Autodiff Upstreaming - enzyme frontend
This is an upstream PR for the `autodiff` rustc_builtin_macro that is part of the autodiff feature.
For the full implementation, see: https://github.com/rust-lang/rust/pull/129175
**Content:**
It contains a new `#[autodiff(<args>)]` rustc_builtin_macro, as well as a `#[rustc_autodiff]` builtin attribute.
The autodiff macro is applied on function `f` and will expand to a second function `df` (name given by user).
It will add a dummy body to `df` to make sure it type-checks. The body will later be replaced by enzyme on llvm-ir level,
we therefore don't really care about the content. Most of the changes (700 from 1.2k) are in `compiler/rustc_builtin_macros/src/autodiff.rs`, which expand the macro. Nothing except expansion is implemented for now.
I have a fallback implementation for relevant functions in case that rustc should be build without autodiff support. The default for now will be off, although we want to flip it later (once everything landed) to on for nightly. For the sake of CI, I have flipped the defaults, I'll revert this before merging.
**Dummy function Body:**
The first line is an `inline_asm` nop to make inlining less likely (I have additional checks to prevent this in the middle end of rustc. If `f` gets inlined too early, we can't pass it to enzyme and thus can't differentiate it.
If `df` gets inlined too early, the call site will just compute this dummy code instead of the derivatives, a correctness issue. The following black_box lines make sure that none of the input arguments is getting optimized away before we replace the body.
**Motivation:**
The user facing autodiff macro can verify the user input. Then I write it as args to the rustc_attribute, so from here on I can know that these values should be sensible. A rustc_attribute also turned out to be quite nice to attach this information to the corresponding function and carry it till the backend.
This is also just an experiment, I expect to adjust the user facing autodiff macro based on user feedback, to improve usability.
As a simple example of what this will do, we can see this expansion:
From:
```
#[autodiff(df, Reverse, Duplicated, Const, Active)]
pub fn f1(x: &[f64], y: f64) -> f64 {
unimplemented!()
}
```
to
```
#[rustc_autodiff]
#[inline(never)]
pub fn f1(x: &[f64], y: f64) -> f64 {
::core::panicking::panic("not implemented")
}
#[rustc_autodiff(Reverse, Duplicated, Const, Active,)]
#[inline(never)]
pub fn df(x: &[f64], dx: &mut [f64], y: f64, dret: f64) -> f64 {
unsafe { asm!("NOP"); };
::core::hint::black_box(f1(x, y));
::core::hint::black_box((dx, dret));
::core::hint::black_box(f1(x, y))
}
```
I will add a few more tests once I figured out why rustc rebuilds every time I touch a test.
Tracking:
- https://github.com/rust-lang/rust/issues/124509
try-job: dist-x86_64-msvc
Special treatment empty tuple when suggest adding a string literal in format macro.
For example:
```rust
let s = "123";
println!({}, "sss", s);
```
Suggest:
`println!("{:?} {} {}", {}, "sss", s);`
fixes#130170
Retire the `unnamed_fields` feature for now
`#![feature(unnamed_fields)]` was implemented in part in #115131 and #115367, however work on that feature has (afaict) stalled and in the mean time there have been some concerns raised (e.g.[^1][^2]) about whether `unnamed_fields` is worthwhile to have in the language, especially in its current desugaring. Because it represents a compiler implementation burden including a new kind of anonymous ADT and additional complication to field selection, and is quite prone to bugs today, I'm choosing to remove the feature.
However, since I'm not one to really write a bunch of words, I'm specifically *not* going to de-RFC this feature. This PR essentially *rolls back* the state of this feature to "RFC accepted but not yet implemented"; however if anyone wants to formally unapprove the RFC from the t-lang side, then please be my guest. I'm just not totally willing to summarize the various language-facing reasons for why this feature is or is not worthwhile, since I'm coming from the compiler side mostly.
Fixes#117942Fixes#121161Fixes#121263Fixes#121299Fixes#121722Fixes#121799Fixes#126969Fixes#131041
Tracking:
* https://github.com/rust-lang/rust/issues/49804
[^1]: https://rust-lang.zulipchat.com/#narrow/stream/213817-t-lang/topic/Unnamed.20struct.2Funion.20fields
[^2]: https://github.com/rust-lang/rust/issues/49804#issuecomment-1972619108
Don't allow the `#[pointee]` attribute where it doesn't belong
Error if the `#[pointee]` attribute is applied to anything but generic type parameters.
