This replaces the single Vec allocation with a series of progressively
larger buckets. With the cfg for parallel enabled but with -Zthreads=1,
this looks like a slight regression in i-count and cycle counts (<0.1%).
With the parallel frontend at -Zthreads=4, this is an improvement (-5%
wall-time from 5.788 to 5.4688 on libcore) than our current Lock-based
approach, likely due to reducing the bouncing of the cache line holding
the lock. At -Zthreads=32 it's a huge improvement (-46%: 8.829 -> 4.7319
seconds).
Subtree sync for rustc_codegen_cranelift
Apart from a perf optimization for some crates (https://github.com/rust-lang/rustc_codegen_cranelift/pull/1541) not much changed this time as the last sync was less than a week ago.
r? `@ghost`
`@rustbot` label +A-codegen +A-cranelift +T-compiler
Rollup of 5 pull requests
Successful merges:
- #132552 (Add v9, v8plus, and leoncasa target feature to sparc and use v8plus in create_object_file)
- #132745 (pointee_info_at: fix logic for recursing into enums)
- #132777 (try_question_mark_nop: update test for LLVM 20)
- #132785 (rustc_target: more target string fixes for LLVM 20)
- #132794 (Use a separate dir for r-a builds consistently in helix config)
r? `@ghost`
`@rustbot` modify labels: rollup
Use a separate dir for r-a builds consistently in helix config
r? `@jieyouxu`
cc `@mrkajetanp`
Previously config used `build-rust-analyzer` for rustfmt and proc macros server, while not using it for actual `x check` commands.
This PR:
- Replaces the build dir with `build/rust-analyzer`
- This is just my preference
- Although I do think this is the nicest option: the directory is already git-ignored, `rm -fr ./build` removes everything, etc
- Uses said directory with the `x check` commands in helix r-a config
- Adds instructions on how to build rustfmt and proc macro server to the config
As of note, this is not what other configs do (like vscode's), however this _is_ what I would actually suggest to people (and what I'm actually using).
rustc_target: more target string fixes for LLVM 20
LLVM continues to clean these up, and we continue to make this consistent. This is similar to 9caced7bad and e985396145.
`@rustbot` label: +llvm-main
try_question_mark_nop: update test for LLVM 20
llvm/llvm-project@dd116369f6 changes the IR of this test in a way that I don't think is bad, but needs adjusting.
r? `@nikic`
`@rustbot` label: +llvm-main
pointee_info_at: fix logic for recursing into enums
Fixes https://github.com/rust-lang/rust/issues/131834
The logic in `pointee_info_at` was likely written at a time when the null pointer optimization was the *only* enum layout optimization -- and as `Variant::Multiple` kept getting expanded, nobody noticed that the logic is now unsound.
The job of this function is to figure out whether there is a dereferenceable-or-null and aligned pointer at a given offset inside a type. So when we recurse into a multi-variant enum, we better make sure that all the other enum variants must be null! This is the part that was forgotten, and this PR adds it.
The reason this didn't explode in many ways so far is that our references only have 1 niche value (null), so it's not possible on stable to have a multi-variant enum with a dereferenceable pointer and other enum variants that are not null. But with `rustc_layout_scalar_valid_range` attributes one can force such a layout, and if `@the8472's` work on alignment niches ever lands, that will make this possible on stable.
Trim and tidy includes in `rustc_llvm`
These includes tend to accumulate over time, and are usually only removed when something breaks in a new LLVM version, so it's nice to clean them up manually once in a while.
General strategy used for this PR:
- Remove all includes from `LLVMWrapper.h` that aren't needed by the header itself, transplanting them to individual source files as necessary.
- For each source file, temporarily remove each include if doing so doesn't cause a compile error.
- If a “required” include looks like it shouldn't be needed, try replacing it with its sub-includes, then trim that list.
- After doing all of the above, go back and re-add any removed include if the file does actually use things defined in that header, even if the header happens to also be included by something else.
use verbose for path separator suggestion
A single `-` of suggestion underlining that is adjacent to a much more significant `^^^` underlying of the LHS path component is hard to distinguish. IMO this presents much more cleanly when it's verbose, especially because it's a *replacment* suggestion.
r? estebank
Don't suggest `.into_iter()` on iterators
This makes the the suggestion to call `.into_iter()` only consider unsatisfied `Iterator` bounds for the receiver type itself. That way, it ignores predicates generated by trying to auto-ref the receiver (the result of which usually won't implement `Iterator`).
Fixes#127511
Unfortunately, the error in that case is still confusing: it labels `Iterator` as an unsatisfied bound because `&impl Iterator: Iterator` can't be satisfied, despite that not being required or helpful. I'd like to handle that in a separate PR. ~~I'm hoping fixing #124802 will fix it too.~~ It doesn't look connected to that issue. Still, I think it'd be clearest to visually distinguish unsatisfied predicates from different attempts at `pick_method`; I'll make a PR for that soon.
Get rid of `check_opaque_type_well_formed`
Instead, replicate it by improving the span of the opaque in `check_opaque_meets_bounds`.
This has two consequences:
1. We now prefer "concrete type differs" errors, since we'll hit those first before we check the opaque is WF.
2. Spans have gotten slightly worse.
Specifically, (2.) could be improved by adding a new obligation cause that explains that the definition's environment has stronger assumptions than the declaration.
r? lcnr
Make `RustString` an extern type to avoid `improper_ctypes` warnings
Currently, any FFI function that uses `&RustString` needs to also add `#[ignore(improper_ctypes)]` to silence a warning.
The warning is not _completely_ bogus, because `RustString` contains `Vec<u8>` and therefore does not have a guaranteed layout. But we have no way of telling the lint that this doesn't matter, because the C++ code only uses that pointer opaquely and never relies on its underlying layout.
Ideally there would be some way to silence `improper_ctypes` at the type-definition site. But because there isn't, casting to and from a separate extern type is better than having to annotate every single use site.
remove support for rustc_safe_intrinsic attribute; use rustc_intrinsic functions instead
This brings us one step closer towards removing support for `extern "rust-intrinsic"` blocks, in favor of `#[rustc_intrinsic]` functions.
Also move `#[rustc_intrinsic]` under the `intrinsics` feature gate, to match the `extern "rust-intrinsic"` style.
Initialize channel `Block`s directly on the heap
The channel's `Block::new` was causing a stack overflow because it held
32 item slots, instantiated on the stack before moving to `Box::new`.
The 32x multiplier made modestly-large item sizes untenable.
That block is now initialized directly on the heap.
Fixes#102246
try-job: test-various
Simplify FFI calls for `-Ztime-llvm-passes` and `-Zprint-codegen-stats`
The existing code for these unstable LLVM-infodump flags was jumping through hoops to pass an allocated C string across the FFI boundary, when it's much simpler to just write to a `&RustString` instead.
coverage: Extract safe FFI wrapper functions to `llvm_cov`
This PR takes all of the inline `unsafe` calls in coverage codegen, and all the safe wrapper functions in `coverageinfo/mod.rs`, and moves them to a new `llvm_cov` submodule that is dedicated to safe FFI wrapper functions. This reduces the mixing of abstraction levels in the rest of coverage codegen.
As a follow-up, this PR also tidies up the names and signatures of several of the coverage FFI functions.
coverage: Simplify parts of coverage graph creation
This is a combination of three semi-related simplifications to how coverage graphs are created, grouped into one PR to avoid conflicts.
There are no observable changes to the output of any of the coverage tests.
