Since the mode is no longer part of `meta_item`, we must insert it manually (otherwise macro expansion with `#[rustc_autodiff]` won't work).
This can be revised later if a more structured representation becomes necessary (using enums, annotated structs, etc).
Some tests are currently failing. I'll address them next.
Most notably, the `FIXME` for suboptimal printing of `use` groups in
`tests/ui/macros/stringify.rs` is fixed. And all other test output
changes result in pretty printed output being closer to the original
formatting in the source code.
Autodiff batching2
~I will rebase it once my first PR landed.~ done.
This autodiff batch mode is more similar to scalar autodiff, since it still only takes one shadow argument.
However, that argument is supposed to be `width` times larger.
r? `@oli-obk`
Tracking:
- https://github.com/rust-lang/rust/issues/124509
- fix errors caused by the move of `ast::Item::ident` (see #138740)
- move the logic of getting `sig`, `vis`, and `ident` from two seperate
`match` statements into one (less repetition especially with the
nested `match`)
Autodiff batching
Enzyme supports batching, which is especially known from the ML side when training neural networks.
There we would normally have a training loop, where in each iteration we would pass in some data (e.g. an image), and a target vector. Based on how close we are with our prediction we compute our loss, and then use backpropagation to compute the gradients and update our weights.
That's quite inefficient, so what you normally do is passing in a batch of 8/16/.. images and targets, and compute the gradients for those all at once, allowing better optimizations.
Enzyme supports batching in two ways, the first one (which I implemented here) just accepts a Batch size,
and then each Dual/Duplicated argument has not one, but N shadow arguments. So instead of
```rs
for i in 0..100 {
df(x[i], y[i], 1234);
}
```
You can now do
```rs
for i in 0..100.step_by(4) {
df(x[i+0],x[i+1],x[i+2],x[i+3], y[i+0], y[i+1], y[i+2], y[i+3], 1234);
}
```
which will give the same results, but allows better compiler optimizations. See the testcase for details.
There is a second variant, where we can mark certain arguments and instead of having to pass in N shadow arguments, Enzyme assumes that the argument is N times longer. I.e. instead of accepting 4 slices with 12 floats each, we would accept one slice with 48 floats. I'll implement this over the next days.
I will also add more tests for both modes.
For any one preferring some more interactive explanation, here's a video of Tim's llvm dev talk, where he presents his work. https://www.youtube.com/watch?v=edvaLAL5RqU
I'll also add some other docs to the dev guide and user docs in another PR.
r? ghost
Tracking:
- https://github.com/rust-lang/rust/issues/124509
- https://github.com/rust-lang/rust/issues/135283
`ast::Item` has an `ident` field.
- It's always non-empty for these item kinds: `ExternCrate`, `Static`,
`Const`, `Fn`, `Mod`, `TyAlias`, `Enum`, `Struct`, `Union`,
`Trait`, `TraitAlias`, `MacroDef`, `Delegation`.
- It's always empty for these item kinds: `Use`, `ForeignMod`,
`GlobalAsm`, `Impl`, `MacCall`, `DelegationMac`.
There is a similar story for `AssocItemKind` and `ForeignItemKind`.
Some sites that handle items check for an empty ident, some don't. This
is a very C-like way of doing things, but this is Rust, we have sum
types, we can do this properly and never forget to check for the
exceptional case and never YOLO possibly empty identifiers (or possibly
dummy spans) around and hope that things will work out.
The commit is large but it's mostly obvious plumbing work. Some notable
things.
- `ast::Item` got 8 bytes bigger. This could be avoided by boxing the
fields within some of the `ast::ItemKind` variants (specifically:
`Struct`, `Union`, `Enum`). I might do that in a follow-up; this
commit is big enough already.
- For the visitors: `FnKind` no longer needs an `ident` field because
the `Fn` within how has one.
- In the parser, the `ItemInfo` typedef is no longer needed. It was used
in various places to return an `Ident` alongside an `ItemKind`, but
now the `Ident` (if present) is within the `ItemKind`.
- In a few places I renamed identifier variables called `name` (or
`foo_name`) as `ident` (or `foo_ident`), to better match the type, and
because `name` is normally used for `Symbol`s. It's confusing to see
something like `foo_name.name`.
Mostly parser: Eliminate code that's been dead / semi-dead since the removal of type ascription syntax
**Disclaimer**: This PR is intended to mostly clean up code as opposed to bringing about behavioral changes. Therefore it doesn't aim to address any of the 'FIXME: remove after a month [dated: 2023-05-02]: "type ascription syntax has been removed, see issue [#]101728"'.
