Tracking Issue for nvptx_target_feature

[jedbrown]

This is a tracking issue for the NVPTX part of #44839 (RFC 2045; rust-lang/rfcs#2045).
The feature gate for the issue is #![feature(nvptx_target_feature)].

About tracking issues

Tracking issues are used to record the overall progress of implementation.
They are also used as hubs connecting to other relevant issues, e.g., bugs or open design questions.
A tracking issue is however not meant for large scale discussion, questions, or bug reports about a feature.
Instead, open a dedicated issue for the specific matter and add the relevant feature gate label.
Discussion comments will get marked as off-topic or deleted.
Repeated discussions on the tracking issue may lead to the tracking issue getting locked.

Steps

• Implementation add nvptx_target_feature #138689
• Stabilization PR (see instructions on rustc-dev-guide)

Unresolved Questions

• Should ptx* and sm_* be target features or should they be target modifiers; cf. https://internals.rust-lang.org/t/why-are-forbidden-target-features-disallowed-in-cfg-target-feature/22944/4.
  • They are exposed as LLVM features, but LLVM does not currently intend these features to be applied as function attributes. As such, we'd prefer that it be erroneous to use #[target_feature(enable = "ptx80")].
  • This relates to how core::arch::nvptx is compiled. If we use #[cfg(target_feature = "ptx80")] for relevant intrinsics, then the user must build their crates using -Zbuild-std=core with appropriate RUSTFLAGS.
• When target features are activated using -Ctarget-feature=+ptx80 or #[target_feature(enable = "ptx80")], all the implied features are also added to LLVM functions. LLVM doesn't officially support that (though it seems to work with LLVM-20), so it would be preferable to only pass the highest level LLVM feature and ensure that is applied at crate/PTX module granularity.

Implementation history

• add nvptx_target_feature #138689

@rustbot label +O-NVPTX +A-target-feature