Bad simplification for n*n when n is Integer

[Adrian Pop]

Consider this:

function eig
  input  Real a[n*n*2];
  input  Integer n;
  output Real vec[2*n];
  output Real mat[n*n*2];
  external "C" eigCompute(a,vec,mat,n) annotation(Library="eigenSolver", Include="#include \"eig.h\"");
end eig;

model testEigenSolver
 parameter Integer n = 3;
 Real a[n*n*2] = {1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10};
 Real vec[2*n];
 Real mat[n*n*2];
equation
 (vec,mat)=eig(a,n);
end testEigenSolver;

The external function doesn't matter, just put something in those vectors.

The old front-end translates n*n to n^2.0

function eig
  input Real[2 * n ^ 2.0] a; // n*n -> n^2.0 is just an abomination!
  input Integer n;
  output Real[2 * n] vec;
  output Real[2 * n ^ 2.0] mat;

  external "C" eigCompute(a, vec, mat, n);
end eig;

which makes it so mat variable will *ALWAYS* be zero after return from eig because the mat variable array in function eig will have mat->dim_size[0] == 0 as in generated omc_eig we have:

  modelica_real tmp1;
  alloc_real_array(&(_vec), 1, (((modelica_integer) 2)) * (_n)); // _vec has no default value.
  tmp1 = _n;
  alloc_real_array(&(_mat), 1, (((modelica_integer) 2)) * ((tmp1 * tmp1))); // _mat has no default value.

The NF does much better, and it works fine:

function eig
  input Real[n * n * 2] a;
  input Integer n;
  output Real[2 * n] vec;
  output Real[n * n * 2] mat;

  external "C" eigCompute(a, vec, mat, n);
end eig;

[Martin Sjölund]

​https://github.com/OpenModelica/OpenModelica/pull/693