Closes#128485
Related to #123430
- fix for divergence
- fix error message
- fix another cranelift test
- fix some cranelift things
- don't set the NORETURN option for naked asm
- fix use of naked_asm! in doc comment
- fix use of naked_asm! in run-make test
- use `span_bug` in unreachable branch
Fix error span if arg to `asm!()` is a macro call
Fixes#129503
When the argument to `asm!()` is a macro call, e.g. `asm!(concat!("abc", "{} pqr"))`, and there's an error in the resulting template string, we do not take into account the presence of this macro call while computing the error span. This PR fixes that. Now we will use the entire thing between the parenthesis of `asm!()` as the error span in this situation e.g. for `asm!(concat!("abc", "{} pqr"))` the error span will be `concat!("abc", "{} pqr")`.
When the template string passed to asm!() is produced by
a macro call like concat!() we were producing wrong error
spans. Now in the case of a macro call we just use the entire
arg to asm!(), macro call and all, as the error span.
- Replace non-standard names like 's, 'p, 'rg, 'ck, 'parent, 'this, and
'me with vanilla 'a. These are cases where the original name isn't
really any more informative than 'a.
- Replace names like 'cx, 'mir, and 'body with vanilla 'a when the lifetime
applies to multiple fields and so the original lifetime name isn't
really accurate.
- Put 'tcx last in lifetime lists, and 'a before 'b.
Simplify some nested `if` statements
Applies some but not all instances of `clippy::collapsible_if`. Some ended up looking worse afterwards, though, so I left those out. Also applies instances of `clippy::collapsible_else_if`
Review with whitespace disabled please.
Fix false positive with `missing_docs` and `#[test]`
Since #130025, the compiler don't ignore missing_docs when compiling the tests. But there is now a false positive warning for every `#[test]`
For example, this code
```rust
//! Crate docs
fn just_a_test() {}
```
Would emit this warning when running `cargo test`
```
warning: missing documentation for a constant
--> src/lib.rs:5:1
|
4 | #[test]
| ------- in this procedural macro expansion
5 | fn just_a_test() {}
| ^^^^^^^^^^^^^^^^^^^
```
Since #130025, the compiler don't ignore missing_docs when compiling the tests.
But there is now a false positive warning for every `#[test]`
For example, this code
```rust
//! Crate docs
fn just_a_test() {}
```
Would emit this warning when running `cargo test`
```
warning: missing documentation for a constant
--> src/lib.rs:5:1
|
4 | #[test]
| ------- in this procedural macro expansion
5 | fn just_a_test() {}
| ^^^^^^^^^^^^^^^^^^^
```
Also emit `missing_docs` lint with `--test` to fulfil expectations
This PR removes the "test harness" suppression of the `missing_docs` lint to be able to fulfil `#[expect]` (expectations) as it is now "relevant".
I think the goal was to maybe avoid false-positive while linting on public items under `#[cfg(test)]` but with effective visibility we should no longer have any false-positive.
Another possibility would be to query the lint level and only emit the lint if it's of expect level, but that is even more hacky.
Fixes https://github.com/rust-lang/rust/issues/130021
try-job: x86_64-gnu-aux
in this commit, `naked_asm!` is an alias for `asm!` with one difference: `options(noreturn)` is always enabled by `naked_asm!`. That makes it future-compatible for when `naked_asm!` starts disallowing `options(noreturn)` later.
debug-fmt-detail option
I'd like to propose a new option that makes `#[derive(Debug)]` generate no-op implementations that don't print anything, and makes `{:?}` in format strings a no-op.
There are a couple of motivations for this:
1. A more thorough stripping of debug symbols. Binaries stripped of debug symbols still retain some of them through `Debug` implementations. It's hard to avoid that without compiler's help, because debug formatting can be used in many places, including dependencies, and their loggers, asserts, panics, etc.
* In my testing it gives about 2% binary size reduction on top of all other binary-minimizing best practices (including `panic_immediate_abort`). There are targets like Web WASM or embedded where users pay attention to binary sizes.
* Users distributing closed-source binaries may not want to "leak" any symbol names as a matter of principle.
