On Thu, Jun 05, 2014 at 10:59:37AM +0200, Christian Hesse wrote:
> David Sterba <dsterba@xxxxxxx> on Wed, 2014/06/04 18:44:
> > On Wed, Jun 04, 2014 at 09:19:26AM +0200, Christian Hesse wrote:
> > > > It seems to be related to default gcc flags from distribution?
> > >
> > > Probably. I did compile with optimization, so adding -O2 may do the trick:
> > >
> > > make CFLAGS="${CFLAGS} -O2" all
> >
> > The warning appears with -O2, so the question is if gcc is not able to
> > reason about the values (ie. a false positive) or if there's a bug that
> > I don't see.
>
> I do not see a bug either. So probably this is a false positive...
Yeah I think so.
> Looks like the warning is triggered as soon as -ftree-vrp is added to CFLAGS.
Good catch
$ make CFLAGS="${CFLAGS} -O2 -Wall -fno-tree-vrp" all
does not print the warning (gcc 4.8.1).
> From gcc man page:
>
> -ftree-vrp
> Perform Value Range Propagation on trees. This is similar to the
> constant propagation pass, but instead of values, ranges of values are
> propagated. This allows the optimizers to remove unnecessary range
> checks like array bound checks and null pointer checks. This is
> enabled by default at -O2 and higher. Null pointer check elimination
> is only done if -fdelete-null-pointer-checks is enabled.
>
> Is it possibly that gcc optimized away any checks?
I'm not sure, that way lies gcc optimization maze.
Looking at the assembly from tree-vrp and no-tree-vrp it's probably the
case, the vrp-optimized version is restructured (and shorter) and looks
almost the same. Some of the bounds checks are missing (at least cannot
be easily paired). Better ask gcc people.
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
