What's the cost of overwriting a variable versus clearing it every time it changes?

As someone who wants to optimize the performance of his code, I want to know if it is more efficient to clear a variable each time I know it will be overwritten (e.g. a "temp" in a loop, or if the name is reused later) OR if the act of telling Matlab to instantiate a variable of the same name every time I will overwrite actually hurts more than it hurts.

I can't reason whether giving Matlab the "freedom" of choosing memory for each instantiation is good or if it is better to overwrite data to a designation of memory that already exists.

P.S. This would be using the "clear" command if that makes any difference.

Thanks!

[Stephen23]

Why not simply read the documentation, which states clearly: "Avoid clearing more code than necessary. Do not use clear all programmatically. For more information, see clear."

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

Answer 1

[per isakson]

Comments

• don't speculate; make a test
• make it a habit to use the function, profile
• see premature optimization

and the answer is that clear hurts badly (in the simple case below)

N = 1e6;
tic
val = 17;
for jj = 1 : N
    val = jj;
end
toc
tic
for jj = 1 : N
    clear val
    val = jj;
end
toc

returns

Elapsed time is 0.005792 seconds.
Elapsed time is 17.567347 seconds.

using R2016a (in response to a comment)

[Walter Roberson]

The "clear" appears to be less expensive in newer versions relative to whatever release dpb is using, but he is right that it is much more expensive.

[James Tursa]

See this related discussion:

https://www.mathworks.com/matlabcentral/answers/350077-order-of-variable-deletion-and-creation