hyperfine-output.txt

$ hyperfine --show-output 'python3 so73815479.py' 'python3 so73815479.py multi' --max-runs 5
Benchmark 1: python3 so73815479.py
multi_proc: False, data dur: 2.83, local dur: 2.31
multi_proc: False, data dur: 2.61, local dur: 2.09
multi_proc: False, data dur: 2.74, local dur: 2.22
multi_proc: False, data dur: 2.75, local dur: 2.26
multi_proc: False, data dur: 2.83, local dur: 2.58
  Time (mean ± σ):      5.180 s ±  0.256 s    [User: 5.875 s, System: 0.196 s]
  Range (min … max):    4.852 s …  5.551 s    5 runs

Benchmark 2: python3 so73815479.py multi
finish new process
multi_proc: True, data dur: 0.77, local dur: 2.29
finish new process
multi_proc: True, data dur: 0.78, local dur: 2.35
finish new process
multi_proc: True, data dur: 0.85, local dur: 2.38
finish new process
multi_proc: True, data dur: 0.77, local dur: 2.29
finish new process
multi_proc: True, data dur: 0.75, local dur: 2.31
  Time (mean ± σ):      3.262 s ±  0.073 s    [User: 6.561 s, System: 0.361 s]
  Range (min … max):    3.201 s …  3.379 s    5 runs

Summary
  'python3 so73815479.py multi' ran
    1.59 ± 0.09 times faster than 'python3 so73815479.py'

so73815479.py

import multiprocessing as mp
import sys

import numpy as np
import time

LOCAL_ITERS = 500
POWER_ITERS = 200
RAND_SIZE = 200


def get_sample():
    x = np.random.rand(RAND_SIZE, RAND_SIZE)
    p = np.random.rand(RAND_SIZE, RAND_SIZE)
    y = 0
    for i in range(POWER_ITERS):
        y += np.power(x, p)
    return y


def data_generator(q_data, total_loops):
    for i in range(total_loops):
        q_data.put(get_sample())
    print('finish new process')


def main(multi_proc=False):
    total_loops = 20

    if multi_proc:
        q_data = mp.Queue()
        new_proc = mp.Process(target=data_generator, args=(q_data, total_loops))
        new_proc.start()
        get_data = q_data.get
    else:
        get_data = get_sample
        new_proc = None
        q_data = None

    get_data_durations = []
    local_durations = []

    for i in range(total_loops):
        get_data_start = time.perf_counter()
        y = get_data()
        get_data_durations.append(time.perf_counter() - get_data_start)

        local_start = time.perf_counter()
        for j in range(LOCAL_ITERS):
            y += np.random.rand(RAND_SIZE, RAND_SIZE)
        local_durations.append(time.perf_counter() - local_start)

    if multi_proc:
        new_proc.join()
        assert q_data.empty()

    data_total = sum(get_data_durations)
    local_total = sum(local_durations)
    print(f'multi_proc: {multi_proc}, data dur: {data_total :.2f}, local dur: {local_total :.2f}')


if __name__ == '__main__':
    main(multi_proc="multi" in sys.argv)