In `copy_nonoverlapping`, use `mul nuw nsw` to compute the byte size by scottmcm · Pull Request #157560 · rust-lang/rust

scottmcm · 2026-06-07T05:21:26Z

Seems like we might as well? Adding these flags means the optimizer can tell the limited range on the count of items -- like how we use these flags (#136575) when calculating size_of_val for a slice.

Today we use a wrapping multiplication, which mean that copy_nonoverlapping::<u32>(src, dst, 0x40000000_00000001) appears like 4 bytes -- a perfectly reasonable size! -- once it gets to the memcpy call.

If I'm understanding https://doc.rust-lang.org/std/ptr/fn.copy_nonoverlapping.html#safety properly, this is just exploiting existing UB, since src and dst must each be inside an allocation, and those allocations can be at most isize::MAX bytes. (Plus, fundamentally, to be non-overlapping there's not enough space in the address space to be bigger than isize::MAX.)

cc @RalfJung to make sure this is ok, as requested last he found out I was newly exploiting some UB in codegen 🙃
r? codegen

RalfJung · 2026-06-07T09:29:55Z

Yeah, in fact both copy and copy_nonoverlapping could use this since we know the result must be at most isize::MAX else it cannot be inbounds.

saethlin · 2026-06-07T13:37:14Z

@bors try @rust-timer queue

In `copy_nonoverlapping`, use `mul nuw nsw` to compute the byte size

rust-bors · 2026-06-07T15:49:27Z

☀️ Try build successful (CI)
Build commit: 1fe6841 (1fe6841f2af59a318002c0c0f68aefac7e31098d, parent: 43a4909ee98ed4d006d9d773f5d94dc58e34f846)

rust-timer · 2026-06-07T16:30:13Z

Finished benchmarking commit (1fe6841): comparison URL.

Overall result: no relevant changes - no action needed

Benchmarking means the PR may be perf-sensitive. Consider adding rollup=never if this change is not fit for rolling up.

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

Instruction count

This perf run didn't have relevant results for this metric.

Max RSS (memory usage)

Results (primary 3.3%, secondary -5.6%)

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

	mean	range	count
Regressions ❌ (primary)	3.3%	[3.3%, 3.3%]	1
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-5.6%	[-5.6%, -5.6%]	1
All ❌✅ (primary)	3.3%	[3.3%, 3.3%]	1

Cycles

Results (secondary -3.5%)

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)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-3.5%	[-3.5%, -3.5%]	1
All ❌✅ (primary)	-	-	0

Binary size

This perf run didn't have relevant results for this metric.

Bootstrap: 516.093s -> 517.507s (0.27%)
Artifact size: 400.83 MiB -> 400.81 MiB (-0.00%)

saethlin · 2026-06-07T17:00:50Z

@bors r+ rollup (because clearly it doesn't impact perf... Right now)

rust-bors · 2026-06-07T17:00:53Z

📌 Commit 4af2ca0 has been approved by saethlin

It is now in the queue for this repository.

@RalfJung

…saethlin In `copy_nonoverlapping`, use `mul nuw nsw` to compute the byte size Seems like we might as well? Adding these flags means the optimizer can tell the limited range on the count of items -- like how we use these flags (rust-lang#136575) when calculating `size_of_val` for a slice. Today we use a wrapping multiplication, which mean that `copy_nonoverlapping::<u32>(src, dst, 0x40000000_00000001)` appears like 4 bytes -- a perfectly reasonable size! -- once it gets to the `memcpy` call. If I'm understanding <https://doc.rust-lang.org/std/ptr/fn.copy_nonoverlapping.html#safety> properly, this is just exploiting existing UB, since `src` and `dst` must each be inside an allocation, and those allocations can be at most `isize::MAX` bytes. (Plus, fundamentally, to be non-overlapping there's not enough space in the address space to be bigger than `isize::MAX`.) cc @RalfJung to make sure this is ok, as requested last he found out I was newly exploiting some UB in codegen 🙃 r? codegen

@RalfJung

…saethlin In `copy_nonoverlapping`, use `mul nuw nsw` to compute the byte size Seems like we might as well? Adding these flags means the optimizer can tell the limited range on the count of items -- like how we use these flags (rust-lang#136575) when calculating `size_of_val` for a slice. Today we use a wrapping multiplication, which mean that `copy_nonoverlapping::<u32>(src, dst, 0x40000000_00000001)` appears like 4 bytes -- a perfectly reasonable size! -- once it gets to the `memcpy` call. If I'm understanding <https://doc.rust-lang.org/std/ptr/fn.copy_nonoverlapping.html#safety> properly, this is just exploiting existing UB, since `src` and `dst` must each be inside an allocation, and those allocations can be at most `isize::MAX` bytes. (Plus, fundamentally, to be non-overlapping there's not enough space in the address space to be bigger than `isize::MAX`.) cc @RalfJung to make sure this is ok, as requested last he found out I was newly exploiting some UB in codegen 🙃 r? codegen

