Skip to content

Add missing runtime tests for SSE alias intrinsics #2041

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

Merged
sayantn merged 11 commits into from
Mar 25, 2026
+114 −74
Merged

Add missing runtime tests for SSE alias intrinsics #2041

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
de85b1c
Add missing runtime tests for alias intrinsics: _mm_cvt_ss2si, _mm_cv…
ArunTamil21 Feb 26, 2026
b397dac
Refactor alias tests using meta function pattern and add missing test…
ArunTamil21 Feb 26, 2026
b5b01c0
Restore deleted test_mm_store_ps
ArunTamil21 Feb 26, 2026
1a3e798
Remove redundant tests for _mm_prefetch and _mm_undefined_ps
ArunTamil21 Feb 27, 2026
b7a5d2d
Refactor alias tests using macros instead of meta functions
ArunTamil21 Mar 12, 2026
6d1930f
Merge upstream/main and resolve conflict in x86-intel.rs
ArunTamil21 Mar 12, 2026
a4d7d29
Fix formatting
ArunTamil21 Mar 12, 2026
8c4d55c
Use macros for test body deduplication in SSE alias tests
ArunTamil21 Mar 17, 2026
66cc054
Fix formatting
ArunTamil21 Mar 17, 2026
4289c08
Use const fn and remove unsafe from alias test wrappers
ArunTamil21 Mar 22, 2026
eec5cd9
Mark alias test wrappers as const fn where supported and clean up ski…
ArunTamil21 Mar 24, 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
183 changes: 113 additions & 70 deletions crates/core_arch/src/x86/sse.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2997,66 +2997,87 @@ mod tests {
}
}

macro_rules! test_mm_cvtss_si32_impl {
($alias:ident) => {
let inputs = &[42.0f32, -3.1, 4.0e10, 4.0e-20, NAN, 2147483500.1];
let result = &[42i32, -3, i32::MIN, 0, i32::MIN, 2147483520];
for i in 0..inputs.len() {
let x = _mm_setr_ps(inputs[i], 1.0, 3.0, 4.0);
let e = result[i];
let r = $alias(x);
assert_eq!(e, r, "TestCase #{} f({:?}) = {}, expected: {}", i, x, r, e);
}
};
}

#[simd_test(enable = "sse")]
fn test_mm_cvtss_si32() {
let inputs = &[42.0f32, -3.1, 4.0e10, 4.0e-20, NAN, 2147483500.1];
let result = &[42i32, -3, i32::MIN, 0, i32::MIN, 2147483520];
for i in 0..inputs.len() {
let x = _mm_setr_ps(inputs[i], 1.0, 3.0, 4.0);
let e = result[i];
let r = _mm_cvtss_si32(x);
assert_eq!(
e, r,
"TestCase #{} _mm_cvtss_si32({:?}) = {}, expected: {}",
i, x, r, e
);
}
test_mm_cvtss_si32_impl!(_mm_cvtss_si32);
}

#[simd_test(enable = "sse")]
fn test_mm_cvt_ss2si() {
test_mm_cvtss_si32_impl!(_mm_cvt_ss2si);
}

macro_rules! test_cvttss_si32_impl {
($alias:ident) => {
let inputs = &[
(42.0f32, 42i32),
(-31.4, -31),
(-33.5, -33),
(-34.5, -34),
(10.999, 10),
(-5.99, -5),
(4.0e10, i32::MIN),
(4.0e-10, 0),
(NAN, i32::MIN),
(2147483500.1, 2147483520),
];
for (i, &(xi, e)) in inputs.iter().enumerate() {
let x = _mm_setr_ps(xi, 1.0, 3.0, 4.0);
let r = $alias(x);
assert_eq!(e, r, "TestCase #{} f({:?}) = {}, expected: {}", i, x, r, e);
}
};
}

#[simd_test(enable = "sse")]
fn test_mm_cvttss_si32() {
let inputs = &[
(42.0f32, 42i32),
(-31.4, -31),
(-33.5, -33),
(-34.5, -34),
(10.999, 10),
(-5.99, -5),
(4.0e10, i32::MIN),
(4.0e-10, 0),
(NAN, i32::MIN),
(2147483500.1, 2147483520),
];
for (i, &(xi, e)) in inputs.iter().enumerate() {
let x = _mm_setr_ps(xi, 1.0, 3.0, 4.0);
let r = _mm_cvttss_si32(x);
assert_eq!(
e, r,
"TestCase #{} _mm_cvttss_si32({:?}) = {}, expected: {}",
i, x, r, e
);
}
test_cvttss_si32_impl!(_mm_cvttss_si32);
}

#[simd_test(enable = "sse")]
const fn test_mm_cvtsi32_ss() {
let a = _mm_setr_ps(5.0, 6.0, 7.0, 8.0);
fn test_mm_cvtt_ss2si() {
test_cvttss_si32_impl!(_mm_cvtt_ss2si);
}

let r = _mm_cvtsi32_ss(a, 4555);
let e = _mm_setr_ps(4555.0, 6.0, 7.0, 8.0);
assert_eq_m128(e, r);
macro_rules! test_mm_cvtsi32_ss_impl {
($alias:ident) => {
let a = _mm_setr_ps(5.0, 6.0, 7.0, 8.0);

let r = _mm_cvtsi32_ss(a, 322223333);
let e = _mm_setr_ps(322223333.0, 6.0, 7.0, 8.0);
assert_eq_m128(e, r);
let r = $alias(a, 4555);
assert_eq_m128(_mm_setr_ps(4555.0, 6.0, 7.0, 8.0), r);

let r = _mm_cvtsi32_ss(a, -432);
let e = _mm_setr_ps(-432.0, 6.0, 7.0, 8.0);
assert_eq_m128(e, r);
let r = $alias(a, 322223333);
assert_eq_m128(_mm_setr_ps(322223333.0, 6.0, 7.0, 8.0), r);

let r = _mm_cvtsi32_ss(a, -322223333);
let e = _mm_setr_ps(-322223333.0, 6.0, 7.0, 8.0);
assert_eq_m128(e, r);
let r = $alias(a, -432);
assert_eq_m128(_mm_setr_ps(-432.0, 6.0, 7.0, 8.0), r);

let r = $alias(a, -322223333);
assert_eq_m128(_mm_setr_ps(-322223333.0, 6.0, 7.0, 8.0), r);
};
}

