This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [Lto-team] Fwd: [tuples] reduced testcase of the libjava build failure
On Fri, May 2, 2008 at 12:00 PM, Rafael Espindola <espindola@google.com> wrote:
> >
> > IMHO if such comparison may trap then we need to gimplify to
> >
> > <bb>
> > ...
> > bool tmp = a < b;
> >
> > <bb>
> > if (tmp)
> > ....
>
> That is interesting. IMHO we should change GIMPLE_COND to have only a
> boolean variable. My only argument for that is that it would make the
> IL more simple, since there would be only one place to look for
> comparisons (on GIMPLE_ASSIGN).
>
> This problems looks like a more convincing argument. What do you think about it?
It also would help SCCVN because all comparisons will have an SSA_NAME
associated. I can imagine (optimization) problems with 1) expand,
2) forwprop (it does COND_EXPR operand combining), 3) vrp and DOM
for the same reason, 4) ifcombine. And of course with all the others
I forgot ;)
IMHO a worthwhile simplification in general, but not necessary for the
immediate problem.
Richard.