Auto merge of #40601 - stjepang:sort-unstable, r=alexcrichton
Implement feature sort_unstable Tracking issue for the feature: #40585 This is essentially integration of [pdqsort](https://github.com/stjepang/pdqsort) into libcore. There's plenty of unsafe blocks to review. The heart of pdqsort is `fn partition_in_blocks` and is probably the most challenging function to understand. It requires some patience, but let me know if you find it too difficult - comments could always be improved. #### Changes * Added `sort_unstable` feature. * Tweaked insertion sort constants for stable sort. Sorting integers is now up to 5% slower, but sorting big elements is much faster (in particular, `sort_large_big_random` is 35% faster). The old constants were highly optimized for sorting integers, so overall the configuration is more balanced now. A minor regression in case of integers is forgivable as we recently had performance improvements (#39538) that completely make up for it. * Removed some uninteresting sort benchmarks. * Added a new sort benchmark for string sorting. #### Benchmarks The following table compares stable and unstable sorting: ``` name stable ns/iter unstable ns/iter diff ns/iter diff % slice::sort_large_ascending 7,240 (11049 MB/s) 7,380 (10840 MB/s) 140 1.93% slice::sort_large_big_random 1,454,138 (880 MB/s) 910,269 (1406 MB/s) -543,869 -37.40% slice::sort_large_descending 13,450 (5947 MB/s) 10,895 (7342 MB/s) -2,555 -19.00% slice::sort_large_mostly_ascending 204,041 (392 MB/s) 88,639 (902 MB/s) -115,402 -56.56% slice::sort_large_mostly_descending 217,109 (368 MB/s) 99,009 (808 MB/s) -118,100 -54.40% slice::sort_large_random 477,257 (167 MB/s) 346,028 (231 MB/s) -131,229 -27.50% slice::sort_large_random_expensive 21,670,537 (3 MB/s) 22,710,238 (3 MB/s) 1,039,701 4.80% slice::sort_large_strings 6,284,499 (38 MB/s) 6,410,896 (37 MB/s) 126,397 2.01% slice::sort_medium_random 3,515 (227 MB/s) 3,327 (240 MB/s) -188 -5.35% slice::sort_small_ascending 42 (1904 MB/s) 41 (1951 MB/s) -1 -2.38% slice::sort_small_big_random 503 (2544 MB/s) 514 (2490 MB/s) 11 2.19% slice::sort_small_descending 72 (1111 MB/s) 69 (1159 MB/s) -3 -4.17% slice::sort_small_random 369 (216 MB/s) 367 (217 MB/s) -2 -0.54% ``` Interesting cases: * Expensive comparison function and string sorting - it's a really close race, but timsort performs a slightly smaller number of comparisons. This is a natural difference of bottom-up merging versus top-down partitioning. * `large_descending` - unstable sort is faster, but both sorts should have equivalent performance. Both just check whether the slice is descending and if so, they reverse it. I blame LLVM for the discrepancy. r? @alexcrichton
This commit is contained in:
bors
commit
cab4bff3de
@ -11,6 +11,7 @@
|
||||
#![deny(warnings)]
|
||||
|
||||
#![feature(rand)]
|
||||
#![feature(sort_unstable)]
|
||||
#![feature(test)]
|
||||
|
||||
extern crate test;
|
||||
|
||||
@ -169,6 +169,7 @@ fn random_inserts(b: &mut Bencher) {
|
||||
}
|
||||
})
|
||||
}
|
||||
|
||||
#[bench]
|
||||
fn random_removes(b: &mut Bencher) {
|
||||
let mut rng = thread_rng();
|
||||
@ -216,65 +217,76 @@ fn gen_mostly_descending(len: usize) -> Vec<u64> {
|
||||
v
|
||||
}
|
||||
|
||||
fn gen_strings(len: usize) -> Vec<String> {
|
||||
let mut rng = thread_rng();
|
||||
let mut v = vec![];
|
||||
for _ in 0..len {
|
||||
let n = rng.gen::<usize>() % 20 + 1;
|
||||
v.push(rng.gen_ascii_chars().take(n).collect());
|
||||
}
|
||||
v
|
||||
}
|
||||
|
||||
fn gen_big_random(len: usize) -> Vec<[u64; 16]> {
|
||||
let mut rng = thread_rng();
|
||||
rng.gen_iter().map(|x| [x; 16]).take(len).collect()
|
||||
}
|
||||
|
||||
fn gen_big_ascending(len: usize) -> Vec<[u64; 16]> {
|
||||
(0..len as u64).map(|x| [x; 16]).take(len).collect()
|
||||
}
|
||||
|
||||
fn gen_big_descending(len: usize) -> Vec<[u64; 16]> {
|
||||
(0..len as u64).rev().map(|x| [x; 16]).take(len).collect()
|
||||
}
|
||||
|
||||
macro_rules! sort_bench {
|
||||
($name:ident, $gen:expr, $len:expr) => {
|
||||
macro_rules! sort {
|
||||
($f:ident, $name:ident, $gen:expr, $len:expr) => {
|
||||
#[bench]
|
||||
fn $name(b: &mut Bencher) {
|
||||
b.iter(|| $gen($len).sort());
|
||||
b.iter(|| $gen($len).$f());
|
||||
b.bytes = $len * mem::size_of_val(&$gen(1)[0]) as u64;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
sort_bench!(sort_small_random, gen_random, 10);
|
||||
sort_bench!(sort_small_ascending, gen_ascending, 10);
|
||||
sort_bench!(sort_small_descending, gen_descending, 10);
|
||||
macro_rules! sort_expensive {
|
||||
($f:ident, $name:ident, $gen:expr, $len:expr) => {
|
||||
#[bench]
|
||||
fn $name(b: &mut Bencher) {
|
||||
b.iter(|| {
|
||||
let mut v = $gen($len);
|
||||
let mut count = 0;
|
||||
v.$f(|a: &u64, b: &u64| {
|
||||
count += 1;
|
||||
if count % 1_000_000_000 == 0 {
|
||||
panic!("should not happen");
|
||||
}
|
||||
(*a as f64).cos().partial_cmp(&(*b as f64).cos()).unwrap()
|
||||
});
|
||||
black_box(count);
|
||||
});
|
||||
b.bytes = $len as u64 * mem::size_of::<u64>() as u64;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
sort_bench!(sort_small_big_random, gen_big_random, 10);
|
||||
sort_bench!(sort_small_big_ascending, gen_big_ascending, 10);
|
||||
sort_bench!(sort_small_big_descending, gen_big_descending, 10);
|
||||
sort!(sort, sort_small_ascending, gen_ascending, 10);
|
||||
sort!(sort, sort_small_descending, gen_descending, 10);
|
||||
sort!(sort, sort_small_random, gen_random, 10);
|
||||
sort!(sort, sort_small_big_random, gen_big_random, 10);
|
||||
sort!(sort, sort_medium_random, gen_random, 100);
|
||||
sort!(sort, sort_large_ascending, gen_ascending, 10000);
|
||||
sort!(sort, sort_large_descending, gen_descending, 10000);
|
||||
sort!(sort, sort_large_mostly_ascending, gen_mostly_ascending, 10000);
|
||||
sort!(sort, sort_large_mostly_descending, gen_mostly_descending, 10000);
|
||||
sort!(sort, sort_large_random, gen_random, 10000);
|
||||
sort!(sort, sort_large_big_random, gen_big_random, 10000);
|
||||
sort!(sort, sort_large_strings, gen_strings, 10000);
|
||||
sort_expensive!(sort_by, sort_large_random_expensive, gen_random, 10000);
|
||||
|
||||
sort_bench!(sort_medium_random, gen_random, 100);
|
||||
sort_bench!(sort_medium_ascending, gen_ascending, 100);
|
||||
sort_bench!(sort_medium_descending, gen_descending, 100);
|
||||
|
||||
sort_bench!(sort_large_random, gen_random, 10000);
|
||||
sort_bench!(sort_large_ascending, gen_ascending, 10000);
|
||||
sort_bench!(sort_large_descending, gen_descending, 10000);
|
||||
sort_bench!(sort_large_mostly_ascending, gen_mostly_ascending, 10000);
|
||||
sort_bench!(sort_large_mostly_descending, gen_mostly_descending, 10000);
|
||||
|
||||
sort_bench!(sort_large_big_random, gen_big_random, 10000);
|
||||
sort_bench!(sort_large_big_ascending, gen_big_ascending, 10000);
|
||||
sort_bench!(sort_large_big_descending, gen_big_descending, 10000);
|
||||
|
||||
#[bench]
|
||||
fn sort_large_random_expensive(b: &mut Bencher) {
|
||||
let len = 10000;
|
||||
b.iter(|| {
|
||||
let mut v = gen_random(len);
|
||||
let mut count = 0;
|
||||
v.sort_by(|a: &u64, b: &u64| {
|
||||
count += 1;
|
||||
if count % 1_000_000_000 == 0 {
|
||||
panic!("should not happen");
|
||||
}
|
||||
(*a as f64).cos().partial_cmp(&(*b as f64).cos()).unwrap()
|
||||
});
|
||||
black_box(count);
|
||||
});
|
||||
b.bytes = len as u64 * mem::size_of::<u64>() as u64;
|
||||
}
|
||||
sort!(sort_unstable, sort_unstable_small_ascending, gen_ascending, 10);
|
||||
sort!(sort_unstable, sort_unstable_small_descending, gen_descending, 10);
|
||||
sort!(sort_unstable, sort_unstable_small_random, gen_random, 10);
|
||||
sort!(sort_unstable, sort_unstable_small_big_random, gen_big_random, 10);
|
||||
sort!(sort_unstable, sort_unstable_medium_random, gen_random, 100);
|
||||
sort!(sort_unstable, sort_unstable_large_ascending, gen_ascending, 10000);
|
||||
sort!(sort_unstable, sort_unstable_large_descending, gen_descending, 10000);
|
||||
sort!(sort_unstable, sort_unstable_large_mostly_ascending, gen_mostly_ascending, 10000);
|
||||
sort!(sort_unstable, sort_unstable_large_mostly_descending, gen_mostly_descending, 10000);
|
||||
sort!(sort_unstable, sort_unstable_large_random, gen_random, 10000);
|
||||
sort!(sort_unstable, sort_unstable_large_big_random, gen_big_random, 10000);
|
||||
sort!(sort_unstable, sort_unstable_large_strings, gen_strings, 10000);
|
||||
sort_expensive!(sort_unstable_by, sort_unstable_large_random_expensive, gen_random, 10000);
|
||||
|
||||
@ -52,6 +52,7 @@
|
||||
#![feature(shared)]
|
||||
#![feature(slice_get_slice)]
|
||||
#![feature(slice_patterns)]
|
||||
#![cfg_attr(not(test), feature(sort_unstable))]
|
||||
#![feature(specialization)]
|
||||
#![feature(staged_api)]
|
||||
#![feature(str_internals)]
|
||||
|
||||
@ -1092,36 +1092,6 @@ impl<T> [T] {
|
||||
merge_sort(self, |a, b| a.lt(b));
|
||||
}
|
||||
|
||||
/// Sorts the slice using `f` to extract a key to compare elements by.
