2022-11-29 Triage Log

A relatively quiet week for performance, with the notable exception of “Avoid GenFuture shim when compiling async constructs #104321” which brought sizeable wins on a number of stress test benchmarks. It probably won't be of huge benefit to most codebases, but should provide smaller wins to folks with large amounts of async-generated futures.

Triage done by @simulacrum. Revision range: a78c9bee4d9d51a3891bd8ecae1f28a93b83653b..8a09420ac48658cad726e0a6997687ceac4151e3

Summary:

(instructions:u)	mean	range	count
Regressions ❌ (primary)	1.0%	[0.3%, 1.5%]	11
Regressions ❌ (secondary)	1.8%	[0.4%, 3.0%]	25
Improvements ✅ (primary)	-1.0%	[-3.8%, -0.3%]	148
Improvements ✅ (secondary)	-6.7%	[-87.7%, -0.3%]	107
All ❌✅ (primary)	-0.9%	[-3.8%, 1.5%]	159

3 Regressions, 3 Improvements, 6 Mixed; 2 of them in rollups 43 artifact comparisons made in total

Regressions

Rollup of 6 pull requests #104758 (Comparison Link)

(instructions:u)	mean	range	count
Regressions ❌ (primary)	0.3%	[0.3%, 0.3%]	1
Regressions ❌ (secondary)	2.1%	[0.6%, 7.1%]	13
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.3%	[0.3%, 0.3%]	1

This PR was a perf regression (see the report here https://github.com/rust-lang/rust/pull/104758#issuecomment-1325100344). It‘s relatively small and mostly limited to secondary workloads so I’m marking as triaged.

Branch Clause from Predicate #104846 (Comparison Link)

(instructions:u)	mean	range	count
Regressions ❌ (primary)	0.8%	[0.2%, 2.4%]	38
Regressions ❌ (secondary)	0.4%	[0.3%, 0.8%]	5
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.8%	[0.2%, 2.4%]	38

Unexpected regression, PR reviewer and author are going to look into it.

Add ConstKind::Expr #99798 (Comparison Link)

(instructions:u)	mean	range	count
Regressions ❌ (primary)	0.3%	[0.2%, 0.5%]	6
Regressions ❌ (secondary)	0.9%	[0.3%, 1.6%]	21
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.2%	[-0.2%, -0.2%]	1
All ❌✅ (primary)	0.3%	[0.2%, 0.5%]	6

Previous perf runs showed slight improvements, but the post-merge one shows regressions. They are mostly among secondary benchmarks.

Improvements

Fix perf regression by correctly matching keywords #104410 (Comparison Link)

(instructions:u)	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.3%	[-0.5%, -0.2%]	32
Improvements ✅ (secondary)	-0.8%	[-1.1%, -0.5%]	8
All ❌✅ (primary)	-0.3%	[-0.5%, -0.2%]	32

Use an IndexVec to cache queries with index-like key #103808 (Comparison Link)

(instructions:u)	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.8%	[-2.3%, -0.2%]	133
Improvements ✅ (secondary)	-1.2%	[-7.8%, -0.2%]	105
All ❌✅ (primary)	-0.8%	[-2.3%, -0.2%]	133

Avoid GenFuture shim when compiling async constructs #104321 (Comparison Link)

(instructions:u)	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-28.0%	[-87.8%, -1.9%]	22
All ❌✅ (primary)	-	-	0

Mixed

optimize field ordering by grouping m*2^n-sized fields with equivalently aligned ones #102750 (Comparison Link)

(instructions:u)	mean	range	count
Regressions ❌ (primary)	0.5%	[0.3%, 0.8%]	4
Regressions ❌ (secondary)	1.2%	[0.3%, 2.3%]	26
Improvements ✅ (primary)	-0.5%	[-2.2%, -0.2%]	44
Improvements ✅ (secondary)	-0.6%	[-1.2%, -0.1%]	22
All ❌✅ (primary)	-0.4%	[-2.2%, 0.8%]	48

Some followup discussion on https://github.com/rust-lang/rust/issues/104807 but overall a roughly neutral change.

privacy: Fix more (potential) issues with effective visibilities #104602 (Comparison Link)

(instructions:u)	mean	range	count
Regressions ❌ (primary)	0.3%	[0.2%, 0.3%]	8
Regressions ❌ (secondary)	2.2%	[0.3%, 3.6%]	17
Improvements ✅ (primary)	-1.0%	[-1.4%, -0.5%]	13
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-0.5%	[-1.4%, 0.3%]	21

This fixes correctness issues, and the regressions are mostly in unused-warnings which is a stress test where 3% doesn't really matter. Plus libc got a bit faster. So overall, not a performance concern.

Manually implement PartialEq for Option<T> and specialize non-nullable types #103556 (Comparison Link)

(instructions:u)	mean	range	count
Regressions ❌ (primary)	2.1%	[2.1%, 2.1%]	1
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.3%	[-0.4%, -0.2%]	2
Improvements ✅ (secondary)	-0.3%	[-0.4%, -0.3%]	2
All ❌✅ (primary)	0.5%	[-0.4%, 2.1%]	3

Relatively few relatively small changes, no further investigation needed.

Stop peeling the last iteration of the loop in Vec::resize_with #104818 (Comparison Link)

(instructions:u)	mean	range	count
Regressions ❌ (primary)	0.8%	[0.8%, 0.8%]	1
Regressions ❌ (secondary)	0.3%	[0.3%, 0.3%]	1
Improvements ✅ (primary)	-0.3%	[-0.3%, -0.3%]	3
Improvements ✅ (secondary)	-0.3%	[-0.4%, -0.1%]	7
All ❌✅ (primary)	-0.0%	[-0.3%, 0.8%]	4

Perf changes are very small, and wins slightly outweigh losses.

Update VecDeque implementation to use head+len instead of head+tail #102991 (Comparison Link)

(instructions:u)	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	0.8%	[0.2%, 1.4%]	4
Improvements ✅ (primary)	-0.3%	[-0.5%, -0.2%]	3
Improvements ✅ (secondary)	-0.3%	[-0.5%, -0.2%]	3
All ❌✅ (primary)	-0.3%	[-0.5%, -0.2%]	3

Performance looks like overall neutral to slight win.

Rollup of 6 pull requests #105008 (Comparison Link)

(instructions:u)	mean	range	count
Regressions ❌ (primary)	0.6%	[0.2%, 0.9%]	8
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.9%	[-0.9%, -0.8%]	2
Improvements ✅ (secondary)	-2.4%	[-2.7%, -2.1%]	6
All ❌✅ (primary)	0.3%	[-0.9%, 0.9%]	10

keccak improvements may be spurious but overall neutral so no further investigation warranted.