Matlab R2015a coder error:Dimension 1 is fixed on the left-hand side but varies on the right ([1024 x 1] ~= [:? x 1]).

I use Matlab R2015a coder to compile a .m function to C code.But get three errors.How can I solve these errors?Thank you for any advice. here is the errors:

• line 117:Dimension 1 is fixed on the left-hand side but varies on the right ([1024 x 1] ~= [:? x 1]).
• line 118:Dimension 2 is fixed on the left-hand side but varies on the right ([60 x 1024] ~= [60 x :?]).
• line 119:Dimension 1 is fixed on the left-hand side but varies on the right ([1024 x 1] ~= [:? x 1]).

and line117-119 is:

gamma = gamma(index);   
Phi = Phi(:,index);            % corresponding columns in Phi
keep_list = keep_list(index);

below is my code:

function [X, gamma_ind, gamma_est, count] = MFOCUSS_edit7(Phi, Y, lambda)
% MFOCUSS algorithm for the MMV model
%    
% ============================== INPUTS ============================== 
%   Phi           : N X M dictionary matrix
%
%   Y             : N X L measurement matrix, i.e. Y = Phi * X + V.
%
%   lambda        : Regularization parameter. Generally it is close to the noise variance. 
%                   In the noiseless cases, simply setting lambda = 1e-10 leads to good performance.
%                   In noisy cases, need the modified L-curve method to find optimal lambda. 
%                    See Ref [1] for details.
%
%  'p'            : p-norm. p lies in [0,1].  Default value: p = 0.8
%
%  'PRUNE_GAMMA'  : Threshold for prunning small gamma_i.
%                   In noisy cases, you can set PRUNE_GAMMA = 1e-3 or 1e-4.
%                   In strongly noisy cases (SNR<=5 dB), suggest to set PRUNE_GAMMA = 0.01;
%                   Default value: MIN_GAMMA = 1e-4. 
%                                   
%  'MAX_ITERS'    : Maximum number of iterations.
%                   Default value: MAX_ITERS = 800
%
%  'EPSILON'      : Threshold to stop the whole algorithm. 
%                   Default value: EPSILON = 1e-8
%
%  'PRINT'        : Display flag. If PRINT = 1: show output; If PRINT = 0: supress output
%                   Default value: PRINT = 0
%
% ==============================  OUTPUTS ============================== 
%   X            : Estimated solution matrix(size: M X L)
%   gamma_ind    : Indexes of nonzero gamma_i
%   gamma_est    : Final values of all the gamma_i (including zeros). An M X 1 vector
%   count        : number of iterations used
%
% ============== Examples of Commands ===============
% [Example 1]
%   lambda = 1e-3;
%   [X,gamma_ind,gamma_est,count] ...
%       = MFOCUSS(Phi,Y,lambda,'p',0.8,'prune_gamma',1e-4,'max_iters',500,'epsilon',1e-8,'print',0);
%
% [Example 2]
%      lambda = 1e-5;  
%      [X,gamma_ind,gamma_est,count] = MFOCUSS(Phi,Y, lambda);
%
% ==============================  Reference =============================
%   [1] Cotter, S.F.;   Rao, B.D.;   Kjersti Engan;   Kreutz-Delgado, K.; 
%       Sparse solutions to linear inverse problems with multiple measurement vectors
%
% ==============================  Author ============================== 
%   Zhilin Zhang ([e-mail])
%   Mainbody was written by David Wipf
% 
% ==============================  Version ============================== 
%   1.0 (05/12/2011)
%
% ==============================  See Also ============================== 
%   TSBL    TMSBL     ARSBL    tMFOCUSS
%

% Dimension of the Problem
[N, M] = size(Phi); 
[N, L] = size(Y);  

% Default Control Parameters 
PRUNE_GAMMA = 1e-4;        % threshold for prunning small gamma_i
p           = 0.8;         % p-norm
EPSILON     = 1e-8;        % threshold for stopping iteration. 
MAX_ITERS   = 800;         % maximum iterations
PRINT       = 0;           % not show progress information

% get input argument values
% if(mod(length(varargin),2)==1)
%     error('Optional parameters should always go by pairs\n');
% else
%     for i=1:2:(length(varargin)-1)
%         switch lower(varargin{i})
%             case 'p'
%                 p = varargin{i+1}; 
%             case 'prune_gamma'
%                 PRUNE_GAMMA = varargin{i+1}; 
%             case 'epsilon'   
%                 EPSILON = varargin{i+1}; 
%             case 'print'    
%                 PRINT = varargin{i+1}; 
%             case 'max_iters'
%                 MAX_ITERS = varargin{i+1};  
%             otherwise
%                 error(['Unrecognized parameter: ''' varargin{i} '''']);
%         end
%     end
% end
% 
% if (PRINT) fprintf('\nRunning M-FOCUSS for the MMV Problem...\n'); end

% Initializations 
gamma = ones(M,1);         % initialization of gamma_i
keep_list = [1:M]';        % record the index of nonzero gamma_i
m = length(keep_list);     % number of nonzero gamma_i
mu = complex(zeros(M,L));           % initialization of the solution matrix
count = 0;                 % record iterations

