[llvm-commits] [llvm] r137672 - /llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp

[Bill Wendling]

On Aug 15, 2011, at 4:00 PM, Eli Friedman wrote:

> On Mon, Aug 15, 2011 at 3:53 PM, Bill Wendling <isanbard at gmail.com> wrote:
>> Author: void
>> Date: Mon Aug 15 17:53:05 2011
>> New Revision: 137672
>> 
>> URL: http://llvm.org/viewvc/llvm-project?rev=137672&view=rev
>> Log:
>> Don't sink the instruction to before a landingpad instruction.
>> 
>> Modified:
>>    llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp
>> 
>> Modified: llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp?rev=137672&r1=137671&r2=137672&view=diff
>> ==============================================================================
>> --- llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp (original)
>> +++ llvm/trunk/lib/Transforms/InstCombine/InstructionCombining.cpp Mon Aug 15 17:53:05 2011
>> @@ -1436,7 +1436,7 @@
>>   }
>> 
>>   BasicBlock::iterator InsertPos = DestBlock->getFirstNonPHI();
>> -
>> +  if (isa<LandingPadInst>(InsertPos)) ++InsertPos; // Skip landingpad inst.
>>   I->moveBefore(InsertPos);
>>   ++NumSunkInst;
>>   return true;
> 
> Is it legal to sink an arbitrary load past a LandingPadInst?
> 
Sure. It's not illegal, at least. The optimizer should be aware that it's taking a 'load' from before an 'invoke' and sinking it into the 'unwind' edge. If it's aware that it's doing that and okay with it, then I don't have a problem allowing it. :)

-bw