[PATCH] D32401: [Devirtualization] insert placement new barrier with -O0

[Piotr Padlewski via Phabricator via cfe-commits]

Prazek added a comment.

In https://reviews.llvm.org/D32401#734921, @rjmccall wrote:

> I continue to be really uncomfortable with the entire design of this optimization, which appears to miscompile code by default, but as long as nobody's suggesting that we actually turn it on by default, I guess it can be your little research-compiler playground.  It might be better to downgrade it to a -cc1 option, though.
>
> This specific change is fine by me.


Can you tell me a little more about what part of the design you dislike? Is it about missing optimizations by introducing the barriers, cost of inserting the barriers or the fact that we have to be cautious to not break anything?


https://reviews.llvm.org/D32401