Problems encountered in Parallel Processing

Dear All,

I tried to use parfor to run my code which is given as follows:

function [A,B] = temp_test(x)
a = [1:2*x]; b = [1:x];
A = zeros(1,x); B = zeros(x,1);
parfor i = 1 : x
    j = x + 1;
    c = func1(a(j),b(i));
    A(i) = c;
    B(i) = a(j) + b(i);
end
function [d] = func1(e,f)
    d = e*f;
end
end

But I was told I can not call func1 like this because it is a nested function. Would anybody tell me how I can fix this problem but I still want to keep the func1 as it is?

Thanks a lot.

Benson

Answer 1

[Matt J]

Is there any reason not to simply make func1 a local function?

function [A,B] = temp_test(x)

    a = [1:2*x]; b = [1:x];
    A = zeros(1,x); B = zeros(x,1);
    parfor i = 1 : x
        j = x + 1;
        c = func1(a(j),b(i));
        A(i) = c;
        B(i) = a(j) + b(i);
    end

end

function [d] = func1(e,f)
    d = e*f;
end

[Benson Gou]

Actually func1 is a very large function. I just created a simple example to show my problem.

Thanks a lot.

Benson

[Matt J]

How does the size of func1 invalidate my advice? A local function can be any size.

[Benson Gou]

The function func1 has hunreds of lines. The code will becomes very hard to debug if I insert all lines of func1 in the loop of parfor. I am wondering if there is a way to solve this problem of nested function while func1 is kept the same way as it is.

[Matt J]

I don't think you read my initial post carefully. If you look, you will see that I did not relocate the code of func1 inside the parfor loop. I simply made it into a local function. You could also move func1 to its own file. This might be a good idea if its code is as long and involved as you say.

[Benson Gou]

Dear Matt,

I am very sorry I did not clearlly read your answer. I tried your changed code and it works very well.

Thanks a lot and have a good weekend!

Benson