From help-octave-request at bevo dot che dot wisc dot edu Thu Jan  8 21:06:11 2004
Subject: Ops on matrix in cell array much slower than on bare matrix
From: "John W. Eaton" <jwe at bevo dot che dot wisc dot edu>
To: gdg at zplane dot com
Cc: help-octave at bevo dot che dot wisc dot edu
Date: Thu, 8 Jan 2004 21:05:53 -0600

On  8-Jan-2004, Glenn Golden <gdg at zplane dot com> wrote:

| # Same thing as above, except using cell array to hold the matrix. Average
| # execution time per assignment is around 100 times larger (2.3 ms) than
| # above.  The slowdown factor is related to the size of the matrix (ysize). 
| # 
| y = cell(1,2);
| y{1} = zeros(ysize,1);
| start = cputime();
| for k = 1:nloops
|     y{1}(1) = 123.4;   

y{1} generates a temporary.  The temporary value is shared with the
original, but a copy is made when you index it for the assignment.

If you really need the speed, then workaround is:

  t = y{1};
  y{1} = [];
  t(1) = 123.4;
  y{1} = t;

Assigning something else to y{1} frees the second reference to the
temporary so that no copy is needed for the assignment.  You won't see
any improvement if you omit that step.

I suppose it would be possible for some extra analysis to make this
happen automatically, but that is not how things are implemented now.

jwe



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------