One Multi Dimensional Array or Several Two Dimensional Arrays ?

Hello everyone, I'm writing a code that includes calling back a specific variable several times, i used Matlab Profiler in order to determine the specific time for every part of the code and it was 95 % of the run time associated with calling back the variable values, Now the variable that's being called back is actually 9 variables or 9 arrays f1(i,j), f2(i,j) , f3 (i,j),... etc. where i,j are changing using a for loop in order to fill the arrays, So i have 9 arrays that are Two Dimensional,

My question is, Which is faster: 1) To run the code as it is using 9 Two Dimensional Arrays f1(i,j), f2(i,j), f3(i,j) ...etc. or 2) To use One Multi Dimensional Array f(i,j,1) , f(i,j,2) , f(i,j,3) ... etc.

Knowing that the functions f1,f2,f3.. etc. are not dependent on each other (can i run in parallel ?).

My Goal is to minimize the code run time.

Thanks in advance

Mahmoud

Answer 1

[Walter Roberson]

Vectorize to do the entire function at one time. If you need to, use ndgrid:

[I, J] = ndgrid(1:maximum_i, 1:maximum_j);
f1(I,J)

[Mahmoud Sedahmed]

Thanks Walter, i think you mean that i can use ndgrid to build the whole array f1(i,j) in a single step, if i'm not getting it wrong, But this is not my case, as in the code the values of f1(i,j) will change in the next loop due to further calculations and so on, so i want to know which is faster, is it to save the arrays as f1(i,j) , f2(i,j) ... etc. or to save only one array in the form f(i,j,1), f(i,j,2) ... etc. in other words, to have several arrays in the work space (like 9 arrays for example) or to have only array in the work space and this array to be multi dimensional array ?

I hope you got me.

Thanks.

[Walter Roberson]

The time difference will be small for that situation. In my tests, the difference was about 1.5% at best, and varied from run to run; small differences in the code being executed would easily have overwhelmed the change.

Basically, unless you are expecting the code to be run for a few days of CPU, don't worry about it; it would take more of your time to work it out than to just execute as-is.

If you really really need to have it run faster, run timing tests on both versions using timeit(). But vectorizing could save you far more execution time than this ever would.

[Mahmoud Sedahmed]

Thanks Walter, unless the form of multidimensional array is saving large run time i won't re-write the code, it would take a lot of time to re-write the code in the new form of multidimensional arrays, and i guess that won't help much. Although i would try vectorizing, Thanks a lot.