NEON fixes.

.github/workflows/ci.yml

@@ -97,7 +97,7 @@ jobs:
           - aarch64-unknown-linux-gnu
           - arm-unknown-linux-gnueabi
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
           toolchain: stable

Cargo.toml

@@ -56,14 +56,12 @@ libm = { version = "0.2", optional = true, default-features = false}
 #rand_xoshiro = "0.6"
 #serde_json = "1.0"
 
-[target.'cfg(target_arch = "x86_64")'.dev-dependencies]
-criterion = { version = "0.4", features = ["html_reports"] }
 rand_xoshiro = "0.6"
 # Set a size_xx feature so that this crate compiles properly with --all-targets --all-features
 rkyv = { version = "0.7", default-features = false, features = ["size_32"] }
 serde_json = "1.0"
 
-[target.'cfg(not(target_arch = "wasm32"))'.dev-dependencies]
+[target.'cfg(target_arch = "x86_64")'.dev-dependencies]
 criterion = { version = "0.5", features = ["html_reports"] }
 
 [target.'cfg(target_arch = "wasm32")'.dev-dependencies]

codegen/templates/vec.rs.tera

@@ -823,7 +823,7 @@ impl {{ self_t }} {
     #[inline]
     #[must_use]
     pub fn element_sum(self) -> {{ scalar_t }} {
-        {% if is_scalar %}
+        {% if is_scalar or is_neon %}
             {% for c in components %}
                 self.{{ c }} {% if not loop.last %} + {% endif %}
             {%- endfor %}
@@ -861,6 +861,8 @@ impl {{ self_t }} {
             {% elif dim == 4 %}
                 self.0.reduce_sum()
             {% endif %}
+        {% else %}
+            unimplemented!()
         {% endif %}
     }
 
@@ -870,7 +872,7 @@ impl {{ self_t }} {
     #[inline]
     #[must_use]
     pub fn element_product(self) -> {{ scalar_t }} {
-        {% if is_scalar %}
+        {% if is_scalar or is_neon %}
             {% for c in components %}
                 self.{{ c }} {% if not loop.last %} * {% endif %}
             {%- endfor %}
@@ -908,6 +910,8 @@ impl {{ self_t }} {
             {% elif dim == 4 %}
                 self.0.reduce_product()
             {% endif %}
+        {% else %}
+            unimplemented!()
         {% endif %}
     }
 
@@ -1653,6 +1657,10 @@ impl {{ self_t }} {
             Self(f32x4_trunc(self.0))
         {% elif is_coresimd %}
             Self(self.0.trunc())
+        {% elif is_neon %}
+            Self(unsafe { vrndq_f32(self.0) })
+        {% else %}
+            unimplemented!()
         {% endif %}
     }
 

codegen/templates/vec_mask.rs.tera

@@ -298,11 +298,20 @@ impl {{ self_t }} {
             }
         {% elif is_coresimd %}
             self.0.set(index, value)
+        {% elif is_neon %}
+            self.0 = match index { 
+                {% for c in components %}
+                    {{ loop.index0 }} => unsafe {
+                        vsetq_lane_u32(MASK[value as usize], self.0, {{ loop.index0 }})
+                    },
+                {%- endfor %}
+                _ => panic!("index out of bounds")
+            }
         {% else %}
             use crate::{{ vec_t }};
             let mut v = {{ vec_t }}(self.0);
             v[index] = f32::from_bits(MASK[value as usize]);
-            *self = Self(v.0);
+            self.0 = v.0;
         {% endif %}
     }
 

src/bool/neon/bvec3a.rs

@@ -106,10 +106,12 @@ impl BVec3A {
     /// Panics if `index` is greater than 2.
     #[inline]
     pub fn set(&mut self, index: usize, value: bool) {
-        use crate::Vec3A;
-        let mut v = Vec3A(self.0);
-        v[index] = f32::from_bits(MASK[value as usize]);
-        *self = Self(v.0);
+        self.0 = match index {
+            0 => unsafe { vsetq_lane_u32(MASK[value as usize], self.0, 0) },
+            1 => unsafe { vsetq_lane_u32(MASK[value as usize], self.0, 1) },
+            2 => unsafe { vsetq_lane_u32(MASK[value as usize], self.0, 2) },
+            _ => panic!("index out of bounds"),
+        }
     }
 