---
By commit:
1. Removes truly dead code:
* Since 1.71 (#109128) `let _ = { f: x };` is a syntax error as opposed to a semantic error which allows the parse-time diagnostic (suggestion) "*struct literal body without path // you might have forgotten […]*" to kick in.
* The analysis-time diagnostic (suggestion) from <=1.70 "*cannot find value \`f\` in this scope // you might have forgotten […]*" is therefore no longer reachable.
2. Updates `is_certainly_not_a_block` to be in line with the current grammar:
* The seq. `{ ident:` is definitely not the start of a block. Before the removal of ty ascr, `{ ident: ty_start` would begin a block expr.
* This shouldn't make more code compile IINM, it should *ultimately* only affect diagnostics.
* For example, `if T { f: () } {}` will now be interpreted as an `if` with struct lit `T { f: () }` as its *condition* (which is banned in the parser anyway) as opposed to just `T` (with the *consequent* being `f : ()` which is also invalid (since 1.71)). The diagnostics are almost the same because we have two separate parse recovery procedures + diagnostics: `StructLiteralNeedingParens` (*invalid struct lit*) before and `StructLiteralNotAllowedHere` (*struct lits aren't allowed here*) now, as you can see from the diff.
* (As an aside, even before this PR, fn `maybe_suggest_struct_literal` should've just used the much older & clearer `StructLiteralNotAllowedHere`)
* NB: This does sadly regress the compiler output for `tests/ui/parser/type-ascription-in-pattern.rs` but that can be fixed in follow-up PRs. It's not super important IMO and a natural consequence.
3. Removes code that's become dead due to the prior commit.
* Basically reverts #106620 + #112475 (without regressing rustc's output!).
* Now the older & more robust parse recovery procedure (cc `StructLiteralNotAllowedHere`) takes care of the cases the removed code used to handle.
* This automatically fixes the suggestions for \[[playground](https://play.rust-lang.org/?version=stable&mode=debug&edition=2024&gist=7e2030163b11ee96d17adc3325b01780)\]:
* `if Ty::<i32> { f: K }.m() {}`: `if Ty::<i32> { SomeStruct { f: K } }.m() {}` (broken) → ` if (Ty::<i32> { f: K }).m() {}`
* `if <T as Trait>::Out { f: K::<> }.m() {}`: `if <T as Trait>(::Out { f: K::<> }).m() {}` (broken) → `if (<T as Trait>::Out { f: K::<> }).m() {}`
4. Merge and simplify UI tests pertaining to this issue, so it's easier to add more regression tests like for the two cases mentioned above.
5. Merge UI tests and add the two regression tests.
Best reviewed commit by commit (on request I'll partially squash after approval).
Rollup of 10 pull requests
Successful merges:
- #137715 (Allow int literals for pattern types with int base types)
- #138002 (Disable CFI for weakly linked syscalls)
- #138051 (Add support for downloading GCC from CI)
- #138231 (Prevent ICE in autodiff validation by emitting user-friendly errors)
- #138245 (stabilize `ci_rustc_if_unchanged_logic` test for local environments)
- #138256 (Do not feed anon const a type that references generics that it does not have)
- #138284 (Do not write user type annotation for const param value path)
- #138296 (Remove `AdtFlags::IS_ANONYMOUS` and `Copy`/`Clone` condition for anonymous ADT)
- #138352 (miri native_calls: ensure we actually expose *mutable* provenance to the memory FFI can access)
- #138354 (remove redundant `body` arguments)
r? `@ghost`
`@rustbot` modify labels: rollup
Prevent ICE in autodiff validation by emitting user-friendly errors
This PR moves `valid_ret_activity` and `valid_input_activity` checks to the macro expansion phase in compiler/rustc_builtin_macros/src/autodiff.rs, replacing the following internal compiler error (ICE):
```
error: internal compiler error:
compiler/rustc_codegen_ssa/src/codegen_attrs.rs:935:13:
Invalid input activity Dual for Reverse mode
```
with a more user-friendly message.
The issue specifically affected the test file `tests/ui/autodiff/autodiff_illegal.rs`, impacting the functions `f5` and `f6`.
The ICE can be reproduced by following [Enzyme's Rustbook](https://enzymead.github.io/rustbook/installation.html) installation guide.
Additionally, this PR adds tests for invalid return activity in `autodiff_illegal.rs`, which previously triggered an unnoticed ICE before these fixes.
r? ``@oli-obk``
`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.
