Why has matlab become extremely inefficient with global variables?

I have a program which I have been running for years on various versions of Matlab. It uses a file of global variables, which I pass between functions. Running profiler:

In Matlab 2012b, the global file took 6 s out of a total 54 s running time

In Matlab 2017a, the global file took 67 s; all the rest was similar to 2012b

In Matlab 2020b, the global file took 320 s; all the rest was similar to 2012b

I realize that using global variable makes me a bad person, but I am not a programmer, don't use classes and just want matlab to help me with my calculations. I don't see why matlab performance should degrade by a factor of 50 in new versions?

[James Lebak]

Hi Neil,

I'm sorry you're experiencing this problem. I did a quick experiment with global variables and I don't see the same slowdown you're describing. Would you be willing to share your global file, or open a service request for this problem?

[Image Analyst]

Like James says (expand the comments), include a .mat file with your global variables. Are you talking about how long it takes to declare global variables in program1.m that have already been read into program1.m? Like

program1.m

g1 = rand(10000); % Somehow create the globa variables and then make them global.
g2 = rand(3000);
tic
global g1
global g2
toc % See how long it takes to put those variables into the "global" workspace.

program2.m

tic
global g1
global g2
toc % Shows 320 seconds.

Or are you perhaps reading the variables in from a mat file or somehow otherwise creating them in program1.m and that's how long it takes to read them in. Exactly how big (how many megabytes) are these variables?

[Neil Marcus]

I did not read from a mat file, and the variables were already defined. A symbolic representation of my issue is: I had a file globals.m defining globals; about 250 scalars, so no big memory issue.

globals.m

global a1

global a2

...

Then I had a program prog2 with various subfunctions using the variables in a loop

prog2.m

globals

a1 = 23;

a2 = 432;

...

for jj = 1:1000

subprog

end

function subprog

globals

a1 = a1+2;

Answer 1

[Jan]

You do not use global variables, but a script. Although you declare the varaibels as globals inside the script, there is no need to do so.

Method 1:

% globals.m
function globals
global a1
global a2
a1 = rand;
a2 = rand;
end

% prog2.m
function prog2
global a1
global a2
% No call of function globals here!
b = a1 + a2;
subprog();
end

function subprog
global a1
a1 = a1+2;
end

I recommend to avoid global variables and scripts. Try this:

Method 2:

% globals.m
function G = globals
G.a1 = rand;
G.a2 = rand;
end

% prog2.m
function prog2
G  = globals();
a1 = G.a1;
a2 = G.a2;

b = a1 + a2;
G = subprog(G);
end

function G = subprog(G)
G.a1 = G.a1+2;
end

I assume the slow down is not cause by using global variable, but scripts. Matlab processes functions for efficiently.

Defining 250 scalars sounds a little bit strange. Did you you the profiler to find the bottleneck of the code in the modern Matlab versions?

[Neil Marcus]

Thanks for the effort

I did use the profiler to check the delay. All the time was on the line "globals" calling the script.

the reason that I used the script, was that I simply didn't want to repeat the 250 lines "global a1, global a2, ..." in every function that used them

I did replace the variables with a global structure (with 250 parts) and this was much more efficient. I still think that there is an issue with the modern matlab being 50 times slower than the old version

[Hiro Yoshino]

You've been using MALTAB for a while, then why don't you try to use "Compatibility report"?

https://www.mathworks.com/help/matlab/matlab_prog/matlab-code-compatibility-report.html

You may find better alternate ways or functions that reproduce what you are seeing with the curret code.

[James Lebak]

@Neil Marcus

We took a look at the sample code from your comment, and we're unable to reproduce the slowdown that you're seeing. If this is still a concern for you or blocking work that you need to do, please open a support ticket so that we can get more details.

Answer 2

[Neil Marcus]

I have made and tested a test example below. In Matlab R2010b (32 bit), tic, for jj = 1:10000, a; end; toc takes 1.87 secs. in matlab 2020b (64 bit) it takes 56 secs

File globals.m

global a1

...

global a25

File b.m

globals

a1 = rand;

...

File a.m

globals

a1 = rand;

...

a24 = rand;

b

[Jan]

The rough paraphrazation of your code is not useful. What happens in the "..."lines? Why do you use a set of scripts to define global variables? Global variables help to share data between functions, while scripts share the variables already with their callers. This means, that the code is a complicated method to impede the efficient access of variables.

Plese post the complete code. Then suggesting an efficient modification does not require to guess your code.

[Steven Lord]

Or if posting the complete code is not an option for whatever reason, do as @James Lebak suggested and "please open a support ticket so that we can get more details." You can open a support ticket from this page.