| Age | Commit message (Collapse) | Author | Lines |
|---|
|
|
|
Refactor the partitioning module to make it easier to introduce new algorithms
I've split the `librustc_mir::monomorphize::partitioning` module into a few files and introduced a `Partitioner` trait which allows us to decouple the partitioning algorithm from the code which integrates it into the query system. This should allow us to introduce new partitioning algorithms much more easily. I've also gone ahead and added a `-Z` flag to control which algorithm is used (currently there is only the `default`).
I left a few comments in places where things might be improved further.
r? @pnkfelix cc @rust-lang/wg-incr-comp
|
|
This commit adjusts the naming of various lang items so that they are
consistent and don't include prefixes containing the target or
"LangItem". In addition, lang item variants are no longer exported from
the `lang_items` module.
Signed-off-by: David Wood <david@davidtw.co>
|
|
|
|
|
|
This will allow us to prototype different partitioning schemes without
adding a lot of extra conditionals everywhere.
|
|
This commit modifies polymorphization's handling of predicates so that
if any generic parameter is used in a predicate then all parameters in
that predicate are used.
Signed-off-by: David Wood <david@davidtw.co>
|
|
This commit adjusts polymorphization's handling of predicates so that
after ensuring that `T` is used in `I: Foo<T>` if `I` is used, it now
ensures that `I` is used if `T` is used in `I: Foo<T>`. This is
necessary to mark generic parameters that only exist in impl parameters
as used - thereby avoiding symbol clashes when using the new mangling
scheme.
Signed-off-by: David Wood <david@davidtw.co>
|
|
This commit constrains the support added for handling unevaluated consts
in polymorphization (introduced in #75260) by:
- Skipping associated constants as this causes cycle errors.
- Skipping promoted constants when they contain `Self` as this ensures
`T` is used in constants of the form `<Self as Foo<T>>`.
Signed-off-by: David Wood <david@davidtw.co>
|
|
polymorphize: unevaluated constants
This PR makes polymorphization visit the promoted MIR of unevaluated constants with available promoted MIR instead of visiting the substitutions of that constant - which will mark all of the generic parameters as used; in addition polymorphization will now visit non-promoted unevaluated constants rather than visit their substs.
r? @lcnr
|
|
By moving `{known,used}_attrs` from `SessionGlobals` to `Session`. This
means they are accessed via the `Session`, rather than via TLS. A few
`Attr` methods and `librustc_ast` functions are now methods of
`Session`.
All of this required passing a `Session` to lots of functions that didn't
already have one. Some of these functions also had arguments removed, because
those arguments could be accessed directly via the `Session` argument.
`contains_feature_attr()` was dead, and is removed.
Some functions were moved from `librustc_ast` elsewhere because they now need
to access `Session`, which isn't available in that crate.
- `entry_point_type()` --> `librustc_builtin_macros`
- `global_allocator_spans()` --> `librustc_metadata`
- `is_proc_macro_attr()` --> `Session`
|
|
This commit makes polymorphization visit non-promoted unevaluated
constants rather than visit their substs directly.
Signed-off-by: David Wood <david@davidtw.co>
|
|
This commit makes polymorphization visited the MIR of unevaluated
constants with available promoted MIR instead of visiting the
substitutions of that constant - which will mark all of the generic
parameters as used.
Signed-off-by: David Wood <david@davidtw.co>
|
|
This commit skips encoding empty polymorphization results - while
polymorphization is disabled, this should be every polymorphization
result; but when polymorphization is re-enabled, this would help with
non-generic functions and those which do use all their parameters (most
functions).
Signed-off-by: David Wood <david@davidtw.co>
|
|
This commit changes polymorphization to return a `FiniteBitSet<u32>`
rather than a `FiniteBitSet<u64>` because most functions do not use
anywhere near sixty-four generic parameters so keeping a `u64` around is
unnecessary in most cases.
Signed-off-by: David Wood <david@davidtw.co>
|
|
|
|
This commit changes the span and content of the "collection encountered
polymorphic constant" bug in monomorphization collection to point to the
use of the constant rather than the definition.
Signed-off-by: David Wood <david@davidtw.co>
|
|
This commit introduces a `FiniteBitSet` type which replaces the manual
bit manipulation which was being performed in polymorphization.
Signed-off-by: David Wood <david@davidtw.co>
|
|
This commit replaces the `-Z polymorphize-errors` debugging flag with a
`#[rustc_polymorphize_error]` attribute for use on functions.
Signed-off-by: David Wood <david@davidtw.co>
|
|
This commit implements the `unused_generic_params` query, an initial
version of polymorphization which detects when an item does not use
generic parameters and is being needlessly monomorphized as a result.
Signed-off-by: David Wood <david@davidtw.co>
|
|
This commit renames `should_monomorphize_locally` to
`should_codegen_locally` which better describes what the function
determines once polymorphization is added.
Signed-off-by: David Wood <david@davidtw.co>
|
|
rustc now generates the coverage map and can support (limited)
coverage report generation, at the function level.
