rodinia-benchmark/opencl/leukocyte/meschach_lib/MACHINES/cornelison

From - Fri Dec  4 08:05:39 1998
Received: from cliffy.statsci.com (root@cliffy.statsci.com [206.63.206.72]) by server.divms.uiowa.edu with id MAA14931 for <dstewart@math.uiowa.edu>; Thu, 3 Dec 1998 12:28:13 -0600 (CST)
Received: from adlib2 (adlib2 [206.63.206.104])
	by cliffy.statsci.com (8.8.8/8.8.8/Hub) with SMTP id KAA28911;
	Thu, 3 Dec 1998 10:28:02 -0800
Message-Id: <3.0.5.32.19981203102809.01371ec0@mailhost.statsci.com>
X-Sender: johnc@mailhost.statsci.com
X-Mailer: QUALCOMM Windows Eudora Pro Version 3.0.5 (32)
Date: Thu, 03 Dec 1998 10:28:09 -0800
To: David Stewart <dstewart@math.uiowa.edu>
From: John Cornelison <johnc@statsci.com>
Subject: Re: VStudio version of Meschach Library available?!
Cc: Zbigniew Leyk <zleyk@wolfram.com>
In-Reply-To: <36669B2E.809B959C@math.uiowa.edu>
References: <3.0.5.32.19981202145120.0106bec0@mailhost.statsci.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
X-Mozilla-Status: 8013
X-Mozilla-Status2: 00000000

Thanks for the fast reply.  I'll send you another note once it's working.
It now compiles fine into a .DLL with the following mods (hacks?!):

The largest problem was figuring out to not include some of the
miscellaneous source files that are used for QA & utility & legacy support.
 Once I read the makefile it was easy to set up.  I'll send you the
necessary project files once I complete things.

As far as coding goes, in machine.h, right near the end, I've added:

#ifndef WIN32
#define LIBEXPORT
#else

/* Comment out the following line if NOT building Meschach.DLL! */
/* TODO: Eventually this should be defined on the compile line only when
building the DLL */
#define DLLCREATION

#ifdef DLLCREATION
	/* Including code is internal to the DLL */
	#define MLibImport _declspec(dllexport)
	#define MLibExport _declspec(dllimport)
	#define LIBEXPORT MLibImport
#else
	/* Including code is external to the DLL */
	#define MLibImport _declspec(dllimport)
	#define MLibExport _declspec(dllexport)
	#define LIBEXPORT MLibExport
#endif

#ifdef WORDS_BIGENDIAN
	/* Win32 is always little endian */
	#undef WORDS_BIGENDIAN
#endif

#ifdef HUGE_VAL
	/* HUGE_VAL later defined in math.h */
	#undef HUGE_VAL
#endif

#if defined(_DEBUG) && defined(_MSC_VER)
	/* WARNING: Disabling the many "signed/unsigned mismatch" warnings so
others stand out! */
	/* These happen when comparing an unsigned int with a regular signed int
but should be fixed! */
	#pragma warning(disable:4018)
#endif
#endif  /* WIN32 */

The only significant thing here is to explicitly define what routines are
being exported (or imported) from the DLL.  This is done with Microsoft's
delightful _declspec(dllexport) & _declspec(dllimport) expressions.  Next I
had to go through all function declarations & insert the LIBEXPORT define.
Ideally this should ONLY be done for routines that are made publicly
available from the library, i.e., not utility routines for internal use
only, but I put it in all the declarations in sparse.h; err.h; iter.h;
matrix.h; sparse2.h; meminfo.h.  I left the complex libraries for another day.

A sample of this is from the matrix.h file:

/* miscellaneous functions */

#ifndef ANSI_C

extern	double LIBEXPORT	square(), cube(), mrand();
extern	void LIBEXPORT	smrand(), mrandlist();
extern  void LIBEXPORT    m_dump(), px_dump(), v_dump(), iv_dump();
extern MAT LIBEXPORT *band2mat();
extern BAND LIBEXPORT *mat2band();

#else

double LIBEXPORT	square(double x), 	/* returns x^2 */
  cube(double x), 		/* returns x^3 */
  mrand(void);                  /* returns random # in [0,1) */

void LIBEXPORT	smrand(int seed),            /* seeds mrand() */
  mrandlist(Real *x, int len);       /* generates len random numbers */

void LIBEXPORT    m_dump(FILE *fp,MAT *a), px_dump(FILE *,PERM *px),
        v_dump(FILE *fp,VEC *x), iv_dump(FILE *fp, IVEC *ix);

