Rollup merge of #39107 - llogiq:branchless_filter_count, r=alexcrichton

branchless .filter(_).count()

I found that the branchless version is only slower if we have little to no branch misses, which usually isn't the case. I notice speedups between -5% (perfect prediction) and 60% (real world data).
This commit is contained in:
Corey Farwell 2017-02-05 09:14:40 -05:00 committed by GitHub
commit 8163b4b1f2
2 changed files with 27 additions and 1 deletions

View File

@ -1086,7 +1086,7 @@ impl<I: Iterator, P> Iterator for Filter<I, P> where P: FnMut(&I::Item) -> bool
#[inline]
fn next(&mut self) -> Option<I::Item> {
for x in self.iter.by_ref() {
for x in &mut self.iter {
if (self.predicate)(&x) {
return Some(x);
}
@ -1099,6 +1099,26 @@ impl<I: Iterator, P> Iterator for Filter<I, P> where P: FnMut(&I::Item) -> bool
let (_, upper) = self.iter.size_hint();
(0, upper) // can't know a lower bound, due to the predicate
}
// this special case allows the compiler to make `.filter(_).count()`
// branchless. Barring perfect branch prediction (which is unattainable in
// the general case), this will be much faster in >90% of cases (containing
// virtually all real workloads) and only a tiny bit slower in the rest.
//
// Having this specialization thus allows us to write `.filter(p).count()`
// where we would otherwise write `.map(|x| p(x) as usize).sum()`, which is
// less readable and also less backwards-compatible to Rust before 1.10.
//
// Using the branchless version will also simplify the LLVM byte code, thus
// leaving more budget for LLVM optimizations.
#[inline]
fn count(mut self) -> usize {
let mut count = 0;
for x in &mut self.iter {
count += (self.predicate)(&x) as usize;
}
count
}
}
#[stable(feature = "rust1", since = "1.0.0")]

View File

@ -191,6 +191,12 @@ fn test_iterator_enumerate_count() {
assert_eq!(xs.iter().count(), 6);
}
#[test]
fn test_iterator_filter_count() {
let xs = [0, 1, 2, 3, 4, 5, 6, 7, 8];
assert_eq!(xs.iter().filter(|&&x| x % 2 == 0).count(), 5);
}
#[test]
fn test_iterator_peekable() {
let xs = vec![0, 1, 2, 3, 4, 5];