Remove fewer Storage calls in `copy_prop` #142531

ohadravid · 2025-06-15T07:51:38Z

Modify the copy_prop MIR optimization pass to remove fewer Storage{Live,Dead} calls, allowing for better optimizations by LLVM - see #141649.

Details

This is my attempt to fix the mentioned issue (this is the first part, I also implemented a similar solution for GVN in this branch).

The idea is to use ~~the MaybeStorageDead~~ a new MaybeUninitializedLocals analysis and remove only the storage calls of heads that are maybe-uninit when the associated local is accessed (or, conversely, keep the storage of heads that are for-sure initialized in every relevant access).

When combined with the GVN change, the final example in the issue (#141649 (comment)) is optimized as expected by LLVM. I also measured the effect on a few functions in rav1d (where I originally saw the issue) and observed reduced stack usage in several of them.

This is my first attempt at working with MIR optimizations, so it's possible this isn't the right approach — but all tests pass, and the resulting diffs appear correct.

r? tmiasko

since he commented on the issue and pointed to these passes.

rustbot · 2025-06-15T07:51:44Z

Some changes occurred to MIR optimizations

cc @rust-lang/wg-mir-opt

matthiaskrgr · 2025-06-15T09:37:15Z

@bors try @rust-timer queue

…try> Remove fewer Storage calls in `copy_prop` Modify the `copy_prop` MIR optimization pass to remove fewer `Storage{Live,Dead}` calls, allowing for better optimizations by LLVM - see #141649. ### Details This is my attempt to fix the mentioned issue (this is the first part, I also implemented a similar solution for GVN in [this branch](https://github.com/rust-lang/rust/compare/master...ohadravid:rust:better-storage-calls-gvn-v2?expand=1)). The idea is to use the `MaybeStorageDead` analysis and remove only the storage calls of `head`s that are maybe-storage-dead when the associated `local` is accessed (or, conversely, keep the storage of `head`s that are for-sure alive in _every_ relevant access). When combined with the GVN change, the final example in the issue (#141649 (comment)) is optimized as expected by LLVM. I also measured the effect on a few functions in `rav1d` (where I originally saw the issue) and observed reduced stack usage in several of them. This is my first attempt at working with MIR optimizations, so it's possible this isn't the right approach — but all tests pass, and the resulting diffs appear correct. r? tmiasko since he commented on the issue and pointed to these passes.

bors · 2025-06-15T09:38:28Z

⌛ Trying commit d24d035 with merge ef7d206...

bors · 2025-06-15T12:05:29Z

☀️ Try build successful - checks-actions
Build commit: ef7d206 (ef7d20666974f0dac45b03e051f2e283f9d9f090)

rust-timer · 2025-06-15T13:31:55Z

Finished benchmarking commit (ef7d206): comparison URL.

Overall result: ❌ regressions - please read the text below

Benchmarking this pull request means it may be perf-sensitive – we'll automatically label it not fit for rolling up. You can override this, but we strongly advise not to, due to possible changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please do so in sufficient writing along with @rustbot label: +perf-regression-triaged. If not, please fix the regressions and do another perf run. If its results are neutral or positive, the label will be automatically removed.

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

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

	mean	range	count
Regressions ❌ (primary)	0.3%	[0.2%, 0.4%]	8
Regressions ❌ (secondary)	0.3%	[0.2%, 0.4%]	7
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.3%	[0.2%, 0.4%]	8

Max RSS (memory usage)

Results (primary 0.7%, secondary 3.4%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	3.5%	[1.8%, 5.0%]	5
Regressions ❌ (secondary)	3.4%	[3.4%, 3.4%]	1
Improvements ✅ (primary)	-3.9%	[-6.5%, -2.0%]	3
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	0.7%	[-6.5%, 5.0%]	8

Cycles

Results (primary -0.6%, secondary -0.1%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	3.8%	[3.8%, 3.8%]	1
Improvements ✅ (primary)	-0.6%	[-0.6%, -0.6%]	1
Improvements ✅ (secondary)	-4.1%	[-4.1%, -4.1%]	1
All ❌✅ (primary)	-0.6%	[-0.6%, -0.6%]	1

