regression on Adobe-C++/loop_unroll

Evan Cheng evan.cheng at apple.com
Tue Feb 11 17:04:42 PST 2014


Thanks for tracking this down!

Evan

On Feb 10, 2014, at 2:48 PM, Chandler Carruth <chandlerc at gmail.com> wrote:

> To close the loop here for those not CC-ed in the PR, I fixed this in r201104. It was "just" a missing preservation of AA in LCSSA that caused globalmodref-aa to be invalidated in the LTO pipeline. =[[[ 4 days to debug, but wow the fix was easy.
> 
> -Chandler
> 
> 
> On Sun, Feb 9, 2014 at 10:28 AM, Chandler Carruth <chandlerc at gmail.com> wrote:
> Evan, see the PR. Essentially, making LICM *better* causes a very strange degradation in DSE. The thing that is found to be loop invariant *is* loop invariant, and ostensibly should be moved into the exit block. I have not yet uncovered the root cause of why this change would cause us to fail to delete dead stores in some other function...
> 
> 
> On Sun, Feb 9, 2014 at 10:19 AM, Evan Cheng <evan.cheng at apple.com> wrote:
> Do you have more detailed information on how this change is causing performance regression?
> 
> Evan
> 
> On Feb 6, 2014, at 11:19 AM, Gerolf Hoflehner <ghoflehner at apple.com> wrote:
> 
>> Hi Chandler
>> 
>> your commit seems to cause a major regression (>20%)  on loop_unroll and other benchmarks as well. This is ( at least)  on x86 under -O3 -flto and seems to be due to the loss of LICM. Please take a look.
>> 
>> Cheers
>> Gerolf
>> 
>> ------------------------------------------------------------------------
>> r200067 | chandlerc | 2014-01-24 20:07:24 -0800 (Fri, 24 Jan 2014) | 44 lines
>> 
>> [LPM] Make LCSSA a utility with a FunctionPass that applies it to all
>> the loops in a function, and teach LICM to work in the presance of
>> LCSSA.
>> 
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> 
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> 
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140211/468ea3db/attachment.html>


More information about the llvm-commits mailing list