[llvm] r237593 - Exploit dereferenceable_or_null attribute in LICM pass

[Philip Reames]

Wasn't quite as simple as it first looked, but this should be fixed in 
237985.  I think this should be fine, but if you observe problems with 
either this change or the original, feel free to revert.  I'm going to 
be offline for the rest of the evening and won't be able to respond 
promptly.

Philip

On 05/21/2015 05:28 PM, Philip Reames wrote:
> Sanjoy is out on vacation, but I'm going to look into this.  I'll be 
> offline for about an hour, but should be able to start investigating 
> by around 7pm.  On the surface, it looks like it should be an easy 
> fix.  Hopefully, I'll have this resolved within a couple of hours.  
> Sorry for the breakage.
>
> Philip
>
> On 05/21/2015 04:38 PM, Nick Lewycky wrote:
>> On 18 May 2015 at 11:07, Sanjoy Das <sanjoy at playingwithpointers.com 
>> <mailto:sanjoy at playingwithpointers.com>> wrote:
>>
>>     Author: sanjoy
>>     Date: Mon May 18 13:07:00 2015
>>     New Revision: 237593
>>
>>     URL: http://llvm.org/viewvc/llvm-project?rev=237593&view=rev
>>     <https://urldefense.proofpoint.com/v2/url?u=http-3A__llvm.org_viewvc_llvm-2Dproject-3Frev-3D237593-26view-3Drev&d=AwMFaQ&c=8hUWFZcy2Z-Za5rBPlktOQ&r=mQ4LZ2PUj9hpadE3cDHZnIdEwhEBrbAstXeMaFoB9tg&m=EZRvoJai0V1ZEULFLl3UFBFw3GpoHtruFDOok84hYrI&s=4QqYr7GqSrZjVxTSY4zWHbjeCHYmt-2bkaDpOtpbsM8&e=>
>>     Log:
>>     Exploit dereferenceable_or_null attribute in LICM pass
>>
>>     Summary:
>>     Allow hoisting of loads from values marked with
>>     dereferenceable_or_null
>>     attribute. For values marked with the attribute perform
>>     context-sensitive analysis to determine whether it's
>>     known-non-null or
>>     not.
>>
>>
>> This commit caused PR23608, see the testcase in that PR.
>>
>>     -/// Only sink or hoist an instruction if it is not a trapping
>>     instruction
>>     +/// Only sink or hoist an instruction if it is not a trapping
>>     instruction,
>>     +/// or if the instruction is known not to trap when moved to the
>>     preheader.
>>      /// or if it is a trapping instruction and is guaranteed to execute.
>>     -///
>>     -static bool isSafeToExecuteUnconditionally(const Instruction &Inst,
>>     +static bool isSafeToExecuteUnconditionally(const Instruction &Inst,
>>                                                 const DominatorTree *DT,
>>     +                                           const
>>     TargetLibraryInfo *TLI,
>>                                                 const Loop *CurLoop,
>>                                                 const LICMSafetyInfo
>>     *SafetyInfo) {
>>     -  // If it is not a trapping instruction, it is always safe to
>>     hoist.
>>     -  if (isSafeToSpeculativelyExecute(&Inst))
>>     +  const Instruction *CtxI =
>>     CurLoop->getLoopPreheader()->getTerminator();
>>
>>
>> A Loop* is not guaranteed to have a preheader, in which case 
>> CurLoop->getLoopPreheader() will return null.
>>
>> Nick
>>
>>     + if (isSafeToSpeculativelyExecute(&Inst, CtxI, DT, TLI))
>>          return true;
>>
>>        return isGuaranteedToExecute(Inst, DT, CurLoop, SafetyInfo);
>>
>>
>>
>>
>> _______________________________________________
>> 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/20150521/5e1c1678/attachment.html>