Skip to content

Commit 1026da5

Browse files
Mark-SimulacrumMark-Simulacrum
authored
Merge pull request #2417 from Mark-Simulacrum/triage
Add triage for this week
2 parents e555238 + 996d692 commit 1026da5

1 file changed

Lines changed: 179 additions & 0 deletions

File tree

‎triage/2026/2026-02-23.md‎

Lines changed: 179 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,179 @@
1+
# 2026-02-24 Triage Log
2+
3+
Overall, a bit more noise than usual this week, but mostly a slight improvement
4+
with several low-level optimizations at MIR and LLVM IR building landing. Also
5+
less commits landing than usual, mostly due to GitHub CI issues during the week.
6+
7+
Triage done by **@simulacrum**.
8+
Revision range: [3c9faa0d..eeb94be7](https://perf.rust-lang.org/?start=3c9faa0d037b9eecda4a440cc482ff7f960fb8a5&end=eeb94be79adc9df7a09ad0b2421f16e60e6d932c&absolute=false&stat=instructions%3Au)
9+
10+
**Summary**:
11+
12+
| (instructions:u) | mean | range | count |
13+
|:----------------------------------:|:-----:|:--------------:|:-----:|
14+
| Regressions ❌ <br /> (primary) | 0.4% | [0.2%, 0.8%] | 4 |
15+
| Regressions ❌ <br /> (secondary) | 0.3% | [0.1%, 0.8%] | 14 |
16+
| Improvements ✅ <br /> (primary) | -0.4% | [-2.2%, -0.2%] | 67 |
17+
| Improvements ✅ <br /> (secondary) | -0.8% | [-5.0%, -0.1%] | 54 |
18+
| All ❌✅ (primary) | -0.4% | [-2.2%, 0.8%] | 71 |
19+
20+
3 Regressions, 4 Improvements, 4 Mixed; 3 of them in rollups
21+
24 artifact comparisons made in total
22+
23+
#### Regressions
24+
25+
Perform many const checks in typeck [#149375](https://github.com/rust-lang/rust/pull/149375) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=b6bed6f2d9f27e15d425958bbd288138ecdff61c&end=7f99507f57e6c4aa0dce3daf6a13cca8cd4dd312&stat=instructions:u)
26+
27+
| (instructions:u) | mean | range | count |
28+
|:----------------------------------:|:----:|:------------:|:-----:|
29+
| Regressions ❌ <br /> (primary) | 0.2% | [0.2%, 0.2%] | 3 |
30+
| Regressions ❌ <br /> (secondary) | 0.2% | [0.2%, 0.2%] | 4 |
31+
| Improvements ✅ <br /> (primary) | - | - | 0 |
32+
| Improvements ✅ <br /> (secondary) | - | - | 0 |
33+
| All ❌✅ (primary) | 0.2% | [0.2%, 0.2%] | 3 |
34+
35+
Most likely expected impact from running more code. Impact is small, and
36+
possibly limited to only secondary benchmarks: primary benchmark impact to libc
37+
*might* be noise, though it's hard to tell.
38+
39+
Rollup of 14 pull requests [#152965](https://github.com/rust-lang/rust/pull/152965) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=d8b2222b11fc03a31d43707c8c9006f025809ece&end=1500f0f47f5fe8ddcd6528f6c6c031b210b4eac5&stat=instructions:u)
40+
41+
| (instructions:u) | mean | range | count |
42+
|:----------------------------------:|:----:|:------------:|:-----:|
43+
| Regressions ❌ <br /> (primary) | 0.4% | [0.2%, 0.8%] | 10 |
44+
| Regressions ❌ <br /> (secondary) | 0.8% | [0.8%, 0.8%] | 1 |
45+
| Improvements ✅ <br /> (primary) | - | - | 0 |
46+
| Improvements ✅ <br /> (secondary) | - | - | 0 |
47+
| All ❌✅ (primary) | 0.4% | [0.2%, 0.8%] | 10 |
48+
49+
Still tracking down full set of root causes, but seems likely that at least the
50+
doc impact was due to changes in ByteStr (probably because rustdoc displays a
51+
lot of `Path`s which use ByteStr under the hood?). Some of the other impact has
52+
unclear origin at this point.
53+
54+
Remove `const FLAGS`. [#152791](https://github.com/rust-lang/rust/pull/152791) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=eeb94be79adc9df7a09ad0b2421f16e60e6d932c&end=b3869b94cd1ed4bfa2eb28f301535d5e9599c713&stat=instructions:u)
55+
56+
| (instructions:u) | mean | range | count |
57+
|:----------------------------------:|:----:|:------------:|:-----:|
58+
| Regressions ❌ <br /> (primary) | 0.2% | [0.1%, 0.5%] | 17 |
