How to 'load' data inside a parfor loop in an effective way?

Let suppose the following general code structure.

parfor i = 1:10000
    result(i) = myfunc(data(i));  
end

or

parfor i = 1:10000
    result(i) = myfunc(data(i).relevantpart);
end

I think, in both case all the data are copied into memory within each loop, and not just data(i) for loop i.

Am I right?

So I want to avoid this.

As far as I see, the only way to do this, if I have separate data variables, like data1, data2, data3, ..., data10000.

But this is very ugly, and to automatically genarate them outside the loop and use them inside the loop, you have to use the function 'eval', which is not recommended in general.

So what is the solution for this general data 'loading/using' problem?

[Stephen23]

"...you have to use the function 'eval', which is not recommended in general."

Not only is it a bad way to write code, eval doesn't really work inside parfor loops:

https://www.mathworks.com/help/parallel-computing/transparency.html

Answer 1

[Matt J]

Am I right?

No. In the code you have shown, the struct array data will be sliced,

https://www.mathworks.com/help/parallel-computing/sliced-variable.html;jsessionid=84aac55b1cc24a4fe4b489b5ae20

[Merse Gaspar]

Is it possible to check sliceness? Compiler can give me an exact answer based on the code, or is it possible to check during the run?

[Matt J]

The Code Analyzer should warn you if you are broadcasting a non-sliced variable.

[Edric Ellis]

You can also view the amount of data sent to/from a parfor loop using ticBytes and tocBytes.