% Learning loop 
while (1)

      % =========== Prune weights as their hyperparameters go to zero ===========
      if (min(gamma) < PRUNE_GAMMA )
          index = find(gamma > PRUNE_GAMMA);

%         tmp = [];
%         for i =1:1:length(index)
%             tmp(i) = gamma(index(i))
%         end
%         gamma = tmp'
        gamma = gamma(index);   
        Phi = Phi(:,index);            % corresponding columns in Phi
        keep_list = keep_list(index);
        m = length(gamma);

          if (m == 0)   break;  end;
      end;

      % ====== Compute new weights ======
      G = repmat(sqrt(gamma)',N,1);
      PhiG = Phi.*G;
      [U,S,V] = svd(PhiG,'econ');
  %     [U,S,V] = svd_edit(PhiG);
      [d1,~] = size(S);
      if (d1 > 1)     diag_S = diag(S);
      else            diag_S = S(1);      end;

      U_scaled = U(:,1:min(N,m)).*repmat((diag_S./(diag_S.^2 + sqrt(lambda) + 1e-16))',N,1);
      Xi = G'.*(V*U_scaled');

      mu_old = mu;
      mu = Xi*Y;

      % *** Update hyperparameters ***
      gamma_old = gamma;
      mu2_bar = sum(abs(mu).^2,2);
      gamma = (mu2_bar/L).^(1-p/2);

      % ========= Check stopping conditions, etc. ========= 
      count = count + 1;
  %     if (PRINT) disp(['iters: ',num2str(count),'   num coeffs: ',num2str(m), ...
  %             '   gamma change: ',num2str(max(abs(gamma - gamma_old)))]); end;
      if (count >= MAX_ITERS) break;  end;

      [m_mu,n_mu] = size(mu);
      [m_mu_old,n_mu_old]= size(mu_old);

%     if (size(mu) == size(mu_old))
    if ((m_mu == m_mu_old) && (n_mu == n_mu_old))
        dmu = max(max(abs(mu_old - mu)));
        if (dmu < EPSILON)  break;  end;
    end;

end;

gamma_ind = sort(keep_list);
gamma_est = zeros(M,1);
gamma_est(keep_list,1) = gamma;  

% expand the final solution
X = complex(zeros(M,L));
X(keep_list,:) = mu;   

% if (PRINT) fprintf('\nM-FOCUSS finished !\n'); end
return;

Answer 1

[Mike Hosea]

Use coder.varsize to create a variable-size copy of the Phi input. Rename the Phi input to something like PhiInput and just before the line "% Dimension of the Problem", add

coder.varsize('Phi');
Phi = PhiInput;

[WillandSmith]

Thank you!That's it.Now the errors are gone.

[Ummi Hani]

Hi, could you explain the meaning of this error. I don't really understand it and I can't apply this answer to my own error

[Walter Roberson]

The statements

gamma = ones(M,1);         % initialization of gamma_i
keep_list = [1:M]';        % record the index of nonzero gamma_i

create gamma and keep_list as fixed length (once M is known), but the statements

        gamma = gamma(index);   
        Phi = Phi(:,index);            % corresponding columns in Phi

redefine gamma and Phi as being sizes defined according to run-time computation and so being of unknown length until the find() and subsequent statements are completed.

In Simulink, variables cannot just change size like that unless special preparations have been made.

You would not normally get a message when you assign a subset of a variable over top of the variable, but in this case Simulink does not know that the results of find() cannot be larger than the original array. Then too, you are doing the size changes within a loop, and Simulink is not smart enough to analyze to prove that the sizes only ever stay the same or go down.

[Jeremy Rutman]

We tried using coder.varsize for our case (which hits the same error, and is caused by trying to set structure fields in a loop). However when we tried this we hit the error "directly accessing a field of a nonscalar struct is not supported for code generation. This limitation also applies to a property of a nonscalar object". Is there some other way to get around this limitation?

[Walter Roberson]

@Jeremy Rutman

Explicity index the struct by (1) to avoid any problem with the code incorrectly inferring that the struct is nonscalar.

[Jeremy Rutman]

The struct is indeed an array struct, does this mean I can't use it for codegen?

When we try

[ud.(fieldName)(1), currentIndex] = fieldAssign(serializedData, ts.(fieldName)(1), currentIndex);
[ud.(fieldName)(2), currentIndex] = fieldAssign(serializedData, ts.(fieldName)(2), currentIndex);
...

the model runs fine, but when we put this into a loop we get an error like the OP's error ,in our case

[1x1]!=[:?x1] . Is there some way to keep the loop ?

for k = 1:arraySize
    [ud.(fieldName)(k), currentIndex] = fieldAssign(serializedData, ts.(fieldName)(k), currentIndex);
end

[Walter Roberson]

What is the code for fieldAssign() ?

[Jeremy Rutman]

fieldAssign() just fills the structure fields with their associated values. Since the fields are of various types we can't just replace the RHS of the assignment with 0 or the like . As mentioned, the assignment works fine when not looped. Are there other workarounds for this type of situation?

[Walter Roberson]

What is the code for fieldAssign() ?

Details matter in this case.