Any additional information related to this implementation, that came to light after the release of this product, will be described below.

1. Additional Systems / Compilers Tested

   This implementation has been successfully tested using the following:

   •    Apple MacPro4.1, Quad-Core Intel Xeon, @ 2.26GHz
        Mac OS X 10.6.8, Darwn Kernel 10.8.0 (Snow Leopard)
        Intel Fortran Composer XE 2011 update 6, (2011.6.038),
         Version 12.1.0.038 Build 20110811
     

   •    Apple Macmini5.1, Intel Core i5, @ 2.30GHz
        Mac OS X 10.7.2, Darwn Kernel 11.2.0 (Lion)
        Intel Fortran Composer XE 2011 update 7, (2011.7.246),
         Version 12.1.1.246 Build 20111011
     

        Apple Macmini5.1, Intel Core i5, @ 2.30GHz
        Mac OS X 10.7.4, Darwin Kernel 11.4.0 (Lion)
        Intel Fortran Composer XE 2013
         Version 13.0.1.119 Build 20121010
     

2. Using the NAG Library with the GNU Fortran compiler

   The NAG libraries supplied with this implementation were built using the Intel Fortran compiler, but with care it is possible to call them from the GNU compiler, gfortran. The information below has been tested using gfortran version 4.6.0 on a Mac running Mac OS X 10.7.0 (Snow Leopard), and all NAG example programs were found to run correctly. It may or may not work for other versions of the operating system or compiler.

   Differences to take into account when using gfortran

   There are several things to bear in mind:
   • This version of the NAG Library uses 64-bit integers - it was built using Intel compiler switch -i8. The equivalent switch for gfortran is -fdefault-integer-8.

   • The modules supplied in the distribution directory nag_interface_blocks were compiled with the Intel compiler and are not suitable for use by the gfortran compiler. Hence it is necessary to recompile the module files with gfortran before you can use them - see below for instructions.

   • The Fortran datatype LOGICAL behaves differently between the two compilers. For the Intel compiler, a LOGICAL value is stored in 8 bytes, with the value .FALSE. being represented by the integer 0, and the value .TRUE. being represented by the integer -1. However, any non-zero value is regarded as being .TRUE.. For gfortran, .FALSE. is represented by 0, and .TRUE. is represented by 1. Any other values stored in the logical datatype and used by gfortran will give undefined behaviour.

     A consequence of these rules is that a gfortran LOGICAL variable can be passed to Intel Fortran and operated on without problem, but not vice versa. Hence, output arguments of type LOGICAL returned by NAG routines may not be interpreted correctly by a gfortran calling program.

     To avoid this problem, it is necessary to pass arguments of type 64-bit INTEGER from gfortran instead of LOGICAL. This affects a small number of NAG routines:

       a00acf (function return value)
       a00adf (argument LICVAL)
       d02lxf (argument START)
       d02ngf (argument LDERIV)
       d02nhf (argument LDERIV)
       d02njf (argument LDERIV)
       d02nnf (argument LDERIV)
       d02nyf (argument ALGEQU)
       d02qff (argument ROOT)
       d02qgf (argument ROOT)
       d02qwf (argument ALTERG)
       f02wdf (argument SVD)
       f06blf (argument FAIL)
       f06clf (argument FAIL)
       f08pkf/DHSEIN (argument SELECT)
       f08pxf/ZHSEIN (argument SELECT)
       f08qkf/DTREVC (argument SELECT)
       g02daf (argument SVD)
       g02ddf (argument SVD)
       g02dnf (argument EST)
       g02gnf (argument EST)
       g02tbf (argument X)
       g05kef (function return value)
       g07gbf (function return value)
       g08ahf (argument TIES)
       g11saf (argument X)
       g11sbf (argument X)
       x02daf (function return value)
       x02djf (function return value)
     

     Routines which have arguments of type LOGICAL that are input only should work correctly without this workaround.

   • A few NAG routines are functions which return a value of type COMPLEX. The mechanism for returning values of this type varies between compilers, and Intel Fortran is incompatible with gfortran. It is necessary instead to call the functions as though they were subroutines, with an extra leading argument of type COMPLEX. NAG routines affected in this way are:

       c06lbf
       f06clf
       f16glf(blas_zsum)
       f06gaf(zdotu)
       f06gbf(zdotc)
       f06grf(zdotui)
       f06gsf(zdotci)
       s01eaf
       s14aff
       s14agf
       s15ddf
       s21daf
     

   • Arguments of type CHARACTER*(*) - i.e. character strings of non-fixed length - also require special treatment. Such character strings are passed to subroutine calls along with an extra "hidden" character length argument. This hidden argument is passed by value (the usual Fortran argument passing mechanism is to pass by reference). In this implementation, the Intel compiler passes these arguments as 64-bit integer values, whereas gfortran passes them as 32-bit integer values even when compiling with the -fdefault-integer-8 switch.

     Passing an incorrect character string length argument must be avoided if the subroutine call is to work correctly. When calling from gfortran it is therefore necessary to pass these arguments explicitly, and make them be 64-bit integers.

     Character strings declared by a NAG routine as being of fixed length (e.g. CHARACTER or CHARACTER*6) cause no problems, because the hidden argument length value is ignored by the called subroutine in any case. Routines containing both fixed-length and non-fixed-length strings need the hidden argument to be explicitly passed for all strings in order to preserve correct ordering of arguments.

   Modified modules and example programs

   A set of NAG Library modules which have been modified to account for the considerations above is available as a compressed tar file. The file also contains an entire set of NAG Library example programs modified where necessary to match the declarations in the module files, with appropriate handling of LOGICAL, COMPLEX and CHARACTER arguments. It is suggested that you use these example programs as templates for your own calls to the NAG Library.

