The R Project SVN R-packages

/* ========================================================================== */
/* === umf_config.h ========================================================= */
/* ========================================================================== */

/* -------------------------------------------------------------------------- */
/* UMFPACK Version 4.5, Copyright (c) 2005 by Timothy A. Davis.  CISE Dept,   */
/* Univ. of Florida.  All Rights Reserved.  See ../Doc/License for License.   */
/* web: http://www.cise.ufl.edu/research/sparse/umfpack                       */
/* -------------------------------------------------------------------------- */

/*
    This file controls the compile-time configuration of UMFPACK.  Modify the
    UFconfig/UFconfig.mk file and this file if necessary, to control these
    options.  The following flags may be given as options to your C compiler
    (as in "cc -DNBLAS", for example).  These flags are normally placed in
    your UMFPACK_CONFIG string, defined in the UFconfig/UFconfig.mk file.

    All of these options, except for the timer, are for accessing the BLAS.

    -DNBLAS

        BLAS mode.  If -DNBLAS is set, then no BLAS will be used.  Vanilla
        C code will be used instead.  This is portable, and easier to
        install, but you won't get the best performance.

        If -DNBLAS is not set, then externally-available BLAS routines
        (dgemm, dger, and dgemv or the equivalent C-BLAS routines) will be
        used.  This will give you the best performance, but perhaps at the
        expense of portability.

        The default is to use the BLAS, for both the C-callable libumfpack.a
        library and the MATLAB mexFunction.  If you have trouble installing
        UMFPACK, set -DNBLAS (but then UMFPACK will be slow).

    -DCBLAS

        If -DCBLAS is set, then the C-BLAS interface to the BLAS is
        used.  If your vendor-supplied BLAS library does not have a C-BLAS
        interface, you can obtain the ATLAS BLAS, available at
        http://www.netlib.org/atlas.

        This flag is ignored if -DNBLAS is set.

    -DLP64

        This should be defined if you are compiling in the LP64 model
        (32 bit int's, 64 bit long's, and 64 bit pointers).  In Solaris,
        this is obtained with the flags -xtarget=ultra -xarch=v9 for
        the cc compiler (for example).

    -DLONGBLAS

        If not defined, then the BLAS are not called in the long integer
        version of UMFPACK (the umfpack_*l_* routines).  The most common
        definitions of the BLAS, unfortunately, use int arguments, and
        are thus not suitable for use in the LP64 model.  Only the Sun
        Performance Library, as far as I can tell, has a version of the
        BLAS that allows long integer (64-bit) input arguments.  This
        flag is set automatically in Sun Solaris if you are using the
        Sun Performance BLAS.  You can set it yourself, too, if your BLAS
        routines can take long integer input arguments.

    -DNSUNPERF

        Applies only to Sun Solaris.  If -DNSUNPERF is set, then the Sun
        Performance Library BLAS will not be used.

        The Sun Performance Library BLAS is used by default when compiling
        the C-callable libumfpack.a library on Sun Solaris.

        This flag is ignored if -DNBLAS is set.

    -DNSCSL

        Applies only to SGI IRIX.  If -DSCSL is set, then the SGI SCSL
        Scientific Library BLAS will not be used.

        The SGI SCSL Scientific Library BLAS is used by default when
        compiling the C-callable libumfpack.a library on SGI IRIX.

        This flag is ignored if -DNBLAS is set.

    -DNPOSIX

        If -DNPOSIX is set, then your Unix operating system is not POSIX-
        compliant, and the POSIX routines sysconf ( ) and times ( )
        routines are not used.  These routines provide CPU time and
        wallclock time information.  If -DNPOSIX is set, then the ANSI
        C clock ( ) routine is used.  If -DNPOSIX is not set, then
        sysconf ( ) and times ( ) are used in umfpack_tic and umfpack_toc.
        See umfpack_tictoc.c for more information.
        The default is to use the POSIX routines, except for Windows,
        which is not POSIX-compliant.

    -DGETRUSAGE

        If -DGETRUSAGE is set, then your system's getrusage ( ) routine
        will be used for getting the process CPU time.  Otherwise the ANSI
        C clock ( ) routine will be used.  The default is to use getrusage
        ( ) on Unix systems, and to use clock on all other architectures.

    -DNO_TIMER

        If -DNO_TIMER is set, then no timing routines are used at all.

    -DNRECIPROCAL

        This option controls a tradeoff between speed and accuracy.  Using
        -DNRECIPROCAL can lead to more accurate results, but with perhaps
        some cost in performance, particularly if floating-point division
        is much more costly than floating-point multiplication.

        This option determines the method used to scale the pivot column.
        If set, or if the absolute value of the pivot is < 1e-12 (or is a
        NaN), then the pivot column is divided by the pivot value.
        Otherwise, the reciprocal of the pivot value is computed, and the
        pivot column is multiplied by (1/pivot).  Multiplying by the
        reciprocal can be slightly less accurate than dividing by the
        pivot, but it is often faster.  See umf_scale.c.

