[tree-ssa] Example in tree-dfa for computing alias sets doesn'tmatch actual results

[Steven Bosscher <s dot bosscher at student dot tudelft dot nl>]

Hi,

In the comments in tree-dfa.c, "Compute alias sets", the example
function,

            foo (int i)
            {
              int *p, a, b;
  
              if (i > 10)
                p = &a;
              else
                p = &b;
 
              *p = 3;
              a = b + 2;
              return *p;
            }

is claimed to be renamed to give:

            foo (int i)
            {
              int *p, a, b;
 
              if (i_2 > 10)
                p_4 = &a;
              else
                p_6 = &b;
              # p_1 = PHI <p_4(1), p_6(2)>;
 
              # a_7 = VDEF <a_3>;
              # b_8 = VDEF <b_5>;
              *p_1 = 3;
              a_9 = b_8 + 2;
 
              # VUSE <a_9>;
              # VUSE <b_8>;
              return *p_1;
            }


In reality, it looks more like a and b are being treated as if they are
global vars:

;; Function foo (foo)
            
foo (i)
{
  int b.1;
  int * p;
  int a;
  int b;
 
  if (i_2 > 10)
    {
      p_3 = &a
    }
  else
    {
      p_4 = &b
    };
  #   p_1 = PHI <p_3(1), p_4(2)>;
 
  #   a_6 = VDEF <a_5>;
  #   b_8 = VDEF <b_7>;
  *p_1 = 3;
 
  #   VUSE <b_8>;
  b.1_9 = b;
 
  #   a_10 = VDEF <a_6>;
  a = b.1_9 + 2;
 
  #   VUSE <a_10>;
  #   VUSE <b_8>;
  return *p_1;;
}

In fact, if a and b are declared as globals in the example, i.e.,

            int a,b;
            foo (int i)
            {
              int *p;
              (etc.)

the diff of the ssa1 dumps only is,

-   int a;
-   int b;


Is this an intended change in behavior?  If so, can someone with better
understanding of virtual ops update the comments in tree-dfa.c?

Gr.
Steven