Skip to content
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

Avoid using simd_fmin/simd_fmax for AVX float min/max comparisons #1155

Merged
merged 1 commit into from
May 9, 2021
Merged

Avoid using simd_fmin/simd_fmax for AVX float min/max comparisons #1155

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
84 changes: 80 additions & 4 deletions crates/core_arch/src/x86/avx.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -196,7 +196,7 @@ pub unsafe fn _mm256_andnot_ps(a: __m256, b: __m256) -> __m256 {
#[cfg_attr(test, assert_instr(vmaxpd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_max_pd(a: __m256d, b: __m256d) -> __m256d {
simd_fmax(a, b)
vmaxpd(a, b)
}

/// Compares packed single-precision (32-bit) floating-point elements in `a`
Expand All @@ -208,7 +208,7 @@ pub unsafe fn _mm256_max_pd(a: __m256d, b: __m256d) -> __m256d {
#[cfg_attr(test, assert_instr(vmaxps))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_max_ps(a: __m256, b: __m256) -> __m256 {
simd_fmax(a, b)
vmaxps(a, b)
}

/// Compares packed double-precision (64-bit) floating-point elements
Expand All @@ -220,7 +220,7 @@ pub unsafe fn _mm256_max_ps(a: __m256, b: __m256) -> __m256 {
#[cfg_attr(test, assert_instr(vminpd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_min_pd(a: __m256d, b: __m256d) -> __m256d {
simd_fmin(a, b)
vminpd(a, b)
}

/// Compares packed single-precision (32-bit) floating-point elements in `a`
Expand All @@ -232,7 +232,7 @@ pub unsafe fn _mm256_min_pd(a: __m256d, b: __m256d) -> __m256d {
#[cfg_attr(test, assert_instr(vminps))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm256_min_ps(a: __m256, b: __m256) -> __m256 {
simd_fmin(a, b)
vminps(a, b)
}

/// Multiplies packed double-precision (64-bit) floating-point elements
Expand Down Expand Up @@ -3034,6 +3034,14 @@ extern "C" {
fn movmskpd256(a: __m256d) -> i32;
#[link_name = "llvm.x86.avx.movmsk.ps.256"]
fn movmskps256(a: __m256) -> i32;
#[link_name = "llvm.x86.avx.min.ps.256"]
fn vminps(a: __m256, b: __m256) -> __m256;
#[link_name = "llvm.x86.avx.max.ps.256"]
fn vmaxps(a: __m256, b: __m256) -> __m256;
#[link_name = "llvm.x86.avx.min.pd.256"]
fn vminpd(a: __m256d, b: __m256d) -> __m256d;
#[link_name = "llvm.x86.avx.max.pd.256"]
fn vmaxpd(a: __m256d, b: __m256d) -> __m256d;
}

#[cfg(test)]
Expand Down Expand Up @@ -3138,6 +3146,23 @@ mod tests {
let r = _mm256_max_pd(a, b);
let e = _mm256_setr_pd(2., 4., 6., 8.);
assert_eq_m256d(r, e);
// > If the values being compared are both 0.0s (of either sign), the
// > value in the second operand (source operand) is returned.
let w = _mm256_max_pd(_mm256_set1_pd(0.0), _mm256_set1_pd(-0.0));
let x = _mm256_max_pd(_mm256_set1_pd(-0.0), _mm256_set1_pd(0.0));
let wu: [u64; 4] = transmute(w);
let xu: [u64; 4] = transmute(x);
assert_eq!(wu, [0x8000_0000_0000_0000u64; 4]);
assert_eq!(xu, [0u64; 4]);
// > If only one value is a NaN (SNaN or QNaN) for this instruction, the
// > second operand (source operand), either a NaN or a valid
// > floating-point value, is written to the result.
let y = _mm256_max_pd(_mm256_set1_pd(f64::NAN), _mm256_set1_pd(0.0));
let z = _mm256_max_pd(_mm256_set1_pd(0.0), _mm256_set1_pd(f64::NAN));
let yf: [f64; 4] = transmute(y);
let zf: [f64; 4] = transmute(z);
assert_eq!(yf, [0.0; 4]);
assert!(zf.iter().all(|f| f.is_nan()), "{:?}", zf);
}

