simdutf8

Bug fixes

• Fix Undefined Behavior in WebAssembly SIMD128 (#86) (thanks @CryZe)
• Documentation and clippy fixes (thanks @rtfeldman, @jqnatividad, @rhysd)

Performance

• WASM: Don't use u8x16_bitmask for ASCII Check (#79) (thanks @CryZe)

Bug fixes

• Fix Undefined Behavior in WebAssembly SIMD128 (#86) (thanks @CryZe)
• Documentation and clippy fixes (thanks @rtfeldman, @jqnatividad, @rhysd)

Performance

• WASM: Don't use u8x16_bitmask for ASCII Check (#79) (thanks @CryZe)

New features

• WASM (wasm32) support thanks to @almann!

Improvements

• Make aarch64 SIMD implementation work on Rust 1.59/1.60 with create feature aarch64_neon
• For Rust Nightly the aarch64 SIMD implementation is enabled out of the box.
• Starting with Rust 1.61 the aarch64 SIMD implementation is expected to be enabled out of the box as well.

Performance

• Prefetch was disabled for aarch64 since the requisite intrinsics have not been stabilized.

New features

• WASM (wasm32) support

Improvements

• Make aarch64 SIMD implementation work on Rust 1.59/1.60 with create feature aarch64_neon
• For Rust Nightly the aarch64 SIMD implementation is enabled out of the box.
• Starting with Rust 1.61 the aarch64 SIMD implementation is expected to be enabled out of the box as well.

Performance

• Prefetch was disabled for aarch64 since the requisite intrinsics have not been stabilized.

New features

• Low-level streaming validation API in simdutf8::basic::imp

New features

• Low-level streaming validation API in simdutf8::basic::imp

New features

• Aarch64 support (e.g. Apple Silicon, Raspberry Pi 4, ...) with nightly Rust and crate feature aarch64_neon

Performance

• Another speedup on pure ASCII data
• Aligned reads have been removed as the performance was worse overall.
• Prefetch is used selectively on AVX 2, where it provides a slight benefit on some Intel CPUs.

Comparison vs v0.1.1 on x86-64

Other

• Refactored SIMD integration to allow easy implementation for new architectures
• Full test coverage
• Thoroughly fuzz-tested

New features

• Aarch64 support (e.g. Apple Silicon, Raspberry Pi 4, ...) with nightly Rust and crate feature aarch64_neon

Performance

• Another speedup on pure ASCII data
• Aligned reads have been removed as the performance was worse overall.
• Prefetch is used selectively on AVX 2, where it provides a slight benefit on some Intel CPUs.

Comparison vs v0.1.1 on x86-64

Other

• Refactored SIMD integration to allow easy implementation for new architectures
• Full test coverage
• Thoroughly fuzz-tested

Performance

• Large speedup on small inputs from delegation to std lib
• Up to 50% better peak throughput on ASCII
• #[inline] main entry points for a small general speedup.

Benchmark against v0.1.0

Other

• Make both Utf8Error variants implement std::error::Error
• Make basic::Utf8Error implement core::fmt::Display
• Document Minimum Supported Rust Version (1.38.0).
• Reduce package size.
• Documentation updates.

Performance

• Large speedup on small inputs from delegation to std lib
• Up to 50% better peak throughput on ASCII
• #[inline] main entry points for a small general speedup.

Benchmark against v0.1.0

Other

• Make both Utf8Error variants implement std::error::Error
• Make basic::Utf8Error implement core::fmt::Display
• Document Minimum Supported Rust Version (1.38.0).
• Reduce package size.
• Documentation updates

Documentation updates only.

0.1.x releases will have API compatibility.

• Documentation updates only.

0.1.x releases will have API compatibility.

Documentation update only.

• Documentation update only.

Documentation update only.

• Documentation update only.

• Initial release.