filled in the triage report.

Author: pnkfelix

triage/2023-07-18.md

@@ -1,6 +1,8 @@
 # 2023-07-18 Triage Log
 
-TODO: Summary
+A lot of spurious results in the regressions this week. However, we did see some
+real gains with PR #113609, with nearly 40 real-world benchmarks improving their
+check-build performance by >=1%.
 
 Triage done by **@pnkfelix**.
 Revision range: [1d4f5aff..6b9236ed](https://perf.rust-lang.org/?start=1d4f5affbdee00c816f961c227c6b28a3e725ce6&end=6b9236ed5a58879969ddcf4c35a6a24f6260e979&absolute=false&stat=instructions%3Au)
@@ -16,7 +18,7 @@ Revision range: [1d4f5aff..6b9236ed](https://perf.rust-lang.org/?start=1d4f5affb
 | All ❌✅ (primary)                 | -1.0% | [-3.7%, 3.0%]  | 57    |
 
 
-5 Regressions, 5 Improvements, 5 Mixed; ??? of them in rollups
+5 Regressions, 5 Improvements, 5 Mixed; 2 of them in rollups
 57 artifact comparisons made in total
 
 #### Regressions
@@ -31,6 +33,10 @@ miri: protect Move() function arguments during the call [#113569](https://github
 | Improvements ✅ <br /> (secondary) | -    | -            | 0     |
 | All ❌✅ (primary)                 | 0.8% | [0.5%, 1.1%] | 13    |
 
+* RalfJ is investigating; has potential fix up in [PR #113630](https://github.com/rust-lang/rust/pull/113630), ...
+* ... but its not totally certain that PR is a real fix (i.e. the regression may already have been masked or otherwise resolved independently).
+* But meanwhile, I am hypothesizing that the regression reported here is spurious ([discussion](https://rust-lang.zulipchat.com/#narrow/stream/247081-t-compiler.2Fperformance/topic/perf.20triage.202023-07-18/near/376402031))
+* marking as triaged
 
 Ignore flaky clippy tests. [#113621](https://github.com/rust-lang/rust/pull/113621) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=1e6db3486da0f6474920477c01fca90453b7c428&end=33a2c2487ac5d9927830ea4c1844335c6b9f77db&stat=instructions:u)
 
@@ -42,6 +48,7 @@ Ignore flaky clippy tests. [#113621](https://github.com/rust-lang/rust/pull/1136
 | Improvements ✅ <br /> (secondary) | -    | -            | 0     |
 | All ❌✅ (primary)                 | 1.0% | [0.6%, 1.2%] | 7     |
 
+* already marked as triaged (its noise)
 
 Rollup of 6 pull requests [#113673](https://github.com/rust-lang/rust/pull/113673) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=a161ab00dbf660dd587ee42a8c855bac94047ddb&end=7bd81ee1902c049691d0a1f03be5558bee51d100&stat=instructions:u)
 
@@ -53,6 +60,7 @@ Rollup of 6 pull requests [#113673](https://github.com/rust-lang/rust/pull/11367
 | Improvements ✅ <br /> (secondary) | -    | -            | 0     |
 | All ❌✅ (primary)                 | 1.2% | [1.2%, 1.2%] | 1     |
 
+* addressed by PR #113697, already marked as triaged.
 
 Add even more GHA log groups [#113514](https://github.com/rust-lang/rust/pull/113514) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=ad963232d9b987d66a6f8e6ec4141f672b8b9900&end=4d6e4260b2de66a356a2536320f339467dff0d2b&stat=instructions:u)
 
@@ -64,6 +72,8 @@ Add even more GHA log groups [#113514](https://github.com/rust-lang/rust/pull/11
 | Improvements ✅ <br /> (secondary) | -    | -            | 0     |
 | All ❌✅ (primary)                 | -    | -            | 0     |
 
+* PR author says "this doesn't affect how the compiler was built at all, the perf regression must be spurious."
+* I agree, marking as triaged
 
 Rollup of 3 pull requests [#113738](https://github.com/rust-lang/rust/pull/113738) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=4124617c6ebe6b5b4df2c8c8400e87389c377dcc&end=2c718d12599ca00dc3957ec604cb3e87fe437124&stat=instructions:u)
 