        This has a small effect on the performance of UMFPACK, at least on
        a Pentium 4M.  It may have a larger effect on other architectures
        where floating-point division is much more costly than floating-
        point multiplication.  The RS 6000 is one such example.

        By default, the method chosen is to multiply by the reciprocal
        (sacrificing accuracy for speed), except when compiling UMFPACK
        as a built-in routine in MATLAB, or when gcc is being used.

        When MATHWORKS is defined, -DNRECIPROCAL is forced on, and the pivot
        column is divided by the pivot value.  The only way of using the
        other method in this case is to edit this file.

        If -DNRECIPROCAL is enabled, then the row scaling factors are always
        applied by dividing each row by the scale factor, rather than
        multiplying by the reciprocal.  If -DNRECIPROCAL is not enabled
        (the default case), then the scale factors are normally applied by
        multiplying by the reciprocal.  If, however, the smallest scale
        factor is tiny, then the scale factors are applied via division.

    -DNO_DIVIDE_BY_ZERO

        If the pivot is zero, and this flag is set, then no divide-by-zero
        occurs.

    You should normally not set these flags yourself:

    -DBLAS_BY_VALUE     if scalars are passed by value, not reference
    -DBLAS_NO_UNDERSCORE    if no underscore should be appended
    -DBLAS_CHAR_ARG     if BLAS options are single char's, not strings

        The BLAS options are normally set automatically.  If your
        architecture cannot be determined (see UMFPACK_ARCHITECTURE, below)
        then you may need to set these flags yourself.

    The following options are controlled by amd_internal.h:

    -DMATLAB_MEX_FILE

        This flag is turned on when compiling the umfpack mexFunction for
        use in MATLAB.  When compiling the MATLAB mexFunction, the MATLAB
        BLAS are used (unless -DNBLAS is set).  The -DCBLAS, -DNSCSL, and
        -DNSUNPERF flags are all ignored.   The -DNRECIPROCAL flag is
        forced on.  Otherwise, [L,U,P,Q,R] = umfpack (A) would return
        either L*U = P*(R\A)*Q or L*U = P*R*A*Q.  Rather than returning a
        flag stating how the scale factors R are to be applied, the umfpack
        mexFunction always takes the more accurate route and returns
        L*U = P*(R\A)*Q.

    -DMATHWORKS

        This flag is turned on when compiling umfpack as a built-in routine
        in MATLAB.  The -DNRECIPROCAL flag is turned on.

    -DNDEBUG

        Debugging mode (if NDEBUG is not defined).  The default, of course,
        is no debugging.  Turning on debugging takes some work (see below).
        If you do not edit this file, then debugging is turned off anyway,
        regardless of whether or not -DNDEBUG is specified in your compiler
        options.
*/

/* ========================================================================== */
/* === AMD configuration ==================================================== */
/* ========================================================================== */

/* NDEBUG, PRINTF defined in amd_internal.h */

/* ========================================================================== */
/* === reciprocal option ==================================================== */
/* ========================================================================== */

/* Force the definition NRECIPROCAL when MATHWORKS or MATLAB_MEX_FILE
 * are defined.  Do not multiply by the reciprocal in those cases. */

#ifndef NRECIPROCAL
#if defined (MATHWORKS) || defined (MATLAB_MEX_FILE)
#define NRECIPROCAL
#endif
#endif

/* ========================================================================== */
/* === Microsoft Windows configuration ====================================== */
/* ========================================================================== */

#ifdef UMF_WINDOWS
/* Windows isn't Unix.  Profound. */
#define NPOSIX
#endif

/* ========================================================================== */
/* === 0-based or 1-based printing ========================================== */
/* ========================================================================== */

#if defined (MATLAB_MEX_FILE) && defined (NDEBUG)
/* In MATLAB, matrices are 1-based to the user, but 0-based internally. */
/* One is added to all row and column indices when printing matrices */
/* for the MATLAB user.  The +1 shift is turned off when debugging. */
#define INDEX(i) ((i)+1)
#else
/* In ANSI C, matrices are 0-based and indices are reported as such. */
/* This mode is also used for debug mode, and if MATHWORKS is defined rather */
/* than MATLAB_MEX_FILE. */
#define INDEX(i) (i)
#endif

/* ========================================================================== */
/* === Timer ================================================================ */
/* ========================================================================== */