|
||||
///
|
||||
/// This sort is stable (i.e. does not reorder equal elements) and `O(n log n)` worst-case.
|
||||
///
|
||||
/// # Current implementation
|
||||
///
|
||||
/// The current algorithm is an adaptive, iterative merge sort inspired by
|
||||
/// [timsort](https://en.wikipedia.org/wiki/Timsort).
|
||||
/// It is designed to be very fast in cases where the slice is nearly sorted, or consists of
|
||||
/// two or more sorted sequences concatenated one after another.
|
||||
///
|
||||
/// Also, it allocates temporary storage half the size of `self`, but for short slices a
|
||||
/// non-allocating insertion sort is used instead.
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// let mut v = [-5i32, 4, 1, -3, 2];
|
||||
///
|
||||
/// v.sort_by_key(|k| k.abs());
|
||||
/// assert!(v == [1, 2, -3, 4, -5]);
|
||||
/// ```
|
||||
#[stable(feature = "slice_sort_by_key", since = "1.7.0")]
|
||||
#[inline]
|
||||
pub fn sort_by_key<B, F>(&mut self, mut f: F)
|
||||
where F: FnMut(&T) -> B, B: Ord
|
||||
{
|
||||
merge_sort(self, |a, b| f(a).lt(&f(b)));
|
||||
}
|
||||
|
||||
/// Sorts the slice using `compare` to compare elements.
|
||||
///
|
||||
/// This sort is stable (i.e. does not reorder equal elements) and `O(n log n)` worst-case.
|
||||
@ -1155,6 +1125,150 @@ impl<T> [T] {
|
||||
merge_sort(self, |a, b| compare(a, b) == Less);
|
||||
}
|
||||
|
||||
/// Sorts the slice using `f` to extract a key to compare elements by.
|
||||
///
|
||||
/// This sort is stable (i.e. does not reorder equal elements) and `O(n log n)` worst-case.
|
||||
///
|
||||
/// # Current implementation
|
||||
///
|
||||
/// The current algorithm is an adaptive, iterative merge sort inspired by
|
||||
/// [timsort](https://en.wikipedia.org/wiki/Timsort).
|
||||
/// It is designed to be very fast in cases where the slice is nearly sorted, or consists of
|
||||
/// two or more sorted sequences concatenated one after another.
|
||||
///
|
||||
/// Also, it allocates temporary storage half the size of `self`, but for short slices a
|
||||
/// non-allocating insertion sort is used instead.
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// let mut v = [-5i32, 4, 1, -3, 2];
|
||||
///
|
||||
/// v.sort_by_key(|k| k.abs());
|
||||
/// assert!(v == [1, 2, -3, 4, -5]);
|
||||
/// ```
|
||||
#[stable(feature = "slice_sort_by_key", since = "1.7.0")]
|
||||
#[inline]
|
||||
pub fn sort_by_key<B, F>(&mut self, mut f: F)
|
||||
where F: FnMut(&T) -> B, B: Ord
|
||||
{
|
||||
merge_sort(self, |a, b| f(a).lt(&f(b)));
|
||||
}
|
||||
|
||||
/// Sorts the slice, but may not preserve the order of equal elements.
|
||||
///
|
||||
/// This sort is unstable (i.e. may reorder equal elements), in-place (i.e. does not allocate),
|
||||
/// and `O(n log n)` worst-case.
|
||||
///
|
||||
/// # Current implementation
|
||||
///
|
||||
/// The current algorithm is based on Orson Peters' [pdqsort][pattern-defeating quicksort],
|
||||
/// which is a quicksort variant designed to be very fast on certain kinds of patterns,
|
||||
/// sometimes achieving linear time. It is randomized but deterministic, and falls back to
|
||||
/// heapsort on degenerate inputs.
|
||||
///
|
||||
/// It is generally faster than stable sorting, except in a few special cases, e.g. when the
|
||||
/// slice consists of several concatenated sorted sequences.
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(sort_unstable)]
|
||||
///
|
||||
/// let mut v = [-5, 4, 1, -3, 2];
|
||||
///
|
||||
/// v.sort_unstable();
|
||||
/// assert!(v == [-5, -3, 1, 2, 4]);
|
||||
/// ```
|
||||
///
|
||||
/// [pdqsort]: https://github.com/orlp/pdqsort
|
||||
// FIXME #40585: Mention `sort_unstable` in the documentation for `sort`.
|
||||
#[unstable(feature = "sort_unstable", issue = "40585")]
|
||||
#[inline]
|
||||
pub fn sort_unstable(&mut self)
|
||||
where T: Ord
|
||||
{
|
||||
core_slice::SliceExt::sort_unstable(self);
|
||||
}
|
||||
|
||||
/// Sorts the slice using `compare` to compare elements, but may not preserve the order of
|
||||
/// equal elements.
|
||||
///
|
||||
/// This sort is unstable (i.e. may reorder equal elements), in-place (i.e. does not allocate),
|
||||
/// and `O(n log n)` worst-case.
|
||||
///
|
||||
/// # Current implementation
|
||||
///
|
||||
/// The current algorithm is based on Orson Peters' [pdqsort][pattern-defeating quicksort],
|
||||
/// which is a quicksort variant designed to be very fast on certain kinds of patterns,
|
||||
/// sometimes achieving linear time. It is randomized but deterministic, and falls back to
|
||||
/// heapsort on degenerate inputs.
|
||||
///
|
||||
/// It is generally faster than stable sorting, except in a few special cases, e.g. when the
|
||||
/// slice consists of several concatenated sorted sequences.
