Issue "Found unsupported class for variable using function 'coder.load'."

Model.mat includes two griddedInterpolant objects, with spline inter- and extrapolation.

function [T_out, V_out] = my_interpolant_eval(A,B,C)

% persistent speichert den Interpolant im Speicher

%

persistent F1;

persistent F2;

if isempty(F1)

data = coder.load('Model.mat');

F1 = data.T_interp;

F2 = data.V_interp;

end

T_out = F1(A,B,C);

V_out = F2(A,B,C);

end

in Codegeneration -->

Found unsupported class for variable using function 'coder.load'.

Value at 'data.T_interp' unsupported for code generation.

Use the command 'whos -file Model.mat' to view the variables in the MAT file.

Not sure what needs to be changed, so it's working.

[Torsten]

If you use the command

whos -file Model.mat

to view the variables in the MAT file, what do you get as output ?

[Michael]

The content of the Mat container is:

whos -file S10_Model.mat

Name Size Bytes Class Attributes

T_interp 1x1 9368 griddedInterpolant

V_interp 1x1 9368 griddedInterpolant

The function works per direct prompt as it should:

[T,V] = my_interpolant_eval(10, 70, 2000)

T =

34.2115

V =

1.0406

[Harald]

Hi,

I have tried to create a simple example to reproduce this and it worked fine. Could you provide more information?

• A sample Model.mat file.
• The context in which you try to use this. I was guessing a MATLAB Function block of a Simulink model, but I might be wrong.

Best wishes,

Harald

[Harald]

Hi again,

I had tried in R2026b and only realized that you are working in R2024a, sorry for this. I could reproduce the problem in R2024a..

If you can move to R2026b, this would be the easiest solution.

In R2024a, I tried a similar approach to what Torsten suggested and it worked fine.

If neither is suiting your needs, please consider contacting Technical Support: https://de.mathworks.com/support/contact_us.html

Best wishes,

Harald

Answer 1

[Torsten]

Answer from AI:

This error occurs because coder.load cannot reconstruct an object instance (like griddedInterpolant) directly from a MAT file during code generation. [1]

Even though the griddedInterpolant class itself supports C/C++ code generation, coder.load is strictly meant to load raw compile-time data constants (such as structures, matrices, and vectors). It treats objects/classes as an unsupported format. [1]

🛠️ The Fix: Separate Data from the Object

To resolve this, you must only save the raw grid points and values inside your .mat file, load them via coder.load, and then instantiate the griddedInterpolant object inside your code. [1]

Step 1: Save only the raw arrays to your MAT file

Instead of saving the entire griddedInterpolant object, save its properties (GridVectors and Values): [1]

matlab

% In your MATLAB base workspace script:
F = griddedInterpolant(X, Y, Z, V, 'linear', 'linear');


% Extract raw data arrays
grid_x = F.GridVectors{1};
grid_y = F.GridVectors{2};
grid_z = F.GridVectors{3};
vals   = F.Values;


% Save ONLY the matrices/vectors
save('ModelData.mat', 'grid_x', 'grid_y', 'grid_z', 'vals');

Step 2: Reconstruct the Object in your Codegen Function

Modify your function to use persistent variables so the object is only created once (during the first call), loading the raw values via coder.load: [1]

matlab

function V_out = my_interpolant_eval(xq, yq, zq)
    
% Keep the object instance in memory across calls
    
persistent F;
    
    
if isempty(F)
        
% 1. Load ONLY the raw numeric vectors/matrices
        data = coder.load('ModelData.mat');
        
        
% 2. Re-instantiate the griddedInterpolant inside the code
        
% Note: 'linear' and extrapolation methods must be compile-time constants!
        F = griddedInterpolant({data.grid_x, data.grid_y, data.grid_z}, data.vals, 'linear', 'linear');
    
end

% Evaluate the interpolant
V_out = F(xq, yq, zq);

end