/*
    If you have the getrusage routine (all Unix systems I've test do), then use
    that.  Otherwise, use the ANSI C clock function.   Note that on many
    systems, the ANSI clock function wraps around after only 2147 seconds, or
    about 36 minutes.  BE CAREFUL:  if you compare the run time of UMFPACK with
    other sparse matrix packages, be sure to use the same timer.  See
    umfpack_tictoc.c for the timer used internally by UMFPACK.  See also
    umfpack_timer.c for the timer used in an earlier version of UMFPACK.
    That timer is still available as a user-callable routine, but it is no
    longer used internally by UMFPACK.
*/

/* Sun Solaris, SGI Irix, Linux, Compaq Alpha, and IBM RS 6000 all have */
/* getrusage.  It's in BSD unix, so perhaps all unix systems have it. */
#if defined (UMF_SOL2) || defined (UMF_SGI) || defined (UMF_LINUX) \
|| defined (UMF_ALPHA) || defined (UMF_AIX)
#define GETRUSAGE
#endif


/* ========================================================================== */
/* === BLAS ================================================================= */
/* ========================================================================== */

/*
    The adventure begins.  Figure out how to call the BLAS ...

    This works, but it is incredibly ugly.  The C-BLAS was supposed to solve
    this problem, and make it easier to interface a C program to the BLAS.
    Unfortunately, the C-BLAS does not have a "long" integer (64 bit) version.
    Various vendors have done their own 64-bit BLAS.  Sun has dgemm_64 routines
    with "long" integers, SGI has a 64-bit dgemm in their scsl_blas_i8 library
    with "long long" integers, and so on.

    Different vendors also have different ways of defining a complex number,
    some using struct's.  That's a bad idea.  See umf_version.h for the better
    way to do it (the method that was also chosen for the complex C-BLAS,
    which is compatible and guaranteed to be portable with ANSI C).

    To make matters worse, SGI's SCSL BLAS has a C-BLAS interface which
    differs from the ATLAS C-BLAS interface (see immediately below);
    although a more recent version of SGI's C-BLAS interface is correct
    if SCSL_VOID_ARGS is defined.
*/


/* -------------------------------------------------------------------------- */
/* Determine which BLAS to use. */
/* -------------------------------------------------------------------------- */

#if defined (MATHWORKS)
#define USE_MATLAB_BLAS

#elif defined (NBLAS)
#define USE_NO_BLAS

#elif defined (MATLAB_MEX_FILE)
#define USE_MATLAB_BLAS

#elif defined (CBLAS)
#define USE_C_BLAS

#elif defined (UMF_SOL2) && !defined (NSUNPERF)
#define USE_SUNPERF_BLAS

#elif defined (UMF_SGI) && !defined (NSCSL)
#define USE_SCSL_BLAS

#else
#define USE_FORTRAN_BLAS
#endif

/* -------------------------------------------------------------------------- */
/* int vs. long integer arguments */
/* -------------------------------------------------------------------------- */

/*
    Determine if the BLAS exists for the long integer version.  It exists if
    LONGBLAS is defined in the Makefile, or if using the BLAS from the
    Sun Performance Library, or SGI's SCSL Scientific Library.
*/

#if defined (USE_SUNPERF_BLAS) || defined (USE_SCSL_BLAS)
#ifndef LONGBLAS
#define LONGBLAS
#endif
#endif

/* do not use the BLAS if Int's are long and LONGBLAS is not defined */
#if defined (LONG_INTEGER) && !defined (LONGBLAS) && !defined (USE_NO_BLAS)
#define USE_NO_BLAS
#endif


/* -------------------------------------------------------------------------- */
/* Use (void *) arguments for the SGI */
/* -------------------------------------------------------------------------- */

#if defined (UMF_SGI)
/*
    Use (void *) pointers for complex types in SCSL.
    The ATLAS C-BLAS, and the SGI C-BLAS differ.  The former uses (void *)
    arguments, the latter uses SCSL_ZOMPLEX_T, which are either scsl_zomplex
    or (void *).  Using (void *) is simpler, and is selected by defining
    SCSL_VOID_ARGS, below.  The cc compiler doesn't complain, but gcc is
    more picky, and generates a warning without this next statement.
    With gcc and the 07/09/98 version of SGI's cblas.h, spurious warnings
    about complex BLAS arguments will be reported anyway.  This is because this
    older version of SGI's cblas.h does not make use of the SCSL_VOID_ARGS
    parameter, which is present in the 12/6/01 version of SGI's cblas.h.  You
    can safely ignore these warnings.
*/
#define SCSL_VOID_ARGS
#endif


/* -------------------------------------------------------------------------- */
/* The BLAS exists, construct appropriate macros */
/* -------------------------------------------------------------------------- */

#if !defined (USE_NO_BLAS)      /* { */

/*
    If the compile-time flag -DNBLAS is defined, then the BLAS are not used,
    portable vanilla C code is used instead, and the remainder of this file
    is ignored.

    Using the BLAS is much faster, but how C calls the Fortran BLAS is
    machine-dependent and thus can cause portability problems.  Thus, use
    -DNBLAS to ensure portability (at the expense of speed).

