[WIP] Implement a "place unification" MIR optimization (aka source/destination propagation aka NRVO). #47954

eddyb · 2018-02-01T23:55:19Z

nothing to see here (PR open for testing purposes)

note to self: DO NOT MERGE without inspecting all FIXME / HACK comments

rust-highfive · 2018-02-01T23:55:24Z

r? @petrochenkov

(rust_highfive has picked a reviewer for you, use r? to override)

eddyb · 2018-02-01T23:55:34Z

r? @nikomatsakis

@bors try

bors · 2018-02-01T23:55:44Z

⌛ Trying commit f88212c with merge 7a4d4c6c9a94179431514d0494125a3339f25718...

bors · 2018-02-02T01:28:19Z

💔 Test failed - status-travis

Mark-Simulacrum · 2018-02-02T01:53:33Z

[01:17:24] �[1m�[31merror: internal compiler error�(B�[m�[1m: librustc_trans/mir/block.rs:838: can't directly store to unaligned value�(B�[m
[01:17:24]   �(B�[m�[1m�[34m--> �(B�[m/cargo/registry/src/github.com-1ecc6299db9ec823/failure-0.1.1/src/error.rs:31:5�(B�[m
[01:17:24]    �(B�[m�[1m�[34m|�(B�[m
[01:17:24] �[1m�[34m31�(B�[m �(B�[m�[1m�[34m| �(B�[m�[1m�[31m/�(B�[m �(B�[m    fn from(failure: F) -> Error {�(B�[m
[01:17:24] �[1m�[34m32�(B�[m �(B�[m�[1m�[34m| �(B�[m�[1m�[31m|�(B�[m �(B�[m        let inner: Inner<F> = {�(B�[m
[01:17:24] �[1m�[34m33�(B�[m �(B�[m�[1m�[34m| �(B�[m�[1m�[31m|�(B�[m �(B�[m            let backtrace = if failure.backtrace().is_none() {�(B�[m
[01:17:24] �[1m�[34m34�(B�[m �(B�[m�[1m�[34m| �(B�[m�[1m�[31m|�(B�[m �(B�[m                Backtrace::new()�(B�[m
[01:17:24] �[1m�[34m...�(B�[m  �(B�[m�[1m�[31m|�(B�[m
[01:17:24] �[1m�[34m38�(B�[m �(B�[m�[1m�[34m| �(B�[m�[1m�[31m|�(B�[m �(B�[m        Error { inner: Box::new(inner) }�(B�[m
[01:17:24] �[1m�[34m39�(B�[m �(B�[m�[1m�[34m| �(B�[m�[1m�[31m|�(B�[m �(B�[m    }�(B�[m
[01:17:24]    �(B�[m�[1m�[34m| �(B�[m�[1m�[31m|_____^�(B�[m
[01:17:24] 
[01:17:24] thread 'rustc' panicked at 'Box<Any>', librustc_errors/lib.rs:482:9
[01:17:24] note: Run with `RUST_BACKTRACE=1` for a backtrace.
[01:17:24] 
[01:17:24] note: the compiler unexpectedly panicked. this is a bug.
[01:17:24] 
[01:17:24] note: we would appreciate a bug report: https://github.com/rust-lang/rust/blob/master/CONTRIBUTING.md#bug-reports
[01:17:24] 
[01:17:24] note: rustc 1.25.0-nightly (7a4d4c6c9 2018-02-01) running on x86_64-unknown-linux-gnu
[01:17:24] 
[01:17:24] �[m�[m�[31m�[1merror:�[m Could not compile `cargo`.
[01:17:24]

eddyb · 2018-02-02T07:22:15Z

@bors try

bors · 2018-02-02T07:22:25Z

⌛ Trying commit 401ced3 with merge 51345ce...

[WIP] Implement a "place unification" MIR optimization (aka source/destination propagation). *nothing to see here* (PR open for testing purposes) note to self: **DO NOT MERGE** without inspecting all `FIXME` / `HACK` comments

bors · 2018-02-02T08:53:26Z

💔 Test failed - status-travis

eddyb · 2018-02-02T08:59:24Z

Looks like banning #[repr(packed)] structs is trickier than I thought.

bors · 2018-02-03T14:40:56Z

☔ The latest upstream changes (presumably #47962) made this pull request unmergeable. Please resolve the merge conflicts.

shepmaster · 2018-02-09T22:16:57Z

nothing to see here

👁 We See All 👁

Like the fact that triage notices this has been open for a week, has some CI failures, and some merge conflicts. Any hopes of fixing those issues in the near future?

eddyb · 2018-02-09T22:18:14Z

@shepmaster I can close while I'm not using this (I just need to remember to reopen before pushing).

eddyb · 2018-02-13T14:49:01Z

@bors try

bors · 2018-02-13T14:49:11Z

⌛ Trying commit 44afedf with merge e8aeb68...

