aggregate alias anomaly

[Jeff Sturm]

On Wed, 18 Feb 2004, Dan Nicolaescu wrote:
> Jeff Sturm <jsturm@one-point.com> writes:
>
>   > Given:
> >
>   > struct X {int x;};
>   > struct Y {int y;};
> >
>   > int f(struct X *x, struct Y *y) {
>   >         x->x = 0;
>   >         y->y = 1;
>   >         return x->x;
>   > }
> >
>   > gcc -O2 -fomit-frame-pointer yields:
> >
>   > f:
>   >         movl    4(%esp), %eax
>   >         movl    8(%esp), %edx
>   >         movl    $0, (%eax)
>   >         movl    $1, (%edx)
>   >         movl    (%eax), %eax
>   >         ret
> >
>   > for every released compiler I tried, plus mainline.  Note the useless load
>   > of x->x, given that x and y cannot possibly have conflicting alias sets.
>
> I am not a language lawyer (nor have any desire to be one), but that
> is not necessarily correct for C. See DR257 (and also the "What is an
> object DR").

I see.  I may have unintentionally found a tree-ssa bug, then.  I'll file
a PR today.

> On the other hand, even if you change struct X and Y so that they
> don't have a common initial sequence:
> struct X {char c; int x;};
> struct Y {int y; short d;};
>
> your function still won't be optimized....
>
> I posted a patch some time ago that solved a problem
> related to this (or the same?):
>
> http://gcc.gnu.org/ml/gcc-patches/2002-03/msg00576.html
>
> This version even has a java example:
>
> http://gcc.gnu.org/ml/gcc-patches/2002-03/msg01869.html
>
> If there's interest I could revive that patch.

Nice.  My example was based on a Java example.

Given that merging tree-ssa appears imminent now, and tree-ssa does a far
better job than mainline of alias analysis for Java, it doesn't seem to be
worth reviving your patch unless it were considered suitable for 3.4
(unlikely).

Thanks,

Jeff