)
fi

AC_MSG_CHECKING([whether we want to use KOJAK instrumentalisation])
AC_ARG_WITH([kojakinst],
  AS_HELP_STRING([--with-kojakinst], [instrumentalise for KOJAK [default=no]]),
  with_kojakinst=$withval, with_kojakinst=no)
if test $with_kojakinst = yes; then
  AC_MSG_RESULT(yes)
  CC="kinst-pomp ${CC}"
else
  AC_MSG_RESULT(no)
fi

AC_MSG_CHECKING(whether we want to use lapack and blas)
AC_ARG_WITH(lapack,
  AS_HELP_STRING([--with-lapack], [enable use of lapack [default=yes]]),
  with_lapack=$withval, with_lapack=yes)
if test "$with_lapack" = yes; then
  AC_MSG_RESULT(yes)
  LAPACKLIB=
  AC_DEFINE(HAVE_LAPACK,1,lapack available)
elif test "$with_lapack" != no; then
  AC_MSG_RESULT(yes)
  LIBS="$withval $LIBS"
  with_lapack=yes
  AC_DEFINE(HAVE_LAPACK,1,lapack available)
else
  AC_MSG_RESULT(no)
  AC_MSG_ERROR([lapack is needed! Will stop here.])
fi

if test $enable_mpi = yes; then
  dnl In general one cannot run mpi programs directly
  dnl thats why we need here cross_compiling=yes
  dnl for non CRAY
  if test "$host_vendor" != "cray"; then
    cross_compiling=yes
  fi
fi

dnl for the case of other configure scripts
dnl AC_CONFIG_SUBDIRS( rng )

dnl check for clock_gettime and set correct library flag if one is required
dnl (as done by AC_CHECK_LIB)
AC_CHECK_FUNCS(clock_gettime, [], [AC_CHECK_LIB(rt, clock_gettime)])


dnl with glibc and c99 mode the timespec required for clock_gettime is only
dnl available in POSIX 199309L compatibility mode when clock_gettime is in librt
if test "$ac_cv_lib_rt_clock_gettime" = "yes"; then
  AC_DEFINE(HAVE_CLOCK_GETTIME,1)
  CFLAGS="$CFLAGS -D_POSIX_C_SOURCE=199309L"
  AC_MSG_NOTICE([Instructing the compiler to use POSIX 199309L])
fi

dnl Checks for lapack and defines proper name mangling scheme for
dnl linking with f77 code
AC_F77_FUNC(zheev)
if test "$zheev" = "zheev"; then
  AC_DEFINE(NOF77_,1,Fortran has no extra _)
fi
AC_SEARCH_LIBS([$zheev],[lapack], [], [AC_MSG_ERROR([Cannot find lapack])])

dnl Checks for header files.
AC_HEADER_STDC
AC_CHECK_HEADERS([float.h libintl.h limits.h stdint.h stdlib.h string.h strings.h sys/time.h unistd.h endian.h])
AC_CHECK_HEADER( getopt.h, [])

dnl Checks for typedefs, structures, and compiler characteristics.
AC_C_CONST
AC_TYPE_OFF_T
AC_TYPE_SIZE_T
AC_HEADER_TIME

dnl Checks for library functions.
AC_SYS_LARGEFILE
AC_FUNC_FSEEKO
AC_FUNC_MALLOC
AC_TYPE_SIGNAL
AC_CHECK_FUNCS([gettimeofday pow sqrt])

dnl We now define some replacement variables
AC_SUBST(OPTARGS)
AC_SUBST(SOPTARGS)
AC_SUBST(INCLUDES)
AC_SUBST(AUTOCONF)
AC_SUBST(SOLVEROUT)
AC_SUBST(CCDEP)
AC_SUBST(CCLD)
AC_SUBST(DEPFLAGS)
AC_SUBST(DEBUG_FLAG)
AC_SUBST(PROFILE_FLAG)
AC_SUBST(XCHANGELIB)
AC_SUBST(XCHANGEDIR)
AC_SUBST(XLIB)
AC_SUBST([LEMON_AVAILABLE])
AC_SUBST(SPI_FILES)

