Added AVX2, some rework and improvements

awxkee · May 28, 2024 · 95c63d5 · 95c63d5
1 parent b8cd9a5
commit 95c63d5
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/src/sse_math.rs b/src/sse_math.rs
@@ -8,15 +8,15 @@ use std::arch::x86::*;
 #[inline]
 #[allow(dead_code)]
 pub unsafe fn _mm_prefer_fma_ps(a: __m128, b: __m128, c: __m128) -> __m128 {
-    return _mm_fmadd_ps(b, c, a);
+    return _mm_add_ps(_mm_mul_ps(b, c), a);
 }
 
 #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
 #[cfg(target_feature = "fma")]
 #[inline]
 #[allow(dead_code)]
 pub unsafe fn _mm_prefer_fma_ps(a: __m128, b: __m128, c: __m128) -> __m128 {
-    return _mm_add_ps(_mm_mul_ps(b, c), a);
+    return _mm_fmadd_ps(b, c, a);
 }
 
 #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]