[PATCH] D20195: [PGO] Add flags to control IRPGO warnings.

Xinliang David Li via llvm-commits llvm-commits at lists.llvm.org
Wed May 11 21:52:00 PDT 2016


So you do turn on pre-inlining in the experiment? Yes if preinlining is on,
there might be mismatching issues for comdat functions. Rong has plan to
address that -- i.e, partial privatizing comdat functions according to
their CFG hash. In other words, for all copies of a comdat func that have
the same inlinings (therefore same cfg hash), they will share same linkage
name (modified from the original name).

Rong can help you check in the patch once he reviewed it.

David


On Wed, May 11, 2016 at 9:45 PM, Jake VanAdrighem <jvanadrighem at gmail.com>
wrote:

> Great, if possible Sean or David could you commit this for me? I don't
> have commit access.
>
> As for your question. Sean and I have actually experienced warnings
> resulting from a difference in inlining choices in different TU's but we
> will definitely be bringing this up very soon as a separate discussion.
> These flags were mostly to deal with `instrprof_error::unknown_function`
> which for our use case tends to be pretty spammy but not very helpful.
>
> On Wed, May 11, 2016 at 9:37 PM, David Li <davidxl at google.com> wrote:
>
>> davidxl accepted this revision.
>> davidxl added a comment.
>> This revision is now accepted and ready to land.
>>
>> lgtm.
>>
>> Question : do you get warnings when there is no source change? This
>> indicates some potential bugs in compiler. If you see them, can you help
>> file bugs?
>>
>>
>> Repository:
>>   rL LLVM
>>
>> http://reviews.llvm.org/D20195
>>
>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160511/216365ad/attachment.html>


More information about the llvm-commits mailing list