[Bug tree-optimization/86605] Suboptimal code for pointer arithmetic with 'this'
rguenth at gcc dot gnu.org
gcc-bugzilla@gcc.gnu.org
Tue Jul 27 06:13:12 GMT 2021
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86605
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |RESOLVED
Resolution|--- |INVALID
--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #4)
> (In reply to Andrew Pinski from comment #3)
> > So on the trunk we get:
> > if (&test != 4B)
> > goto <bb 4>; [70.00%]
> > else
> > goto <bb 3>; [30.00%]
> >
> > <bb 3> [local count: 322122544]:
> >
> > <bb 4> [local count: 1073741824]:
> > # iftmp.0_3 = PHI <10(2), 20(3)>
> >
> > I know we fold &v != 0B but it seems like if the constant is within a few
> > bytes of 0, we should fold it similar. Shouldn't we?
>
> Interesting we fold "&test != 4B" on aarch64-linux-gnu but not on
> x86_64-linux-gnu.
>
> If we change 4 to 16, we don't fold on either.
Probably sth do to with alignment.
Seriously, I don't think we can choose some arbitrary constant and say
everything < N is equal to "NULL".
So I don't think optimizing this as wanted by the reporter is valid.
More information about the Gcc-bugs
mailing list