[Bug debug/64511] [5 Regression] ICE at -O3 with -g enabled on x86_64-linux-gnu

[rguenther at suse dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64511

--- Comment #6 from rguenther at suse dot de <rguenther at suse dot de> ---
On Tue, 13 Jan 2015, jakub at gcc dot gnu.org wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64511
> 
> Jakub Jelinek <jakub at gcc dot gnu.org> changed:
> 
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>                  CC|                            |jakub at gcc dot gnu.org
> 
> --- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
> x % x for x == 0 is undefined behavior, so perhaps with the exception of
> sanitization we can just assume it is not 0 and thus fold x % x to 0.
> Or does e.g. Ada/Java require something different?

See the inconsistent handing of % vs. / in fold (we fold 0 % x to 0
but preserve literal 0 % 0 - we don't fold 0 / x to 0).

In my view we should just go ahead and fold, though with 
-fnon-call-exceptions one may do

 try {
  x % x
  abort ();
 } catch (...) {
 }

and throw from inside a trap handler.  Which means that one can
catch undefined behavior.  IMHO -fnon-call-exceptions and sub-flags
like -ftrapv tell GCC to treat the specific undefined behavior as
defined - namely trapping.  We don't have a special flag for
divide-by-zero-traps though (it's not exactly "overflow").

So maybe with making the folding more consistent we should simply
add such flag, -fdivide-by-zero-traps which also makes modulo-by-zero
trap?