@Johannes – you might try to employ in-place data manipulation. See chapter 9 of my performance-tuning book for details.

thank you much for your many insights into Matlab performance. I am currently working with very large datasets in Matlab (20 GB and more) which I need to reshape, crop, filter, … . My code uses dedicated functions for most operations for clarity and reusability. I pass structures containing the data set to those functions and return an updated version of that structure back to the invoking function. As a simple example:

A.data = rand(50,50);
A = my_function(A);

where my_function would then do something like:

function A = my_function(A)
A.data = filter(A.data);
end

This does not seem to be memory efficient since a copy of the original data is created and hence I need twice as much memory at some point which becomes a problem if the data array is bigger than half my physical memory. I was wondering if you have an idea on how to solve this? Would it be wise to store my large dataset in a global variable and operate on that? Are there potential drawbacks to this approach?

Best,
Johannes

@Steve – As I noted in my article: In order to benefit from in-place optimizations of function results, we must both use the same variable in the caller workspace (x = function1(x)) and also ensure that the called function is optimizable (e.g., function x = function1(x)) – if any of these two requirements is not met then in-place function-call optimization is not performed.

The way I understand it, based on Loren’s comment back in 2007, is that funcB would NOT be optimized, unless the JIT was improved in recent years to take this into account. If this is the case, then perhaps a kind MathWorker could enlighten us with a follow-up comment here. Otherwise we should assume that what I wrote above is still correct.

aMatrix = funcA(aMatrix); % assignment to same var

function [param] = funcA(param)
% param in and param out are the same var
param = param * 10;
end

verses:

aMatrix = funcB(aMatrix); % assignment to same var

function [param_out] = funcB(param_in)
% param in and param out are Not the same var
param_out = param_in * 10;
end