How to properly define input types for Matlab Coder?

I'm struggling to make a mex file work properly (see more here) so I decided to try and use Matlab Coder to do that hoping that would help. Using Matlab Coder I need to define the inputs, but I encountered 2 issues:

1. How do I define a variable to be 3 dimensional array (the only options I see there are: scalar, vector or matrix)?
2. How do I define an array with dimensions equal to a certain variable (that is also an input)?

Additional details:

The Matlab code I'm trying to convert is:

function [V,delta_V_rms,delta_V_abs,delta_V_max] = potentials_evo(L,t_max,N_eff_occ_sites_bounded,i_eff_V_bounded,j_eff_V_bounded,k_eff_V_bounded,system,over_relax_factor,V,delta_V_rms,delta_V_abs,delta_V_max)


% evolution of the potentials %
% note : for the index directions with periodic boundary conditions: index=mod(index-1,L)+1 . for index=index+1 it is mod(index,L)+1 , and for index=index-1 it is mod(index-2,L)+1 %          
for t=1:t_max
    for q=1:N_eff_occ_sites_bounded % this is set instead of running i=2:(L-1), j=1:L , k=1:L and ending up going over sites that are 0 in our effective system % 
        i=i_eff_V_bounded(q);
        j=j_eff_V_bounded(q);
        k=k_eff_V_bounded(q);
        V0=V(i,j,k);    
        V1=( V(i+1,j,k)+V(i-1,j,k)+V(i,mod(j,L)+1,k)+V(i,mod(j-2,L)+1,k)+V(i,j,mod(k,L)+1)+V(i,j,mod(k-2,L)+1) )/( system(i+1,j,k)+system(i-1,j,k)+system(i,mod(j,L)+1,k)+system(i,mod(j-2,L)+1,k)+system(i,j,mod(k,L)+1)+system(i,j,mod(k-2,L)+1) ); % evolving the potential as the average of its occupied neighbors %
        V(i,j,k)=V0+(V1-V0)*over_relax_factor; % evolving the potentials in time with the over relaxation factor %
        delta_V_rms(t)=delta_V_rms(t)+(V1-V0)^2; % for each t at a given p, we sum over (V1-V0)^2 in order to eventually calculate delta_V_rms_avg %
        delta_V_abs(t)=delta_V_abs(t)+abs(V1-V0); % for each t at a given p, we sum over |V1-V0| in order to eventually calculate delta_V_abs_avg %                
        delta_V_max(t)=max(abs(V1-V0),delta_V_max(t)); % for each t at a given p, we take the max of |V1-V0| from all the sites in order to eventually calculate delta_V_max_avg % 
    end  
end


end

How do I define input for variables like V which is an LxLxL array?

[Adam]

For the first question, just select m x n and then start typing 'x...' after the n when filling in the values. I was confused by this initially too until I realised that!

I don't think you can define sizes in terms of other variables, but I may be wrong, I very rarely use Coder.

[tensorisation]

I find it hard to believe that you can't do that, since it can be done in C, and it's very useful. Without that, how could I properly define an input?

[Adam]

Well, you can leave it as m x n, which will allow it to be a 2d input of any size, which, when the variable is created to be passed in, could come from another variable in code. You just can't hard-define it in the Coder input argument definition to be of a size that is defined by some other variable.

[tensorisation]

Can you elaborate and clarify what exactly do you mean by: "you can leave it as m x n, which will allow it to be a 2d input of any size, which, when the variable is created to be passed in, could come from another variable in code.". I'm still not sure what to do.

Also, when I right-click the a variable in the screen where I'm suppost to define inputs, I noticed there a 'Bounded (fixed-size)' marked, and there is also an unmarked 'Bounded (variable-size)' and 'Unbounded'. I tried clicking these options but nothing happens and the 'Bounded (fixed-size)' option is still marked. Are these options relevant to this, and how do I use them?

[Adam]

Ah yes, it's a while since I used it. If you enter :Inf as the dimension size then it is allowed to be anything, so

:Inf x :Inf x :Inf

will be a 3d array of any size. You would then validate the actual size somewhere else if you wish to do this. Usually I have a .m wrapper file around mex functions I create that does any input validation I need so that the C or C++ code just works on the general case, unless I can hard-code sizes (e.g. if something should always be a 3x3x3 array)

Answer 1

[Soumya Paliwal]

1) How do I define a variable to be 3 dimensional array (the only options I see there are: scalar, vector or matrix)?

• To define a 3D input using the MATLAB Coder app, first select the datatype, then select any one of the options provided (scalar, vector, matrix). After specifying the first two dimensions, leave a space and type the third dimension. The app will understand that the variable is a 3D variable.

2) How do I define an array with dimensions equal to a certain variable (that is also an input)?

• If the reference variable is already defined in the workspace, then in the MATLAB Coder app you can specify your variable to be

        coder.typeof(<variable_in_workspace>)

• Alternatively, when specifying the type and value of a variable, you can choose 'Define by Example' and choose the reference variable (present in the workspace) from the list generated.
• This will ensure that your input argument is of the same data size and same dimensions as the reference variable.
• For more information on coder.typeof, please refer https://www.mathworks.com/help/fixedpoint/ref/coder.typeof.html