Numbapro trying to vectorize a jit'd function

Question

I'm trying to understand the semantics behind jit/vectorize in numbapro by trying a few simple algorithms. This is the miller-rabin primality test, which I want to subsequently run multiple times.

This works fine

from numbapro import jit , guvectorize , vectorize, uint64, int32, int8, bool_
import random
@jit(bool_(uint64,int32),target='cpu')
def is_prime(n,k):
    if n % 2 == 0:
        return False
    # n-1 as 2^s * d
    dn = n - 1
    s1 = 0
    while dn % 2 == 0:
        dn = dn >> 1
        s1 += 1
    for j in range(k):
        a1 = random.randint(2,n-2)
        x = pow(a1,dn) % n
        if x == 1 or x == n - 1:
            continue
        for i in range(s1):
            x = (x * x) % n
            if x == 1:
                return False
            if x == n - 1:
                break
    return True

But replacing the decorator with

@vectorize(bool_(uint64,int32),target='cpu')

gives an error

Traceback (most recent call last):
  File "h:\users\mushfaque.cradle\documents\visual studio 2013\Projects\EulerPraxis\EulerPraxis\EulerPraxis.py", line 12, in <module>
    @vectorize(int8(uint64,int32),target='cpu')
  File "H:\Apps\Anaconda3\lib\site-packages\numba\npyufunc\decorators.py", line 67, in wrap
    for fty in ftylist:
TypeError: 'NotImplementedType' object is not callable

I understand that vectorize should be used on ufuncs, but what am I missing to make this a ufunc?

Answer 1

I've solved this.

• @vectorize([int32(uint64,int32)],target='cpu') fixes the first issue. Note the '[]' around the type annotation
• bool_ as a return type is unsupported (bug) which has been reported. In the mean time use int32
• A more subtle issue is that randint throws an exception if you use n = 3 . Numba doesn't handle exceptions in code yet so there should be an extra 'if' to catch that possibility