import mobula

@mobula.op.register
class MyFirstOP:
    def forward(self, x, y):
        return x + y
    def backward(self, dy): 
        return [dy, dy]
    def infer_shape(self, in_shape):
        assert in_shape[0] == in_shape[1]
        return in_shape, [in_shape[0]]

# MXNet
import mxnet as mx
a = mx.nd.array([1, 2, 3])
b = mx.nd.array([4, 5, 6])
c = MyFirstOP(a, b)
print (c) # [5, 7, 9]

# PyTorch
import torch
a = torch.tensor([1, 2, 3])
b = torch.tensor([4, 5, 6])
c = MyFirstOP(a, b)
print (c) # [5, 7, 9]

# NumPy
import numpy as np
a = np.array([1, 2, 3])
b = np.array([4, 5, 6])
op = MyFirstOP[np.ndarray]()
c = op(a, b)
print (c) # [5, 7, 9]

# CuPy
import cupy as cp
a = cp.array([1, 2, 3])
b = cp.array([4, 5, 6])
op = MyFirstOP[cp.ndarray]()
c = op(a, b)
print(c) # [5, 7, 9]

# Use ROIAlign operator
import mxnet as mx
import numpy as np
import mobula

# Load ROIAlign Module
mobula.op.load('ROIAlign')

ctx = mx.cpu(0)
dtype = np.float32
N, C, H, W = 2, 3, 4, 4

data = mx.nd.array(np.arange(N*C*H*W).astype(dtype).reshape((N,C,H,W)))
rois = mx.nd.array(np.array([[0, 1, 1, 3, 3]], dtype = dtype))

data.attach_grad()
with mx.autograd.record():
    # mx.nd.NDArray and mx.sym.Symbol are both available as the inputs.
    output = mobula.op.ROIAlign(data = data, rois = rois, pooled_size = (2,2), spatial_scale = 1.0, sampling_ratio = 1)

print (output.asnumpy(), data.grad.asnumpy())

import mobula
# Import Custom Operator Dynamically
mobula.op.load('./AdditionOP')

import mxnet as mx
a = mx.nd.array([1,2,3])
b = mx.nd.array([4,5,6])
c = mobula.op.AdditionOP(a, b)

print ('a + b = c \n {} + {} = {}'.format(a.asnumpy(), b.asnumpy(), c.asnumpy()))