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(sort_unstable)]
|
||||
///
|
||||
/// let mut v = [5, 4, 1, 3, 2];
|
||||
/// v.sort_unstable_by(|a, b| a.cmp(b));
|
||||
/// assert!(v == [1, 2, 3, 4, 5]);
|
||||
///
|
||||
/// // reverse sorting
|
||||
/// v.sort_unstable_by(|a, b| b.cmp(a));
|
||||
/// assert!(v == [5, 4, 3, 2, 1]);
|
||||
/// ```
|
||||
///
|
||||
/// [pdqsort]: https://github.com/orlp/pdqsort
|
||||
// FIXME #40585: Mention `sort_unstable_by` in the documentation for `sort_by`.
|
||||
#[unstable(feature = "sort_unstable", issue = "40585")]
|
||||
#[inline]
|
||||
pub fn sort_unstable_by<F>(&mut self, compare: F)
|
||||
where F: FnMut(&T, &T) -> Ordering
|
||||
{
|
||||
core_slice::SliceExt::sort_unstable_by(self, compare);
|
||||
}
|
||||
|
||||
/// Sorts the slice using `f` to extract a key to compare elements by, but may not preserve the
|
||||
/// order of equal elements.
|
||||
///
|
||||
/// This sort is unstable (i.e. may reorder equal elements), in-place (i.e. does not allocate),
|
||||
/// and `O(n log n)` worst-case.
|
||||
///
|
||||
/// # Current implementation
|
||||
///
|
||||
/// The current algorithm is based on Orson Peters' [pdqsort][pattern-defeating quicksort],
|
||||
/// which is a quicksort variant designed to be very fast on certain kinds of patterns,
|
||||
/// sometimes achieving linear time. It is randomized but deterministic, and falls back to
|
||||
/// heapsort on degenerate inputs.
|
||||
///
|
||||
/// It is generally faster than stable sorting, except in a few special cases, e.g. when the
|
||||
/// slice consists of several concatenated sorted sequences.
|
||||
///
|
||||
/// # Examples
|
||||
///
|
||||
/// ```
|
||||
/// #![feature(sort_unstable)]
|
||||
///
|
||||
/// let mut v = [-5i32, 4, 1, -3, 2];
|
||||
///
|
||||
/// v.sort_unstable_by_key(|k| k.abs());
|
||||
/// assert!(v == [1, 2, -3, 4, -5]);
|
||||
/// ```
|
||||
///
|
||||
/// [pdqsort]: https://github.com/orlp/pdqsort
|
||||
// FIXME #40585: Mention `sort_unstable_by_key` in the documentation for `sort_by_key`.
|
||||
#[unstable(feature = "sort_unstable", issue = "40585")]
|
||||
#[inline]
|
||||
pub fn sort_unstable_by_key<B, F>(&mut self, f: F)
|
||||
where F: FnMut(&T) -> B,
|
||||
B: Ord
|
||||
{
|
||||
core_slice::SliceExt::sort_unstable_by_key(self, f);
|
||||
}
|
||||
|
||||
/// Copies the elements from `src` into `self`.
|
||||
///
|
||||
/// The length of `src` must be the same as `self`.
|
||||
@ -1553,28 +1667,20 @@ unsafe fn merge<T, F>(v: &mut [T], mid: usize, buf: *mut T, is_less: &mut F)
|
||||
fn merge_sort<T, F>(v: &mut [T], mut is_less: F)
|
||||
where F: FnMut(&T, &T) -> bool
|
||||
{
|
||||
// Slices of up to this length get sorted using insertion sort.
|
||||
const MAX_INSERTION: usize = 20;
|
||||
// Very short runs are extended using insertion sort to span at least this many elements.
|
||||
const MIN_RUN: usize = 10;
|
||||
|
||||
// Sorting has no meaningful behavior on zero-sized types.
|
||||
if size_of::<T>() == 0 {
|
||||
return;
|
||||
}
|
||||
|
||||
// FIXME #12092: These numbers are platform-specific and need more extensive testing/tuning.
|
||||
//
|
||||
// If `v` has length up to `max_insertion`, simply switch to insertion sort because it is going
|
||||
// to perform better than merge sort. For bigger types `T`, the threshold is smaller.
|
||||
//
|
||||
// Short runs are extended using insertion sort to span at least `min_run` elements, in order
|
||||
// to improve performance.
|
||||
let (max_insertion, min_run) = if size_of::<T>() <= 2 * mem::size_of::<usize>() {
|
||||
(64, 32)
|
||||
} else {
|
||||
(32, 16)
|
||||
};
|
||||
|
||||
let len = v.len();
|
||||
|
||||
// Short arrays get sorted in-place via insertion sort to avoid allocations.
|
||||
if len <= max_insertion {
|
||||
if len <= MAX_INSERTION {
|
||||
if len >= 2 {
|
||||
for i in (0..len-1).rev() {
|
||||
insert_head(&mut v[i..], &mut is_less);
|
||||
@ -1618,7 +1724,7 @@ fn merge_sort<T, F>(v: &mut [T], mut is_less: F)
|
||||
|
||||
// Insert some more elements into the run if it's too short. Insertion sort is faster than
|
||||
// merge sort on short sequences, so this significantly improves performance.
|
||||
while start > 0 && end - start < min_run {
|
||||
while start > 0 && end - start < MIN_RUN {
|
||||
start -= 1;
|
||||
insert_head(&mut v[start..end], &mut is_less);
|
||||
}
|
||||
|
||||
@ -399,9 +399,10 @@ fn test_sort() {
|
||||
}
|
||||
}
|
||||
|
||||
// shouldn't panic
|
||||
let mut v: [i32; 0] = [];
|
||||
v.sort();
|
||||
// Should not panic.
|
||||
[0i32; 0].sort();
|
||||
[(); 10].sort();
|
||||
[(); 100].sort();
|
||||
|
||||
let mut v = [0xDEADBEEFu64];
|
||||
v.sort();
|
||||
@ -441,13 +442,6 @@ fn test_sort_stability() {
|
||||
}
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_sort_zero_sized_type() {
|
||||
// Should not panic.
|
||||
[(); 10].sort();
|
||||
[(); 100].sort();
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_concat() {
|
||||
let v: [Vec<i32>; 0] = [];
|
||||
|
||||
@ -71,26 +71,27 @@
|
||||
#![feature(asm)]
|
||||
#![feature(associated_type_defaults)]
|
||||
#![feature(cfg_target_feature)]
|
||||
#![feature(cfg_target_has_atomic)]
|
||||
#![feature(concat_idents)]
|
||||
#![feature(const_fn)]
|
||||
#![feature(cfg_target_has_atomic)]
|
||||
#![feature(custom_attribute)]
|
||||
#![feature(fundamental)]
|
||||
#![feature(i128_type)]
|
||||
#![feature(inclusive_range_syntax)]
|
||||
#![feature(intrinsics)]
|
||||
#![feature(lang_items)]
|
||||
#![feature(never_type)]
|
||||
#![feature(no_core)]
|
||||
#![feature(on_unimplemented)]
|
||||
#![feature(optin_builtin_traits)]
|
||||
#![feature(unwind_attributes)]
|
||||
#![feature(prelude_import)]
|
||||
#![feature(repr_simd, platform_intrinsics)]
|
||||
#![feature(rustc_attrs)]
|
||||
#![feature(specialization)]
|
||||
#![feature(staged_api)]
|
||||
#![feature(unboxed_closures)]
|
||||
#![feature(never_type)]
|
||||
#![feature(i128_type)]
|
||||
#![feature(prelude_import)]
|
||||
#![feature(untagged_unions)]
|
||||
#![feature(unwind_attributes)]
|
||||
|
||||
#[prelude_import]
|
||||
#[allow(unused)]
|
||||
|
||||
@ -1,4 +1,4 @@
|
||||
// Copyright 2012-2014 The Rust Project Developers. See the COPYRIGHT
|
||||
// Copyright 2012-2017 The Rust Project Developers. See the COPYRIGHT
|
||||
// file at the top-level directory of this distribution and at
|
||||
// http://rust-lang.org/COPYRIGHT.
