A Deployable Cross-Platform SIMD RNG Library for C++ (With Bnchmks)

Name: A Deployable Cross-Platform SIMD RNG Library for C++ (With Bnchmks)
Availability: InStock
Author: whisprer

by whisprer·Feb 14, 2026·1 point·0 comments

Visit Project View on HN

AI Analysis

●●SolidNiche GemWizardry

SIMD RNG for C++ faster than std::mt19937, but most projects don't need this speed.

Strengths

•True cross-platform SIMD with graceful fallback (Windows/Linux/macOS); rare to see done right
•Deterministic reproducible streams and zero external dependencies make it deployable
•Legitimate performance gain in bulk generation; benchmarks included and transparent

Weaknesses

•Niche audience—most applications don't need hand-tuned RNG performance over stdlib
•No active adoption signal or proof of production use; 2 stars suggests early-stage
•Lacks API documentation and integration examples for potential users

Post Description

I’ve built a fully cross-platform SIMD-accelerated C++ pseudo-random number generation library designed for real-world deployment (Windows, Linux, macOS).

Repository: https://github.com/whisprer/c-simd-rng-lib/

The goal was to solve a practical gap:

Most high-performance SIMD RNG implementations are either:

academic prototypes

single-architecture

non-portable

incomplete

or not packaged for real deployment

This library provides:

• AVX2 / AVX-512 accelerated paths (with graceful fallback) • Deterministic, reproducible streams • Clean API surface • Zero external runtime dependencies • Works across Win / Linux / macOS • Production-ready build setup

In bulk generation scenarios it significantly outperforms std::mt19937 and standard <random> engines, and benchmarks competitively (or faster) than other SIMD-enabled RNG libraries.

A separate benchmarking repository contains:

• Full comparison suite • Competing library benchmarks • Throughput numbers • Architectural breakdowns • Methodology + raw results

Benchmark repo: https://github.com/whisprer/benchmark/

The focus is high-throughput generation for simulations, Monte Carlo, procedural systems, and statistical workloads where large batches matter more than single draws.

Interested in feedback from folks working in HPC, simulation, game engines, or scientific computing.