Tag Archives: OpenGL


Play with Particles III – Visualize using OpenGL

In the last two posts we implemented three different solvers for the n-body problem, one on the CPU and two different GPU versions. In this post we visualize the n-body simulation graphically with OpenGL to illustrate the cross platform capabilities of Alea GPU and its interoperability with OpenGL.

A Simple OpenGL Example with OpenTK

There exist essentially two different 3d visualization technologies: Microsoft’s DirectX and OpenGL. DirectX is targeting the Windows platform. Examples showing the interoperability of Alea GPU and DirectX can be found in the sample gallery of Alea GPU. The benefit of OpenGL is platform independence. A good starting point is the OpenGL tutorial.

In this blog article we use OpenGL through OpenTK which wraps OpenGL for .NET. You might find the OpenTK documentation, the OpenTK tutorial, how to set up OpenTK, and this OpenTK example useful resources.

We introduce OpenTK with a simplistic example, which renders a triangle. Add the OpenTK NuGet package to your project and reference System.Drawing. Then open the following namespaces:

Create a new class OpenTKExample which inherits from the OpenTK class GameWindow:

Then overwrite the following methods:

On load set the Background to DarkBlue:

On resize of the window use the whole area to paint and set the projection:

On render the frame is where all the drawing happens. Clear the buffer and draw a triangle and three points with different colors. Instead of a triangle we could also draw many different other figures such as points, lines, etc. End the triangle mode and swap the buffers.

Add a function creating an instance of our OpenTKExample class and running it:

The result is a colored triangle on a dark blue background:


We will use the same structure in order to display our particles.

Displaying Particles Directly on the GPU

The main difference between the simple example above and the n-body simulation is that in case of the n-body simulation the data already resides in GPU memory and we do not want copy it from GPU to CPU and back to an OpenGL buffer to finally display the particles. This needs some infrastructure code. First we show how to create two buffers accessible from OpenGL and Alea GPU, in which we save our positions:

  1. Generate an array consisting of GLuint.
  2. Create a buffer using GL.GenBuffers.
  3. For every element of the array:
    1. bind the buffer;
    2. allocate the memory;
    3. get the buffer-size;
    4. unbind the buffer;
    5. register the buffer with cuGLRegisterBufferObject available from Alea GPU.

We now have two buffers which can be accessed from OpenTK and Alea GPU. We also need some resource pointers corresponding to the buffers. We obtain them by calling the Alea GPU function cuGraphicsGLRegisterBuffer inside a cuSafeCall:

If we work with the buffers outside of OpenTK we need to lock their positions. We therefore write a function lockPos which locks the positions, calls a function f on the positions and unlocks the positions again:

To share the buffers we require an Alea.Worker on the same context as used by OpenTK. The following function creates a CUDA context on the machine’s default device:

We use this function to create an Alea.Worker on the same context using the same CUDA device.

We can now initialize the positions using the lockPos function and our newly generated worker:

Recall that we read from oldPos and write to newPos in the GPU implementation. We need to swap the buffers before each integration step using the function swapPos:

In the OnRenderFrame method we swap the buffers and perform an integration step:

We bind the buffer and draw the positions:

Implementation Details

We point out some implementation details. To use the different GPU implementations and test them for different block sizes we introduce a queue of ISimulator objects. During simulation we walk through the queue with an “S” key down event.

We create the simulators and put them into the queue. Note that we also return a dispose function to clean up the simulators at the end:

Here is the logic to switch between simulators:

We use Matrix4.LookAt to inform OpenTK that our viewing position is (0,0,50) and that the viewing direction is along the z axis:

These additional comments should be helpful to understand how the positions are displayed using OpenTK and how the data is directly read from GPU memory. The previous blog posts explain the physics, the CPU implementation, the two GPU implementations and their differences. All that remains is to run the example.