[llvm-dev] UB and known bits

[John Regehr via llvm-dev]

> My personal opinion here is that we create two different interfaces into computeKnownBits: one which reasons that
> operating on undef is OK and one which is not comfortable with operating on undef.
> The one uncomfortable with operating on undef would be ideal for stuff like isSafeToSpeculativelyExecute while the
> other would be great for isKnownToBeAPowerOfTwo.

This sounds easy and like the right thing to do.

I'm working on a patch that fixes up some of the imprecisions, but so far 
I'm not doing any of the UB-based ones.

John