From lamson%sierra.tcpip@ge-crd.arpa Wed Dec  9 12:04:40 1987
Return-Path: <lamson%sierra.tcpip@ge-crd.arpa>
Received: from dasher.mcs.anl by rudolph.mcs.anl (3.2/SMI-3.2)
	id AA27029; Wed, 9 Dec 87 12:04:38 CST
Received: from anl-mcs.ARPA by dasher.mcs.anl (3.2/SMI-3.2)
	id AA22847; Wed, 9 Dec 87 12:02:53 CST
Received: from GE-CRD.ARPA (ge-crd.arpa.ARPA) by anl-mcs.ARPA (4.12/4.9)
	id AA06786; Wed, 9 Dec 87 12:04:04 cst
Message-Id: <8712091804.AA06786@anl-mcs.ARPA>
Date:  9 Dec 87 13:01 EST
From: lamson%sierra.tcpip@ge-crd.arpa
Subject: sciport library for netlib distribution
To: dongarra@anl-mcs.ARPA
Status: RO

Received: by sierra.moose (3.2/1.1x Steinmetz)
	id AA03229; Wed, 9 Dec 87 10:41:26 EST
Date: Wed, 9 Dec 87 10:41:26 EST
From: scott h lamson <lamson@sierra>
Posted-Date: Wed, 9 Dec 87 10:41:26 EST
Message-Id: <8712091541.AA03229@sierra.moose>
To: dongarra@anl-mcs.ARPA
Subject: sciport library for netlib distribution


Jack,

On monday, I mailed you a tar file on a SUN tape cartridge containing
SCIPORT, which is CRAY's SCILIB implemented in portable fortran.
There are both single precision and double precision versions of
SCIPORT with identical calling interfaces.  IE, in the double
precision version, all single precision BLAS, Eispack, and Linpack
routines are inlcuded but written in double precision (or conversly
the double precision BLAS, Eispack & Linpack have had their names
changed to single precision notation) so users using single precision
on the CRAY but double precision on a 32 bit machine do not have to
change these names.
     The source files are grouped into blas, matrix utilities,
search utilities, recurrence functions, etc.  It may be preferable to
split them up into individual files.  There also are files to provide
on line documentation for VAX/VMS (sciport.hlp) and UNIX (sciport.1).
The files that start with dr_ are test drivers.  We run the test
drivers on the CRAY (linking with SCILIB) and using files like
blas.dat as input.  The driver creates a new file with all input
values but with the expected results included in the file.  Then we
can link the test drivers with sciport on another system and the test
drivers will compare the expected results in the file with the data
produced by the sciport utilities (to some tolerance) and note any
inconsistencies.
     All scilib utilities up thru COS 1.13 are included.  If we extend
this to later releases, I will forward them to you.  Let me know if
I can be of any assistance.

     I am also interested in offering GE/CRD as a possible test site
for LAPACK, and am discussing this with management here.  We have done
some work with remote procedure calls for linpack utilities and a
delaunay triangulation routine I wrote.  It has enabled us to run
program on a workstation but have computationally intense modules
(linpack or delaunay) run on the CRAY-2 at Minnesota Supercomputer
Center.  It would be interesting to do this with LAPACK if I can
convince CRD to support it.

	                                        scott



        Scott|  ARPA:      lamson@ge-crd.arpa
       Lamson|  UUCP:      uunet!steinmetz!sierra!lamson
(518)387-5795|  GE DECnet: qtmvax::lamson



.