valtree performance tuning
Summary: This PR makes type checking of code with many type-level constants faster.
After https://github.com/rust-lang/rust/pull/136180 was merged, we observed a small perf regression (https://github.com/rust-lang/rust/pull/136318#issuecomment-2635562821). This happened because that PR introduced additional copies in the fast reject code path for consts, which is very hot for certain crates: 6c1d960d88/compiler/rustc_type_ir/src/fast_reject.rs (L486-L487)
This PR improves the performance again by properly interning the valtrees so that copying and comparing them becomes faster. This will become especially useful with `feature(adt_const_params)`, so the fast reject code doesn't have to do a deep compare of the valtrees.
Note that we can't just compare the interned consts themselves in the fast reject, because sometimes `'static` lifetimes in the type are be replaced with inference variables (due to canonicalization) on one side but not the other.
A less invasive alternative that I considered is simply avoiding copies introduced by https://github.com/rust-lang/rust/pull/136180 and comparing the valtrees it in-place (see commit: 9e91e50ac5 / perf results: https://github.com/rust-lang/rust/pull/136593#issuecomment-2642303245), however that was still measurably slower than interning.
There are some minor regressions in secondary benchmarks: These happen due to changes in memory allocations and seem acceptable to me. The crates that make heavy use of valtrees show no significant changes in memory usage.
It is speculated that these two can be conceptually merged, and it can
start by ripping out rustc's notion of the PtxKernel call convention.
Leave the ExternAbi for now, but the nvptx target now should see it as
just a different way to spell Conv::GpuKernel.
Implement MIR lowering for unsafe binders
This is the final bit of the unsafe binders puzzle. It implements MIR, CTFE, and codegen for unsafe binders, and enforces that (for now) they are `Copy`. Later on, I'll introduce a new trait that relaxes this requirement to being "is `Copy` or `ManuallyDrop<T>`" which more closely models how we treat union fields.
Namely, wrapping unsafe binders is now `Rvalue::WrapUnsafeBinder`, which acts much like an `Rvalue::Aggregate`. Unwrapping unsafe binders are implemented as a MIR projection `ProjectionElem::UnwrapUnsafeBinder`, which acts much like `ProjectionElem::Field`.
Tracking:
- https://github.com/rust-lang/rust/issues/130516
Insert null checks for pointer dereferences when debug assertions are enabled
Similar to how the alignment is already checked, this adds a check
for null pointer dereferences in debug mode. It is implemented similarly
to the alignment check as a `MirPass`.
This inserts checks in the same places as the `CheckAlignment` pass and additionally
also inserts checks for `Borrows`, so code like
```rust
let ptr: *const u32 = std::ptr::null();
let val: &u32 = unsafe { &*ptr };
```
will have a check inserted on dereference. This is done because null references
are UB. The alignment check doesn't cover these places, because in `&(*ptr).field`,
the exact requirement is that the final reference must be aligned. This is something to
consider further enhancements of the alignment check.
For now this is implemented as a separate `MirPass`, to make it easy to disable
this check if necessary.
This is related to a 2025H1 project goal for better UB checks in debug
mode: https://github.com/rust-lang/rust-project-goals/pull/177.
r? `@saethlin`
Similar to how the alignment is already checked, this adds a check
for null pointer dereferences in debug mode. It is implemented similarly
to the alignment check as a MirPass.
This is related to a 2025H1 project goal for better UB checks in debug
mode: https://github.com/rust-lang/rust-project-goals/pull/177.
Fix deduplication mismatches in vtables leading to upcasting unsoundness
We currently have two cases where subtleties in supertraits can trigger disagreements in the vtable layout, e.g. leading to a different vtable layout being accessed at a callsite compared to what was prepared during unsizing. Namely:
### #135315
In this example, we were not normalizing supertraits when preparing vtables. In the example,
```
trait Supertrait<T> {
fn _print_numbers(&self, mem: &[usize; 100]) {
println!("{mem:?}");
}
}
impl<T> Supertrait<T> for () {}
trait Identity {
type Selff;
}
impl<Selff> Identity for Selff {
type Selff = Selff;
}
trait Middle<T>: Supertrait<()> + Supertrait<T> {
fn say_hello(&self, _: &usize) {
println!("Hello!");
}
}
impl<T> Middle<T> for () {}
trait Trait: Middle<<() as Identity>::Selff> {}
impl Trait for () {}
fn main() {
(&() as &dyn Trait as &dyn Middle<()>).say_hello(&0);
}
```
When we prepare `dyn Trait`, we see a supertrait of `Middle<<() as Identity>::Selff>`, which itself has two supertraits `Supertrait<()>` and `Supertrait<<() as Identity>::Selff>`. These two supertraits are identical, but they are not duplicated because we were using structural equality and *not* considering normalization. This leads to a vtable layout with two trait pointers.
