Move DepKind and query TLS to rustc_query_system by cjgillot · Pull Request #86038 · rust-lang/rust

cjgillot · 2021-06-05T20:52:55Z

r? @ghost

cjgillot · 2021-06-05T20:53:22Z

@bors try @rust-timer queue

rust-timer · 2021-06-05T20:53:24Z

Awaiting bors try build completion.

@rustbot label: +S-waiting-on-perf

bors · 2021-06-05T20:53:33Z

⌛ Trying commit 84a156cb46350ea4ca35884c4a910b62343f0677 with merge b1ce53c0cc21a9301621cb4076f77bf3ec601d7b...

bors · 2021-06-05T21:53:57Z

☀️ Try build successful - checks-actions
Build commit: b1ce53c0cc21a9301621cb4076f77bf3ec601d7b (b1ce53c0cc21a9301621cb4076f77bf3ec601d7b)

rust-timer · 2021-06-05T21:53:59Z

Queued b1ce53c0cc21a9301621cb4076f77bf3ec601d7b with parent 9104c89, future comparison URL.

rust-timer · 2021-06-06T00:28:11Z

Finished benchmarking try commit (b1ce53c0cc21a9301621cb4076f77bf3ec601d7b): comparison url.

Benchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. Please note that if the perf results are neutral, you should likely undo the rollup=never given below by specifying rollup- to bors.

Importantly, though, if the results of this run are non-neutral do not roll this PR up -- it will mask other regressions or improvements in the roll up.

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

mati865 · 2021-06-06T00:34:41Z

31% diesel-doc regression is red herring: #85993 (comment)

compiler/rustc_query_system/src/dep_graph/dep_kind.rs

compiler/rustc_query_system/src/decl.rs

bjorn3 · 2021-07-01T19:11:05Z

compiler/rustc_query_system/src/tls.rs

Could these Options be removed and instead clear the ImplicitContext in the tls when they would be None?

I don't think so. The implicitctxt is still required to detect cycles and to print query stack even when the dependencies are ignored.

And the other fields? At least one can be made non-optional, right? I don't see what the difference between no implicitctxt and an empty implicitctxt would be.

Indeed, query should not need to be an option.

JohnCSimon · 2021-07-31T20:54:31Z

Ping from triage:
@cjgillot assigning back to author to resolve merge conflicts

camelid · 2021-08-20T00:22:29Z

triage: Is this PR experimental? If so, you can replace S-waiting-on-author with S-experimental to prevent it from being triaged.

bors · 2022-02-24T22:57:42Z

☀️ Try build successful - checks-actions
Build commit: 38e911f90695efec7806bf6f6c1dccc01a05c130 (38e911f90695efec7806bf6f6c1dccc01a05c130)

rust-timer · 2022-02-24T22:57:44Z

Queued 38e911f90695efec7806bf6f6c1dccc01a05c130 with parent 4b043fa, future comparison URL.

rust-timer · 2022-02-25T00:40:48Z

Finished benchmarking commit (38e911f90695efec7806bf6f6c1dccc01a05c130): comparison url.

Summary: This benchmark run shows 98 relevant improvements 🎉 to instruction counts.

Arithmetic mean of relevant improvements: -0.7%
Largest improvement in instruction counts: -3.9% on full builds of keccak check

If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf.

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 led to changes in compiler perf.

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

Mark-Simulacrum

Would be nice to squash the commits in, and modulo some nits this seems good to me. I think in a fully ideal world the query system would be less tightly coupled to the particular queries, but in practice we both care about performance and the abstractions actually hurt us more than help, I suspect.

Mark-Simulacrum · 2022-02-25T01:48:13Z

compiler/rustc_middle/src/ty/context.rs

nit: "current" TyCtxt seems a little misleading -- we really expect just one, right? I wonder if there's some slight alternate framing that would work well here.

FWIW, I think the thread_local nature here may be important for things like unit tests and such which want to spin up their own TyCtxt's -- I'm not sure we should drop the TLS variable. At the very least, we should make sure we don't have UB or so if there's conflicts.

There may be multiple TyCtxt in the same process if you use rustc_driver on multiple threads.

Mark-Simulacrum · 2022-02-25T01:49:39Z

compiler/rustc_middle/src/ty/context.rs

Minor nit: Maybe TLV should be an AtomicPtr? I'm not sure if that causes us lifetime troubles.

