MathKeisan User's Guide

---

Linking MathKeisan

Linking for SX

To link, use the SUPER-UX command f90, or the cross-compile command sxf90. If your code is compiled with the f90 -dw flag (the default), link with the libraries listed in Table 1. If your code is compiled with the f90 -ew flag, link with the libraries in Table 2. Link the libraries in the order shown, or use the f90 -h lib_cyclic flag. Please see Table 3 for the integer and real data types used in the libraries.

Link errors may occur if the MathKeisan libraries cannot be found. The default directory containing MathKeisan libraries is /usr/opt/mathkeisan/inst/lib0 on SX machines, and /SX/opt/mathkeisan/inst/lib0 on cross-compile machines. If the libraries are not in these directories, please specify their location using the link flag -Ldir, where dir is the full path name of the directory containing MathKeisan libraries.

When linking with MathKeisan Libraries (except METIS, PARMETIS, CBLAS and BLACS) using the C/C++ compiler, append -f90lib compile/link option on the command line.
Use -f90lib dw when linking in -dw MathKeisan Libraries and use -f90lib ew when linking in -ew MathKeisan Libraries.

When linking with f90, if any of the symbols __c_cexp, __c_csqrt, __c_npow, __c_pow, __c_ppow, __c_sincos, __c_spow, __c_stkmem, __c_stkmem2, __c_tpow, __c_vcexp, __c_vcsqrt, __c_vpow, __c_vsassign4, __c_vsincos, __c_wcexp, __c_wcsqrt, __c_wpow, __c_wsincos appear unresolved, then append -lcpp to the link command line after all MathKeisan libraries.

Libraries for SX

The MathKeisan SX libraries for different data types are listed in Table 3.

Libraries in Table 3 are for either 4-byte (32-bit) integer data type (Fortran integer(kind=4)) or 8-byte (64-bit) integer date type (Fortran integer(kind=8)).

Code compiled with the f90 default -dw uses 4-byte integers, and should be linked to libraries shown under the heading "4-byte integer".

Code compiled with f90 -ew uses 8-byte integers, and should be linked to libraries shown under the heading "8-byte integer".

The floating-point data type is either implicit, which is indicated when the "Prefix" column is empty, or is determined by the prefix of the subprogram name, as follows:

Note that in Table 4, the subprograms for 8-byte integer data type (f90 -ew) have 8-byte real data type for both the S and D prefixes, and 16-byte complex data type for both the C and Z prefixes.

Linking for Itanium® Processor Family

MathKeisan for IPF contains archive libraries (.a), and shared libraries (.so). Table 5 lists link flags for .a libraries using full library path names. Table 6 is for .a libraries using the link flag -static. Table 7 is for .so libraries. The macro $(LIBDIR) in the tables should be set to the MathKeisan lib directory. The default is $(LIBDIR) = /opt/MathKeisan/lib. For the libraries ScaLAPACK, BLACS, and ParMETIS, the macros $(MPI) and $(MPILIBDIR) should be set to the MPI directories for either NEC MPI or MPICH. For NEC MPI, the defaults are $(MPI) = NECMPI, and $(MPILIBDIR) = /usr/lib/libmpi.a. For MPICH, the defaults are $(MPI) = MPICH_ch_p4s, and $(MPILIBDIR) = /opt/mpich/lib/libmpich.a. See Table 8 for the integer and real data types used in the libraries.

If you see the run-time message error while loading shared libraries, you may need to add the MathKeisan lib directory to your LD_LIBRARY_PATH environment variable. The following C shell commands will add the default MathKeisan lib directory /opt/MathKeisan/lib to LD_LIBRARY_PATH. You can put these commands in your .cshrc file, or enter them at the command line:

if !($?LD_LIBRARY_PATH) then
    setenv LD_LIBRARY_PATH /opt/MathKeisan/lib
else
    setenv LD_LIBRARY_PATH /opt/MathKeisan/lib:$LD_LIBRARY_PATH
endif

You can put the following Bash shell command in your .profile file, or enter it at the command line:

export LD_LIBRARY_PATH="/opt/MathKeisan/lib:$LD_LIBRARY_PATH"

Libraries for Itanium® Processor Family

The MathKeisan IPF libraries for different data types are listed in Table 8.

Libraries in Table 8 are for 4-byte (32-bit) integer data type (Fortran integer(kind=4)). The floating-point data type is either implicit, which is indicated when the "Prefix" column is empty, or is determined by the prefix of the subprogram name, as follows: