A couple of scalar optimizations speeding up quantization by about 1.6x. Overall encoder speedup is around 3%.

Change-Id: I19981d1ef0b33e4e5732739574f367fe82771a84