optimize round()

NamorNiradnug · May 4, 2024 · 16d710f · 16d710f
1 parent fd0448b
commit 16d710f
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 2 deletions.
diff --git a/src/math/f32/trig.rs b/src/math/f32/trig.rs
@@ -30,7 +30,10 @@ where
     abs_x = abs_x
         .simd_lt(Simd::splat(INPUT_LIMIT))
         .select(abs_x, Simd::default());
-    let quadrants_float = (abs_x * Simd::splat(FRAC_2_PI)).round();
+    // (abs_x * Simd::splat(FRAC_2_PI)).round() generates more instruntions and hence is slower
+    let quadrants_float = abs_x
+        .mul_add(Simd::splat(FRAC_2_PI), Simd::splat(0.5))
+        .trunc();
 
     // SAFETY: INPUT_LIMIT guaratees that `quadrants_float` are representable in u32
     let quadrants = unsafe { quadrants_float.to_int_unchecked::<i32>().cast() };

diff --git a/src/math/f64/trig.rs b/src/math/f64/trig.rs
@@ -26,7 +26,10 @@ where
     abs_x = abs_x
         .simd_lt(Simd::splat(INPUT_LIMIT))
         .select(abs_x, Simd::default());
-    let quadrants_float = (abs_x * Simd::splat(FRAC_2_PI)).round();
+    // (abs_x * Simd::splat(FRAC_2_PI)).round() generates more instruntions and hence is slower
+    let quadrants_float = abs_x
+        .mul_add(Simd::splat(FRAC_2_PI), Simd::splat(0.5))
+        .trunc();
 
     // SAFETY: INPUT_LIMIT guarantees that values in `quadrants_float` are representable in u64
     let quadrants = unsafe { quadrants_float.to_int_unchecked::<i64>().cast() };