glam_det 2.0.0-beta.2

// 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::nums::*;
use crate::{BVec4x4, UnitVec4x4, Vec2x4, Vec3x4};

use crate::Vec4;

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

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

/// Creates a 4-dimensional 4 lanes SOA vector.
#[inline]
pub const fn vec4x4(x: f32x4, y: f32x4, z: f32x4, w: f32x4) -> Vec4x4 {
    Vec4x4::new(x, y, z, w)
}

/// A 4-dimensional 4 lanes SOA vector.
/// This type is 16 byte aligned, same as f32x4.
#[derive(Clone, Copy, PartialEq)]
#[cfg_attr(feature = "cuda", repr(align(16)))]
#[repr(C)]
pub struct Vec4x4 {
    pub x: f32x4,
    pub y: f32x4,
    pub z: f32x4,
    pub w: f32x4,
}

impl Vec4x4 {
    /// All zeroes.
    pub const ZERO: Self = Self::splat(f32x4::ZERO);

    /// All ones.
    pub const ONE: Self = Self::splat(f32x4::ONE);

    /// All negative ones.
    pub const NEG_ONE: Self = Self::splat(f32x4::NEG_ONE);

    /// All NAN.
    pub const NAN: Self = Self::splat(f32x4::NAN);

    /// A unit-length vector pointing along the positive X axis.
    pub const X: Self = Self::new(f32x4::ONE, f32x4::ZERO, f32x4::ZERO, f32x4::ZERO);

    /// A unit-length vector pointing along the positive Y axis.
    pub const Y: Self = Self::new(f32x4::ZERO, f32x4::ONE, f32x4::ZERO, f32x4::ZERO);

    /// A unit-length vector pointing along the positive Z axis.
    pub const Z: Self = Self::new(f32x4::ZERO, f32x4::ZERO, f32x4::ONE, f32x4::ZERO);

    /// A unit-length vector pointing along the positive W axis.
    pub const W: Self = Self::new(f32x4::ZERO, f32x4::ZERO, f32x4::ZERO, f32x4::ONE);

    /// A unit-length vector pointing along the negative X axis.
    pub const NEG_X: Self = Self::new(f32x4::NEG_ONE, f32x4::ZERO, f32x4::ZERO, f32x4::ZERO);

    /// A unit-length vector pointing along the negative Y axis.
    pub const NEG_Y: Self = Self::new(f32x4::ZERO, f32x4::NEG_ONE, f32x4::ZERO, f32x4::ZERO);

    /// A unit-length vector pointing along the negative Z axis.
    pub const NEG_Z: Self = Self::new(f32x4::ZERO, f32x4::ZERO, f32x4::NEG_ONE, f32x4::ZERO);

    /// A unit-length vector pointing along the negative W axis.
    pub const NEG_W: Self = Self::new(f32x4::ZERO, f32x4::ZERO, f32x4::ZERO, f32x4::NEG_ONE);

    /// The unit axes.
    pub const AXES: [Self; 4] = [Self::X, Self::Y, Self::Z, Self::W];

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

    /// Creates a vector with all elements set to `v`.
    #[inline]
    pub const fn splat(v: f32x4) -> Self {
        Self {
            x: v,
            y: v,
            z: v,
            w: 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: BVec4x4, if_true: Self, if_false: Self) -> Self {
        Self {
            x: if_true.x.select(mask.x, if_false.x),
            y: if_true.y.select(mask.y, if_false.y),
            z: if_true.z.select(mask.z, if_false.z),
            w: if_true.w.select(mask.w, if_false.w),
        }
    }

    /// Creates a new vector from an array.
    #[inline]
    pub const fn from_array(a: [f32x4; 4]) -> Self {
        Self::new(a[0], a[1], a[2], a[3])
    }

    /// `[x, y, z, w]`
    #[inline]
    pub const fn to_array(&self) -> [f32x4; 4] {
        [self.x, self.y, self.z, self.w]
    }

