[PATCH] D32305: [CodeExtractor] Remove a bunch of unneeded constructors

Xinliang David Li via llvm-commits llvm-commits at lists.llvm.org
Thu Apr 20 12:21:47 PDT 2017


On Thu, Apr 20, 2017 at 12:00 PM, Davide Italiano via Phabricator <
reviews at reviews.llvm.org> wrote:

> davide added a comment.
>
> In https://reviews.llvm.org/D32305#732439, @davidxl wrote:
>
> > lgtm.
> >
> > by the way,  what is your PM setup for partial inlining?
>
>
> Unfortunately, once I start playing with it, I hit this crash and I
> started debugging, so I didn't had yet time to put at different places in
> the pipeline.
> You have already tried some configurations? Aside, I noticed partial
> inlining isn't using BB frequency informations (which I found odd, but
> probably  it's because nobody cared).
>

It should use frequency information -- one step a time :)


> I imagine at least a simple heuristic where you split functions comparing
> as GCC does, IIRC (or at least, used to do at some point).
>

Partial inlining will always lead to reduced call overhead with very small
impact on size -- so even without using BFI, it should just do fine. My
pending patch to improve partial inining will limit the size impact at
least.

David


> There was some initial work on this, but, alas, it's unfinished
> https://reviews.llvm.org/D22744
>
>
> Repository:
>   rL LLVM
>
> https://reviews.llvm.org/D32305
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170420/5ed8e6b8/attachment.html>


More information about the llvm-commits mailing list