Is there a way to improve max & min functions in matlab?

Hello everyone, I'm using max and min functions on a vector of (just) 2 element in a very very long for cycle, and it is simply too slow. Is there a way to avoid using this functions, or, at least, improve them? Thanks in advance for any answer! And sorry for my english, I'm italian =) !

[Stephen23]

http://www.mathworks.com/help/matlab/matlab_prog/techniques-for-improving-performance.html

Answer 1

[dpb]

Well, the first question would be can you vectorize the loop and then do the min/max operation over the array result instead of just a 2-vector?

For two values only you can write an explicit expression--one possibility is to assume one, then check--

mxv=v(1); if v(2)>v(1), mxv=v(2); end

or just the obvious if ... else ... end

Ways you can play with logical addressing as well...altho you have to be certain to not end up with a null return if aren't careful in implementation.

I don't know how much it might speed up anything, though it would eliminate the function overhead. I'd've thunk the JIT optimizer might would've inlined that by now, however, unless it can't tell the size of the input. Or mayhaps TMW figured the case wasn't frequent enough to have analyzed for.

ADDENDUM

Meant to put this first, but better late than never... :)

Before you start such micro-optimizing, have you profiled the app to actually locate these as bottlnecks?

[John D'Errico]

The important thing is, no, you are probably not going to improve max and min, unless you decide to write compiled C code to do the operation, without any tests, and checks, etc. Then you can save some, since these checks will surely be using more time than the actual operation. But only if you will write the code.

But FAR better is to recognize that often in these cases, as dpb points out, that you might be able to reformulate the problem. Don't loop at all if you can avoid it. If you can do so, then you will gain far more speed this way.

[dpb]

+27.3 :)

[Greg Heath]

OK! I give up !

 +27.3 ==> Increase in speed??

Greg

[dpb]

Nah, nothing but a big +1 head nod in agreement...