Port the remaining SIMD intrinsics to const-eval

[sayantn]

successor to #146568, this refactors some implementations and ports the implementation of simd_fma and simd_relaxed_fmato rustc_const_eval

Also adds some remaining f16/f128 support in these intrinsics

r? @RalfJung

[rustbot]

Some changes occurred to the CTFE / Miri interpreter

cc @rust-lang/miri

The Miri subtree was changed

cc @rust-lang/miri

Some changes occurred to the CTFE machinery

cc @RalfJung, @oli-obk, @lcnr

[sayantn]

In a similar fashion to #146568, I have not made the signatures const. If this gets merged quickly enough, I will just add these to #147521, otherwise new PR it is

[RalfJung]

Thanks! Here are some first comments.

All of the new intrinsics Miri supports with this will also need Miri tests.

View changes since this review

[RalfJung]

successor to #146568, this refactors some implementations and implements the following SIMD intrinsics in const-eval

Please update the PR description: This doesn't actually add anything in const-eval, it adds it to the interpreter and then only exposes this in Miri.

[RalfJung]

Thanks!
Still missing tests for the new f16/f128 support.

View changes since this review

[sayantn]

Do we really need f16/f128 tests? It's painful to add them because portable_simd doesn't implement SimdElement for f16 and f128, so we have to roll our own repr_simd types

[RalfJung]

Why do we add f16/f128 support if portable_simd can't even use it?

tests/pass/intrinsics/portable-simd.rs already declares two custom repr(simd) types. That only takes a few lines of code. Declaring two more shouldn't be so bad. We only need smoke tests, but I'd rather not have entirely dead codepaths here.

[sayantn]

I am adding them mostly for completeness sake, but stdarch will need simd_fma with f16.

Declaring two more shouldn't be so bad.

Ok I will add 👍

[sayantn]

@rustbot ready

added tests for f16 and f128 (just copied f32 and f64 tests)