|
||||
//
|
||||
@ -51,6 +51,8 @@ use mem;
|
||||
use marker::{Copy, Send, Sync, Sized, self};
|
||||
use iter_private::TrustedRandomAccess;
|
||||
|
||||
mod sort;
|
||||
|
||||
#[repr(C)]
|
||||
struct Repr<T> {
|
||||
pub data: *const T,
|
||||
@ -71,86 +73,119 @@ pub trait SliceExt {
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn split_at(&self, mid: usize) -> (&[Self::Item], &[Self::Item]);
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn iter(&self) -> Iter<Self::Item>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn split<P>(&self, pred: P) -> Split<Self::Item, P>
|
||||
where P: FnMut(&Self::Item) -> bool;
|
||||
where P: FnMut(&Self::Item) -> bool;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn splitn<P>(&self, n: usize, pred: P) -> SplitN<Self::Item, P>
|
||||
where P: FnMut(&Self::Item) -> bool;
|
||||
where P: FnMut(&Self::Item) -> bool;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn rsplitn<P>(&self, n: usize, pred: P) -> RSplitN<Self::Item, P>
|
||||
where P: FnMut(&Self::Item) -> bool;
|
||||
where P: FnMut(&Self::Item) -> bool;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn windows(&self, size: usize) -> Windows<Self::Item>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn chunks(&self, size: usize) -> Chunks<Self::Item>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn get<I>(&self, index: I) -> Option<&I::Output>
|
||||
where I: SliceIndex<Self::Item>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn first(&self) -> Option<&Self::Item>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn split_first(&self) -> Option<(&Self::Item, &[Self::Item])>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn split_last(&self) -> Option<(&Self::Item, &[Self::Item])>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn last(&self) -> Option<&Self::Item>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
unsafe fn get_unchecked<I>(&self, index: I) -> &I::Output
|
||||
where I: SliceIndex<Self::Item>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn as_ptr(&self) -> *const Self::Item;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn binary_search<Q: ?Sized>(&self, x: &Q) -> Result<usize, usize>
|
||||
where Self::Item: Borrow<Q>,
|
||||
Q: Ord;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn binary_search_by<'a, F>(&'a self, f: F) -> Result<usize, usize>
|
||||
where F: FnMut(&'a Self::Item) -> Ordering;
|
||||
|
||||
#[stable(feature = "slice_binary_search_by_key", since = "1.10.0")]
|
||||
fn binary_search_by_key<'a, B, F, Q: ?Sized>(&'a self, b: &Q, f: F) -> Result<usize, usize>
|
||||
where F: FnMut(&'a Self::Item) -> B,
|
||||
B: Borrow<Q>,
|
||||
Q: Ord;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn len(&self) -> usize;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn is_empty(&self) -> bool { self.len() == 0 }
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn get_mut<I>(&mut self, index: I) -> Option<&mut I::Output>
|
||||
where I: SliceIndex<Self::Item>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn iter_mut(&mut self) -> IterMut<Self::Item>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn first_mut(&mut self) -> Option<&mut Self::Item>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn split_first_mut(&mut self) -> Option<(&mut Self::Item, &mut [Self::Item])>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn split_last_mut(&mut self) -> Option<(&mut Self::Item, &mut [Self::Item])>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn last_mut(&mut self) -> Option<&mut Self::Item>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn split_mut<P>(&mut self, pred: P) -> SplitMut<Self::Item, P>
|
||||
where P: FnMut(&Self::Item) -> bool;
|
||||
where P: FnMut(&Self::Item) -> bool;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn splitn_mut<P>(&mut self, n: usize, pred: P) -> SplitNMut<Self::Item, P>
|
||||
where P: FnMut(&Self::Item) -> bool;
|
||||
where P: FnMut(&Self::Item) -> bool;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn rsplitn_mut<P>(&mut self, n: usize, pred: P) -> RSplitNMut<Self::Item, P>
|
||||
where P: FnMut(&Self::Item) -> bool;
|
||||
where P: FnMut(&Self::Item) -> bool;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn chunks_mut(&mut self, chunk_size: usize) -> ChunksMut<Self::Item>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn swap(&mut self, a: usize, b: usize);
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn split_at_mut(&mut self, mid: usize) -> (&mut [Self::Item], &mut [Self::Item]);
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn reverse(&mut self);
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
unsafe fn get_unchecked_mut<I>(&mut self, index: I) -> &mut I::Output
|
||||
where I: SliceIndex<Self::Item>;
|
||||
|
||||
#[stable(feature = "core", since = "1.6.0")]
|
||||
fn as_mut_ptr(&mut self) -> *mut Self::Item;
|
||||
|
||||
@ -165,8 +200,22 @@ pub trait SliceExt {
|
||||
|
||||
#[stable(feature = "clone_from_slice", since = "1.7.0")]
|
||||
fn clone_from_slice(&mut self, src: &[Self::Item]) where Self::Item: Clone;
|
||||
|
||||
#[stable(feature = "copy_from_slice", since = "1.9.0")]
|
||||
fn copy_from_slice(&mut self, src: &[Self::Item]) where Self::Item: Copy;
|
||||
|
||||
#[unstable(feature = "sort_unstable", issue = "40585")]
|
||||
fn sort_unstable(&mut self)
|
||||
where Self::Item: Ord;
|
||||
|
||||
#[unstable(feature = "sort_unstable", issue = "40585")]
|
||||
fn sort_unstable_by<F>(&mut self, compare: F)
|
||||
where F: FnMut(&Self::Item, &Self::Item) -> Ordering;
|
||||
|
||||
#[unstable(feature = "sort_unstable", issue = "40585")]
|
||||
fn sort_unstable_by_key<B, F>(&mut self, f: F)
|
||||
where F: FnMut(&Self::Item) -> B,
|
||||
B: Ord;
|
||||
}
|
||||
|
||||
// Use macros to be generic over const/mut
|
||||
@ -238,7 +287,9 @@ impl<T> SliceExt for [T] {
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn split<P>(&self, pred: P) -> Split<T, P> where P: FnMut(&T) -> bool {
|
||||
fn split<P>(&self, pred: P) -> Split<T, P>
|
||||
where P: FnMut(&T) -> bool
|
||||
{
|
||||
Split {
|
||||
v: self,
|
||||
pred: pred,
|
||||
@ -247,8 +298,8 @@ impl<T> SliceExt for [T] {
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn splitn<P>(&self, n: usize, pred: P) -> SplitN<T, P> where
|
||||
P: FnMut(&T) -> bool,
|
||||
fn splitn<P>(&self, n: usize, pred: P) -> SplitN<T, P>
|
||||
where P: FnMut(&T) -> bool
|
||||
{
|
||||
SplitN {
|
||||
inner: GenericSplitN {
|
||||
@ -260,8 +311,8 @@ impl<T> SliceExt for [T] {
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn rsplitn<P>(&self, n: usize, pred: P) -> RSplitN<T, P> where
|
||||
P: FnMut(&T) -> bool,
|
||||
fn rsplitn<P>(&self, n: usize, pred: P) -> RSplitN<T, P>
|
||||
where P: FnMut(&T) -> bool
|
||||
{
|
||||
RSplitN {
|
||||
inner: GenericSplitN {
|
||||
@ -422,13 +473,15 @@ impl<T> SliceExt for [T] {
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn split_mut<P>(&mut self, pred: P) -> SplitMut<T, P> where P: FnMut(&T) -> bool {
|
||||
fn split_mut<P>(&mut self, pred: P) -> SplitMut<T, P>
|
||||
where P: FnMut(&T) -> bool
|
||||
{
|
||||
SplitMut { v: self, pred: pred, finished: false }
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn splitn_mut<P>(&mut self, n: usize, pred: P) -> SplitNMut<T, P> where
|
||||
P: FnMut(&T) -> bool
|
||||
fn splitn_mut<P>(&mut self, n: usize, pred: P) -> SplitNMut<T, P>
|
||||
where P: FnMut(&T) -> bool
|
||||
{
|
||||
SplitNMut {
|
||||
inner: GenericSplitN {
|
||||
@ -450,7 +503,7 @@ impl<T> SliceExt for [T] {
|
||||
invert: true
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn chunks_mut(&mut self, chunk_size: usize) -> ChunksMut<T> {
|
||||
@ -512,7 +565,10 @@ impl<T> SliceExt for [T] {
|
||||
m >= n && needle == &self[m-n..]
|
||||
}
|
||||
|
||||
fn binary_search<Q: ?Sized>(&self, x: &Q) -> Result<usize, usize> where T: Borrow<Q>, Q: Ord {
|
||||
fn binary_search<Q: ?Sized>(&self, x: &Q) -> Result<usize, usize>
|
||||
where T: Borrow<Q>,
|
||||
Q: Ord
|
||||
{
|
||||
self.binary_search_by(|p| p.borrow().cmp(x))
|
||||
}
|
||||
|
||||
@ -548,6 +604,28 @@ impl<T> SliceExt for [T] {
|
||||
{
|
||||
self.binary_search_by(|k| f(k).borrow().cmp(b))
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn sort_unstable(&mut self)
|
||||
where Self::Item: Ord
|
||||
{
|
||||
sort::quicksort(self, |a, b| a.lt(b));
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn sort_unstable_by<F>(&mut self, mut compare: F)
|
||||
where F: FnMut(&Self::Item, &Self::Item) -> Ordering
|
||||
{
|
||||
sort::quicksort(self, |a, b| compare(a, b) == Ordering::Less);
|
||||
}
|
||||
|
||||
#[inline]
|
||||
fn sort_unstable_by_key<B, F>(&mut self, mut f: F)
|
||||
where F: FnMut(&Self::Item) -> B,
|
||||
B: Ord
|
||||
{
|
||||
sort::quicksort(self, |a, b| f(a).lt(&f(b)));
|
||||
}
|
||||
}
|
||||
|
||||
#[stable(feature = "rust1", since = "1.0.0")]
|
||||
@ -2175,6 +2253,15 @@ pub unsafe fn from_raw_parts_mut<'a, T>(p: *mut T, len: usize) -> &'a mut [T] {
|
||||
mem::transmute(Repr { data: p, len: len })
|
||||
}
|
||||
|
||||
// This function is public only because there is no other way to unit test heapsort.
