Add amdgpu target
Add amdgpu target to rustc and enable the LLVM target.
Fix compiling `core` with the amdgpu:
The amdgpu backend makes heavy use of different address spaces. This
leads to situations, where a pointer in one addrspace needs to be casted
to a pointer in a different addrspace. `bitcast` is invalid for this
case, `addrspacecast` needs to be used.
Fix compilation failures that created bitcasts for such cases by
creating pointer casts (which creates an `addrspacecast` under the hood)
instead.
MCP: https://github.com/rust-lang/compiler-team/issues/823
Tracking issue: #135024
Kinda related to the original amdgpu tracking issue #51575 (though that one has been closed for a while).
Rollup of 7 pull requests
Successful merges:
- #135488 (Stabilize `vec_pop_if`)
- #136068 (crashes: more tests)
- #136694 (Update minifier version to `0.3.4`)
- #136722 (Visit all debug info in MIR Visitor)
- #136746 (Emit an error if `-Zdwarf-version=1` is requested)
- #136760 (Fix unwrap error in overflowing int literal)
- #136782 (Fix mistake in x86_64-unknown-freebsd platform description)
r? `@ghost`
`@rustbot` modify labels: rollup
Visit all debug info in MIR Visitor
I've been experimenting with simplifying debug info in MIR inliner, and discovered that MIR Visitor doesn't reliably visit all spans. This PR adds the missing visitor calls.
Update minifier version to `0.3.4`
It fixes a bug where a whitespace would get removed in `a [attribute]` (you're not forced to add a tag before an attribute selector).
r? ````@notriddle````
DWARF 1 is very different than DWARF 2+ (see the commentary in
https://gcc.gnu.org/onlinedocs/gcc/Debugging-Options.html#index-gdwarf)
and LLVM does not really seem to support DWARF 1 as Clang does not offer
a `-gdwarf-1` flag and `llc` will just generate DWARF 2 with the version
set to 1: https://godbolt.org/z/s85d87n3a.
Since this isn't actually supported (and it's not clear it would be
useful anyway), report that DWARF 1 is not supported if it is requested.
Also add a help message to the error saying which versions are supported.
Update bootstrap compiler and rustfmt
The rustfmt version we previously used formats things differently from what the latest nightly rustfmt does. This causes issues for subtrees that get formatted both in-tree and in their own repo. Updating the rustfmt used in-tree solves those issues. Also bumped the bootstrap compiler as the stage0 update command always updates both at the same
time.
Rollup of 5 pull requests
Successful merges:
- #134679 (Windows: remove readonly files)
- #136213 (Allow Rust to use a number of libc filesystem calls)
- #136530 (Implement `x perf` directly in bootstrap)
- #136601 (Detect (non-raw) borrows of null ZST pointers in CheckNull)
- #136659 (Pick the max DWARF version when LTO'ing modules with different versions )
r? `@ghost`
`@rustbot` modify labels: rollup
Small resolve refactor
I was looking into how resolve works in order to find a good way for clippy to shorten paths in messages and suggestions, and found a needless `.collect()` and a recursive function that could be written as a loop, also removed a panicky code path.
transmutability: fix ICE when passing wrong ADT to ASSUME
- Remove an incorrect assert that the `ASSUME` parameter has the type `Assume` and delay a bug instead.
- Since we checked the type of `ASSUME` is `Assume` (an ADT), its valtree must be a branch, so we can just unwrap it.
r? ```@jswrenn```
Make `AsyncFnOnce`, `AsyncFnMut`, `AsyncFn` non-`#[fundamental]`
Address the issue #136723 on nightly (the issue will only *actually* be fixed with a beta backport).
Document `Sum::sum` returns additive identities for `[]`
Because the neutral element of `<fNN as iter::Sum>` was changed to `neg_zero`, the documentation needed to be updated, as it was reporting inadequate information about what should be expected from the return.
Relevant Commit: 4908188518
Relevant Pull Request: https://github.com/rust-lang/rust/pull/129321
---
The referenced commit causes unintended side effects on presentation layer applications like using Tera templates, for example. I'm not sure what the motivation was behind the original change, but it seems like more discussion should be put into this issue and potentially have that change reverted.
compiler: mostly-finish `rustc_abi` updates
This almost-finishes all the updates in the compiler to use `rustc_abi` and removes some of the reexports of `rustc_abi` items in `rustc_target` that were previously available.
r? ```@compiler-errors```
Clean up `HashMap` and `HashSet` docs.