[WIP] Implement a "place unification" MIR optimization (aka source/destination propagation). *nothing to see here* (PR open for testing purposes) note to self: **DO NOT MERGE** without inspecting all `FIXME` / `HACK` comments

bors · 2018-02-13T16:34:50Z

💔 Test failed - status-travis

bors · 2018-02-17T11:36:46Z

☔ The latest upstream changes (presumably #47926) made this pull request unmergeable. Please resolve the merge conflicts.

bors · 2018-03-01T06:21:00Z

☀️ Test successful - status-travis
State: approved= try=True

Mark-Simulacrum · 2018-03-01T14:29:29Z

Perf started.

eddyb · 2018-03-02T09:04:46Z

Latest perf results look much better, worse regression is inflate still but now only a bit under 30x.

pietroalbini · 2018-03-12T10:17:23Z

Ping from triage! Do you still need this @eddyb?

nikomatsakis · 2018-03-12T16:46:44Z

I'm going to go ahead and close this -- @eddyb feel free to re-open. I'm assuming we don't want to merge? (Title is WIP..)

MSxDOS · 2020-01-05T14:22:21Z

@eddyb Have you made any progress with this?

eddyb · 2020-01-16T23:11:12Z

@MSxDOS I may have found a way to make this cheap enough to always run it.
I'll probably get back to it in the coming months.

mark-i-m · 2020-03-05T01:58:03Z

Any updates?

Dylan-DPC-zz · 2020-03-24T18:41:15Z

Closing this due to inactivity

@eddyb

Implement a generic Destination Propagation optimization on MIR This takes the work that was originally started by @eddyb in rust-lang#47954, and then explored by me in rust-lang#71003, and implements it in a general (ie. not limited to acyclic CFGs) and dataflow-driven way (so that no additional infrastructure in rustc is needed). The pass is configured to run at `mir-opt-level=2` and higher only. To enable it by default, some followup work on it is still needed: * Performance needs to be evaluated. I did some light optimization work and tested against `tuple-stress`, which caused trouble in my last attempt, but didn't go much in depth here. * We can also enable the pass only at `opt-level=2` and higher, if it is too slow to run in debug mode, but fine when optimizations run anyways. * Debuginfo needs to be fixed after locals are merged. I did not look into what is required for this. * Live ranges of locals (aka `StorageLive` and `StorageDead`) are currently deleted. We either need to decide that this is fine, or if not, merge the variable's live ranges (or remove these statements entirely – rust-lang#68622). Some benchmarks of the pass were done in rust-lang#72635.

Implement a generic Destination Propagation optimization on MIR This takes the work that was originally started by `@eddyb` in rust-lang#47954, and then explored by me in rust-lang#71003, and implements it in a general (ie. not limited to acyclic CFGs) and dataflow-driven way (so that no additional infrastructure in rustc is needed). The pass is configured to run at `mir-opt-level=2` and higher only. To enable it by default, some followup work on it is still needed: * Performance needs to be evaluated. I did some light optimization work and tested against `tuple-stress`, which caused trouble in my last attempt, but didn't go much in depth here. * We can also enable the pass only at `opt-level=2` and higher, if it is too slow to run in debug mode, but fine when optimizations run anyways. * Debuginfo needs to be fixed after locals are merged. I did not look into what is required for this. * Live ranges of locals (aka `StorageLive` and `StorageDead`) are currently deleted. We either need to decide that this is fine, or if not, merge the variable's live ranges (or remove these statements entirely – rust-lang#68622). Some benchmarks of the pass were done in rust-lang#72635.

rust-highfive assigned petrochenkov Feb 1, 2018

rust-highfive assigned nikomatsakis and unassigned petrochenkov Feb 1, 2018

shepmaster added the S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. label Feb 3, 2018

eddyb closed this Feb 9, 2018

eddyb reopened this Feb 13, 2018

eddyb force-pushed the copy-elision branch 2 times, most recently from aee57ce to 44afedf Compare February 13, 2018 13:20

eddyb force-pushed the copy-elision branch from 44afedf to 932b415 Compare February 14, 2018 14:54

jonas-schievink mentioned this pull request Feb 17, 2018

AVX512 code generated for i32 array sum is worse than code by clang 5 #48287

Closed

eddyb force-pushed the copy-elision branch from 932b415 to dcee10c Compare February 20, 2018 02:50

pietroalbini closed this Mar 20, 2018

This was referenced Apr 1, 2018

Consider aggressively avoiding generating memcpys for moves if we can avoid it #13707

