A PackedVec
stores vectors of integers
efficiently while providing an API similar to Vec
. The basic idea is to store
each element using the minimum number of bits needed to represent every element
in the Vec
. For example, if we have a Vec<u64>
with elements [20, 30, 140],
every element wastes most of its 64 bits: 7 bits is sufficient to represent the
range of elements in the vector. Given this input vector, PackedVec
stores
each elements using exactly 7 bits, saving substantial memory. For vectors which
often contain small ranges of numbers, and which are created rarely, but read
from frequently, this can be a significant memory and performance win.
PackedVec
has two main API differences from Vec
: a PackedVec
is created
from a Vec
; and a PackedVec
returns values rather than references. Both
points can be seen in this example:
use packedvec::PackedVec;
let v = vec![-1, 30, 120];
let pv = PackedVec::new(v.clone());
assert_eq!(pv.get(0), Some(-1));
assert_eq!(pv.get(2), Some(120));
assert_eq!(pv.get(3), None);
assert_eq!(v.iter().cloned().collect::<Vec<_>>(), pv.iter().collect::<Vec<_>>());