59+
| Regressions ❌ <br /> (secondary) | 0.3% | [0.1%, 0.6%] | 37 |
60+
| Improvements ✅ <br /> (primary) | - | - | 0 |
61+
| Improvements ✅ <br /> (secondary) | - | - | 0 |
62+
| All ❌✅ (primary) | 0.2% | [0.1%, 0.5%] | 17 |
63+
64+
Expected regressions in several benchmarks, simplifying code judged more
65+
important here than the slight perf loss.
66+
67+
#### Improvements
68+
69+
Simplify the canonical enum clone branches to a copy statement [#148034](https://github.com/rust-lang/rust/pull/148034) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=fbd6934114e905d3cc61adbbd7e4a355eac5d0d7&end=41198cb9659f3cc7c4a1a3ee97bc4af78bee403e&stat=instructions:u)
70+
71+
| (instructions:u) | mean | range | count |
72+
|:----------------------------------:|:-----:|:--------------:|:-----:|
73+
| Regressions ❌ <br /> (primary) | 0.7% | [0.7%, 0.7%] | 1 |
74+
| Regressions ❌ <br /> (secondary) | 0.2% | [0.2%, 0.2%] | 1 |
75+
| Improvements ✅ <br /> (primary) | -0.6% | [-2.1%, -0.2%] | 9 |
76+
| Improvements ✅ <br /> (secondary) | -0.5% | [-0.8%, -0.2%] | 14 |
77+
| All ❌✅ (primary) | -0.5% | [-2.1%, 0.7%] | 10 |
78+
79+
Two regressions look like noise. Nice win!
80+
81+
Rollup of 6 pull requests [#152934](https://github.com/rust-lang/rust/pull/152934) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=99246f40931ce1d1d29517b61bd70dd9ed4fbe5d&end=5fb2ff8611e5a4af4dc85977cfdecfbf3ffa6ade&stat=instructions:u)
82+
83+
| (instructions:u) | mean | range | count |
84+
|:----------------------------------:|:-----:|:--------------:|:-----:|
85+
| Regressions ❌ <br /> (primary) | - | - | 0 |
86+
| Regressions ❌ <br /> (secondary) | - | - | 0 |
87+
| Improvements ✅ <br /> (primary) | -0.5% | [-0.5%, -0.5%] | 2 |
88+
| Improvements ✅ <br /> (secondary) | -0.4% | [-0.6%, -0.2%] | 3 |
89+
| All ❌✅ (primary) | -0.5% | [-0.5%, -0.5%] | 2 |
90+
91+
Changes look maybe slightly above noise threshold but not by much. Possibly
92+
just noise, possibly due to giving LLVM more information in "Tighten the !range
93+
bounds on alignments in vtables" [#152929](https://github.com/rust-lang/rust/pull/152929).
94+
95+
Rollup of 18 pull requests [#152825](https://github.com/rust-lang/rust/pull/152825) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=e0cb264b814526acb82def4b5810e394a2ed294f&end=fbd6934114e905d3cc61adbbd7e4a355eac5d0d7&stat=instructions:u)
96+
97+
| (instructions:u) | mean | range | count |
98+
|:----------------------------------:|:-----:|:--------------:|:-----:|
99+
| Regressions ❌ <br /> (primary) | - | - | 0 |
100+
| Regressions ❌ <br /> (secondary) | 0.2% | [0.2%, 0.2%] | 1 |
101+
| Improvements ✅ <br /> (primary) | -0.2% | [-0.3%, -0.2%] | 5 |
102+
| Improvements ✅ <br /> (secondary) | -0.3% | [-0.6%, -0.1%] | 5 |
103+
| All ❌✅ (primary) | -0.2% | [-0.3%, -0.2%] | 5 |
104+
105+
The one regression looks like noise, so moved to improvements in this report.
106+
The rollup is large enough that the source of those improvements is hard to
107+
narrow down.
108+
109+
Bring back `enum DepKind`. [#152747](https://github.com/rust-lang/rust/pull/152747) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=ef70767064ab87b0a41400f69e1dc0b55c8d5284&end=59fd4ef94daa991e6797b5aa6127e824f3067def&stat=instructions:u)
110+
111+
| (instructions:u) | mean | range | count |
112+
|:----------------------------------:|:-----:|:--------------:|:-----:|
113+
| Regressions ❌ <br /> (primary) | - | - | 0 |
114+
| Regressions ❌ <br /> (secondary) | 0.8% | [0.8%, 0.8%] | 1 |
115+
| Improvements ✅ <br /> (primary) | -0.3% | [-0.7%, -0.2%] | 34 |
116+
| Improvements ✅ <br /> (secondary) | -0.3% | [-0.7%, -0.1%] | 10 |
117+
| All ❌✅ (primary) | -0.3% | [-0.7%, -0.2%] | 34 |
118+
119+
One regression is noise. Moving to improvements.
120+
121+
#### Mixed
122+