INCLUDES="$INCLUDES -I\$(HOME)/include/ -I. -I\${abs_top_builddir}/  -I\${abs_top_srcdir}/ -I${lime_dir}/include/ -I${lemon_dir}/include/"
DEPFLAGS="$DEPFLAGS"

  AC_MSG_CHECKING(what alignment we want for arrays)
  AC_ARG_WITH(alignment,
    [AS_HELP_STRING([--with-alignment[=n]], [align arrays to 0, 16, 32, 64 or 128 bits [default=auto]])],
    withalign=$withval, withalign=auto)
  if test "$withalign" = "no"; then
    withalign=0
  fi
  if test "$withalign" = "0"; then
    AC_MSG_RESULT(none)
    AC_DEFINE(ALIGN_BASE, 0x00, [Align base])
    AC_DEFINE(ALIGN, [])
  elif test $withalign = 16; then
    AC_MSG_RESULT(16 bits)
    AC_DEFINE(ALIGN_BASE, 0x0f, [Align base])
    AC_DEFINE(ALIGN, [__attribute__ ((aligned (16)))])
  elif test $withalign = 32; then
    AC_MSG_RESULT(32 bits)
    AC_DEFINE(ALIGN_BASE, 0x1f, [Align base])
    AC_DEFINE(ALIGN, [__attribute__ ((aligned (32)))])
  elif test $withalign = 64; then
    AC_MSG_RESULT(64 bits)
    AC_DEFINE(ALIGN_BASE, 0x3f, [Align base])
    AC_DEFINE(ALIGN, [__attribute__ ((aligned (64)))])
  elif test $withalign = 128; then
    AC_MSG_RESULT(128 bits)
    AC_DEFINE(ALIGN_BASE, 0x7f, [Align base])
    AC_DEFINE(ALIGN, [__attribute__ ((aligned (128)))])
  elif test $withalign = auto; then
    withautoalign=0
    AC_MSG_RESULT(auto)
    AC_DEFINE(ALIGN_BASE, 0x00, [Align base])
    AC_DEFINE(ALIGN, [], [])
  else
    AC_MSG_RESULT(Unusable value for array alignment)
    AC_MSG_ERROR([Only alignment to 0, 16, 32, 64 or 128 bits, or auto alignment available])
  fi


dnl in the following we check for extra options
if test "$host_cpu" = "i686" || test "$host_cpu" = "x86_64"; then

  AC_MSG_CHECKING(whether we want to use P4 instructions)
  AC_ARG_ENABLE(p4,
    AS_HELP_STRING([--enable-p4],[enable use of P4 instructions [default=no]]),
    enable_p4=$enableval, enable_p4=no)
  if test $enable_p4 = yes; then
    AC_MSG_RESULT(yes)
    AC_DEFINE(P4,1,Use Pentium4 instructions)
    if test $withalign = auto; then
      if test $withautoalign = 0 || test $withautoalign = 16 ||test $withautoalign = 32; then
        AC_MSG_RESULT(changing array alignment to 64 bits for P4 instructions)
        AC_DEFINE(ALIGN_BASE, 0x3f, [Align base])
        AC_DEFINE(ALIGN, [__attribute__ ((aligned (64)))])
        withautoalign=64
      fi
    elif test $withalign = 0 || test $withalign = 16 ||test $withalign = 32; then
      AC_MSG_ERROR([alignment incompatible with P4 instructions (64 bits required)!])
    fi
  else
    AC_MSG_RESULT(no)
  fi

  AC_MSG_CHECKING(whether we want to use Opteron instructions)
  AC_ARG_ENABLE(opteron,
    AS_HELP_STRING([--enable-opteron], [enable use of Opteron instructions [default=no]]),
    enable_opteron=$enableval, enable_opteron=no)
  if test $enable_opteron = yes; then
    AC_MSG_RESULT(yes)
    AC_DEFINE(OPTERON,1,Use Opteron instructions)
    if test $withalign = auto; then
      if test $withautoalign = 0 || test $withautoalign = 16; then
        AC_MSG_RESULT(changing array alignment to 32 bits for Opteron instructions)
        AC_DEFINE(ALIGN_BASE, 0x1f, [Align base])
        AC_DEFINE(ALIGN, [__attribute__ ((aligned (32)))])
        withautoalign=0x1f
      fi
    elif test $withalign = 0 || test $withalign = 16; then
      AC_MSG_ERROR([alignment incompatible with Opteron instructions (32 bits required)!])
    fi
  else
    AC_MSG_RESULT(no)
  fi

