Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

More layout sanity checks #99983

Merged
merged 5 commits into from
Aug 7, 2022
Merged

Conversation

RalfJung
Copy link
Member

r? @eddyb

@rustbot rustbot added the T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. label Jul 31, 2022
@rust-highfive rust-highfive added the S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. label Jul 31, 2022
@RalfJung RalfJung force-pushed the more-layout-checks branch from fea513d to c618abe Compare July 31, 2022 13:32
@rust-log-analyzer

This comment was marked as resolved.

@RalfJung RalfJung force-pushed the more-layout-checks branch from b180d53 to 6e27150 Compare August 1, 2022 13:32
@rust-log-analyzer

This comment has been minimized.

@@ -124,6 +124,7 @@ mod erase_regions;
mod generics;
mod impls_ty;
mod instance;
mod layout_sanity_check;
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

(this is most of a comment first written when I started a review on the original state of the PR but didn't submit it right away)

You could deal with the large file by putting the checks in ty/layout_checks.rs or ty/layout/checks.rs (the latter would mean mod checks; in layout.rs and would be able to do use super::*; to improve everything including private items - I have mixed feelings but it's practical I think).

(thoughts unactionable in this PR follow below)

Honestly we should consider to what crate to move the entire layout module, since it's mostly just providing queries, and it feels awkward where it is.

I also wish it was rustc_middle::ty::abi or rustc_middle::abi (could even go into mir but I doubt that makes as much sense) for consistency with rustc_target::abi but that's another mess.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I wish we used the term 'abi' less, not more, given how ambiguous it is.^^

Comment on lines 24 to 25
/// Yields non-1-ZST fields of the type
fn non_zst_fields<'tcx, 'a>(
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could this have 1zst/1_zst or align1zst/align1_zst in the name instead of zst?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It doesn't actually produce 1-ZST though, I had to change it to produce all ZST. Will adjust the comment.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

FWIW this function seems useful to have elsewhere as well? Not sure where to put it though.

@eddyb
Copy link
Member

eddyb commented Aug 6, 2022

Hmm, so you know how tools like pointee_info_at dig into a type to find primitive leaves?

It might be easier to make an abstraction like this, if only for the layout checks (note: "cases" below refers to enum Variants::Multiple and union fields):

// HACK(eddyb) used below to disambiguate between the two uses of `Size`,
// but we should probably have this as a completely separate type, with:
// `Size`: measured in bytes and/or bits; Add+Mul ops (Sub not needed)
// `Offset`: measured in bytes (no "bit offsets"); Add+Sub ops (Mul not needed)
type Offset = Size;

// HACK(eddyb) necessary method sadly missing from `std::ops::Range`.
impl<T> Range<T> {
    fn includes(&self, other: Range<T>) -> bool {
        self.start <= other.start && other.end <= self.end
    }
}

struct ScalarLeaf {
    /// Where in the parent layout this leaf is.
    ///
    /// Must be the right size, i.e. `extent.end - extent.start == primitive`.
    //
    // FIXME(eddyb) does the above condition make `extent.end` redundant?
    // (move to a `fn extent(&self, cx: ...) -> Range<Offset>` helper method?)
    extent: Range<Offset>,

    // FIXME(eddyb) instead of ignoring validity ranges, should they be merged
    // across e.g. `enum` variants?
    primitive: Primitive,
}

struct ScalarLeafWithinError {
    /// No leaves found in `search_extent`.
    OnlyPadding,

    /// A leaf intersects `search_extent` without being contained in it,
    /// i.e. the leaf is "straddling" `search_extent`'s start or end.
    PartialIntersection(ScalarLeaf),

    /// Two leaves found within `search_extent` that differ in their `extent`,
    /// `primitive`, or both.
    Mismatch(ScalarLeaf, ScalarLeaf),
}