When we upcast to `dyn Middle<()>`, those two supertraits are now the same, leading to a vtable layout with only one trait pointer. This leads to an offset error, and we call the wrong method.
### #135316
This one is a bit more interesting, and is the bulk of the changes in this PR. It's a bit similar, except it uses binder equality instead of normalization to make the compiler get confused about two vtable layouts. In the example,
```
trait Supertrait<T> {
fn _print_numbers(&self, mem: &[usize; 100]) {
println!("{mem:?}");
}
}
impl<T> Supertrait<T> for () {}
trait Trait<T, U>: Supertrait<T> + Supertrait<U> {
fn say_hello(&self, _: &usize) {
println!("Hello!");
}
}
impl<T, U> Trait<T, U> for () {}
fn main() {
(&() as &'static dyn for<'a> Trait<&'static (), &'a ()>
as &'static dyn Trait<&'static (), &'static ()>)
.say_hello(&0);
}
```
When we prepare the vtable for `dyn for<'a> Trait<&'static (), &'a ()>`, we currently consider the PolyTraitRef of the vtable as the key for a supertrait. This leads two two supertraits -- `Supertrait<&'static ()>` and `for<'a> Supertrait<&'a ()>`.
However, we can upcast[^up] without offsetting the vtable from `dyn for<'a> Trait<&'static (), &'a ()>` to `dyn Trait<&'static (), &'static ()>`. This is just instantiating the principal trait ref for a specific `'a = 'static`. However, when considering those supertraits, we now have only one distinct supertrait -- `Supertrait<&'static ()>` (which is deduplicated since there are two supertraits with the same substitutions). This leads to similar offsetting issues, leading to the wrong method being called.
[^up]: I say upcast but this is a cast that is allowed on stable, since it's not changing the vtable at all, just instantiating the binder of the principal trait ref for some lifetime.
The solution here is to recognize that a vtable isn't really meaningfully higher ranked, and to just treat a vtable as corresponding to a `TraitRef` so we can do this deduplication more faithfully. That is to say, the vtable for `dyn for<'a> Tr<'a>` and `dyn Tr<'x>` are always identical, since they both would correspond to a set of free regions on an impl... Do note that `Tr<for<'a> fn(&'a ())>` and `Tr<fn(&'static ())>` are still distinct.
----
There's a bit more that can be cleaned up. In codegen, we can stop using `PolyExistentialTraitRef` basically everywhere. We can also fix SMIR to stop storing `PolyExistentialTraitRef` in its vtable allocations.
As for testing, it's difficult to actually turn this into something that can be tested with `rustc_dump_vtable`, since having multiple supertraits that are identical is a recipe for ambiguity errors. Maybe someone else is more creative with getting that attr to work, since the tests I added being run-pass tests is a bit unsatisfying. Miri also doesn't help here, since it doesn't really generate vtables that are offset by an index in the same way as codegen.
r? `@lcnr` for the vibe check? Or reassign, idk. Maybe let's talk about whether this makes sense.
<sup>(I guess an alternative would also be to not do any deduplication of vtable supertraits (or only a really conservative subset) rather than trying to normalize and deduplicate more faithfully here. Not sure if that works and is sufficient tho.)</sup>
cc `@steffahn` -- ty for the minimizations
cc `@WaffleLapkin` -- since you're overseeing the feature stabilization :3
Fixes#135315Fixes#135316
miri: optimize zeroed alloc
When allocating zero-initialized memory in MIR interpretation, rustc allocates zeroed memory, marks it as initialized and then re-zeroes it. Remove the last step.
I don't expect this to have much of an effect on performance normally, but in my case in which I'm creating a large allocation via mmap it gets in the way.