@@ -145,6 +155,9 @@ Eliminate ZST allocations in `Box` and `Vec` [#113113](https://github.com/rust-l
 | Improvements ✅ <br /> (secondary) | -     | -              | 0     |
 | All ❌✅ (primary)                 | 0.6%  | [-1.7%, 3.5%]  | 5     |
 
+* regressions here were anticipated and unavoidable. This is a bug fix.
+* Marking as triaged.
+
 
 Enable MIR reference propagation by default [#109025](https://github.com/rust-lang/rust/pull/109025) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=5767cad9b8b58a5c0521c5ddd652a24d0c65f466&end=079e544174b79c372b4b4b473a01d699f128c2de&stat=instructions:u)
 
@@ -156,6 +169,9 @@ Enable MIR reference propagation by default [#109025](https://github.com/rust-la
 | Improvements ✅ <br /> (secondary) | -0.6% | [-1.1%, -0.4%] | 6     |
 | All ❌✅ (primary)                 | 0.8%  | [-1.0%, 2.5%]  | 18    |
 
+* This is turning on a MIR pass at lower optimizations levels, so its expected that it would cause the compiler to do more work.
+* its clear from the perf runs on the PR itself that the PR author already put in much effort to make the pass faster than it had started out.
+* marking as triaged.
 
 Resurrect: rustc_target: Add alignment to indirectly-passed by-value types, correcting the alignment of byval on x86 in the process. [#112157](https://github.com/rust-lang/rust/pull/112157) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=4d6e4260b2de66a356a2536320f339467dff0d2b&end=7a17f577b3d437179cad254e299b2ace972487c5&stat=instructions:u)
 
@@ -167,6 +183,9 @@ Resurrect: rustc_target: Add alignment to indirectly-passed by-value types, corr
 | Improvements ✅ <br /> (secondary) | -2.8% | [-2.9%, -2.8%] | 6     |
 | All ❌✅ (primary)                 | 1.7%  | [1.5%, 2.0%]   | 6     |
 
+* The six primary regressions were all to variants of diesel (all of check/debug/opt) for variious full and incr-full scenarios.
+* It isn't noise, there seems to be a clear cliff that starts with this PR when looking at the graph starting from 2023-0702.
+* not marking as triaged yet, but was tempted to do so, because this PR is a prerequiste for unlocking various memcpy optimizations added by pcwalton to LLVM
 
 Remove unneeded handling for `ExternalLocation::Unknown` in rustdoc render context [#113697](https://github.com/rust-lang/rust/pull/113697) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=425726d46b637f94995ca1c95f4f16cdb903d30e&end=4c8bb79d9f565115637cc6da739f8389e79f3a29&stat=instructions:u)
 
@@ -178,6 +197,9 @@ Remove unneeded handling for `ExternalLocation::Unknown` in rustdoc render conte
 | Improvements ✅ <br /> (secondary) | -     | -              | 0     |
 | All ❌✅ (primary)                 | -1.1% | [-1.1%, -1.1%] | 1     |
 
+* this was mentioned up above when I was talking about PR #113673
+* the secondary regressions are all to ctfe-stress-5, which lqd says was noisy at this time.
+* marked as triaged.
 
 Add support for allocators in `Rc` & `Arc` [#89132](https://github.com/rust-lang/rust/pull/89132) [(Comparison Link)](https://perf.rust-lang.org/compare.html?start=1787f312907eee9a4a9ac010985d725ad9a03ab6&end=da6b55cc5eaf76ed6acb7dc2f7d611e32af7c9a7&stat=instructions:u)
 
@@ -189,6 +211,11 @@ Add support for allocators in `Rc` & `Arc` [#89132](https://github.com/rust-lang
 | Improvements ✅ <br /> (secondary) | -0.7% | [-0.8%, -0.5%] | 4     |
 | All ❌✅ (primary)                 | -0.3% | [-1.3%, 0.8%]  | 2     |
 
+Visiting for perf-triage
+
+* primary regression was to image-0.24.1 opt full by 0.79%
+* I think this is just noise. From the [graph](https://perf.rust-lang.org/index.html?start=2023-06-17&end=2023-07-18&benchmark=image-0.24.1&profile=opt&scenario=full&stat=instructions%3Au&kind=raw), it seems like image has unpredictably jumped up and down between two plateaus since PR #113113 (a PR discussed up above that changed low level allocation procotol code in Box and Vec, and thus might be expected to have some weird follow-on effects).
+* marked as triaged
 
 #### Untriaged Pull Requests