Module simd 

SIMD-accelerated engines for the standard and URL-safe alphabets.

These are gated behind the simd-unsafe feature because they use unsafe. Three engines are provided:

• Simd detects the best available instruction set at runtime and falls back to the scalar GeneralPurpose engine when none is available. It requires std for the detection.
• Avx2 and Neon target a specific instruction set with no runtime detection, so they can be used in no_std builds when the target is known to support the instructions.

Only the STANDARD and URL_SAFE alphabets are accelerated (they share indices 0..=61 and differ only at 62/63). Each engine therefore has dedicated standard / url_safe constructors rather than taking an arbitrary Alphabet; use GeneralPurpose for any other alphabet.

The kernels follow Wojciech Mula’s vectorized base64 algorithms (http://0x80.pl/notesen/2016-01-17-sse-base64-decoding.html and the companion encoding note). AVX2 uses the multiply-based bit (de)interleave; NEON, which lacks the relevant multiplies, uses the shift/mask variant. Both share the same per-alphabet lookup tables, expressed as the associated constants of the SimdAlphabet trait so the kernels can inline them.

Modules

avx2 🔒

Structs

Avx2

A base64 engine that unconditionally uses AVX2, without runtime detection.

Simd

A base64 engine that uses the best SIMD instruction set detected at runtime, falling back to the scalar GeneralPurpose engine.

Enums

Backend 🔒

Which instruction set an engine dispatches to.

SimdKind 🔒

Which accelerated alphabet family an engine uses. Selects the kernel monomorphization at runtime.

Standard 🔒

The STANDARD alphabet family (+// at 62/63).

UrlSafe 🔒

The URL_SAFE alphabet family (-/_ at 62/63).

Constants

BITPOS_LUT 🔒

pshufb decode validity table: maps a high nibble to a single set bit. Shared by both alphabets.

SIMD_MIN_INPUT_DECODE 🔒

SIMD_MIN_INPUT_ENCODE 🔒

Minimum input length before a SIMD path is used. Below these the setup cost outweighs the gain; encode needs more data than decode to break even.

Traits

SimdAlphabet 🔒

A base64 alphabet family the SIMD kernels can accelerate.

Functions

avx2_decode 🔒 ^⚠

Dispatch the AVX2 decode kernel to the right alphabet monomorphization.

avx2_encode 🔒 ^⚠

Dispatch the AVX2 encode kernel to the right alphabet monomorphization.