|
||||
#[unstable(feature = "sort_internals", reason = "internal to sort module", issue = "0")]
|
||||
#[doc(hidden)]
|
||||
pub fn heapsort<T, F>(v: &mut [T], mut is_less: F)
|
||||
where F: FnMut(&T, &T) -> bool
|
||||
{
|
||||
sort::heapsort(v, &mut is_less);
|
||||
}
|
||||
|
||||
//
|
||||
// Comparison traits
|
||||
//
|
||||
699
src/libcore/slice/sort.rs
Normal file
699
src/libcore/slice/sort.rs
Normal file
@ -0,0 +1,699 @@
|
||||
// Copyright 2017 The Rust Project Developers. See the COPYRIGHT
|
||||
// file at the top-level directory of this distribution and at
|
||||
// http://rust-lang.org/COPYRIGHT.
|
||||
//
|
||||
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
|
||||
// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
|
||||
// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
|
||||
// option. This file may not be copied, modified, or distributed
|
||||
// except according to those terms.
|
||||
|
||||
//! Slice sorting
|
||||
//!
|
||||
//! This module contains an sort algorithm based on Orson Peters' pattern-defeating quicksort,
|
||||
//! published at: https://github.com/orlp/pdqsort
|
||||
//!
|
||||
//! Unstable sorting is compatible with libcore because it doesn't allocate memory, unlike our
|
||||
//! stable sorting implementation.
|
||||
|
||||
use cmp;
|
||||
use mem;
|
||||
use ptr;
|
||||
|
||||
/// Holds a value, but never drops it.
|
||||
#[allow(unions_with_drop_fields)]
|
||||
union NoDrop<T> {
|
||||
value: T
|
||||
}
|
||||
|
||||
/// When dropped, copies from `src` into `dest`.
|
||||
struct CopyOnDrop<T> {
|
||||
src: *mut T,
|
||||
dest: *mut T,
|
||||
}
|
||||
|
||||
impl<T> Drop for CopyOnDrop<T> {
|
||||
fn drop(&mut self) {
|
||||
unsafe { ptr::copy_nonoverlapping(self.src, self.dest, 1); }
|
||||
}
|
||||
}
|
||||
|
||||
/// Shifts the first element to the right until it encounters a greater or equal element.
|
||||
fn shift_head<T, F>(v: &mut [T], is_less: &mut F)
|
||||
where F: FnMut(&T, &T) -> bool
|
||||
{
|
||||
let len = v.len();
|
||||
unsafe {
|
||||
// If the first two elements are out-of-order...
|
||||
if len >= 2 && is_less(v.get_unchecked(1), v.get_unchecked(0)) {
|
||||
// Read the first element into a stack-allocated variable. If a following comparison
|
||||
// operation panics, `hole` will get dropped and automatically write the element back
|
||||
// into the slice.
|
||||
let mut tmp = NoDrop { value: ptr::read(v.get_unchecked(0)) };
|
||||
let mut hole = CopyOnDrop {
|
||||
src: &mut tmp.value,
|
||||
dest: v.get_unchecked_mut(1),
|
||||
};
|
||||
ptr::copy_nonoverlapping(v.get_unchecked(1), v.get_unchecked_mut(0), 1);
|
||||
|
||||
for i in 2..len {
|
||||
if !is_less(v.get_unchecked(i), &tmp.value) {
|
||||
break;
|
||||
}
|
||||
|
||||
// Move `i`-th element one place to the left, thus shifting the hole to the right.
|
||||
ptr::copy_nonoverlapping(v.get_unchecked(i), v.get_unchecked_mut(i - 1), 1);
|
||||
hole.dest = v.get_unchecked_mut(i);
|
||||
}
|
||||
// `hole` gets dropped and thus copies `tmp` into the remaining hole in `v`.
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// Shifts the last element to the left until it encounters a smaller or equal element.
|
||||
fn shift_tail<T, F>(v: &mut [T], is_less: &mut F)
|
||||
where F: FnMut(&T, &T) -> bool
|
||||
{
|
||||
let len = v.len();
|
||||
unsafe {
|
||||
// If the last two elements are out-of-order...
|
||||
if len >= 2 && is_less(v.get_unchecked(len - 1), v.get_unchecked(len - 2)) {
|
||||
// Read the last element into a stack-allocated variable. If a following comparison
|
||||
// operation panics, `hole` will get dropped and automatically write the element back
|
||||
// into the slice.
|
||||
let mut tmp = NoDrop { value: ptr::read(v.get_unchecked(len - 1)) };
|
||||
let mut hole = CopyOnDrop {
|
||||
src: &mut tmp.value,
|
||||
dest: v.get_unchecked_mut(len - 2),
|
||||
};
|
||||
ptr::copy_nonoverlapping(v.get_unchecked(len - 2), v.get_unchecked_mut(len - 1), 1);
|
||||
|
||||
for i in (0..len-2).rev() {
|
||||
if !is_less(&tmp.value, v.get_unchecked(i)) {
|
||||
break;
|
||||
}
|
||||
|
||||
// Move `i`-th element one place to the right, thus shifting the hole to the left.
|
||||
ptr::copy_nonoverlapping(v.get_unchecked(i), v.get_unchecked_mut(i + 1), 1);
|
||||
hole.dest = v.get_unchecked_mut(i);
|
||||
}
|
||||
// `hole` gets dropped and thus copies `tmp` into the remaining hole in `v`.
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// Partially sorts a slice by shifting several out-of-order elements around.
|
||||
///
|
||||
/// Returns true if the slice is sorted at the end. This function is `O(n)` worst-case.
|
||||
#[cold]
|
||||
fn partial_insertion_sort<T, F>(v: &mut [T], is_less: &mut F) -> bool
|
||||
where F: FnMut(&T, &T) -> bool
|
||||
{
|
||||
// Maximum number of adjacent out-of-order pairs that will get shifted.
|
||||
const MAX_STEPS: usize = 5;
|
||||
// If the slice is shorter than this, don't shift any elements.
|
||||
const SHORTEST_SHIFTING: usize = 50;
|
||||
|
||||
let len = v.len();
|
||||
let mut i = 1;
|
||||
|
||||
for _ in 0..MAX_STEPS {
|
||||
unsafe {
|
||||
// Find the next pair of adjacent out-of-order elements.
|
||||
while i < len && !is_less(v.get_unchecked(i), v.get_unchecked(i - 1)) {
|
||||
i += 1;
|
||||
}
|
||||
}
|
||||
|
||||
// Are we done?
|
||||
if i == len {
|
||||
return true;
|
||||
}
|
||||
|
||||
// Don't shift elements on short arrays, that has a performance cost.
|
||||
if len < SHORTEST_SHIFTING {
|
||||
return false;
|
||||
}
|
||||
|
||||
// Swap the found pair of elements. This puts them in correct order.
|
||||
v.swap(i - 1, i);
|
||||
|
||||
// Shift the smaller element to the left.
|
||||
shift_tail(&mut v[..i], is_less);
|
||||
// Shift the greater element to the right.
|
||||
shift_head(&mut v[i..], is_less);
|
||||
}
|
||||
|
||||
// Didn't manage to sort the slice in the limited number of steps.
|
||||
false
|
||||
}
|
||||
|
||||
/// Sorts a slice using insertion sort, which is `O(n^2)` worst-case.
|
||||
fn insertion_sort<T, F>(v: &mut [T], is_less: &mut F)
|
||||
where F: FnMut(&T, &T) -> bool
|
||||
{
|
||||
for i in 2..v.len()+1 {
|
||||
shift_tail(&mut v[..i], is_less);
|
||||
}
|
||||
}
|
||||
|
||||
/// Sorts `v` using heapsort, which guarantees `O(n log n)` worst-case.
|
||||
#[cold]
|
||||
pub fn heapsort<T, F>(v: &mut [T], is_less: &mut F)
|
||||
where F: FnMut(&T, &T) -> bool
|
||||
{
|
||||
// This binary heap respects the invariant `parent >= child`.
|
||||
let mut sift_down = |v: &mut [T], mut node| {
|
||||
loop {
|
||||
// Children of `node`:
|
||||
let left = 2 * node + 1;
|
||||
let right = 2 * node + 2;
|
||||
|
||||
// Choose the greater child.
|
||||
let greater = if right < v.len() && is_less(&v[left], &v[right]) {
|
||||
right
|
||||
} else {
|
||||
left
|
||||
};
|
||||
|
||||
// Stop if the invariant holds at `node`.
|
||||
if greater >= v.len() || !is_less(&v[node], &v[greater]) {
|
||||
break;
|
||||
}
|
||||
|
||||
// Swap `node` with the greater child, move one step down, and continue sifting.
|
||||
v.swap(node, greater);
|
||||
node = greater;
|
||||
}
|
||||
};
|
||||
|
||||
// Build the heap in linear time.
|
||||
for i in (0 .. v.len() / 2).rev() {
|
||||
sift_down(v, i);
|
||||
}
|
||||
|
||||
// Pop maximal elements from the heap.
