reflect changes in std::simd

src/simd.rs

@@ -3,6 +3,9 @@ use std::simd::Simd;
 #[cfg(target_feature="avx")]
 use std::arch::x86_64::*;
 
+#[cfg(not(target_feature="avx"))]
+use std::simd::{SimdFloat, SimdPartialOrd};
+
 // Number of single-precision floating-point numbers per vector
 #[cfg(target_feature="avx")]
 pub const LANES : usize = 8;
@@ -24,6 +27,7 @@ pub(crate) use simd_load;
 //   https://stackoverflow.com/questions/13879609
 //   https://stackoverflow.com/questions/41303780
 #[cfg(target_feature="avx")]
+#[inline]
 pub fn horizontal_sum(x : Simd<f32, 8>) -> f32
 {
   unsafe {
@@ -40,6 +44,7 @@ pub fn horizontal_sum(x : Simd<f32, 8>) -> f32
 }
 
 #[cfg(not(target_feature="avx"))]
+#[inline]
 pub fn horizontal_sum(x : Simd<f32, LANES>) -> f32
 {
   return x.reduce_sum();
@@ -80,6 +85,7 @@ pub fn horizontal_sum(x : Simd<f32, LANES>) -> f32
 //   I'm not certain.
 //
 #[cfg(target_feature="avx")]
+#[inline]
 pub fn relu_ps(x : Simd<f32, 8>) -> Simd<f32, 8>
 {
   let x = __m256::from(x);
@@ -88,13 +94,15 @@ pub fn relu_ps(x : Simd<f32, 8>) -> Simd<f32, 8>
 }
 
 #[cfg(not(target_feature="avx"))]
+#[inline]
 pub fn relu_ps(x : Simd<f32, LANES>) -> Simd<f32, LANES>
 {
-  return x.max(x * Simd::splat(0.03125));
+  return x.simd_max(x * Simd::splat(0.03125));
 }
 
 // The same caveat in re broadcast/set1 as above applies here.
 #[cfg(target_feature="avx")]
+#[inline]
 pub fn d_relu_ps(x : Simd<f32, 8>) -> Simd<f32, 8>
 {
   unsafe {
@@ -109,7 +117,8 @@ pub fn d_relu_ps(x : Simd<f32, 8>) -> Simd<f32, 8>
 }
 
 #[cfg(not(target_feature="avx"))]
+#[inline]
 pub fn d_relu_ps(x : Simd<f32, LANES>) -> Simd<f32, LANES>
 {
-  return x.lanes_lt(Simd::splat(0.0)).select(Simd::splat(0.03125), Simd::splat(1.0));
+  return x.simd_lt(Simd::splat(0.0)).select(Simd::splat(0.03125), Simd::splat(1.0));
 }