Undefined function ' ' for input arguments of type 'gpuArray'. - Why ?

So Im using Paralell Computing Toolbox v. 23.2, on R2023b and having this problem when passing an array from one function to another, and can't seem to understand what is causing this.

My nested loop looks like this, everything is initiated correctly in its running, except for when I add the nu_t term.

for t_n = 1:t-1
        for i = 2:length(X) - 1
            for j = 2:length(Y) - 1
                nu_t = addTurbulentViscotiy(U_temp, C_s, Delta)
            end
        end
        U_temp = applyNoSlipBoundary(U_temp, t_n);
        P_temp = applyNeumannBoundary(P_temp, t_n);
end

And the function that doesnt work looks like this:

function [nu_t] = addTurbulentViscotiy(U, C_s, Delta)

    % Computing the strain rate tensor components for S
    S_11     = (U(t_n, i+1, j, 1) - U(t_n, i-1, j, 1)) / (2 * Delta); 
    S_22     = (U(t_n, i, j+1, 2) - U(t_n, i, j-1, 2)) / (2 * Delta); 
    S_12     = 0.5 * ((U(t_n, i, j+1, 1) - U(t_n, i, j-1, 1)) / (2 * Delta) ...
                + (U(t_n, i+1, j, 2) - U(t_n, i-1, j, 2)) / (2 * Delta)); 
    S_21 = S_12;

    % Calculating the magnitude of S
    S_mag   = sqrt(S_11.^2 + S_22.^2 + 2* S12.^2);

    % Calculating the turbulent Viscocity
    nu_t = (C_s*Delta).^2*S_mag
end

However, these function work. Its the same values being passed:

function [U_bound] = applyNoSlipBoundary(U_field, t_n)
    U_field(t_n + 1, 1, :, :) = 0; % Top boundary
    U_field(t_n + 1, end, :, :) = 0; % Bottom boundary
    U_field(t_n + 1, :, 1, :) = 0; % Left boundary
    U_field(t_n + 1, :, end, :) = 0; % Right boundary
    U_bound = U_field;
end

% Apply Neumann boundary conditions (zero gradient)
function [P_bound] = applyNeumannBoundary(P_temp, t_n)
    P_temp(t_n + 1, 1, :) = P_temp(t_n + 1, 2, :); % Top boundary
    P_temp(t_n + 1, end, :) = P_temp(t_n + 1, end-1, :); % Bottom boundary
    P_temp(t_n + 1, :, 1) = P_temp(t_n + 1, :, 2); % Left boundary
    P_temp(t_n + 1, :, end) = P_temp(t_n + 1, :, end-1); % Right boundary
    P_bound = P_temp;
end

The error gotten is

Undefined function 'NSsolverTest2' for input arguments of type 'gpuArray'.

The functions are in the same file, altough not nested.

Why do they work in one function but not the other? What is causing this I looked through the documentations but can't find anything there. The only difference is where in the loop theyre being called. Anybody that could help me figure this out?

Thanks in advance!

[Joss Knight]

Are you sure this function is spelled correctly where you are calling it? Try selecting the function and hitting Ctrl-D.

Answer 1

[Nipun]

Hi Elias,

I understand that you are having trouble with the "Parallel Computing Toolbox" in MATLAB when passing arrays between functions, particularly when adding the "nu_t" term in your nested loop. The error "Undefined function 'NSsolverTest2' for input arguments of type 'gpuArray'" indicates that the function "addTurbulentViscosity" may not be compatible with GPU arrays. Here's how you can address this issue:

1. Ensure GPU Compatibility: Ensure that the functions and operations within "addTurbulentViscosity" are compatible with GPU arrays.
2. Debugging with CPU Arrays: Test the code with CPU arrays to identify if the issue is specifically related to GPU arrays.
3. Check Function Scope: Make sure the function "addTurbulentViscosity" is accessible within the scope where it is called.

Here's a modified version of your code with added checks and changes for GPU compatibility:

% Main loop with added checks
for t_n = 1:t-1
    for i = 2:length(X) - 1
        for j = 2:length(Y) - 1
            nu_t = addTurbulentViscosity(U_temp, C_s, Delta, t_n, i, j);
        end
    end
    U_temp = applyNoSlipBoundary(U_temp, t_n);
    P_temp = applyNeumannBoundary(P_temp, t_n);
end

% Updated function with additional arguments for indices
function [nu_t] = addTurbulentViscosity(U, C_s, Delta, t_n, i, j)
    % Ensure U is on the CPU if it's a GPU array
    if isa(U, 'gpuArray')
        U = gather(U);
    end
    
    % Computing the strain rate tensor components for S
    S_11 = (U(t_n, i+1, j, 1) - U(t_n, i-1, j, 1)) / (2 * Delta); 
    S_22 = (U(t_n, i, j+1, 2) - U(t_n, i, j-1, 2)) / (2 * Delta); 
    S_12 = 0.5 * ((U(t_n, i, j+1, 1) - U(t_n, i, j-1, 1)) / (2 * Delta) ...
                + (U(t_n, i+1, j, 2) - U(t_n, i-1, j, 2)) / (2 * Delta)); 
    S_21 = S_12;
    
    % Calculating the magnitude of S
    S_mag = sqrt(S_11.^2 + S_22.^2 + 2 * S_12.^2);
    
    % Calculating the turbulent Viscocity
    nu_t = (C_s * Delta).^2 * S_mag;
end

% Ensure other functions are also compatible with CPU arrays
function [U_bound] = applyNoSlipBoundary(U_field, t_n)
    if isa(U_field, 'gpuArray')
        U_field = gather(U_field);
    end
    U_field(t_n + 1, 1, :, :) = 0; % Top boundary
    U_field(t_n + 1, end, :, :) = 0; % Bottom boundary
    U_field(t_n + 1, :, 1, :) = 0; % Left boundary
    U_field(t_n + 1, :, end, :) = 0; % Right boundary
    U_bound = U_field;
end

function [P_bound] = applyNeumannBoundary(P_temp, t_n)
    if isa(P_temp, 'gpuArray')
        P_temp = gather(P_temp);
    end
    P_temp(t_n + 1, 1, :) = P_temp(t_n + 1, 2, :); % Top boundary
    P_temp(t_n + 1, end, :) = P_temp(t_n + 1, end-1, :); % Bottom boundary
    P_temp(t_n + 1, :, 1) = P_temp(t_n + 1, :, 2); % Left boundary
    P_temp(t_n + 1, :, end) = P_temp(t_n + 1, :, end-1); % Right boundary
    P_bound = P_temp;
end

For more information on gpuArray, refer to the following MathWorks documentation: https://www.mathworks.com/help/parallel-computing/gpuarray.html

Hope this helps.

Regards,

Nipun