https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92298
Luca Rocca <disquisitiones at gmail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|RESOLVED |UNCONFIRMED
Resolution|INVALID |---
--- Comment #2 from Luca Rocca <disquisitiones at gmail dot com> ---
Division by 0 is always undefined, regardless of the numerator.
So, X / X should not be simplified if we cannot exclude that X = 0.
Then if X = 0 we should expect an exception triggered at runtime,
as we have for example for 1 / 0.
Consider the relevant section of the file gcc-9.2.0/gcc/match.pd:
/* X / X is one. */
(simplify
(div @0 @0)
/* But not for 0 / 0 so that we can get the proper warnings and errors.
And not for _Fract types where we can't build 1. */
(if (!integer_zerop (@0) && !ALL_FRACT_MODE_P (TYPE_MODE (type)))
{ build_one_cst (type); }))
It seems that the intention is in fact to perform the simplification
except for the case 0 / 0, but for some reason this is not implemented
correctly.
Consider also for comparison the approach of gcc up to gcc-6.4.0,
reading this comment from the corresponding file gcc-6.4.0/gcc/match.pd:
/* Make sure to preserve divisions by zero. This is the reason why
we don't simplify x / x to 1 or 0 / x to 0. */
For the same code, GCC up to 6.4.0 does not perform the simplification and when
X = 0 we have an exception raised at runtime as expected
