Decoration of function

OrchesPy provides a decorator orchespy.decorator(). orchespy.device() enables to decorate a function to execute on a device you want.

• Example 1:

>>> from orchespy import device
>>> from orchespy.devicetype import CUDAGPU
>>> import numpy as np
>>>
>>> @device(CUDAGPU)
... def exec_on_gpu(x, y):
...     return x * y
>>>
>>> x = np.linspace(1, 5, 5)
>>> y = np.linspace(1, 5, 5)
>>> z = exec_on_gpu(x, y)
>>> type(z)   
<class 'cupy.core.core.ndarray'>   # ndarray of cupy

x and y are numpy.ndarray. The function exec_on_gpu() will execute on CUDA GPU. You don’t need to convert x and y to cupy.ndarray. OrchesPy will automatically convert them to the target ndarray type.

Note

Since the function exec_on_gpu() is executed on CUDA GPU, an ndarray of CuPy is returned. The data of ndarray is transferred automatically to CUDA GPU.

• Example 2:

>>> from orchespy import device
>>> from orchespy.devicetype import VE
>>>
>>> @device(VE, numpy_module_arg='xp')
... def create_array_on_VE(xp):
...     return xp.random.rand(2, 2)
>>>
>>> x = create_array_on_VE()
>>> type(x)   
<class 'nlcpy.core.core.ndarray'>   # ndarray of nlcpy

You can pass the package of NumPy-like library for a device to a decorated function. In Example 2, the argument xp is set to nlcpy. nlcpy.random.rand(), the array creation function in NLCPy, is called at the line xp.random.rand().

• Example 3:

>>> from orchespy import device
>>> import numpy as np
>>> import nlcpy as vp
>>> @device(VE(1))
... def exec_on_ve(x, y):
...     return x * y
...
>>> x = np.ones((2,2))
>>> y = np.ones((2,2))
>>> vp.venode.VE(1).use()
<VE node logical_id=1, physical_id=1>
>>> mul = exec_on_ve(x, y)
>>> mul.venode.id
1

If you have multiple identical devices, you can specify any device for the decorator. Also, when operating within the device decorator function, It is necessary to switch the output device in advance.