Sync rustc_codegen_cranelift
Bunch of bug fixes this time. Also an update to Cranelift 0.93 which adds a brand new optimization pass which cg_clif exposes when using `--release`. And various improvements to cg_clif's test suite, making it faster to run. And finally two small perf improvements.
r? `@ghost`
`@rustbot` label +A-codegen +A-cranelift +T-compiler
Since ascii chars are already handled by a special case in the
`to_lower` and `to_upper` functions, there's no need to waste space on
them in the LUTs.
When bootstrap compiles native dependencies like LLVM, it should set
CMAKE_SYSTEM_NAME for the target system; otherwise cmake may not
identify that it is cross-compiling.
In particular, when building a Linux rustc on a macOS host, cmake was
including `-isysroot /path/to/macOS.sdk` options that caused things to
break. By setting `CMAKE_SYSTEM_NAME=Linux` when building for Linux
targets, the macOS SDK is no longer passed as sysroot to the compiler.
Rollup of 7 pull requests
Successful merges:
- #108991 (add `enable-warnings` flag for llvm, and disable it by default.)
- #109109 (Use `unused_generic_params` from crate metadata)
- #109111 (Create dirs for build_triple)
- #109136 (Simplify proc macro signature validity check)
- #109150 (Update cargo)
- #109154 (Fix MappingToUnit to support no span of arg_ty)
- #109157 (Remove mw from review rotation for a while)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
This patched has been necessary for subtree syncs from the start, but
previously it was necessary to locally install tqc's patched git
version, which hasn't been updated for quite a while. I made a small
change to allow downloading it as script without requiring an entire git
installation for the patched version.
This removes a dependency on ntapi 0.3.x which failed to compiled with
latest beta.
rust-lang/rustc-perf@93dc60d995 removed
that dependency in the upstream rustc-perf.
Install binaries to the miri toolchain's sysroot
The default install produces this behavior:
```
$ cargo +miri miri --version
miri 0.1.0 (0ba1f4a0 2023-03-05)
$ cargo +nightly miri --version
miri 0.1.0 (0ba1f4a0 2023-03-05)
```
Which is not good. We've effectively erased the toolchain selection, and users may reasonably conclude that their rustup install is broken.
After this change, we now get this:
```
$ cargo +miri miri --version
miri 0.1.0 (0ba1f4a0 2023-03-05)
$ cargo +nightly miri --version
miri 0.1.0 (f63ccaf 2023-03-06)
```
Thanks `@jyn514` who all but wrote this for me.
Update cargo
14 commits in 7d3033d2e59383fd76193daf9423c3d141972a7d..4a3c588b1f0a8e2dc8dd8789dbf3b6a71b02ed49
2023-03-08 17:05:08 +0000 to 2023-03-14 14:05:36 +0000
- ci: make clean-test-output a script for reuse (rust-lang/cargo#11848)
- Accurately show status when downgrading dependencies (rust-lang/cargo#11839)
- docs(contrib): Move Design Principles earlier in the book (rust-lang/cargo#11842)
- docs(contrib): Point compilation docs to doc comments (rust-lang/cargo#11841)
- `cargo install --git` multiple packages with binaries found hint (rust-lang/cargo#11835)
- Disable flaky auth tests when `gitoxide` runs them (rust-lang/cargo#11830)
- Add some documentation on writing cross-compilation tests (rust-lang/cargo#11825)
- chore: Use sparse protocol on stable CI (rust-lang/cargo#11829)
- Notice for potential unexpected shell expansions in help text of `cargo-add` (rust-lang/cargo#11826)
- Add tracking issue to gitoxide unstable docs (rust-lang/cargo#11822)
- Bump crates-io to 0.36.0 (rust-lang/cargo#11820)
- Bump to 0.71.0; update changelog (rust-lang/cargo#11815)
- docs(contrib): Move overview to lib (rust-lang/cargo#11809)
- Fix semver check for 1.68 (rust-lang/cargo#11817)
r? `@ghost`
Simplify proc macro signature validity check
Use an `ObligationCtxt` instead of `normalize_erasing_regions` + `DeepRejectCtxt`. This should both give us a more accurate error message, and also avoid issues like not-well-formed proc macro signatures. Also, let's fall back on the regular type mismatch error reporting for making these diagnostic notes, instead of hard-coding a bunch of specific diagnostics.
Fixes#109129
Use `unused_generic_params` from crate metadata
Due to the way that `separate_provide_extern` interacted with the implementation of `<ty::InstanceDef<'tcx> as Key>::query_crate_is_local`, we actually never hit the foreign provider for `unused_generic_params`.
Additionally, since the *local* provider of `unused_generic_params` calls `should_polymorphize`, which always returns false if the def-id is foreign, this means that we never actually polymorphize monomorphic instances originating from foreign crates.
We don't actually encode `unused_generic_params` for items where all generics are used, so I had to tweak the foreign provider to fall back to `ty::UnusedGenericParams::new_all_used()` to avoid more ICEs when the above bugs were fixed.
add `enable-warnings` flag for llvm, and disable it by default.
This flag allows to turn off warnings of llvm compilation for people who are not interested on those warnings.
fix: Do not retry inlay hint requests
Should close https://github.com/rust-lang/rust-analyzer/issues/13372, retrying the way its currently implemented is not ideal as we do not adjust offsets in the requests, but doing that is a major PITA, so this should at least work around one of the more annoying issues stemming from it.
Ensure `ptr::read` gets all the same LLVM `load` metadata that dereferencing does
I was looking into `array::IntoIter` optimization, and noticed that it wasn't annotating the loads with `noundef` for simple things like `array::IntoIter<i32, N>`. Trying to narrow it down, it seems that was because `MaybeUninit::assume_init_read` isn't marking the load as initialized (<https://rust.godbolt.org/z/Mxd8TPTnv>), which is unfortunate since that's basically its reason to exist.
The root cause is that `ptr::read` is currently implemented via the *untyped* `copy_nonoverlapping`, and thus the `load` doesn't get any type-aware metadata: no `noundef`, no `!range`. This PR solves that by lowering `ptr::read(p)` to `copy *p` in MIR, for which the backends already do the right thing.
Fortuitiously, this also improves the IR we give to LLVM for things like `mem::replace`, and fixes a couple of long-standing bugs where `ptr::read` on `Copy` types was worse than `*`ing them.
Zulip conversation: <https://rust-lang.zulipchat.com/#narrow/stream/219381-t-libs/topic/Move.20array.3A.3AIntoIter.20to.20ManuallyDrop/near/341189936>
cc `@erikdesjardins` `@JakobDegen` `@workingjubilee` `@the8472`
Fixes#106369Fixes#73258
