X-Recipient:	archive-cygwin AT delorie DOT com
X-SWARE-Spam-Status:	No, hits=-2.8 required=5.0 tests=AWL,BAYES_00
X-Spam-Check-By:	sourceware.org
Message-ID:	<474871.34239.qm@web25501.mail.ukl.yahoo.com>
Date:	Tue, 3 Nov 2009 23:29:39 +0000 (GMT)
From:	Marco Atzeri <marco_atzeri AT yahoo DOT it>
Subject:	source package for atlas 3.9.16 on cygwin 1-7
To:	cygwin AT cygwin DOT com
MIME-Version:	1.0
X-IsSubscribed:	yes
Mailing-List:	contact cygwin-help AT cygwin DOT com; run by ezmlm
List-Id:	<cygwin.cygwin.com>
List-Subscribe:	<mailto:cygwin-subscribe AT cygwin DOT com>
List-Archive:	<http://sourceware.org/ml/cygwin/>
List-Post:	<mailto:cygwin AT cygwin DOT com>
List-Help:	<mailto:cygwin-help AT cygwin DOT com>, <http://sourceware.org/ml/#faqs>
Sender:	cygwin-owner AT cygwin DOT com
Mail-Followup-To:	cygwin AT cygwin DOT com
Delivered-To:	mailing list cygwin AT cygwin DOT com

Hi All,
I prepared a (very preliminary) source package for=20
Automatically Tuned Linear Algebra Software (ATLAS).
It is a fast alternative to the reference BLAS/LAPACK build.
=20
ATLAS 3.9.16 packages, source and Core232SSE3 lib,=20
for cygwin-1.7 are at=20=20
http://matzeri.altervista.org/cygwin-1.7/atlas/index.html

The source package should allow to build any other=20
specific core version, if not please let me know.

Of course you need cygport ;-) and gcc-4 + gfortran

The SciViews  bechmark=20
http://www.sciviews.org/benchmark/index.html
was used to compare the blas/lapack reference with Atlas=20
one, and the outcome is:

=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
Lapack 3.2.1 vs Atlas-3.9.16 with  Core232SSE3 on Cygwin-1.7

   I. Matrix calculation
   ---------------------
700x700 cross-product matrix (b =3D a' * a)___________ (sec):      0.3564  =
0.0653

Linear regression over a 600x600 matrix (c =3D a \\ b')(sec):      0.1269  =
0.0554

   II. Matrix functions
   --------------------
Determinant of a 650x650 random matrix______________ (sec):      0.1586  0.=
0643

Cholesky decomposition of a 900x900 matrix__________ (sec):      0.2319  0.=
0686

Inverse of a 400x400 random matrix__________________ (sec):      0.1030  0.=
0455

=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
So at least a factor 2 up to a factor 6 improvement.


I am very interested to receive feedback on the source package
and the performance comparison for the various core.

At the end of the build the lib package is called=20
libatlas0_RenameWithCore-3.9.16-1.tar.bz2=20

and it contains the dll's

$ tar -tjf libatlas0_RenameWithCore-3.9.16-1.tar.bz2=20
usr/bin/
usr/bin/cygatlas-0.dll
usr/bin/cygblas-0.dll
usr/bin/cygclapack-0.dll
usr/bin/cyglapack-0.dll

that you can copy on /usr/bin or on usr/local/bin as you prefer

The lapack dll are behind in the path=20

$ cygcheck -l liblapack0=20
/etc/profile.d/lapack0.csh
/etc/profile.d/lapack0.sh
/usr/lib/lapack/cygblas-0.dll
/usr/lib/lapack/cyglapack-0.dll

so they will be overcome, without being removed.
=20=20
Regards
Marco Atzeri

cygwin package mantainer for lapack




--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

- Raw text -