Crate tensorlogic_oxicuda_rng 

tensorlogic-oxicuda-rng

GPU-accelerated random number generation for TensorLogic, with a pure-Rust CPU fallback.

Features

Feature	Description
cpu (default)	PCG-XSH-RR generator with Box-Muller transform, zero external dependencies
gpu	oxicuda-rand GPU backend (requires an NVIDIA GPU + CUDA driver at runtime)

Quick start

use tensorlogic_oxicuda_rng::{RngEngine, RngEngineKind};

let mut rng = RngEngine::new(RngEngineKind::Philox, 42).unwrap();

// f32 variants
let mut out = vec![0f32; 1024];
rng.uniform_f32(&mut out).unwrap();

let mut normal_out = vec![0f32; 1024];
rng.normal_f32(&mut normal_out, 0.0, 1.0).unwrap();

// f64 variants (52-bit mantissa precision)
let mut out64 = vec![0f64; 1024];
rng.uniform_f64(&mut out64).unwrap();

let mut normal_out64 = vec![0f64; 1024];
rng.normal_f64(&mut normal_out64, 0.0, 1.0).unwrap();

// Streaming API — large buffers without single allocation
rng.fill_uniform_chunked(1_000_000, 4096, &mut |chunk: &[f32]| {
    let _ = chunk; // process chunk
}).unwrap();

let mut bernoulli_out = vec![0u8; 1024];
rng.bernoulli(&mut bernoulli_out, 0.3).unwrap();

Thread safety

On the CPU path (default), RngEngine is Send + Sync. On the GPU path (feature = "gpu"), RngEngine is Send but not Sync.

Policy notes

• No rand, rand_distr, or ndarray imports — the PCG generator and Box-Muller transform are implemented from scratch in engine.
• scirs2-core is listed as an optional dependency under the cpu feature to satisfy workspace policy; the actual random primitives live in engine::CpuRngState for zero-dependency builds.

Re-exports

pub use engine::RngEngine;

pub use engine::RngEngineKind;

pub use error::RngError;

Modules

engine

Core RNG engine: a unified GPU + CPU dual-path random number generator.

error

Error types for the tensorlogic-oxicuda-rng crate.