NVPTX: Avoid PassMode::Direct for args in C abi
Fixes#117480
I must admit that I'm confused about `PassMode` altogether, is there a good sum-up threads for this anywhere? I'm especially confused about how "indirect" and "byval" goes together. To me it seems like "indirect" basically means "use a indirection through a pointer", while "byval" basically means "do not use indirection through a pointer".
The return used to keep `PassMode::Direct` for small aggregates. It turns out that `make_indirect` messes up the tests and one way to fix it is to keep `PassMode::Direct` for all aggregates. I have mostly seen this PassMode mentioned for args. Is it also a problem for returns? When experimenting with `byval` as an alternative i ran into [this assert](61a3eea804/compiler/rustc_codegen_llvm/src/abi.rs (L463C22-L463C22))
I have added tests for the same kind of types that is already tested for the "ptx-kernel" abi. The tests cannot be enabled until something like #117458 is completed and merged.
CC: ``@RalfJung`` since you seem to be the expert on this and have already helped me out tremendously
CC: ``@RDambrosio016`` in case this influence your work on `rustc_codegen_nvvm`
``@rustbot`` label +O-NVPTX
Relax restrictions on multiple sanitizers
Most combinations of LLVM sanitizers are legal-enough to enable simultaneously. This change will allow simultaneously enabling ASAN and shadow call stacks on supported platforms.
I used this python script to generate the mutually-exclusive sanitizer combinations:
```python
#!/usr/bin/python3
import subprocess
flags = [
["-fsanitize=address"],
["-fsanitize=leak"],
["-fsanitize=memory"],
["-fsanitize=thread"],
["-fsanitize=hwaddress"],
["-fsanitize=cfi", "-flto", "-fvisibility=hidden"],
["-fsanitize=memtag", "--target=aarch64-linux-android", "-march=armv8a+memtag"],
["-fsanitize=shadow-call-stack"],
["-fsanitize=kcfi", "-flto", "-fvisibility=hidden"],
["-fsanitize=kernel-address"],
["-fsanitize=safe-stack"],
["-fsanitize=dataflow"],
]
for i in range(len(flags)):
for j in range(i):
command = ["clang++"] + flags[i] + flags[j] + ["-o", "main.o", "-c", "main.cpp"]
completed = subprocess.run(command, stderr=subprocess.DEVNULL)
if completed.returncode != 0:
first = flags[i][0][11:].replace('-', '').upper()
second = flags[j][0][11:].replace('-', '').upper()
print(f"(SanitizerSet::{first}, SanitizerSet::{second}),")
```
Refactor documentation for Apple targets
Refactor the documentation for Apple targets in `rustc`'s platform support page to make it clear what the supported OS version is and which environment variables are being read (`*_DEPLOYMENT_TARGET` and `SDKROOT`). This fixes https://github.com/rust-lang/rust/issues/124215.
Note that I've expanded the `aarch64-apple-ios-sim` maintainers `@badboy` and `@deg4uss3r` to include being maintainer of all `*-apple-ios-*` targets. If you do not wish to be so, please state that, then I'll explicitly note that in the docs.
Additionally, I've added myself as co-maintainer of most of these targets.
r? `@thomcc`
I think the documentation you've previously written on tvOS is great, have mostly modified it to have a more consistent formatting with the rest of the Apple target.
I recognize that there's quite a few changes here, feel free to ask about any of them!
---
CC `@simlay` `@Nilstrieb`
`@rustbot` label O-apple
the `rust.lld` config enables rustc's `CFG_USE_SELF_CONTAINED_LINKER` env var, and we:
- set the linker-flavor to use lld
- enable the self-contained linker
this makes the target use the rust-lld linker by default
Most combinations of LLVM sanitizers are legal-enough to enable
simultaneously. This change will allow simultaneously enabling ASAN and
shadow call stacks on supported platforms.
Refactor Apple `target_abi`
This was bundled together with `Arch`, which complicated a few code paths and meant we had to do more string matching than necessary.
CC `@BlackHoleFox` as you've worked on the Apple target spec before
Related: Is there a reason why `Target`/`TargetOptions` use `StaticCow` for so many things, instead of an enum with defined values (and perhaps a catch-all case for custom target json files)? Tagging `@Nilstrieb,` as you might know?