Example:
$ BUILD=$HOME/rust/build/x86_64-unknown-linux-gnu
$ $BUILD/stage1/bin/rustc -Zinstrument-coverage \
$HOME/rust/src/test/run-make-fulldeps/instrument-coverage/main.rs
$ LLVM_PROFILE_FILE="main.profraw" ./main
called
$ $BUILD/llvm/bin/llvm-profdata merge -sparse main.profraw -o main.profdata
$ $BUILD/llvm/bin/llvm-cov show --instr-profile=main.profdata main
1| 1|pub fn will_be_called() {
2| 1| println!("called");
3| 1|}
4| |
5| 0|pub fn will_not_be_called() {
6| 0| println!("should not have been called");
7| 0|}
8| |
9| 1|fn main() {
10| 1| let less = 1;
11| 1| let more = 100;
12| 1|
13| 1| if less < more {
14| 1| will_be_called();
15| 1| } else {
16| 1| will_not_be_called();
17| 1| }
18| 1|}
|
|
Support const args in type dependent paths (Take 2)
once more, except it is sound this time :smiling_face_with_three_hearts: previously #71154
-----
```rust
#![feature(const_generics)]
struct A;
impl A {
fn foo<const N: usize>(&self) -> usize { N }
}
struct B;
impl B {
fn foo<const N: usize>(&self) -> usize { 42 }
}
fn main() {
let a = A;
a.foo::<7>();
}
```
When calling `type_of` for generic const arguments, we now use the `TypeckTables` of the surrounding body to get the expected type.
This alone causes cycle errors though, as we now have `typeck_tables_of(main)` -> `...` ->
`type_of(main_ANON0 := 7)` -> `typeck_tables_of(main)` :zap: (see https://github.com/rust-lang/rust/issues/68400#issuecomment-611760290)
To prevent this we must not call `type_of(const_arg)` during `typeck_tables_of`. This is achieved by
calling `type_of(param_def_id)` instead.
We have to somehow remember the `DefId` of the param through all of typeck, which is done using the
struct `ty::WithOptConstParam<DefId>`, which replaces `DefId` where needed and contains an `Option<DefId>` to
be able to store the const parameter in case it exists.
Queries which are currently cached on disk are split into two variants: `query_name`(cached) and `query_name_(of|for)_const_arg`(not cached), with `query_name_of_const_arg` taking a pair `(did, param_did): (LocalDefId, DefId)`.
For some queries a method `query_name_of_opt_const_arg` is added to `TyCtxt` which takes a `ty::WithOptConstParam` and either calls `query_name` or `query_name_of_const_arg` depending on the value of `const_param_did`.
r? @eddyb @varkor
|
|
|
|
|
|
|
|
This eliminates a bunch of `Symbol::intern()` and `Symbol::as_str()`
calls, which is good, because they require locking the interner.
Note that the unsafety in `from_cycle_error()` is identical to the
unsafety on other adjacent impls.
|
|
|
|
r=ecstatic-morse
Point at the call span when overflow occurs during monomorphization
This improves the output for issue #72577, but there's still more work
to be done.
Currently, an overflow error during monomorphization results in an error
that points at the function we were unable to monomorphize. However, we
don't point at the call that caused the monomorphization to happen. In
the overflow occurs in a large recursive function, it may be difficult
to determine where the issue is.
This commit tracks and `Span` information during collection of
`MonoItem`s, which is used when emitting an overflow error. `MonoItem`
itself is unchanged, so this only affects
`src/librustc_mir/monomorphize/collector.rs`
|
|
This improves the output for issue #72577, but there's still more work
to be done.
Currently, an overflow error during monomorphization results in an error
that points at the function we were unable to monomorphize. However, we
don't point at the call that caused the monomorphization to happen. In
the overflow occurs in a large recursive function, it may be difficult
to determine where the issue is.
This commit tracks and `Span` information during collection of
`MonoItem`s, which is used when emitting an overflow error. `MonoItem`
itself is unchanged, so this only affects
`src/librustc_mir/monomorphize/collector.rs`
|
|
typo fix
Co-authored-by: Jonas Schievink <jonasschievink@gmail.com>
|
|
This commit fixes typos in the doc comments of 'librustc_mir/monomorphize/collector.rs'
|
|
r=Mark-Simulacrum
Change how compiler-builtins gets many CGUs
This commit intends to fix an accidental regression from #70846. The
goal of #70846 was to build compiler-builtins with a maximal number of
CGUs to ensure that each module in the source corresponds to an object
file. This high degree of control for compiler-builtins is desirable to
ensure that there's at most one exported symbol per CGU, ideally
enabling compiler-builtins to not conflict with the system libgcc as
often.
