[Bug rtl-optimization/116053] [14/15 Regression] during RTL pass: cprop_hardreg ICE: verify_flow_info failed: missing REG_EH_REGION note at the end of bb 2 with -O1 -finstrument-functions -fno-forward-propagate -fno-delete-dead-exceptions -fnon-call-exceptions

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116053

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |needs-bisection

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note I think a bisect here will just show where this latent bug shows up (GCC
13 register allocator gets it diffenent and uses the sp register as the
address).

I think after the call to df_analyze (that is in the loop) in
pass_cprop_hardreg::execute we need to call purge_all_dead_edges.

Or in df_analyze needs to call purge_dead_edge on the bb if it deletes an
instruction with a REG_EH_REGION note on it.

Note I do think we have a missed optimization at the gimple level. if the first
argument to memset/memcpy is an ADDR_EXPR, we change the assignment.

That is:
  x = *(__int128 *) __builtin_memset (&x, 0, 10);
Should be done as:
__builtin_memset (&x, 0, 10)
  x = *(__int128 *) &x;

And then this will be optimized away earlier.
Note this is a very much an artificial testcase and I highly doubt this will
show up normally because -fnon-call-exceptions is not used that much and even
less if fno-dead-exceptions.