[llvm] [RISCV] Add groupid/bitmask for RISC-V extension (PR #94440)

[Craig Topper via llvm-commits]

topperc wrote:

> > I'm not sure we should allocate bits for extensions that aren't supported by hwprobe. Without hwprobe or another enumeration method, the multiversion function that uses a feature that isn't in hwprobe will always never be called. Does that give users false hope?
> > Should we try to keep the most common features in the same group as long as we can? Like F and V in the same group.
> 
> hwprobe is Linux-specific, this proposed interface (which I don’t love, but at least it’s not hwprobe) is not. And it can still fail at run time if you run on an OS that doesn’t yet support the extension.

Do you envision that other OSes that aren't Linux will support more extensions than what hwprobe supports? Or that hwprobe will grow to support all extensions?

I'm just not sure we should we allocate bit positions that no OS will expose.

https://github.com/llvm/llvm-project/pull/94440