There is only one GlobalCtxt ever.

cjgillot · 2022-04-16T15:48:38Z

I finally got back to this PR.

In this new version, I stop moving the macros to rustc_query_system, and side with @Zoxc's suggestion to make DepKind a newtype over u16. This may be useful if we want to define queries in crates other than rustc_middle one day.

The global pointer to GlobalCtxt is now a thread-local *const GlobalCtxt<'static>. Since all the jobs in the compilation thread pool have the same GlobalCtxt, this should not trigger any issue, while at the same time allowing to run two instances of rustc in different threads as asked by @bjorn3.

@bors try @rust-timer queue

rust-timer · 2022-04-16T15:48:39Z

Awaiting bors try build completion.

@rustbot label: +S-waiting-on-perf

bors · 2022-04-16T15:48:46Z

⌛ Trying commit 4387490 with merge 0794d7f4f502d8a0e9a4a71f7669e4adda756b81...

bors · 2022-04-16T17:22:27Z

☀️ Try build successful - checks-actions
Build commit: 0794d7f4f502d8a0e9a4a71f7669e4adda756b81 (0794d7f4f502d8a0e9a4a71f7669e4adda756b81)

rust-timer · 2022-04-16T17:22:29Z

Queued 0794d7f4f502d8a0e9a4a71f7669e4adda756b81 with parent febce1f, future comparison URL.

rust-timer · 2022-04-16T19:18:12Z

Finished benchmarking commit (0794d7f4f502d8a0e9a4a71f7669e4adda756b81): comparison url.

Summary:

Primary benchmarks: mixed results
Secondary benchmarks: mixed results

	Regressions 😿 (primary)	Regressions 😿 (secondary)	Improvements 🎉 (primary)	Improvements 🎉 (secondary)	All 😿 🎉 (primary)
count¹	10	21	9	35	19
mean²	1.0%	2.6%	-0.4%	-0.6%	0.4%
max	3.4%	8.2%	-0.7%	-1.6%	3.4%

If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf.

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-review -S-waiting-on-perf +perf-regression

number of relevant changes ↩
the arithmetic mean of the percent change ↩

cjgillot · 2022-04-16T21:44:07Z

@bors try @rust-timer queue

rust-timer · 2022-04-16T21:44:08Z

Awaiting bors try build completion.

@rustbot label: +S-waiting-on-perf

bors · 2022-04-16T21:44:15Z

⌛ Trying commit 8b8c12e with merge 8227b19194e5b9a2926d15786b99e6cc9f73389f...

bors · 2022-04-16T23:34:38Z

☀️ Try build successful - checks-actions
Build commit: 8227b19194e5b9a2926d15786b99e6cc9f73389f (8227b19194e5b9a2926d15786b99e6cc9f73389f)

rust-timer · 2022-04-16T23:34:40Z

Queued 8227b19194e5b9a2926d15786b99e6cc9f73389f with parent 878c783, future comparison URL.

rust-timer · 2022-04-17T01:11:21Z

Finished benchmarking commit (8227b19194e5b9a2926d15786b99e6cc9f73389f): comparison url.

Summary:

Primary benchmarks: mixed results
Secondary benchmarks: mixed results

	Regressions 😿 (primary)	Regressions 😿 (secondary)	Improvements 🎉 (primary)	Improvements 🎉 (secondary)	All 😿 🎉 (primary)
count¹	4	13	28	42	32
mean²	2.0%	3.7%	-0.4%	-0.5%	-0.1%
max	3.2%	7.9%	-0.7%	-0.9%	3.2%

If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf.

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-review -S-waiting-on-perf +perf-regression

number of relevant changes ↩
the arithmetic mean of the percent change ↩

Zoxc · 2022-04-17T05:01:23Z

compiler/rustc_query_system/src/tls.rs

+/// query. It is updated when
+/// executing a new query.
+#[derive(Clone, Default)]
+struct ImplicitCtxt<'a> {


You should keep the tcx field as a *const (). That lets you move more of the tls module intact and the parallel compiler doesn't regress. with_opt can convert it back in rustc_middle.

You'll want to add an assertion about GlobalCtxt being Sync too, since the type is now erased from ImplicitCtxt.

Zoxc · 2022-04-17T05:11:34Z

compiler/rustc_interface/src/util.rs

+    // Extend the lifetime of the GlobalCtxt so the new thread can know of it.
+    // The current thread will not free it, it is deadlocked.
+    let tcx: &'static GlobalCtxt<'static> =
+        tls::with(|tcx| &*(*tcx as *const GlobalCtxt<'_>).cast());


This is incorrect since it's only set on one of Rayon's threads. The deadlock handler can be called on any thread. The current state of things is also broken to be fair.