Binary size

Results (primary 0.0%, secondary 0.0%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	0.2%	[0.0%, 0.8%]	10
Regressions ❌ (secondary)	0.1%	[0.0%, 0.1%]	5
Improvements ✅ (primary)	-0.2%	[-0.8%, -0.0%]	8
Improvements ✅ (secondary)	-0.2%	[-0.2%, -0.2%]	1
All ❌✅ (primary)	0.0%	[-0.8%, 0.8%]	18

Bootstrap: 757.399s -> 756.065s (-0.18%)
Artifact size: 372.20 MiB -> 372.12 MiB (-0.02%)

ohadravid · 2025-06-15T14:55:18Z

@matthiaskrgr - I updated the impl to stop re-checking once a head is found to be maybe-dead, which should be a bit better

matthiaskrgr · 2025-06-15T15:06:54Z

@bors try @rust-timer queue

bors · 2025-06-15T15:08:08Z

⌛ Trying commit 905e968 with merge c0a2949...

…try> Remove fewer Storage calls in `copy_prop` Modify the `copy_prop` MIR optimization pass to remove fewer `Storage{Live,Dead}` calls, allowing for better optimizations by LLVM - see #141649. ### Details This is my attempt to fix the mentioned issue (this is the first part, I also implemented a similar solution for GVN in [this branch](https://github.com/rust-lang/rust/compare/master...ohadravid:rust:better-storage-calls-gvn-v2?expand=1)). The idea is to use the `MaybeStorageDead` analysis and remove only the storage calls of `head`s that are maybe-storage-dead when the associated `local` is accessed (or, conversely, keep the storage of `head`s that are for-sure alive in _every_ relevant access). When combined with the GVN change, the final example in the issue (#141649 (comment)) is optimized as expected by LLVM. I also measured the effect on a few functions in `rav1d` (where I originally saw the issue) and observed reduced stack usage in several of them. This is my first attempt at working with MIR optimizations, so it's possible this isn't the right approach — but all tests pass, and the resulting diffs appear correct. r? tmiasko since he commented on the issue and pointed to these passes.

cjgillot · 2025-06-15T15:45:26Z

Should this check happen in Replacer::visit_local, and move the replacement of storage statements to a dedicated cleanup visitor?

bors · 2025-06-15T17:41:36Z

☀️ Try build successful - checks-actions
Build commit: c0a2949 (c0a294957df10fc3880e1677c72c0cf122485509)

ohadravid · 2025-06-15T18:12:43Z

Should this check happen in Replacer::visit_local

I'm not sure how to make this work: using ResultsCursor requires a &body, but it's not possible to have that while running a MutVisitor since it requires a &mut body.

Is there a different way to do this?

rust-timer · 2025-06-15T20:15:45Z

Finished benchmarking commit (c0a2949): comparison URL.

Overall result: ❌✅ regressions and improvements - please read the text below

Benchmarking this pull request means it may be perf-sensitive – we'll automatically label it not fit for rolling up. You can override this, but we strongly advise not to, due to possible changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please do so in sufficient writing along with @rustbot label: +perf-regression-triaged. If not, please fix the regressions and do another perf run. If its results are neutral or positive, the label will be automatically removed.

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

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

	mean	range	count
Regressions ❌ (primary)	0.3%	[0.2%, 0.4%]	9
Regressions ❌ (secondary)	0.3%	[0.2%, 0.4%]	7
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.2%	[-0.2%, -0.2%]	1
All ❌✅ (primary)	0.3%	[0.2%, 0.4%]	9

Max RSS (memory usage)

Results (primary -0.1%, secondary -1.3%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	4.2%	[3.4%, 5.8%]	4
Regressions ❌ (secondary)	3.1%	[3.1%, 3.1%]	1
Improvements ✅ (primary)	-4.4%	[-6.6%, -1.8%]	4
Improvements ✅ (secondary)	-5.8%	[-5.8%, -5.8%]	1
All ❌✅ (primary)	-0.1%	[-6.6%, 5.8%]	8

Cycles

Results (secondary -1.0%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	2.3%	[2.3%, 2.3%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-2.6%	[-2.6%, -2.5%]	2
All ❌✅ (primary)	-	-	0

