Re: Alias analysis

[Richard Henderson <rth at cygnus dot com>]

On Sun, Jan 17, 1999 at 10:47:13PM -0700, Jeffrey A Law wrote:
>       if (GET_CODE (y) == AND
>           && (GET_CODE (XEXP (y, 1)) != CONST_INT
>               || GET_MODE_UNIT_SIZE (x_mode) < -INTVAL (XEXP (y, 1))))
>         return 1;
>     }
> 
> x_base & y_base point to different SYMBOL_REFs, but none of the AND checks
> apply and we fall through.
> 
> I think the fix is simply a "return 0" before the close brace.

Ho hum.  That would be my bug.

Interesting that I didn't see a Spec95 regression because of it.

I wonder if those extra clauses do so much good to overcome the
obvious symbol deficiency, or if the problem is hidden by the
fact that (Fortran) COMMON symbol references all resolve to the
same base symbol.

Thanks for finding it.


r~