MAT LIBEXPORT *band2mat(BAND *bA, MAT *A);
BAND LIBEXPORT *mat2band(MAT *A, int lb,int ub, BAND *bA);

#endif

It is sort of ugly, and I probably messed up all your formatting but it at
least works and won't break things on other platforms.  The only warnings
that now appear during a full compilation (again, I've left out all the
z*.c routines) are:

Deleting intermediate files and output files for project 'Meschach - Win32
Debug'.
--------------------Configuration: Meschach - Win32 Debug--------------------
Compiling...
bdfactor.c
bkpfacto.c
chfactor.c
copy.c
err.c
D:\Meschach\Source\err.c(328) : warning C4113: 'void (__cdecl *)()' differs
in parameter lists from 'void (__cdecl *)(int )'
D:\Meschach\Source\err.c(337) : warning C4113: 'void (__cdecl *)()' differs
in parameter lists from 'void (__cdecl *)(int )'
fft.c
givens.c
hessen.c
hsehldr.c
init.c
iter0.c
iternsym.c
itersym.c
D:\Meschach\Source\itersym.c(424) : warning C4113: 'int (__cdecl *)()'
differs in parameter lists from 'int (__cdecl *)(const void *,const void *)'
ivecop.c
lufactor.c
machine.c
matlab.c
matop.c
matrixio.c
meminfo.c
memory.c
memstat.c
mfunc.c
D:\Meschach\Source\mfunc.c(174) : warning C4244: '=' : conversion from
'double ' to 'int ', possible loss of data
norm.c
otherio.c
pxop.c
qrfactor.c
schur.c
D:\Meschach\Source\schur.c(158) : warning C4101: 't' : unreferenced local
variable
solve.c
sparse.c
sparseio.c
spbkp.c
D:\Meschach\Source\spbkp.c(1244) : warning C4113: 'int (__cdecl *)()'
differs in parameter lists from 'int (__cdecl *)(const void *,const void *)'
spchfctr.c
splufctr.c
sprow.c
spswap.c
submat.c
svd.c
symmeig.c
update.c
vecop.c
version.c
Linking...
   Creating library Debug/Meschach.lib and object Debug/Meschach.exp
Creating browse info file...

Meschach.dll - 0 error(s), 6 warning(s)



At 08:07 AM 12/3/98 -0600, David Stewart wrote:
>John Cornelison wrote:
>> 
>> Hi, I'm starting to get the library to work in the Win32 environment and
>> was curious if you are aware of anyone else that may have done this
already.
>> 
>> Specifically I'd appreciate pointers to anyone that has wrapped this into a
>> Visual Studio project.  I know Visual Studio is super platform specific so
>> is not something you are likely to promote, but knowing about this would
>> save us some time.
>> 
>> Also I understand from the readme that you have an electronic form of the
>> manual I can ask for.  If easy, I would appreciate a copy of that.  We are
>> using the sparse matrix processing functionality for a small research
>> project we're considering and getting improvements of many magnitudes over
>> our previous techniques.
>> 
>> Thanks in advance for doing all this incredible work!
>> 
>> -John
>
>You can get a **partial** on-line manual via 
>
>	http://www.math.uiowa.edu/~dstewart/meschach/
>
>However, this doesn't include the sparse matrix stuff.
>There have been some others who have worked on MS-DOS ports of
>Meschach.  However, I haven't been keeping up with the status of these
>efforts.  And I ceratinly haven't heard of anyone using it in
>Visual Studio.  Zbigniew and I are glad to hear of another happy
>user though!
>
>Neither Zbigniew nor I have had much time to devote to Meschach
>in the past few years, which has (unfortunately) stalled its
>development...  You should also look for matrix re-ordering techniques
>(I have some code for the Gibbs-Poole-Stockmeyer method for
>Meschach, if you would like that.  Someone else developed it,
>so I can't say much regarding it; the algorithm is not the best
>matrix re-ordering technique for general sparse matrices, but
>it definitely better than standard orderings for most problems.)

Sure, if convenient, I would appreciate getting this.

>Also: please note the new email addresses!
>
>	-- David Stewart

Thanks again for developing the library!

----------------------------------------------------------------------------
http://www.MathSoft.com, Data Analysis Products Division
1700 Westlake Av N, Suite 500, Seattle, WA 98109-3012
206.283.8802x243, 283.8691fax, johnc@statsci.com, NIC JHC7
----------------------------------------------------------------------------