Copy-on-Write

If a function does not modify an input argument, MATLAB does not make a copy of the values contained in the input variable.

For example, suppose that you pass a large array to a function.

A = rand(1e7,1);
B = f1(A);

The function f1 multiplies each element in the input array X by 1.1 and assigns the result to the variable Y.

function Y = f1(X)
Y = X.*1.1; % X is a shared copy of A
end

Because the function does not modify the input values, the local variable X and the variable A in the caller's workspace share the data. After f1 executes, the values assigned to A have not changed. The variable B in the caller's workspace contains the result of the element-wise multiplication. The input is passed by value. However, no copy is made when calling f1.

The function f2 does modify its local copy of the input variable, causing the local copy to be unshared with input A. The value of X in the function is now an independent copy of the input variable A in the caller's workspace. When f2 returns the result to the caller's workspace, the local variable X is destroyed.

A = rand(1e7,1);
B = f2(A);

function Y = f2(X)
X = X.*1.1; % X is an independent copy of A
Y = X;      % Y is a shared copy of X
end

Passing Inputs as MATLAB Expressions

You can use the value returned from a function as an input argument to another function. For example, use the rand function to create the input for the function f2 directly.

B = f2(rand(1e7,1));

The only variable holding the value returned by rand is the temporary variable X in the workspace of the function f2. There is no shared or independent copy of these values in the caller's workspace. Directly passing function outputs saves the time and memory required to create a copy of the input values in the called function. This approach makes sense when the input values are not used again.

Assigning In-Place

When you do not need to preserve the original input values, you can assign the output of a function to the same variable that you provided as input.

A = f2(A);

In-place assignment follows the copy-on-write behavior described previously: modifying the input variable values results in a temporary copy of those values.

MATLAB can apply memory optimizations under certain conditions. Consider the following example. The canBeOptimized function creates a large array of random numbers in the variable A. Then it calls the local function fLocal, passing A as the input, and assigning the output of the local function to the same variable name.

function canBeOptimized
A = rand(1e7,1);
A = fLocal(A);
end
function X = fLocal(X)
X = X.*1.1;
end

Because the call to the local function, A = fLocal(A), assigns the output to the variable A, MATLAB does not need to preserve the original value of A during execution of the function. Modifications made to X inside fLocal do not result in a copy of the data. The assignment X = X.*1.1 modifies X in place, without allocating a new array for the result of the multiplication. Eliminating the copy in the local function saves memory and improves execution speed for large arrays.

However, MATLAB cannot apply this optimization if the assignment in the local function requires array indexing. For example, modifying the cell array created in updateCells requires indexing into X in the local function gLocal. For every loop iteration i, the looped assignment in the form X{i} = X{i}*1.1 results in a temporary variable the same size as X{i} for evaluating and storing the value of X{i}*1.1. MATLAB destroys the temporary variable after assigning its value to X{i}.

function updateCells
C = num2cell(rand(1e7,1));
C = gLocal(C);
end
function X = gLocal(X)
for i = 1:length(X)
    X{i} = X{i}*1.1;
end
end

Several additional restrictions apply. MATLAB cannot apply memory optimization when it is possible to use the variable after the function throws an error. Therefore, this optimization is not applied in scripts, on the command line, in calls to eval, or to code inside try/catch blocks. Also, MATLAB does not apply memory optimization when the original variable is directly accessible during execution of the called function. For example, if fLocal was a nested function, MATLAB could not apply the optimization because variables can be shared with the parent function. Finally, MATLAB does not apply memory optimization when the assigned variable is declared as global or persistent.

Debugging Code That Uses In-Place Assignment

When MATLAB applies in-place optimization to an assignment statement, the variable on the left side of the assignment is set to a temporary state that makes it inaccessible before MATLAB executes the right side of the assignment statement. If MATLAB stops in the debugger before the result of executing the right-side of the statement has been assigned to the variable, examining the left-side variable can produce an error indicating that the variable is unavailable.

For example, this function has a mismatch in the dimensions of variables A and B.

function A  = inPlace
A = rand(100);
B = rand(99);
dbstop if error
A = A.*B;
end

Executing the function throws an error and stops in the debugger.

inPlace

Arrays have incompatible sizes for this operation.

Error in inPlace (line 5)
A = A.*B;

Attempting to see the value of the variable A while in debug mode results in an error because the variable is temporarily unavailable.

K>> A

Variable "A" is inaccessible. When a variable appears on both sides of
an assignment statement, the variable may become temporarily
unavailable during processing.

To gain more flexibility when debugging, refactor your code to remove the in-place assignment. For example, assign the result to another variable.

function A  = inPlace
A = rand(100);
B = rand(99);
dbstop if error
% Assign result to C instead of A
C = A.*B;
A = C;
end

Then the variable A is visible while in the debugger.