An example of how to use orthogonal projection to speed up some mathematical CUDA kernels. Checkout this article I wrote for a detailed explanation of the math and implementation behind it.

Some CUDA kernels perform heavy scientific computing tasks, this sample project shows how to find alternative functions and speed up the kernel.

To approximate functions with polynomial, use opc.py (Orthogonal Projection Calculator). It's able to calculate the derivative, integral of a polynomial; get an orthonormal basis of an inner product space of polynomials; or find polynomial approximation of any continuous real-valued function.

To get a detailed list of its commands, use python opc.py --help. Make sure you have Python 3 and SymPy installed.

The CUDA program in this project is an implementation of a specific example I mentioned in the article. It contains many testing cases to prove the claims in the article.

The supplementary_resources folder contains some additional materials to help you understand orthogonal projection visually (a macOS Grapher file), and a text file with a sample output of Orthogonal Projection Calculator when approximating a function.