Support optional arguments axis, dtype and out for numpy.cumsum

[yujianll]

Reporting a bug

• [y ] I am using the latest released version of Numba (most recent is visible in
  the change log (https://github.com/numba/numba/blob/master/CHANGE_LOG).
• [y ] I have included below a minimal working reproducer (if you are unsure how
  to write one see http://matthewrocklin.com/blog/work/2018/02/28/minimal-bug-reports).

I somehow cannot use numpy.cumsum in no python mode.
Reproducer:

from numba import jit
import numpy as np

@jit(nopython=True)
def error():
    weights = np.ones((100, 100))
    np.cumsum(weights, 0, float, weights)

error()

[esc]

@yujianll thanks for taking the time to report this and helping to improve Numba. As documented on:

https://numba.pydata.org/numba-doc/dev/reference/numpysupported.html

cumsum in Numba currently only supports the first argument. I.e. none of axis, dtype or out are implemented.

If you try the following it should work:

from numba import jit
import numpy as np

@jit(nopython=True)
def error():
    weights = np.ones((100, 100))
    np.cumsum(weights)

error()

[esc]

I have rename the issues and labeled it as a feature request accordingly.

[esc]

@yujianll if you need axis support immediately, I would suggest to implement it yourself for the given dimensionality in your case. Numba is great at optimizing for loops and a simple implementation should run in reasonable time.

[yujianll]

@esc Thanks! Do you mean a for loop to each column of the array would be fast as well?

[esc]

@yujianll hmmm, not sure, maybe with fortran order that would work well?

[yujianll]

@esc OK, make sense! So convert the array to fortran order (numba supports np.asfortranarray), and then manipulate array on 0 axis?

[yujianll]

@esc I cannot see this. But thanks!

[esc]

@yujianll sorry, wrong link:

#4628

[yujianll]

@esc Thanks!

[maxnoe]

I also just ran into this, support for the axis keyword would be appreciated!

[AhmetCanSolak]

I would also appreciate axis keyword support

[ch3rn0v]

+1 for axis support

[AhmetCanSolak]

I do not know the codebase in depth and how to implement this feature but if there can be some guidance I am more than happy to put work on this. @esc do you have any pointers about where to start?

[trevor-pope]

+1 for support on this

[sliemelela]

+1 for axis support

[YuanJ-i]

+1 for axis support