Closed
Description
Annoying issues, but not hard blockers:
- split Index into Index, IndexMut and IndexAssign
- make Vec implement the slice methods
- Constructing an iterator from a slice doesn't optimise completely
This is a significant performance improvement for any vectors that are grown, and will reduce heap allocation since zero-length vectors do not allocate and it does not have a header.
test bench::old_push ... bench: 43200 ns/iter (+/- 804)
test bench::old_push_preallocated ... bench: 42948 ns/iter (+/- 4043)
test bench::push ... bench: 7190 ns/iter (+/- 245)
test bench::push_preallocated ... bench: 6606 ns/iter (+/- 170)
#[cfg(test)]
extern mod extra;
#[cfg(test)]
mod bench {
use std;
use std::vec_ng::Vec;
use extra::test::BenchHarness;
static size: uint = 10000;
#[bench]
fn push(bh: &mut BenchHarness) {
bh.iter(|| {
let mut xs = Vec::with_capacity(16);
for i in range(0, size) {
xs.push(i);
}
});
}
#[bench]
fn old_push(bh: &mut BenchHarness) {
bh.iter(|| {
let mut xs = std::vec::with_capacity(16);
for i in range(0, size) {
xs.push(i);
}
});
}
#[bench]
fn push_preallocated(bh: &mut BenchHarness) {
bh.iter(|| {
let mut xs = Vec::with_capacity(size);
for i in range(0, size) {
xs.push(i);
}
});
}
#[bench]
fn old_push_preallocated(bh: &mut BenchHarness) {
bh.iter(|| {
let mut xs = std::vec::with_capacity(size);
for i in range(0, size) {
xs.push(i);
}
});
}
}I moved the priority tag here from the old issue, as this is the same core issue but without the distractions of past issues and discussions that are no longer relevant.