  AC_MSG_CHECKING(whether we want to use SSE2 instructions)
  AC_ARG_ENABLE(sse2,
    AS_HELP_STRING([--enable-sse2], [enable use of SSE2 instructions [default=no]]),
    enable_sse2=$enableval, enable_sse2=no)
  if test $enable_sse2 = yes; then
    AC_MSG_RESULT(yes)
    if test $withalign = auto; then
      if test $withautoalign = 0 || test $withautoalign = 16; then
        AC_MSG_WARN(changing array alignment to 32 bits for SSE2 instructions)
        AC_DEFINE(ALIGN_BASE, 0x1f, [Align base])
        AC_DEFINE(ALIGN, [__attribute__ ((aligned (32)))])
        withautoalign=0x32
      fi
    elif test $withalign = 0 || test $withalign = 16; then
      AC_MSG_ERROR([alignment incompatible with SSE2 instructions (32 bits required)]!)
    fi
  else
    AC_MSG_RESULT(no)
  fi

  AC_MSG_CHECKING(whether we want to use SSE3 instructions)
  AC_ARG_ENABLE(sse3,
    AS_HELP_STRING([--enable-sse3], [enable use of SSE3 instructions [default=no]]),
    enable_sse3=$enableval, enable_sse3=no)
  if test $enable_sse3 = yes; then
    AC_MSG_RESULT(yes)
    if test $withalign = auto; then
      if test $withautoalign = 0 || test $withautoalign = 16; then
        AC_MSG_RESULT(changing array alignment to 32 bits for SSE3 instructions)
        AC_DEFINE(ALIGN_BASE, 0x1f, [Align base])
        AC_DEFINE(ALIGN, [__attribute__ ((aligned (32)))])
        withautoalign=32
      fi
    elif test $withalign = 0 || test $withalign = 16; then
      AC_MSG_ERROR([alignment incompatible with SSE2 instructions (32 bits required)])
    fi
  else
    AC_MSG_RESULT(no)
  fi
fi

AC_MSG_CHECKING(whether we want to use gprof as profiler)
AC_ARG_WITH(gprof,
  AS_HELP_STRING([--with-gprof], [use of gprof profiler [default=no]]),
  enable_gprof=$withval, enable_gprof=no)
if test $enable_gprof = yes; then
  AC_MSG_RESULT(yes)
    if test "$host_cpu" = "powerpc" && test "$host_vendor" = "ibm"; then
      PROFILE_FLAG="-pg -qfullpath -g"
    else
      PROFILE_FLAG="-pg -g"
    fi
else
  AC_MSG_RESULT(no)
  PROFILE_FLAG=
fi

AC_MSG_CHECKING(whether we shall use rts dram window)
AC_ARG_WITH([bgldram],
  AS_HELP_STRING([--with-bgldram], [use BGL dram window (BGL only!) [default=yes]]),
  with_bgldram=$withval, with_bgldram=yes)
if test $with_bgldram = yes; then
  AC_MSG_RESULT(yes)
  AC_DEFINE(_USE_BGLDRAM,1,use BGL dram window)
else
  AC_MSG_RESULT(no)
fi

dnl Now we have to set all Flags and compiler properly
XLCGREP=`$CC -V 2>&1 | grep -i xlc`
if test "$XLCGREP" != ""; then
  XLC="yes"
  AC_DEFINE(XLC,1,Are we using the IBM xlc compiler?)
fi
PGCC=`$CC -V 2>&1 | grep pgcc`
ICC=`$CC -V 2>&1 | grep -i intel`

dnl first for PC's
if test "$host_cpu" = "i686" || test "$host_cpu" = "x86_64"; then
dnl the GNU compiler
  if test "$GCC" = yes && test "$ICC" = ""; then
    DEPFLAGS="-MM"
    CFLAGS="$CFLAGS -pedantic -Wall"
    OPTARGS='-O'
    SOPTARGS='-O'

