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

[Craig Topper via llvm-commits]

topperc wrote:

If I understand things. Adding this to optimizeCondBranch and relying on the call from PeepholeOptimizer doesn't get any optimizations on real world code because the problem we're trying to solve is created long after after the PeepholeOptimizer. It effects some tests in lit, but that's primarily things that should have been folded in IR like a branch on `fcmp false`.

We still need a late optimization pass to run some kind of branch cleanup to get the real world cases created by tail duplication, machine block placement, machine copy propagation.

So I guess my question is, what do we gain by spreading the logic around instead of doing everything in a late pass like [#117060](https://github.com/llvm/llvm-project/pull/117060)?

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