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