#[simd_test(enable = "avx")]
Expand All @@ -3147,6 +3172,23 @@ mod tests {
let r = _mm256_max_ps(a, b);
let e = _mm256_setr_ps(2., 4., 6., 8., 10., 12., 14., 16.);
assert_eq_m256(r, e);
// > If the values being compared are both 0.0s (of either sign), the
// > value in the second operand (source operand) is returned.
let w = _mm256_max_ps(_mm256_set1_ps(0.0), _mm256_set1_ps(-0.0));
let x = _mm256_max_ps(_mm256_set1_ps(-0.0), _mm256_set1_ps(0.0));
let wu: [u32; 8] = transmute(w);
let xu: [u32; 8] = transmute(x);
assert_eq!(wu, [0x8000_0000u32; 8]);
assert_eq!(xu, [0u32; 8]);
// > If only one value is a NaN (SNaN or QNaN) for this instruction, the
// > second operand (source operand), either a NaN or a valid
// > floating-point value, is written to the result.
let y = _mm256_max_ps(_mm256_set1_ps(f32::NAN), _mm256_set1_ps(0.0));
let z = _mm256_max_ps(_mm256_set1_ps(0.0), _mm256_set1_ps(f32::NAN));
let yf: [f32; 8] = transmute(y);
let zf: [f32; 8] = transmute(z);
assert_eq!(yf, [0.0; 8]);
assert!(zf.iter().all(|f| f.is_nan()), "{:?}", zf);
}

#[simd_test(enable = "avx")]
Expand All @@ -3156,6 +3198,23 @@ mod tests {
let r = _mm256_min_pd(a, b);
let e = _mm256_setr_pd(1., 3., 5., 7.);
assert_eq_m256d(r, e);
// > If the values being compared are both 0.0s (of either sign), the
// > value in the second operand (source operand) is returned.
let w = _mm256_min_pd(_mm256_set1_pd(0.0), _mm256_set1_pd(-0.0));
let x = _mm256_min_pd(_mm256_set1_pd(-0.0), _mm256_set1_pd(0.0));
let wu: [u64; 4] = transmute(w);
let xu: [u64; 4] = transmute(x);
assert_eq!(wu, [0x8000_0000_0000_0000u64; 4]);
assert_eq!(xu, [0u64; 4]);
// > If only one value is a NaN (SNaN or QNaN) for this instruction, the
// > second operand (source operand), either a NaN or a valid
// > floating-point value, is written to the result.
let y = _mm256_min_pd(_mm256_set1_pd(f64::NAN), _mm256_set1_pd(0.0));
let z = _mm256_min_pd(_mm256_set1_pd(0.0), _mm256_set1_pd(f64::NAN));
let yf: [f64; 4] = transmute(y);
let zf: [f64; 4] = transmute(z);
assert_eq!(yf, [0.0; 4]);
assert!(zf.iter().all(|f| f.is_nan()), "{:?}", zf);
}

#[simd_test(enable = "avx")]
Expand All @@ -3165,6 +3224,23 @@ mod tests {
let r = _mm256_min_ps(a, b);
let e = _mm256_setr_ps(1., 3., 5., 7., 9., 11., 13., 15.);
assert_eq_m256(r, e);
// > If the values being compared are both 0.0s (of either sign), the
// > value in the second operand (source operand) is returned.
let w = _mm256_min_ps(_mm256_set1_ps(0.0), _mm256_set1_ps(-0.0));
let x = _mm256_min_ps(_mm256_set1_ps(-0.0), _mm256_set1_ps(0.0));
let wu: [u32; 8] = transmute(w);
let xu: [u32; 8] = transmute(x);
assert_eq!(wu, [0x8000_0000u32; 8]);
assert_eq!(xu, [0u32; 8]);
// > If only one value is a NaN (SNaN or QNaN) for this instruction, the
// > second operand (source operand), either a NaN or a valid
// > floating-point value, is written to the result.
let y = _mm256_min_ps(_mm256_set1_ps(f32::NAN), _mm256_set1_ps(0.0));
let z = _mm256_min_ps(_mm256_set1_ps(0.0), _mm256_set1_ps(f32::NAN));
let yf: [f32; 8] = transmute(y);
let zf: [f32; 8] = transmute(z);
assert_eq!(yf, [0.0; 8]);
assert!(zf.iter().all(|f| f.is_nan()), "{:?}", zf);
}

#[simd_test(enable = "avx")]
Expand Down