[PATCH] Inliner Enhancement

[Yin Ma]

Hi Jiangning

 

We have seen good performance improvement on SPEC from our internal build.

I will take a look if there is any porting issue there. 

 

For the compile time increase, it is expected in some cases because implementing

Deferred inlining in any way will only increase threshold compared with baseline 

inliner.

 

To remedy this issue, the cost used in deferred inlining should be tuned to 

Be consistent with inline threshold in my opinion.

 

Yin 

 

From: Jiangning Liu [mailto:liujiangning1 at gmail.com] 
Sent: Friday, March 20, 2015 12:29 AM
To: reviews+d8408+public+fc653c157840723f
Cc: Chandler Carruth; Ana Pazos; Hal Finkel; Yin Ma; Renato Golin; Amara Emerson; llvm-commits at cs.uiuc.edu for LLVM
Subject: Re: [PATCH] Inliner Enhancement

 

Hi Yin,

 

I checked your patch,

 

(1) Compile-time has huge regression. For llvm bootstrap on x86, I can see +100% compile-time increase, which is unacceptable.

(2) For performance, I can only see +0.05% performance gain for SPEC2000, so I think it's a kind of noise.

 

Thanks,

-Jiangning

 

2015-03-20 7:31 GMT+08:00 Yin Ma <yinma at codeaurora.org <mailto:yinma at codeaurora.org> >:

Hi Jiangning,

I have posted my implementation of deferred inlining here.
http://reviews.llvm.org/D8475

Please review and if possible, give a try with your other tuning.

Yin


REPOSITORY
  rL LLVM

http://reviews.llvm.org/D8408

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/



 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150320/f1a01096/attachment.html>