Auto vectorizing optimizers have gotten quite good. If you are using integers it often just happens whether you think about it or not. With floats unless you specify fast math you will need to use wide types to let it know you don't care about floating point addition order.