`Vec::resize` for bytes should be a single `memset` #120050

scottmcm · 2024-01-17T07:29:23Z

Really I just started by trying to see if specializing iter::repeat_n would help the perf issue that kept me from removing Vec::extend_with last time I tried, but I noticed in the process that a resize for bytes doesn't set all the new space with a single memset: https://play.rust-lang.org/?version=nightly&mode=release&edition=2021&gist=35175ec844b46fcd95e2d0aad526859e

So using repeat_n to implement it -- like VecDeque uses, with the specialization for next to avoid a branch -- means that the optimizer notices the resize can set all the values with a single memset.

rustbot · 2024-01-17T07:29:32Z

r? @m-ou-se

(rustbot has picked a reviewer for you, use r? to override)

scottmcm · 2024-01-17T08:40:42Z

@bors try @rust-timer queue

`Vec::resize` for bytes should be a single `memset` Really I just started by trying to see if specializing `iter::repeat_n` would help the perf issue that kept me from removing `Vec::extend_with` last time I tried, but I noticed in the process that a resize for bytes doesn't set all the new space with a single `memset`: <https://play.rust-lang.org/?version=nightly&mode=release&edition=2021&gist=35175ec844b46fcd95e2d0aad526859e> So using `repeat_n` to implement it -- like `VecDeque` uses, with the specialization for `next` to avoid a branch -- means that the optimizer notices the resize can set all the values with a single memset.

bors · 2024-01-17T08:41:52Z

⌛ Trying commit 8652062 with merge 0db48a7...

bors · 2024-01-17T10:07:59Z

☀️ Try build successful - checks-actions
Build commit: 0db48a7 (0db48a7c3c2ca5e498da14eae53a7474e16886df)

rust-timer · 2024-01-17T11:52:44Z

Finished benchmarking commit (0db48a7): comparison URL.

Overall result: ❌✅ regressions and improvements - ACTION NEEDED

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

This is a highly reliable metric that was used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	0.4%	[0.2%, 0.5%]	15
Regressions ❌ (secondary)	3.3%	[3.1%, 3.7%]	6
Improvements ✅ (primary)	-1.5%	[-1.5%, -1.5%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.2%	[-1.5%, 0.5%]	16

Max RSS (memory usage)

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	2.9%	[0.3%, 5.5%]	2
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-8.3%	[-12.7%, -4.7%]	4
Improvements ✅ (secondary)	-4.9%	[-4.9%, -4.9%]	1
All ❌✅ (primary)	-4.6%	[-12.7%, 5.5%]	6

Cycles

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	20.2%	[18.6%, 22.7%]	6
Improvements ✅ (primary)	-1.5%	[-1.5%, -1.5%]	1
Improvements ✅ (secondary)	-2.9%	[-4.3%, -2.3%]	5
All ❌✅ (primary)	-1.5%	[-1.5%, -1.5%]	1

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 665.423s -> 664.993s (-0.06%)
Artifact size: 308.34 MiB -> 308.35 MiB (0.01%)

Kobzol · 2024-01-17T11:55:01Z

Looks like something bad has happened in borrow checking of wg-grammar (https://perf.rust-lang.org/detailed-query.html?commit=0db48a7c3c2ca5e498da14eae53a7474e16886df&benchmark=wg-grammar-check&scenario=full&base_commit=f45fe573a587f5d82e9e78cbece64bad4b3709f8.

scottmcm · 2024-01-17T22:17:52Z

@bors try @rust-timer queue

bors · 2024-01-17T22:19:01Z

⌛ Trying commit 6c32590 with merge 27371af...

`Vec::resize` for bytes should be a single `memset` Really I just started by trying to see if specializing `iter::repeat_n` would help the perf issue that kept me from removing `Vec::extend_with` last time I tried, but I noticed in the process that a resize for bytes doesn't set all the new space with a single `memset`: <https://play.rust-lang.org/?version=nightly&mode=release&edition=2021&gist=35175ec844b46fcd95e2d0aad526859e> So using `repeat_n` to implement it -- like `VecDeque` uses, with the specialization for `next` to avoid a branch -- means that the optimizer notices the resize can set all the values with a single memset.

bors · 2024-01-17T23:44:35Z

☀️ Try build successful - checks-actions
Build commit: 27371af (27371af9b8c50c2f3b165000d3bcd9684c4a0807)

rust-timer · 2024-01-18T02:32:19Z

Finished benchmarking commit (27371af): comparison URL.