    Preferences:

    *** The best interface to use, regardless of the option you select
        below, is the standard C-BLAS interface.  Not all BLAS libraries
        use this interface.  The only problem with this interface is that
        it does not extend to the LP64 model.  The C-BLAS does not provide
        for a 64-bit integer.  In addition, SGI's older cblas.h can cause
        spurious warnings when using the C-BLAS interface.

    1) often the most preferred (but see option (3)):  use the
        optimized vendor-supplied library (such as the Sun Performance
        Library, or IBM's ESSL).  This is often the fastest, but might not
        be portable and might not always be available.  When compiling a
        MATLAB mexFunction it might be difficult get the mex compiler
        script to recognize the vendor- supplied BLAS.  Note that the
        freely-available BLAS (option 3) can be faster than the vendor-
        specific BLAS.  You are encourage to try both option (1) and (3).

    2) When compiling the UMFPACK mexFunction to use UMFPACK in MATLAB, use
        the BLAS provided by The Mathworks, Inc.  This assumes you are using
        MATLAB V6 or higher, since the BLAS are not incorporated in V5 or
        earlier versions.  On my Sun workstation, the MATLAB BLAS gave
        slightly worse performance than the Sun Perf. BLAS.  The advantage
        of using the MATLAB BLAS is that it's available on any computer that
        has MATLAB V6 or higher.  I have not tried using MATLAB BLAS outside
        of a mexFunction in a stand-alone C code, but MATLAB (V6) allows for
        this.  This is well worth trying if you have MATLAB and don't want
        to bother installing the ATLAS BLAS (option 3a, below).  The only
        glitch to this is that MATLAB does not provide a portable interface
        to the BLAS (an underscore is required for some but not all
        architectures).  For Windows and MATLAB 6.0 or 6.1, you also need
        to copy the libmwlapack.dll file into your MATLAB installation
        directory; see the User Guide for details.

        In the current distribution, the only BLAS that the UMFPACK
        mexFunction will use is the internal MATLAB BLAS.  It's possible to
        use other BLAS, but handling the porting of using the mex compiler
        with different BLAS libraries is not trivial.

        As of MATLAB 6.5, the BLAS used internally in MATLAB is the ATLAS
        BLAS.

    3) Use a freely-available high-performance BLAS library:

        (a) The BLAS by Kazashige Goto and Robert van de Geijn, at
        http://www.cs.utexas.edu/users/flame/goto.  This BLAS increased
        the performance of UMFPACK by almost 50% as compared to the
        ATLAS BLAS (v3.2).

        (b) The ATLAS BLAS, available at http://www.netlib.org/atlas,
        by R. Clint Whaley, Antoine Petitet, and Jack Dongarra.
        This has a standard C interface, and thus the interface to it is
        fully portable.  Its performance rivals, and sometimes exceeds,
        the vendor-supplied BLAS on many computers.

        (b) The Fortran RISC BLAS by Michel Dayde', Iain Duff, Antoine
        Petitet, and Abderrahim Qrichi Aniba, available via anonymous
        ftp to ftp.enseeiht.fr in the pub/numerique/BLAS/RISC directory,
        See M. J. Dayde' and I. S. Duff, "The RISC BLAS:  A blocked
        implementation of level 3 BLAS for RISC processors, ACM Trans.
        Math. Software, vol. 25, no. 3., Sept. 1999.  This will give
        you good performance, but with the same C-to-Fortran portability
        problems as option (1).

    4) Use UMFPACK's built-in vanilla C code by setting -DNBLAS at compile
        time.  The key advantage is portability, which is guaranteed if you
        have an ANSI C compliant compiler.  You also don't need to download
        any other package - UMFPACK is stand-alone.  No Fortran is used
        anywhere in UMFPACK.  UMFPACK will be much slower than when using
        options (1) through (3), however.

    5) least preferred:  use the standard Fortran implementation of the
        BLAS, also available at Netlib (http://www.netlib.org/blas).  This
        will be no faster than option (4), and not portable because of
        C-to-Fortran calling conventions.  Don't bother trying option (5).

    The mechanics of how C calls the BLAS on various computers are as follows:

    * C-BLAS (from the ATLAS library, for example):
        The same interface is used on all computers.

    * Defaults for calling the Fortran BLAS:
        add underscore, pass scalars by reference, use string arguments.

    * The Fortran BLAS on Sun Solaris (when compiling the MATLAB mexFunction
        or when using the Fortran RISC BLAS), SGI IRIX, Linux, and Compaq
        Alpha: use defaults.

    * Sun Solaris (when using the C-callable Sun Performance library):
        no underscore, pass scalars by value, use character arguments.

    * The Fortran BLAS (ESSL Library) on the IBM RS 6000, and HP Unix:
        no underscore, pass scalars by reference, use string arguments.

