Add demangling for f16 and f128

[tgross35]

k was selected for f16/half and q for f128/quad.

These are different from LLVM's defaults which are Dh for half (link) and g for quad (link). I don't know if there is any reason to mirror LLVM here because it would mean adding multi character tags for primitives. I just used q because it it's more obviously in line with f for float/f32 and d for double/f64.

Part of rust-lang/rust#114607

[tgross35]

The f16 & f128 RFC was merged so I think it is OK to decide on mangling. Marking it as ready for review @Mark-Simulacrum

[tgross35]

I think we probably don't need different mangling for PowerPC that LLVM seems to have but could use confirmation https://github.com/llvm/llvm-project/blob/fdef7952cbc26fd31d0d92a4f14dde58bc461fe9/clang/lib/Basic/Targets/PPC.h#L351 patch llvm/llvm-project@0461534

[tgross35]

@rcvalle / @michaelwoerister, we need this mangling at some point - any guidance on how to proceed?

Itanium uses Dh for f16 but I think that may conflict with some dyn signature.

[michaelwoerister]

We'll need to take a look at what the grammar allows. Is this implemented in the compiler already? Will these types be primitive types in Rust?

[michaelwoerister]

I just saw the links in the PR message. Will read those 🙂

[tgross35]

We'll need to take a look at what the grammar allows. Is this implemented in the compiler already? Will these types be primitive types in Rust?

The primitives exist in the compiler but not yet fully exposed (rust-lang/rust#121926 coming soon). Currently it does the itanium mangling and just ICEs with v0 mangling.

[michaelwoerister]

k and q seem like good picks to me.

@ehuss suggested here that we extend the mangling scheme by updating the official docs. I think that is a good idea. Would you mind opening a PR that adds the two cases to the basic types section and then we can do a compiler team FCP to sign off on the change.

[rcvalle]

I added a comment at rust-lang/rust#121728 (comment), but this is for CFI encoding and cross-language CFI compatibility and independent/dissociated to Rust mangling/demangling and a different encoding can be used for it.

[tgross35]

Thanks for the pointers, I opened rust-lang/rust#122106 to follow up

[michaelwoerister]

I think this can be closed now, right @tgross35?

[tgross35]

Totally forgot abut this, thanks!