This is a workaround for bad codegen (Rust bug, LLVM bug) for the any() and all() reductions for NEON-backed SIMD vectors on 32-bit ARM. On other platforms these delegate to any() and all() in core::simd.
The plan is to abandon this crate once the LLVM bug is fixed or core::simd works around the LLVM bug.
The code is forked from the packed_simd crate.
This crate requires Nightly Rust as it depends on the portable_simd feature.
MIT OR Apache-2.0, since that's how packed_simd is licensed.