[llvm] r246304 - [CodeGen] isInTailCallPosition didn't consider readnone tailcalls

[David Majnemer via llvm-commits]

On Fri, Aug 28, 2015 at 10:16 AM, Reid Kleckner <rnk at google.com> wrote:

> Test case?
>

Forgot to 'git add', added in r246306.


> Also, I think you can do a more precise analysis here. We should be able
> to tail call in this example:
>
> i32* @f() {
>   %a = tail call i32* @foo_readnone()
>   store i32 0, i32* %b
>   ret i32* %a
> }
>
> I think in the readnone case you just need to check that the return value
> is used by nothing more than the ReturnInst.
>

I think that's possible, I'm willing to live with the current state of
things until I'm finished adding a 'pure' attribute to LLVM IR.


>
> On Fri, Aug 28, 2015 at 9:44 AM, David Majnemer via llvm-commits <
> llvm-commits at lists.llvm.org> wrote:
>
>> Author: majnemer
>> Date: Fri Aug 28 11:44:09 2015
>> New Revision: 246304
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=246304&view=rev
>> Log:
>> [CodeGen] isInTailCallPosition didn't consider readnone tailcalls
>>
>> A readnone tailcall may still have a chain of computation which follows
>> it that would invalidate a tailcall lowering.  Don't skip the analysis
>> in such cases.
>>
>> This fixes PR24613.
>>
>> Modified:
>>     llvm/trunk/lib/Analysis/ValueTracking.cpp
>>     llvm/trunk/lib/CodeGen/Analysis.cpp
>>
>> Modified: llvm/trunk/lib/Analysis/ValueTracking.cpp
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/ValueTracking.cpp?rev=246304&r1=246303&r2=246304&view=diff
>>
>> ==============================================================================
>> --- llvm/trunk/lib/Analysis/ValueTracking.cpp (original)
>> +++ llvm/trunk/lib/Analysis/ValueTracking.cpp Fri Aug 28 11:44:09 2015
>> @@ -3147,7 +3147,8 @@ bool llvm::isSafeToSpeculativelyExecute(
>>          LI->getPointerOperand(), LI->getAlignment(), DL, CtxI, DT, TLI);
>>    }
>>    case Instruction::Call: {
>> -    if (cast<CallInst>(Inst)->doesNotAccessMemory())
>> +    auto *CI = cast<CallInst>(Inst);
>> +    if (CI->doesNotAccessMemory() && !CI->isMustTailCall())
>>        return true;
>>      return false; // The called function could have undefined behavior or
>>                    // side-effects.
>>
>> Modified: llvm/trunk/lib/CodeGen/Analysis.cpp
>> URL:
>> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/Analysis.cpp?rev=246304&r1=246303&r2=246304&view=diff
>>
>> ==============================================================================
>> --- llvm/trunk/lib/CodeGen/Analysis.cpp (original)
>> +++ llvm/trunk/lib/CodeGen/Analysis.cpp Fri Aug 28 11:44:09 2015
>> @@ -506,18 +506,16 @@ bool llvm::isInTailCallPosition(Immutabl
>>
>>    // If I will have a chain, make sure no other instruction that will
>> have a
>>    // chain interposes between I and the return.
>> -  if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
>> -      !isSafeToSpeculativelyExecute(I))
>> -    for (BasicBlock::const_iterator BBI = std::prev(ExitBB->end(), 2);;
>> --BBI) {
>> -      if (&*BBI == I)
>> -        break;
>> -      // Debug info intrinsics do not get in the way of tail call
>> optimization.
>> -      if (isa<DbgInfoIntrinsic>(BBI))
>> -        continue;
>> -      if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
>> -          !isSafeToSpeculativelyExecute(BBI))
>> -        return false;
>> -    }
>> +  for (BasicBlock::const_iterator BBI = std::prev(ExitBB->end(), 2);;
>> --BBI) {
>> +    if (&*BBI == I)
>> +      break;
>> +    // Debug info intrinsics do not get in the way of tail call
>> optimization.
>> +    if (isa<DbgInfoIntrinsic>(BBI))
>> +      continue;
>> +    if (BBI->mayHaveSideEffects() || BBI->mayReadFromMemory() ||
>> +        !isSafeToSpeculativelyExecute(BBI))
>> +      return false;
>> +  }
>>
>>    const Function *F = ExitBB->getParent();
>>    return returnTypeIsEligibleForTailCall(
>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at lists.llvm.org
>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150828/c6ed51f9/attachment.html>