Autodiff Upstreaming - rustc_codegen_ssa, rustc_middle
This PR should not be merged until the rustc_codegen_llvm part is merged.
I will also alter it a little based on what get's shaved off from the cg_llvm PR,
and address some of the feedback I received in the other PR (including cleanups).
I am putting it already up to
1) Discuss with `@jieyouxu` if there is more work needed to add tests to this and
2) Pray that there is someone reviewing who can tell me why some of my autodiff invocations get lost.
Re 1: My test require fat-lto. I also modify the compilation pipeline. So if there are any other llvm-ir tests in the same compilation unit then I will likely break them. Luckily there are two groups who currently have the same fat-lto requirement for their GPU code which I have for my autodiff code and both groups have some plans to enable support for thin-lto. Once either that work pans out, I'll copy it over for this feature. I will also work on not changing the optimization pipeline for functions not differentiated, but that will require some thoughts and engineering, so I think it would be good to be able to run the autodiff tests isolated from the rest for now. Can you guide me here please?
For context, here are some of my tests in the samples folder: https://github.com/EnzymeAD/rustbook
Re 2: This is a pretty serious issue, since it effectively prevents publishing libraries making use of autodiff: https://github.com/EnzymeAD/rust/issues/173. For some reason my dummy code persists till the end, so the code which calls autodiff, deletes the dummy, and inserts the code to compute the derivative never gets executed. To me it looks like the rustc_autodiff attribute just get's dropped, but I don't know WHY? Any help would be super appreciated, as rustc queries look a bit voodoo to me.
Tracking:
- https://github.com/rust-lang/rust/issues/124509
r? `@jieyouxu`
When encountering unexpected closure return type, point at return type/expression
```
error[E0271]: expected `{closure@fallback-closure-wrap.rs:18:40}` to be a closure that returns `()`, but it returns `!`
--> $DIR/fallback-closure-wrap.rs:19:9
|
LL | let error = Closure::wrap(Box::new(move || {
| -------
LL | panic!("Can't connect to server.");
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ expected `()`, found `!`
|
= note: expected unit type `()`
found type `!`
= note: required for the cast from `Box<{closure@$DIR/fallback-closure-wrap.rs:18:40: 18:47}>` to `Box<dyn FnMut()>`
```
```
error[E0271]: expected `{closure@dont-ice-for-type-mismatch-in-closure-in-async.rs:6:10}` to be a closure that returns `bool`, but it returns `Option<()>`
--> $DIR/dont-ice-for-type-mismatch-in-closure-in-async.rs:6:16
|
LL | call(|| -> Option<()> {
| ---- ------^^^^^^^^^^
| | |
| | expected `bool`, found `Option<()>`
| required by a bound introduced by this call
|
= note: expected type `bool`
found enum `Option<()>`
note: required by a bound in `call`
--> $DIR/dont-ice-for-type-mismatch-in-closure-in-async.rs:3:25
|
LL | fn call(_: impl Fn() -> bool) {}
| ^^^^ required by this bound in `call`
```
```
error[E0271]: expected `{closure@f670.rs:28:13}` to be a closure that returns `Result<(), _>`, but it returns `!`
--> f670.rs:28:20
|
28 | let c = |e| -> ! {
| -------^
| |
| expected `Result<(), _>`, found `!`
...
32 | f().or_else(c);
| ------- required by a bound introduced by this call
-Ztrack-diagnostics: created at compiler/rustc_trait_selection/src/error_reporting/traits/fulfillment_errors.rs:1433:28
|
= note: expected enum `Result<(), _>`
found type `!`
note: required by a bound in `Result::<T, E>::or_else`
--> /home/gh-estebank/rust/library/core/src/result.rs:1406:39
|
1406 | pub fn or_else<F, O: FnOnce(E) -> Result<T, F>>(self, op: O) -> Result<T, F> {
| ^^^^^^^^^^^^ required by this bound in `Result::<T, E>::or_else`
```
CC #111539.
float::min/max: mention the non-determinism around signed 0
Turns out this can actually produce different results on different machines [in practice](https://github.com/rust-lang/rust/issues/83984#issuecomment-2623859230); that seems worth documenting. I assume LLVM will happily const-fold these operations so so there could be different results for the same input even on the same machine, depending on whether things get const-folded or not.
