rustdoc: use the next solver for blanket impl synthesis #125907

fmease · 2024-06-03T01:38:52Z

Presumably due to better caching behavior, switching from the old to the next solver drastically improves the compile time on certain inputs. See #114891.
Fixes #114891.

Furthermore use an ObligationCtxt instead of operating on an InferCtxt directly and don't drop the obligations generated by the type relating.

For context, originally I just wanted to submit the infcx→ocx change. However, that regressed tests/rustdoc-ui/ice-blanket-impl-52873.rs (pass→overflow error) because with ocx.select_where_possible we would no longer suppress / defatalize (canonical) overflow errors. CC #54199 which introduced that special case. Obviously in the next solver overflows are non-fatal incl. ice-blanket-impl-52873.rs. Hence the switch now.

S-blocked Status: Marked as blocked ❌ on something else such as an RFC or other implementation work. on perf improvements for the next solver.

fmease · 2024-06-03T02:10:45Z

@bors try @rust-timer queue

fmease · 2024-06-03T02:12:33Z

I guess I'll also do a crater run 🤔? I don't want to accidentally make rustdoc hang or ice for half the ecosystem.

fmease · 2024-06-03T02:55:21Z

Let's try again. Bors, please respond this time around!

@bors try @rust-timer queue

fmease · 2024-06-03T02:59:05Z

@bors try

bors · 2024-06-03T03:01:17Z

⌛ Trying commit 737b8ac with merge dcbecf2...

…t, r=<try> rustdoc: use the next solver for blanket impl synthesis Furthermore use an `ObligationCtxt` instead of operating on an `InferCtxt` directly and don't drop the obligations generated by the type relating (fixes a FIXME). Originally I just wanted to submit the infcx→ocx change. However, that regressed `tests/rustdoc-ui/ice-blanket-impl-52873.rs` (pass→overflow error) because with `ocx.select_where_possible` we would no longer suppress / defatalize (canonical) overflow errors on the last obligation we register. CC rust-lang#54199 which introduced that special case. Obviously in the next solver overflows are non-fatal incl. `ice-blanket-impl-52873.rs`. Hence the switch now. Note that I wanted to hold off on switching to the new solver since it makes rust-lang#114891 go from long I-compiletime to I-compilemem + I-hang + eventual death by the OOM killer. So I don't know maybe we should block this PR on me finding a rustc reproducer for the rustdoc issue rust-lang#114891 (this is on my agenda) to be able to properly investigate the underlying next-solver issue.

bors · 2024-06-03T04:39:50Z

☀️ Try build successful - checks-actions
Build commit: dcbecf2 (dcbecf205cd3ecd75a91d7290ac77e62f9c4bad2)

rust-timer · 2024-06-03T06:07:41Z

Finished benchmarking commit (dcbecf2): comparison URL.

Overall result: ❌ regressions - ACTION NEEDED

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

This is a highly reliable metric that was used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	988.1%	[0.6%, 18344.5%]	19
Regressions ❌ (secondary)	5.4%	[1.0%, 11.2%]	5
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	988.1%	[0.6%, 18344.5%]	19

Max RSS (memory usage)

Results (primary 23.3%, secondary 5.2%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	23.3%	[3.5%, 54.7%]	11
Regressions ❌ (secondary)	5.2%	[4.4%, 6.1%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	23.3%	[3.5%, 54.7%]	11

Cycles

Results (primary 824.6%, secondary 7.0%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	824.6%	[2.2%, 13651.3%]	17
Regressions ❌ (secondary)	11.5%	[10.8%, 12.2%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-2.0%	[-2.0%, -2.0%]	1
All ❌✅ (primary)	824.6%	[2.2%, 13651.3%]	17

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 667.872s -> 669.107s (0.18%)
Artifact size: 318.92 MiB -> 318.92 MiB (0.00%)

fmease · 2024-06-03T06:12:08Z

okay lol, that is crazy

…ket-impls, r=GuillaumeGomez rustdoc: add a regression test for a former blanket impl synthesis ICE Fixes rust-lang#119792 (also passes in rust-lang#125907 in case you were wondering). r? rustdoc

fmease · 2024-08-15T11:20:01Z

@bors try @rust-timer queue

…t, r=<try> rustdoc: use the next solver for blanket impl synthesis Furthermore use an `ObligationCtxt` instead of operating on an `InferCtxt` directly and don't drop the obligations generated by the type relating. Originally I just wanted to submit the infcx→ocx change. However, that regressed `tests/rustdoc-ui/ice-blanket-impl-52873.rs` (pass→overflow error) because with `ocx.select_where_possible` we would no longer suppress / defatalize (canonical) overflow errors. CC rust-lang#54199 which introduced that special case. Obviously in the next solver overflows are non-fatal incl. `ice-blanket-impl-52873.rs`. Hence the switch now. ~~Note that I wanted to hold off on switching to the new solver since it makes rust-lang#114891 go from long I-compiletime to I-compilemem + I-hang + eventual death by the OOM killer. So I don't know maybe we should block this PR on me finding a rustc reproducer for the rustdoc issue rust-lang#114891 (this is on my agenda) to be able to properly investigate the underlying next-solver issue.~~ Fixes rust-lang#114891.

bors · 2024-08-15T11:21:11Z

⌛ Trying commit e1e3180 with merge c9f5be2...

bors · 2024-08-15T13:17:36Z

☀️ Try build successful - checks-actions
Build commit: c9f5be2 (c9f5be263f484d4930b7c37ef7bd737451891d8c)

rust-timer · 2024-08-15T14:47:25Z

Finished benchmarking commit (c9f5be2): comparison URL.

