glam_det 2.0.0

// Copyright (C) 2020-2025 glam-det authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

// Generated from vec.rs.tera template. Edit the template, not the generated file.

use crate::bool::simd_alias::BVec3A;
use crate::f32::simd_alias::{UnitVec3A, Vec4};
use crate::{Vec2, Vec3};

#[cfg(not(target_arch = "spirv"))]
use core::fmt;
use core::iter::{Product, Sum};
use core::ops::*;

use core::arch::wasm32::*;

use crate::nums::*;

use auto_ops_det::impl_op_ex;
use core::ops;

union UnionCast {
    a: [f32; 4],
    v: Vec3A,
}

/// Creates a 3-dimensional vector.
#[inline]
pub const fn vec3a(x: f32, y: f32, z: f32) -> Vec3A {
    Vec3A::new(x, y, z)
}

/// A 3-dimensional vector with SIMD support.
///
/// This type is 16 byte aligned. A SIMD vector type is used for storage on supported platforms for
/// better performance than the `Vec3` type.
///
/// It is possible to convert between `Vec3` and `Vec3A` types using `From` trait implementations.
#[derive(Clone, Copy)]
#[repr(transparent)]
pub struct Vec3A(pub(crate) v128);

impl Vec3A {
    /// All zeroes.
    pub const ZERO: Self = Self::splat(0.0_f32);

    /// All ones.
    pub const ONE: Self = Self::splat(1.0_f32);

    /// All negative ones.
    pub const NEG_ONE: Self = Self::splat(-1.0_f32);

    /// All NAN.
    pub const NAN: Self = Self::splat(f32::NAN);

    /// A unit-length vector pointing along the positive X axis.
    pub const X: Self = Self::new(1.0_f32, 0.0_f32, 0.0_f32);

    /// A unit-length vector pointing along the positive Y axis.
    pub const Y: Self = Self::new(0.0_f32, 1.0_f32, 0.0_f32);

    /// A unit-length vector pointing along the positive Z axis.
    pub const Z: Self = Self::new(0.0_f32, 0.0_f32, 1.0_f32);

    /// A unit-length vector pointing along the negative X axis.
    pub const NEG_X: Self = Self::new(-1.0_f32, 0.0_f32, 0.0_f32);

    /// A unit-length vector pointing along the negative Y axis.
    pub const NEG_Y: Self = Self::new(0.0_f32, -1.0_f32, 0.0_f32);

    /// A unit-length vector pointing along the negative Z axis.
    pub const NEG_Z: Self = Self::new(0.0_f32, 0.0_f32, -1.0_f32);

    /// The unit axes.
    pub const AXES: [Self; 3] = [Self::X, Self::Y, Self::Z];

    /// Creates a new vector.
    #[inline]
    pub const fn new(x: f32, y: f32, z: f32) -> Self {
        Self(f32x4(x, y, z, z))
    }

    /// Creates a vector with all elements set to `v`.
    #[inline]
    pub const fn splat(v: f32) -> Self {
        unsafe { UnionCast { a: [v; 4] }.v }
    }

    /// Creates a vector from the elements in `if_true` and `if_false`, selecting which to use
    /// for each element of `self`.
    ///
    /// A true element in the mask uses the corresponding element from `if_true`, and false
    /// uses the element from `if_false`.
    #[inline]
    pub fn select(mask: BVec3A, if_true: Self, if_false: Self) -> Self {
        Self(v128_bitselect(if_true.0, if_false.0, mask.0))
    }

    /// Creates a new vector from an array.
    #[inline]
    pub const fn from_array(a: [f32; 3]) -> Self {
        Self::new(a[0], a[1], a[2])
    }

    /// `[x, y, z]`
    #[inline]
    pub const fn to_array(&self) -> [f32; 3] {
        unsafe { *(self as *const Vec3A as *const [f32; 3]) }
    }

    /// Creates a vector from the first 3 values in `slice`.
    ///
    /// # Panics
    ///
    /// Panics if `slice` is less than 3 elements long.
    #[inline]
    pub const fn from_slice(slice: &[f32]) -> Self {
        Self::new(slice[0], slice[1], slice[2])
    }

    /// Writes the elements of `self` to the first 3 elements in `slice`.
    ///
    /// # Panics
    ///
    /// Panics if `slice` is less than 3 elements long.
    #[inline]
    pub fn write_to_slice(self, slice: &mut [f32]) {
        slice[0] = self.x;
        slice[1] = self.y;
        slice[2] = self.z;
    }

