Replacement operation not supported by ncnn partial operators

resolvent ： take forward The parameter tensor is segmented in advance as a parameter , Input becomes multiple ,

At the last output, the tensor of each segmentation is calculated separately .

Such as ：forward(self,data)  data_shape:3,3,64,64

        x=data[0]+data[1]+data[2] #(3,64,64)

        return x

It is amended as follows forward(self,data1,data2,data3) #1,3,64,64

                x=data1+data2+data3 #1,3,64,64

                x=x.squeeze(0)#(3,64,64)

                return x

2、repeat I won't support it

When repeating operations on a channel , switch views .

Such as ：forward(self,data)  data_shape:(1,3,64,64)

        x=data.repeat(3,1,1,1)#(3,3,64,64)

        return x

It can be used torch.cat Replace ：

forward(self,data)  data_shape:(1,3,64,64)

        ls=[]

        for i in range(3):# The number of times to repeat

                 ls.append(data)

        x=torch.cat(ls,dim=0)#(3,3,64,64)

        return x

3、view、reshap onnx No, but simplify After that you can

To be continued ..