CPU efficient
n Vector quantization by
table lookup
n 16 memory accesses to
find best matching
vector.