x86: Correctly pass larger structs/types in registers with -Zregparm #147628
Conversation
rustbot
added
S-waiting-on-review
|
rustbot has assigned @jdonszelmann. Use |
This comment has been minimized.
This comment has been minimized.
|
The |
|
If this fixes that issue, I'd love to see a test for that issue which now passes. At least some amount of testing seems wise. I'm aware this is an unstable option I know little about otherwise, but since there is a specific test failing, I think adding it as a test is the least we can do before merging this fix so we won't forget about this failure in the future. |
|
@rustbot author |
rustbot
added
S-waiting-on-author
|
Reminder, once the PR becomes ready for a review, use |
|
Does this make us match Clang's LLVM IR? There are two tests that should probably be updated, |
The -Zregparm option for 32-bit x86 modifies the calling convention to pass a number of arguments in registers instead of on the stack. (This is primarily used by the Linux kernel.) Currently, rustc will only pass integers of size <= 32 bits in registers, but gcc (and clang) will pass larger integers (e.g., 64-bit integers) across two registers if possible. This is not the case with fastcall/vectorcall. Note that gcc/clang with -mregparm will also pass some structs in registers: this patch does not attempt to fix that case.
|
This PR was rebased onto a different master 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. |
5 participants
Fixes #145694. I think this should be the last compiler fix needed to unblock 32-bit x86 support for Rust-for-linux (Rust-for-Linux/linux#78)
Tracking issue for -Zregparm: #131749
The -Zregparm option for 32-bit x86 modifies the calling convention to pass a number of arguments in registers instead of on the stack. (This is primarily used by the Linux kernel.)
Currently, rustc will only pass primitive integer types in registers, which seems to match the gcc documentation[1], which says that arguments "of integral type" are passed as registers. This also matches the fastcall and vectorcall conventions on windows.
However, it seems that any type — most particularly structs — should be passed as a register if possible. This matches what clang and gcc do, and so avoids an ABI mismatch when linking with C code (which, after all, is the whole point of supporting -Zregparm).
Note that I haven't tested this particularly thoroughly outside the Linux kernel usecase, and in particular haven't tried to implement any 'assembly-llvm' tests, nor test for regressions against fastcall/vectorcall.