This program is part of my master's thesis which concerns real-time fluid simulation using OpenCL and OpenGL.

• OpenCL 1.2 to provide GPU acceleration for simulation part of the program
• OpenGL 3.3 for visualising simulation output
• Qt 5.4 to provide rendering surface and other goodies (like resources embedded in executable, loading OpenGL extensions and context creation, ...)
• Boost.Compute, which simplifies work with OpenCL API and provides some handy parallel algorithms (like sorting)

• To compile the project you can use the provided qmake file, which can be used directly by typing qmake followed by make on command line or from Qt Creator.

• Program execution can be tweaked by additional command line arguments:

  • -particle-count <number-of-particles> determines the number of particles that will be used for simulation in mode 1. <number-of-particles> is an integer representing the number of particles.

  • -grid-size <dimensions> determines dimensions of the uniform grid that is used for neighbour search acceleration. <dimensions> is a string in format WIDTHxHEIGHTxDEPTH, where WIDTH, HEIGHT and DEPTH are integer numbers determining respectively the width, the height and the depth of the specified uniform grid (these numbers must be powers of 2). Another option is to write -grid-size SIZE, where again SIZE is an integer power of 2 determining the size of the uniform grid. In this case, the width, the height and the depth of the uniform grid are all set to SIZE.

  • -benchmark when specified, this option runs the program in command line mode and performs a comprehesive suite of stress tests to evaluate its performance for a variety of uniform grid and particle count settings. Beware that this option will make the program run for quite a while and may significantly load your system.

  • -benchmark-rendering this is a smaller set of stress tests, that will run several iterations of the main program loop and print simple statistics on the time spent simulating and rendering the fluid for given particle count and grid size settings. This option also makes the program run without creating a window.

• simulation is done using Smoothed Particle Hydrodynamics
• rendering is done using Screen Space Fluid Rendering with Curvature Flow
• with proper hardware (in my case Radeon R9 280X, Intel Core i5 4690K, 8GB of RAM) and latest drivers (in my case Catalyst 15.7.1), the program is capable of simulating over 1,000,000 particles at around 30-40 FPS (depending on how is the fluid spread across the domain) and over 4,000,000 particles at about 8-12 FPS. These numbers are including full-fledged visualization.