[llvm] [RISCV] Add a pass to eliminate special copies in order to facilitate shrink-wrap optimization (PR #140716)

[Philip Reames via llvm-commits]

preames wrote:

Can you say a bit more about the motivation for this patch?  Was this noticed from workload analysis?  If so, which workload?  Have you done any broader performance comparison to justify this?

The heuristic in this patch seems somewhat unprincipled (i.e. why only copies feeding branches?).  That doesn't mean we can't accept it, but it does mean we need to know why this helps in practice.  

https://github.com/llvm/llvm-project/pull/140716