fix ci

Author: Human9000-bit

crates/yscv-imgproc/src/tests/morphology.rs

@@ -422,7 +422,11 @@ fn binary_box_matches_generic_ones_kernel() {
         state = state
             .wrapping_mul(6364136223846793005)
             .wrapping_add(1442695040888963407);
-        data.push(if (state >> 33) % 3 == 0 { 1.0f32 } else { 0.0 });
+        data.push(if (state >> 33).is_multiple_of(3) {
+            1.0f32
+        } else {
+            0.0
+        });
     }
     let img = Tensor::from_vec(vec![h, w, 1], data).unwrap();
     for ksize in [1usize, 3, 5, 9] {

crates/yscv-kernels/src/ops/simd/activation.rs

@@ -1210,9 +1210,9 @@ unsafe fn sigmoid_slice_neon(input: &[f32], output: &mut [f32]) {
     let len = input.len();
     let in_ptr = input.as_ptr();
     let out_ptr = output.as_mut_ptr();
-    let one = vdupq_n_f32(1.0);
     let mut i = 0usize;
     unsafe {
+        let one = vdupq_n_f32(1.0);
         while i + 16 <= len {
             let x0 = vld1q_f32(in_ptr.add(i));
             let x1 = vld1q_f32(in_ptr.add(i + 4));

crates/yscv-kernels/src/ops/simd/exp.rs

@@ -25,9 +25,8 @@ use std::arch::x86_64::{
 };
 #[cfg(target_arch = "x86_64")]
 use std::arch::x86_64::{
-    __m512, _mm512_add_epi32, _mm512_add_ps, _mm512_castsi512_ps, _mm512_cvtepi32_ps,
-    _mm512_cvtps_epi32, _mm512_div_ps, _mm512_fmadd_ps, _mm512_loadu_ps, _mm512_max_ps,
-    _mm512_min_ps, _mm512_mul_ps, _mm512_scalef_ps, _mm512_set1_epi32, _mm512_set1_ps,
+    __m512, _mm512_add_ps, _mm512_cvtepi32_ps, _mm512_cvtps_epi32, _mm512_div_ps, _mm512_fmadd_ps,
+    _mm512_loadu_ps, _mm512_max_ps, _mm512_min_ps, _mm512_mul_ps, _mm512_scalef_ps, _mm512_set1_ps,
     _mm512_storeu_ps, _mm512_sub_ps,
 };
 
@@ -292,22 +291,6 @@ pub(crate) unsafe fn fast_exp_sigmoid_avx(x: __m256) -> __m256 {
 // AVX-512 fast-exp helper (16-wide)
 // ===========================================================================
 
-/// Schraudolph 1999 bit-trick exp for AVX-512.
-#[cfg(target_arch = "x86_64")]
-#[allow(unsafe_code)]
-#[allow(unsafe_op_in_unsafe_fn)]
-#[target_feature(enable = "avx512f")]
-#[inline]
-pub(crate) unsafe fn fast_exp_bittrick_avx512(x: __m512) -> __m512 {
-    let scale = _mm512_set1_ps(12102203.0);
-    let offset = _mm512_set1_epi32(1065353216);
-    let clamp_lo = _mm512_set1_ps(-87.0);
-    let clamp_hi = _mm512_set1_ps(88.0);
-    let x_clamped = _mm512_max_ps(_mm512_min_ps(x, clamp_hi), clamp_lo);
-    let val = _mm512_cvtps_epi32(_mm512_mul_ps(x_clamped, scale));
-    _mm512_castsi512_ps(_mm512_add_epi32(val, offset))
-}
-
 /// Polynomial exp for AVX-512: same range reduction and coefficients as AVX.
 #[cfg(target_arch = "x86_64")]
 #[allow(unsafe_code)]