Add `-lmingwex` second time in `mingw_libs`
Upcoming mingw-w64 releases will contain small math functions refactor which moved implementation around. As a result functions like `lgamma`
now depend on libraries in this order:
`libmingwex.a` -> `libmsvcrt.a` -> `libmingwex.a`.
Fixes#124221
Refactor float `Primitive`s to a separate `Float` type
Now there are 4 of them, it makes sense to refactor `F16`, `F32`, `F64` and `F128` out of `Primitive` and into a separate `Float` type (like integers already are). This allows patterns like `F16 | F32 | F64 | F128` to be simplified into `Float(_)`, and is consistent with `ty::FloatTy`.
As a side effect, this PR also makes the `Ty::primitive_size` method work with `f16` and `f128`.
Tracking issue: #116909
`@rustbot` label +F-f16_and_f128
rustc: Some small changes for the wasm32-wasip2 target
This commit has a few changes for the wasm32-wasip2 target. The first two are aimed at improving the compatibility of using `clang` as an external linker driver on this target. The default target to LLVM is updated to match the Rust target and additionally the `-fuse-ld=lld` argument is dropped since that otherwise interferes with clang's own linker detection. The only linker on wasm targets is LLD but on the wasip2 target a wrapper around LLD, `wasm-component-ld`, is used to drive the process and perform steps necessary for componentization.
The final commit changes the output of all objects on the wasip2 target to being PIC by default. This improves compatibilty with shared libaries but notably does not mean that there's a turnkey solution for shared libraries. The hope is that by having the standard libray work both with and without dynamic libraries will make experimentation easier.
This commit changes the new `wasm32-wasip2` target to being PIC by
default rather than the previous non-PIC by default. This change is
intended to make it easier for the standard library to be used in a
shared object in its precompiled form. This comes with a hypothetical
modest slowdown but it's expected that this is quite minor in most use
cases or otherwise wasm compilers and/or optimizing runtimes can elide
the cost.
LLVM has updated data layouts to specify `Fn32` on 64-bit ARM to avoid
C++ accidentally underaligning functions when trying to comply with
member function ABIs.
This should only affect Rust in cases where we had a similar bug (I
don't believe we have one), but our data layout must match to generate
code.
As a compatibility adaptatation, if LLVM is not version 19 yet, `Fn32`
gets voided from the data layout.
See llvm/llvm-project#90415
- Fixed std support in top-level docs.
- Added `*-apple-darwin` docs.
- Added `i686-apple-darwin` docs.
- Moved `aarch64-apple-ios-sim` to `*-apple-ios` and document all the
iOS targets there.
- Added `*-apple-ios-macabi` docs.
- Add myself (madsmtm) as co-maintainer of most of these targets.
Set non-leaf frame pointers on Fuchsia targets
This is part of our work to enable shadow call stack sanitization on Fuchsia, see [this Fuchsia issue](https://g-issues.fuchsia.dev/issues/327643884).
r? ``@tmandry``
Change `SIGPIPE` ui from `#[unix_sigpipe = "..."]` to `-Zon-broken-pipe=...`
In the stabilization [attempt](https://github.com/rust-lang/rust/pull/120832) of `#[unix_sigpipe = "sig_dfl"]`, a concern was [raised ](https://github.com/rust-lang/rust/pull/120832#issuecomment-2007394609) related to using a language attribute for the feature: Long term, we want `fn lang_start()` to be definable by any crate, not just libstd. Having a special language attribute in that case becomes awkward.
So as a first step towards the next stabilization attempt, this PR changes the `#[unix_sigpipe = "..."]` attribute to a compiler flag `-Zon-broken-pipe=...` to remove that concern, since now the language is not "contaminated" by this feature.
Another point was [also raised](https://github.com/rust-lang/rust/pull/120832#issuecomment-1987023484), namely that the ui should not leak **how** it does things, but rather what the **end effect** is. The new flag uses the proposed naming. This is of course something that can be iterated on further before stabilization.
Tracking issue: https://github.com/rust-lang/rust/issues/97889
This commit changes the LLVM target of for the Rust `wasm32-wasip2`
target to `wasm32-wasip2` as well. LLVM does a bit of detection on the
target string to know when to call `wasm-component-ld` vs `wasm-ld` so
otherwise clang is invoking the wrong linker.
