[PATCH] D19369: VirtRegMap.cpp should not just remove an identity copy which also has a impl-def of a super-reg

[Quentin Colombet via llvm-commits]

Also, side question, how do we end up with this representation?
Do we use an undef lane or something, that’s why the verifier complains?

> On Apr 21, 2016, at 9:47 AM, Quentin Colombet <qcolombet at apple.com> wrote:
> 
> qcolombet added a subscriber: qcolombet.
> qcolombet added a comment.
> 
> We could fix MachineInstr:: isIdentityCopy instead. Indeed, the next optimization that removes them will hit the same issue. Moreover, I do not think we are supposed to have implicit_def around after TwoAddrPass (not entirely sure about that though).
> 
> Q.
> 
> 
> http://reviews.llvm.org/D19369
> 
> 
>