Fortify and re-enable NRVO MIR opt

[cjgillot]

Fixes #111005
Fixes #110902

This PR re-enables NRVO opt that had been disabled in #111007

To look for RVO opportunities, we walk the MIR backwards from return terminators. In addition to the former implementation, we look at all the traversed statements and terminators for writes. If a local is written-to or moved-from, we discard it from the RVO candidates (assigned_locals bitset). If we see an indirect write or move, we discard all borrowed locals from candidates.

cc @JakobDegen

[rustbot]

r? @b-naber

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

[rustbot]

Some changes occurred to MIR optimizations

cc @rust-lang/wg-mir-opt

[cjgillot]

@bors try @rust-timer queue

[bors]

⌛ Trying commit 75d82bd with merge 549ef8c...

[bors]

☀️ Try build successful - checks-actions
Build commit: 549ef8c (549ef8c3a56f8dd29dad5eee971226d19b7088e9)

[bors]

☀️ Try build successful - checks-actions
Build commit: 549ef8c (549ef8c3a56f8dd29dad5eee971226d19b7088e9)

[rust-timer]

Finished benchmarking commit (549ef8c): comparison URL.

Overall result: ❌✅ regressions and improvements - 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)	5.2%	[0.6%, 9.2%]	6
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.3%	[-0.7%, -0.2%]	156
Improvements ✅ (secondary)	-0.4%	[-0.6%, -0.2%]	53
All ❌✅ (primary)	-0.1%	[-0.7%, 9.2%]	162

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)	1.9%	[0.0%, 3.1%]	4
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-7.0%	[-10.7%, -2.8%]	3
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-1.9%	[-10.7%, 3.1%]	7

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)	5.7%	[1.0%, 10.3%]	6
Regressions ❌ (secondary)	2.4%	[2.2%, 2.6%]	4
Improvements ✅ (primary)	-1.8%	[-1.8%, -1.8%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	4.6%	[-1.8%, 10.3%]	7

Binary size

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)	0.2%	[0.0%, 0.4%]	17
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.2%	[-1.4%, -0.0%]	39
Improvements ✅ (secondary)	-0.2%	[-0.2%, -0.2%]	1
All ❌✅ (primary)	-0.1%	[-1.4%, 0.4%]	56

Bootstrap: 624.124s -> 624.52s (0.06%)
Artifact size: 270.68 MiB -> 270.53 MiB (-0.06%)

[RalfJung]

Will this ever lead to MIR that takes a reference to the return place? E.g. this

fn foo(f: fn(&i16), k: i16) -> i16 {
    let mut x = k;
    x += 2;
    f(&x);
    x
}

could replace all x by RET. Will this pass do that?

EDIT: looks like it will

[tmiasko]

#![allow(internal_features, unused_assignments)]
#![feature(custom_mir, core_intrinsics)]
extern crate core;
use core::intrinsics::mir::*;

#[custom_mir(dialect = "runtime", phase = "initial")]
pub fn f() -> u32 {
    mir!({
        let x = 1;
        RET = 2;
        RET = x;
        Return()
    })
}
fn main() {
    assert_eq!(f(), 1);
}

$ rustc +stage1 a.rs -Zmir-opt-level=0 && ./a
$ rustc +stage1 a.rs -Zmir-opt-level=0 -Zmir-enable-passes=+RenameReturnPlace && ./a
thread 'main' panicked at a.rs:16:5:
assertion `left == right` failed
  left: 2
 right: 1
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

[cjgillot]

Indeed, that's even worse. This does not appear solvable without reimplementing DestProp here.

[tmiasko]

The approach seems plausible if limited to a MIR that contains a single occurrence of a return place - as a destination of an assignment.

[RalfJung]

Fortify and re-enable RNVO MIR opt

Should this say NRVO?