Trait std::simd::StdFloat

source ·

pub trait StdFloat: Sealed + Sized {
    // Required method
    fn fract(self) -> Self;

    // Provided methods
    fn mul_add(self, a: Self, b: Self) -> Self { ... }
    fn sqrt(self) -> Self { ... }
    fn ceil(self) -> Self { ... }
    fn floor(self) -> Self { ... }
    fn round(self) -> Self { ... }
    fn trunc(self) -> Self { ... }
}

🔬This is a nightly-only experimental API. (portable_simd #86656)

Expand description

This trait provides a possibly-temporary implementation of float functions that may, in the absence of hardware support, canonicalize to calling an operating system’s math.h dynamically-loaded library (also known as a shared object). As these conditionally require runtime support, they should only appear in binaries built assuming OS support: std.

However, there is no reason SIMD types, in general, need OS support, as for many architectures an embedded binary may simply configure that support itself. This means these types must be visible in core but have these functions available in std.

f32 and f64 achieve a similar trick by using “lang items”, but due to compiler limitations, it is harder to implement this approach for abstract data types like Simd. From that need, this trait is born.

It is possible this trait will be replaced in some manner in the future, when either the compiler or its supporting runtime functions are improved. For now this trait is available to permit experimentation with SIMD float operations that may lack hardware support, such as mul_add.

Required Methods§

source

fn fract(self) -> Self

🔬This is a nightly-only experimental API. (portable_simd #86656)

Returns the floating point’s fractional value, with its integer part removed.

Provided Methods§

source

fn mul_add(self, a: Self, b: Self) -> Self

🔬This is a nightly-only experimental API. (portable_simd #86656)

Fused multiply-add. Computes (self * a) + b with only one rounding error, yielding a more accurate result than an unfused multiply-add.

Using mul_add may be more performant than an unfused multiply-add if the target architecture has a dedicated fma CPU instruction. However, this is not always true, and will be heavily dependent on designing algorithms with specific target hardware in mind.

source