|
||||
for i in (1 .. v.len()).rev() {
|
||||
v.swap(0, i);
|
||||
sift_down(&mut v[..i], 0);
|
||||
}
|
||||
}
|
||||
|
||||
/// Partitions `v` into elements smaller than `pivot`, followed by elements greater than or equal
|
||||
/// to `pivot`.
|
||||
///
|
||||
/// Returns the number of elements smaller than `pivot`.
|
||||
///
|
||||
/// Partitioning is performed block-by-block in order to minimize the cost of branching operations.
|
||||
/// This idea is presented in the [BlockQuicksort][pdf] paper.
|
||||
///
|
||||
/// [pdf]: http://drops.dagstuhl.de/opus/volltexte/2016/6389/pdf/LIPIcs-ESA-2016-38.pdf
|
||||
fn partition_in_blocks<T, F>(v: &mut [T], pivot: &T, is_less: &mut F) -> usize
|
||||
where F: FnMut(&T, &T) -> bool
|
||||
{
|
||||
// Number of elements in a typical block.
|
||||
const BLOCK: usize = 128;
|
||||
|
||||
// The partitioning algorithm repeats the following steps until completion:
|
||||
//
|
||||
// 1. Trace a block from the left side to identify elements greater than or equal to the pivot.
|
||||
// 2. Trace a block from the right side to identify elements smaller than the pivot.
|
||||
// 3. Exchange the identified elements between the left and right side.
|
||||
//
|
||||
// We keep the following variables for a block of elements:
|
||||
//
|
||||
// 1. `block` - Number of elements in the block.
|
||||
// 2. `start` - Start pointer into the `offsets` array.
|
||||
// 3. `end` - End pointer into the `offsets` array.
|
||||
// 4. `offsets - Indices of out-of-order elements within the block.
|
||||
|
||||
// The current block on the left side (from `l` to `l.offset(block_l)`).
|
||||
let mut l = v.as_mut_ptr();
|
||||
let mut block_l = BLOCK;
|
||||
let mut start_l = ptr::null_mut();
|
||||
let mut end_l = ptr::null_mut();
|
||||
let mut offsets_l: [u8; BLOCK] = unsafe { mem::uninitialized() };
|
||||
|
||||
// The current block on the right side (from `r.offset(-block_r)` to `r`).
|
||||
let mut r = unsafe { l.offset(v.len() as isize) };
|
||||
let mut block_r = BLOCK;
|
||||
let mut start_r = ptr::null_mut();
|
||||
let mut end_r = ptr::null_mut();
|
||||
let mut offsets_r: [u8; BLOCK] = unsafe { mem::uninitialized() };
|
||||
|
||||
// FIXME: When we get VLAs, try creating one array of length `min(v.len(), 2 * BLOCK)` rather
|
||||
// than two fixed-size arrays of length `BLOCK`. VLAs might be more cache-efficient.
|
||||
|
||||
// Returns the number of elements between pointers `l` (inclusive) and `r` (exclusive).
|
||||
fn width<T>(l: *mut T, r: *mut T) -> usize {
|
||||
assert!(mem::size_of::<T>() > 0);
|
||||
(r as usize - l as usize) / mem::size_of::<T>()
|
||||
}
|
||||
|
||||
loop {
|
||||
// We are done with partitioning block-by-block when `l` and `r` get very close. Then we do
|
||||
// some patch-up work in order to partition the remaining elements in between.
|
||||
let is_done = width(l, r) <= 2 * BLOCK;
|
||||
|
||||
if is_done {
|
||||
// Number of remaining elements (still not compared to the pivot).
|
||||
let mut rem = width(l, r);
|
||||
if start_l < end_l || start_r < end_r {
|
||||
rem -= BLOCK;
|
||||
}
|
||||
|
||||
// Adjust block sizes so that the left and right block don't overlap, but get perfectly
|
||||
// aligned to cover the whole remaining gap.
|
||||
if start_l < end_l {
|
||||
block_r = rem;
|
||||
} else if start_r < end_r {
|
||||
block_l = rem;
|
||||
} else {
|
||||
block_l = rem / 2;
|
||||
block_r = rem - block_l;
|
||||
}
|
||||
debug_assert!(block_l <= BLOCK && block_r <= BLOCK);
|
||||
debug_assert!(width(l, r) == block_l + block_r);
|
||||
}
|
||||
|
||||
if start_l == end_l {
|
||||
// Trace `block_l` elements from the left side.
|
||||
start_l = offsets_l.as_mut_ptr();
|
||||
end_l = offsets_l.as_mut_ptr();
|
||||
let mut elem = l;
|
||||
|
||||
for i in 0..block_l {
|
||||
unsafe {
|
||||
// Branchless comparison.
|
||||
*end_l = i as u8;
|
||||
end_l = end_l.offset(!is_less(&*elem, pivot) as isize);
|
||||
elem = elem.offset(1);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if start_r == end_r {
|
||||
// Trace `block_r` elements from the right side.
|
||||
start_r = offsets_r.as_mut_ptr();
|
||||
end_r = offsets_r.as_mut_ptr();
|
||||
let mut elem = r;
|
||||
|
||||
for i in 0..block_r {
|
||||
unsafe {
|
||||
// Branchless comparison.
|
||||
elem = elem.offset(-1);
|
||||
*end_r = i as u8;
|
||||
end_r = end_r.offset(is_less(&*elem, pivot) as isize);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Number of out-of-order elements to swap between the left and right side.
|
||||
let count = cmp::min(width(start_l, end_l), width(start_r, end_r));
|
||||
|
||||
if count > 0 {
|
||||
macro_rules! left { () => { l.offset(*start_l as isize) } }
|
||||
macro_rules! right { () => { r.offset(-(*start_r as isize) - 1) } }
|
||||
|
||||
// Instead of swapping one pair at the time, it is more efficient to perform a cyclic
|
||||
// permutation. This is not strictly equivalent to swapping, but produces a similar
|
||||
// result using fewer memory operations.
|
||||
unsafe {
|
||||
let tmp = ptr::read(left!());
|
||||
ptr::copy_nonoverlapping(right!(), left!(), 1);
|
||||
|
||||
for _ in 1..count {
|
||||
start_l = start_l.offset(1);
|
||||
ptr::copy_nonoverlapping(left!(), right!(), 1);
|
||||
start_r = start_r.offset(1);
|
||||
ptr::copy_nonoverlapping(right!(), left!(), 1);
|
||||
}
|
||||
|
||||
ptr::copy_nonoverlapping(&tmp, right!(), 1);
|
||||
mem::forget(tmp);
|
||||
start_l = start_l.offset(1);
|
||||
start_r = start_r.offset(1);
|
||||
}
|
||||
}
|
||||
|
||||
if start_l == end_l {
|
||||
// All out-of-order elements in the left block were moved. Move to the next block.
|
||||
l = unsafe { l.offset(block_l as isize) };
|
||||
}
|
||||
|
||||
if start_r == end_r {
|
||||
// All out-of-order elements in the right block were moved. Move to the previous block.
|
||||
r = unsafe { r.offset(-(block_r as isize)) };
|
||||
}
|
||||
|
||||
if is_done {
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
// All that remains now is at most one block (either the left or the right) with out-of-order
|
||||
// elements that need to be moved. Such remaining elements can be simply shifted to the end
|
||||
// within their block.
|
||||
|
||||
if start_l < end_l {
|
||||
// The left block remains.
|
||||
// Move it's remaining out-of-order elements to the far right.
|
||||
debug_assert_eq!(width(l, r), block_l);
|
||||
while start_l < end_l {
|
||||
unsafe {
|
||||
end_l = end_l.offset(-1);
|
||||
ptr::swap(l.offset(*end_l as isize), r.offset(-1));
|
||||
r = r.offset(-1);
|
||||
}
|
||||
}
|
||||
width(v.as_mut_ptr(), r)
|
||||
} else if start_r < end_r {
|
||||
// The right block remains.
|
||||
// Move it's remaining out-of-order elements to the far left.
|
||||
debug_assert_eq!(width(l, r), block_r);
|
||||
while start_r < end_r {
|
||||
unsafe {
|
||||
end_r = end_r.offset(-1);
|
||||
ptr::swap(l, r.offset(-(*end_r as isize) - 1));
|
||||
l = l.offset(1);
|
||||
}
|
||||
}
|
||||
width(v.as_mut_ptr(), l)
|
||||
} else {
|
||||
// Nothing else to do, we're done.
|
||||
width(v.as_mut_ptr(), l)
|
||||
}
|
||||
}
|
||||
|
||||
/// Partitions `v` into elements smaller than `v[pivot]`, followed by elements greater than or
|
||||
/// equal to `v[pivot]`.
|
||||
///
|
||||
/// Returns a tuple of:
|
||||
///
|
||||
/// 1. Number of elements smaller than `v[pivot]`.
|
||||
/// 2. True if `v` was already partitioned.
