Performance Study of Various Programming Techniques on MicroPython

Say, I want to find the dot product of two vectors a and b. Which of the following implementations is faster?

def dot_product_by_accumulate(a, b):
    s = 0
    for i in range(0, len(a)):
        s += a[i] * b[i]
    return s

def dot_product_by_zip(a, b):
    return sum([x*y for x,y in zip(a,b)])

The first uses an imperative style; the second uses a functional style. Which is faster on a given MicroPython platform?

This type of questions may not be of much concern on a high level, but it can make a big difference in heavy-lifting parts of some algorithms. When it is done repeatedly for each pixel on a 480x360 image, small differences add up.

Faced with this uncertainty, I set out to test the speed of various programming techniques. Experiments are run on OpenMV, a Pyboard with a STM32H743VI ARM Cortex M7 processor at 480 MHz.

Lessons learned

1. Imperatively accumulating a sum is faster than sum([... zip(...)])

2. [func(x) for x in list] is faster than map(func, list)

3. [0] * length to create a list is the fastest

4. a ** b is faster than pow(a, b)

5. Accessing an element in an array is slower than accessing an element in a list. However, an array with float elements uses far less memory than a corresponding list.

These knowledge informs my programming style in these projects:

• micropython-vec - Vector Operations on MicroPython

• micropython-mtx - Fast Matrix Multiplication and Linear Solver on MicroPython

• OpenRV - Robot Vision routines for OpenMV