2018-10-03 18:59:01 +00:00
|
|
|
# from: http://stackoverflow.com/questions/2196693/improving-numpy-performance
|
|
|
|
|
# pythran export specialconvolve(uint32 [][])
|
|
|
|
|
# setup: import numpy as np ; r = np.arange(100*10000, dtype=np.uint32).reshape(1000,1000) # noqa
|
|
|
|
|
# run: specialconvolve(r)
|
|
|
|
|
|
2018-04-05 22:07:33 +00:00
|
|
|
|
|
|
|
|
def specialconvolve(a):
|
|
|
|
|
# sorry, you must pad the input yourself
|
2018-10-03 18:59:01 +00:00
|
|
|
rowconvol = a[1:-1, :] + a[:-2, :] + a[2:, :]
|
|
|
|
|
colconvol = rowconvol[:, 1:-1] + rowconvol[:, :-2] + \
|
|
|
|
|
rowconvol[:, 2:] - 9 * a[1:-1, 1:-1]
|
2018-04-05 22:07:33 +00:00
|
|
|
return colconvol
|