    if test $enable_sse3 = yes; then
      echo Using SSE3 and SSE2 macros!
      AC_DEFINE(SSE3,1,Compile with SSE3 support)
      DEPFLAGS="$DEPFLAGS -DSSE3"
      if test "$host_cpu" = "x86_64"; then
        CFLAGS="$CFLAGS -mfpmath=387"
      fi
    elif test $enable_sse2 = yes; then
      DEPFLAGS="$DEPFLAGS -DSSE2"
      AC_DEFINE(SSE2,1,Compile with SSE2 support)
      if test "$host_cpu" = "x86_64"; then
        CFLAGS="$CFLAGS -mfpmath=387"
      fi
    fi

    if test "$host_cpu" = "x86_64"; then
      AC_DEFINE(_x86_64,1,x86 64 Bit architecture)
    fi
    CCDEP="$CC"
    if test $enable_mpi = yes; then
      CCDEP="gcc"
    fi
    DEBUG_FLAG="-g"
dnl other compilers
  else
dnl check for pgcc
    if test "$PGCC" != ""; then
      DEPFLAGS="-M"
      echo "We are using the Portland Group C compiler!"
      OPTARGS="-O2"
      SOPTARGS="-O2"
      DEBUG_FLAG="-g"
      PROFILE_FLAG="-p -g"
      CCDEP="$CC"

dnl check for icc
    elif test "$ICC" != ""; then
      echo "We are using the Intel C compiler!"
      DEPFLAGS="-M"
      OPTARGS="-O3"
      SOPTARGS="-O3"
      DEBUG_FLAG="-g"
      PROFILE_FLAG="-p -g"
      CCDEP="$CC"

    else
      DEPFLAGS="-M"
      CFLAGS="$CFLAGS -O"
      DEBUG_FLAG="-g"
      CCDEP="$CC"
    fi
  fi

# The MareNostrum: powerpc on a linux system
# this will also evaluate to "true" on BG/Q with XLC
elif test "$host_cpu" = "powerpc64" && test "$host_vendor" = "unknown" && test "$host_os" = "linux-gnu"; then

  DEBUGFLAG="-g"
  if test "$XLC" = "yes"; then
    CFLAGS="-qsrcmsg $CFLAGS"
    DEBUGFLAG="$DEBUGFLAG -qfullpath"
  fi

  OPTARGS="$OPTARGS -O3"
  SOPTARGS="$OPTARGS"
  if test "$CCDEP" = "gcc"; then
    DEPFLAGS="-MM"
  else
    DEPFLAGS="-M"
  fi

#The BLue Gene/L
elif test "$host_cpu" = "powerpc" && test "$host_vendor" = "ibm" && test "$host_os" = "blrts"; then
  if test "$with_bgldram" = yes; then
    if (test -e /bgl/local/bin/blrts_gcc); then
      BLRTSGCC=/bgl/local/bin/blrts_gcc
    elif (test -e /bgl/BlueLight/ppcfloor/blrts-gnu/bin/powerpc-bgl-blrts-gnu-gcc); then
      BLRTSGCC=/bgl/BlueLight/ppcfloor/blrts-gnu/bin/powerpc-bgl-blrts-gnu-gcc
    else
      AC_MSG_ERROR([Sorry, don't know where to find blrts_gcc, see README.bgl!])
    fi
    CCLD="$BLRTSGCC -Xlinker --script=./elf32ppcblrts.x"
    if (!(test -s ./elf32ppcblrts.x)); then
      AC_MSG_ERROR([Sorry, elf32ppcblrts.x is missing, see README.bgl!])
    fi
  fi
  DEBUGFLAG="-g"
  OPTARGS="-O3"
  SOPTARGS="-O3"
  AC_DEFINE(BGL,1,[Optimize for Blue Gene/L])
  if test $withalign = auto; then
    if test $withautoalign = 0; then
      AC_MSG_RESULT(changing array alignment to 16 bits for BGL instructions)
      AC_DEFINE(ALIGN_BASE, 0x0f, [Align base])
      AC_DEFINE(ALIGN, [__attribute__ ((aligned (16)))])
      withautoalign=16
    fi
  elif test $withalign = 0; then
    AC_MSG_ERROR([alignment incompatible with BGL instructions (16 bits required)!])
  fi

