The values are typically 16 bit signed values, ie. centred around zero, so you can't just shift them as it would make the values fly all over the place. You need to multiply them by between 0.0 and 1.0, to do it properly. I don't know much about low-level speed these days, so I'm unsure whether it's faster to convert to a float, perform the multiplication, then convert back, or to do it all in integers (eg. multiply by 100, then divide by a number between 1 and 100). I'm sure there are other tricks with bit manipulation or special CPU instructions but I'd recommend you keep it simple to begin with.