How can I guarantee Function input Data to be in a certain shape for Coder translated C Code?

I have the following Matlab Function:

function[Product] = computeDotProduct( vecA, vecB )

    Product = double(0.0);
    errors = char('');

    %%Check Shape of Input Data
    if( isrow(vecA) == false || iscolumn(vecB) == false)
        errors = 'Input Vectors have wrong shape!';
        return;
    end

    %%Compute the dot Product.
    Product = vecA * vecB;

    errors = 'OK';
  end

I then use the Matlab Coder Tool box to translate the function into C Code with the following command:

codegen computeDotProduct.m -args {coder.typeof( double( 0.0 ), [ 1, Inf ], [ 0, 1 ] ), coder.typeof( double( 0.0 ), [ Inf,1 ], [ 1, 0 ] )}

As specified, the first argument vecA must be a row vector, vecB a column vector, respectively. The generated C Function has the following Signature:

void computeDotProduct(const emlrtStack *sp,
                       const emxArray_real_T *vecA,
                       const emxArray_real_T *vecB,
                       real_T *Product,
                       char_T errors_data[],
                       int32_T errors_size[2] )

where the two input structs vecA and vecB may be of arbitrary Size, because emxArray_real_T is not specifically a row or column vector. The body of the function looks like:

    ...
    static const char_T cv0[2] = { 'O', 'K' };
    ...

    /*     %%Check Shape of Input Data */
    /*     %%Compute the dot Product. */
    st.site = &emlrtRSI;
    b_st.site = &c_emlrtRSI;
    if (!(vecA->size[1] == vecB->size[0])) {
      if ((vecA->size[1] == 1) || (vecB->size[0] == 1)) {
        emlrtErrorWithMessageIdR2012b(&b_st, &b_emlrtRTEI,
          "Coder:toolbox:mtimes_noDynamicScalarExpansion", 0);
      } else {
        emlrtErrorWithMessageIdR2012b(&b_st, &c_emlrtRTEI, "Coder:MATLAB:innerdim",
          0);
      }
    }

    if ((vecA->size[1] == 1) || (vecB->size[0] == 1)) {
      ...
      }
    } else {
      ...
    }
    ...  
    for (i0 = 0; i0 < 2; i0++) {
      errors_data[i0] = cv0[i0];
    }

There is indeed a check for Dimensions, but the Shape-Check I put in was dropped by the Coder. So, my question is: Is it possible to prevent the coder from removing the input check but still keeping the coder.typeof statements as above?

Answer 1

[Tony Mohan Varghese]

Unbounded variable size inputs are defined as structures in the generated C code and it will be named as emxArray_real_T.

As you can see, the data and size are defined separately in the structure emxArray_real_T.

The shape-check that you specified is done by the following line in the generated code:

if ((vecA->size[1] == 1) || (vecB->size[0] == 1)) {
    Product = 0.0;
    for (i0 = 0; i0 < vecA->size[1]; i0++) {
      Product += vecA->data[vecA->size[0] * i0] * vecB->data[i0];
    }
  } else {
    Product = 0.0;
    for (i0 = 0; i0 < vecA->size[1]; i0++) {
      Product += vecA->data[vecA->size[0] * i0] * vecB->data[i0];
    }
  }

vecA->size[1] corresponds to the row of the vector A and similarily for vector B.