Observation regarding imports and performance

[josie87]

Hello,

I was just playing with the well known synthetic uPy benchmark:

Code: Select all

@micropython.viper
def performanceTest():
    millis = pyb.millis
    endTime = int(millis()) + 10000
    count = 0
    while int(millis()) < endTime:
        count += 1
    print("Count: ", count)
performanceTest()

Score: 5.433.881

Code: Select all

from pyb import millis
@micropython.viper
def performanceTest():
    #millis = pyb.millis
    endTime = int(millis()) + 10000
    count = 0
    while int(millis()) < endTime:
        count += 1
    print("Count: ", count)
performanceTest()

Score: 3.033.640

Code: Select all

@micropython.viper
def performanceTest():
    #millis = pyb.millis
    from pyb import millis
    endTime = int(millis()) + 10000
    count = 0
    while int(millis()) < endTime:
        count += 1
    print("Count: ", count)
performanceTest()

Score: 5.433.740

I get the same behaviour when using @micropython.native or no decorator.

My current strategy is to put my imports on top of the file, since its more readable.

Is the conclusion that uPy is usually faster when placing the imports inside the function/method?

Btw: Tested on my STM32F4 Discovery (I own a pyboard, but borrowed it to my brother). Since the chip is the same it should show similar performance.

[pfalcon]

Is the conclusion that uPy is usually faster when placing the imports inside the function/method?

Nope, that conclusion is wrong, or more exactly, it's sideways. MicroPython (and whole lot of other scripting languages) have more performance when using local variables, than with globals (that's because global variables are accessed by name, while locals are usually optimized to be access by stack index). If you do import within function, you just put module object into local variable.

So, if you really need to have more performance, then "millis = pyb.millis" (like shown commented in your code) should be enough, and is the clearest way to do it.

[josie87]

Thanks