Tracking Issue for `float_mul_add_relaxed`

[scottmcm]

Feature gate: #![feature(float_mul_add_relaxed)]

This is a tracking issue for the mul_add_relaxed method on floating-point types which is either mul_add (1 rounding) or *+ (2 roundings), nondeterministically.

Public API

// in std
impl f32/f64/... {
    pub fn mul_add_relaxed(self, multiplicand: Self, addend: Self) -> Self;
}

Steps / History

(Remember to update the S-tracking-* label when checking boxes.)

• ACP: Add an fN::mul_add variant that might round twice libs-team#712
• Implementation: Add mul_add_relaxed methods for floating-point types #151793
• Final comment period (FCP)¹
• Stabilization PR

Unresolved Questions

• We would like to see some documentation added to the effect of "If you want some target-specific performance optimization possibilities but need a guarantee of either the fused or unfused behavior, use this. If you want even more performance and aren't concerned about error bounds, use the algebraic_* functions such as algebraic_mul and algebraic_add.".

Footnotes

1. https://std-dev-guide.rust-lang.org/feature-lifecycle/stabilization.html ↩

[scottmcm]

E-help-wanted: Anyone feel like making the PR to add this API? The intrinsic was added in #124874, so it should be relatively straight-forward to add the unstable functions calling it.

[RalfJung]

Should this be available in const? That will take a a bit more care due to the non-determinism: we'll need a machine hook making the decision so that Miri can remain non-deterministic, but const makes some fixed choice.

[bjoernager]

I don't believe anyone would be against making this a const fn item, it's likely just a matter of who is willing to implement it. FWIW regular mul_add is already const fn.