Expose the `vfp2sp` target feature on 32-bit Arm targets

[tgross35]

vfp2sp is the LLVM target feature that enables hard float operations on f32. It is implied by vfp2 (f32 and f64 operations) which is in the baseline for targets like armv7-unknown-linux-gnueabihf, but vfp2sp is also in the baseline of a number of targets that don't have vfp2 such as thumbv7em-none-eabihf.

We should expose this so it's possible to gate 32-bit hard float operations on cfg(target_feature = "vfp2sp").

Some context at rust-lang/compiler-builtins#1300 (comment).

@rustbot ping arm-maintainers

[rustbot]

Dear ARM Maintainers, your feedback has been requested on this issue/PR.
Thanks! <3

cc @adamgemmell @davidtwco @Jamesbarford @lqd

[adamgemmell]

I expect this feature name is probably fine, but note there's some disagreement on what arm's floating point feature names should be (see this thread #149512 (comment)). I'm planning to tackle this problem properly in a few months.

[tgross35]

I agree that it should be fine but from that thread, @RalfJung would you find vfp2sp problematic? The only thing it should imply is fpregs, IIUC.

[RalfJung]

So far nobody said anything that made my alarm bells go off, but I also don't know much about this. :)

We used the C names for everything so far so I think we can continue doing so for now. It's all unstable anyway so @adamgemmell can still do a renaming/cleanup later even if we initially land this as vfp2sp.