In the stabilization attempt of `#[unix_sigpipe = "sig_dfl"]`, a concern
was raised related to using a language attribute for the feature: Long
term, we want `fn lang_start()` to be definable by any crate, not just
libstd. Having a special language attribute in that case becomes
awkward.
So as a first step towards towards the next stabilization attempt, this
PR changes the `#[unix_sigpipe = "..."]` attribute to a compiler flag
`-Zon-broken-pipe=...` to remove that concern, since now the language
is not "contaminated" by this feature.
Another point was also raised, namely that the ui should not leak
**how** it does things, but rather what the **end effect** is. The new
flag uses the proposed naming. This is of course something that can be
iterated on further before stabilization.
Upcoming mingw-w64 releases will contain small math functions refactor which moved implementation around.
As a result functions like `lgamma`
now depend on libraries in this order:
`libmingwex.a` -> `libmsvcrt.a` -> `libmingwex.a`.
Fixes#124221
Introduce perma-unstable `wasm-c-abi` flag
Now that `wasm-bindgen` v0.2.88 supports the spec-compliant C ABI, the idea is to switch to that in a future version of Rust. In the meantime it would be good to let people test and play around with it.
This PR introduces a new perma-unstable `-Zwasm-c-abi` compiler flag, which switches to the new spec-compliant C ABI when targeting `wasm32-unknown-unknown`.
Alternatively, we could also stabilize this and then deprecate it when we switch. I will leave this to the Rust maintainers to decide.
This is a companion PR to #117918, but they could be merged independently.
MCP: https://github.com/rust-lang/compiler-team/issues/703
Tracking issue: https://github.com/rust-lang/rust/issues/122532
To the best of my ability I believe that this is no longer necessary. I
don't fully recall why this was first added but I believe it had to do
with symbols all being exported by default and this was required to undo
that. Regardless nowadays the default output of rustc seems suitable so
it seems best to keep wasm in line with other targets.
Linker flavors next steps: linker features
This is my understanding of the first step towards `@petrochenkov's` vision for the future of linker flavors, described in https://github.com/rust-lang/rust/pull/119906#issuecomment-1895693162 and the discussion that followed.
To summarize: having `Cc` and `Lld` embedded in linker flavors creates tension about naming, and a combinatorial explosion of flavors for each new linker feature we'd want to use. Linker features are an extension mechanism that is complementary to principal flavors, with benefits described in #119906.
The most immediate use of this flag would be to turn self-contained linking on and off via features instead of flavors. For example, `-Clinker-features=+/-lld` would toggle using lld instead of selecting a precise flavor, and would be "generic" and work cross-platform (whereas linker flavors are currently more tied to targets). Under this scheme, MCP510 is expected to be `-Clink-self-contained=+linker -Zlinker-features=+lld -Zunstable-options` (though for the time being, the original flags using lld-cc flavors still work).
I purposefully didn't add or document CLI support for `+/-cc`, as it would be a noop right now. I only expect that we'd initially want to stabilize `+/-lld` to begin with.
r? `@petrochenkov`
You had requested that minimal churn would be done to the 230 target specs and this does none yet: the linker features are inferred from the flavor since they're currently isomorphic. We of course expect this to change sooner rather than later.
In the future, we can allow targets to define linker features independently from their flavor, and remove the cc and lld components from the flavors to use the features instead, this actually doesn't need to block stabilization, as we discussed.
(Best reviewed per commit)
Re-enable `has_thread_local` for i686-msvc
A few years back, `has_thread_local` was disabled as a workaround for a compiler issue. While the exact cause was never tracked down, it was suspected to be caused by the compiler inlining a thread local access across a dylib boundary. This should be fixed now so let's try again.
While they're isomorphic, we can flip the lld component where
applicable, so that downstream doesn't have to check both the flavor and
the linker features.
They are a flexible complementary mechanism to linker flavors,
that also avoid the combinatorial explosion of mapping linking features
to actual linker flavors.
Reduce Size of `ModifierInfo`
I added `ModifierInfo` in #121940 and had used a `u64` for the `size` field even though the largest value it holds is `512`.
This PR changes the type of the `size` field to `u16`.
