[llvm] [RISCV] Optimize conditional branches that can be statically evaluated (PR #131684)

[Philip Reames via llvm-commits]

preames wrote:

> > If you just make the first branch unconditional in optimizeCondBranch will later optimizations delete dead branches after it?
> 
> I'm running into crashes and incorrect CodeGen if I do that, and its not obvious why after I tried to resolve it for about an hour. I will have to take a deeper look if we decide that we'd like to fuse `trySimplifyCondBr` and `optimizeCondBranch`.

I found myself asking the same question Craig did here.  I have a feeling here that the scope of this patch has increased enough that you should go take another look at this question.  I don't have any more than a feeling here, but to my understanding, leaving an unconditional branch followed by a conditional one is supposed to be legal, and cleaned up later.  Just a guess, did you maybe leave the successors in an invalid state?  If you're not careful, it would be easy to delete a successor edge with this code structure.  

https://github.com/llvm/llvm-project/pull/131684