Various refactors to the LTO handling code (part 2)
Continuing from https://github.com/rust-lang/rust/pull/143388 this removes a bit of dead code and moves the LTO symbol export calculation from individual backends to cg_ssa.
For example, transmuting between `bool` and `Ordering` doesn't need two `assume`s because one range is a superset of the other.
Multiple are still used for things like `char` <-> `NonZero<u32>`, which overlap but where neither fully contains the other.
When combined with 143720, this means `rvalue_creates_operand` can just return `true` for *every* `Rvalue`. (A future PR could consider removing it, though just letting it optimize out is fine for now.)
It's nicer anyway, IMHO, because it avoids needing the layout checks to be consistent in the two places, and thus is an overall reduction in code. Plus it's a more helpful building block when used in other places this way.
In optimized builds GVN gets rid of these already, but in `opt-level=0` we actually make `alloca`s for this, which particularly impacts `?`-style things that use actually-only-one-variant types like this.
Certain LLVM intrinsics, such as `llvm.wasm.throw`, can unwind. Marking
them as nounwind causes us to skip cleanup of locals and optimize out
`catch_unwind` under inlining or when `llvm.wasm.throw` is used directly
by user code.
The motivation for forcibly marking llvm.* as nounwind is no longer
present: most intrinsics are linked as `extern "C"` or other
non-unwinding ABIs, so we won't codegen `invoke` for them anyway.
gpu offload host code generation
r? ghost
This will generate most of the host side code to use llvm's offload feature.
The first PR will only handle automatic mem-transfers to and from the device.
So if a user calls a kernel, we will copy inputs back and forth, but we won't do the actual kernel launch.
Before merging, we will use LLVM's Info infrastructure to verify that the memcopies match what openmp offloa generates in C++. `LIBOMPTARGET_INFO=-1 ./my_rust_binary` should print that a memcpy to and later from the device is happening.
A follow-up PR will generate the actual device-side kernel which will then do computations on the GPU.
A third PR will implement manual host2device and device2host functionality, but the goal is to minimize cases where a user has to overwrite our default handling due to performance issues.
I'm trying to get a full MVP out first, so this just recognizes GPU functions based on magic names. The final frontend will obviously move this over to use proper macros, like I'm already doing it for the autodiff work.
This work will also be compatible with std::autodiff, so one can differentiate GPU kernels.
Tracking:
- https://github.com/rust-lang/rust/issues/131513
Fix `-Ctarget-feature`s getting ignored after `crt-static`
The current behaviour introduced by commit a50a3b8e31 would discard any target features specified after `crt-static` (the only member of `RUSTC_SPECIFIC_FEATURES`). This is because it returned instead of continuing processing the next feature.
I wasn't entirely sure how the regression test should look like, but this one should do. If anyone has some suggestions, I'm happy to learn, it's my first test :)
I've confirmed that the test fails without the fix on `powerpc64le-unknown-linux-musl` and `x86_64-unknown-linux-gnu`.
cc ``@RalfJung``
The conversation in 143502 made be realize how easy this is to handle, since the only possibilty is ZSTs -- everything else ends up with the destination being `LocalKind::Memory` and thus doesn't call `codegen_rvalue_operand` at all.
This gets us perilously close to a world where `rvalue_creates_operand` only ever returns true. I'll try out such a world next :)
Simplify discriminant codegen for niche-encoded variants which don't wrap across an integer boundary
Inspired by rust-lang/rust#139729, this attempts to be a much-simpler and more-localized change while still making a difference. (Specifically, this does not try to solve the problem with select-sinking, leaving that to be fixed by https://github.com/llvm/llvm-project/issues/134024 -- once it gets released -- instead of in rustc's codegen.)
What this *does* improve is checking for the variant in a 3+ variant enum when that variant is the type providing the niche. Something like `if let Foo::WithBool(_) = ...` previously compiled to `ugt(add(x, -2), 2)`, which is non-trivial to think about because it's depending on the unsigned wrapping to shift the 0/1 up above 2. With this PR it compiles to just `ult(x, 2)`, which is probably what you'd have written yourself if you were doing it by hand to look for "is this byte a bool?".
That's done by leaving most of the codegen alone, but adding a couple new special cases to the `is_niche` check. The default looks at the relative discriminant, but in the common cases where there's no wraparound involved, we can just check the original value, rather than the offsetted one.
The first commit just adds some tests, so the best way to see the effect of this change is to look at the second commit and how it updates the test expectations.
The current behaviour introduced by commit
a50a3b8e31 would discard any
target features specified after crt-static (the only member of
RUSTC_SPECIFIC_FEATURES). This is because it returned instead of
continuing processing the next flag.
Signed-off-by: Jens Reidel <adrian@travitia.xyz>
pass --gc-sections if -Zexport-executable-symbols is enabled and improve tests
Exported symbols are added as GC roots in linking, so `--gc-sections` won't hurt `-Zexport-executable-symbols`.
