Force fallback to slow round on wasm32

Lokathor · Oct 27, 2022 · 1ba208b · 1ba208b
1 parent fc05722
commit 1ba208b
Show file tree

Hide file tree

Showing 4 changed files with 34 additions and 30 deletions.
diff --git a/src/f32x4_.rs b/src/f32x4_.rs
@@ -628,13 +628,14 @@ impl f32x4 {
         mask.blend(self, f)
       } else if #[cfg(target_feature="simd128")] {
         Self { simd: f32x4_nearest(self.simd) }
-      } else if #[cfg(feature="std")] {
-        Self { arr: [
-          self.arr[0].round(),
-          self.arr[1].round(),
-          self.arr[2].round(),
-          self.arr[3].round(),
-        ]}
+      } else if #[cfg(all(feature="std", not(target_arch="wasm32")))] {
+        let data = self.to_array();
+        Self::new([
+          data[0].round(),
+          data[1].round(),
+          data[2].round(),
+          data[3].round(),
+        ])
       } else {
         // Note(Lokathor): This software fallback is probably very slow compared
         // to having a hardware option available, even just the sse2 version is

diff --git a/src/f32x8_.rs b/src/f32x8_.rs
@@ -750,17 +750,18 @@ impl f32x8 {
         Self { sse0: round_m128::<{round_op!(Nearest)}>(self.sse0), sse1: round_m128::<{round_op!(Nearest)}>(self.sse1) }
       } else if #[cfg(target_feature="simd128")] {
         Self { simd0: f32x4_nearest(self.simd0), simd1: f32x4_nearest(self.simd1) }
-      } else if #[cfg(feature="std")] {
-        Self { arr: [
-          self.arr[0].round(),
-          self.arr[1].round(),
-          self.arr[2].round(),
-          self.arr[3].round(),
-          self.arr[4].round(),
-          self.arr[5].round(),
-          self.arr[6].round(),
-          self.arr[7].round(),
-        ]}
+      } else if #[cfg(all(feature="std", not(target_arch="wasm32")))] {
+        let data = self.to_array();
+        Self::new([
+          data[0].round(),
+          data[1].round(),
+          data[2].round(),
+          data[3].round(),
+          data[4].round(),
+          data[5].round(),
+          data[6].round(),
+          data[7].round(),
+        ])
       } else {
         // Note(Lokathor): This software fallback is probably very slow compared
         // to having a hardware option available, even just the sse2 version is

diff --git a/src/f64x2_.rs b/src/f64x2_.rs
@@ -588,11 +588,12 @@ impl f64x2 {
         Self { sse: round_m128d::<{round_op!(Nearest)}>(self.sse) }
       } else if #[cfg(target_feature="simd128")] {
         Self { simd: f64x2_nearest(self.simd) }
-      } else if #[cfg(feature="std")] {
-        Self { arr: [
-          self.arr[0].round(),
-          self.arr[1].round(),
-        ]}
+      } else if #[cfg(all(feature="std", not(target_arch="wasm32")))] {
+        let data = self.to_array();
+        Self::new([
+          data[0].round(),
+          data[1].round(),
+        ])
       } else {
         let sign_mask = f64x2::from(-0.0);
         let magic = f64x2::from(f64::from_bits(0x43300000_00000000));

diff --git a/src/f64x4_.rs b/src/f64x4_.rs
@@ -682,13 +682,14 @@ impl f64x4 {
         Self { sse0: round_m128d::<{round_op!(Nearest)}>(self.sse0), sse1: round_m128d::<{round_op!(Nearest)}>(self.sse1) }
       } else if #[cfg(target_feature="simd128")] {
         Self { simd0: f64x2_nearest(self.simd0), simd1: f64x2_nearest(self.simd1) }
-      } else if #[cfg(feature="std")] {
-        Self { arr: [
-          self.arr[0].round(),
-          self.arr[1].round(),
-          self.arr[2].round(),
-          self.arr[3].round(),
-        ]}
+      } else if #[cfg(all(feature="std", not(target_arch="wasm32")))] {
+        let data = self.to_array();
+        Self::new([
+          data[0].round(),
+          data[1].round(),
+          data[2].round(),
+          data[3].round(),
+        ])
       } else {
           let sign_mask = f64x4::from(-0.0);
           let magic = f64x4::from(f64::from_bits(0x43300000_00000000));