In #70846, however, only part of the compiler understands that
compiler-builtins is built with many CGUs. The rest of the compiler
thinks it's building with `sess.codegen_units()`. Notably the
calculation of `sess.lto()` consults `sess.codegen_units()`, which when
there's only one CGU it disables ThinLTO. This means that
compiler-builtins is built without ThinLTO, which is quite harmful to
performance! This is the root of the cause from #73135 where intrinsics
were found to not be inlining trivial functions.
The fix applied in this commit is to remove the special-casing of
compiler-builtins in the compiler. Instead the build system is now
responsible for special-casing compiler-builtins. It doesn't know
exactly how many CGUs will be needed but it passes a large number that
is assumed to be much greater than the number of source-level modules
needed. After reading the various locations in the compiler source, this
seemed like the best solution rather than adding more and more special
casing in the compiler for compiler-builtins.
Closes #73135
|
|
remove visit_terminator_kind from MIR visitor
For some reason, we had both `visit_terminator` and `visit_terminator_kind`. In contrast, for `Statement` we just have `visit_statement`. So this cleans things up by removing `visit_terminator_kind` and porting its users to `visit_terminator`.
|
|
Fix link error with #[thread_local] introduced by #71192
r? @oli-obk
|
|
|
|
|
|
This commit intends to fix an accidental regression from #70846. The
goal of #70846 was to build compiler-builtins with a maximal number of
CGUs to ensure that each module in the source corresponds to an object
file. This high degree of control for compiler-builtins is desirable to
ensure that there's at most one exported symbol per CGU, ideally
enabling compiler-builtins to not conflict with the system libgcc as
often.
In #70846, however, only part of the compiler understands that
compiler-builtins is built with many CGUs. The rest of the compiler
thinks it's building with `sess.codegen_units()`. Notably the
calculation of `sess.lto()` consults `sess.codegen_units()`, which when
there's only one CGU it disables ThinLTO. This means that
compiler-builtins is built without ThinLTO, which is quite harmful to
performance! This is the root of the cause from #73135 where intrinsics
were found to not be inlining trivial functions.
The fix applied in this commit is to remove the special-casing of
compiler-builtins in the compiler. Instead the build system is now
responsible for special-casing compiler-builtins. It doesn't know
exactly how many CGUs will be needed but it passes a large number that
is assumed to be much greater than the number of source-level modules
needed. After reading the various locations in the compiler source, this
seemed like the best solution rather than adding more and more special
casing in the compiler for compiler-builtins.
Closes #73135
|
|
Fix #[thread_local] statics as asm! sym operands
The `asm!` RFC specifies that `#[thread_local]` statics may be used as `sym` operands for inline assembly.
This also fixes a regression in the handling of `#[thread_local]` during monomorphization which caused link-time errors with multiple codegen units, most likely introduced by #71192.
r? @oli-obk
|
|
Rename traits::Vtable to ImplSource.
Originally suggested by @eddyb.
r? @nikomatsakis
|
|
|
|
|
|
|
|
|
|
|
|
Make TLS accesses explicit in MIR
r? @rust-lang/wg-mir-opt
cc @RalfJung @vakaras for miri thread locals
cc @bjorn3 for cranelift
fixes #70685
|
|
r=varkor
mir: adjust conditional in recursion limit check
Fixes #67552.
This PR adjusts the condition used in the recursion limit check of
the monomorphization collector, from `>` to `>=`.
In #67552, the test case had infinite indirect recursion, repeating a
handful of functions (from the perspective of the monomorphization
collector): `rec` -> `identity` -> `Iterator::count` -> `Iterator::fold`
-> `Iterator::next` -> `rec`.
During this process, `resolve_associated_item` was invoked for
`Iterator::fold` (during the construction of an `Instance`), and
ICE'd due to substitutions needing inference. However, previous
iterations of this recursion would have called this function for
`Iterator::fold` - and did! - and succeeded in doing so (trivially
checkable from debug logging, `()` is present where `_` is in the substs
of the failing execution).
The expected outcome of this test case would be a recursion limit error
(which is present when the `identity` fn indirection is removed), and
the recursion depth of `rec` is increasing (other functions finish
collecting their neighbours and thus have their recursion depths reset).
When the ICE occurs, the recursion depth of `rec` is 256 (which matches
the recursion limit), which suggests perhaps that a different part of
the compiler is using a `>=` comparison and returning a different result
on this recursion rather than what it returned in every previous
recursion, thus stopping the monomorphization collector from reporting
an error on the next recursion, where `recursion_depth_of_rec > 256`
would have been true.
With grep and some educated guesses, we can determine that
the recursion limit check at line 818 in
`src/librustc_trait_selection/traits/project.rs` is the other check that
is using a different comparison. Modifying either comparison to be `>` or
`>=` respectively will fix the error, but changing the monomorphization
collector produces the nicer error.
|
|
Properly handle InlineAsmOperand::SymFn when collecting monomorphized items
Fixes #72484
|
|
|
|
This commit introduces a `Limit` type which is used to ensure that all
comparisons against limits within the compiler are consistent (which can
result in ICEs if they aren't).
Signed-off-by: David Wood <david@davidtw.co>
|
