FAST

Zero-cost, hyper-optimized deterministic computation.

Features

Requires nightly rust.

• ⚡️ Zero-overhead abstractions
• 🚀 SIMD-first design
• 🎯 Cache-optimal algorithms
• 🧮 Vectorized operations
• 🎲 Flexible random number generation
• 🛡️ Compile-time validation

Performance

Raw computational speed is a core principle:

• Direct SIMD mapping
• Cache-line alignment
• Vectorized processing
• Minimal runtime overhead
• Deterministic and repeatable results

Installation

[dependencies]
fast = "0.1.0"

Why Fast?

• SIMD Vectorization: Leverages SIMD intrinsics for parallel operations on data arrays, boosting performance by processing multiple elements simultaneously.

• Compile-Time Shuffle: Utilizes compile-time constants and const fn to define SIMD data shuffle patterns, minimizing runtime overhead.

• PCG Randomness: Employs a SIMD-optimized PCG generator for fast, high-quality pseudorandom number generation. Weyl sequences are added for increased entropy.

• Probability Distributions: Provides optimized sampling for Normal, Exponential, Gamma, Poisson, and Beta distributions using math intrinsics.

• Distribution Transforms: Enables composition of distributions via transforms, additions, mixes, and multiplies allowing for the building of complex simulations.

Requirements

• Rust 1.70+
• CPU with SIMD support

Platform Support

• x86_64 (AVX, AVX2, AVX-512)
• ARM (NEON, SVE)

Safety

• Compile-time validation
• Bounds checking elimination
• Alignment verification
• Type safety guarantees
• Architecture validation

Contributing

FAST is a low-level crate for high-performance computation. We welcome:

1. Performance improvements
2. Architecture optimizations
3. New algorithms
4. Documentation
5. Benchmarks

License

Apache 2.0 / MIT dual license

FAQ

Why FAST?

• Zero Cost: No runtime overhead
• SIMD First: Built for vector instructions
• Cache Optimal: Memory layout tuned
• Repeatable: Results are deterministic given the same inputs.

FAST vs Others?

• Focus on raw performance
• SIMD by default
• Designed for low-level control
• Minimal abstractions

What about random numbers?

• Provides flexible and performant random number generation
• Useful for simulation and statistical applications