[PATCH] D23464: [PR28367][Reassociation] Avoid iterator invalidation when negating value.

[Daniel Berlin via llvm-commits]

FWIW: if we want reassociate solely for the purpose of "better CSE", and in
particular, by either GVN or EarlyCSE, you may be better off moving it into
those passes.

Reassociate essentially flies blind. It tries to canonicalize to some
 thing, hoping that this is the thing that exposes the most redundancies.

EarlyCSE/GVN know what is already redundant and what is not, and can
evaluate whether reassociating actually enables an elimination or not.

They can also incorporate forward propagation, etc.




On Wed, Aug 17, 2016 at 10:29 AM, Chad Rosier via llvm-commits <
llvm-commits at lists.llvm.org> wrote:

> mcrosier abandoned this revision.
> mcrosier added a comment.
>
> Neither Eli or I are particularly thrilled with this solution and Eli
> convinced me reverting r258830 would fix the regression.  Therefore, I'm
> abandoning this change and have reverted r258830 in r278938.
>
>
> https://reviews.llvm.org/D23464
>
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160819/74cca4c7/attachment.html>