Convert datetime64 and timedelta64 to integer (in nanoseconds)

Numba Support: How do I do ...?
1

Hi all, is it possible to convert np.datetime64 and np.timedelta64 into integer within Numba?

import numpy as np
import pandas as pd
from numba import njit

@njit
def f(dt):
    return int(dt)

f(pd.Timestamp.now().to_datetime64())
TypingError: Failed in nopython mode pipeline (step: nopython frontend)
No implementation of function Function(<class 'int'>) found for signature:
 
 >>> int(datetime64[ns])
 
There are 2 candidate implementations:
  - Of which 2 did not match due to:
  Overload of function 'int': File: numba/core/typing/builtins.py: Line 925.
    With argument(s): '(datetime64[ns])':
   No match.
2

Edit: Gitter discussion here

This isn’t exactly what you asked, but with some sketchy assumptions (64-bit cython where id() is a pointer to the PyObject) and the great @stuartarchibald’s suggestion here you can dig the raw integer value directly from the PyObject.

I’d guess that this is made more complicated by the fact that we’re working with a scalar object; I think this would likelier be cleaner if you had a numpy array of these, then they’d all be lined up as integers in memory with a well-defined size and base address…

import numpy as np
from numba import njit, carray, int64
from numba.extending import intrinsic

@intrinsic
def address_as_void_pointer(typingctx, src):
    """ returns a void pointer from a given memory address """
    from numba.core import types, cgutils
    sig = types.voidptr(src)

    def codegen(cgctx, builder, sig, args):
        return builder.inttoptr(args[0], cgutils.voidptr_t)

    return sig, codegen

@njit
def f(dt_ptr):
    return carray(address_as_void_pointer(dt_ptr), 3, int64)[2]

dt = np.datetime64('2005-02-25')
print('python says', dt, dt.astype(np.int64), id(dt))
print('numba says', f(id(dt)))

Edit 2: Output is

python says 2005-02-25 12839 2851846899696
numba says 12839