Last few month I was participating in the NVidia CUDA lecture & seminar course hosted by the NVidia company together with CMC department of the MSU. NVidia offered certificate to every participant who will successfully solve all the proposed assignments or use CUDA in his (or her, I’ve seen some girl who did it) course project. There were no place for CUDA in my course project (at least, I though so), so I’ve written the following 5 programs in C++/CUDA.

1. Find all the roots of the piecewise-linear function with A LOT of points. It was the first and the most simple one.
2. Calculate haar wavelet coefficients for the 1D signal.
3. Implement simple real-time raytracer with support for planes, spheres and triangles (also with texturing).
4. Numerical heat equation solution.
5. Optimization of a given a program that performs image filtering using convolution with large kernel.

Need to say, some of this assignments were quite interesting. Problems that have pretty simple general parallel algorithm can be very hard to implement on stream computing architecture efficiently. Anyway, performance boost worth it.

So, now I’m happy owner of that piece of paper (blame Steve Jobs for the bad iphone camera quality):