Being able to do things like:
# a is some np.ndarray
pinned_a = cuda.pinned_array_like(a)
# As opposed to `my_zs = cuda.pinned_array(10); my_zs[:] = 0`
my_zs = cuda.pinned_zeros(10)
# As opposed to `my_zs = cuda.pinned_array(10); my_zs[:] = 1`
my_ones = cuda.pinned_ones(10)
In general if we could create pinned arrays in more ways similar to how NumPy ndarray objects can be created, this would provide a slightly nicer API to work with.
Being able to do things like:
In general if we could create pinned arrays in more ways similar to how NumPy
ndarrayobjects can be created, this would provide a slightly nicer API to work with.