Open

Optimization opportunity for constructor functions #49539

Closed

eddyb mentioned this pull request Nov 22, 2018

rustc_mir: implement an "lvalue reuse" optimization (aka destination propagation aka NRVO). #46321

Closed

eddyb mentioned this pull request May 22, 2019

Generator optimization: Overlap locals that never have storage live at the same time #60187

Merged

10 tasks

eddyb changed the title ~~[WIP] Implement a "place unification" MIR optimization (aka source/destination propagation).~~ [WIP] Implement a "place unification" MIR optimization (aka source/destination propagation aka NRVO). Jun 16, 2019

This was referenced Jun 16, 2019

Unnecessary memcpy when returning a struct #57077

Closed

Do move forwarding on MIR #32966

Open

Large structs constructed on stack rust-random/rand#817

Closed

bwesterb mentioned this pull request Jul 7, 2019

Add fixed_result_into to prevent copy RustCrypto/hashes#86

Closed

pnkfelix mentioned this pull request Oct 24, 2019

Attempt to make Box::default() construct in-place + Box::new_in_place() #64062

Closed

eddyb reopened this Jan 16, 2020

MSxDOS mentioned this pull request Jan 28, 2020

Stack overflow with Boxed array #53827

Open

Dylan-DPC-zz closed this Mar 24, 2020

slanterns mentioned this pull request Apr 10, 2020

[WIP] NRVO pass for acyclic CFGs #71003

Closed

ecstatic-morse mentioned this pull request May 16, 2020

Dumb NRVO #72205

Merged

jonas-schievink mentioned this pull request May 26, 2020

Implement a generic Destination Propagation optimization on MIR #72632

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[WIP] Implement a "place unification" MIR optimization (aka source/destination propagation aka NRVO). #47954

[WIP] Implement a "place unification" MIR optimization (aka source/destination propagation aka NRVO). #47954

eddyb commented Feb 1, 2018 •

edited

Loading

rust-highfive commented Feb 1, 2018

eddyb commented Feb 1, 2018

bors commented Feb 1, 2018

bors commented Feb 2, 2018

Mark-Simulacrum commented Feb 2, 2018

eddyb commented Feb 2, 2018

bors commented Feb 2, 2018

bors commented Feb 2, 2018

eddyb commented Feb 2, 2018

bors commented Feb 3, 2018

shepmaster commented Feb 9, 2018 •

edited

Loading

eddyb commented Feb 9, 2018

eddyb commented Feb 13, 2018

bors commented Feb 13, 2018

bors commented Feb 13, 2018

bors commented Feb 17, 2018

bors commented Mar 1, 2018

Mark-Simulacrum commented Mar 1, 2018

eddyb commented Mar 2, 2018

pietroalbini commented Mar 12, 2018

nikomatsakis commented Mar 12, 2018

MSxDOS commented Jan 5, 2020

eddyb commented Jan 16, 2020

mark-i-m commented Mar 5, 2020

Dylan-DPC-zz commented Mar 24, 2020

[WIP] Implement a "place unification" MIR optimization (aka source/destination propagation aka NRVO). #47954

[WIP] Implement a "place unification" MIR optimization (aka source/destination propagation aka NRVO). #47954

Conversation

eddyb commented Feb 1, 2018 • edited Loading

rust-highfive commented Feb 1, 2018

eddyb commented Feb 1, 2018

bors commented Feb 1, 2018

bors commented Feb 2, 2018

Mark-Simulacrum commented Feb 2, 2018

eddyb commented Feb 2, 2018

bors commented Feb 2, 2018

bors commented Feb 2, 2018

eddyb commented Feb 2, 2018

bors commented Feb 3, 2018

shepmaster commented Feb 9, 2018 • edited Loading

eddyb commented Feb 9, 2018

eddyb commented Feb 13, 2018

bors commented Feb 13, 2018

bors commented Feb 13, 2018

bors commented Feb 17, 2018

bors commented Mar 1, 2018

Mark-Simulacrum commented Mar 1, 2018

eddyb commented Mar 2, 2018

pietroalbini commented Mar 12, 2018

nikomatsakis commented Mar 12, 2018

MSxDOS commented Jan 5, 2020

eddyb commented Jan 16, 2020

mark-i-m commented Mar 5, 2020

Dylan-DPC-zz commented Mar 24, 2020

eddyb commented Feb 1, 2018 •

edited

Loading

shepmaster commented Feb 9, 2018 •

edited

Loading