[mir-opt] Allow SRoA when an object is transmuted to a field type#158291
Draft
scottmcm wants to merge 1 commit into
Draft
[mir-opt] Allow SRoA when an object is transmuted to a field type#158291scottmcm wants to merge 1 commit into
scottmcm wants to merge 1 commit into
Conversation
Collaborator
|
Some changes occurred to MIR optimizations cc @rust-lang/wg-mir-opt |
rustbot
added
S-waiting-on-review
This was referenced Jun 24, 2026
JonathanBrouwer
added a commit
to JonathanBrouwer/rust
that referenced
this pull request
Jul 1, 2026
…d, r=aapoalas
Implement `ptr::{read,write}_unaligned` via `repr(packed)`
We already support doing unaligned reads & writes via `repr(packed)` fields, so this just uses that support from the backend rather than needing to thinks about `memcpy` and `intrinsics::forget` and such.
Turns out in codegen this ends up essentially identical because the packed type read gets `BackendRepr::Memory` and thus the read/write of the packed type *is* an `align 1` memcpy, just like the library code before this PR. But doing this gives much simpler MIR and rust-lang#158291 will make it better than the previous version by not needing to emit the `memcpy` at all for things like `read_unaligned::<u32>`.
First commit are some tests that pass on master; Second commit changes the implementation and shows the corresponding test changes.
r? libs
jhpratt
added a commit
to jhpratt/rust
that referenced
this pull request
Jul 1, 2026
…d, r=aapoalas
Implement `ptr::{read,write}_unaligned` via `repr(packed)`
We already support doing unaligned reads & writes via `repr(packed)` fields, so this just uses that support from the backend rather than needing to thinks about `memcpy` and `intrinsics::forget` and such.
Turns out in codegen this ends up essentially identical because the packed type read gets `BackendRepr::Memory` and thus the read/write of the packed type *is* an `align 1` memcpy, just like the library code before this PR. But doing this gives much simpler MIR and rust-lang#158291 will make it better than the previous version by not needing to emit the `memcpy` at all for things like `read_unaligned::<u32>`.
First commit are some tests that pass on master; Second commit changes the implementation and shows the corresponding test changes.
r? libs
scottmcm
force-pushed
the
even-better-read-unaligned
branch
from
7ed426a to
fd33c1f
Compare
Collaborator
|
This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed. Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers. |
scottmcm
force-pushed
the
even-better-read-unaligned
branch
from
fd33c1f to
a576ef9
Compare
scottmcm
commented
Jul 2, 2026
Comment on lines
-42
to
+41
| StorageLive(_4); | ||
| _4 = copy (*_3); | ||
| _5 = copy _4 as T (Transmute); | ||
| StorageDead(_4); | ||
| _4 = copy ((*_3).0: T); |
Member
Author
There was a problem hiding this comment.
The library change to use packed landed in #158427, so now this shows the critical change from this optimization...
Comment on lines
-25
to
+26
| // CHECK-NOT: !noundef | ||
| // CHECK-NOT: !range | ||
| // CHECK-SAME: !range [[R16:![0-9]+]] | ||
| // CHECK-SAME: !noundef |
Member
Author
There was a problem hiding this comment.
...which then means these loads get metadata on them without codegen changes.
Member
Author
|
The more I poked at this the less I'm confident that SRoA is the correct answer here, especially since I tried going a bit further and ended up just getting things like this bb0: {
StorageLive(_3);
_3 = PtrMetadata(copy _1);
StorageLive(_5);
StorageLive(_4);
_4 = &raw const (*_1);
- _5 = copy _4 as std::ptr::NonNull<[T]> (Transmute);
+ _5 = copy _4 as (*const [T]) is !null (Transmute);
StorageDead(_4);
StorageLive(_7);
StorageLive(_6);
_6 = copy _5 as *mut [T] (Transmute);
_7 = copy _6 as *mut T (PtrToPtr);
StorageDead(_6);
- _8 = copy _7 as std::ptr::NonNull<T> (Transmute);
+ _8 = copy _7 as (*const T) is !null (Transmute);
StorageDead(_7);
StorageDead(_5);
switchInt(const <T as std::mem::SizedTypeProperties>::IS_ZST) -> [0: bb1, otherwise: bb2];
}which is kinda the opposite of the direction in #153085 which I think would be nicer to see when reading the MIR. |
rustbot
added
S-waiting-on-author
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Inspired by #158202 (comment), this allows SRoA to expand a local consumed by a
CastKind::Transmuteif that transmute is to the type of a field, since if it's to the type of a field we can just have it take the local for that field like we would if it was a field projection.Future PRs could do more complex stuff, like using layout to always allow SRoA of
xinTransmute(x)when x is https://doc.rust-lang.org/nightly/nightly-rustc/rustc_middle/ty/layout/type.TyAndLayout.html#method.non_1zst_field, but this is enough for the specific case I needed.r? mir-opt