Improve sift_down performance in BinaryHeap #81127
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
r? @shepmaster (rust-highfive has picked a reviewer for you, use r? to override) |
rust-highfive
added
the
S-waiting-on-review
JohnCSimon
added
S-waiting-on-review
JohnTitor
added a commit
to JohnTitor/rust
that referenced
this pull request
Feb 21, 2021
…e, r=Mark-Simulacrum Document BinaryHeap unsafe functions `BinaryHeap` contains some private safe functions but that are actually unsafe to call. This PR marks them `unsafe` and documents all the `unsafe` function calls inside them. While doing this I might also have found a bug: some "SAFETY" comments in `sift_down_range` and `sift_down_to_bottom` are valid only if you assume that `child` doesn't overflow. However it may overflow if `end > isize::MAX` which can be true for ZSTs (but I think only for them). I guess the easiest fix would be to skip any sifting if `mem::size_of::<T> == 0`. Probably conflicts with rust-lang#81127 but solving the eventual merge conflict should be pretty easy.
JohnTitor
added a commit
to JohnTitor/rust
that referenced
this pull request
Feb 21, 2021
…e, r=Mark-Simulacrum Document BinaryHeap unsafe functions `BinaryHeap` contains some private safe functions but that are actually unsafe to call. This PR marks them `unsafe` and documents all the `unsafe` function calls inside them. While doing this I might also have found a bug: some "SAFETY" comments in `sift_down_range` and `sift_down_to_bottom` are valid only if you assume that `child` doesn't overflow. However it may overflow if `end > isize::MAX` which can be true for ZSTs (but I think only for them). I guess the easiest fix would be to skip any sifting if `mem::size_of::<T> == 0`. Probably conflicts with rust-lang#81127 but solving the eventual merge conflict should be pretty easy.
JohnTitor
added a commit
to JohnTitor/rust
that referenced
this pull request
Feb 21, 2021
…e, r=Mark-Simulacrum Document BinaryHeap unsafe functions `BinaryHeap` contains some private safe functions but that are actually unsafe to call. This PR marks them `unsafe` and documents all the `unsafe` function calls inside them. While doing this I might also have found a bug: some "SAFETY" comments in `sift_down_range` and `sift_down_to_bottom` are valid only if you assume that `child` doesn't overflow. However it may overflow if `end > isize::MAX` which can be true for ZSTs (but I think only for them). I guess the easiest fix would be to skip any sifting if `mem::size_of::<T> == 0`. Probably conflicts with rust-lang#81127 but solving the eventual merge conflict should be pretty easy.
|
☔ The latest upstream changes (presumably #82359) made this pull request unmergeable. Please resolve the merge conflicts. |
Because child > 0, the two statements are equivalent, but using saturating_sub and <= yields in faster code. This is most notable in the binary_heap::bench_into_sorted_vec benchmark, which shows a speedup of 1.26x, which uses sift_down_range internally. The speedup of pop (that uses sift_down_to_bottom internally) is much less significant as the sifting method is not called in a loop.
hanmertens
force-pushed
the
binary_heap_sift_down_perf
branch
from
bdf2962 to
095bf01
Compare
|
r? @dtolnay |
dtolnay
approved these changes
Mar 9, 2021
There was a problem hiding this comment.
Thanks @hanmertens. I am not familiar with the BinaryHeap implementation but I am prepared to accept this on the basis of the into_sorted_vec benchmark. I confirmed that the behavior is logically identical to before in both places.
-
If
end >= 2thenchild < end - 1is equivalent tochild <= end - 2is equivalent tochild <= end.saturating_sub(2). -
If
end == 1thenchild < end - 1is false whilechild <= end.saturating_sub(2)is equivalent tochild == 0. However it's a loop invariant thatchild == 2 * hole.pos() + 1 > 0sochild != 0andchild <= end.saturating_sub(2)is also false. -
If
end == 0then contradiction because it's a precondition of the function that0 <= pos < end, andendis not mutated.
|
@bors r+ |
|
📌 Commit 095bf01 has been approved by |
bors
added
S-waiting-on-bors
m-ou-se
added a commit
to m-ou-se/rust
that referenced
this pull request
Mar 9, 2021
…rf, r=dtolnay Improve sift_down performance in BinaryHeap Replacing `child < end - 1` with `child <= end.saturating_sub(2)` in `BinaryHeap::sift_down_range` (surprisingly) results in a significant speedup of `BinaryHeap::into_sorted_vec`. The same substitution can be done for `BinaryHeap::sift_down_to_bottom`, which causes a slight but probably statistically insignificant speedup for `BinaryHeap::pop`. It's interesting that benchmarks aside from `bench_into_sorted_vec` are barely affected, even those that do use `sift_down_*` methods internally. | Benchmark | Before (ns/iter) | After (ns/iter) | Speedup | |--------------------------|------------------|-----------------|---------| | bench_find_smallest_1000<sup>1</sup> | 392,617 | 385,200 | 1.02 | | bench_from_vec<sup>1</sup> | 506,016 | 504,444 | 1.00 | | bench_into_sorted_vec<sup>1</sup> | 476,869 | 384,458 | 1.24 | | bench_peek_mut_deref_mut<sup>3</sup> | 518,753 | 519,792 | 1.00 | | bench_pop<sup>2</sup> | 446,718 | 444,409 | 1.01 | | bench_push<sup>3</sup> | 772,481 | 770,208 | 1.00 | <sup>1</sup>: internally calls `sift_down_range` <sup>2</sup>: internally calls `sift_down_to_bottom` <sup>3</sup>: should not be affected
This was referenced Mar 9, 2021
bors
added a commit
to rust-lang-ci/rust
that referenced
this pull request
Mar 9, 2021
Rollup of 8 pull requests Successful merges: - rust-lang#81127 (Improve sift_down performance in BinaryHeap) - rust-lang#81879 (Added #[repr(transparent)] to core::cmp::Reverse) - rust-lang#82048 (or-patterns: disallow in `let` bindings) - rust-lang#82731 (Bump libc dependency of std to 0.2.88.) - rust-lang#82799 (Add regression test for rust-lang#75525) - rust-lang#82841 (Change x64 size checks to not apply to x32.) - rust-lang#82883 (Update Cargo) - rust-lang#82887 (Update CONTRIBUTING.md) Failed merges: r? `@ghost` `@rustbot` modify labels: rollup
# for free
to join this conversation on GitHub.
Already have an account?
# to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Replacing
child < end - 1withchild <= end.saturating_sub(2)inBinaryHeap::sift_down_range(surprisingly) results in a significant speedup ofBinaryHeap::into_sorted_vec. The same substitution can be done forBinaryHeap::sift_down_to_bottom, which causes a slight but probably statistically insignificant speedup forBinaryHeap::pop. It's interesting that benchmarks aside frombench_into_sorted_vecare barely affected, even those that do usesift_down_*methods internally.1: internally calls
sift_down_range2: internally calls
sift_down_to_bottom3: should not be affected