Rollup of 25 pull requests Successful merges: - #157447 (Move cross crate tests into the appropriate folder) - #145108 (Resolver: Batched Import Resolution) - #156119 (Further optimize `SliceIndex<str>` impl for `Range<usize>`) - #157224 (Manually unroll loop in `str::floor_char_boundary`) - #157289 (Add infallible primitive type lookups to template arg resolver) - #157540 (Cleanup and optimize `render_impls`) - #157444 (Couple of work product cleanups) - #157543 (Reorganize `tests/ui/issues` [5/N]) - #153513 (Syntactically reject equality predicates) - #155797 (LineWriter: cap write_vectored newline scan to avoid quadratic write_all_vectored) - #156155 (macros: report unbound metavariables directly) - #156188 (riscv: promote d, e, and f target_features to CfgStableToggleUnstable) - #156666 (Clarify meaning of ranges in pointer offset docs) - #157078 (Document equivalence of `highest_one` and `ilog2` methods on integers) - #157129 (ci: update download-artifact action to v8) - #157169 (triagebot: Update messages to direct changes to appropriate repositories) - #157323 (Document Repeat::last panic behavior) - #157370 (Clarify MaybeUninit::zeroed padding docs) - #157399 (Silence llbc's output by default to prevent rustc's linker output warning) - #157500 (Improve documentation of `align_of` and `Alignment`.) - #157545 (Suggest using comma to separate valid attribute list items) - #157559 (chore: Update annotate-snippets to 0.12.16) - #157560 (In `copy_nonoverlapping`, use `mul nuw nsw` to compute the byte size) - #157580 (Importing suggestion reported twice when reporting privacy error) - #157581 (Test fixup)

scottmcm · 2026-06-08T00:14:16Z

because clearly it doesn't impact perf... Right now

Yeah, it enables fewer optimizations than I'd hoped right now; see llvm/llvm-project#202240 and #157589

@RalfJung

Rollup merge of #157560 - scottmcm:mul_nuw_nsw_in_memcpy, r=saethlin In `copy_nonoverlapping`, use `mul nuw nsw` to compute the byte size Seems like we might as well? Adding these flags means the optimizer can tell the limited range on the count of items -- like how we use these flags (#136575) when calculating `size_of_val` for a slice. Today we use a wrapping multiplication, which mean that `copy_nonoverlapping::<u32>(src, dst, 0x40000000_00000001)` appears like 4 bytes -- a perfectly reasonable size! -- once it gets to the `memcpy` call. If I'm understanding <https://doc.rust-lang.org/std/ptr/fn.copy_nonoverlapping.html#safety> properly, this is just exploiting existing UB, since `src` and `dst` must each be inside an allocation, and those allocations can be at most `isize::MAX` bytes. (Plus, fundamentally, to be non-overlapping there's not enough space in the address space to be bigger than `isize::MAX`.) cc @RalfJung to make sure this is ok, as requested last he found out I was newly exploiting some UB in codegen 🙃 r? codegen

Rollup of 25 pull requests Successful merges: - rust-lang/rust#157447 (Move cross crate tests into the appropriate folder) - rust-lang/rust#145108 (Resolver: Batched Import Resolution) - rust-lang/rust#156119 (Further optimize `SliceIndex<str>` impl for `Range<usize>`) - rust-lang/rust#157224 (Manually unroll loop in `str::floor_char_boundary`) - rust-lang/rust#157289 (Add infallible primitive type lookups to template arg resolver) - rust-lang/rust#157540 (Cleanup and optimize `render_impls`) - rust-lang/rust#157444 (Couple of work product cleanups) - rust-lang/rust#157543 (Reorganize `tests/ui/issues` [5/N]) - rust-lang/rust#153513 (Syntactically reject equality predicates) - rust-lang/rust#155797 (LineWriter: cap write_vectored newline scan to avoid quadratic write_all_vectored) - rust-lang/rust#156155 (macros: report unbound metavariables directly) - rust-lang/rust#156188 (riscv: promote d, e, and f target_features to CfgStableToggleUnstable) - rust-lang/rust#156666 (Clarify meaning of ranges in pointer offset docs) - rust-lang/rust#157078 (Document equivalence of `highest_one` and `ilog2` methods on integers) - rust-lang/rust#157129 (ci: update download-artifact action to v8) - rust-lang/rust#157169 (triagebot: Update messages to direct changes to appropriate repositories) - rust-lang/rust#157323 (Document Repeat::last panic behavior) - rust-lang/rust#157370 (Clarify MaybeUninit::zeroed padding docs) - rust-lang/rust#157399 (Silence llbc's output by default to prevent rustc's linker output warning) - rust-lang/rust#157500 (Improve documentation of `align_of` and `Alignment`.) - rust-lang/rust#157545 (Suggest using comma to separate valid attribute list items) - rust-lang/rust#157559 (chore: Update annotate-snippets to 0.12.16) - rust-lang/rust#157560 (In `copy_nonoverlapping`, use `mul nuw nsw` to compute the byte size) - rust-lang/rust#157580 (Importing suggestion reported twice when reporting privacy error) - rust-lang/rust#157581 (Test fixup)