     #[inline]

src/bool/neon/bvec4a.rs

@@ -112,10 +112,13 @@ impl BVec4A {
     /// Panics if `index` is greater than 3.
     #[inline]
     pub fn set(&mut self, index: usize, value: bool) {
-        use crate::Vec4;
-        let mut v = Vec4(self.0);
-        v[index] = f32::from_bits(MASK[value as usize]);
-        *self = Self(v.0);
+        self.0 = match index {
+            0 => unsafe { vsetq_lane_u32(MASK[value as usize], self.0, 0) },
+            1 => unsafe { vsetq_lane_u32(MASK[value as usize], self.0, 1) },
+            2 => unsafe { vsetq_lane_u32(MASK[value as usize], self.0, 2) },
+            3 => unsafe { vsetq_lane_u32(MASK[value as usize], self.0, 3) },
+            _ => panic!("index out of bounds"),
+        }
     }
 
     #[inline]

src/bool/sse2/bvec3a.rs

@@ -103,7 +103,7 @@ impl BVec3A {
         use crate::Vec3A;
         let mut v = Vec3A(self.0);
         v[index] = f32::from_bits(MASK[value as usize]);
-        *self = Self(v.0);
+        self.0 = v.0;
     }
 
     #[inline]

src/bool/sse2/bvec4a.rs

@@ -109,7 +109,7 @@ impl BVec4A {
         use crate::Vec4;
         let mut v = Vec4(self.0);
         v[index] = f32::from_bits(MASK[value as usize]);
-        *self = Self(v.0);
+        self.0 = v.0;
     }
 
     #[inline]

src/bool/wasm32/bvec3a.rs

@@ -94,7 +94,7 @@ impl BVec3A {
         use crate::Vec3A;
         let mut v = Vec3A(self.0);
         v[index] = f32::from_bits(MASK[value as usize]);
-        *self = Self(v.0);
+        self.0 = v.0;
     }
 
     #[inline]

src/bool/wasm32/bvec4a.rs

@@ -95,7 +95,7 @@ impl BVec4A {
         use crate::Vec4;
         let mut v = Vec4(self.0);
         v[index] = f32::from_bits(MASK[value as usize]);
-        *self = Self(v.0);
+        self.0 = v.0;
     }
 
     #[inline]

src/f32/neon/vec3a.rs

@@ -272,14 +272,18 @@ impl Vec3A {
     /// In other words, this computes `self.x + self.y + ..`.
     #[inline]
     #[must_use]
-    pub fn element_sum(self) -> f32 {}
+    pub fn element_sum(self) -> f32 {
+        self.x + self.y + self.z
+    }
 
     /// Returns the product of all elements of `self`.
     ///
     /// In other words, this computes `self.x * self.y * ..`.
     #[inline]
     #[must_use]
-    pub fn element_product(self) -> f32 {}
+    pub fn element_product(self) -> f32 {
+        self.x * self.y * self.z
+    }
 
     /// Returns a vector mask containing the result of a `==` comparison for each element of
     /// `self` and `rhs`.
@@ -664,7 +668,9 @@ impl Vec3A {
     /// always truncated towards zero.
     #[inline]
     #[must_use]
-    pub fn trunc(self) -> Self {}
+    pub fn trunc(self) -> Self {
+        Self(unsafe { vrndq_f32(self.0) })
+    }
 
     /// Returns a vector containing the fractional part of the vector as `self - self.trunc()`.
     ///

src/f32/neon/vec4.rs

@@ -259,14 +259,18 @@ impl Vec4 {
     /// In other words, this computes `self.x + self.y + ..`.
     #[inline]
     #[must_use]
-    pub fn element_sum(self) -> f32 {}
+    pub fn element_sum(self) -> f32 {
+        self.x + self.y + self.z + self.w
+    }
 
     /// Returns the product of all elements of `self`.
     ///
     /// In other words, this computes `self.x * self.y * ..`.
     #[inline]
     #[must_use]
-    pub fn element_product(self) -> f32 {}
+    pub fn element_product(self) -> f32 {
+        self.x * self.y * self.z * self.w
+    }
 
     /// Returns a vector mask containing the result of a `==` comparison for each element of
     /// `self` and `rhs`.
@@ -654,7 +658,9 @@ impl Vec4 {
     /// always truncated towards zero.
     #[inline]
     #[must_use]
-    pub fn trunc(self) -> Self {}
+    pub fn trunc(self) -> Self {
+        Self(unsafe { vrndq_f32(self.0) })
+    }
 
     /// Returns a vector containing the fractional part of the vector as `self - self.trunc()`.
     ///