MATLAB Answers

Hesham
0

Alternative to using global for a very large array

Asked by Hesham
on 16 Sep 2019
Latest activity Commented on by Hesham
on 17 Sep 2019
Hi
I have a quite large array of data (100s of MB) that I am using as a LUT in multiple functions. If I declare it as a parameter, it will be copied locally by each function, which is an unjustified waste of memory. The solution I am currently using is to define it as global. I know that globals are highly discouraged, but is there a smarter alternative for my case?

  1 Comment

As per isakson explained to you, you try to FIX something that does not exists and hurt your code.
MATLAB does not copy the matrix if you call it as parameter, or even assign it to new variable
BIG = rand(1e4);
B = BIG;
C = foo(BIG);
function C = foo(A)
C = A;
end
A, B, C use a shared-memory of data.
So remove this global and program normally as you would and let MATLAB does the work.
NOTE: However if you do this there is 2 big matrices in memory after the third statement
BIG = rand(1e4);
B = BIG;
B(1) = 10;

Sign in to comment.

Products


Release

R2017a

1 Answer

Answer by per isakson
on 17 Sep 2019
Edited by per isakson
on 17 Sep 2019
 Accepted Answer

"unjustified waste of memory" Matlab is smarter than that, see Avoid Unnecessary Copies of Data.
If LUT is look-up-table and the called function doesn't change the LUT-value no copies are made.
See also Memory Management for Functions and Variables. This blog is more that ten years and Matlab evolves, but it's worth reading.

  1 Comment

Sign in to comment.