123+
Revert "Fix an ICE in the vtable iteration for a trait reference" [#152738](https://github.com/rust-lang/rust/pull/152738) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=1210e9fa3ee9c9712fa694f90f7a032f2f0fd786&end=3f6250a7bb79e600d9036bbc7c8f65af43933643&stat=instructions:u)
124+
125+
| (instructions:u) | mean | range | count |
126+
|:----------------------------------:|:-----:|:--------------:|:-----:|
127+
| Regressions ❌ <br /> (primary) | - | - | 0 |
128+
| Regressions ❌ <br /> (secondary) | 0.2% | [0.2%, 0.3%] | 4 |
129+
| Improvements ✅ <br /> (primary) | - | - | 0 |
130+
| Improvements ✅ <br /> (secondary) | -0.9% | [-1.3%, -0.5%] | 16 |
131+
| All ❌✅ (primary) | - | - | 0 |
132+
133+
Already triaged in advance (https://github.com/rust-lang/rust/pull/152738#issuecomment-3915321344):
134+
135+
> Revert of #152287 which landed in the rollup #152399 and should have seen the opposite effect.
136+
137+
No further action needed.
138+
139+
Audit Symbols [#152624](https://github.com/rust-lang/rust/pull/152624) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=1c00e989ca032d57e815e930fad00b61e65a1826&end=3fc37321c49cb8aed32da426de6145e6d57105da&stat=instructions:u)
140+
141+
| (instructions:u) | mean | range | count |
142+
|:----------------------------------:|:-----:|:--------------:|:-----:|
143+
| Regressions ❌ <br /> (primary) | 0.2% | [0.2%, 0.2%] | 1 |
144+
| Regressions ❌ <br /> (secondary) | 0.2% | [0.1%, 0.2%] | 2 |
145+
| Improvements ✅ <br /> (primary) | - | - | 0 |
146+
| Improvements ✅ <br /> (secondary) | -0.5% | [-0.7%, -0.3%] | 4 |
147+
| All ❌✅ (primary) | 0.2% | [0.2%, 0.2%] | 1 |
148+
149+
Perf regressions look close to noise thresholds (though maybe not entirely
150+
noise). Regardless I don't think it's worth trying to iterate on the specific
151+
set of symbols too much, it's effectively neutral (mean including non-relevant
152+
is +0.01% on instructions, +0.00% on cycles).
153+
154+
Just pass `Layout` directly to `box_new_uninit` [#152737](https://github.com/rust-lang/rust/pull/152737) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=5fb2ff8611e5a4af4dc85977cfdecfbf3ffa6ade&end=d8b2222b11fc03a31d43707c8c9006f025809ece&stat=instructions:u)
155+
156+
| (instructions:u) | mean | range | count |
157+
|:----------------------------------:|:-----:|:--------------:|:-----:|
158+
| Regressions ❌ <br /> (primary) | 1.3% | [1.3%, 1.3%] | 1 |
159+
| Regressions ❌ <br /> (secondary) | 0.7% | [0.5%, 0.8%] | 6 |
160+
| Improvements ✅ <br /> (primary) | -0.8% | [-1.1%, -0.5%] | 7 |
161+
| Improvements ✅ <br /> (secondary) | -1.8% | [-3.8%, -0.4%] | 9 |
162+
| All ❌✅ (primary) | -0.5% | [-1.1%, 1.3%] | 8 |
163+
164+
Some nice wins, a few regressions. Overall seems like a positive change. See
165+
also the note in the PR
166+
(https://github.com/rust-lang/rust/pull/152737#issuecomment-3941795762).
167+
168+
GVN: consider constants of primitive types as deterministic [#149366](https://github.com/rust-lang/rust/pull/149366) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=d3e8bd94cc042c65e30a6c60146a8a00531d1292&end=eeb94be79adc9df7a09ad0b2421f16e60e6d932c&stat=instructions:u)
169+
170+
| (instructions:u) | mean | range | count |
171+
|:----------------------------------:|:-----:|:--------------:|:-----:|
172+
| Regressions ❌ <br /> (primary) | 0.7% | [0.6%, 0.8%] | 3 |
173+
| Regressions ❌ <br /> (secondary) | - | - | 0 |
174+
| Improvements ✅ <br /> (primary) | -0.3% | [-0.7%, -0.2%] | 15 |
175+
| Improvements ✅ <br /> (secondary) | -0.3% | [-0.3%, -0.3%] | 1 |
176+
| All ❌✅ (primary) | -0.1% | [-0.7%, 0.8%] | 18 |
177+
178+
Some nice wins on a number of benchmarks. The regressions look potentially real
179+
but are generally outweighed by the improvements.

0 commit comments

Comments
 (0)