Get rid of `mir::Const::from_ty_const`
This function is strange, because it turns valtrees into `mir::Const::Value`, but the rest of the const variants stay as type system consts.
All of the callsites except for one in `instsimplify` (array length simplification of `ptr_metadata` call) just go through the valtree arm of the function, so it's easier to just create a `mir::Const` directly for those.
For the instsimplify case, if we have a type system const we should *keep* having a type system const, rather than turning it into a `mir::Const::Value`; it doesn't really matter in practice, though, bc `usize` has no padding, but it feels more principled.
The amdgpu-kernel calling convention was reverted in commit
f6b21e90d1 due to inactivity in the amdgpu
target.
Introduce a `gpu-kernel` calling convention that translates to
`ptx_kernel` or `amdgpu_kernel`, depending on the target that rust
compiles for.
Begin to implement type system layer of unsafe binders
Mostly TODOs, but there's a lot of match arms that are basically just noops so I wanted to split these out before I put up the MIR lowering/projection part of this logic.
r? oli-obk
Tracking:
- https://github.com/rust-lang/rust/issues/130516
`rustc_span::symbol` defines some things that are re-exported from
`rustc_span`, such as `Symbol` and `sym`. But it doesn't re-export some
closely related things such as `Ident` and `kw`. So you can do `use
rustc_span::{Symbol, sym}` but you have to do `use
rustc_span::symbol::{Ident, kw}`, which is inconsistent for no good
reason.
This commit re-exports `Ident`, `kw`, and `MacroRulesNormalizedIdent`,
and changes many `rustc_span::symbol::` qualifiers in `compiler/` to
`rustc_span::`. This is a 200+ net line of code reduction, mostly
because many files with two `use rustc_span` items can be reduced to
one.
Remove queries from the driver interface
All uses of driver queries in the public api of rustc_driver have been removed in https://github.com/rust-lang/rust/pull/134130 already. This removes driver queries from rustc_interface and does a couple of cleanups around TyCtxt construction and entering enabled by this removal.
Finishes the removal of driver queries started with https://github.com/rust-lang/rust/pull/126834.
Some more refactorings towards removing driver queries
Follow up to https://github.com/rust-lang/rust/pull/127184
## Custom driver breaking change
The `after_analysis` callback is changed to accept `TyCtxt` instead of `Queries`. The only safe query in `Queries` to call at this point is `global_ctxt()` which allows you to enter the `TyCtxt` either way. To fix your custom driver, replace the `queries: &'tcx Queries<'tcx>` argument with `tcx: TyCtxt<'tcx>` and remove your `queries.global_ctxt().unwrap().enter(|tcx| { ... })` call and only keep the contents of the closure.
## Custom driver deprecation
The `after_crate_root_parsing` callback is now deprecated. Several custom drivers are incorrectly calling `queries.global_ctxt()` from inside of it, which causes some driver code to be skipped. As such I would like to either remove it in the future or if custom drivers still need it, change it to accept an `&rustc_ast::Crate` instead.
take 2
open up coroutines
tweak the wordings
the lint works up until 2021
We were missing one case, for ADTs, which was
causing `Result` to yield incorrect results.
only include field spans with significant types
deduplicate and eliminate field spans
switch to emit spans to impl Drops
Co-authored-by: Niko Matsakis <nikomat@amazon.com>
collect drops instead of taking liveness diff
apply some suggestions and add explantory notes
small fix on the cache
let the query recurse through coroutine
new suggestion format with extracted variable name
fine-tune the drop span and messages
bugfix on runtime borrows
tweak message wording
filter out ecosystem types earlier
apply suggestions
clippy
check lint level at session level
further restrict applicability of the lint
translate bid into nop for stable mir
detect cycle in type structure
the behavior of the type system not only depends on the current
assumptions, but also the currentnphase of the compiler. This is
mostly necessary as we need to decide whether and how to reveal
opaque types. We track this via the `TypingMode`.
[StableMIR] API to retrieve definitions from crates
Add functions to retrieve function definitions and static items from all crates (local and external).
For external crates, we're still missing items from trait implementation and primitives.
r? ````@compiler-errors:```` Do you know what is the best way to retrieve the associated items for primitives and trait implementations for external crates? Thanks!
Add functions to retrieve function definitions and static items from
all crates (local and external).
For external crates, add a query to retrieve the number of defs in a
foreign crate.
