[PATCH] D29826: [SLP] General improvements of SLP vectorization process.

[Alexey Bataev via Phabricator via llvm-commits]

ABataev added a comment.

In https://reviews.llvm.org/D29826#676819, @mkuper wrote:

> Thanks, Alexey.
>  Could you explain why this is better than just changing the main iteration to go backwards? Wouldn't it solve the same problem, without adding the distinction between different kinds of root nodes?
>
> (Adding Ayal and Gil to this one too.)


I tried the reversed order of the instruction analysis. I don't remember exactly why, but I decided not to continue to work on this. IIRC, there were troubles with the handling of PHIs.


https://reviews.llvm.org/D29826