simdnoise 2.2.2

SIMD accelerate noise library with runtime feature detection
Documentation

SIMDNoise

Super fast SIMD noise library for Rust. PRs welcome!
Available on crates.io.
Documentation.

Features

  • SSE2, SSE41, and AVX2 instruction sets, along with non SIMD fallback
  • AVX2 version also leverages FMA3
  • Runtime detection picks the best available instruction set
  • Simplex noise, fractal brownian motion, turbulence, and ridge
  • 1D, 2D, 3D, and 4D

Benchmarks

Intel(R) Core(TM) i7-6700 CPU @ 3.40GHz Single Threaded

1D 100,000 points of FBM Noise, 3 Octaves

  • scalar ... bench: 1,694,281 ns/iter (+/- 51,360)
  • sse2 ... bench: 829,633 ns/iter (+/- 41,989)
  • sse41 ... bench: 657,276 ns/iter (+/- 14,358)
  • avx2 ... bench: 294,431 ns/iter (+/- 10,268)

2D 1000x1000 FBM Noise, 3 Octaves

  • scalar ... bench: 74,686,044 ns/iter (+/- 3,053,838)
  • sse2 ... bench: 23,619,783 ns/iter (+/- 1,008,879)
  • sse41 ... bench: 21,847,769 ns/iter (+/- 914,364)
  • avx2 ... bench: 11,791,738 ns/iter (+/- 446,718)

3D 64x64x64 FBM Noise, 3 Octaves

  • scalar ... bench: 22,219,344 ns/iter (+/- 817,769)
  • sse2 ... bench: 10,331,856 ns/iter (+/- 450,920)
  • sse41 ... bench: 9,766,523 ns/iter (+/- 604,034)
  • avx2 ... bench: 5,566,535 ns/iter (+/- 181,791)

4D 24x24x24x24 FBM Noise, 3 Octaves

  • scalar ... bench: 48,324,536 ns/iter (+/- 1,813,984)
  • sse2 ... bench: 26,955,224 ns/iter (+/- 1,253,751)
  • sse41 ... bench: 25,792,680 ns/iter (+/- 749,234)
  • avx2 ... bench: 13,080,348 ns/iter (+/- 491,006)

Todo

  • AVX512 support
  • ARM NEON support
  • Voroni, Cell, and other noise types

Get a block of noise with runtime SIMD detection

The library will, at runtime, pick the fastest available options between SSE2, SSE41, and AVX2

use simdnoise::*;

// Set up noise type and parameters
let noise_type = simdnoise::NoiseType::FBM {
      freq: 0.04,
      lacunarity: 0.5,
      gain: 2.0,
      octaves: 3,
}; 

// Get a block of 2d 800x600 noise, with no scaling of resulting values
// min and max values are returned so you can apply your own scaling
let (an_f32_vec,min,max) = simdnoise::get_2d_noise(0.0, 800, 0.0, 600, noise_type);

// Get a block of 200x200x200 3d noise
let (an_f32_vec,min,max) = simdnoise::get_3d_noise(0.0, 200, 0.0, 200,0.0, 200, noise_type);

// Get a block of noise scaled between -1 and 1
let an_f32_vec = simdnoise::get_2d_scaled_noise(0.0, 800, 0.0, 600, noise_type,-1.0,1.0);

Call noise functions directly

Sometimes you need something other than a block, like the points on the surface of a sphere. Sometimes you may want to use SSE41 even with AVX2 is available


// get a block of 100x100 sse41 noise, skip runtime detection
let (noise,min,max) = simdnoise::sse41::get_2d_noise(0.0,100,0.0,100,noise_type);

// send your own SIMD x,y values to the noise functions directly
unsafe {
  // sse2 simplex noise
  let x = _mm_set1_ps(5.0);
  let y = _mm_set1_ps(10.0);
  let f : __m128 = simdnoise::sse2::simplex_2d(x,y);
  
  // avx2 turbulence
  let x = _mm256_set1_ps(5.0);
  let y = _mm256_set1_ps(10.0);
  let freq = _mm_256_set1_ps(1.0);
  let lacunarity = _mm256_set1_ps(0.5);
  let gain = _mm256_set1_ps(2.0);
  let octaves = 3;
  let f_turbulence : __m256 = simdnoise::avx2::turbulence_2d(x,y,freq,lacunarity,gain,octaves);
    
}