    /// Internal method for creating a 3D vector from a 4D vector, discarding `w`.
    #[allow(dead_code)]
    #[inline]
    pub(crate) fn from_vec4(v: Vec4) -> Self {
        Self(v.0)
    }

    /// Creates a 4D vector from `self` and the given `w` value.
    #[inline]
    pub fn extend(self, w: f32) -> Vec4 {
        Vec4::new(self.x, self.y, self.z, w)
    }

    /// Creates a 2D vector from the `x` and `y` elements of `self`, discarding `z`.
    ///
    /// Truncation may also be performed by using `self.xy()` or `Vec2::from()`.
    #[inline]
    pub fn truncate(self) -> Vec2 {
        use crate::swizzles::Vec3Swizzles;
        self.xy()
    }

    /// Computes the dot product of `self` and `rhs`.
    #[inline]
    pub(crate) fn dot(self, rhs: Self) -> f32 {
        crate::wasm32::dot3(self.0, rhs.0)
    }

    /// Computes the cross product of `self` and `rhs`.
    #[inline]
    pub(crate) fn cross(self, rhs: Self) -> Self {
        let lhszxy = i32x4_shuffle::<2, 0, 1, 1>(self.0, self.0);
        let rhszxy = i32x4_shuffle::<2, 0, 1, 1>(rhs.0, rhs.0);
        let lhszxy_rhs = f32x4_mul(lhszxy, rhs.0);
        let rhszxy_lhs = f32x4_mul(rhszxy, self.0);
        let sub = f32x4_sub(lhszxy_rhs, rhszxy_lhs);
        Self(i32x4_shuffle::<2, 0, 1, 1>(sub, sub))
    }

    /// Returns a vector containing the minimum values for each element of `self` and `rhs`.
    ///
    /// In other words this computes `[self.x.min(rhs.x), self.y.min(rhs.y), ..]`.
    #[inline]
    pub fn min(self, rhs: Self) -> Self {
        Self(f32x4_pmin(self.0, rhs.0))
    }

    /// Returns a vector containing the maximum values for each element of `self` and `rhs`.
    ///
    /// In other words this computes `[self.x.max(rhs.x), self.y.max(rhs.y), ..]`.
    #[inline]
    pub fn max(self, rhs: Self) -> Self {
        Self(f32x4_pmax(self.0, rhs.0))
    }

    /// Component-wise clamping of values, similar to [`f32::clamp`].
    ///
    /// Each element in `min` must be less-or-equal to the corresponding element in `max`.
    ///
    /// # Panics
    ///
    /// Will panic if `min` is greater than `max` when `glam_assert` is enabled.
    #[inline]
    pub fn clamp(self, min: Self, max: Self) -> Self {
        glam_assert!(min.cmple(max).all(), "clamp: expected min <= max");
        self.max(min).min(max)
    }

    /// Returns the horizontal minimum of `self`.
    ///
    /// In other words this computes `min(x, y, ..)`.
    #[inline]
    pub fn min_element(self) -> f32 {
        let v = self.0;
        let v = f32x4_pmin(v, i32x4_shuffle::<2, 2, 1, 1>(v, v));
        let v = f32x4_pmin(v, i32x4_shuffle::<1, 0, 0, 0>(v, v));
        f32x4_extract_lane::<0>(v)
    }

    /// Returns the horizontal maximum of `self`.
    ///
    /// In other words this computes `max(x, y, ..)`.
    #[inline]
    pub fn max_element(self) -> f32 {
        let v = self.0;
        let v = f32x4_pmax(v, i32x4_shuffle::<2, 2, 0, 0>(v, v));
        let v = f32x4_pmax(v, i32x4_shuffle::<1, 0, 0, 0>(v, v));
        f32x4_extract_lane::<0>(v)
    }

    /// Returns a vector mask containing the result of a `==` comparison for each element of
    /// `self` and `rhs`.
    ///
    /// In other words, this computes `[self.x == rhs.x, self.y == rhs.y, ..]` for all
    /// elements.
    #[inline]
    pub fn cmpeq(self, rhs: Self) -> BVec3A {
        BVec3A(f32x4_eq(self.0, rhs.0))
    }

    /// Returns a vector mask containing the result of a `!=` comparison for each element of
    /// `self` and `rhs`.
    ///
    /// In other words this computes `[self.x != rhs.x, self.y != rhs.y, ..]` for all
    /// elements.
    #[inline]
    pub fn cmpne(self, rhs: Self) -> BVec3A {
        BVec3A(f32x4_ne(self.0, rhs.0))
    }

