⚡ use faster SIMD intrinsics

jvdd · Feb 7, 2023 · df40466 · df40466
1 parent bae6139
commit df40466
Showing 1 changed file with 5 additions and 5 deletions.
diff --git a/src/simd/simd_f32.rs b/src/simd/simd_f32.rs
@@ -12,10 +12,12 @@ use std::arch::x86_64::*;
 use super::task::{max_index_value, min_index_value};
 
 const XOR_VALUE: i32 = 0x7FFFFFFF;
+const BIT_SHIFT: i32 = 31;
 
+#[inline(always)]
 fn _ord_i32_to_f32(ord_i32: i32) -> f32 {
     // TODO: more efficient transformation -> can be decreasing order as well
-    let v = ((ord_i32 >> 31) & XOR_VALUE) ^ ord_i32;
+    let v = ((ord_i32 >> BIT_SHIFT) & XOR_VALUE) ^ ord_i32;
     unsafe { std::mem::transmute::<i32, f32>(v) }
 }
 
@@ -28,14 +30,13 @@ mod avx2 {
 
     const LANE_SIZE: usize = AVX2::LANE_SIZE_32;
     const XOR_MASK: __m256i = unsafe { std::mem::transmute([XOR_VALUE; LANE_SIZE]) };
-    const BIT_SHIFT: i32 = 31;
 
     #[inline(always)]
     unsafe fn _f32_as_m256i_to_i32ord(f32_as_m256i: __m256i) -> __m256i {
         // on a scalar: ((v >> 31) & 0x7FFFFFFF) ^ v
         let sign_bit_shifted = _mm256_srai_epi32(f32_as_m256i, BIT_SHIFT);
         let sign_bit_masked = _mm256_and_si256(sign_bit_shifted, XOR_MASK);
-        _mm256_xor_epi32(sign_bit_masked, f32_as_m256i)
+        _mm256_xor_si256(sign_bit_masked, f32_as_m256i)
     }
 
     #[inline(always)]
@@ -55,7 +56,7 @@ mod avx2 {
 
         #[inline(always)]
         unsafe fn _mm_loadu(data: *const f32) -> __m256i {
-            _f32_as_m256i_to_i32ord(_mm256_loadu_epi32(data as *const i32))
+            _f32_as_m256i_to_i32ord(_mm256_loadu_si256(data as *const __m256i))
         }
 
         #[inline(always)]
@@ -259,7 +260,6 @@ mod sse {
 
     const LANE_SIZE: usize = SSE::LANE_SIZE_32;
     const XOR_MASK: __m128i = unsafe { std::mem::transmute([XOR_VALUE; LANE_SIZE]) };
-    const BIT_SHIFT: i32 = 31;
 
     #[inline(always)]
     unsafe fn _f32_as_m128i_to_i32ord(f32_as_m128i: __m128i) -> __m128i {