Binary size

Results (primary -0.0%, secondary 0.0%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	0.2%	[0.0%, 0.8%]	10
Regressions ❌ (secondary)	0.1%	[0.0%, 0.1%]	5
Improvements ✅ (primary)	-0.2%	[-0.8%, -0.0%]	8
Improvements ✅ (secondary)	-0.2%	[-0.2%, -0.2%]	1
All ❌✅ (primary)	-0.0%	[-0.8%, 0.8%]	18

Bootstrap: 756.494s -> 757.685s (0.16%)
Artifact size: 372.15 MiB -> 372.11 MiB (-0.01%)

compiler/rustc_mir_transform/src/copy_prop.rs

tmiasko · 2025-06-18T13:33:27Z

compiler/rustc_mir_dataflow/src/impls/initialized.rs

+            // StorageDead makes a local uninitialized.
+            mir::StatementKind::StorageDead(local) => {
+                state.insert(local);
+            }


This should also include StorageLive. A variant of the earlier example:

#![feature(custom_mir, core_intrinsics)] use std::intrinsics::mir::*; #[custom_mir(dialect = "runtime")] pub fn live_twice<T: Copy>(_1: T) -> T { mir! { let _2: T; let _3: T; { StorageLive(_2); Call(_2 = opaque(Move(_1)), ReturnTo(bb1), UnwindUnreachable()) } bb1 = { let _3 = Move(_2); StorageLive(_2); Call(RET = opaque(_3), ReturnTo(bb2), UnwindUnreachable()) } bb2 = { StorageDead(_2); Return() } } } #[inline(never)] fn opaque<T>(a: T) -> T { a }

I'll add the StorageLive check anyway, but this won't compile since

broken MIR in Item(...) at bb1[0]: StorageLive(_2) which already has storage here

This MIR is valid, although we wouldn't like to generate such a code in the first place. mir-opt tests enable an additional lint that detects suspicious MIR code patterns.. You can disable it with -Zlint-mir=false.

tmiasko · 2025-06-18T13:37:19Z

compiler/rustc_mir_transform/src/copy_prop.rs

+                // We need to determine if we can keep the head's storage statements (which enables better optimizations).
+                // For every local's usage location, if the head is maybe-uninitialized, we'll need to remove it's storage statements.
+                head_storage_to_check.insert(head);


Add a comment that this approach requires local not to be borrowed, since otherwise we cannot easily identify when it is used.

Please rebase past #142571, when it lands, and double check that we actually enforce this. I suspect this pull request also fixes failure from x86_64-gnu-tools cc @cjgillot.

tmiasko · 2025-06-18T13:43:50Z

compiler/rustc_mir_transform/src/copy_prop.rs

+            // Debug builds have no use for the storage statements, so avoid extra work.
+            let storage_to_remove = if tcx.sess.emit_lifetime_markers() {
+                let maybe_uninit = MaybeUninitializedLocals::new()
+                    .iterate_to_fixpoint(tcx, body, Some("mir_opt::copy_prop"))
+                    .into_results_cursor(body);
+
+                let mut storage_checker =
+                    StorageChecker { maybe_uninit, head_storage_to_check, storage_to_remove };
+
+                storage_checker.visit_body(body);
+
+                storage_checker.storage_to_remove
+            } else {
+                // Conservatively remove all storage statements for the head locals.
+                head_storage_to_check
+            };
+            StorageRemover { tcx, storage_to_remove }.visit_body_preserves_cfg(body);


Would it be possible to organize it so thatStorageChecker runs before Replacer and all changes are performed all at once inside Replacer?

I think it is quite confusing when changes are applied in a piecewise manner, first in Replaced and then StorageRemover, since it is not immediately clear what is and what isn't replaced already.

tmiasko · 2025-06-18T13:49:23Z

tests/mir-opt/copy-prop/copy_prop_storage_dead_twice.rs

@@ -0,0 +1,38 @@
+// skip-filecheck
+// EMIT_MIR_FOR_EACH_PANIC_STRATEGY


EMIT_MIR_FOR_EACH_PANIC_STRATEGY is used when output differs between panic strategies. I don't think we need it in this test. Please remove it.

… to remove fewer storage statements

rustbot assigned tmiasko Jun 15, 2025

rustbot added S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels Jun 15, 2025