    /// Returns a vector mask containing the result of a `>=` comparison for each element of
    /// `self` and `rhs`.
    ///
    /// In other words this computes `[self.x >= rhs.x, self.y >= rhs.y, ..]` for all
    /// elements.
    #[inline]
    pub fn cmpge(self, rhs: Self) -> BVec3A {
        BVec3A(f32x4_ge(self.0, rhs.0))
    }

    /// Returns a vector mask containing the result of a `>` comparison for each element of
    /// `self` and `rhs`.
    ///
    /// In other words this computes `[self.x > rhs.x, self.y > rhs.y, ..]` for all
    /// elements.
    #[inline]
    pub fn cmpgt(self, rhs: Self) -> BVec3A {
        BVec3A(f32x4_gt(self.0, rhs.0))
    }

    /// Returns a vector mask containing the result of a `<=` comparison for each element of
    /// `self` and `rhs`.
    ///
    /// In other words this computes `[self.x <= rhs.x, self.y <= rhs.y, ..]` for all
    /// elements.
    #[inline]
    pub fn cmple(self, rhs: Self) -> BVec3A {
        BVec3A(f32x4_le(self.0, rhs.0))
    }

    /// Returns a vector mask containing the result of a `<` comparison for each element of
    /// `self` and `rhs`.
    ///
    /// In other words this computes `[self.x < rhs.x, self.y < rhs.y, ..]` for all
    /// elements.
    #[inline]
    pub fn cmplt(self, rhs: Self) -> BVec3A {
        BVec3A(f32x4_lt(self.0, rhs.0))
    }

    /// Returns a vector containing the absolute value of each element of `self`.
    #[inline]
    pub fn abs(self) -> Self {
        Self(f32x4_abs(self.0))
    }

    /// Returns a vector with elements representing the sign of `self`.
    ///
    /// - `1.0` if the number is positive, `+0.0` or `INFINITY`
    /// - `-1.0` if the number is negative, `-0.0` or `NEG_INFINITY`
    /// - `NAN` if the number is `NAN`
    ///
    /// # Warning
    ///
    /// Because of the fact that `-Vec3A::ZERO` output `Vec3A::ZERO`,
    /// so the sign of `-Vec3A::ZERO` is `1.0`, which is different from the behavior of `std::f32`.
    /// This phenomenon exists if some of vector elements is zero.
    #[inline]
    pub fn signum(self) -> Self {
        let mask = self.cmpge(Self::ZERO);
        let result = Self::select(mask, Self::ONE, Self::NEG_ONE);
        let mask = self.is_nan_mask();
        Self::select(mask, self, result)
    }

    /// Returns `true` if, and only if, all elements are finite.  If any element is either
    /// `NaN`, positive or negative infinity, this will return `false`.
    #[inline]
    pub fn is_finite(self) -> bool {
        self.x.is_finite() && self.y.is_finite() && self.z.is_finite()
    }

    /// Returns `true` if any elements are `NaN`.
    #[inline]
    pub fn is_nan(self) -> bool {
        self.is_nan_mask().any()
    }

    /// Performs `is_nan` on each element of self, returning a vector mask of the results.
    ///
    /// In other words, this computes `[x.is_nan(), y.is_nan(), z.is_nan(), w.is_nan()]`.
    #[inline]
    pub fn is_nan_mask(self) -> BVec3A {
        BVec3A(f32x4_ne(self.0, self.0))
    }

    /// Returns a vector containing the nearest integer to a number for each element of `self`.
    /// Round half-way cases away from 0.0.
    #[inline]
    pub fn round(self) -> Self {
        Self(f32x4_nearest(self.0))
    }

    /// Returns a vector containing the largest integer less than or equal to a number for each
    /// element of `self`.
    #[inline]
    pub fn floor(self) -> Self {
        Self(f32x4_floor(self.0))
    }

    /// Returns a vector containing the smallest integer greater than or equal to a number for
    /// each element of `self`.
    #[inline]
    pub fn ceil(self) -> Self {
        Self(f32x4_ceil(self.0))
    }

    /// Returns a vector containing the fractional part of the vector, e.g. `self -
    /// self.floor()`.
    ///
    /// Note that this is fast but not precise for large numbers.
    #[inline]
    pub fn fract(self) -> Self {
        self - self.floor()
    }

    /// Returns a vector containing `e^self` (the exponential function) for each element of
    /// `self`.
    #[inline]
    pub fn exp(self) -> Self {
        Self::new(self.x.expf(), self.y.expf(), self.z.expf())
    }