Overall result: ❌ regressions - ACTION NEEDED

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

This is a highly reliable metric that was used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	9.7%	[0.8%, 46.3%]	17
Regressions ❌ (secondary)	2.0%	[0.4%, 5.1%]	5
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	9.7%	[0.8%, 46.3%]	17

Max RSS (memory usage)

Results (primary 4.9%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	4.9%	[1.6%, 15.5%]	9
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	4.9%	[1.6%, 15.5%]	9

Cycles

Results (primary 11.2%, secondary 4.3%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	11.2%	[2.3%, 33.7%]	12
Regressions ❌ (secondary)	4.3%	[3.6%, 5.1%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	11.2%	[2.3%, 33.7%]	12

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 751.515s -> 752.426s (0.12%)
Artifact size: 339.18 MiB -> 339.16 MiB (-0.00%)

GuillaumeGomez · 2024-08-15T14:49:24Z

That's quite a big regression. Do you have ideas on how to limit it or is it not possible for the moment?

lcnr · 2024-08-15T16:24:21Z

the following PRs will likely reduce the perf impact here:

lcnr · 2024-08-15T16:25:23Z

That's quite a big regression. Do you have ideas on how to limit it or is it not possible for the moment?

we still haven't optimized the new solver at all really apart from making sure the architecture itself is able to be performant, so there are still a lot of low-hanging fruits in the new solver.

GuillaumeGomez · 2024-08-15T18:33:26Z

Thanks for the information! Don't hesitate if there is anything I can help with.

fmease · 2024-10-16T12:19:45Z

@bors try @rust-timer queue

…t, r=<try> rustdoc: use the next solver for blanket impl synthesis Presumably due to better caching behavior, switching from the old to the next solver *drastically* improves the compile time on certain inputs. See rust-lang#114891. Fixes rust-lang#114891. Furthermore use an `ObligationCtxt` instead of operating on an `InferCtxt` directly and don't drop the obligations generated by the type relating. For context, originally I just wanted to submit the infcx→ocx change. However, that regressed `tests/rustdoc-ui/ice-blanket-impl-52873.rs` (pass→overflow error) because with `ocx.select_where_possible` we would no longer suppress / defatalize (canonical) overflow errors. CC rust-lang#54199 which introduced that special case. Obviously in the next solver overflows are non-fatal incl. `ice-blanket-impl-52873.rs`. Hence the switch now. https://github.com/rust-lang/rust/labels/S-blocked on perf improvements for the next solver.

bors · 2024-10-16T12:20:57Z

⌛ Trying commit c32dd09 with merge 9daeecb...

bors · 2024-10-16T14:04:48Z

☀️ Try build successful - checks-actions
Build commit: 9daeecb (9daeecbe7b57dd50c95c08cc8ea2ecdf72c1876e)

rust-timer · 2024-10-16T15:22:20Z

Finished benchmarking commit (9daeecb): comparison URL.

Overall result: ❌✅ regressions and improvements - please read the text below

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

This is the most reliable metric that we have; it was used to determine the overall result at the top of this comment. However, even this metric can sometimes exhibit noise.

	mean	range	count
Regressions ❌ (primary)	10.0%	[0.3%, 59.6%]	19
Regressions ❌ (secondary)	2.3%	[0.5%, 5.8%]	5
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-1.6%	[-1.6%, -1.6%]	1
All ❌✅ (primary)	10.0%	[0.3%, 59.6%]	19

Max RSS (memory usage)

Results (primary 5.1%, secondary -2.3%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	5.1%	[1.6%, 15.0%]	12
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-2.3%	[-2.4%, -2.1%]	2
All ❌✅ (primary)	5.1%	[1.6%, 15.0%]	12

Cycles

Results (primary 9.6%, secondary 4.7%)

This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.

	mean	range	count
Regressions ❌ (primary)	9.6%	[0.6%, 42.0%]	17
Regressions ❌ (secondary)	4.7%	[4.4%, 4.9%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	9.6%	[0.6%, 42.0%]	17

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 780.385s -> 781.632s (0.16%)
Artifact size: 333.66 MiB -> 333.65 MiB (-0.00%)

fmease assigned GuillaumeGomez and lcnr Jun 3, 2024

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-rustdoc Relevant to the rustdoc team, which will review and decide on the PR/issue. labels Jun 3, 2024

fmease added the S-blocked Status: Marked as blocked ❌ on something else such as an RFC or other implementation work. label Jun 3, 2024

fmease mentioned this pull request Jun 3, 2024

rustdoc: add a regression test for a former blanket impl synthesis ICE #125909

Merged

This comment was marked as resolved.

Sign in to view

rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jun 3, 2024

This comment was marked as resolved.

Sign in to view