Optimizing your Simulations

Vectors

To make your code perform better, try to pack as many quantities as possible into vectors. Especially on GPUs this dramatically increases the performance of your code.

Example (Three different scalar quantities):

(x,y,z) = eq.generate_quantities(3)
eq.prepare_quantity_for_recursive_definition(x, 4.0, 0, 1)
eq.prepare_quantity_for_recursive_definition(y, 2.0, 0, 1)
eq.prepare_quantity_for_recursive_definition(z, 3.0, 0, 1)
eq.define_quantity_recursively(x, sigma * (y.d(0)-x.d(0)))
eq.define_quantity_recursively(y, x.d(0)*(rho - z.d(0))-y.d(0))
eq.define_quantity_recursively(z, x.d(0)*y.d(0) - z.d(0)*beta)

Example (One vector quantity):

(xyz,) = eq.generate_quantities(1)
eq.define_quantity(xyz,np.array([2.1, 3.2, 4.4]), 0, 1)(tf.pack([sigma * (xyz.d(0)[1]-xyz.d(0)[0]),
                                                                    xyz.d(0)[0] * (rho - xyz.d(0)[2]) - xyz.d(0)[1],
                                                                    xyz.d(0)[0]*xyz.d(0)[1] - beta * xyz.d(0)[2]]))

You can see the performance increase for yourself by comparing the performance of the two Lorenz Attractor examples, the lorenz_attractor_recursive.py script uses three different quantities, while lorenz_attractor.py uses a vector quantity.