https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93745
--- Comment #14 from rguenther at suse dot de <rguenther at suse dot de> --- On Fri, 28 Feb 2020, law at redhat dot com wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93745 > > --- Comment #12 from Jeffrey A. Law <law at redhat dot com> --- > It would seem like C ought to be able to set the flag across the board. But > Richi would know best if this is going to run afoul of of the alias oracle > implementation & underlying gimple semantics. > > I vaguely recall a discussion on the lists WRT how to represent > placement-new. > I didn't follow it at the time. Without reviewing the thread it does seem > helpful if we could see the placement-new in the IL. Note for C++ placement new isn't required. At least older standards simply talk about "reuse" of storage when POD types are involved. Note a representation of "placement new" is awkward - it has to be present in the virtual use-def chain since otherwise it's not a barrier for alias purposes. In the end it will hinder more optimizations than it will enable and you'll need to add it before all POD stores because of the above mentioned issues.
