Enum F32Transcendental 

pub enum F32Transcendental {
    ScalarLibm,
    AccelerateVForce,
    SleefU10,
}

Which sin/exp implementation an elementwise parity probe evaluates.

The reference stack does not call the scalar libm for large tensors: its CPU elementwise kernels dispatch through a vectorized Vectorized<float>, whose transcendentals are ~1-ulp routines rather than correctly-rounded ones. codec_snake_bisect established that the SnakeBeta seam’s entire residual divergence lives in exactly these two functions — every other operation in that expression is a correctly-rounded f32 *, + or / with no freedom at all — so identifying which vectorized routine the pinned oracle used is the whole remaining question there.

Variants

ScalarLibm

Rust’s scalar f32::sin / f32::exp, i.e. the platform libm. The production reference.

AccelerateVForce

macOS Accelerate vForce (vvsinf / vvexpf), selected only by the parity harness.

On every other target this deliberately falls back to Self::ScalarLibm.

SleefU10

SLEEF’s 1-ulp Sleef_sinf_u10 / Sleef_expf_u10, ported to safe Rust in crate::sleef.

This is the routine an AArch64 Vectorized<float> actually dispatches to, so it is the candidate the vForce probe was only ever standing in for — and unlike vForce it is portable and could therefore be adopted into production if it measures exact.

Trait Implementations

impl Clone for F32Transcendental

fn clone(&self) -> F32Transcendental

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for F32Transcendental

impl Debug for F32Transcendental

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Eq for F32Transcendental

impl PartialEq for F32Transcendental

fn eq(&self, other: &F32Transcendental) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for F32Transcendental