The actual ABI implication here is that in some cases the values
are required to be "consecutive", i.e. must either all be passed
in registers or all on stack (without padding).
Adjust the code to either use Uniform::new() or Uniform::consecutive()
depending on which behavior is needed.
Then, when lowering this in LLVM, skip the [1 x i128] to i128
simplification if is_consecutive is set. i128 is the only case
I'm aware of where this is problematic right now. If we find
other cases, we can extend this (either based on target information
or possibly just by not simplifying for is_consecutive entirely).
When passing a 16 (or higher) aligned struct by value on ppc64le,
it needs to be passed as an array of `i128` rather than an array
of `i64`. This will force the use of an even starting register.
For the case of a 16 byte struct with alignment 16 it is important
that `[1 x i128]` is used instead of `i128` -- apparently, the
latter will get treated similarly to `[2 x i64]`, not exhibiting
the correct ABI. Add a `force_array` flag to `Uniform` to support
this.
The relevant clang code can be found here:
fe2119a7b0/clang/lib/CodeGen/Targets/PPC.cpp (L878-L884)fe2119a7b0/clang/lib/CodeGen/Targets/PPC.cpp (L780-L784)
I think the corresponding psABI wording is this:
> Fixed size aggregates and unions passed by value are mapped to as
> many doublewords of the parameter save area as the value uses in
> memory. Aggregrates and unions are aligned according to their
> alignment requirements. This may result in doublewords being
> skipped for alignment.
In particular the last sentence.
Fixes https://github.com/rust-lang/rust/issues/122767.
Fix incorrect 'llvm_target' value used on watchOS target
## Issue
`xcodebuild -create-xcframework` command doesn't recognize static libraries that are built on "arm64_32-apple-watchos" target.
Here are steps to reproduce the issue on a Mac:
1. Install nightly toolchain `nightly-2024-03-27`. Needs this specific version, because newer nightly versions are broken on watchos target.
1. Create an empty library: `mkdir watchos-lib && cd watchos-lib && cargo init --lib`.
1. Add configuration `lib.crate-type=["staticlib"]` to Cargo.toml.
1. Build the library: `cargo +nightly-2024-03-27 build --release -Zbuild-std --target arm64_32-apple-watchos`
1. Run `xcodebuild -create-xcframework` to put the static library into a xcframework, which results in an error:
```
$ xcodebuild -create-xcframework -library target/arm64_32-apple-watchos/release/libwatchos_lib.a -output test.xcframework
error: unable to determine the platform for the given binary '.../watchos-lib/target/arm64_32-apple-watchos/release/libwatchos_lib.a'; check your deployment version settings
```
## Fix
The root cause of this error is `xcodebuild` couldn't read `LC_BUILD_VERSION` from the static library to determine the library's target platform. And the reason it's missing is that an incorrect `llvm_target` value is used in `arm64_32-apple-watchos` target. The expected value is `<arch>-apple-watchos<major>.<minor>.0`, i.e. "arm64_32-apple-watchos8.0.0".
The [.../apple/mod.rs](43f4f2a3b1/compiler/rustc_target/src/spec/base/apple/mod.rs (L321)) file contains functions that construct such string. There is an existing function `watchos_sim_llvm_target` which returns llvm target value for watchOS simulator. But there is none for watchOS device. This PR adds that missing function to align watchOS with other Apple platform targets.
To verify the fix, you can simply build a toolchain on this PR branch and repeat the steps above using the built local toolchain to verify the `xcodebuild -create-xcframework` command can create a xcframework successfully.
Furthermore, you can verify `LC_BUILD_VERSION` contains correct info by using the simple shell script below to print `LC_BUILD_VERSION` of the static library that's built on watchos target:
```shell
bin=target/arm64_32-apple-watchos/release/libwatchos_lib.a
file=$(ar -t "$bin" | grep -E '\.o$' | head -n 1)
ar -x "$bin" "$file"
vtool -show-build-version "$file"
```
Here is an example output from my machine:
```
watchos_rust-495d6aaf3bccc08d.watchos_rust.35ba42bf9255ca9d-cgu.0.rcgu.o:
Load command 1
cmd LC_BUILD_VERSION
cmdsize 24
platform WATCHOS
minos 8.0
sdk n/a
ntools 0
```
