Extract suitable code from rustc_query_impl into a new crate rustc_query_misc

[Zoxc]

This extracts code from rustc_query_impl into a new crate rustc_query_misc in order to reduce the compile time of rustc_query_impl. The compile time of the new crate is roughly 60% of the remaining rustc_query_impl .

I've picked code which should not impact performance or result in duplicate generic code generation. Encoding and decoding of query results and profiling handling is moved along with some other minor methods which use dynamic dispatch.

r? @cjgillot

[rustbot]

These commits modify the Cargo.lock file. Unintentional changes to Cargo.lock can be introduced when switching branches and rebasing PRs.

If this was unintentional then you should revert the changes before this PR is merged.
Otherwise, you can ignore this comment.

[cjgillot]

I'm not sure I understand the benefit. This crate is still on the critical path middle -> query_misc -> query_impl, so at worst we could lose some codegen parallelism.

[bjorn3]

Thanks to build pipelining cargo can invoke rustc for a dependent crate while dependencies are still busy with codegen. Only crate metadata writing needs to have finished.

[Zoxc]

It's a benefit when compiling with codegen-units=1 since rustc_query_impl and rustc_query_misc can each get a core for code generation instead of rustc_query_impl being stuck on one. I use that configuration locally for performance benchmarks and CI should switch to that soon as it compiles faster and results in higher code quality.

[bors]

☔ The latest upstream changes (presumably #115326) made this pull request unmergeable. Please resolve the merge conflicts.

[cjgillot]

It's a benefit when compiling with codegen-units=1 since rustc_query_impl and rustc_query_misc can each get a core for code generation instead of rustc_query_impl being stuck on one. I use that configuration locally for performance benchmarks and CI should switch to that soon as it compiles faster and results in higher code quality.

So we split the crate in 2 codegen units, to counteract the fact we ask for only 1?

The split does not follow a functional separation of the crate. The large macro is still there, structured the same way. How should we chose in which crate code goes?

Is there a way to get rid of the dependency between query_misc and query_impl?

[Zoxc]

So we split the crate in 2 codegen units, to counteract the fact we ask for only 1?

Basically, but we can do it without a performance hit, unlike the automatic partitioning.

The split does not follow a functional separation of the crate. The large macro is still there, structured the same way. How should we chose in which crate code goes?

In general, hot direct calls stays in the same crate, while cold or indirect calls can cross crate boundaries.

Is there a way to get rid of the dependency between query_misc and query_impl?

Yes, but it's not very beneficial as they mostly spend their time in code generation. It's also would require another copy of the macro mentioned.

[bors]

☔ The latest upstream changes (presumably #115518) made this pull request unmergeable. Please resolve the merge conflicts.

[Zoxc]

Here's a link to the new compilation timings. rustc_middle is back on top as the biggest crate. rustc_borrowck is the new tail.

[bors]

☔ The latest upstream changes (presumably #115820) made this pull request unmergeable. Please resolve the merge conflicts.

[bors]

☔ The latest upstream changes (presumably #115920) made this pull request unmergeable. Please resolve the merge conflicts.

[wesleywiser]

@bors try @rust-timer queue

[bors]

⌛ Trying commit c9ce621 with merge 3d352fd...

[cjgillot]

Making my review in #115118 (comment) more explicit.

I think this PR introduces complexity with no reason beyond bootstrap time.
The query system is already very complex. Splitting into an additional crate should be justified by a reduction in that complexity.

I would accept a PR splitting rustc_query_impl along logical / API boundaries.
Even better, moving code between rustc_query_system and rustc_query_impl to help modularisation and decoupling.

[bors]

☀️ Try build successful - checks-actions
Build commit: 3d352fd (3d352fd278737661390c7291be0afe22e8971a0c)

[rust-timer]

Finished benchmarking commit (3d352fd): 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)	1.5%	[0.2%, 7.3%]	21
Regressions ❌ (secondary)	2.1%	[0.1%, 6.3%]	73
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	1.5%	[0.2%, 7.3%]	21

Max RSS (memory usage)

Results

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)	-	-	0
Regressions ❌ (secondary)	5.6%	[5.6%, 5.6%]	1
Improvements ✅ (primary)	-3.1%	[-3.1%, -3.1%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-3.1%	[-3.1%, -3.1%]	1

Cycles

Results

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)	4.8%	[0.7%, 8.4%]	5
Regressions ❌ (secondary)	3.0%	[1.6%, 4.8%]	27
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	4.8%	[0.7%, 8.4%]	5

Binary size

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

Bootstrap: 621.055s -> 622.704s (0.27%)
Artifact size: 271.92 MiB -> 272.25 MiB (0.12%)

[apiraino]

Flipping the review switch to the author, I think there are some design questions at this comment on how to proceed.

@rustbot author

[JohnCSimon]

Ping from triage:
@Zoxc - can you please address the questions form @apiraino ? Thank you.