[PATCH] D71733: [NFC][InlineCost] Factor cost modeling out of CallAnalyzer traversal.

Mircea Trofin via llvm-commits llvm-commits at lists.llvm.org
Tue Jan 7 11:15:17 PST 2020


The visit methods now delegate benefit computation the derived class. There
is no cost tracking in the CallAnalyzer anymore (unless I'm missing
something)

This would allow for an alternative benefit computation where combining the
various features of the code discovered through traversal is done
differently.

On Tue, Jan 7, 2020 at 11:11 AM Easwaran Raman <eraman at google.com> wrote:

> Sorry for the delay. The stated intention is to keep the legality part in
> CallAnalyzer and move the benefit part to InlineCostCallAnalyzer.  But if
> you consider any of the visit methods in CallAnalyzer, they actually
> compute the benefit - tells whether the instruction can be simplified or
> removed by inlining the callee. So what is the division of work
> between CallAnalyzer and InlineCostCallAnalyzer?
>
> On Mon, Jan 6, 2020 at 9:07 AM David Li via Phabricator <
> reviews at reviews.llvm.org> wrote:
>
>> davidxl added a comment.
>>
>> this looks good to me. Easwaran, do you have any comments on the
>> refactoring?
>>
>>
>> Repository:
>>   rG LLVM Github Monorepo
>>
>> CHANGES SINCE LAST ACTION
>>   https://reviews.llvm.org/D71733/new/
>>
>> https://reviews.llvm.org/D71733
>>
>>
>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200107/78a7371e/attachment.html>


More information about the llvm-commits mailing list