[PATCH] fix aliasing for chars in global structures - take II

[Dan Nicolaescu]

Fergus Henderson <fjh@cs.mu.oz.au> writes:

  > On 22-Mar-2002, Dan Nicolaescu <dann@godzilla.ICS.UCI.EDU> wrote:
  > > 
  > > Currently references to a "char" member of a structure are put in
  > > alias set zero, ie like they are a char*. 
  > 
  > Right.  That matches what the standard specifies.
  > 
  > Consider the following program:
  > 
  > 	struct first  {  char f1; int x; };
  > 	struct second {  char f2; double y; };
  > 
  > 	void 
  > 	g1 (struct first *ps1, struct second *ps2)
  > 	{
  > 	  ps1->f1++;
  > 	  ps2->f2++;
  > 	  ps1->f1++;
  > 	  ps2->f2++;
  > 	}
  > 
  > 	int 
  > 	main (void)
  > 	{
  > 	  void *p = calloc(sizeof(struct first) + sizeof(struct second), 1);
  > 	  g1 (p, p);
  > 	  printf("%d\n", ((struct first *)p)->f1);
  > 	  return 0;
  > 	}
  > 
  > Isn't that program strictly conforming C, which should output `4'?

It does. 
Here is the generated SPARC assembly

g1:
        !#PROLOGUE# 0
        !#PROLOGUE# 1
        ldub    [%o0], %o3
        add     %o3, 1, %o3
        stb     %o3, [%o0]
        ldub    [%o1], %o2
        add     %o2, 1, %o2
        stb     %o2, [%o1]
        ldub    [%o0], %o3
        add     %o3, 1, %o3
        stb     %o3, [%o0]
        ldub    [%o1], %o0
        add     %o0, 1, %o0
        retl
        stb     %o0, [%o1]


  > And won't your patch mis-optimize it to output something
  > different?

I don't think so. The the references to ps1->f1 and ps2->f2 will be in
the same alias set. Look at the "g3" example in my previous mail.