New mir-opt pass to simplify gotos with const values (reopening #77486) by simonvandel · Pull Request #80475 · rust-lang/rust

simonvandel · 2020-12-29T15:35:38Z

Reopening PR #77486

This pass optimizes the following sequence

bb2:{
_2 = const true;
goto -> bb3;}bb3:{
switchInt(_2) -> [false: bb4, otherwise: bb5];}

into

bb2:{
_2 = const true;
goto -> bb5;}

rust-highfive · 2020-12-29T15:35:42Z

r? @ecstatic-morse

(rust-highfive has picked a reviewer for you, use r? to override)

simonvandel · 2020-12-29T15:42:42Z

CC @oli-obk from #77486 (comment)
The MIR still contains two branches in the final MIR, so the optimization in this PR still messes this up. I'm wondering if this can be considered a bug in MatchBranchSimplification that it does not handle the _0 = const () case.
Tangentially, is _0 = const () ever needed in a function that returns unit? I think it would only make sense if _0 is considered uninitialized at the beginning of the function. But then, it would probably be a simplification to just set _0 = const () at the beginning of bb0.

_0 = const () is never needed, and I thought it's actually eliminated by InstCombine or sth. Not sure why there's still one around at this point. No ZST assignment is ever needed, you can create references to locals before they are assigned if the local is a ZST (as long as it was StorageLived I think).

oli-obk · 2021-01-01T17:07:17Z

cc @tmiasko if you have the time (and interest), could you have a look at this PR?

ecstatic-morse · 2021-01-01T19:25:41Z

I'll set aside some time in the next week or so to review this. In the meantime, if you could squash your commits to remove refactorings/fixes to newly added code, it would make things a bit easier for me.

tmiasko · 2021-01-01T20:48:09Z

Compared to the earlier version the transform now supports switch blocks with statements. As a result it has to establish that the definition place = const reaches the switch, whereas previously it was trivially true.
Based on past experiences, I feel that doing this for an arbitrary place without a dedicated analysis will be challenging. If place is indirect it could be invalidated by other assignments, it can also contain indexing projections referencing other locals that could be modified, etc.
I would probably start with locals without any projections, and in addition to the existing check, also verify that either they never have their address taken or there are no indirect assignment in the switch block.
An example that doesn't work correctly right now:
pubfnmain(){letmut a = &mut0;if*a == 0{*a = 0;()};let b = &mut a;**b = 1;match*a {1 => {}, _ => panic!(),}}

Yeah, you are right. I'll revert the newest changes so that we only support statement-less targets. Then follow ups can improve the situation

bors · 2021-01-14T17:53:06Z

☔ The latest upstream changes (presumably #79328) made this pull request unmergeable. Please resolve the merge conflicts.

simonvandel · 2021-02-12T21:07:56Z

Fixed merge conflicts.

Hi @tmiasko, sorry for the long wait. Would you be able to do a review?

oli-obk · 2021-02-13T12:26:28Z

@bors try @rust-timer queue

rust-timer · 2021-02-13T12:26:29Z

Awaiting bors try build completion.

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

bors · 2021-02-13T12:26:37Z

⌛ Trying commit 5734754fa175f74a0081ea2823b387013c54ee10 with merge 57512cbea2265af4686716b44fd0bb86358fbcdb...

simonvandel · 2021-02-13T12:29:09Z

@oli-obk the pass currently bails if mir opt level is less than 3 so I would expect no observable change in a perf run.If you want, I can remove the bail out so a perf run can be made.

oli-obk · 2021-02-13T13:03:19Z

oh... oops, didn't re-check, just read your comment and started the perf run. Do you still want that perf run? If so, yea let's change it back

oli-obk · 2021-02-13T13:03:29Z

@bors try-

simonvandel · 2021-02-13T13:14:47Z

@oli-obk Added a commit to disable the mir-opt-level check. Can you start the run again?

oli-obk · 2021-02-13T14:12:41Z

@bors try @rust-timer queue

rust-timer · 2021-02-13T14:12:42Z

Awaiting bors try build completion.

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

Fixesrust-lang#77355

oli-obk · 2021-02-23T10:40:50Z

Yea, oops, should've rechecked.