/// Find the unique `Scalar` (but ignoring validity ranges, so... `Primitive`¹?)
/// leaf field wholly contained within `search_extent` in `layout`.
/// 
/// An `Ok(leaf)` result indicates a leaf field that is:
/// * "wholly contained within": `search_extent.includes(leaf.extent)`
/// * "unique": `layout` always contains (across all `enum`/`union` cases):
///   * `Scalar(leaf.primitive)` exactly filling `leaf.extent`
///     * some `enum`/`union` cases may contain padding at `leaf.extent` instead,
///       but all other `enum`/`union` cases must agree on the same `leaf`
///     * `Scalar` validity might vary, but the `Primitive` must match exactly
///   * only padding, in `search_extent` "around" `leaf.extent`,
///     i.e. neither of these extents contain any leaves of their own:
///     * `search_extent.start..leaf.extent.start` (before `leaf`)
///     * `leaf.extent.end..search_extent.end` (after `leaf`)
/// 
/// ¹ I wish `Primitive` was called `Scalar` but what would we call `Scalar`?
///   (maybe `Primitive` should be renamed to `ScalarType`/`ScalarKind`?)
fn scalar_leaf_within(
    layout: &TyAndLayout<'tcx>,
    search_extent: Range<Offset>,
) -> Result<ScalarLeaf, ScalarLeafWithinError> {
    fn try_for_each_leaf_intersecting<E>(
        base_offset: Offset,
        layout: &TyAndLayout<'tcx>,
        filter_extent: Range<Offset>,
        each: &mut impl FnMut(ScalarLeaf) -> Result<(), E>
    ) -> Result<(), E> {
        for (i, field_offset) in layout.fields.enumerate() {
            if field_offset < filter_extent.end {
                let field = layout.field(i);
                if filter_extent.start < field_offset + field.size {
                    try_for_each_leaf_intersecting(
                        base_offset + field_offset,
                        field,
                        filter_extent,
                        each,
                    )?;
                }
            }
        }
        for v in 0..layout.variants.len() {
            try_for_each_leaf_intersecting(base_offset, layout.for_variant(v), filter_extent, each)?;
        }

        if layout.fields.is_empty() && !layout.is_zst() {
            match layout.abi {
                Abi::Scalar(scalar) => each(ScalarLeaf {
                    extent: base_offset..base_offset+scalar.size(),
                    primitive: scalar.primitive,
                })?,

                // FIXME(eddyb) can we guarantee all non-ZST leaves are `Scalar`?
                _ => unreachable!(),
            }
        }

        Ok(())
    }

    let mut found = None;
    try_for_each_leaf_intersecting(
        Offset::ZERO,
        layout,
        search_extent,
        |candidate| match found {
            _ if !search_extent.includes(&candidate.extent) => {
                Err(ScalarLeafError::PartialIntersection(candidate))
            }
            Some(previous) if candidate != previous => {
                Err(ScalarLeafError::Mismatch(previous, candidate))
            }
            _ => {
                found = Some(candidate);
                Ok(())
            }
        }
    )?;
    Ok(found.ok_or(ScalarLeafWithinError::OnlyPadding))
}

With that abstraction in hand, we can define the conditions for:

  • Abi::Scalar(scalar):
// HACK(eddyb) needed to deal with `Offset` vs `Size` type differences.
let layout_extent = Offset::ZERO..Offset::ZERO+layout.size;

assert_eq!(
    scalar_leaf_within(layout, layout_extent),
    Ok(ScalarLeaf {
        extent: layout_extent,
        primitive: scalar.primitive,
    }),
);
  • Abi::ScalarPair(a, b):
// HACK(eddyb) needed to deal with `Offset` vs `Size` type differences.
let layout_extent = Offset::ZERO..Offset::ZERO+layout.size;
let a_extent = Offset::ZERO..Offset::ZERO+a.size;
// FIXME(eddyb) there should be a nicer way to make `start..start+len` ranges.
let b_extent = a_extent.end.align_to(b.align)..;
let b_extent = b_extent.start..b_extent.start+b.size;

assert_eq!(
    scalar_leaf_within(layout, layout_extent.start..b_extent.start),
    Ok(ScalarLeaf {
        extent: a_extent,
        primitive: a.primitive,
    }),
);
assert_eq!(
    scalar_leaf_within(layout, b_extent.start..layout_extent.end),
    Ok(ScalarLeaf {
        extent: b_extent,
        primitive: b.primitive,
    }),
);

