From: "Tim Davis" <davis@cise.ufl.edu>
Path: news.mathworks.com!newsfeed-00.mathworks.com!webcrossing
Newsgroups: comp.soft-sys.matlab
Subject: Re: Wishlist for R2007b
Message-ID: <ef57195.12@webcrossing.raydaftYaTP>
Date: Thu, 17 May 2007 15:20:50 -0400
References: <ef57195.-1@webcrossing.raydaftYaTP> <MPG.20b52ece5ac3d2e798973b@news.mathworks.com> <1179378005.215820.316490@u30g2000hsc.googlegroups.com> <ef57195.8@webcrossing.raydaftYaTP> <1179428426.369097.298150@w5g2000hsg.googlegroups.com>
Lines: 23
NNTP-Posting-Host: 70.171.50.145
MIME-Version: 1.0
Content-Type: text/plain; charset="ISO-8859-1"
Content-Transfer-Encoding: 8bit
Xref: news.mathworks.com comp.soft-sys.matlab:409686



spasmous wrote:
>...
> Hi Tim, thank you for your input. Just for the record I'm basing my
> own comments on comparisons of A * B and A .* B (different
> operations). Whereas A * B is BLAS and The Mathworks is probably
> wise
> to stay away, A .* B is definitely within their scope. When one is
> complex and the other is real, the operation is more expensive than
> pure complex-complex. I'm not really complaining as much as
> explaining ;) Bug reports were submitted a year or two ago...

Oh, I see. A dot-times B (A.*B) doesn't exist since it's not linear
algebra ;-) so I know nothing about how it should work. You're
right, A.*B is BLAS-less.

In MATLAB 7.4, A.*B when A and B are both real take 0.4 seconds on
my laptop. When A is real and B is complex, the time goes up to 0.7
seconds. B.*A takes the same time (both A and B are 3000-by-3000).
The real-complex case runs at 25 MFlops, which is slow but then the
operation is memory-bound. When both are complex, the time is 0.8
seconds. These times sound reasonable to me. Am I wrong?

I have a 2Ghz Pentium 4 Mobile, with 1GB of memory.