simplification of SIN

[Daniel Franke]

Hi all.

Could somebody knowledgeable of simplifying be so kind to comment on why SIN 
is not simplified if called with a constant array argument?

Or phrased differently, why simplification is attempted on the whole array (x-
>expr_type == EXPR_ARRAY in gfc_simplify_sin), instead of the single, 
constant, elements?

Is this necessary for some reason or "just" a missed optimization?

Thanks

	Daniel


==============================================
$> cat scalar.f90
  REAL :: a
  a = sin(0.0)
end

$> gfortran-svn -fdump-tree-original scalar.f90
$> cat scalar.f90.003t.original
MAIN__ ()
{
  real(kind=4) a;
  static integer(kind=4) options.0[8] = {68, 255, 0, 0, 0, 1, 0, 1};

  _gfortran_set_options (8, (void *) &options.0);
  a = 0.0;
}

==============================================
$> cat array.f90
  REAL :: a(1)
  a = sin([0.0])
end

$> gfortran-svn -fdump-tree-original array.f90
$> cat array.f90.003t.original
MAIN__ ()
{
  real(kind=4) a[1];
  static integer(kind=4) options.0[8] = {68, 255, 0, 0, 0, 1, 0, 1};

  _gfortran_set_options (8, (void *) &options.0);
  {
    static real(kind=4) A.1[1] = {0.0};

    {
      integer(kind=4) S.2;

      S.2 = 0;
      while (1)
        {
          if (S.2 > 0) goto L.1;
          a[S.2] = __builtin_sinf (A.1[S.2]);
          S.2 = S.2 + 1;
        }
      L.1:;
    }
  }
}
==============================================
If used as initialization expression, gfc_simplify_sin() is called 2 + SIZE(a) 
times, then being simplified as expected:

$> cat array_2.f90
  REAL :: a(1) = sin([0.0])
  REAL b
  b = a(1)
end

$> gfortran-svn -fdump-tree-original array_2.f90
$> cat array_2.f90.003t.original
MAIN__ ()
{
  static real(kind=4) a[1] = {0.0};
  real(kind=4) b;
  static integer(kind=4) options.0[8] = {68, 255, 0, 0, 0, 1, 0, 1};

  _gfortran_set_options (8, (void *) &options.0);
  b = a[0];
}