Several non-noise changes this week, with both improvements and regresions coming as a result. Overall compiler performance is roughly neutral across the week.
Triage done by @simulacrum. Revision range: a77f76e2..c65b2dc9
Summary:
| (instructions:u) | mean | range | count |
|---|---|---|---|
| Regressions ❌ (primary) | 0.4% | [0.2%, 1.4%] | 104 |
| Regressions ❌ (secondary) | 2.4% | [0.2%, 23.7%] | 81 |
| Improvements ✅ (primary) | -3.8% | [-26.1%, -0.3%] | 10 |
| Improvements ✅ (secondary) | -1.6% | [-4.6%, -0.5%] | 12 |
| All ❌✅ (primary) | 0.1% | [-26.1%, 1.4%] | 114 |
2 Regressions, 2 Improvements, 3 Mixed; 1 of them in rollups 51 artifact comparisons made in total
Use fulfillment in method probe, not evaluation #122317 (Comparison Link)
| (instructions:u) | mean | range | count |
|---|---|---|---|
| Regressions ❌ (primary) | 0.5% | [0.2%, 1.3%] | 38 |
| Regressions ❌ (secondary) | 4.2% | [0.5%, 23.6%] | 39 |
| Improvements ✅ (primary) | - | - | 0 |
| Improvements ✅ (secondary) | - | - | 0 |
| All ❌✅ (primary) | 0.5% | [0.2%, 1.3%] | 38 |
Some additional attempts to fix perf were done in a follow-up PR (https://github.com/rust-lang/rust/pull/124303) but did not pan out to be successful. It looks like this is a correctness fix, so we'll need to accept the regressions for now.
Stabilize Utf8Chunks #123909 (Comparison Link)
| (instructions:u) | mean | range | count |
|---|---|---|---|
| Regressions ❌ (primary) | 0.5% | [0.2%, 1.1%] | 11 |
| Regressions ❌ (secondary) | 0.8% | [0.3%, 1.2%] | 19 |
| Improvements ✅ (primary) | - | - | 0 |
| Improvements ✅ (secondary) | - | - | 0 |
| All ❌✅ (primary) | 0.5% | [0.2%, 1.1%] | 11 |
The regressions are in doc builds, but are not really expected from what is a relatively small change. Further investigation feels warranted (see https://github.com/rust-lang/rust/pull/123909#issuecomment-2082668500).
Rollup of 5 pull requests #124289 (Comparison Link)
| (instructions:u) | mean | range | count |
|---|---|---|---|
| Regressions ❌ (primary) | - | - | 0 |
| Regressions ❌ (secondary) | - | - | 0 |
| Improvements ✅ (primary) | -0.3% | [-0.3%, -0.2%] | 7 |
| Improvements ✅ (secondary) | -0.2% | [-0.2%, -0.2%] | 2 |
| All ❌✅ (primary) | -0.3% | [-0.3%, -0.2%] | 7 |
Unclear whether this is a genuine improvement. Performance seems to have re-regressed in #123126 (see Mixed results below), so this may just be bimodality of some kind.
Set writable and dead_on_unwind attributes for sret arguments #121298 (Comparison Link)
| (instructions:u) | mean | range | count |
|---|---|---|---|
| Regressions ❌ (primary) | 0.5% | [0.5%, 0.5%] | 1 |
| Regressions ❌ (secondary) | - | - | 0 |
| Improvements ✅ (primary) | -3.1% | [-26.0%, -0.3%] | 12 |
| Improvements ✅ (secondary) | -1.6% | [-4.4%, -0.5%] | 11 |
| All ❌✅ (primary) | -2.8% | [-26.0%, 0.5%] | 13 |
Enable CrateNum query feeding via TyCtxt #123126 (Comparison Link)
| (instructions:u) | mean | range | count |
|---|---|---|---|
| Regressions ❌ (primary) | 0.3% | [0.2%, 0.6%] | 19 |
| Regressions ❌ (secondary) | 0.4% | [0.3%, 0.5%] | 5 |
| Improvements ✅ (primary) | - | - | 0 |
| Improvements ✅ (secondary) | -0.7% | [-0.7%, -0.7%] | 2 |
| All ❌✅ (primary) | 0.3% | [0.2%, 0.6%] | 19 |
This looks like it‘s mostly just a regression to incremental. The PR description notes this is expected and sounds like there’s follow-up work planned (unclear whether it will help with performance though).
Stop using LLVM struct types for alloca #122053 (Comparison Link)
| (instructions:u) | mean | range | count |
|---|---|---|---|
| Regressions ❌ (primary) | 0.2% | [0.2%, 0.3%] | 8 |
| Regressions ❌ (secondary) | 0.4% | [0.2%, 1.1%] | 17 |
| Improvements ✅ (primary) | -1.9% | [-1.9%, -1.9%] | 1 |
| Improvements ✅ (secondary) | -0.4% | [-0.4%, -0.4%] | 1 |
| All ❌✅ (primary) | -0.0% | [-1.9%, 0.3%] | 9 |
Instruction counts are predominantly affected by some shuffling inside LLVM, but cycles are largely unaffected. See detailed analysis in https://github.com/rust-lang/rust/pull/122053#issuecomment-2074850501.
Abort a process when FD ownership is violated #124210 (Comparison Link)
| (instructions:u) | mean | range | count |
|---|---|---|---|
| Regressions ❌ (primary) | 0.4% | [0.4%, 0.4%] | 2 |
| Regressions ❌ (secondary) | - | - | 0 |
| Improvements ✅ (primary) | -0.8% | [-0.8%, -0.8%] | 1 |
| Improvements ✅ (secondary) | -0.1% | [-0.1%, -0.1%] | 2 |
| All ❌✅ (primary) | 0.0% | [-0.8%, 0.4%] | 3 |
Based on the self profile results I suspect this is caused by us generating more code in the downstream crate(s) as a result of the late-bound ub checks.