c_variadic: provide our own va_arg implementation for more targets
#150094
Conversation
rustbot
added
A-LLVM
|
|
| Arch::LoongArch32 => emit_ptr_va_arg( | ||
| bx, | ||
| addr, | ||
| target_ty, | ||
| if target_ty_size > 2 * 4 { PassMode::Indirect } else { PassMode::Direct }, | ||
| SlotSize::Bytes4, | ||
| AllowHigherAlign::Yes, | ||
| ForceRightAdjust::No, | ||
| ), | ||
| Arch::LoongArch64 => emit_ptr_va_arg( | ||
| bx, | ||
| addr, | ||
| target_ty, | ||
| if target_ty_size > 2 * 8 { PassMode::Indirect } else { PassMode::Direct }, | ||
| SlotSize::Bytes8, | ||
| AllowHigherAlign::Yes, | ||
| ForceRightAdjust::No, | ||
| ), |
There was a problem hiding this comment.
zero-sized types can't reach this logic currently, but we could handle them. This logic is very similar for riscv, but it has an annoying exception for the 32-bit embedded ABI which lowers alignment for f64 and i64/u64 from 8 to 4. Apparently that may change as things are ratified.
| Arch::AmdGpu => emit_ptr_va_arg( | ||
| bx, | ||
| addr, | ||
| target_ty, | ||
| PassMode::Direct, | ||
| SlotSize::Bytes4, | ||
| AllowHigherAlign::No, | ||
| ForceRightAdjust::No, | ||
| ), |
There was a problem hiding this comment.
| Arch::Nvptx64 => emit_ptr_va_arg( | ||
| bx, | ||
| addr, | ||
| target_ty, | ||
| PassMode::Direct, | ||
| SlotSize::Bytes1, | ||
| AllowHigherAlign::Yes, | ||
| ForceRightAdjust::No, | ||
| ), |
There was a problem hiding this comment.
| Arch::Wasm32 | Arch::Wasm64 => emit_ptr_va_arg( | ||
| bx, | ||
| addr, | ||
| target_ty, | ||
| if layout.is_aggregate() || layout.is_zst() || layout.is_1zst() { | ||
| PassMode::Indirect | ||
| } else { | ||
| PassMode::Direct | ||
| }, | ||
| SlotSize::Bytes4, | ||
| AllowHigherAlign::Yes, | ||
| ForceRightAdjust::No, | ||
| ), |
There was a problem hiding this comment.
curiously the slot size is always 4, independent of whether wasm32 or wasm64 is used. I suppose that may change in the future if/when wasm64 gets more attention.
| Arch::CSky => emit_ptr_va_arg( | ||
| bx, | ||
| addr, | ||
| target_ty, | ||
| PassMode::Direct, | ||
| SlotSize::Bytes4, | ||
| AllowHigherAlign::Yes, | ||
| ForceRightAdjust::No, | ||
| ), |
There was a problem hiding this comment.
|
☔ The latest upstream changes (presumably #150089) made this pull request unmergeable. Please resolve the merge conflicts. |
folkertdev
force-pushed
the
more-va-arg
branch
from
1c516dc to
9d0af69
Compare
|
This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed. Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers. |
4 participants
tracking issue: #44930
Provide our own implementations in order to guarantee the behavior of
va_arg. We will only be able to stabilizec_variadicon targets where we know and guarantee the properties ofva_arg.r? workingjubilee