`@nikic` I remember there was an LLVM soundness fix regarding scalar evolution for loops that had to recognize certain operations as non-deterministic... it seems to me that pass would also have to avoid predicting the result of `llvm.{min,max}num`, for the same reason?
r? `@tgross35`
Cc `@rust-lang/libs-api`
If this lands we should also make Miri non-deterministic here.
Fixes https://github.com/rust-lang/rust/issues/83984
add constraint graph to polonius MIR dump
Another easy one while I work on diagnostics. This PR adds a mermaid visualization of the polonius constraint graph to the polonius MIR dump.
Adding kills is left to a future PR (until they're encoded in edges directly or I set up recording debugging info in and out of the analysis), because right now they're only computed during traversal.
[Here's](https://gistpreview.github.io/?096b0131e8258f9a3125c55c7ac369bc) how that looks.
r? `@matthewjasper` but as always feel free to reroll.
Remove `NamedVarMap`.
`NamedVarMap` is extremely similar to `ResolveBoundVars`. The former contains two `UnordMap<ItemLocalId, T>` fields (obscured behind `ItemLocalMap` typedefs). The latter contains two
`SortedMap<ItemLocalId, T>` fields. We construct a `NamedVarMap` and then convert it into a `ResolveBoundVars` by sorting the `UnordMap`s, which is unnecessary busywork.
This commit removes `NamedVarMap` and constructs a `ResolveBoundVars` directly. `SortedMap` and `NamedVarMap` have slightly different perf characteristics during construction (e.g. speed of insertion) but this code isn't hot enough for that to matter.
A few details to note.
- A `FIXME` comment is removed.
- The detailed comments on the fields of `NamedVarMap` are copied to `ResolveBoundVars` (which has a single, incorrect comment).
- `BoundVarContext::map` is renamed.
- `ResolveBoundVars` gets a derived `Default` impl.
r? `@jackh726`
Stabilize `const_black_box`
This has been unstably const since #92226, but a tracking issue was never created. Per [discussion on Zulip][zulip], there should not be any blockers to making this const-stable. The function does not provide any functionality at compile time but does allow code reuse between const- and non-const functions, so stabilize it here.
[zulip]: https://rust-lang.zulipchat.com/#narrow/channel/146212-t-compiler.2Fconst-eval/topic/const_black_box
It's a function that prints numbers with underscores inserted for
readability (e.g. "1_234_567"), used by `-Zmeta-stats` and
`-Zinput-stats`. It's the only thing in `rustc_middle::util::common`,
which is a bizarre location for it.
This commit:
- moves it to `rustc_data_structures`, a more logical crate for it;
- puts it in a module `thousands`, like the similar crates.io crate;
- renames it `format_with_underscores`, which is a clearer name;
- rewrites it to be more concise;
- slightly improves the testing.
It's a function that does stuff with MIR and yet it weirdly has its own
module in `rustc_middle::util`. This commit moves it into
`rustc_middle::mir`, a more sensible home.
- Use `object` based test logic instead of processing `nm`
human-readable textual output.
- Try to expand test coverage to not be limited to only linux + x86_64.
Co-authored-by: binarycat <binarycat@envs.net>
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
This makes it possible to start incrementally replacing our debuginfo bindings
with the ones in the LLVM-C API, all of which operate on `LLVMDIBuilderRef`.
Instead re-export `rustc_hir_analysis::collect::suggest_impl_trait`,
which is the only thing from the module used in another crate. This
fixes a `FIXME` comment. Also adjust some visibilities to satisfy the
`unreachable_pub` lint.
This changes requires downgrading a link in a comment on `FnCtxt`
because `collect` is no longer public and rustdoc complains otherwise.
This is annoying but I can't see how to avoid it.
`delegation.rs` has three builders: `GenericsBuilder`,
`PredicatesBuilder`, and `GenericArgsBuilder`. The first two builders
have just two optional parameters, and the third one has zero. Each
builder is used within a single function. The code is over-engineered.
This commit removes the builders, replacing each with with a single
`build_*` function. This makes the code shorter and simpler.
There is a comment `Delegation to inherent methods is not yet
supported.` that appears three times mid-pattern and somehow inhibits
rustfmt from formatting the enclosing `match` statement. This commit
moves them to the top of the pattern, which enables more formatting.
This comment made sense when this crate was called `rustc_typeck`, but
makes less sense now that it's called `rustc_hir_analysis`. Especially
given that `check_drop_impl` is only called within the crate.
