Why is the code so slow?

Hello,

could please someone explain to me, why the second part of this code is so slow?

clear
% A = zeros(360,100000);
tic
for m = 1 : 1000
 b =rand(360,1);
    B(:,m) = b;

end
toc

iter = 1000;
L = 360;
A = ClassTestZeit('foo', 5, 4, L, iter);
tic
for m = 1: iter

   A = A.func1(L,m);
end
toc

Here is the class:

classdef ClassTestZeit
    properties (SetAccess = public, GetAccess = public)
        name string
        mu(1,1) double
        ku(1,1) double
        matrix(:,:) double

    end

    methods
        function obj = ClassTestZeit(name, mu, ku, t, iter)
            obj.name = name;
            obj.mu = mu;
            obj.ku = ku;
            obj.matrix = zeros(t,iter);
        end

        function obj = func1(obj,L, iter)
            
                obj.matrix(:,iter) = obj.func2(L);
        end

        function out = func2(~, L)
             out = rand(L,1);
        end
    end

end

The property "matrix" is preallocated, but it still talks much longer, then the first part.

What I am doing wrong? is there a way to speed it up, and still use OOP.

I hope, that somebody cann help my.

[Image Analyst]

Why are you overwriting the class variable instance "A" with a Lx1 vector of random numbers? The functions of A won't work after that.

[Julian]

Sorry, i don't get it.

I'm writing in each column a vector with random numbers. Whats the right way to do it?

[Walter Roberson]

It is a value class and each time a revised object is being returned and overwrites the previous object. This is semantically valid, even if inefficient.

Answer 1

[Walter Roberson]

1. every interaction with an object other than copying the object as a whole, requires making function calls. MATLAB does not internally compile the references down to minimal form like a C++ compiler could potentially do. At any point during execution you might potentially inject a breakpoint into a class method and the breakpoint must be taken: the methods must be executed as-written, not distilled down to their essence like a compiler potentially could.
2. you are using a value class, so your object is "copy on write". During any one call into an object method, the first time you modify any property of the object, MATLAB has to clone the object infrastructure (but it can increment reference counts of stored property values and store pointers to them, instead of having to deep copy all of the values associated with the object.) This is going to be slower than a plain loop.

If you ever need to hold on to the "old" version of the object along with the new version, you might indeed need a value class. But if every time that you call a method that modifies an object property, your invoking code overwrites the object with the result, then consider using a handle class instead of a value class.

[Julian]

Ok thanks for your help. Now i tried to use a handle class.

That's how I changed the code:

iter = 1000;
L = 360;
A = ClassTestZeit('foo', 5, 4, L, iter);
tic
for m = 1: iter
    A.func1(L,m);
end
toc

classdef ClassTestZeit < handle
    properties (SetAccess = public, GetAccess = public)
        name string
        mu(1,1) double
        ku(1,1) double
        matrix(:,:) double
    end

    methods
        function obj = ClassTestZeit(name, mu, ku, t, iter)
            obj.name = name;
            obj.mu = mu;
            obj.ku = ku;
            obj.matrix = zeros(t,iter);
        end

        function func1(obj, L, iter)
            obj.matrix(:,iter) = obj.func2(L);
        end

        function out = func2(~, L)
            out = rand(L,1);
        end
    end
end

But it didn't changed anything regarding time.