How to implement the rmse as a performance function for a neural network

I would like to use the rmse as the performance function for my neural network, instead of the mse. I use the default code from the NN toolbox in Matlab. As the rmse isn't implemented in Matlab I try to set it up myself (see net.performFcn) . To do so I need access to the output of the NN in the workspace, which appears only after training and testing the NN. However the performance function is needet to train the network.. How can I work around this? Or what do I need to do to implement the rmse properly? I already searched Matlab answers and the documentation.

...
% Setup Division of Data for Training, Validation, Testing
net.divideParam.trainRatio = 70/100;
net.divideParam.valRatio = 15/100;
net.divideParam.testRatio = 15/100;

net.trainParam.max_fail = 6;
net.trainParam.epochs = n;

Z=cell2mat(T);%target

net.performParam.normalization = 'standard';
net.performFcn=sqrt(sum((Z(:)-t1_Output1(:)).^2)/numel(Z));

% Train the Network
[net,tr] = train(net,x,t,xi,ai);

% Test the Network
y = net(x,xi,ai);
e = gsubtract(t,y);
performance_ol = perform(net,t,y);
...

[victor mukungunugwa]

See this example

Anomaly Detection in Industrial Machinery Using Three-Axis Vibration Data - MATLAB & Simulink - MathWorks United Kingdom