We can not use the Rayon "Task Local Value" either, since the deadlock handler can also be called from a thread with no rustc task is running. My solution to this was to point all threads to a lock which holds the GlobalCtxt when available. It was reverted in #74969.

rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jun 5, 2021

This comment has been minimized.

Sign in to view

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-perf Status: Waiting on a perf run to be completed. labels Jun 6, 2021

bjorn3 reviewed Jun 6, 2021

View reviewed changes

compiler/rustc_query_system/src/dep_graph/dep_kind.rs Outdated Show resolved Hide resolved

bjorn3 reviewed Jun 6, 2021

View reviewed changes

compiler/rustc_query_system/src/decl.rs Outdated Show resolved Hide resolved

cjgillot force-pushed the tls branch from 84a156c to cc7b1c9 Compare June 8, 2021 18:02

cjgillot force-pushed the tls branch from cc7b1c9 to 18e281b Compare July 1, 2021 17:59

bjorn3 reviewed Jul 1, 2021

View reviewed changes

cjgillot force-pushed the tls branch 3 times, most recently from a5bfc56 to a91b82e Compare July 6, 2021 18:51

cjgillot force-pushed the tls branch from a91b82e to 79732e0 Compare July 15, 2021 18:32

JohnCSimon added S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Jul 31, 2021

cjgillot force-pushed the tls branch from 79732e0 to 5a1d2ae Compare August 1, 2021 08:03

cjgillot force-pushed the tls branch from 5a1d2ae to c1fe2b5 Compare August 20, 2021 21:55

rust-lang deleted a comment from bors Aug 21, 2021

rustbot removed S-waiting-on-perf Status: Waiting on a perf run to be completed. perf-regression Performance regression. labels Feb 25, 2022

Mark-Simulacrum removed the perf-regression-triaged The performance regression has been triaged. label Feb 25, 2022

Mark-Simulacrum reviewed Feb 25, 2022

View reviewed changes

Mark-Simulacrum added S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Feb 26, 2022

cjgillot added 4 commits April 15, 2022 18:25

Remove with_related_context.

4b3745f

There is only one GlobalCtxt ever.

Move query TLS to rustc_query_system.

353dbdf

Define DepKind in rustc_query_system.

c989f89

Add inline.

4387490

Add inline.

8b8c12e

Zoxc reviewed Apr 17, 2022

View reviewed changes

cjgillot mentioned this pull request Jan 14, 2024

Move TLS to rustc_query_system #119966

Closed

Uh oh!

Conversation

cjgillot commented Jun 5, 2021

Uh oh!

cjgillot commented Jun 5, 2021

Uh oh!

rust-timer commented Jun 5, 2021

Uh oh!

bors commented Jun 5, 2021

Uh oh!

This comment has been minimized.

bors commented Jun 5, 2021

Uh oh!

rust-timer commented Jun 5, 2021

Uh oh!

rust-timer commented Jun 6, 2021

Uh oh!

mati865 commented Jun 6, 2021

Uh oh!

Uh oh!

Uh oh!

bjorn3 Jul 1, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

JohnCSimon commented Jul 31, 2021

Uh oh!

camelid commented Aug 20, 2021

Uh oh!

bors commented Feb 24, 2022

Uh oh!

rust-timer commented Feb 24, 2022

Uh oh!

rust-timer commented Feb 25, 2022

Uh oh!

Mark-Simulacrum left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cjgillot commented Apr 16, 2022

Uh oh!

rust-timer commented Apr 16, 2022

Uh oh!

bors commented Apr 16, 2022

Uh oh!

bors commented Apr 16, 2022

Uh oh!

rust-timer commented Apr 16, 2022

Uh oh!

rust-timer commented Apr 16, 2022

Footnotes

Uh oh!

cjgillot commented Apr 16, 2022

Uh oh!

rust-timer commented Apr 16, 2022

Uh oh!

bors commented Apr 16, 2022

Uh oh!

bors commented Apr 16, 2022

Uh oh!

rust-timer commented Apr 16, 2022

Uh oh!

rust-timer commented Apr 17, 2022

Footnotes

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bjorn3 Jul 1, 2021 •

edited

Loading