Performance of dpnp is 350x slower than numpy on Monte Carlo Pi example

[samaid]

import numpy as np
import time

BATCH_SIZE = 2048 # Draw random pairs (x, y) in batches of BATCH_SIZE elements
N_BATCHES = 1024 # Total number of batches
N = N_BATCHES * BATCH_SIZE # Total number of random pairs (x, y)


def monte_carlo_pi_batch():
 x = np.random.random(BATCH_SIZE)
 y = np.random.random(BATCH_SIZE)
 acc = np.count_nonzero(x * x + y * y <= 1.0)
 return 4.0 * acc / BATCH_SIZE


def monte_carlo_pi():
 s = 0.0
 for i in range(N_BATCHES):
 s += monte_carlo_pi_batch()
 return s/N_BATCHES


def main():
 print("Using device ...")
 t1 = time.time()
 pi = monte_carlo_pi()
 t2 = time.time()
 print("Pi =", pi)
 print("Done in ", t2 - t1, "seconds...")


if __name__ == "__main__":
 main()

[antonwolfy]

The issue was resolved in #1278.