This should cover any field hierarchy, but here are some extra thoughts:

  • I wasted spent a lot of time on designing scalar_leaf_within's interface,
    but try_for_each_leaf_intersecting might be better to use directly?
    • its filter_extent could be removed if we outright just want to iterate all
      leaves and for e.g. Abi::ScalarPair(a, b) we check whether every found leaf
      is either matching a or b (in extent and primitive)
  • this isn't trying to tackle Scalar validity ranges, but I don't think that's
    possible without making scalar_leaf_within recursive (which I avoided above
    to make it simpler, as I'm effectively flattening all the leaves together),
    and having logic for merging across enum variants and union fields
    • even then, you'd still have to handle e.g. NonZero* types and niches
  • we should probably have a separate Offset from Size no matter what
  • the whole "extent" terminology was an attempt at minimizing ambiguity/confusion
    over ideas like "a subslice of a layout" (as in, the subset of a layout's
    fields that occupy bytes exclusively in a subrange of 0..layout.size),
    but I'm not sure it's optimal at all, ideas welcome

Copy link
Member

@eddyb eddyb left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

r=me if you don't want to take up getting The Fun Layout Search pseudocode working (would still be good to have an issue tracking how good these checks are etc.)

@RalfJung
Copy link
Member Author

RalfJung commented Aug 7, 2022

I opened an issue at #100231, also linking specifically to your proposed checks.

@bors r=eddyb rollup=iffy

@bors
Copy link
Contributor

bors commented Aug 7, 2022

📌 Commit 27013d2 has been approved by eddyb

It is now in the queue for this repository.

@bors bors 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 Aug 7, 2022
@bors
Copy link
Contributor

bors commented Aug 7, 2022

⌛ Testing commit 27013d2 with merge 5a9c3a2...

@bors
Copy link
Contributor

bors commented Aug 7, 2022

☀️ Test successful - checks-actions
Approved by: eddyb
Pushing 5a9c3a2 to master...

@bors bors added the merged-by-bors This PR was explicitly merged by bors. label Aug 7, 2022
@bors bors merged commit 5a9c3a2 into rust-lang:master Aug 7, 2022
@rustbot rustbot added this to the 1.65.0 milestone Aug 7, 2022
@rust-timer
Copy link
Collaborator

Finished benchmarking commit (5a9c3a2): comparison url.

Instruction count

This benchmark run did not return any relevant results for this metric.

Max RSS (memory usage)

Results
  • Primary benchmarks: no relevant changes found
  • Secondary benchmarks: 😿 relevant regression found
mean1 max count2
Regressions 😿
(primary)
N/A N/A 0
Regressions 😿
(secondary)
2.5% 2.5% 1
Improvements 🎉
(primary)
N/A N/A 0
Improvements 🎉
(secondary)
N/A N/A 0
All 😿🎉 (primary) N/A N/A 0

Cycles

Results
  • Primary benchmarks: no relevant changes found
  • Secondary benchmarks: 😿 relevant regression found
mean1 max count2
Regressions 😿
(primary)
N/A N/A 0
Regressions 😿
(secondary)
2.8% 2.8% 1
Improvements 🎉
(primary)
N/A N/A 0
Improvements 🎉
(secondary)
N/A N/A 0
All 😿🎉 (primary) N/A N/A 0

If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf.

@rustbot label: -perf-regression

Footnotes

  1. the arithmetic mean of the percent change 2

  2. number of relevant changes 2

@RalfJung RalfJung deleted the more-layout-checks branch August 8, 2022 13:03
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
merged-by-bors This PR was explicitly merged by bors. S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue.
Projects
None yet
Development

Successfully merging this pull request may close these issues.

7 participants