Fall 2013
The first part of this project involves a N-body simulation of planets revolving the sun. Each planet is assigned a random velocity. At every frame, acceleration is calculated using the law of universal gravitation.
Each planet is represented as a sphere lit by the star at position (0,0,0).
Basic features include:
- Force calculation between all bodies using global and shared memory
- Height field rendering
- Fragment shader rendering of spherical planets
For part two, I implemented a flocking simulation using the methods described here. Each boid is affected by the following:
- Cohesion -- boid steers to move to the center of mass of all boids in its neighborhood
- Alignmnet -- boid heads toward the average direction of its neighbors
- Separation -- boid steers to avoid being too close to its neighbors
- Avoidance -- boid tries to avoid a boundary. In my implementation, I created an imaginary cube to constrain the boids.
- Wander -- adds a random velocity (constrained to a unit sphere) to each boid to create interesting movement
Here are some results:
Here is a video of the simulation. There are 5000 boids in the scene. The framerate is about 60% slower because of the screen capture.
Comparing the runtime of global versus shared memory with varying numbers of planets from Part 1. The block size is 128.
| Number of bodies | Global Memory | Shared Memory |
|---|---|---|
| 10 | 1.17 | 1.13 |
| 50 | 3.75 | 3.67 |
| 100 | 6.89 | 6.55 |
| 200 | 12.47 | 13.17 |
| 500 | 29.39 | 30.83 |
| 1000 | 58.67 | 60.7 |
| 2000 | 115.61 | 120.48 |
Comparing the effect of block size on shared and global memory. Tests are run with 500 bodies.
| Block Size (500 bodies) | Global Memory | Shared Memory |
|---|---|---|
| 16 | 74.52 | 74.82 |
| 32 | 37.47 | 37.92 |
| 64 | 29.64 | 31.6 |
| 128 | 29.39 | 30.88 |
| 256 | 29.35 | 31.22 |
In both cases, global and shared memory produced similar results. Surprisingly, shared memory is actually slower than global memory for larger number of bodies and for all block sizes. The only time that shared memory improved performance is when there are less than 100 bodies in the scene.