half-rs
half-rs copied to clipboard
starkat99
Half-precision floating point types f16 and bf16 for Rust.
this code: ```rust let x = f16::from_f32(std::f32::consts::PI); println!("{:.1e}", x); println!("{:.1e}", f32::from(x)); ``` produces this ``` 3.140625e0 3.1e0 ``` `f16::debug` just uses `fmt(self.to_f32(), f)` so I don't know what might be...
Preface: Thanks for this crate - it's super fundamental to the ecosystem! ❤️ There are too many fuzzing crates! This adds support for [rust-fuzz/arbitrary](https://github.com/rust-fuzz/arbitrary) behind a feature flag. Context: https://github.com/gfx-rs/wgpu/pull/5701...
Bincode 2.0 will move away from using Serde's serialize/deserialize into it's own Decode/Encode structs, and this adds a bincode feature flag that supports them. Some notes: - Yes it's a...
NEON intrinsics are not supported on big endian targets. Change to use fall back types for big endian aarch64. tested on `aarch64_be-unknown-linux-gnu_ilp32` and `aarch64_be-unknown-linux-gnu`
`f16::from_f64` with hardware intrinsics gives wrong (or at least different) result for corner cases
This got caught by RustPython's regression tests that we port from CPython. If this is expected/acceptable behavior, that's fine, since `from_f64_const` works anyways. Here's the failing test cases adapted into...