    * The Fortran BLAS on Windows:
        no underscore, pass scalars by reference, use string arguments.
        If you compile the umfpack mexFunction using umfpack_make, and are
        using the lcc compiler bundled with MATLAB, then you must first
        copy the umfpack\lcc_lib\libmwlapack.lib file into the
        <matlab>\extern\lib\win32\lcc\ directory, where <matlab> is the
        directory in which MATLAB is installed.  Next, type mex -setup
        at the MATLAB prompt, and ask MATLAB to select the lcc compiler.
        MATLAB has built-in BLAS, but it cannot be accessed by a program
        compiled by lcc without first copying this file.
*/



/* -------------------------------------------------------------------------- */
#ifdef USE_C_BLAS   /* { */
/* -------------------------------------------------------------------------- */


/* -------------------------------------------------------------------------- */
/* use the C-BLAS (any computer) */
/* -------------------------------------------------------------------------- */

/*
    C-BLAS is the default interface, with the following exceptions.  Solaris
    uses the Sun Performance BLAS for libumfpack.a (the C-callable library).
    SGI IRIX uses the SCSL BLAS for libumfpack.a.  All architectures use
    MATLAB's internal BLAS for the mexFunction on any architecture.  These
    options are set in the UFconfig/UFconfig.mk file.

    If you use the ATLAS C-BLAS, then be sure to set the -I flag to
    -I/path/ATLAS/include, where /path/ATLAS is the ATLAS installation
    directory.  You do not need to do this for the SGI, which has a
    /usr/include/cblas.h.
*/

#include "cblas.h"

#ifdef COMPLEX
#define BLAS_GEMM_ROUTINE cblas_zgemm
#define BLAS_TRSM_ROUTINE cblas_ztrsm
#define BLAS_TRSV_ROUTINE cblas_ztrsv
#define BLAS_GEMV_ROUTINE cblas_zgemv
#define BLAS_GER_ROUTINE  cblas_zgeru
#define BLAS_SCAL_ROUTINE cblas_zscal
#define BLAS_COPY_ROUTINE cblas_zcopy
#define BLAS_DECLARE_SCALAR(x) double x [2]
#define BLAS_ASSIGN(x,xr,xi) { x [0] = xr ; x [1] = xi ; }
#else
#define BLAS_GEMM_ROUTINE cblas_dgemm
#define BLAS_TRSM_ROUTINE cblas_dtrsm
#define BLAS_TRSV_ROUTINE cblas_dtrsv
#define BLAS_GEMV_ROUTINE cblas_dgemv
#define BLAS_GER_ROUTINE  cblas_dger
#define BLAS_SCAL_ROUTINE cblas_dscal
#define BLAS_COPY_ROUTINE cblas_dcopy
#define BLAS_DECLARE_SCALAR(x) double x
#define BLAS_ASSIGN(x,xr,xi) { x = xr ; }
#endif

#define BLAS_LOWER CblasLower
#define BLAS_UNIT_DIAGONAL CblasUnit
#define BLAS_RIGHT CblasRight
#define BLAS_NO_TRANSPOSE CblasNoTrans
#define BLAS_TRANSPOSE CblasTrans
#define BLAS_COLUMN_MAJOR_ORDER CblasColMajor,
#define BLAS_SCALAR(x) x
#define BLAS_INT_SCALAR(n) n
#define BLAS_ARRAY(a) a



/* -------------------------------------------------------------------------- */
#else   /* } USE_C_BLAS { */
/* -------------------------------------------------------------------------- */

/* -------------------------------------------------------------------------- */
/* use Fortran (or other architecture-specific) BLAS */
/* -------------------------------------------------------------------------- */

/* No such argument when not using the C-BLAS */
#define BLAS_COLUMN_MAJOR_ORDER

/* Determine which architecture we're on and set options accordingly. */
/* The default, if nothing is defined is to add an underscore, */
/* pass scalars by reference, and use string arguments. */

/* ---------------------------------- */
/* Sun Performance BLAS */
/* ---------------------------------- */

#ifdef USE_SUNPERF_BLAS
#ifdef _SUNPERF_H
/* <sunperf.h> has been included somehow anyway, outside of umf_config.h */
#error "sunperf.h must NOT be #include'd.  See umf_config.h for details."
#endif
#define BLAS_BY_VALUE
#define BLAS_NO_UNDERSCORE
#define BLAS_CHAR_ARG
#endif  /* USE_SUNPERF_BLAS */

/* ---------------------------------- */
/* SGI SCSL BLAS */
/* ---------------------------------- */

#ifdef USE_SCSL_BLAS
#if defined (LP64)
#include <scsl_blas_i8.h>
#else
#include <scsl_blas.h>
#endif
#define BLAS_BY_VALUE
#define BLAS_NO_UNDERSCORE
#endif  /* USE_SCSL_BLAS */