|
||||
fn partition<T, F>(v: &mut [T], pivot: usize, is_less: &mut F) -> (usize, bool)
|
||||
where F: FnMut(&T, &T) -> bool
|
||||
{
|
||||
let (mid, was_partitioned) = {
|
||||
// Place the pivot at the beginning of slice.
|
||||
v.swap(0, pivot);
|
||||
let (pivot, v) = v.split_at_mut(1);
|
||||
let pivot = &mut pivot[0];
|
||||
|
||||
// Read the pivot into a stack-allocated variable for efficiency. If a following comparison
|
||||
// operation panics, the pivot will be automatically written back into the slice.
|
||||
let mut tmp = NoDrop { value: unsafe { ptr::read(pivot) } };
|
||||
let _pivot_guard = CopyOnDrop {
|
||||
src: unsafe { &mut tmp.value },
|
||||
dest: pivot,
|
||||
};
|
||||
let pivot = unsafe { &tmp.value };
|
||||
|
||||
// Find the first pair of out-of-order elements.
|
||||
let mut l = 0;
|
||||
let mut r = v.len();
|
||||
unsafe {
|
||||
// Find the first element greater then or equal to the pivot.
|
||||
while l < r && is_less(v.get_unchecked(l), pivot) {
|
||||
l += 1;
|
||||
}
|
||||
|
||||
// Find the last element smaller that the pivot.
|
||||
while l < r && !is_less(v.get_unchecked(r - 1), pivot) {
|
||||
r -= 1;
|
||||
}
|
||||
}
|
||||
|
||||
(l + partition_in_blocks(&mut v[l..r], pivot, is_less), l >= r)
|
||||
|
||||
// `_pivot_guard` goes out of scope and writes the pivot (which is a stack-allocated
|
||||
// variable) back into the slice where it originally was. This step is critical in ensuring
|
||||
// safety!
|
||||
};
|
||||
|
||||
// Place the pivot between the two partitions.
|
||||
v.swap(0, mid);
|
||||
|
||||
(mid, was_partitioned)
|
||||
}
|
||||
|
||||
/// Partitions `v` into elements equal to `v[pivot]` followed by elements greater than `v[pivot]`.
|
||||
///
|
||||
/// Returns the number of elements equal to the pivot. It is assumed that `v` does not contain
|
||||
/// elements smaller than the pivot.
|
||||
fn partition_equal<T, F>(v: &mut [T], pivot: usize, is_less: &mut F) -> usize
|
||||
where F: FnMut(&T, &T) -> bool
|
||||
{
|
||||
// Place the pivot at the beginning of slice.
|
||||
v.swap(0, pivot);
|
||||
let (pivot, v) = v.split_at_mut(1);
|
||||
let pivot = &mut pivot[0];
|
||||
|
||||
// Read the pivot into a stack-allocated variable for efficiency. If a following comparison
|
||||
// operation panics, the pivot will be automatically written back into the slice.
|
||||
let mut tmp = NoDrop { value: unsafe { ptr::read(pivot) } };
|
||||
let _pivot_guard = CopyOnDrop {
|
||||
src: unsafe { &mut tmp.value },
|
||||
dest: pivot,
|
||||
};
|
||||
let pivot = unsafe { &tmp.value };
|
||||
|
||||
// Now partition the slice.
|
||||
let mut l = 0;
|
||||
let mut r = v.len();
|
||||
loop {
|
||||
unsafe {
|
||||
// Find the first element greater that the pivot.
|
||||
while l < r && !is_less(pivot, v.get_unchecked(l)) {
|
||||
l += 1;
|
||||
}
|
||||
|
||||
// Find the last element equal to the pivot.
|
||||
while l < r && is_less(pivot, v.get_unchecked(r - 1)) {
|
||||
r -= 1;
|
||||
}
|
||||
|
||||
// Are we done?
|
||||
if l >= r {
|
||||
break;
|
||||
}
|
||||
|
||||
// Swap the found pair of out-of-order elements.
|
||||
r -= 1;
|
||||
ptr::swap(v.get_unchecked_mut(l), v.get_unchecked_mut(r));
|
||||
l += 1;
|
||||
}
|
||||
}
|
||||
|
||||
// We found `l` elements equal to the pivot. Add 1 to account for the pivot itself.
|
||||
l + 1
|
||||
|
||||
// `_pivot_guard` goes out of scope and writes the pivot (which is a stack-allocated variable)
|
||||
// back into the slice where it originally was. This step is critical in ensuring safety!
|
||||
}
|
||||
|
||||
/// Scatters some elements around in an attempt to break patterns that might cause imbalanced
|
||||
/// partitions in quicksort.
|
||||
#[cold]
|
||||
fn break_patterns<T>(v: &mut [T]) {
|
||||
let len = v.len();
|
||||
|
||||
if len >= 8 {
|
||||
// A random number will be taken modulo this one. The modulus is a power of two so that we
|
||||
// can simply take bitwise "and", thus avoiding costly CPU operations.
|
||||
let modulus = (len / 4).next_power_of_two();
|
||||
debug_assert!(modulus >= 1 && modulus <= len / 2);
|
||||
|
||||
// Pseudorandom number generation from the "Xorshift RNGs" paper by George Marsaglia.
|
||||
let mut random = len;
|
||||
random ^= random << 13;
|
||||
random ^= random >> 17;
|
||||
random ^= random << 5;
|
||||
random &= modulus - 1;
|
||||
debug_assert!(random < len / 2);
|
||||
|
||||
// The first index.
|
||||
let a = len / 4 * 2;
|
||||
debug_assert!(a >= 1 && a < len - 2);
|
||||
|
||||
// The second index.
|
||||
let b = len / 4 + random;
|
||||
debug_assert!(b >= 1 && b < len - 2);
|
||||
|
||||
// Swap neighbourhoods of `a` and `b`.
|
||||
for i in 0..3 {
|
||||
v.swap(a - 1 + i, b - 1 + i);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// Chooses a pivot in `v` and returns the index and true if the slice is likely already sorted.
|
||||
///
|
||||
/// Elements in `v` might be reordered in the process.
|
||||
fn choose_pivot<T, F>(v: &mut [T], is_less: &mut F) -> (usize, bool)
|
||||
where F: FnMut(&T, &T) -> bool
|
||||
{
|
||||
// Minimum length to choose the median-of-medians method.
|
||||
// Shorter slices use the simple median-of-three method.
|
||||
const SHORTEST_MEDIAN_OF_MEDIANS: usize = 50;
|
||||
// Maximum number of swaps that can be performed in this function.
|
||||
const MAX_SWAPS: usize = 4 * 3;
|
||||
|
||||
let len = v.len();
|
||||
|
||||
// Three indices near which we are going to choose a pivot.
|
||||
let mut a = len / 4 * 1;
|
||||
let mut b = len / 4 * 2;
|
||||
let mut c = len / 4 * 3;
|
||||
|
||||
// Counts the total number of swaps we are about to perform while sorting indices.
|
||||
let mut swaps = 0;
|
||||
|
||||
if len >= 8 {
|
||||
// Swaps indices so that `v[a] <= v[b]`.
|
||||
let mut sort2 = |a: &mut usize, b: &mut usize| unsafe {
|
||||
if is_less(v.get_unchecked(*b), v.get_unchecked(*a)) {
|
||||
ptr::swap(a, b);
|
||||
swaps += 1;
|
||||
}
|
||||
};
|
||||
|
||||
// Swaps indices so that `v[a] <= v[b] <= v[c]`.
|
||||
let mut sort3 = |a: &mut usize, b: &mut usize, c: &mut usize| {
|
||||
sort2(a, b);
|
||||
sort2(b, c);
|
||||
sort2(a, b);
|
||||
};
|
||||
|
||||
if len >= SHORTEST_MEDIAN_OF_MEDIANS {
|
||||
// Finds the median of `v[a - 1], v[a], v[a + 1]` and stores the index into `a`.
|
||||
let mut sort_adjacent = |a: &mut usize| {
|
||||
let tmp = *a;
|
||||
sort3(&mut (tmp - 1), a, &mut (tmp + 1));
|
||||
};
|
||||
|
||||
// Find medians in the neighborhoods of `a`, `b`, and `c`.
|
||||
sort_adjacent(&mut a);
|
||||
sort_adjacent(&mut b);
|
||||
sort_adjacent(&mut c);
|
||||
}
|
||||
|
||||
// Find the median among `a`, `b`, and `c`.
|
||||
sort3(&mut a, &mut b, &mut c);
|
||||
}
|
||||
|
||||
if swaps < MAX_SWAPS {
|
||||
(b, swaps == 0)
|
||||
} else {
|
||||
// The maximum number of swaps was performed. Chances are the slice is descending or mostly
|
||||
// descending, so reversing will probably help sort it faster.
|
||||
v.reverse();
|
||||
(len - 1 - b, true)
|
||||
}
|
||||
}
|
||||
|
||||
/// Sorts `v` recursively.
