Currently float are converted to 16b uint in input part
using src depth (32 bits) in hScale16To19 and hScale16to15,
make an invalid shift for the data

So shift the value when using float input
like 16 bpc uint.