  if test "$XLC" = "yes"; then
    CFLAGS="-qsrcmsg $CFLAGS"
    OPTARGS="$OPTARGS -qarch=440d -qtune=440"
    SOPTARGS="$SOPTARGS -qarch=440d -qtune=440"
    DEBUGFLAG="$DEBUGFLAG -qfullpath"
# OPTARGS="-qhot" leads to wrong code
  fi
  LIBS="-lmpich.rts -lfmpich.rts -lmsglayer.rts -lrts.rts -ldevices.rts $LIBS"
  LDFLAGS="$LDFLAGS -L/bgl/BlueLight/ppcfloor/bglsys/lib"
  if test $with_lapack = yes; then
    LIBS="-lesslbg -llapack.rts -lesslbg -lxlf90 -lxlfmath -lxl -lxlopt $LIBS"
    LDFLAGS="$LDFLAGS -L/opt/ibmcmp/xlf/bg/10.1/blrts_lib -L/bgl/local/lib/ -L/opt/ibmmath/lib/"
  fi

  if test "$CCDEP" = "gcc"; then
    DEPFLAGS="-MM"
  else
    DEPFLAGS="-M"
  fi
  CPPFLAGS="-I/bgl/BlueLight/ppcfloor/bglsys/include"
  INCLUDES="$INCLUDES -I/bgl/BlueLight/ppcfloor/bglsys/include/"

#The BLue Gene/P
elif test "$host_cpu" = "powerpc" && test "$host_vendor" = "ibm" && test "$host_os" = "bprts"; then
  CFLAGS="$CFLAGS"
  DEBUGFLAG="-g"
  OPTARGS="-O3"
  SOPTARGS="-O3"
  AC_DEFINE(BGL,1,[Optimize for Blue Gene/L])
  AC_DEFINE(BGP,1,[Optimize for Blue Gene/P])
  if test $withalign = auto; then
    if test $withautoalign = 0; then
      AC_MSG_RESULT(changing array alignment to 16 bits for BGP instructions)
      AC_DEFINE(ALIGN_BASE, 0x0f, [Align base])
      AC_DEFINE(ALIGN, [__attribute__ ((aligned (16)))])
      withautoalign=16
    fi
  elif test $withalign = 0; then
    AC_MSG_ERROR([alignment incompatible with BGP instructions (16 bits required)!])
  fi

  if test "$XLC" = "yes"; then
    CFLAGS="-qsrcmsg $CFLAGS"
    OPTARGS="$OPTARGS -qarch=450d -qtune=450"
    SOPTARGS="$SOPTARGS -qarch=450d -qtune=450"
    DEBUGFLAG="$DEBUGFLAG -qfullpath"
# OPTARGS="-qhot" leads to wrong code
  fi
#  LIBS="-lxlf90_r -lxlomp_ser -lxl -lxlopt -lxlfmath -ldl -lrt -lpthread  $LIBS"
#  LDFLAGS="$LDFLAGS -L/bgsys/local/lib/ -L/opt/ibmcmp/xlf/bg/11.1/lib  -L/bgsys/drivers/ppcfloor/comm/"
#  if test $with_lapack = yes; then
#    LIBS="-lesslbg -llapack -lesslbg $LIBS"
#    LDFLAGS="$LDFLAGS -L/opt/ibmmath/lib/"
#  fi

  if test "$CCDEP" = "gcc"; then
    DEPFLAGS="-MM"
  else
    DEPFLAGS="-M"
  fi
  CPPFLAGS="-I/bgsys/drivers/ppcfloor/arch/include/ -I/bgsys/drivers/ppcfloor/comm/include"
  INCLUDES="$INCLUDES -I/bgsys/local/include/ -I/bgsys/drivers/ppcfloor/arch/include/ -I/bgsys/drivers/ppcfloor/comm/include"



# The IBM Power PC
elif test "$host_cpu" = "powerpc" && test "$host_vendor" = "ibm"; then
  CFLAGS="$CFLAGS -q64 -qsrcmsg"
  LDFLAGS="$LDFLAGS -q64"
  OPTARGS="-O3 -qhot"
  SOPTARGS="-O3 -qhot"
  DEBUG_FLAG="-qfullpath -g"
  if test "$CCDEP" = "gcc"; then
    DEPFLAGS="-MM"
  else
    DEPFLAGS="-M"
  fi

