Use WebAssembly SIMD instructions

WebAssembly SIMD is almost here. Chrome 91 releases today. It is the
first browser to ship WASM SIMD. Firefox 89 will follow in a bit over a
week as well. Rust also intends to stabilize these instructions very
soon. Until then this PR will stay WIP.

Author: CryZe

src/lib.rs

@@ -28,6 +28,8 @@ See the `examples/` directory for usage examples.
 #![allow(clippy::neg_cmp_op_on_partial_ord)]
 #![allow(clippy::too_many_arguments)]
 
+#![cfg_attr(target_arch = "wasm32", feature(wasm_simd))]
+
 #[cfg(not(any(feature = "std", feature = "libm")))]
 compile_error!("You have to activate either the `std` or the `libm` feature.");
 

src/wide/f32x4_t.rs

@@ -14,6 +14,27 @@ cfg_if::cfg_if! {
         #[derive(Default, Clone, Copy, PartialEq, Debug)]
         #[repr(C, align(16))]
         pub struct f32x4(m128);
+    } else if #[cfg(all(feature = "simd", target_feature = "simd128"))] {
+        use core::arch::wasm32::*;
+
+        // repr(transparent) allows for directly passing the v128 on the WASM stack.
+        #[derive(Clone, Copy, Debug)]
+        #[repr(transparent)]
+        pub struct f32x4(v128);
+
+        impl Default for f32x4 {
+            fn default() -> Self {
+                Self::splat(0.0)
+            }
+        }
+
+        impl PartialEq for f32x4 {
+            fn eq(&self, other: &Self) -> bool {
+                unsafe {
+                    i32x4_all_true(f32x4_eq(self.0, other.0))
+                }
+            }
+        }
     } else {
         #[derive(Default, Clone, Copy, PartialEq, Debug)]
         #[repr(C, align(16))]
@@ -33,6 +54,10 @@ impl f32x4 {
         cfg_if::cfg_if! {
             if #[cfg(all(feature = "simd", target_feature = "sse"))] {
                 Self(max_m128(self.0, rhs.0))
+            } else if #[cfg(all(feature = "simd", target_feature = "simd128"))] {
+                unsafe {
+                    Self(f32x4_max(self.0, rhs.0))
+                }
             } else {
                 Self([
                     self.0[0].max(rhs.0[0]),
@@ -48,6 +73,10 @@ impl f32x4 {
         cfg_if::cfg_if! {
             if #[cfg(all(feature = "simd", target_feature = "sse"))] {
                 Self(min_m128(self.0, rhs.0))
+            } else if #[cfg(all(feature = "simd", target_feature = "simd128"))] {
+                unsafe {
+                    Self(f32x4_min(self.0, rhs.0))
+                }
             } else {
                 Self([
                     self.0[0].min(rhs.0[0]),
@@ -79,6 +108,10 @@ impl core::ops::Add for f32x4 {
         cfg_if::cfg_if! {
             if #[cfg(all(feature = "simd", target_feature = "sse"))] {
                 Self(add_m128(self.0, rhs.0))
+            } else if #[cfg(all(feature = "simd", target_feature = "simd128"))] {
+                unsafe {
+                    Self(f32x4_add(self.0, rhs.0))
+                }
             } else {
                 Self([
                     self.0[0] + rhs.0[0],
@@ -104,6 +137,10 @@ impl core::ops::Sub for f32x4 {
         cfg_if::cfg_if! {
             if #[cfg(all(feature = "simd", target_feature = "sse"))] {
                 Self(sub_m128(self.0, rhs.0))
+            } else if #[cfg(all(feature = "simd", target_feature = "simd128"))] {
+                unsafe {
+                    Self(f32x4_sub(self.0, rhs.0))
+                }
             } else {
                 Self([
                     self.0[0] - rhs.0[0],
@@ -123,6 +160,10 @@ impl core::ops::Mul for f32x4 {
         cfg_if::cfg_if! {
             if #[cfg(all(feature = "simd", target_feature = "sse"))] {
                 Self(mul_m128(self.0, rhs.0))
+            } else if #[cfg(all(feature = "simd", target_feature = "simd128"))] {
+                unsafe {
+                    Self(f32x4_mul(self.0, rhs.0))
+                }
             } else {
                 Self([
                     self.0[0] * rhs.0[0],