Re: TREE_ADDRESSABLE problem

[Jan Hubicka <jh at suse dot cz>]

> On Thu, Nov 20, 2003 at 12:02:56AM +0100, Jan Hubicka wrote:
> > Hi,
> > the following testcase taken from testustie:
> > 
> > typedef struct {
> >   int a, b, c, d, e, f;
> > } A;
> > 
> > void foo (A *v, int w1, int x, int *y, int *z)
> > {
> > }
> > 
> > void
> > bar (A *v, int x, int y, int w2, int h)
> > {
> >   if (v->a != x || v->b != y) {
> >     int oldw = w2;
> >     int oldh = h;
> >     int e = v->e;
> >     int f = v->f;
> >     int dx, dy;
> >     foo(v, 0, 0, &w2, &h);
> >     dx = (oldw - w2) * (double) e/2.0;
> >     dy = (oldh - h) * (double) f/2.0;
> >     x += dx;
> >     y += dy;
> >     v->a = x;
> >     v->b = y;
> >     v->c = w2;
> >     v->d = h;
> >   }
> > }
> > 
> > int main ()
> > {
> >   A w = { 100, 110, 20, 30, -1, -1 };
> >   bar (&w,400,420,50,70);
> >   if (w.d != 70)
> >     abort();
> >   exit(0);
> > }
> > 
> > Does not pass TREE_ADDRESSABLE check with my checking code with -O3.
> > The problem is that we first compile function bar, inline foo and
> > elliminate reference to w2, clear TREE_ADDRESSABLE in must_alias pass.
> > Then we inline bar body into main but TREE_ADDRESSABLE is not set even
> > though it is used.
> 
> Huh?  (1) w2 in bar and w2 in bar-inlined-in-main are two different
No, it is not.  We first optimize bar and then produce
bar-inlined-in-main.
> decls.  (2) w2 isn't used after optimization, ever, not even in main.
The checking fails before optimization is done, when the address is
still present.
Somehow we manage to save copy of the original unoptimized bar for
in-lining, but we manage to not save the variables.  Very odd.  I didn't
know we are donig something like that...
I am attaching relevant dumps.

Honza
> 
> 
> r~

Attachment: a.c.t10.cfg
Description: Text document

Attachment: a.c.t30.optimized
Description: Text document