Move llvm.x86.* shims into shims::x86 and implement _addcarry_u32 and _subborrow_u{32,64}
#3075
Conversation
|
☔ The latest upstream changes (presumably #3072) made this pull request unmergeable. Please resolve the merge conflicts. |
| { | ||
| let [] = this.check_shim(abi, Abi::C { unwind: false }, link_name, args)?; | ||
| this.yield_active_thread(); | ||
| } | ||
| "llvm.aarch64.isb" if this.tcx.sess.target.arch == "aarch64" => { |
There was a problem hiding this comment.
| "llvm.aarch64.isb" if this.tcx.sess.target.arch == "aarch64" => { | |
| // FIXME: Move these to an `arm` submodule. | |
| "llvm.aarch64.isb" if this.tcx.sess.target.arch == "aarch64" => { |
tests/pass/intrinsics-x86.rs
Outdated
| pub fn main() { | ||
| assert_eq!(adc(0, 1, 1), (0, 2)); | ||
| assert_eq!(adc(1, 1, 1), (0, 3)); | ||
| assert_eq!(adc(2, 1, 1), (0, 3)); |
There was a problem hiding this comment.
| assert_eq!(adc(2, 1, 1), (0, 3)); | |
| assert_eq!(adc(2, 1, 1), (0, 3)); // any non-zero carry acts as 1! |
| @@ -1,3 +1,51 @@ | |||
| #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] | |||
There was a problem hiding this comment.
Ah, here's another way to handle tests that should run only on x86. Might be good to also port the other tests (that use the long ignore- list) to this pattern.
There was a problem hiding this comment.
For SSE and SSE2 that will work fine because they are enabled by default. But other features require -C target-feature=+feature, which will generate warnings on non-x86.
Maybe a new directive could be added to compiletest ui_test? Something like @only-x86-family
There was a problem hiding this comment.
I've done it for SSE and SSE2 for the time being.
eduardosm
force-pushed
the
x86-addcarry-subborrow
branch
from
f5a5d3e
to
fed4244
Compare
|
It looks to me like you addressed Ralf's comments above but GitHub doesn't seem to have noticed 🤔 |
|
@eduardosm Can you squash the commits down a bit? I think your original 4 commits are a good organization, but the rest don't stand on their own. Then this LGTM (and I would rather like to sync these changes into rust-lang/rust). |
…` to make it generic
The input carry is an 8-bit value that is interpreted as 1 when it is non-zero. The output carry is an 8-bit value that will be 0 or 1. https://www.intel.com/content/www/us/en/docs/cpp-compiler/developer-guide-reference/2021-8/addcarry-u32-addcarry-u64.html
eduardosm
force-pushed
the
x86-addcarry-subborrow
branch
from
86c4f91
to
ce3a919
Compare
|
Ready! |
|
@bors r+ |
|
☀️ Test successful - checks-actions |
This PR moves all
llvm.x86.*shims intoshims::x86and addsllvm.x86.addcarry.32,llvm.x86.subborrow.32andllvm.x86.subborrow.64.Additionally, it fixes the input carry semantics of
llvm.x86.addcarry.32. The input carry is an 8-bit value that is interpreted as 1 when it is non-zero.https://www.intel.com/content/www/us/en/docs/cpp-compiler/developer-guide-reference/2021-8/addcarry-u32-addcarry-u64.html