    /// Returns a vector containing each element of `self` raised to the power of `n`.
    #[inline]
    pub fn powf(self, n: f32) -> Self {
        Self::new(self.x.powff(n), self.y.powff(n), self.z.powff(n))
    }

    /// Returns a vector containing the reciprocal `1.0/n` of each element of `self`.
    #[inline]
    pub fn recip(self) -> Self {
        Self(f32x4_div(Self::ONE.0, self.0))
    }

    /// Computes the length of `self`.
    #[doc(alias = "magnitude")]
    #[inline]
    pub fn length(self) -> f32 {
        self.dot(self).sqrtf()
    }

    /// Computes the squared length of `self`.
    ///
    /// This is faster than `length()` as it avoids a square root operation.
    #[doc(alias = "magnitude2")]
    #[inline]
    pub fn length_squared(self) -> f32 {
        self.dot(self)
    }

    /// Computes `1.0 / length()`.
    ///
    /// For valid results, `self` must _not_ be of length zero.
    #[inline]
    pub fn length_recip(self) -> f32 {
        self.length().recip()
    }

    /// Computes the Euclidean distance between two points in space.
    #[inline]
    pub fn distance(self, rhs: Self) -> f32 {
        (self - rhs).length()
    }

    /// Compute the squared euclidean distance between two points in space.
    #[inline]
    pub fn distance_squared(self, rhs: Self) -> f32 {
        (self - rhs).length_squared()
    }

    /// Returns `self` normalized to length 1.0.
    ///
    /// For valid results, `self` must _not_ be of length zero or infinite, nor very close to zero.
    ///
    /// See also [`Self::try_normalize`] and [`Self::normalize_or_zero`].
    ///
    /// # Panics
    ///
    /// Will panic if `self` is zero or infinite length when `glam_assert` is enabled.
    #[must_use]
    #[inline]
    pub fn normalize(self) -> Self {
        let length_wide = f32x4_sqrt(crate::wasm32::dot3_into_v128(self.0, self.0));
        #[cfg(any(
            all(debug_assertions, feature = "debug-glam-assert"),
            feature = "glam-assert"
        ))]
        let length = f32x4_extract_lane::<0>(length_wide);
        glam_assert!(length.gt(&0.0_f32), "Trying to normalize {:?}", self);
        glam_assert!(self.is_finite(), "Trying to normalize {:?}", self);
        glam_assert!(length.is_finite(), "Trying to normalize {:?}", self);
        #[allow(clippy::let_and_return)]
        let normalized = Self(f32x4_div(self.0, length_wide));
        glam_assert!(normalized.is_finite(), "Trying to normalize {:?}", self);
        normalized
    }

    /// Returns `self` normalized to length 1.0  and length of `self`
    ///
    /// For valid results, `self` must _not_ be of length zero or infinite, nor very close to zero.
    ///
    /// See also [`Self::normalize`].
    ///
    /// # Panics
    ///
    /// Will panic if `self` is zero or infinite length when `glam_assert` is enabled.
    #[must_use]
    #[inline]
    pub fn normalize_and_length(self) -> (Self, f32) {
        let length_wide = f32x4_sqrt(crate::wasm32::dot3_into_v128(self.0, self.0));
        let length = f32x4_extract_lane::<0>(length_wide);
        glam_assert!(length.gt(&0.0_f32), "Trying to normalize {:?}", self);
        glam_assert!(self.is_finite(), "Trying to normalize {:?}", self);
        glam_assert!(length.is_finite(), "Trying to normalize {:?}", self);
        #[allow(clippy::let_and_return)]
        let normalized = Self(f32x4_div(self.0, length_wide));
        glam_assert!(normalized.is_finite(), "Trying to normalize {:?}", self);
        (normalized, length)
    }

    /// Returns `self` normalized to a unit vector.
    ///
    /// For valid results, `self` must _not_ be of length zero or infinite, nor very close to zero.
    ///
    /// See also [`Self::normalize`].
    ///
    /// # Panics
    ///
    /// Will panic if `self` is zero or infinite length when `glam_assert` is enabled.
    #[must_use]
    #[inline]
    pub fn normalize_to_unit(self) -> UnitVec3A {
        self.normalize().as_unit_vec3a_unchecked()
    }

