random number generators perform unnecessary copy

[braingram]

Taking generate as an example:

GalSim/galsim/random.py

Lines 265 to 271 in ee177aa

	array_1d = np.ascontiguousarray(array.ravel(),dtype=float)
	#assert(array_1d.strides[0] == array_1d.itemsize)
	_a = array_1d.__array_interface__['data'][0]
	self._rng.generate(len(array_1d), _a)
	if array_1d.data != array.data:
	# array_1d is not a view into the original array. Need to copy back.
	np.copyto(array, array_1d.reshape(array.shape), casting='unsafe')

The result of ascontinguousarray can share memory with the input array yet the if array_1d.data != array.data: check can fail resulting in an unnecessary copy.

The issue can be illustrated in the following code:

import numpy as np
array = np.empty((10, 20, 30), dtype=float)
array_1d = np.ascontiguousarray(array.ravel(),dtype=float)

# data is not the same
assert array.data != array_1d.data

# because array_1d is a view of array
assert array.data == array_1d.base.data  # note the use of base here

# arrays share memory
assert np.shares_memory(array, array_1d)

# double check
array_1d[:] = 2
assert np.all(array == 2)

Replacing if array_1d.data != array.data: with if not np.shares_memory(array, array_1d): should fix the issue.

[rmjarvis]

Sure. I'm not sure it's an "issue" exactly. But I'll admit the change would be marginally more efficient. Feel free to submit a PR.