/* ---------------------------------- */
/* IBM AIX, Windows, and HP Fortran BLAS */
/* ---------------------------------- */

#if defined (UMF_AIX) || defined (UMF_WINDOWS) || defined (UMF_HP)
#define BLAS_NO_UNDERSCORE
#endif


/* -------------------------------------------------------------------------- */
/* BLAS names */
/* -------------------------------------------------------------------------- */

#if defined (LP64) && defined (USE_SUNPERF_BLAS) && defined (LONG_INTEGER)

/* 64-bit sunperf BLAS, for Sun Solaris only */
#ifdef COMPLEX
#define BLAS_GEMM_ROUTINE zgemm_64
#define BLAS_TRSM_ROUTINE ztrsm_64
#define BLAS_TRSV_ROUTINE ztrsv_64
#define BLAS_GEMV_ROUTINE zgemv_64
#define BLAS_GER_ROUTINE  zgeru_64
#define BLAS_SCAL_ROUTINE zscal_64
#define BLAS_COPY_ROUTINE zcopy_64
#else
#define BLAS_GEMM_ROUTINE dgemm_64
#define BLAS_TRSM_ROUTINE dtrsm_64
#define BLAS_TRSV_ROUTINE dtrsv_64
#define BLAS_GEMV_ROUTINE dgemv_64
#define BLAS_GER_ROUTINE  dger_64
#define BLAS_SCAL_ROUTINE dscal_64
#define BLAS_COPY_ROUTINE dcopy_64
#endif  /* COMPLEX */

#else

#ifdef COMPLEX

/* naming convention (use underscore, or not) */
#ifdef BLAS_NO_UNDERSCORE
#define BLAS_GEMM_ROUTINE zgemm
#define BLAS_TRSM_ROUTINE ztrsm
#define BLAS_TRSV_ROUTINE ztrsv
#define BLAS_GEMV_ROUTINE zgemv
#define BLAS_GER_ROUTINE  zgeru
#define BLAS_SCAL_ROUTINE zscal
#define BLAS_COPY_ROUTINE zcopy
#else
/* default:  add underscore */
#define BLAS_GEMM_ROUTINE zgemm_
#define BLAS_TRSM_ROUTINE ztrsm_
#define BLAS_TRSV_ROUTINE ztrsv_
#define BLAS_GEMV_ROUTINE zgemv_
#define BLAS_GER_ROUTINE  zgeru_
#define BLAS_SCAL_ROUTINE zscal_
#define BLAS_COPY_ROUTINE zcopy_
#endif

#else

/* naming convention (use underscore, or not) */
#ifdef BLAS_NO_UNDERSCORE
#define BLAS_GEMM_ROUTINE dgemm
#define BLAS_TRSM_ROUTINE dtrsm
#define BLAS_TRSV_ROUTINE dtrsv
#define BLAS_GEMV_ROUTINE dgemv
#define BLAS_GER_ROUTINE  dger
#define BLAS_SCAL_ROUTINE dscal
#define BLAS_COPY_ROUTINE dcopy
#else
/* default:  add underscore */
#define BLAS_GEMM_ROUTINE dgemm_
#define BLAS_TRSM_ROUTINE dtrsm_
#define BLAS_TRSV_ROUTINE dtrsv_
#define BLAS_GEMV_ROUTINE dgemv_
#define BLAS_GER_ROUTINE  dger_
#define BLAS_SCAL_ROUTINE dscal_
#define BLAS_COPY_ROUTINE dcopy_
#endif

#endif  /* COMPLEX */

#endif /* LP64 && USE_SUNPERF_BLAS */


/* -------------------------------------------------------------------------- */
/* BLAS real or complex floating-point scalars */
/* -------------------------------------------------------------------------- */

#ifdef COMPLEX

/*
    The SunPerf BLAS expects to see a doublecomplex scalar, but it
    also will accept an array of size 2.  See the manual, normally at
    file:///opt/SUNWspro/WS6U1/lib/locale/C/html/manuals/perflib/user_guide
    /plug_using_perflib.html .  This manual is inconsistent with the man pages
    for zgemm, zgemv, and zgeru and also inconsistent with the <sunperf.h>
    include file.  Use this instead, for SunPerf (only works if you do NOT
    include sunperf.h).  Fortunately, this file (umf_config.h) is not included
    in any user code that calls UMFPACK.  Thus, the caller may include
    sunperf.h in his or her own code, and that is safely ignored here.
    SGI's SCSL BLAS has yet a different kind of struct, but we can use a
    double array of size 2 instead (since SCSL_VOID_ARGS is defined).
    Most BLAS expect complex scalars as pointers to double arrays of size 2.
*/