#[simd_test(enable = "sse")]
const fn test_mm_cvtsi32_ss() {
test_mm_cvtsi32_ss_impl!(_mm_cvtsi32_ss);
}

#[simd_test(enable = "sse")]
fn test_mm_cvt_si2ss() {
test_mm_cvtsi32_ss_impl!(_mm_cvt_si2ss);
}

#[simd_test(enable = "sse")]
Expand All @@ -3071,18 +3092,24 @@ mod tests {
assert_eq_m128(r, _mm_setr_ps(4.25, 0.0, 0.0, 0.0));
}

macro_rules! test_mm_set1_ps_impl {
($alias:ident) => {
let r = $alias(black_box(4.25));
assert_eq!(get_m128(r, 0), 4.25);
assert_eq!(get_m128(r, 1), 4.25);
assert_eq!(get_m128(r, 2), 4.25);
assert_eq!(get_m128(r, 3), 4.25);
};
}

#[simd_test(enable = "sse")]
const fn test_mm_set1_ps() {
let r1 = _mm_set1_ps(black_box(4.25));
let r2 = _mm_set_ps1(black_box(4.25));
assert_eq!(get_m128(r1, 0), 4.25);
assert_eq!(get_m128(r1, 1), 4.25);
assert_eq!(get_m128(r1, 2), 4.25);
assert_eq!(get_m128(r1, 3), 4.25);
assert_eq!(get_m128(r2, 0), 4.25);
assert_eq!(get_m128(r2, 1), 4.25);
assert_eq!(get_m128(r2, 2), 4.25);
assert_eq!(get_m128(r2, 3), 4.25);
test_mm_set1_ps_impl!(_mm_set1_ps);
}

#[simd_test(enable = "sse")]
const fn test_mm_set_ps1() {
test_mm_set1_ps_impl!(_mm_set_ps1);
}

#[simd_test(enable = "sse")]
Expand Down Expand Up @@ -3171,11 +3198,22 @@ mod tests {
assert_eq_m128(r, _mm_setr_ps(42.0, 0.0, 0.0, 0.0));
}

macro_rules! test_mm_load1_ps_impl {
($alias:ident) => {
let a = 42.0f32;
let r = unsafe { $alias(ptr::addr_of!(a)) };
assert_eq_m128(r, _mm_setr_ps(42.0, 42.0, 42.0, 42.0));
};
}

#[simd_test(enable = "sse")]
const fn test_mm_load1_ps() {
let a = 42.0f32;
let r = unsafe { _mm_load1_ps(ptr::addr_of!(a)) };
assert_eq_m128(r, _mm_setr_ps(42.0, 42.0, 42.0, 42.0));
test_mm_load1_ps_impl!(_mm_load1_ps);
}

#[simd_test(enable = "sse")]
const fn test_mm_load_ps1() {
test_mm_load1_ps_impl!(_mm_load_ps1);
}

#[simd_test(enable = "sse")]
Expand Down Expand Up @@ -3227,19 +3265,24 @@ mod tests {
assert_eq!(vals[2], 0.0);
}

macro_rules! test_mm_store1_ps_impl {
($alias:ident) => {
let mut vals = Memory { data: [0.0f32; 4] };
let a = _mm_setr_ps(1.0, 2.0, 3.0, 4.0);
let p = vals.data.as_mut_ptr();
unsafe { $alias(p, *black_box(&a)) };
assert_eq!(vals.data, [1.0, 1.0, 1.0, 1.0]);
};
}

#[simd_test(enable = "sse")]
const fn test_mm_store1_ps() {
let mut vals = Memory { data: [0.0f32; 4] };
let a = _mm_setr_ps(1.0, 2.0, 3.0, 4.0);

// guaranteed to be aligned to 16 bytes
let p = vals.data.as_mut_ptr();

unsafe {
_mm_store1_ps(p, *black_box(&a));
}
test_mm_store1_ps_impl!(_mm_store1_ps);
}

assert_eq!(vals.data, [1.0, 1.0, 1.0, 1.0]);
#[simd_test(enable = "sse")]
const fn test_mm_store_ps1() {
test_mm_store1_ps_impl!(_mm_store_ps1);
}

#[simd_test(enable = "sse")]
Expand Down
5 changes: 1 addition & 4 deletions crates/stdarch-verify/tests/x86-intel.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -214,7 +214,7 @@ fn verify_all_signatures() {
"_m_prefetchrs",
// CMPXCHG
"cmpxchg16b",
// Undefined
// Undefined,
"_mm_undefined_ps",
"_mm_undefined_pd",
"_mm_undefined_si128",
Expand Down Expand Up @@ -250,9 +250,6 @@ fn verify_all_signatures() {
"_mm_cvt_ss2si",
"_mm_cvtt_ss2si",
"_mm_cvt_si2ss",
"_mm_set_ps1",
"_mm_load_ps1",
"_mm_store_ps1",
"_mm_bslli_si128",
"_mm_bsrli_si128",
"_bextr2_u32",
Expand Down