1 use core::borrow::Borrow;
4 use core::num::Wrapping;
5 use test::{black_box, Bencher};
8 fn bench_rposition(b: &mut Bencher) {
9 let it: Vec<usize> = (0..300).collect();
11 it.iter().rposition(|&x| x <= 150);
16 fn bench_skip_while(b: &mut Bencher) {
29 fn bench_multiple_take(b: &mut Bencher) {
30 let mut it = (0..42).cycle();
32 let n = it.next().unwrap();
34 it.clone().take(it.next().unwrap()).all(|_| true);
39 fn scatter(x: i32) -> i32 {
44 fn bench_max_by_key(b: &mut Bencher) {
47 it.map(black_box).max_by_key(|&x| scatter(x))
51 // https://www.reddit.com/r/rust/comments/31syce/using_iterators_to_find_the_index_of_the_min_or/
53 fn bench_max_by_key2(b: &mut Bencher) {
54 fn max_index_iter(array: &[i32]) -> usize {
55 array.iter().enumerate().max_by_key(|&(_, item)| item).unwrap().0
58 let mut data = vec![0; 1638];
61 b.iter(|| max_index_iter(&data));
65 fn bench_max(b: &mut Bencher) {
68 it.map(black_box).map(scatter).max()
72 pub fn copy_zip(xs: &[u8], ys: &mut [u8]) {
73 for (a, b) in ys.iter_mut().zip(xs) {
78 pub fn add_zip(xs: &[f32], ys: &mut [f32]) {
79 for (a, b) in ys.iter_mut().zip(xs) {
85 fn bench_zip_copy(b: &mut Bencher) {
86 let source = vec![0u8; 16 * 1024];
87 let mut dst = black_box(vec![0u8; 16 * 1024]);
88 b.iter(|| copy_zip(&source, &mut dst))
92 fn bench_zip_add(b: &mut Bencher) {
93 let source = vec![1.; 16 * 1024];
94 let mut dst = vec![0.; 16 * 1024];
95 b.iter(|| add_zip(&source, &mut dst));
98 /// `Iterator::for_each` implemented as a plain loop.
99 fn for_each_loop<I, F>(iter: I, mut f: F)
109 /// `Iterator::for_each` implemented with `fold` for internal iteration.
110 /// (except when `by_ref()` effectively disables that optimization.)
111 fn for_each_fold<I, F>(iter: I, mut f: F)
116 iter.fold((), move |(), item| f(item));
120 fn bench_for_each_chain_loop(b: &mut Bencher) {
123 let iter = (0i64..1000000).chain(0..1000000).map(black_box);
124 for_each_loop(iter, |x| acc += x);
130 fn bench_for_each_chain_fold(b: &mut Bencher) {
133 let iter = (0i64..1000000).chain(0..1000000).map(black_box);
134 for_each_fold(iter, |x| acc += x);
140 fn bench_for_each_chain_ref_fold(b: &mut Bencher) {
143 let mut iter = (0i64..1000000).chain(0..1000000).map(black_box);
144 for_each_fold(iter.by_ref(), |x| acc += x);
149 /// Helper to benchmark `sum` for iterators taken by value which
150 /// can optimize `fold`, and by reference which cannot.
151 macro_rules! bench_sums {
152 ($bench_sum:ident, $bench_ref_sum:ident, $iter:expr) => {
154 fn $bench_sum(b: &mut Bencher) {
155 b.iter(|| -> i64 { $iter.map(black_box).sum() });
159 fn $bench_ref_sum(b: &mut Bencher) {
160 b.iter(|| -> i64 { $iter.map(black_box).by_ref().sum() });
167 bench_flat_map_ref_sum,
168 (0i64..1000).flat_map(|x| x..x+1000)
172 bench_flat_map_chain_sum,
173 bench_flat_map_chain_ref_sum,
174 (0i64..1000000).flat_map(|x| once(x).chain(once(x)))
179 bench_enumerate_ref_sum,
180 (0i64..1000000).enumerate().map(|(i, x)| x * i as i64)
184 bench_enumerate_chain_sum,
185 bench_enumerate_chain_ref_sum,
186 (0i64..1000000).chain(0..1000000).enumerate().map(|(i, x)| x * i as i64)
191 bench_filter_ref_sum,
192 (0i64..1000000).filter(|x| x % 3 == 0)
196 bench_filter_chain_sum,
197 bench_filter_chain_ref_sum,
198 (0i64..1000000).chain(0..1000000).filter(|x| x % 3 == 0)
202 bench_filter_map_sum,
203 bench_filter_map_ref_sum,
204 (0i64..1000000).filter_map(|x| x.checked_mul(x))
208 bench_filter_map_chain_sum,
209 bench_filter_map_chain_ref_sum,
210 (0i64..1000000).chain(0..1000000).filter_map(|x| x.checked_mul(x))
216 (0i64..1000000).fuse()
220 bench_fuse_chain_sum,
221 bench_fuse_chain_ref_sum,
222 (0i64..1000000).chain(0..1000000).fuse()
227 bench_inspect_ref_sum,
228 (0i64..1000000).inspect(|_| {})
232 bench_inspect_chain_sum,
233 bench_inspect_chain_ref_sum,
234 (0i64..1000000).chain(0..1000000).inspect(|_| {})
239 bench_peekable_ref_sum,
240 (0i64..1000000).peekable()
244 bench_peekable_chain_sum,
245 bench_peekable_chain_ref_sum,
