[PATCH] Modularizing LICM

Nema, Ashutosh Ashutosh.Nema at amd.com
Mon Feb 16 21:04:11 PST 2015


Thanks Hal, your comments looks fine will incorporate.

Regards,
Ashutosh

-----Original Message-----
From: hfinkel at anl.gov [mailto:hfinkel at anl.gov] 
Sent: Monday, February 16, 2015 7:59 PM
To: Nema, Ashutosh; listmail at philipreames.com; bruno.cardoso at gmail.com; hfinkel at anl.gov
Cc: llvm-commits at cs.uiuc.edu
Subject: Re: [PATCH] Modularizing LICM

In http://reviews.llvm.org/D7291#124183, @ashutosh.nema wrote:

> Thanks Hal, Philip & Bruno for review.
>
> Incorporated review comments.
>
> Changes in this patch:
>
> 1. Introduced 'LICMSafetyInfo', earlier 'MayThrow' & 'HeaderMayThrow' was exposed outside. But now 'SinkRegion', 'HoistRegion' and 'PromoteAliasSet' will compute safety information.


But this means that each of these functions now needs to re-scan the loop to compute the safety information. I was hoping to avoid this extra expense. Please expose LICMSafetyInfo, and the utility function to compute it.


REPOSITORY
  rL LLVM

================
Comment at: include/llvm/Transforms/Utils/LoopUtils.h:76
@@ +75,3 @@
+/// DataLayout, TargetLibraryInfo, Loop and AliasSetTracker as argument 
+/// and returns changed status.
+bool SinkRegion(DomTreeNode *, AliasAnalysis *, LoopInfo *, 
+DominatorTree *,
----------------
Please describe here, and in the function descriptions below, what is expected to be in the AliasSetTracker.

http://reviews.llvm.org/D7291

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






More information about the llvm-commits mailing list