Fearless SIMD 1.0 brings unsafe-free vectorization to stable Rust

The Rust SIMD library Fearless SIMD has reached 1.0 after eight years of development, offering portable vector abstractions, autovectorization, function multiversioning, and safe access to platform intrinsics under one roof. Its central pitch is eliminating the thousands of unsafe blocks that riddle competing SIMD crates: the project confines all memory-unsafe operations to two small, auditable building blocks — a kernel! macro that leverages the compiler’s target-feature v1.1 support to call intrinsics without unsafe, and a bytemuck/zerocopy-style safe transmute module that wraps SIMD loads and stores. Because Rust’s type system guarantees the rest of the code is safe as long as those two pieces are, the maintainers argue the whole crate is provably memory-safe.

On performance, the library exposes both precise (cross-platform-identical) and fast (platform-dependent) variants of edge-case-sensitive operations like swizzles and floating-point max, lets algorithms target native hardware vector widths, and permits dropping to zero-overhead intrinsics where portable abstractions fall short. A companion crate, fearless_simd_macros 0.1, adds a #[simd] attribute that handles multiversioning and inlining automatically, avoiding the leaky trade-offs of prior approaches. Ergonomics is the one area the team expects to keep evolving, potentially via the Struct Target Features RFC.

The release commits to three years of security updates and API stability, with paths to adopt future Rust and hardware features (f16, SVE, RISC-V Vector) without breaking changes. Rather than competing with the still-unstabilized std::simd, the project plans to build on it once available while continuing to supply the ecosystem-level pieces the standard library won’t. Adoption is already meaningful: 30 direct dependents and over a thousand crates relying on it transitively.