# The CRAY
elif test "$host_vendor" = "cray"; then
  echo
  echo "Hey, we are on a cray, you should take some time for this..."
  echo "get yourself a coffee or so!"
  echo
  CFLAGS="$CFLAGS -dp"
  AC_DEFINE(CRAY,1,We are on a CRAY)
  OPTARGS="-O3"
  SOPTARGS="-O3"
  DEBUG_FLAG="-g"
  CCDEP="$CC"
  DEPFLAGS="-M"

else
  AC_CHECK_PROG(CCDEP, gcc, "gcc", "$CC")
  if test "$CCDEP" = "gcc"; then
    DEPFLAGS="-MM"
  else
    DEPFLAGS="-M"
  fi
  OPTARGS=
  SOPTARGS=
fi


AC_MSG_CHECKING(whether we want to switch on optimisation)
AC_ARG_ENABLE(optimize,
  AS_HELP_STRING([--enable-optimize], [enable optimisation [default=yes]]),
  enable_optimize=$enableval, enable_optimize=yes)
if test $enable_optimize = no; then
  AC_MSG_RESULT(no)
  OPTARGS=
  SOPTARGS=
else
  AC_MSG_RESULT(yes)
fi

AC_MSG_CHECKING(whether we want to use a copy of the gauge field)
AC_ARG_ENABLE(gaugecopy,
  AS_HELP_STRING([--enable-gaugecopy], [enable use of a copy of the gauge field [default=yes]]),
  enable_gaugecopy=$enableval, enable_gaugecopy=yes)
if test $enable_gaugecopy = yes; then
  AC_MSG_RESULT(yes)
  AC_DEFINE(_GAUGE_COPY,1,Construct an extra copy of the gauge fields)
else
  AC_MSG_RESULT(no)
fi

AC_MSG_CHECKING(whether we want to use a Dirac Op. with halfspinor exchange)
AC_ARG_ENABLE(halfspinor,
  AS_HELP_STRING([--enable-halfspinor], [use a Dirac Op. with halfspinor exchange [default=yes]]),
  enable_halfspinor=$enableval, enable_halfspinor=yes)
if test $enable_halfspinor = yes; then
  AC_MSG_RESULT(yes)
  AC_DEFINE(_USE_HALFSPINOR,1,Exchange only a halfspinor in the Dirac Operator)
  if test $enable_gaugecopy = no; then
    AC_MSG_WARN([switching on gaugecopy for Dirac operator with halfspinor!])
    AC_DEFINE(_GAUGE_COPY,1,Construct an extra copy of the gauge fields)
  fi
else
  AC_MSG_RESULT(no)
fi

AC_MSG_CHECKING(whether we want to use shmem API)
AC_ARG_ENABLE(shmem,
  AS_HELP_STRING([--enable-shmem],[use shmem API [default=no]]),
  enable_shmem=$enableval, enable_shmem=no)
if test $enable_shmem = yes; then
  AC_MSG_RESULT(yes)
  AC_DEFINE(_USE_SHMEM,1,Use shmem API)
  LIBS="$LIBS -lsma"
else
  AC_MSG_RESULT(no)
fi

AC_MSG_CHECKING(whether we want to use timeslice-splitted communications)
AC_ARG_ENABLE(tsplitpar,
  AS_HELP_STRING([--enable-tsplitpar],[enable timeslice-splitted communications [default=no]]),
  enable_tsp=$enableval, enable_tsp=no)
if test $enable_tsp = yes; then
  AC_MSG_RESULT(yes)
  AC_DEFINE(_USE_TSPLITPAR,1,timeslice-splitted communications)
else
 AC_MSG_RESULT(no)
fi

AC_MSG_CHECKING(whether we want to compute the LapH eigenvalues)
AC_ARG_ENABLE(laph,
  AS_HELP_STRING([--enable-laph], [enable computation of LapH eigensystem [default=no]]),
  enable_laph=$enableval, enable_laph=no)
if test $enable_laph = yes; then
  AC_MSG_RESULT(yes)
  AC_DEFINE(WITHLAPH,1,LapH eigensystem)
else
 AC_MSG_RESULT(no)
fi