    /// Returns `self` normalized to a unit vector and length of `self`.
    ///
    /// For valid results, `self` must _not_ be of length zero, nor very close to zero.
    ///
    /// See also [`Self::normalize_to_unit`].
    ///
    /// # Panics
    ///
    /// Will panic if `self` is zero length when `glam_assert` is enabled.
    #[must_use]
    #[inline]
    pub fn normalize_to_unit_and_length(self) -> (UnitVec3A, f32) {
        let res = self.normalize_and_length();
        (res.0.as_unit_vec3a_unchecked(), res.1)
    }

    /// Returns `self` normalized to length 1.0 if length is greater then "min_len"
    /// and is finite, else returns `None`.
    ///
    /// See also [`Self::normalize_or_zero`].
    #[must_use]
    #[inline]
    pub fn try_normalize(self, min_len: f32) -> Option<Self> {
        let length = self.length();
        if length.is_finite() && length > min_len {
            Some(self * length.recip())
        } else {
            None
        }
    }

    /// Returns `self` normalized to unit vector if length is greater then "min_len"
    /// and is finite, else returns `None`.
    ///
    /// See also [`Self::try_normalize`].
    #[must_use]
    #[inline]
    pub fn try_normalize_to_unit(self, min_len: f32) -> Option<UnitVec3A> {
        let length = self.length();
        if length.is_finite() && length > min_len {
            Some((self * length.recip()).as_unit_vec3a_unchecked())
        } else {
            None
        }
    }

    /// Returns `self` normalized to unit vector and length of `self` if length is greater then "min_len"
    /// and is finite, else returns `None`.
    ///
    /// See also [`Self::try_normalize_to_unit`].
    #[must_use]
    #[inline]
    pub fn try_normalize_to_unit_and_length(self, min_len: f32) -> Option<(UnitVec3A, f32)> {
        let length = self.length();
        if length.is_finite() && length > min_len {
            Some(((self * length.recip()).as_unit_vec3a_unchecked(), length))
        } else {
            None
        }
    }

    /// Returns `self` normalized to length 1.0 if possible, else returns zero.
    ///
    /// In particular, if the input is zero (or very close to zero), or non-finite,
    /// the result of this operation will be zero.
    ///
    /// See also [`Self::try_normalize`].
    #[must_use]
    #[inline]
    pub fn normalize_or_zero(self) -> Self {
        let rcp = self.length_recip();
        if rcp.is_finite() && rcp > 0.0 {
            self * rcp
        } else {
            Self::ZERO
        }
    }

    /// Returns whether `self` is length `1.0` or not.
    ///
    /// Uses a precision threshold of `1e-6`.
    #[inline]
    pub fn is_normalized(self) -> bool {
        // TODO: do something with epsilon
        (self.length_squared() - 1.0_f32).absf() <= 1e-4
    }

    /// Returns the vector projection of `self` onto `rhs`.
    ///
    /// `rhs` must be of non-zero length.
    ///
    /// # Panics
    ///
    /// Will panic if `rhs` is zero length when `glam_assert` is enabled.
    #[must_use]
    #[inline]
    pub fn project_onto(self, rhs: Self) -> Self {
        let other_len_sq_rcp = rhs.dot(rhs).recip();
        glam_assert!(
            other_len_sq_rcp.is_finite(),
            "Trying to project onto infinite rhs = {:?}",
            rhs
        );
        rhs * self.dot(rhs) * other_len_sq_rcp
    }

    /// Returns the vector rejection of `self` from `rhs`.
    ///
    /// The vector rejection is the vector perpendicular to the projection of `self` onto
    /// `rhs`, in rhs words the result of `self - self.project_onto(rhs)`.
    ///
    /// `rhs` must be of non-zero length.
    ///
    /// # Panics
    ///
    /// Will panic if `rhs` has a length of zero when `glam_assert` is enabled.
    #[must_use]
    #[inline]
    pub fn reject_from(self, rhs: Self) -> Self {
        self - self.project_onto(rhs)
    }

    /// Returns the vector projection of `self` onto `rhs`.
    #[must_use]
    #[inline]
    pub fn project_onto_normalized(self, rhs: UnitVec3A) -> Self {
        rhs.as_vec3a() * self.dot(rhs.as_vec3a())
    }

    /// Returns the vector rejection of `self` from `rhs`.
    #[must_use]
    #[inline]
    pub fn reject_from_normalized(self, rhs: UnitVec3A) -> Self {
        self - self.project_onto_normalized(rhs)
    }

