[PATCH] [LPM] A targeted but somewhat horrible fix to the legacy pass manager's querying of the pass registry.

Owen Anderson resistor at mac.com
Tue Jan 27 21:33:19 PST 2015


After some more investigation, I think the initializeAllAnalysisInfo() track is a red herring.  It was actually hot in the prior profile as well, just buried behind locking-related things.  I think this change should go ahead.

—Owen

> On Jan 27, 2015, at 5:26 PM, Chandler Carruth <chandlerc at gmail.com> wrote:
> 
> To summarize the discussion on IRC:
> 
> This hot code path doesn't really seem to be due to my patch. If it is, I can't explain why. I definitely can't explain why without some steps to reproduce.
> 
> This patch essentially erases all the performance hits I can find any way to measure of the pthread locks, so I'll wait for more information from you or the others that have measured more significant issues here. Let me know what you'd like me to do next.
> 
> On Tue, Jan 27, 2015 at 4:56 PM, Owen Anderson <resistor at mac.com <mailto:resistor at mac.com>> wrote:
> Hi Chandler,
> 
> In my measurements, this captures about half of the regression, i.e. the regression goes from ~20% to ~10%.  While the reader/writer locks are gone from the profile, several other PassManager related pieces of code have jumped up to take their place:
> 
> —Owen
> 
> - F362108: ThreadProfile2.png <http://reviews.llvm.org/F362108 <http://reviews.llvm.org/F362108>>
> 
> 
> http://reviews.llvm.org/D7213 <http://reviews.llvm.org/D7213>
> 
> EMAIL PREFERENCES
>   http://reviews.llvm.org/settings/panel/emailpreferences/ <http://reviews.llvm.org/settings/panel/emailpreferences/>
> 
> 
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu <mailto:llvm-commits at cs.uiuc.edu>
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits <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/20150127/df0d10da/attachment.html>


More information about the llvm-commits mailing list