Implement SSA-based reference propagation

[cjgillot]

Rust has a tendency to create a lot of short-lived borrows, in particular for method calls. This PR aims to remove those short-lived borrows with a const-propagation dedicated to pointers to local places.

This pass aims to transform the following pattern:

 _1 = &raw? mut? PLACE;
_3 = *_1;
_4 = &raw? mut? *_1;

Into

 _1 = &raw? mut? PLACE;
_3 = PLACE;
_4 = &raw? mut? PLACE;

where PLACE is a direct or an indirect place expression.

By removing indirection, this pass should help both dest-prop and const-prop to handle more cases.
This optimization is distinct from const-prop and dataflow const-prop since the borrow-reborrow patterns needs to preserve borrowck invariants, especially the uniqueness property of mutable references.

The pointed-to places are computed using a SSA analysis. We suppose that removable borrows are typically temporaries from autoref, so they are by construction assigned only once, and a SSA analysis is enough to catch them. For each local, we store both where and how it is used, in order to efficiently compute the all-or-nothing property. Thanks to Derefer, we only have to track locals, not places in general.

---

There are 3 properties that need to be upheld for this transformation to be legal:

• place constness: PLACE must refer to the same memory wherever it appears;
• pointer liveness: we must not introduce dereferences of dangling pointers;
• &mut borrow uniqueness.

Constness

If PLACE is an indirect projection, if its of the form (*LOCAL).PROJECTIONS where:

• LOCAL is SSA;
• all projections in PROJECTIONS are constant (no dereference and no indexing).

If PLACE is a direct projection of a local, we consider it as constant if:

• the local is always live, or it has a single StorageLive that dominates all uses;
• all projections are constant.

Liveness

When performing a substitution, we must take care not to introduce uses of dangling locals.

Using a dangling borrow is UB. Therefore, we assume that for any use of *x, where x is a borrow, the pointed-to memory is live.

Limitations:

• occurrences of *x in an &raw mut? *x are accepted;
• raw pointers are allowed to be dangling.

In those 2 case, we do not substitute anything, to be on the safe side.

Open question: we do not differentiate borrows of ZST and non-ZST. The UB rules may be
different depending on the layout. Having a different treatment would effectively prevent this
pass from running on polymorphic MIR, which defeats the purpose of MIR opts.

Uniqueness

For &mut borrows, we also need to preserve the uniqueness property:
we must avoid creating a state where we interleave uses of *_1 and _2.
To do it, we only perform full substitution of mutable borrows:
we replace either all or none of the occurrences of *_1.

Some care has to be taken when _1 is copied in other locals.

 _1 = &raw? mut? _2;
_3 = *_1;
_4 = _1
_5 = *_4

In such cases, fully substituting _1 means fully substituting all of the copies.

For immutable borrows, we do not need to preserve such uniqueness property,
so we perform all the possible substitutions without removing the _1 = &_2 statement.

[rustbot]

r? @petrochenkov

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

[petrochenkov]

r? rust-lang/compiler

[cjgillot]

@bors try @rust-timer queue

[bors]

⌛ Trying commit 46516095d7a9cb208c8f17a01574a9a8bc180833 with merge fea7409a5e37c7f115c6b24d93e12e52fe364c27...

[bors]

☀️ Try build successful - checks-actions
Build commit: fea7409a5e37c7f115c6b24d93e12e52fe364c27 (fea7409a5e37c7f115c6b24d93e12e52fe364c27)

[rust-timer]

Finished benchmarking commit (fea7409a5e37c7f115c6b24d93e12e52fe364c27): 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)	0.7%	[0.2%, 1.8%]	25
Regressions ❌ (secondary)	0.5%	[0.2%, 0.8%]	17
Improvements ✅ (primary)	-0.8%	[-1.6%, -0.3%]	16
Improvements ✅ (secondary)	-1.0%	[-1.9%, -0.5%]	19
All ❌✅ (primary)	0.1%	[-1.6%, 1.8%]	41

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)	3.5%	[2.5%, 4.5%]	2
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-2.1%	[-3.8%, -0.1%]	5
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-0.5%	[-3.8%, 4.5%]	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)	0.9%	[0.8%, 1.1%]	4
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.9%	[0.8%, 1.1%]	4

[cjgillot]

For immutable borrows and AddressOf, we probably don't need to mark the local as non-SSA: that borrow cannot be used, even indirectly to modify the local. Recording a direct use should be enough.

[rustbot]

Some changes occurred to MIR optimizations

cc @rust-lang/wg-mir-opt

[bors]

⌛ Testing commit d30bbe1857fbdac2f4c03a974b22419b02b3c514 with merge bdd6f2fd486b691ba79fff426393a0a45417becb...

[bors]

💔 Test failed - checks-actions

[cjgillot]

@bors r=JakobDegen

[bors]

📌 Commit bde213c has been approved by JakobDegen

It is now in the queue for this repository.

[bors]

⌛ Testing commit bde213c with merge 50dff95...

[bors]

☀️ Test successful - checks-actions
Approved by: JakobDegen
Pushing 50dff95 to master...

[rust-timer]

Finished benchmarking commit (50dff95): comparison URL.

Overall result: ❌✅ regressions and improvements - ACTION NEEDED

Next Steps: If you can justify the regressions found in this perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please open an issue or create a new PR that fixes the regressions, add a comment linking to the newly created issue or PR, and then add the perf-regression-triaged label to this PR.

@rustbot label: +perf-regression
cc @rust-lang/wg-compiler-performance

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)	0.8%	[0.2%, 1.3%]	2
Regressions ❌ (secondary)	0.4%	[0.4%, 0.4%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.6%	[-0.9%, -0.3%]	7
All ❌✅ (primary)	0.8%	[0.2%, 1.3%]	2

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)	3.9%	[3.9%, 3.9%]	1
Regressions ❌ (secondary)	0.8%	[0.8%, 0.8%]	1
Improvements ✅ (primary)	-3.1%	[-3.9%, -2.2%]	4
Improvements ✅ (secondary)	-3.2%	[-3.2%, -3.2%]	1
All ❌✅ (primary)	-1.7%	[-3.9%, 3.9%]	5

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)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-1.3%	[-1.3%, -1.3%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-1.3%	[-1.3%, -1.3%]	1

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%, 1.3%]	45
Regressions ❌ (secondary)	0.9%	[0.7%, 1.1%]	6
Improvements ✅ (primary)	-0.1%	[-0.1%, -0.1%]	3
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.2%	[-0.1%, 1.3%]	48

Bootstrap: 657.982s -> 659.227s (0.19%)

[Mark-Simulacrum]

Regressions appear to mostly be bimodality.

[RalfJung]

There is no _2 is the examples above so I don't understand what this comment refers to.

[RalfJung]

This seems to be incomplete -- the transformation also kicks in when _3 or _4 apply extra projections to *_1. This is causing some trouble with Stacked Borrows, though arguably that's a SB bug.