#define BLAS_DECLARE_SCALAR(x) double x [2]
#define BLAS_ASSIGN(x,xr,xi) { x [0] = xr ; x [1] = xi ; }
#define BLAS_SCALAR(x) x

#else

#define BLAS_DECLARE_SCALAR(x) double x
#define BLAS_ASSIGN(x,xr,xi) { x = xr ; }
#ifdef BLAS_BY_VALUE
#define BLAS_SCALAR(x) x
#else
#define BLAS_SCALAR(x) &(x)
#endif

#endif /* COMPLEX */


/* -------------------------------------------------------------------------- */
/* BLAS integer scalars */
/* -------------------------------------------------------------------------- */

/*
    Fortran requires integers to be passed by reference.
    The SCSL BLAS requires long long arguments in LP64 mode.
*/

#if defined (USE_SCSL_BLAS) && defined (LP64)
#define BLAS_INT_SCALAR(n) ((long long) n)
#else
#ifdef BLAS_BY_VALUE
#define BLAS_INT_SCALAR(n) n
#else
#define BLAS_INT_SCALAR(n) &(n)
#endif
#endif


/* -------------------------------------------------------------------------- */
/* BLAS strings */
/* -------------------------------------------------------------------------- */

/*
    The Sun Performance BLAS wants a character instead of a string.
*/

#ifdef BLAS_CHAR_ARG
#define BLAS_NO_TRANSPOSE 'N'
#define BLAS_TRANSPOSE 'T'
#define BLAS_LEFT 'L'
#define BLAS_RIGHT 'R'
#define BLAS_LOWER 'L'
#define BLAS_UNIT_DIAGONAL 'U'
#else
#define BLAS_NO_TRANSPOSE "N"
#define BLAS_TRANSPOSE "T"
#define BLAS_LEFT "L"
#define BLAS_RIGHT "R"
#define BLAS_LOWER "L"
#define BLAS_UNIT_DIAGONAL "U"
#endif


/* -------------------------------------------------------------------------- */
/* BLAS arrays */
/* -------------------------------------------------------------------------- */

/*
    The complex SunPerf BLAS expects to see a doublecomplex array of size s.
    This is broken (see above, regarding complex scalars in sunperf.h).
    For SunPerf BLAS, just pass a pointer to the array, and ignore sunperf.h.
    With sunperf.h, you would need:

    #define BLAS_ARRAY(a) ((doublecomplex *)(a))

    SGI's SCSL BLAS has yet a different kind of struct, but we can use a
    double array of size 2 instead (since SCSL_VOID_ARGS is defined).

    The real versions all use just a (double *) pointer.

    In all cases, no typecast is required.  This will break if <sunperf.h> is
    included.

    If you have read this far, I hope you see now why (void *) a much better
    choice for complex BLAS prototypes, and why double x [2] is better than
    an architecture dependent struct { double real ; double imag ; }
    type definition.

*/

#define BLAS_ARRAY(a) (a)


/* -------------------------------------------------------------------------- */
#endif /* USE_C_BLAS } */
/* -------------------------------------------------------------------------- */





/* -------------------------------------------------------------------------- */
/* BLAS macros, for all interfaces */
/* -------------------------------------------------------------------------- */

/*
   All architecture dependent issues have now been taken into consideration,
   and folded into the macros BLAS_DECLARE_SCALAR, BLAS_ASSIGN, BLAS_*_ROUTINE,
   BLAS_COLUMN_MAJOR_ORDER, BLAS_NO_TRANSPOSE, BLAS_TRANSPOSE, BLAS_SCALAR,
   BLAS_INT_SCALAR, BLAS_ARRAY, and Int.

   You will note that there is not a *** single *** name, declaration, or
   argument to the BLAS which is not somehow different in one or more versions
   of the BLAS!
*/


/* C = C - A*B', where:
 * A is m-by-k with leading dimension ldac
 * B is k-by-n with leading dimension ldb
 * C is m-by-n with leading dimension ldac */
#define BLAS_GEMM(m,n,k,A,B,ldb,C,ldac) \
{ \
    BLAS_DECLARE_SCALAR (alpha) ; \
    BLAS_DECLARE_SCALAR (beta) ; \
    BLAS_ASSIGN (alpha, -1.0, 0.0) ; \
    BLAS_ASSIGN (beta, 1.0, 0.0) ; \
    (void) BLAS_GEMM_ROUTINE (BLAS_COLUMN_MAJOR_ORDER \
    BLAS_NO_TRANSPOSE, BLAS_TRANSPOSE, \
    BLAS_INT_SCALAR (m), BLAS_INT_SCALAR (n), BLAS_INT_SCALAR (k), \
    BLAS_SCALAR (alpha), \
    BLAS_ARRAY (A), BLAS_INT_SCALAR (ldac), \
    BLAS_ARRAY (B), BLAS_INT_SCALAR (ldb), BLAS_SCALAR (beta), \
    BLAS_ARRAY (C), BLAS_INT_SCALAR (ldac)) ; \
}

