Skip to content

fft with 1.16.6 is faster than with 1.18.0 #17130

Open
@Intro1997

Description

@Intro1997

Reproducing code example:

import random
import time
# import matplotlib
import numpy as np 
from scipy.fftpack import fft
# import cv2

coordinate = np.ndarray((1920, 1080, 2))
print(coordinate.shape)

for i in range(1920):
    for j in range(1080):
      coordinate[i][j] = (random.randint(0, 1920), random.randint(0, 1080))

print(coordinate[0][0])
# test = ([2.+4.j, 3.+3.j, 4.+4.j, 5.+3.j, 4.+2.j, 3.+1.j, 2.+2.j, 1.+3.j])
start_time = time.time();
res = np.fft.fft(coordinate)
end_time = time.time();
print("spend time: ", (end_time - start_time));

Error message:

The result is :

% python2 ./test.py
(1920, 1080, 2)
[ 1022.   435.]
('spend time: ', 0.06316184997558594)
% python3 ./test.py
(1920, 1080, 2)
[1143.  678.]
spend time:  0.18218493461608887

Numpy/Python version information:

The pythons' numpy version are both 1.19.1
Python2: 2.7.16
Python3: 3.7.3

macOS: 10.15.6

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions