albop
commented
5 years ago Thank you! This seems to be numba issue. I have made a smaller example here : (btw it's easier to replicate if you copy paste code instead of image)
from numba import jit, typeof, generated_jit
import numpy as np
sig = numba.typeof(grid)
x1 = np.linspace(0,1,100)
x2 = np.linspace(0,1,100)
grid = (x1,x2)
# this works:
@jit(nopython=True)
def dosomething(x):
return x + x1.sum()
print( dosomething(0.1) ) # -returns: 50.1
x1[0] += 10
print( dosomething(0.1) ) # still returns: 50.1
# (a frozen copy of x1 is made in the njitted space)
# this doesn't work: it looks like there is an issue with freezing a tuple of array
@jit(nopython=True)
def dosomething(x):
return x + gg[0].sum()
dosomething(0.2) # fails with horrible error message
# it works with basic tuples
basic_tuple = (0.1, 0.2)
@jit(nopython=True)
def dosomething(x):
return x + basic_tuple[1]
dosomething(0.3)
# it doesn't work with lists either
basic_tuple = [x1, x2]
@jit(nopython=True)
def dosomething(x):
return x + basic_tuple[0].sum()
natashawatkins
Seems that
mlinterphas problems when being jitted and passed global variables...Interestingly, moving
grid = (x1, x2)inside works, wherex1andx2are global variablesAlso had an issue with trying to jit a function inside a function with a class argument...