   The tar file contains the following directory structure:

     postrelease -|- nag_interface_blocks_gfortran (.f90 module files)
                  |
                  |- examples -|- source_gfortran  (modified example programs)
   

   Note that the data files used by the example programs do not need to change - the ones distributed with the library are fine for use with gfortran.

   Compiling the NAG Library modules for use with gfortran

   To compile the interface blocks, go into directory nag_interface_blocks_gfortran and use commands like these (N.B. compilation order is important):

     gfortran -fdefault-integer-8 -c nag_precisions.f90
     gfortran -fdefault-integer-8 -c nag_blas_consts.f90
     gfortran -fdefault-integer-8 -c nag_a_ib.f90
     gfortran -fdefault-integer-8 -c nag_c_ib.f90
     gfortran -fdefault-integer-8 -c nag_d_ib.f90
     gfortran -fdefault-integer-8 -c nag_e_ib.f90
     gfortran -fdefault-integer-8 -c nag_f_ib.f90
     gfortran -fdefault-integer-8 -c nag_g_ib.f90
     gfortran -fdefault-integer-8 -c nag_h_ib.f90
     gfortran -fdefault-integer-8 -c nag_m_ib.f90
     gfortran -fdefault-integer-8 -c nag_p_ib.f90
     gfortran -fdefault-integer-8 -c nag_s_ib.f90
     gfortran -fdefault-integer-8 -c nag_x_ib.f90
     gfortran -fdefault-integer-8 -c nag_long_names.f90
     gfortran -fdefault-integer-8 -c nag_library.f90
   

   Accessing the NAG Library from gfortran

   In this section we use the same notation [INSTALL_DIR] to describe the installed location of the NAG Library material as used in the Users' Note. We further assume that the modules compiled for gfortran as described above are in directory [INSTALL_DIR]/nag_interface_blocks_gfortran.

   To use the NAG Fortran Library and the supplied MKL libraries, you may link in the following manner:

      gfortran -fdefault-integer-8 driver.f \
          -I[INSTALL_DIR]/nag_interface_blocks_gfortran \
          [INSTALL_DIR]/lib/libnag_mkl.a \
          [INSTALL_DIR]/mkl_intel64/libmkl_intel_ilp64.a \
          -L[INSTALL_DIR]/mkl_intel64 -lmkl_intel_thread -lmkl_core \
          [INSTALL_DIR]/rtl/libsvml.a [INSTALL_DIR]/rtl/libimf.a \
          [INSTALL_DIR]/rtl/libifcoremt.a [INSTALL_DIR]/rtl/libirc.a -lpthread
   

   where driver.f is your application program;
   or

      gfortran -fdefault-integer-8 driver.f \
          -I[INSTALL_DIR]/nag_interface_blocks_gfortran \
          [INSTALL_DIR]/lib/libnag_mkl.dylib -L[INSTALL_DIR]/mkl_intel64 \
          -lmkl_intel_ilp64 -lmkl_intel_thread -lmkl_core \
          -L[INSTALL_DIR]/rtl -liomp5 -lsvml -limf -lifcoremt -lirc -lpthread
   

   if the shareable library is required.

   However, if you prefer to link to a version of the NAG Library which does not require the use of MKL you may wish to use the self-contained libraries as follows:

      gfortran -fdefault-integer-8 driver.f \
          -I[INSTALL_DIR]/nag_interface_blocks_gfortran \
          [INSTALL_DIR]/lib/libnag_nag.a [INSTALL_DIR]/rtl/libsvml.a \
          [INSTALL_DIR]/rtl/libimf.a [INSTALL_DIR]/rtl/libifcoremt.a \
          [INSTALL_DIR]/rtl/libirc.a
   

   or

      gfortran -fdefault-integer-8 driver.f \
          -I[INSTALL_DIR]/nag_interface_blocks_gfortran \
          [INSTALL_DIR]/lib/libnag_nag.dylib \
          -L[INSTALL_DIR]/rtl -lsvml -limf -lifcoremt -lirc
   

   if the shareable library is required.

   If your application has been linked with the shareable NAG and MKL libraries then the environment variable DYLD_LIBRARY_PATH must be set (or extended) to allow run time linkage.

   In the C shell, type:

      setenv DYLD_LIBRARY_PATH \
          [INSTALL_DIR]/lib:[INSTALL_DIR]/mkl_intel64:[INSTALL_DIR]/rtl
   

   to set DYLD_LIBRARY_PATH, or

      setenv DYLD_LIBRARY_PATH \
          [INSTALL_DIR]/lib:[INSTALL_DIR]/mkl_intel64:[INSTALL_DIR]/rtl:\
          ${DYLD_LIBRARY_PATH}
   

   to extend DYLD_LIBRARY_PATH if you already have it set.

   In the Bourne shell, type:

      DYLD_LIBRARY_PATH=[INSTALL_DIR]/lib:[INSTALL_DIR]/mkl_intel64:\
          [INSTALL_DIR]/rtl
      export DYLD_LIBRARY_PATH
   

   to set DYLD_LIBRARY_PATH, or

      DYLD_LIBRARY_PATH=[INSTALL_DIR]/lib:[INSTALL_DIR]/mkl_intel64:\
          [INSTALL_DIR]/rtl:${DYLD_LIBRARY_PATH}
      export DYLD_LIBRARY_PATH
   

   to extend DYLD_LIBRARY_PATH if you already have it set.