------- Additional Comments From law at redhat dot com 2005-02-01 14:49 -------
Subject: Re: [4.0 Regression] A side effect
is missed in 0 % a++.
On Mon, 2005-01-31 at 22:36 -0800, Richard Henderson wrote:
> On Mon, Jan 31, 2005 at 08:45:34PM -0700, Jeffrey A Law wrote:
> > + /* X % 0, return X % 0 unchanged so that we can get the
> > + proper warnings and errors. */
> > if (integer_zerop (arg1))
> > return t;
> >
> > + /* 0 % X is always zero, but be sure to preserve any side
> > + effects in X. Place this after checking for X == 0. */
> > + if (integer_zerop (arg0))
> > + return omit_one_operand (type, integer_zero_node, arg1);
>
> Not ok yet. You have to *know* that arg1 is not zero. Otherwise
> you're still potentially removing a division-by-zero.
Would would mean that we really can't do anything with DIV/MOD when
the numerator is a constant and the denominator is an unknown.
What I think we really want is to know whether or not the language
requires trapping on a DIV/MOD by zero. If so, then disable this class
of optimizations, otherwise we have a non-conformant program.
jeff
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19723
