[llvm-commits] CVS: llvm/lib/VMCore/ConstantFolding.cpp

[Reid Spencer]

On Thu, 2006-11-30 at 17:33 -0800, Chris Lattner wrote:
> Reid,
> 
> On Nov 30, 2006, at 4:25 PM, Anton Korobeynikov wrote:
> > Introducing external weak linkage. Darwin codegen should be added  
> > later.
> >
> > Index: llvm/lib/VMCore/ConstantFolding.cpp
> > diff -u llvm/lib/VMCore/ConstantFolding.cpp:1.100 llvm/lib/VMCore/ 
> > ConstantFolding.cpp:1.101
> > --- llvm/lib/VMCore/ConstantFolding.cpp:1.100	Sun Nov 26 19:05:10 2006
> > +++ llvm/lib/VMCore/ConstantFolding.cpp	Thu Nov 30 18:25:12 2006
> > @@ -894,11 +894,8 @@
> >
> >    case Instruction::PtrToInt:
> >      // Cast of a global address to boolean is always true.
> > -    if (isa<GlobalValue>(V)) {
> > -      if (DestTy == Type::BoolTy)
> > -        // FIXME: When we support 'external weak' references, we  
> > have to
> > -        // prevent this transformation from happening.  This code  
> > will need
> > -        // to be updated to ignore external weak symbols when we  
> > support it.
> > +    if (const GlobalValue *GV = dyn_cast<GlobalValue>(V)) {
> > +      if (DestTy == Type::BoolTy && !GV->hasExternalWeakLinkage())
> >          return ConstantBool::getTrue();
> >
> 
> This code is wrong, but it's not anton's fault.  This is fallout from  
> the cast patch.  ptrtoint(&GV, bool) is not guaranteed to be true,  
> even with non-extern-weak symbols, as the ptrtoint truncates, it  
> doesn't compare to null any more.
> 
> This code should be removed, 

Okay, I agree. It should be removed.

> and code should be added to  
> ConstantFolding.cpp in the setne(&GV, null)/seteq(&GV, null) codepath  
> to handle this.

Which code path?  You mean you generally want a setne(&GV, null) to be
folded to ConstBool::True and seteq(&GV, null) to be folded to
ConstBool::False ?

> 
> Thanks,
> 
> -Chris
> 
>