Skip to content

PowerPC: Add intrinsics support vec_add(vector_double, vector_double) #2161

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
lei137 wants to merge 3 commits into
base: main
Choose a base branch
from
+34 −1
Open

PowerPC: Add intrinsics support vec_add(vector_double, vector_double) #2161

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
221954c
Add intrinsics support on Linux on Power for vec_add(vector_double, v…
lei137 Jun 15, 2026
581f358
fix ambiguous export issue with vec_add
lei137 Jun 15, 2026
e67122b
fmt
lei137 Jun 15, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion crates/core_arch/src/powerpc/altivec.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -411,7 +411,7 @@ unsafe extern "unadjusted" {
}

#[macro_use]
mod sealed {
pub(crate) mod sealed {
use super::*;

#[unstable(feature = "stdarch_powerpc", issue = "111145")]
Expand Down
33 changes: 33 additions & 0 deletions crates/core_arch/src/powerpc/vsx.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,7 @@

use crate::core_arch::powerpc::*;
use crate::core_arch::simd::*;
use crate::intrinsics::simd::simd_add;

#[cfg(test)]
use stdarch_test::assert_instr;
Expand Down Expand Up @@ -171,6 +172,27 @@ mod sealed {
vec_mergeeo! { vector_unsigned_int, mergee, mergeo }
vec_mergeeo! { vector_bool_int, mergee, mergeo }
vec_mergeeo! { vector_float, mergee, mergeo }

#[inline]
#[target_feature(enable = "vsx")]
#[cfg_attr(test, assert_instr(xvadddp))]
pub(crate) unsafe fn vec_add_double_double(
a: vector_double,
b: vector_double,
) -> vector_double {
simd_add(a, b)
}
}

// Implement AltiVec's VectorAdd trait for vector_double to enable vec_add support
#[unstable(feature = "stdarch_powerpc", issue = "111145")]
impl crate::core_arch::powerpc::altivec::sealed::VectorAdd<vector_double> for vector_double {
type Result = vector_double;
#[inline]
#[target_feature(enable = "vsx")]
unsafe fn vec_add(self, other: vector_double) -> Self::Result {

@folkertdev folkertdev Jun 16, 2026 •
edited by rustbot
Loading

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

is this the only function you'd want to do this for? I'd guess not?

View changes since the review

@lei137 lei137 Jun 17, 2026

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorry, I do not understand what you mean. Can you please elaborate?

@folkertdev folkertdev Jun 17, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

well what about vec_sub and other vec_* methods? Should they not also follow this pattern, or is vec_add special somehow?

@lei137 lei137 Jun 17, 2026

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah yes! I do have a few more other ones to add. I tried to limit this PR to just vec_add as it is my first patch so I was just trying to get a feel on how these should be done.

I have a list of intrinsics that I need to add and the plan was to follow up with sets of 3/4 intrinsics based on type/complexity afterwards. Is there a preferred PR size/breakdown?

sealed::vec_add_double_double(self, other)
}
}
Comment on lines +187 to 196

@folkertdev folkertdev Jun 16, 2026 •
edited by rustbot
Loading

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

So is the argument here that if you got your hands on a vector_double then vsx must be enabled?

Because normally adding stricter target features on a trait method is error-prone: there is now an additional safety requirement on that trait method, which is that if that particular instance is called, the target feature must be enabled.

View changes since the review

@lei137 lei137 Jun 17, 2026

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes that was the idea. On PPC, vec_add(vector_double, vector_double) should result in code gen xvadddp, which is only available if vsx feature is enabled.

@folkertdev folkertdev Jun 17, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hmm yeah the s390x implementation has the same sort of problem with float in their case. Many instructions for f32 were only added in a later target feature. Actually this is fine for vec_add because the implementation is just simd_add, which will generate xvadddp when the right target features are available, or some fallback thing (using scalars, probably) when that target feature is not available.


/// Vector permute.
Expand Down Expand Up @@ -255,4 +277,15 @@ mod tests {
test_vec_xxpermdi! {test_vec_xxpermdi_i64x2, i64x2, vector_signed_long, [0], [-1], [2], [-3]}
test_vec_xxpermdi! {test_vec_xxpermdi_m64x2, m64x2, vector_bool_long, [false], [true], [false], [true]}
test_vec_xxpermdi! {test_vec_xxpermdi_f64x2, f64x2, vector_double, [0.0], [1.0], [2.0], [3.0]}

#[simd_test(enable = "vsx")]
fn test_vec_add_f64x2_f64x2() {
let a = vector_double::from(f64x2::from_array([1.0, 2.0]));
let b = vector_double::from(f64x2::from_array([3.0, 4.0]));
let expected = vector_double::from(f64x2::from_array([4.0, 6.0]));

unsafe {
assert_eq!(f64x2::from(vec_add(a, b)), f64x2::from(expected));
}
}
}