|
||||
///
|
||||
/// If the slice had a predecessor in the original array, it is specified as `pred`.
|
||||
///
|
||||
/// `limit` is the number of allowed imbalanced partitions before switching to `heapsort`. If zero,
|
||||
/// this function will immediately switch to heapsort.
|
||||
fn recurse<'a, T, F>(mut v: &'a mut [T], is_less: &mut F, mut pred: Option<&'a T>, mut limit: usize)
|
||||
where F: FnMut(&T, &T) -> bool
|
||||
{
|
||||
// Slices of up to this length get sorted using insertion sort.
|
||||
const MAX_INSERTION: usize = 20;
|
||||
|
||||
// True if the last partitioning was reasonably balanced.
|
||||
let mut was_balanced = true;
|
||||
// True if the last partitioning didn't shuffle elements (the slice was already partitioned).
|
||||
let mut was_partitioned = true;
|
||||
|
||||
loop {
|
||||
let len = v.len();
|
||||
|
||||
// Very short slices get sorted using insertion sort.
|
||||
if len <= MAX_INSERTION {
|
||||
insertion_sort(v, is_less);
|
||||
return;
|
||||
}
|
||||
|
||||
// If too many bad pivot choices were made, simply fall back to heapsort in order to
|
||||
// guarantee `O(n log n)` worst-case.
|
||||
if limit == 0 {
|
||||
heapsort(v, is_less);
|
||||
return;
|
||||
}
|
||||
|
||||
// If the last partitioning was imbalanced, try breaking patterns in the slice by shuffling
|
||||
// some elements around. Hopefully we'll choose a better pivot this time.
|
||||
if !was_balanced {
|
||||
break_patterns(v);
|
||||
limit -= 1;
|
||||
}
|
||||
|
||||
// Choose a pivot and try guessing whether the slice is already sorted.
|
||||
let (pivot, likely_sorted) = choose_pivot(v, is_less);
|
||||
|
||||
// If the last partitioning was decently balanced and didn't shuffle elements, and if pivot
|
||||
// selection predicts the slice is likely already sorted...
|
||||
if was_balanced && was_partitioned && likely_sorted {
|
||||
// Try identifying several out-of-order elements and shifting them to correct
|
||||
// positions. If the slice ends up being completely sorted, we're done.
|
||||
if partial_insertion_sort(v, is_less) {
|
||||
return;
|
||||
}
|
||||
}
|
||||
|
||||
// If the chosen pivot is equal to the predecessor, then it's the smallest element in the
|
||||
// slice. Partition the slice into elements equal to and elements greater than the pivot.
|
||||
// This case is usually hit when the slice contains many duplicate elements.
|
||||
if let Some(p) = pred {
|
||||
if !is_less(p, &v[pivot]) {
|
||||
let mid = partition_equal(v, pivot, is_less);
|
||||
|
||||
// Continue sorting elements greater than the pivot.
|
||||
v = &mut {v}[mid..];
|
||||
continue;
|
||||
}
|
||||
}
|
||||
|
||||
// Partition the slice.
|
||||
let (mid, was_p) = partition(v, pivot, is_less);
|
||||
was_balanced = cmp::min(mid, len - mid) >= len / 8;
|
||||
was_partitioned = was_p;
|
||||
|
||||
// Split the slice into `left`, `pivot`, and `right`.
|
||||
let (left, right) = {v}.split_at_mut(mid);
|
||||
let (pivot, right) = right.split_at_mut(1);
|
||||
let pivot = &pivot[0];
|
||||
|
||||
// Recurse into the shorter side only in order to minimize the total number of recursive
|
||||
// calls and consume less stack space. Then just continue with the longer side (this is
|
||||
// akin to tail recursion).
|
||||
if left.len() < right.len() {
|
||||
recurse(left, is_less, pred, limit);
|
||||
v = right;
|
||||
pred = Some(pivot);
|
||||
} else {
|
||||
recurse(right, is_less, Some(pivot), limit);
|
||||
v = left;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// Sorts `v` using pattern-defeating quicksort, which is `O(n log n)` worst-case.
|
||||
pub fn quicksort<T, F>(v: &mut [T], mut is_less: F)
|
||||
where F: FnMut(&T, &T) -> bool
|
||||
{
|
||||
// Sorting has no meaningful behavior on zero-sized types.
|
||||
if mem::size_of::<T>() == 0 {
|
||||
return;
|
||||
}
|
||||
|
||||
// Limit the number of imbalanced partitions to `floor(log2(len)) + 1`.
|
||||
let limit = mem::size_of::<usize>() * 8 - v.len().leading_zeros() as usize;
|
||||
|
||||
recurse(v, &mut is_less, None, limit);
|
||||
}
|
||||
@ -19,18 +19,20 @@
|
||||
#![feature(decode_utf8)]
|
||||
#![feature(fixed_size_array)]
|
||||
#![feature(flt2dec)]
|
||||
#![feature(fmt_internals)]
|
||||
#![feature(libc)]
|
||||
#![feature(nonzero)]
|
||||
#![feature(rand)]
|
||||
#![feature(raw)]
|
||||
#![feature(sip_hash_13)]
|
||||
#![feature(slice_patterns)]
|
||||
#![feature(sort_internals)]
|
||||
#![feature(sort_unstable)]
|
||||
#![feature(step_by)]
|
||||
#![feature(test)]
|
||||
#![feature(try_from)]
|
||||
#![feature(unicode)]
|
||||
#![feature(unique)]
|
||||
#![feature(fmt_internals)]
|
||||
|
||||
extern crate core;
|
||||
extern crate test;
|
||||
|
||||
@ -8,7 +8,9 @@
|
||||
// option. This file may not be copied, modified, or distributed
|
||||
// except according to those terms.
|
||||
|
||||
use core::slice::heapsort;
|
||||
use core::result::Result::{Ok, Err};
|
||||
use rand::{Rng, XorShiftRng};
|
||||
|
||||
#[test]
|
||||
fn test_binary_search() {
|
||||
@ -139,9 +141,6 @@ fn test_chunks_mut_last() {
|
||||
assert_eq!(c2.last().unwrap()[0], 4);
|
||||
}
|
||||
|
||||
|
||||
|
||||
|
||||
#[test]
|
||||
fn test_windows_count() {
|
||||
let v: &[i32] = &[0, 1, 2, 3, 4, 5];
|
||||
@ -224,3 +223,57 @@ fn get_unchecked_mut_range() {
|
||||
assert_eq!(v.get_unchecked_mut(1..4), &mut [1, 2, 3][..]);
|
||||
}
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn sort_unstable() {
|
||||
let mut v = [0; 600];
|
||||
let mut tmp = [0; 600];
|
||||
let mut rng = XorShiftRng::new_unseeded();
|
||||
|
||||
for len in (2..25).chain(500..510) {
|
||||
let v = &mut v[0..len];
|
||||
let tmp = &mut tmp[0..len];
|
||||
|
||||
for &modulus in &[5, 10, 100, 1000] {
|
||||
for _ in 0..100 {
|
||||
for i in 0..len {
|
||||
v[i] = rng.gen::<i32>() % modulus;
|
||||
}
|
||||
|
||||
// Sort in default order.
|
||||
tmp.copy_from_slice(v);
|
||||
tmp.sort_unstable();
|
||||
assert!(tmp.windows(2).all(|w| w[0] <= w[1]));
|
||||
|
||||
// Sort in ascending order.
|
||||
tmp.copy_from_slice(v);
|
||||
tmp.sort_unstable_by(|a, b| a.cmp(b));
|
||||
assert!(tmp.windows(2).all(|w| w[0] <= w[1]));
|
||||
|
||||
// Sort in descending order.
|
||||
tmp.copy_from_slice(v);
|
||||
tmp.sort_unstable_by(|a, b| b.cmp(a));
|
||||
assert!(tmp.windows(2).all(|w| w[0] >= w[1]));
|
||||
|
||||
// Test heapsort using `<` operator.
|
||||
tmp.copy_from_slice(v);
|
||||
heapsort(tmp, |a, b| a < b);
|
||||
assert!(tmp.windows(2).all(|w| w[0] <= w[1]));
|
||||
|
||||
// Test heapsort using `>` operator.
|
||||
tmp.copy_from_slice(v);
|
||||
heapsort(tmp, |a, b| a > b);
|
||||
assert!(tmp.windows(2).all(|w| w[0] >= w[1]));
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Should not panic.
|
||||
[0i32; 0].sort_unstable();
|
||||
[(); 10].sort_unstable();
|
||||
[(); 100].sort_unstable();
|
||||
|
||||
let mut v = [0xDEADBEEFu64];
|
||||
v.sort_unstable();
|
||||
assert!(v == [0xDEADBEEF]);
|
||||
}
|
||||
|
||||
Loading…
Reference in New Issue
Block a user