    /// Performs a linear interpolation between `self` and `rhs` based on the value `s`.
    ///
    /// When `s` is `0.0`, the result will be equal to `self`.  When `s` is `1.0`, the result
    /// will be equal to `rhs`. When `s` is outside of range `[0, 1]`, the result is linearly
    /// extrapolated.
    #[doc(alias = "mix")]
    #[inline]
    pub fn lerp(self, rhs: Self, s: f32) -> Self {
        self + ((rhs - self) * s)
    }

    /// Returns true if the absolute difference of all elements between `self` and `rhs` is
    /// less than or equal to `max_abs_diff`.
    ///
    /// This can be used to compare if two vectors contain similar elements. It works best when
    /// comparing with a known value. The `max_abs_diff` that should be used depends on
    /// the values being compared against.
    ///
    /// For more see
    /// [comparing floating point numbers](https://randomascii.wordpress.com/2012/02/25/comparing-floating-point-numbers-2012-edition/).
    #[inline]
    pub fn abs_diff_eq(self, rhs: Self, max_abs_diff: f32) -> bool {
        self.sub(rhs).abs().cmple(Self::splat(max_abs_diff)).all()
    }

    /// Returns a vector with a length no less than `min` and no more than `max`
    ///
    /// # Panics
    ///
    /// Will panic if `min` is greater than `max` when `glam_assert` is enabled.
    #[inline]
    pub fn clamp_length(self, min: f32, max: f32) -> Self {
        glam_assert!(min <= max);
        let length_sq = self.length_squared();
        if length_sq < min * min {
            self * (length_sq.sqrtf().recip() * min)
        } else if length_sq > max * max {
            self * (length_sq.sqrtf().recip() * max)
        } else {
            self
        }
    }

    /// Returns a vector with a length no more than `max`
    #[inline]
    pub fn clamp_length_max(self, max: f32) -> Self {
        let length_sq = self.length_squared();
        if length_sq > max * max {
            self * (length_sq.sqrtf().recip() * max)
        } else {
            self
        }
    }

    /// Returns a vector with a length no less than `min`
    #[inline]
    pub fn clamp_length_min(self, min: f32) -> Self {
        let length_sq = self.length_squared();
        if length_sq < min * min {
            self * (length_sq.sqrtf().recip() * min)
        } else {
            self
        }
    }

    /// Fused multiply-add. Computes `(self * a) + b` element-wise with only one rounding
    /// error, yielding a more accurate result than an unfused multiply-add.
    ///
    /// Using `mul_add` *may* be more performant than an unfused multiply-add if the target
    /// architecture has a dedicated fma CPU instruction. However, this is not always true,
    /// and will be heavily dependant on designing algorithms with specific target hardware in
    /// mind.
    #[inline]
    pub fn mul_add(self, a: Self, b: Self) -> Self {
        Self::new(
            self.x.mul_addf(a.x, b.x),
            self.y.mul_addf(a.y, b.y),
            self.z.mul_addf(a.z, b.z),
        )
    }

    /// Returns the angle (in radians) between two vectors.
    ///
    /// The input vectors do not need to be unit length however they must be non-zero.
    #[inline]
    pub fn angle_between(self, rhs: Self) -> f32 {
        use crate::FloatEx;
        self.dot(rhs)
            .div(self.length_squared().mul(rhs.length_squared()).sqrtf())
            .acos_approx()
    }

    /// Returns some vector that is orthogonal to the given one.
    ///
    /// The input vector must be finite and non-zero.
    #[inline]
    pub fn any_orthogonal_vector(&self) -> Self {
        // This can probably be optimized
        if self.x.absf() > self.y.absf() {
            Self::new(-self.z, 0.0, self.x) // self.cross(Self::Y)
        } else {
            Self::new(0.0, self.z, -self.y) // self.cross(Self::X)
        }
    }

    /// Casts all elements of `self` to `f64`.
    #[inline]
    pub fn as_dvec3(&self) -> crate::DVec3 {
        crate::DVec3::new(self.x as f64, self.y as f64, self.z as f64)
    }

    /// Casts all elements of `self` to `i32`.
    #[inline]
    pub fn as_ivec3(&self) -> crate::IVec3 {
        crate::IVec3::new(self.x as i32, self.y as i32, self.z as i32)
    }

