how can I specify the memory address of a Numpy array using ctypes?

Question

I would like to construct a Numpy array that starts at a specific memory address. How would I do that? I assume the solution involves ctypes but I can't figure it out from the docs.

More Details and Context

I would like to create a number of arrays whose values are all at sequential memory addresses. For example, two size 2 arrays, where the 1st mem address of the 1st array is N and the last mem address of the 2nd array is N + 3. I could achieve the same affect by making one size 4 array and slicing it... but I want to call the array constructor a separate time for each "slice." I am trying to track a bug in a certain C library.

I don't like asking "What have you tried so far?", but still people tend me to do so. — Naman
– Naman, Commented Dec 19, 2016 at 12:54
np.ndarray can make an array with an existing data buffer. — hpaulj
– hpaulj, Commented Dec 19, 2016 at 14:09
"but I want to call the array constructor a separate time for each "slice."" - this is exactly what slicing will do anyway, other than the sliced version will have .base is not None — Eric
– Eric, Commented Dec 19, 2016 at 21:30

hpaulj · Accepted Answer · 2016-12-20 01:53:32Z

4

Make an array of 10 bytes:

In [287]: x = np.arange(10, dtype=np.uint8)
In [288]: x
Out[288]: array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9], dtype=uint8)
In [289]: x.__array_interface__
Out[289]: 
{'data': (155596184, False),
 'descr': [('', '|u1')],
 'shape': (10,),
 'strides': None,
 'typestr': '|u1',
 'version': 3}
In [290]: x.data
Out[290]: <memory at 0xaec88f34>

Make another array with ndarray, using the same data buffer, but with an offset:

In [291]: y=np.ndarray(shape=(3,), dtype=x.dtype, buffer=x.data, offset=3)
In [292]: y
Out[292]: array([3, 4, 5], dtype=uint8)
In [293]: y.__array_interface__
Out[293]: 
{'data': (155596187, False),
 'descr': [('', '|u1')],
 'shape': (3,),
 'strides': None,
 'typestr': '|u1',
 'version': 3}

this looks the same as a 3 element slice:

In [294]: z=x[3:6]
In [295]: z
Out[295]: array([3, 4, 5], dtype=uint8)
In [296]: z.__array_interface__
Out[296]: 
{'data': (155596187, False),
 'descr': [('', '|u1')],
 'shape': (3,),
 'strides': None,
 'typestr': '|u1',
 'version': 3}

If I knew more about specifying a memory address I probably could get by with just the buffer parameter, and default 0 offset. But from your additions, the use of offset might be just what you want.

edited Dec 20, 2016 at 1:53

answered Dec 20, 2016 at 1:46

hpaulj

233k14 gold badges260 silver badges392 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

Trekkie Over a year ago

Thank you for responding. This looks promising, and I will play with it tomorrow at work. Ideally, I would like to be able to specify an array to start a particular pointer... so for instance the first memory address after the tail end of the first array..I think I can work with this. I'm really trying to avoid the weeds of going into the C library I'm dealing with, but it may be unavoidable. Here is a link to the issue I'm trying to trace in PyFFTW. I posted this example with a script that reproduces the error. Maybe you might want to look at it. :) github.com/pyFFTW/pyFFTW/issues/139

Trekkie Over a year ago

PyFFTW is a Python wrapper around the FFTW Fast Fourier Transform library.

Collectives™ on Stack Overflow

how can I specify the memory address of a Numpy array using ctypes?

1 Answer 1

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related