2023-10-03 Triage Log

A week completely free of pure regressions! The compiler has definitely come out of this week a decent amount faster and less memory hungry than before with the large gain coming from building the compiler with a single CGU on x64 Linux. This not only allows LLVM to do more optimizations across the entire compiler, but should hopefully also result in less non-deterministic performance regressions in the future. This improvement largely comes only at the expense of a few more minutes spent when bootstrapping the compiler.

Triage done by @rylev. Revision range: 27b4eb..9998f4

Summary:

(instructions:u)meanrangecount
Regressions ❌
(primary)		1.0%[0.3%, 6.1%]41
Regressions ❌
(secondary)		2.0%[0.9%, 7.8%]21
Improvements ✅
(primary)		-1.3%[-5.1%, -0.2%]134
Improvements ✅
(secondary)		-1.8%[-6.9%, -0.2%]175
All ❌✅ (primary)-0.8%[-5.1%, 6.1%]175

0 Regressions, 2 Improvements, 4 Mixed; 0 of them in rollups 74 artifact comparisons made in total

Regressions

Improvements

rustdoc: speed up processing of cross-crate fns to fix a perf regression #116195 (Comparison Link)

(instructions:u)meanrangecount
Regressions ❌
(primary)		--0
Regressions ❌
(secondary)		--0
Improvements ✅
(primary)		-0.3%[-0.5%, -0.2%]6
Improvements ✅
(secondary)		-0.6%[-0.8%, -0.3%]11
All ❌✅ (primary)-0.3%[-0.5%, -0.2%]6

Cleanup number handling in match exhaustiveness #116281 (Comparison Link)

(instructions:u)meanrangecount
Regressions ❌
(primary)		--0
Regressions ❌
(secondary)		--0
Improvements ✅
(primary)		-0.7%[-1.5%, -0.3%]3
Improvements ✅
(secondary)		--0
All ❌✅ (primary)-0.7%[-1.5%, -0.3%]3

Mixed

Don't use a thread to load the dep graph #116109 (Comparison Link)

(instructions:u)meanrangecount
Regressions ❌
(primary)		0.8%[0.5%, 1.1%]2
Regressions ❌
(secondary)		0.5%[0.1%, 1.2%]3
Improvements ✅
(primary)		-0.2%[-0.3%, -0.2%]4
Improvements ✅
(secondary)		--0
All ❌✅ (primary)0.1%[-0.3%, 1.1%]6
  • A very small regression that basically results in net zero change to performance.

Build rustc with a single CGU on x64 Linux #115554 (Comparison Link)

(instructions:u)meanrangecount
Regressions ❌
(primary)		0.8%[0.3%, 2.6%]39
Regressions ❌
(secondary)		1.6%[0.7%, 2.5%]23
Improvements ✅
(primary)		-1.2%[-3.9%, -0.2%]89
Improvements ✅
(secondary)		-1.6%[-3.5%, -0.4%]124
All ❌✅ (primary)-0.6%[-3.9%, 2.6%]128
  • The change is overwhelmingly positive (especially when taking other measures like max RSS and artifact size) making the small regressions acceptable
  • Take a look at cycles for a nearly completely regression free view of the positive nature of this change.

Rebased: Mark drop calls in landing pads cold instead of noinline #102099 (Comparison Link)

(instructions:u)meanrangecount
Regressions ❌
(primary)		1.5%[0.3%, 5.9%]12
Regressions ❌
(secondary)		2.7%[0.4%, 7.7%]5
Improvements ✅
(primary)		-0.5%[-4.5%, -0.2%]78
Improvements ✅
(secondary)		-0.9%[-6.9%, -0.1%]65
All ❌✅ (primary)-0.2%[-4.5%, 5.9%]90
  • Largely positive with a few large regressions caused by more time spent in LLVM. This seems to be expected and acceptable.

Make subtyping explicit in MIR #115025 (Comparison Link)

(instructions:u)meanrangecount
Regressions ❌
(primary)		0.6%[0.4%, 0.8%]2
Regressions ❌
(secondary)		0.5%[0.1%, 1.1%]17
Improvements ✅
(primary)		-0.5%[-0.6%, -0.5%]2
Improvements ✅
(secondary)		-0.5%[-0.5%, -0.5%]1
All ❌✅ (primary)0.0%[-0.6%, 0.8%]4
  • Small regression that is basically only seen in some secondary benchmarks. Unlikely to be worth deep investigation.