T in (tensor(float16),  tensor(float)tensor(double)tensor(bfloat16)) : Constrain input and output types to float tensors.