This commit makes some small, pedantic changes to the docs for `HashMap` and `HashSet`, which fixes that:
* "HashMap" is not always formatted as code (as in `HashMap`), and that
* `HashSet` sometimes references `HashMap` instead of itself.
Add a comment pointing to ICE-136223
Fixes#136223
## Steps how the ICE happen
This explanation is based on the test case `&Some(Some(x)) = &Some(&mut Some(0))`.
The case should fail with E0596 error, but it catches the debug assertion instead.
1. For the first `&`: In check_pat_ref(), the value max_ref_mutbl becomes MutblCap::Not ([here](fdd1a3b026/compiler/rustc_hir_typeck/src/pat.rs (L2394-L2396))). Once max_ref_mutbl becomes Not, it will never be back to MutblCap::Mut.
2. For `&mut`: In peel_off_references(), because Some(x) doesn't have `&` nor `&mut`, `&mut` in `&mut Some(0)` is not consumed then default_binding_mode (def_br) becomes `ByRef::Yes(Mutability::Mut)` (around [here](fdd1a3b026/compiler/rustc_hir_typeck/src/pat.rs (L519-L536))). This will be inherited to the next step. So this pattern has the mismatch between `def_br=Yes(Mut)` and `max_ref_mutbl=Not` now.
3. For the value `0`: Because of the step 2, the default_binding_mode is `Yes(Mut)`, but max_ref_mutbl is `Not` from the step 1. It causes the assertion error [here](fdd1a3b026/compiler/rustc_hir_typeck/src/pat.rs (L427-L430)).
## What this PR fixes
Step 1 has happened from [this commit](e2f3ce9568) by deleting `no_ref_mut_behind_and` from the if block. In my understanding, after RFC3627 is released, step 1 should happen not only 2024 edition but also other editions to track MutblCap value. But for now, it should not happen for non-2024 edition. So I put it back.
NOTE: I think there is another solution - We should return an E0596 error in calc_default_binding_mode() instead of the debug assertion. Since the assertion is caused by the mismatch between `def_br = Yes(Mut)` and `max_ref_mutbl = Not`, but in my understanding this violation is the same as E0596. check_pat_ident() does returns E0596 by a similar reason [here](fdd1a3b026/compiler/rustc_hir_typeck/src/pat.rs (L837-L856)).
Pick the max DWARF version when LTO'ing modules with different versions
Currently, when rustc compiles code with `-Clto` enabled that was built
with different choices for `-Zdwarf-version`, a warning will be
reported. It's very easy to observe this by compiling most anything (eg,
"hello world") and specifying `-Clto -Zdwarf-version=5` since the
standard library is distributed with `-Zdwarf-version=4`.
This behavior isn't actually useful for a few reasons:
- From observation, LLVM chooses to pick the highest DWARF version
anyway after issuing the warning.
- Clang specifies that in this case, the max version should be picked
without a warning and as a general principle, we want to support
x-lang LTO with Clang which implies using the same module flag merge
behaviors.
- Debuggers need to be able to handle a variety of versions within the
same debugging session as you can easily have some parts of a binary
(or some dynamic libraries within an application) all compiled with
different DWARF versions.
This commit changes the module flag merge behavior to match Clang and
use the highest version of DWARF. It also adds a test to ensure this
behavior is respected in the case of two crates being LTO'd together and
adds a test to ensure no warning is printed.
Fixes#130041 which fails due to these warnings being printed
cc #103057
Detect (non-raw) borrows of null ZST pointers in CheckNull
Fixes#136568. Ensures that we check that borrows of derefs are non-null in the `CheckNull` pass **even if** it's a ZST pointee.
I'm actually surprised that this is UB in Miri, but if it's certainly UB, then this PR modifies the null check to be stricter. I couldn't find anywhere in https://doc.rust-lang.org/reference/behavior-considered-undefined.html that discusses this case specifically, but I didn't read it too closely, or perhaps it's just missing a bullet point.
On the contrary, if this is actually erroneous UB in Miri, then I'm happy to close this (and perhaps fix the null check in Miri to exclude ZSTs?)
