https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58681
--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> --- So the reason why this was not fixed by r12-5301-g045206450386 (which is unlike PR 46555) is because here we have VOP that is the same edges so the forwarder block was removed after cddce3 and not kept around. In PR 46555, the VOP for the forwarder block got the vop phi in it and cleanupcfg did know how to remove forwarder blocks with phis just yet (well until r16-5258-g1d8e2d51e5c5cb). Anyways my patch set ending at https://gcc.gnu.org/pipermail/gcc-patches/2025-December/703029.html fixes this correctly by not doing a cleanup cfg between the creating of the forwarder blocks and expand.