@bors r+

bors · 2021-02-23T10:40:51Z

📌 Commit a6dccfe has been approved by oli-obk

bors · 2021-02-23T12:43:23Z

⌛ Testing commit a6dccfe with merge 8aa2df00d8834c83aacfb232a83b3c160eeb4dae...

rust-log-analyzer · 2021-02-23T14:38:39Z

A job failed! Check out the build log: (web)(plain)

Click to see the possible cause of the failure (guessed by this bot)

bors · 2021-02-23T14:38:40Z

💔 Test failed - checks-actions

bjorn3 · 2021-02-23T15:16:38Z

@bors retry
Seems a spurious issue.

bors · 2021-02-23T20:10:04Z

⌛ Testing commit a6dccfe with merge c1357d951bac5da7640b3cc5adac992277a89c22...

bors · 2021-02-24T00:25:04Z

💥 Test timed out

rust-log-analyzer · 2021-02-24T00:33:05Z

A job failed! Check out the build log: (web)(plain)

Click to see the possible cause of the failure (guessed by this bot)

bjorn3 · 2021-02-24T07:12:56Z

@bors retry

Only a single runner timed out. The rest passed.

bors · 2021-02-24T07:24:00Z

⌛ Testing commit a6dccfe with merge 6b56603...

bors · 2021-02-24T10:04:55Z

☀️ Test successful - checks-actions
Approved by: oli-obk
Pushing 6b56603 to master...

rust-highfive assigned ecstatic-morse Dec 29, 2020

rust-highfive added the S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. label Dec 29, 2020

simonvandel commented Dec 29, 2020
View reviewed changes

simonvandel force-pushed the fix-77355 branch from 05b4e79 to 6a73f1cCompare December 29, 2020 19:30

simonvandel mentioned this pull request Dec 29, 2020
Remove MIR assignments to ZST types #80493
Closed

camelid added the A-mir-opt Area: MIR optimizations label Dec 30, 2020

oli-obk reviewed Jan 1, 2021
View reviewed changes

Comment threadcompiler/rustc_mir/src/transform/const_goto.rs Outdated

tmiasko reviewed Jan 1, 2021
View reviewed changes

simonvandel force-pushed the fix-77355 branch from 6a73f1c to c6293b5Compare January 3, 2021 15:54

simonvandel mentioned this pull request Jan 4, 2021
Rust insufficiently optimizes loop { match { } } state machines #80630
Open

JohnCSimon added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Feb 1, 2021

simonvandel force-pushed the fix-77355 branch from c6293b5 to 5734754Compare February 12, 2021 20:24

bjorn3 reviewed Feb 12, 2021
View reviewed changes

Comment threadcompiler/rustc_mir/src/transform/const_goto.rs Outdated

rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Feb 13, 2021

New mir-opt pass to simplify gotos with const values
a6dccfe
Fixesrust-lang#77355

simonvandel force-pushed the fix-77355 branch from 0a560ff to a6dccfeCompare February 22, 2021 20:04

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. labels Feb 23, 2021

bors added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. labels Feb 23, 2021

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Feb 23, 2021

bors added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. and removed S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. labels Feb 24, 2021

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Feb 24, 2021

bors added the merged-by-bors This PR was explicitly merged by bors. label Feb 24, 2021

bors merged commit 6b56603 into rust-lang:masterFeb 24, 2021

rustbot added this to the 1.52.0 milestone Feb 24, 2021

bors mentioned this pull request Feb 24, 2021
Remove storage markers if they won't be used during code generation #78360
Merged

Uh oh!

Uh oh!

Conversation

simonvandel commented Dec 29, 2020

Uh oh!

rust-highfive commented Dec 29, 2020

Uh oh!

simonvandelDec 29, 2020

Choose a reason for hiding this comment

Uh oh!

oli-obkDec 29, 2020

Choose a reason for hiding this comment

Uh oh!

oli-obk commented Jan 1, 2021

Uh oh!

Uh oh!

ecstatic-morse commented Jan 1, 2021

Uh oh!

tmiaskoJan 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!

simonvandelJan 3, 2021

Choose a reason for hiding this comment

Uh oh!

bors commented Jan 14, 2021

Uh oh!

simonvandel commented Feb 12, 2021

Uh oh!

Uh oh!

oli-obk commented Feb 13, 2021

Uh oh!

rust-timer commented Feb 13, 2021

Uh oh!

bors commented Feb 13, 2021

Uh oh!

simonvandel commented Feb 13, 2021

Uh oh!

oli-obk commented Feb 13, 2021

Uh oh!

oli-obk commented Feb 13, 2021

Uh oh!

simonvandel commented Feb 13, 2021

Uh oh!

oli-obk commented Feb 13, 2021

Uh oh!

rust-timer commented Feb 13, 2021

Uh oh!

oli-obk commented Feb 23, 2021

Uh oh!

bors commented Feb 23, 2021

Uh oh!

bors commented Feb 23, 2021

Uh oh!

rust-log-analyzer commented Feb 23, 2021

Uh oh!

bors commented Feb 23, 2021

Uh oh!

bjorn3 commented Feb 23, 2021

Uh oh!

bors commented Feb 23, 2021

Uh oh!

bors commented Feb 24, 2021

Uh oh!

rust-log-analyzer commented Feb 24, 2021

Uh oh!

bjorn3 commented Feb 24, 2021

Uh oh!

bors commented Feb 24, 2021

Uh oh!

bors commented Feb 24, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

12 participants

tmiaskoJan 1, 2021•
edited
Loading