Re: integer_onep vs. signed 1-bit bitfields

[Marc Glisse <marc dot glisse at inria dot fr>]

On Sat, 2 Nov 2013, Jakub Jelinek wrote:

> On Sat, Nov 02, 2013 at 05:38:53PM +0000, Richard Sandiford wrote:
> > OK, thanks.  I should have realised this earlier, but we have:
> >
> > /* Return 1 if EXPR is the integer constant one or the corresponding
> >    complex constant.  */
> >
> > int
> > integer_onep (const_tree expr)
> > ...
> > /* Return 1 if EXPR is the integer constant minus one.  */
> >
> > int
> > integer_minus_onep (const_tree expr)
> >
> > which makes them sound like a pair.  But integer_minus_onep returns
> > true for any all-ones INTEGER_CST (regardless of sign):
> >
> >   if (TREE_CODE (expr) == COMPLEX_CST)
> >     return (integer_all_onesp (TREE_REALPART (expr))
> > 	    && integer_zerop (TREE_IMAGPART (expr)));
> >   else
> >     return integer_all_onesp (expr);
> >
> > So a nonzero 1-bit unsigned bitfield is both integer_onep and
> > integer_minus_onep, but a 1-bit signed bitfield is only
> > integer_minus_onep.  Should integer_minus_onep be changed so
> > that it always returns false for unsigned types?
>
> Yeah, I was wondering about that too.  I think integer_minus_onep
> is a fairly new thing, I guess if you want to change it, somebody has to go
> through all users and see what they want.
> Returning false for unsigned types is natural thing to do though.

See the conversation around:
http://gcc.gnu.org/ml/gcc-patches/2013-05/msg00053.html
for the introduction of this function.

Richard already asked the same question then, and the patch renaming some 
integer_all_onesp to integer_minus_onep was forgotten. Personnally, I am 
fine with integer_minus_onep returning true on -1U. Without the renaming 
patch, there is a single use of integer_minus_onep (in fold-const.c), so 
it shouldn't be hard to check all users, and IIRC we want it to return 
true for -1U in this case (turning x*-1 into -x).

--
Marc Glisse