Skip to content

Latest commit

 

History

History
110 lines (85 loc) · 6.29 KB

README.md

File metadata and controls

110 lines (85 loc) · 6.29 KB

Depreciated

HastyNoise is now depreciated, it has been replaced by FastNoise 2.

license hunter discord
travis appveyor

Wiki

Documentation

HastyNoise

Hasty Noise is a noise library. It aims to provide high performance noise through the use of SIMD instructions on a variety of platforms. Vectorisation of the code allows noise functions to process data in sets of 4/8/16 increasing performance by 700% in some cases (Simplex).

preview

The library compiles a shared library for each of the SIMD instructions sets. During compile time the library will build the highest level of SIMD supported by the compilier and at runtime all of the SIMD libraries built are loaded and the highest supported instruction set is reported. By default the library will use the highest level of SIMD detected. If no support is found it will fallback to standard types (float/int).

"We must not be hasty." - Treebeard

Features

Supported Noise Fractal Perturb
White no yes
Value yes yes
Perlin yes yes
Simplex yes yes
OpenSimplex2 yes yes
Cubic yes yes
Cellular (Worley) no yes
Fractal Types Perturb Cellular Value Celllar Distance
FBM Gradient Value Eucliden
Billow Gradient Fractal Distance Manhattan
RidgidMulti Normalize Value Distance 2 Natural
Gradient Normalize Distance 2
Gradient Fractal Normalize Distance 2 Add
Distance 2 Sub
Distance 2 Mul
Distance 2 Div
Lookup
Distance 2 Cave

Supported Instruction/Compilers/Platforms

Instructions Compilers Platforms
ARM NEON* MSVC v140/v150 Windows
AVX-512F GCC 7 Linux Linux
AVX2 - FMA3 Clang Linux/MacOSX Android*
SSE4.1 MacOSX
SSE2 iOs*

* needs work

Origins

This is an altered version of FastNoiseSIMD. All of the macros in the original version have been replace with templates allowing for a little better debugging. Also the library has been altered to create multiple shared libraries each compiled with the proper SIMD Instructions. The shared libraries are dynamically loaded by the main library.

Related repositories

Performance Comparisons

Timings below timings are to generate 100x 64x64x64 (~26.2M) points of noise on a single thread.

  • CPU: Intel Core i7-5820K @ 3.3Ghz
  • OS: Windows 10
  • Compiler: Visual Studio 2017 x64
Noise Type AVX512 AVX2 SSE4.1 SSE2 None
White Noise 9ms 18ms 21ms 48ms
Value 114ms 243ms 282ms 2071ms
Perlin 193ms 416ms 534ms 2816ms
Simplex 198ms 372ms 474ms 2769ms
OpenSimplex2 218ms 451ms 471ms 6683ms
Cellular 915ms 2095ms 2218ms 16388ms
Cubic 668ms 1370ms 2336ms 5698ms
  • CPU: Intel Xeon W-2125 @ 4.0Ghz
  • OS: Ubuntu 18.04
  • Compiler: gcc 7+ -O3
Noise Type AVX512 AVX2 SSE4.1 SSE2 None
White Noise 3ms 4ms 8ms 18ms 21ms
Value 50ms 80ms 161ms 263ms 509ms
Perlin 62ms 163ms 366ms 445ms 1556ms
Simplex 59ms 155ms 300ms 339ms 1129ms
OpenSimplex2 82ms 175ms 313ms 421ms 1683ms
Cellular 328ms 580ms 1316ms 1358ms 6145ms
Cubic 257ms 490ms 984ms 1790ms 2567ms

Examples

preview preview preview preview preview preview preview