[PATCH] ccp for V_MUST_DEF operands

[Diego Novillo]

On Fri, 2004-07-16 at 10:26, Brian Booth wrote:
> On Thu, 2004-07-15 at 09:18, Diego Novillo wrote:
> > Could you add a test case?  Something like this should suffice:
> > 
> > --------------------------------------------------------------------
> > int G;
> > 
> > foo (int i)
> > {
> >    if (i > 0)
> >      G = 3;
> >    else
> >      G = 3;
> >    if (G != 3)
> >      link_error ();
> > }
> > 
> > main ()
> > {
> >    foo (0);
> >    return 0;
> > }
> > --------------------------------------------------------------------
> 
> I'm withdrawing this patch for the time being because it does not
> propagate G in the testcase above. The problem has to do with performing
> a meet operation on UNKNOWN_VAL. More specifically, the case CONSTANT M
> UNKNOWN_VAL = VARYING is causing G not to be propagated since the PHI
> node for G is simulated before one of the assignments to G in the
> if-else block. This is causing the PHI node to be marked as VARYING and
> ensures that it is not simulated again.
> 
What about initially setting phi_val to UNDEFINED if 
get_value (PHI_RESULT (phi)) returns UNKNOWN_VAL?

So, the first time we visit the PHI we get UNDEFINED M 3 = 3.  The
second time we'll get 3 M 3 = 3.

If the else branches didn't have an assignment to G, we would get
UNDEFINED M 3 = 3.  But the second round would give us 3 M UNKNOWN_VAL =
VARYING.


Diego.