How to speed up normalization of an array

I have an array say selfn of size 1000X3. I need to normalize it and here is the code I am using:

selfn = selfn ./ repmat( sqrt( sum( selfn.^2, 2 ) ), 1, 3 );

Now this is a bottleneck in my code. Itried using bsxfun as:

mynews=ones(size (selfn)); selfn=bsxfun(@rdivide,selfn,bsxfun(@times,mynews,bsxfun(@power,sum(bsxfun(@power,selfn,2),2),0.5)));

But this doesn't make much change to the time it takes. I was wondering if I can use hypot to speed up this part. Any thoughts or suggestions on further speed improvement? I would really appreciate it if anyone can help.

Answer 1

[W. Owen Brimijoin]

You don't have to use bsxfun three times in that line, just the once. Try this simpler line:

selfn = bsxfun(@rdivide,selfn,sqrt(sum(selfn.^2,2)));

That is ever so slightly faster than your repmat solution (on my machine the elapsed time is 0.001946 versus 0.002416 seconds), but I fear you aren't going to get a whole lot quicker than that. Anyone else know how to beat that time?

[Arti Siddhpura]

Thanks W. Owen Brimijoin. And yeah you are right, it doesn't change elapsed time much.