https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118205
Lewis Hyatt <lhyatt at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Summary|[15 Regression] ICE: in |[15 Regression] ICE: in
|as_a, at is-a.h:255 with -O |as_a, at is-a.h:255 with -O
|-floop-parallelize-all |-floop-parallelize-all
|-ftree-parallelize-loops=2 |-ftree-parallelize-loops=2
| |since r15-6001
Last reconfirmed| |2024-12-27
URL| |https://gcc.gnu.org/piperma
| |il/gcc-patches/2024-Decembe
| |r/672271.html
Ever confirmed|0 |1
--- Comment #4 from Lewis Hyatt <lhyatt at gcc dot gnu.org> ---
I think the issue is related to the undefined behavior with the zero-sized
array in this testcase. It leads to trying to call reduction_phi() to look up a
stmt that is not actually a phi. Prior to r15-6001, that was OK and just
returned NULL; in r15-6001 I did not consider that case that the PHI arg to
reduction_phi() may not be a gphi. One-line change to check for this case
submitted here:
https://gcc.gnu.org/pipermail/gcc-patches/2024-December/672271.html.