    /// Casts all elements of `self` to `u32`.
    #[inline]
    pub fn as_uvec3(&self) -> crate::UVec3 {
        crate::UVec3::new(self.x as u32, self.y as u32, self.z as u32)
    }
}

impl Default for Vec3A {
    #[inline]
    fn default() -> Self {
        Self::ZERO
    }
}

impl PartialEq for Vec3A {
    #[inline]
    fn eq(&self, rhs: &Self) -> bool {
        self.cmpeq(*rhs).all()
    }
}

impl_op_ex!(/ |a: &Vec3A, b: &Vec3A| -> Vec3A {
            Vec3A(f32x4_div(a.0, b.0))

});

impl_op_ex!(/= |a: &mut Vec3A, b: &Vec3A| {
            a.0 = f32x4_div(a.0, b.0);
});

impl_op_ex!(/ |a: &Vec3A, b: &f32| -> Vec3A {
            Vec3A(f32x4_div(a.0, f32x4_splat(*b)))
});

impl_op_ex!(/= |a: &mut Vec3A, b: &f32| {
            a.0 = f32x4_div(a.0, f32x4_splat(*b));

});

impl_op_ex!(/ |a: &f32, b: &Vec3A| -> Vec3A {
            Vec3A(f32x4_div(f32x4_splat(*a), b.0))
});

impl_op_ex!(*|a: &Vec3A, b: &Vec3A| -> Vec3A { Vec3A(f32x4_mul(a.0, b.0)) });

impl_op_ex!(*= |a: &mut Vec3A, b: &Vec3A| {
            a.0 = f32x4_mul(a.0, b.0);
});

impl_op_ex!(*|a: &Vec3A, b: &f32| -> Vec3A { Vec3A(f32x4_mul(a.0, f32x4_splat(*b))) });

impl_op_ex!(*= |a: &mut Vec3A, b: &f32| {
            a.0 = f32x4_mul(a.0, f32x4_splat(*b));

});

impl_op_ex!(*|a: &f32, b: &Vec3A| -> Vec3A { Vec3A(f32x4_mul(f32x4_splat(*a), b.0)) });

impl_op_ex!(+ |a: &Vec3A, b: &Vec3A| -> Vec3A {
            Vec3A(f32x4_add(a.0, b.0))

});

impl_op_ex!(+= |a: &mut Vec3A, b: &Vec3A| {
            a.0 = f32x4_add(a.0, b.0);
});

impl_op_ex!(+ |a: &Vec3A, b: &f32| -> Vec3A {
            Vec3A(f32x4_add(a.0, f32x4_splat(*b)))
});

impl_op_ex!(+= |a: &mut Vec3A, b: &f32| {
            a.0 = f32x4_add(a.0, f32x4_splat(*b));

});

impl_op_ex!(+ |a: &f32, b: &Vec3A| -> Vec3A {
            Vec3A(f32x4_add(f32x4_splat(*a), b.0))
});

impl_op_ex!(-|a: &Vec3A, b: &Vec3A| -> Vec3A { Vec3A(f32x4_sub(a.0, b.0)) });

impl_op_ex!(-= |a: &mut Vec3A, b: &Vec3A| {
            a.0 = f32x4_sub(a.0, b.0);
});

impl_op_ex!(-|a: &Vec3A, b: &f32| -> Vec3A { Vec3A(f32x4_sub(a.0, f32x4_splat(*b))) });

impl_op_ex!(-= |a: &mut Vec3A, b: &f32| {
            a.0 = f32x4_sub(a.0, f32x4_splat(*b));

});

impl_op_ex!(-|a: &f32, b: &Vec3A| -> Vec3A { Vec3A(f32x4_sub(f32x4_splat(*a), b.0)) });

impl_op_ex!(% |a: &Vec3A, b: &Vec3A| -> Vec3A {
            let n = f32x4_floor(f32x4_div(a.0, b.0));
            Vec3A(f32x4_sub(a.0, f32x4_mul(n, b.0)))
});

impl_op_ex!(%= |a: &mut Vec3A, b: &Vec3A| {
            *a = a.rem(*b)
});

impl_op_ex!(% |a: &Vec3A, b: &f32| -> Vec3A {
            a.rem(Vec3A::splat(*b))
});

impl_op_ex!(%= |a: &mut Vec3A, b: &f32| {
            *a = a.rem(Vec3A::splat(*b))
});

impl_op_ex!(% |a: &f32, b: &Vec3A| -> Vec3A {
            Vec3A::splat(*a).rem(b)
});

impl<'a> Sum<&'a Self> for Vec3A {
    #[inline]
    fn sum<I>(iter: I) -> Self
    where
        I: Iterator<Item = &'a Self>,
    {
        iter.fold(Self::ZERO, |a, &b| Self::add(a, b))
    }
}

