[llvm] [BOLT] Remap instruction labels during tail duplication (PR #219920)

[via llvm-commits]

WangJee wrote:

> I don't remember if I already asked this, I thought I did, but have you folks measured any wins with tail duplication? That is disabled by default because it is quite hard to pull wins out of this in x86.

Yes, this came up in the issue discussion. In our testing, we did not observe any significant performance gains from tail duplication on RISC-V.
Given these results and the correctness bug on RISC-V, would simply skipping this pass on RISC-V, even when explicitly enabled, be a reasonable fix?

https://github.com/llvm/llvm-project/pull/219920