Cookbook

We start with importing the CrossPy library, as well as NumPy and CuPy for demostration.

import crosspy as xp
import numpy as np
import cupy as cp

Create arrays

We create a NumPy array and a CuPy and then merge them into a CrossPy array.

x_cpu = np.array([1, 2, 3])
x_cpu

array([1, 2, 3])

x_gpu = cp.array([4, 5])
x_gpu

array([4, 5])

x_gpu.device

<CUDA Device 0>

x_cross = xp.array([x_cpu, x_gpu], axis=0)
x_cross

array([1, 2, 3])@<CPU 0>; array([4, 5])@<CUDA Device 0>

x_cpu.shape

(3,)

x_gpu.shape

(2,)

x_cross.shape

(5,)

x = xp.array([x_gpu, x_gpu], axis=0)
x

array([4, 5])@<CUDA Device 0>; array([4, 5])@<CUDA Device 0>

x.shape

(4,)

x = xp.array([x_gpu, x_gpu], axis=0)
x

array([4, 5])@<CUDA Device 0>; array([4, 5])@<CUDA Device 0>

x.shape

(4,)

# xp.array([x_cpu, x_cpu.T])

Indexing

x = x_cross[0] # slicing with a single integer
x

1

x.shape

()

x = x_cross[[0, 3]] # slicing with a list of integers
x

array([1])@<CPU 0>; array([4])@<CUDA Device 0>

x.shape

(2,)

# x = x_cross[0:2] # slicing with Python built-in slices
# x

x.shape

(2,)

from crosspy import cpu, gpu

# a = xp.array(range(6), distribution=[cpu(0), gpu(0), gpu(1)])
# a

# a.device

# from crosspy import PartitionScheme
# partition = PartitionScheme(6, default_device=cpu(0))

# partition[(0, 4, 5)] = cpu(0)
# partition[1:3] = gpu(0)
# partition[3] = gpu(1)

# a = xp.array(range(6), distribution=partition)
# a

# a.device