[clang] [llvm] [AArch64][llvm] The preferred disassembly for `bti` is now `bti r` (PR #171819)

[Jonathan Thackray via cfe-commits]

jthackray wrote:

> Changing the preferred disassembly for bti instructions is okay, I guess? Maybe a little inconvenient for anyone who wants to build code with `-no-integrated-as`, though... maybe worth considering adding a flag to force downgraded assembler output? Not sure how common that sort of workflow is these days.
> 
> From what I can see of the Arm documentation, FEAT_BTIE is a feature which exists, and processors that implement this feature have additional functionality related to BTI. This patch completely removes any awareness from the compiler of FEAT_BTIE. Are you saying that nothing in the compiler ever needs to be aware of whether a processor implements FEAT_BTIE?

We've agreed with our GNU/gcc team that we'll keep the `+btie` flag, as we already have a `+bti` flag. Gcc doesn't have a `+bti` and won't add a `+btie` flag either, as I understand it. Whilst it's a minor difference (we want to minimise these) they're advanced flags that most people are unlikely to use.

https://github.com/llvm/llvm-project/pull/171819