osino - An Open Simplex Noise implementation for Enoki/CUDA.

This is an Enoki-port of the OpenSimplex.java code by Stefan Gustavson.

It is derived from my C-port sino but improves in several ways:

• Uses Enoki and CUDA to run on the GPU.
• Avoids memory gather by having purely procedural pseudo random directions from Murmur2 hashing.
• More varied noise due to using pseudo random directions instead of using 12 cardinal directions that are shuffled.

It comes with a Marching Cubes implementation to extract a surface, so that a 3D density field can be visualized.

License

3-clause BSD

Dependencies

• Enoki
• CUDA
• ThreadTracer

Building

Install cuda development environment. I used /usr/local/cuda for the destination.

Get the source and dependencies, recursively:

$ git clone --recursive git@github.com:stolk/osino.git

Build enoki:

$ cd externals/enoki
$ mkdir build
$ cd build
$ CXX=clang++-8 CC=clang-8 cmake -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc -DCMAKE_BUILD_TYPE=Debug -DENOKI_CUDA=ON ..
$ make
$ cd ../../..

Before building osino, edit the Makefile to set your compiler. Then use make to build the example.

Test osino:

$ make output.obj

API

The sample coordinates and noise values are contained in Enoki dynamic arrays with this signature:

typedef enoki::CUDAArray<float>    FV;  // Flt vector

To get noise values for a specified set of coordinates, use:

FV osino_2d(FV x, FV y);
FV osino_3d(FV x, FV y, FV z);

Or if you want multi octave noise (also called fractal noise) then use:

FV osino_2d_4o(FV x, FV y);
FV osino_3d_4o(FV x, FV y, FV z);

Note that to make it worthwhile to have a round trip to the GPU, you need to compute a lot of values in one go. Which means millions of noise values, not thousands. Otherwise the communication overhead would defeat the purpose of doing this GPU-side. In that case, you are better off using the AVX backend of Enoki.