Use `mul nuw nsw` in `intrinsics::copy` Essentially the same as rust-lang#157560, just for `copy` instead of `copy_nonoverlapping`. > Yeah, in fact both copy and copy_nonoverlapping could use this since we know the result must be at most isize::MAX else it cannot be inbounds. > ~ rust-lang#157560 (comment) r? saethlin

Rollup merge of #157588 - scottmcm:copy-nsuw, r=saethlin Use `mul nuw nsw` in `intrinsics::copy` Essentially the same as #157560, just for `copy` instead of `copy_nonoverlapping`. > Yeah, in fact both copy and copy_nonoverlapping could use this since we know the result must be at most isize::MAX else it cannot be inbounds. > ~ #157560 (comment) r? saethlin

In copy_nonoverlapping, use mul nuw nsw to compute the byte count

4af2ca0

rustbot assigned saethlin Jun 7, 2026

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 7, 2026

This comment has been minimized.

Sign in to view

rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jun 7, 2026

rust-bors Bot pushed a commit that referenced this pull request Jun 7, 2026

Auto merge of #157560 - scottmcm:mul_nuw_nsw_in_memcpy, r=<try>

1fe6841

In `copy_nonoverlapping`, use `mul nuw nsw` to compute the byte size

This comment has been minimized.

Sign in to view

rustbot removed the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jun 7, 2026

rust-bors Bot 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 Jun 7, 2026

JonathanBrouwer mentioned this pull request Jun 7, 2026

Rollup of 21 pull requests #157582

Closed

This was referenced Jun 7, 2026

Rollup of 21 pull requests #157583

Closed

Rollup of 25 pull requests #157586

Merged

This was referenced Jun 7, 2026

Use mul nuw nsw in intrinsics::copy #157588

Merged

Collapse icmps due to nuw+noundef llvm/llvm-project#202240

Open

rust-bors Bot merged commit 753665a into rust-lang:main Jun 8, 2026
13 checks passed

rustbot added this to the 1.98.0 milestone Jun 8, 2026

scottmcm deleted the mul_nuw_nsw_in_memcpy branch June 8, 2026 01:13

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

In `copy_nonoverlapping`, use `mul nuw nsw` to compute the byte size#157560

In `copy_nonoverlapping`, use `mul nuw nsw` to compute the byte size#157560
rust-bors[bot] merged 1 commit into
rust-lang:mainfrom
scottmcm:mul_nuw_nsw_in_memcpy

scottmcm commented Jun 7, 2026 •

edited by rustbot

Loading

Uh oh!

RalfJung commented Jun 7, 2026

Uh oh!

saethlin commented Jun 7, 2026

Uh oh!

This comment has been minimized.

This comment has been minimized.

rust-bors Bot commented Jun 7, 2026

Uh oh!

This comment has been minimized.

rust-timer commented Jun 7, 2026

Uh oh!

saethlin commented Jun 7, 2026

Uh oh!

rust-bors Bot commented Jun 7, 2026

Uh oh!

scottmcm commented Jun 8, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Uh oh!

Conversation

scottmcm commented Jun 7, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

RalfJung commented Jun 7, 2026

Uh oh!

saethlin commented Jun 7, 2026

Uh oh!

This comment has been minimized.

This comment has been minimized.

rust-bors Bot commented Jun 7, 2026

Uh oh!

This comment has been minimized.

rust-timer commented Jun 7, 2026

Overall result: no relevant changes - no action needed

Instruction count

Max RSS (memory usage)

Cycles

Binary size

Uh oh!

saethlin commented Jun 7, 2026

Uh oh!

rust-bors Bot commented Jun 7, 2026

Uh oh!

scottmcm commented Jun 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

scottmcm commented Jun 7, 2026 •

edited by rustbot

Loading

scottmcm commented Jun 8, 2026 •

edited

Loading