/* A = A - x*y', where:
 * A is m-by-n with leading dimension d
   x is a column vector with stride 1
   y is a column vector with stride 1 */
#define BLAS_GER(m,n,x,y,A,d) \
{ \
    Int one = 1 ; \
    BLAS_DECLARE_SCALAR (alpha) ; \
    BLAS_ASSIGN (alpha, -1.0, 0.0) ; \
    (void) BLAS_GER_ROUTINE (BLAS_COLUMN_MAJOR_ORDER \
    BLAS_INT_SCALAR (m), BLAS_INT_SCALAR (n), \
    BLAS_SCALAR (alpha), \
    BLAS_ARRAY (x), BLAS_INT_SCALAR (one), \
    BLAS_ARRAY (y), BLAS_INT_SCALAR (one), \
    BLAS_ARRAY (A), BLAS_INT_SCALAR (d)) ; \
}

/* y = y - A*x, where A is m-by-n with leading dimension d,
   x is a column vector with stride 1
   y is a column vector with stride 1 */
#define BLAS_GEMV(m,n,A,x,y,d) \
{ \
    Int one = 1 ; \
    BLAS_DECLARE_SCALAR (alpha) ; \
    BLAS_DECLARE_SCALAR (beta) ; \
    BLAS_ASSIGN (alpha, -1.0, 0.0) ; \
    BLAS_ASSIGN (beta, 1.0, 0.0) ; \
    (void) BLAS_GEMV_ROUTINE (BLAS_COLUMN_MAJOR_ORDER \
    BLAS_NO_TRANSPOSE, \
    BLAS_INT_SCALAR (m), BLAS_INT_SCALAR (n), \
    BLAS_SCALAR (alpha), \
    BLAS_ARRAY (A), BLAS_INT_SCALAR (d), \
    BLAS_ARRAY (x), BLAS_INT_SCALAR (one), BLAS_SCALAR (beta), \
    BLAS_ARRAY (y), BLAS_INT_SCALAR (one)) ; \
}


/* solve Lx=b, where:
 * B is a column vector (m-by-1) with leading dimension d
 * A is m-by-m with leading dimension d */
#define BLAS_TRSV(m,A,b,d) \
{ \
    Int one = 1 ; \
    (void) BLAS_TRSV_ROUTINE (BLAS_COLUMN_MAJOR_ORDER \
    BLAS_LOWER, BLAS_NO_TRANSPOSE, BLAS_UNIT_DIAGONAL, \
    BLAS_INT_SCALAR (m), \
    BLAS_ARRAY (A), BLAS_INT_SCALAR (d), \
    BLAS_ARRAY (b), BLAS_INT_SCALAR (one)) ; \
}

/* solve XL'=B where:
 * B is m-by-n with leading dimension ldb
 * A is n-by-n with leading dimension lda */
#define BLAS_TRSM_RIGHT(m,n,A,lda,B,ldb) \
{ \
    BLAS_DECLARE_SCALAR (alpha) ; \
    BLAS_ASSIGN (alpha, 1.0, 0.0) ; \
    (void) BLAS_TRSM_ROUTINE (BLAS_COLUMN_MAJOR_ORDER \
    BLAS_RIGHT, BLAS_LOWER, BLAS_TRANSPOSE, BLAS_UNIT_DIAGONAL, \
    BLAS_INT_SCALAR (m), BLAS_INT_SCALAR (n), \
    BLAS_SCALAR (alpha), \
    BLAS_ARRAY (A), BLAS_INT_SCALAR (lda), \
    BLAS_ARRAY (B), BLAS_INT_SCALAR (ldb)) ; \
}

/* x = s*x, where x is a stride-1 vector of length n */
#define BLAS_SCAL(n,s,x) \
{ \
    Int one = 1 ; \
    BLAS_DECLARE_SCALAR (alpha) ; \
    BLAS_ASSIGN (alpha, REAL_COMPONENT (s), IMAG_COMPONENT (s)) ; \
    (void) BLAS_SCAL_ROUTINE ( \
    BLAS_INT_SCALAR (n), BLAS_SCALAR (alpha), \
    BLAS_ARRAY (x), BLAS_INT_SCALAR (one)) ; \
}

/* x = y, where x and y are a stride-1 vectors of length n */
#define BLAS_COPY(n,x,y) \
{ \
    Int one = 1 ; \
    (void) BLAS_COPY_ROUTINE ( \
    BLAS_INT_SCALAR (n), \
    BLAS_ARRAY (x), BLAS_INT_SCALAR (one), \
    BLAS_ARRAY (y), BLAS_INT_SCALAR (one)) ; \
}

#endif  /* !defined (USE_NO_BLAS) } */