On Thu, Jan 31, 2019 at 4:02 PM Jakub Jelinek <ja...@redhat.com> wrote:
>
> On Thu, Jan 31, 2019 at 03:56:12PM +0100, Richard Biener wrote:
> > Don't you alreday have
> >
> > @@ -4200,10 +4202,34 @@ pass_sprintf_length::execute (function *fun)
> > init_target_to_host_charmap ();
> >
> > calculate_dominance_info (CDI_DOMINATORS);
> > + bool optimizing_late = optimize > 0 && fold_return_value;
> > + if (optimizing_late)
> > + {
> > + /* ?? We should avoid changing the CFG as much as possible.
> > ...
> > + loop_optimizer_init (LOOPS_HAVE_PREHEADERS);
> > + scev_initialize ();
> > + }
> >
> > so loops are only initialized if fold_return_value is true? Ah - but that's
> > the pass parameter from params.def rather than the flag to enable
> > the folding... So indeed, change it to include && flag_printf_return_value
>
> fold_return_value is not the same thing as flag_printf_return_value,
> the former is just a bool whether it is the -O0 or -O1+ version of the pass.
> So, optimizing_late doesn't make much sense, one can use optimize > 0
> directly instead.
>
> If changing the above to && flag_printf_return_value then people will
> complain that they get the false positive warning with -Wall
> -fno-printf-return-value.
Sure - too bad then. Another option would be to make SCEV / niter analysis
"safe" to be called with LOOPS_AVOID_CFG_MANIPULATIONs (multiple
latches, multiple entries into headers, etc.).
Richard.
> Jakub