AC_MSG_CHECKING(whether we want to use CUDA GPU)
AC_ARG_ENABLE(gpu,
  AS_HELP_STRING([--enable-gpu],[use GPU [default=no]]),
  usegpu=$enableval, usegpu=no)
if test $usegpu = yes; then
  AC_MSG_RESULT(yes)
  AC_DEFINE(HAVE_GPU,1,Using CUDA GPU)
  NVCC="nvcc"
  USESUBDIRS="$USESUBDIRS GPU"
  GPUDIR="GPU"
  LIBS="$LIBS -lcuda -lcudart -lcublas"
  
  AC_MSG_CHECKING([where to search for CUDA libs])
  AC_ARG_WITH(cuda,
    AS_HELP_STRING([--with-cuda[=dir]], [use CUDA GPU with lib dir [default=/usr/local/cuda/lib]]),
    cuda_dir=$withval, cuda_dir="/usr/local/cuda/lib")
  AC_MSG_RESULT($cuda_dir)
  if test $usegpu = yes; then  
    LDFLAGS="$LDFLAGS -L$cuda_dir"
  fi


  AC_MSG_CHECKING([CUDA compile args])
  AC_ARG_WITH(cudacompileargs,
    AS_HELP_STRING([--with-cudacompileargs[=string]], [use CUDA compile args [default="--gpu-architecture sm_13 --use_fast_math -O3"]]),
    cuda_compileargs=$withval, cuda_compileargs="--gpu-architecture sm_13 --use_fast_math -O3")
  AC_MSG_RESULT($cuda_compileargs)
  if test $usegpu = yes; then  
    GPUCFLAGS="$GPUCFLAGS $cuda_compileargs"
  fi
  if test $enable_mpi = yes; then
    GPUMPICOMPILER="--compiler-bindir mpicc"
    if test $withmpidimension != 1; then
      AC_MSG_ERROR(ERROR! The GPU Code is only parallelized in t-direction so far!)
    fi
  else
    GPUMPICOMPILER=""
  fi
else
  AC_MSG_RESULT(no)
  NVCC=""
fi


AC_SUBST(USESUBDIRS)
AC_SUBST(NVCC)
AC_SUBST(GPUDIR)
AC_SUBST(GPUCFLAGS)
AC_SUBST(GPUMPICOMPILER)


AC_MSG_CHECKING(checking consistency)
if test $enable_mpi = yes ; then
 if test $enable_iig = yes && test $withpersistent = yes ; then
  AC_MSG_ERROR(ERROR! indexindepgeom is not compatible with persistent communications )
 fi
 if test $enable_iig = yes && test $enable_shmem = yes ; then
   AC_MSG_ERROR(ERROR! indexindepgeom is not compatible with shmem API )
 fi
 if test $enable_tsp = yes && test $enable_iig = no; then
   AC_MSG_ERROR(ERROR! tsplitpar needs indexindepgeom)
 fi
 if test $enable_tsp = yes && test $enable_sse2 != yes ; then
   AC_MSG_ERROR(ERROR! tsplitpar needs at least SSE2 )
 fi
 if test $enable_tsp = yes && test $enable_gaugecopy != yes ; then
   AC_MSG_ERROR(ERROR! tsplitpar needs gaugecopy)
 fi
 if test $enable_laph = yes && test $enable_tsp != yes ; then
   AC_MSG_ERROR(ERROR! laph needs tsplitpar)
 fi
fi

if test ! -e lib; then
  mkdir lib
fi

dnl create the test and tests directory here
if test ! -e test; then
  mkdir test
fi

if test ! -e tests; then
  mkdir tests
fi

if test ! -e tests/regressions; then
  mkdir tests/regressions
fi


LIBS="-lhmc -lsolver -llinalg -lhmc -lio $LIBS"
AUTOCONF=autoconf

for i in $USESUBDIRS
do
  make_files="$make_files $i/Makefile"
done

AC_CONFIG_FILES([Makefile $make_files])

AC_OUTPUT