Fixes the run-make test to work on Linux. Enable the ui test on more targets.
cc rust-lang/rust#84161
fix `-Zsanitizer=kcfi` on `#[naked]` functions
fixes https://github.com/rust-lang/rust/issues/143266
With `-Zsanitizer=kcfi`, indirect calls happen via generated intermediate shim that forwards the call. The generated shim preserves the attributes of the original, including `#[unsafe(naked)]`. The shim is not a naked function though, and violates its invariants (like having a body that consists of a single `naked_asm!` call).
My fix here is to match on the `InstanceKind`, and only use `codegen_naked_asm` when the instance is not a `ReifyShim`. That does beg the question whether there are other `InstanceKind`s that could come up. As far as I can tell the answer is no: calling via `dyn` seems to work find, and `#[track_caller]` is disallowed in combination with `#[naked]`.
r? codegen
````@rustbot```` label +A-naked
cc ````@maurer```` ````@rcvalle````
Various refactors to the LTO handling code
In particular reducing the sharing of code paths between fat and thin-LTO and making the fat LTO implementation more self-contained. This also moves some autodiff handling out of cg_ssa into cg_llvm given that Enzyme only works with LLVM anyway and an implementation for another backend may do things entirely differently. This will also make it a bit easier to split LTO handling out of the coordinator thread main loop into a separate loop, which should reduce the complexity of the coordinator thread.
Rollup of 9 pull requests
Successful merges:
- rust-lang/rust#143403 (Port several trait/coherence-related attributes the new attribute system)
- rust-lang/rust#143633 (fix: correct assertion to check for 'noinline' attribute presence before removal)
- rust-lang/rust#143647 (Clarify and expand documentation for std::sys_common dependency structure)
- rust-lang/rust#143716 (compiler: doc/comment some codegen-for-functions interfaces)
- rust-lang/rust#143747 (Add target maintainer information for aarch64-unknown-linux-musl)
- rust-lang/rust#143759 (Fix typos in function names in the `target_feature` test)
- rust-lang/rust#143767 (Bump `src/tools/x` to Edition 2024 and some cleanups)
- rust-lang/rust#143769 (Remove support for SwitchInt edge effects in backward dataflow)
- rust-lang/rust#143770 (build-helper: clippy fixes)
r? `@ghost`
`@rustbot` modify labels: rollup
Make UB transmutes really UB in LLVM
Ralf suggested in <https://github.com/rust-lang/rust/pull/143410#discussion_r2184928123> that UB transmutes shouldn't be trapping, which happened for the one path *that* PR was changing, but there's another path as well, so *this* PR changes that other path to match.
r? codegen
use `--dynamic-list` for exporting executable symbols
closesrust-lang/rust#101610
cc rust-lang/rust#84161https://sourceware.org/binutils/docs-2.39/ld/VERSION.html:
> --dynamic-list=dynamic-list-file
Specify the name of a dynamic list file to the linker. This is typically used when creating shared libraries to specify a list of global symbols whose references shouldn’t be bound to the definition within the shared library, or creating dynamically linked executables to specify a list of symbols which should be added to the symbol table in the executable. This option is only meaningful on ELF platforms which support shared libraries.
`ld.lld --help`:
> --dynamic-list=<file>: Similar to --export-dynamic-symbol-list. When creating a shared object, this additionally implies -Bsymbolic but does not set DF_SYMBOLIC
> --export-dynamic-symbol-list=file: Read a list of dynamic symbol patterns. Apply --export-dynamic-symbol on each pattern
> --export-dynamic-symbol=glob: (executable) Put matched symbols in the dynamic symbol table. (shared object) References to matched non-local STV_DEFAULT symbols shouldn't be bound to definitions within the shared object. Does not imply -Bsymbolic.
> --export-dynamic: Put symbols in the dynamic symbol table
Use `--dynamic-list` because it's older than `--export-dynamic-symbol-list` (binutils 2.35)
try-job: dist-i586-gnu-i586-i686-musl
Let `rvalue_creates_operand` return true for *all* `Rvalue::Aggregate`s
~~Draft for now because it's built on Ralf's rust-lang/rust#143291~~
Inspired by https://github.com/rust-lang/rust/pull/138759#discussion_r2156375342 where I noticed that we were nearly at this point, plus the comments I was writing in rust-lang/rust#143410 that reminded me a type-dependent `true` is fine.
This PR splits the `OperandRef::builder` logic out to a separate type, with the updates needed to handle SIMD as well. In doing so, that makes the existing `Aggregate` path in `codegen_rvalue_operand` capable of handing SIMD values just fine.
As a result, we no longer need to do layout calculations for aggregate result types when running the analysis to determine which things can be SSA in codegen.