On the double contrary, if this is still an "open question", I'm also happy to close this and wait for a decision to be made.
r? ``@saethlin`` cc ``@RalfJung`` (perhaps you feel strongly about this change)
Allow Rust to use a number of libc filesystem calls
This allows Rust on Fuchsia to use a number of function calls from libc:
* dirfd
* fdatasync
* flock with LOCK_EX, LOCK_SH, LOCK_NB, LOCK_UN
* fstatat
cc #120426
try-job: dist-various-2
Windows: remove readonly files
When calling `remove_file`, we shouldn't fail to delete readonly files. As the test makes clear, this make the Windows behaviour consistent with other platforms. This also makes us internally consistent with `remove_dir_all`.
try-job: x86_64-msvc-ext1
The rustfmt version we previously used formats things differently from
what the latest nightly rustfmt does. This causes issues for subtrees
that get formatted both in-tree and in their own repo. Updating the
rustfmt used in-tree solves those issues. Also bumped the bootstrap
compiler as the stage0 update command always updates both at the same
time.
Because the neutral element of `<fNN as iter::Sum>` was changed to
`neg_zero`, the documentation needed to be updated, as it was reporting
inadequate information about what should be expected from the return.
Co-authored-by: Jubilee <workingjubilee@gmail.com>
ci: Use largedisk for loongarch
`dist-loongarch64-linux` and `dist-loongarch64-musl` seem to run out of disk space when adding the amdgpu LLVM backend. There are changes underway to free more disk space, but that may take a while, so it was suggested to switch the affected jobs to a largedisk runner until that is ready.
`dist-powerpc64-linux` is another job that ran out of disk space one time, but it succeeded the other times, so I did not include it here.
amdgpu target PR with more references to the failures: #134740
r? ```@marcoieni```
Make empty-line-after an early clippy lint
r? ```@y21```
95% a refiling of https://github.com/rust-lang/rust-clippy/pull/13658 but for correctness it needed 2 extra methods in `rust_lint` which made it much easier to apply on `rust-lang/rust` than `rust-lang/rust-clippy`.
Commits have been thoroughly reviewed on `rust-lang/clippy already`. The last two review comments there (about using `Option` and popping for assoc items have been applied here.
Generate correct terminate block under Wasm EH
This fixes failing LLVM assertions during insnsel.
Improves #135665.
r? bjorn3
^ you reviewed the PR bringing Wasm EH in, I assume this is within your area of expertise?
Optimize `Rc::<str>::default()` implementation
This PR lets `impl Default for Rc<str>` re-use the implementation for `Rc::<[u8]>::default()`. The previous version only calculted the memory layout at runtime, even though it should be known at compile time, resulting in an additional function call.
The same optimization is done for `Rc<CStr>`.
Generated byte code: <https://godbolt.org/z/dfq73jsoP>.
Resolves <https://github.com/rust-lang/rust/issues/135784>.
Cc `@Billy-Sheppard.`
std: move `io` module out of `pal`, get rid of `sys_common::io`
Part of #117276.
This does two related things:
1. It moves the platform-specific definitions for `IoSlice`, `IoSliceMut` and `is_terminal` out of `pal` and into `sys` and unifies some of them.
2. It gets rid of `sys_common::io`, moving the non-platform-specific test helpers into `std::test_helpers` and the buffer size definition to the new `sys::io` module.
Subtree sync for rustc_codegen_cranelift
The main highlights this time are a Cranelift update and a fix for a warning that the x87 feature is not enabled.
r? `@ghost`
`@rustbot` label +A-codegen +A-cranelift +T-compiler<!-- homu-ignore:start -->
Currently, when rustc compiles code with `-Clto` enabled that was built
with different choices for `-Zdwarf-version`, a warning will be
reported. It's very easy to observe this by compiling most anything (eg,
"hello world") and specifying `-Clto -Zdwarf-version=5` since the
standard library is distributed with `-Zdwarf-version=4`.
This behavior isn't actually useful for a few reasons:
- from observation, LLVM chooses to pick the highest DWARF version
anyway after issuing the warning
- Clang specifies that in this case, the max version should be picked
without a warning and as a general principle, we want to support
x-lang LTO with Clang which implies using the same module flag merge
behaviors
- Debuggers need to be able to handle a variety of versions withing the
same debugging session as you can easily have some parts of a binary
(or some dynamic libraries within an application) all compiled with
different DWARF versions
This commit changes the module flag merge behavior to match Clang and
use the highest version of DWARF. It also adds a test to ensure this
behavior is respected in the case of two crates being LTO'd together and
updates the test added in the previous commit to ensure no warning is
printed.