246 (0i64..1000000).chain(0..1000000).peekable()
252 (0i64..1000000).skip(1000)
256 bench_skip_chain_sum,
257 bench_skip_chain_ref_sum,
258 (0i64..1000000).chain(0..1000000).skip(1000)
262 bench_skip_while_sum,
263 bench_skip_while_ref_sum,
264 (0i64..1000000).skip_while(|&x| x < 1000)
268 bench_skip_while_chain_sum,
269 bench_skip_while_chain_ref_sum,
270 (0i64..1000000).chain(0..1000000).skip_while(|&x| x < 1000)
274 bench_take_while_chain_sum,
275 bench_take_while_chain_ref_sum,
276 (0i64..1000000).chain(1000000..).take_while(|&x| x < 1111111)
280 bench_cycle_take_sum,
281 bench_cycle_take_ref_sum,
282 (0..10000).cycle().take(1000000)
286 bench_cycle_skip_take_sum,
287 bench_cycle_skip_take_ref_sum,
288 (0..100000).cycle().skip(1000000).take(1000000)
292 bench_cycle_take_skip_sum,
293 bench_cycle_take_skip_ref_sum,
294 (0..100000).cycle().take(1000000).skip(100000)
298 bench_skip_cycle_skip_zip_add_sum,
299 bench_skip_cycle_skip_zip_add_ref_sum,
300 (0..100000).skip(100).cycle().skip(100)
301 .zip((0..100000).cycle().skip(10))
307 // Checks whether Skip<Zip<A,B>> is as fast as Zip<Skip<A>, Skip<B>>, from
308 // https://users.rust-lang.org/t/performance-difference-between-iterator-zip-and-skip-order/15743
310 fn bench_zip_then_skip(b: &mut Bencher) {
311 let v: Vec<_> = (0..100_000).collect();
312 let t: Vec<_> = (0..100_000).collect();
319 .take_while(|t| *t.0 < 10100)
320 .map(|(a, b)| *a + *b)
322 assert_eq!(s, 2009900);
326 fn bench_skip_then_zip(b: &mut Bencher) {
327 let v: Vec<_> = (0..100_000).collect();
328 let t: Vec<_> = (0..100_000).collect();
334 .zip(t.iter().skip(10000))
335 .take_while(|t| *t.0 < 10100)
336 .map(|(a, b)| *a + *b)
338 assert_eq!(s, 2009900);
343 fn bench_filter_count(b: &mut Bencher) {
344 b.iter(|| (0i64..1000000).map(black_box).filter(|x| x % 3 == 0).count())
348 fn bench_filter_ref_count(b: &mut Bencher) {
349 b.iter(|| (0i64..1000000).map(black_box).by_ref().filter(|x| x % 3 == 0).count())
353 fn bench_filter_chain_count(b: &mut Bencher) {
354 b.iter(|| (0i64..1000000).chain(0..1000000).map(black_box).filter(|x| x % 3 == 0).count())
358 fn bench_filter_chain_ref_count(b: &mut Bencher) {
360 (0i64..1000000).chain(0..1000000).map(black_box).by_ref().filter(|x| x % 3 == 0).count()
365 fn bench_partial_cmp(b: &mut Bencher) {
366 b.iter(|| (0..100000).map(black_box).partial_cmp((0..100000).map(black_box)))
370 fn bench_chain_partial_cmp(b: &mut Bencher) {
372 (0..50000).chain(50000..100000).map(black_box).partial_cmp((0..100000).map(black_box))
377 fn bench_lt(b: &mut Bencher) {
378 b.iter(|| (0..100000).map(black_box).lt((0..100000).map(black_box)))
382 fn bench_trusted_random_access_adapters(b: &mut Bencher) {
383 let vec1: Vec<_> = (0usize..100000).collect();
384 let vec2 = black_box(vec1.clone());
390 .map(|(idx, e)| idx.wrapping_add(e))
391 .zip(vec2.iter().copied())
392 .map(|(a, b)| a.wrapping_add(b))
394 let mut acc: usize = 0;
395 let size = iter.size();
397 // SAFETY: TRA requirements are satisfied by 0..size iteration and then dropping the
399 acc = acc.wrapping_add(unsafe { iter.__iterator_get_unchecked(i) });
405 /// Exercises the iter::Copied specialization for slice::Iter
407 fn bench_copied_chunks(b: &mut Bencher) {
408 let v = vec![1u8; 1024];
411 let mut iter = black_box(&v).iter().copied();
412 let mut acc = Wrapping(0);
413 // This uses a while-let loop to side-step the TRA specialization in ArrayChunks
414 while let Ok(chunk) = iter.next_chunk::<{ mem::size_of::<u64>() }>() {
415 let d = u64::from_ne_bytes(chunk);
416 acc += Wrapping(d.rotate_left(7).wrapping_add(1));
422 /// Exercises the TrustedRandomAccess specialization in ArrayChunks
424 fn bench_trusted_random_access_chunks(b: &mut Bencher) {
425 let v = vec![1u8; 1024];
430 // this shows that we're not relying on the slice::Iter specialization in Copied
431 .map(|b| *b.borrow())
432 .array_chunks::<{ mem::size_of::<u64>() }>()
434 let d = u64::from_ne_bytes(ary);
435 Wrapping(d.rotate_left(7).wrapping_add(1))
437 .sum::<Wrapping<u64>>()