2. Adds ability to test whether code depends on specifics of the `Debug` format implementation in unwise ways (e.g. trying to get data unavailable via public interface, or using it as a serialization format). Because current Rust's debug implementation doesn't change, there's a risk of it becoming a fragile de-facto API that [won't be possible to change in the future](https://www.hyrumslaw.com/). An option that "breaks" it can act as a [grease](https://www.rfc-editor.org/rfc/rfc8701.html).
This implementation is a `-Z fmt-debug=opt` flag that takes:
* `full` — the default, current state.
* `none` — makes derived `Debug` and `{:?}` no-ops. Explicit `impl Debug for T` implementations are left unharmed, but `{:?}` format won't use them, so they may get dead-code eliminated if they aren't invoked directly.
* `shallow` — makes derived `Debug` print only the type's name, without recursing into fields. Fieldless enums print their variant names. `{:?}` works.
The `shallow` option is a compromise between minimizing the `Debug` code, and compatibility. There are popular proc-macro crates that use `Debug::fmt` as a way to convert enum values into their Rust source code.
There's a corresponding `cfg` flag: `#[cfg(fmt_debug = "none")]` that can be used in user code to react to this setting to minimize custom `Debug` implementations or remove unnecessary formatting helper functions.
derive(SmartPointer): assume pointee from the single generic and better error messages
Fix#129465
Actually RFC says that `#[pointee]` can be inferred when there is no ambiguity, or there is only one generic type parameter so to say.
cc ```@Darksonn```
r? ```@compiler-errors```
Add `#[warn(unreachable_pub)]` to a bunch of compiler crates
By default `unreachable_pub` identifies things that need not be `pub` and tells you to make them `pub(crate)`. But sometimes those things don't need any kind of visibility. So they way I did these was to remove the visibility entirely for each thing the lint identifies, and then add `pub(crate)` back in everywhere the compiler said it was necessary. (Or occasionally `pub(super)` when context suggested that was appropriate.) Tedious, but results in more `pub` removal.
There are plenty more crates to do but this seems like enough for a first PR.
r? `@compiler-errors`
Stabilize `unsafe_attributes`
# Stabilization report
## Summary
This is a tracking issue for the RFC 3325: unsafe attributes
We are stabilizing `#![feature(unsafe_attributes)]`, which makes certain attributes considered 'unsafe', meaning that they must be surrounded by an `unsafe(...)`, as in `#[unsafe(no_mangle)]`.
RFC: rust-lang/rfcs#3325
Tracking issue: #123757
## What is stabilized
### Summary of stabilization
Certain attributes will now be designated as unsafe attributes, namely, `no_mangle`, `export_name`, and `link_section` (stable only), and these attributes will need to be called by surrounding them in `unsafe(...)` syntax. On editions prior to 2024, this is simply an edition lint, but it will become a hard error in 2024. This also works in `cfg_attr`, but `unsafe` is not allowed for any other attributes, including proc-macros ones.
```rust
#[unsafe(no_mangle)]
fn a() {}
#[cfg_attr(any(), unsafe(export_name = "c"))]
fn b() {}
```
For a table showing the attributes that were considered to be included in the list to require unsafe, and subsequent reasoning about why each such attribute was or was not included, see [this comment here](https://github.com/rust-lang/rust/pull/124214#issuecomment-2124753464)
## Tests
The relevant tests are in `tests/ui/rust-2024/unsafe-attributes` and `tests/ui/attributes/unsafe`.
Use more slice patterns inside the compiler
Nothing super noteworthy. Just replacing the common 'fragile' pattern of "length check followed by indexing or unwrap" with slice patterns for legibility and 'robustness'.
r? ghost
improve error message when `global_asm!` uses `asm!` operands
follow-up to https://github.com/rust-lang/rust/pull/128207
what was
```
error: expected expression, found keyword `in`
--> src/lib.rs:1:31
|
1 | core::arch::global_asm!("{}", in(reg));
| ^^ expected expression
```
becomes
```
error: the `in` operand cannot be used with `global_asm!`
--> $DIR/parse-error.rs:150:19
|
LL | global_asm!("{}", in(reg));
| ^^ the `in` operand is not meaningful for global-scoped inline assembly, remove it
```
the span of the error is just the keyword, which means that we can't create a machine-applicable suggestion here. The alternative would be to attempt to parse the full operand, but then if there are syntax errors in the operand those would be presented to the user, even though the parser already knows that the output won't be valid. Also that would require more complexity in the parser.
So I think this is a nice improvement at very low cost.
