numba vectorize returning list/array

[cristipurdel]

Feature request

Reporting a bug

• 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).
• 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).

[cristipurdel]

If my code is wrong, please correct it, otherwise make a request
I need this option not to run multiple times the vectorize function
Could not find any tutorials on how to do this, hopefully is also 'cuda' future proof
Using the latest numba 0.38

Found this linked and will cross-link
Can this be as feature implemented or it does not make sense?
https://groups.google.com/a/continuum.io/forum/#!msg/numba-users/4B2elGzEFw8/XfSMDZLuAgAJ

import numpy as np
import time
import pdb
from numba import vectorize

@vectorize(['float32(float32, float32)'], target='parallel')
def VectorPowv(a, b):
    x = a ** b
    y = a * b
    out = []
    out.append(x)
    out.append(y)
    return out

def main():  
    N = 20000000
    A = B = np.array(np.random.sample(N), dtype=np.float32)
    A = A[:,np.newaxis]
    B = B[:,np.newaxis]    
    pdb.set_trace() 
    start = time.time()
    C = VectorPowv(A, B)
    print("C[:5] = " + str(C[:5]))
    print("C[-5:] = " + str(C[-5:]))
    vector_add_time = time.time() - start    
    print("VectorPow took for % seconds" % vector_add_time)

if __name__=='__main__':
    main()
    

Traceback

dtypenums.append(np.dtype(signature.return_type.name).num)

TypeError: data type "pyobject" not understood

[stuartarchibald]

Thanks for the report. What is mentioned on the mailing list is correct. If you want to mutate a number of arrays at once then perhaps try a gufunc via @guvectorize? http://numba.pydata.org/numba-doc/dev/user/vectorize.html#the-guvectorize-decorator, or just use a @jit function if you don't need all the ufunc mechanics.

[cristipurdel]

It is working with guvectorize, but I would have wanted this to work with vectorize.
I find @jit a bit difficult to work