Overall result: ❌✅ regressions and improvements - ACTION NEEDED

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

This is a highly reliable metric that was used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	0.4%	[0.3%, 0.7%]	13
Regressions ❌ (secondary)	2.1%	[1.9%, 2.3%]	6
Improvements ✅ (primary)	-0.5%	[-0.9%, -0.3%]	4
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.2%	[-0.9%, 0.7%]	17

Max RSS (memory usage)

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	4.3%	[0.2%, 8.4%]	2
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-2.5%	[-3.7%, -1.4%]	2
Improvements ✅ (secondary)	-1.2%	[-1.5%, -0.7%]	3
All ❌✅ (primary)	0.9%	[-3.7%, 8.4%]	4

Cycles

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	1.1%	[1.1%, 1.1%]	1
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-1.1%	[-1.1%, -1.1%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.0%	[-1.1%, 1.1%]	2

Binary size

Results

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	0.7%	[0.4%, 0.9%]	29
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.7%	[0.4%, 0.9%]	29

Bootstrap: 663.454s -> 664.862s (0.21%)
Artifact size: 308.30 MiB -> 308.31 MiB (0.01%)

scottmcm · 2024-07-04T23:34:19Z

@bors try @rust-timer queue

bors · 2024-08-04T00:00:24Z

☔ The latest upstream changes (presumably #128614) made this pull request unmergeable. Please resolve the merge conflicts.

Dylan-DPC · 2024-08-05T11:35:52Z

@scottmcm the try run didn't get executed so you will have to resolve conflicts and run it again

scottmcm · 2024-09-18T05:03:45Z

@bors try @rust-timer queue

bors · 2024-09-18T05:04:55Z

⌛ Trying commit 1697af2 with merge f5c3755...

`Vec::resize` for bytes should be a single `memset` Really I just started by trying to see if specializing `iter::repeat_n` would help the perf issue that kept me from removing `Vec::extend_with` last time I tried, but I noticed in the process that a resize for bytes doesn't set all the new space with a single `memset`: <https://play.rust-lang.org/?version=nightly&mode=release&edition=2021&gist=35175ec844b46fcd95e2d0aad526859e> So using `repeat_n` to implement it -- like `VecDeque` uses, with the specialization for `next` to avoid a branch -- means that the optimizer notices the resize can set all the values with a single memset.

scottmcm · 2024-09-18T05:07:23Z

library/alloc/src/vec/mod.rs

+            // Because there's no user code being run here, we can skip it for ZSTs.
+            // That helps tests in debug mode that do things like `vec![(); HUGE]`.
+            // See <https://github.com/rust-lang/rust/pull/118094>
+            if !T::IS_ZST {


cc @JarvisCraft in case you have thoughts on this approach, since I removed your specialization from #118094 in this PR

Thanks for tagging me! I've commented below on what I am unsure about, though the optimization really looks promising.

bors · 2024-09-18T06:54:28Z

☀️ Try build successful - checks-actions
Build commit: f5c3755 (f5c37550f884726ba52e559b2750e8585e7c12da)

rust-timer · 2024-09-18T08:08:07Z

Finished benchmarking commit (f5c3755): comparison URL.

Overall result: ❌✅ regressions and improvements - ACTION NEEDED

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

This is a highly reliable metric that was used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	0.3%	[0.2%, 0.4%]	7
Regressions ❌ (secondary)	0.5%	[0.2%, 0.8%]	11
Improvements ✅ (primary)	-0.4%	[-0.4%, -0.4%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.2%	[-0.4%, 0.4%]	8

Max RSS (memory usage)

Results (primary 0.9%, secondary -2.5%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	5.8%	[3.8%, 7.8%]	2
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-2.4%	[-3.5%, -1.4%]	3
Improvements ✅ (secondary)	-2.5%	[-2.5%, -2.5%]	1
All ❌✅ (primary)	0.9%	[-3.5%, 7.8%]	5

Cycles

Results (secondary 2.5%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	2.5%	[2.4%, 2.6%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Binary size

Results (primary 0.2%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	0.3%	[0.1%, 0.4%]	28
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.1%	[-0.2%, -0.0%]	11
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.2%	[-0.2%, 0.4%]	39

Bootstrap: 768.777s -> 770.588s (0.24%)
Artifact size: 341.28 MiB -> 341.28 MiB (-0.00%)

JarvisCraft · 2024-09-18T17:31:50Z

library/alloc/src/vec/spec_extend_elem.rs

+    T: Copy,
+{
+    fn spec_extend_elem(&mut self, n: usize, value: T) {
+        self.extend_elem_copy(n, value)


I am not sure if this specialization is correct for all cases of vec![ZST; N].

At the moment:

impl Copy for Foo permits <Foo as Clone>::clone to have side-effects.

The docs of vec! explicitly mention that it works via Clone.

Which means that at the moment for vec![ZST; N] any effects of ZST::clone are observed N times as can be seen in the example. With this change, they won't since the specialization skips any calls to this method.

This is the reason why I've only implemented the specialization for () previously.

As mentioned in #118094 (comment), this kind of change is still allowed, although I expect that there must be an explicit proof that this is a valid optimization and an update to vec!'s doc is probably required to explicitly state that such optimization may occur.

An alternative may be to add a perma-unstable fn to Clone like is_trivially_cloneable() defaulting to false and only overridable by rustc on derive, but this of course is more tedious.

rustbot assigned m-ou-se Jan 17, 2024

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-libs Relevant to the library team, which will review and decide on the PR/issue. labels Jan 17, 2024