impl<'a> Product<&'a Self> for Vec3A {
    #[inline]
    fn product<I>(iter: I) -> Self
    where
        I: Iterator<Item = &'a Self>,
    {
        iter.fold(Self::ONE, |a, &b| Self::mul(a, b))
    }
}

impl Neg for Vec3A {
    type Output = Self;
    #[inline]
    fn neg(self) -> Self {
        Self(f32x4_neg(self.0))
    }
}

impl Index<usize> for Vec3A {
    type Output = f32;
    #[inline]
    fn index(&self, index: usize) -> &Self::Output {
        match index {
            0 => &self.x,
            1 => &self.y,
            2 => &self.z,
            _ => panic!("index out of bounds"),
        }
    }
}

impl IndexMut<usize> for Vec3A {
    #[inline]
    fn index_mut(&mut self, index: usize) -> &mut Self::Output {
        match index {
            0 => &mut self.x,
            1 => &mut self.y,
            2 => &mut self.z,
            _ => panic!("index out of bounds"),
        }
    }
}

#[cfg(not(target_arch = "spirv"))]
impl fmt::Display for Vec3A {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        write!(f, "[{}, {}, {}]", self.x, self.y, self.z)
    }
}

#[cfg(not(target_arch = "spirv"))]
impl fmt::Debug for Vec3A {
    fn fmt(&self, fmt: &mut fmt::Formatter<'_>) -> fmt::Result {
        fmt.debug_tuple(stringify!(Vec3A))
            .field(&self.x)
            .field(&self.y)
            .field(&self.z)
            .finish()
    }
}

impl From<Vec3A> for v128 {
    #[inline]
    fn from(t: Vec3A) -> Self {
        t.0
    }
}

impl From<v128> for Vec3A {
    #[inline]
    fn from(t: v128) -> Self {
        Self(t)
    }
}

impl From<[f32; 3]> for Vec3A {
    #[inline]
    fn from(a: [f32; 3]) -> Self {
        Self::new(a[0], a[1], a[2])
    }
}

impl From<Vec3A> for [f32; 3] {
    #[inline]
    fn from(v: Vec3A) -> Self {
        unsafe { *(&v.0 as *const v128 as *const [f32; 3]) }
    }
}

impl From<(f32, f32, f32)> for Vec3A {
    #[inline]
    fn from(t: (f32, f32, f32)) -> Self {
        Self::new(t.0, t.1, t.2)
    }
}

impl From<Vec3A> for (f32, f32, f32) {
    #[inline]
    fn from(v: Vec3A) -> Self {
        unsafe { *(&v.0 as *const v128 as *const Self) }
    }
}

impl From<Vec3> for Vec3A {
    #[inline]
    fn from(v: Vec3) -> Self {
        Self::new(v.x, v.y, v.z)
    }
}

impl From<Vec4> for Vec3A {
    /// Creates a `Vec3A` from the `x`, `y` and `z` elements of `self` discarding `w`.
    ///
    /// On architectures where SIMD is supported such as SSE2 on `x86_64` this conversion is a noop.
    #[inline]
    fn from(v: Vec4) -> Self {
        Self(v.0)
    }
}

impl From<Vec3A> for Vec3 {
    #[inline]
    fn from(v: Vec3A) -> Self {
        unsafe { *(&v.0 as *const v128 as *const Self) }
    }
}

impl From<(Vec2, f32)> for Vec3A {
    #[inline]
    fn from((v, z): (Vec2, f32)) -> Self {
        Self::new(v.x, v.y, z)
    }
}

impl Deref for Vec3A {
    type Target = crate::deref::Vec3<f32>;
    #[inline]
    fn deref(&self) -> &Self::Target {
        unsafe { &*(self as *const Self).cast() }
    }
}

impl DerefMut for Vec3A {
    #[inline]
    fn deref_mut(&mut self) -> &mut Self::Target {
        unsafe { &mut *(self as *mut Self).cast() }
    }
}

#[cfg(not(target_arch = "spirv"))]
impl AsRef<[f32; 3]> for Vec3A {
    #[inline]
    fn as_ref(&self) -> &[f32; 3] {
        unsafe { &*(self as *const Vec3A as *const [f32; 3]) }
    }
}

#[cfg(not(target_arch = "spirv"))]
impl AsMut<[f32; 3]> for Vec3A {
    #[inline]
    fn as_mut(&mut self) -> &mut [f32; 3] {
        unsafe { &mut *(self as *mut Vec3A as *mut [f32; 3]) }
    }
}