    /// Creates a vector from the first 4 values in `slice`.
    ///
    /// # Panics
    ///
    /// Panics if `slice` is less than 4 elements long.
    #[inline]
    pub const fn from_slice(slice: &[f32x4]) -> Self {
        Self::new(slice[0], slice[1], slice[2], slice[3])
    }

    /// Writes the elements of `self` to the first 4 elements in `slice`.
    ///
    /// # Panics
    ///
    /// Panics if `slice` is less than 4 elements long.
    #[inline]
    pub fn write_to_slice(self, slice: &mut [f32x4]) {
        slice[0] = self.x;
        slice[1] = self.y;
        slice[2] = self.z;
        slice[3] = self.w;
    }

    /// Creates a 3D vector from the `x`, `y` and `z` elements of `self`, discarding `w`.
    ///
    /// Truncation to `Vec3x4` may also be performed by using `self.xyz()` or `Vec3x4::from()`.
    #[inline]
    pub fn truncate(self) -> Vec3x4 {
        use crate::swizzles::Vec4Swizzles;
        self.xyz()
    }

    /// Computes the dot product of `self` and `rhs`.
    #[inline]
    pub(crate) fn dot(self, rhs: Self) -> f32x4 {
        (self.x * rhs.x) + (self.y * rhs.y) + (self.z * rhs.z) + (self.w * rhs.w)
    }

    /// 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 {
            x: self.x.minf(rhs.x),

            y: self.y.minf(rhs.y),

            z: self.z.minf(rhs.z),

            w: self.w.minf(rhs.w),
        }
    }

    /// 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 {
            x: self.x.maxf(rhs.x),

            y: self.y.maxf(rhs.y),

            z: self.z.maxf(rhs.z),

            w: self.w.maxf(rhs.w),
        }
    }

    /// 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) -> f32x4 {
        self.x.minf(self.y.minf(self.z.minf(self.w)))
    }

    /// Returns the horizontal maximum of `self`.
    ///
    /// In other words this computes `max(x, y, ..)`.
    #[inline]
    pub fn max_element(self) -> f32x4 {
        self.x.maxf(self.y.maxf(self.z.maxf(self.w)))
    }

    /// 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) -> BVec4x4 {
        BVec4x4::new(
            self.x.eq(rhs.x),
            self.y.eq(rhs.y),
            self.z.eq(rhs.z),
            self.w.eq(rhs.w),
        )
    }

    /// 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) -> BVec4x4 {
        BVec4x4::new(
            self.x.ne(rhs.x),
            self.y.ne(rhs.y),
            self.z.ne(rhs.z),
            self.w.ne(rhs.w),
        )
    }

    /// 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) -> BVec4x4 {
        BVec4x4::new(
            self.x.ge(rhs.x),
            self.y.ge(rhs.y),
            self.z.ge(rhs.z),
            self.w.ge(rhs.w),
        )
    }

    /// 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) -> BVec4x4 {
        BVec4x4::new(
            self.x.gt(rhs.x),
            self.y.gt(rhs.y),
            self.z.gt(rhs.z),
            self.w.gt(rhs.w),
        )
    }

    /// 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) -> BVec4x4 {
        BVec4x4::new(
            self.x.le(rhs.x),
            self.y.le(rhs.y),
            self.z.le(rhs.z),
            self.w.le(rhs.w),
        )
    }

    /// 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) -> BVec4x4 {
        BVec4x4::new(
            self.x.lt(rhs.x),
            self.y.lt(rhs.y),
            self.z.lt(rhs.z),
            self.w.lt(rhs.w),
        )
    }

    /// Returns a vector containing the absolute value of each element of `self`.
    #[inline]
    pub fn abs(self) -> Self {
        Self {
            x: self.x.absf(),
            y: self.y.absf(),
            z: self.z.absf(),
            w: self.w.absf(),
        }
    }

    /// 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 `-Vec4x4::ZERO` output `Vec4x4::ZERO`,
    /// so the sign of `-Vec4x4::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 {
        Self {
            x: self.x.signumf(),
            y: self.y.signumf(),
            z: self.z.signumf(),
            w: self.w.signumf(),
        }
    }

    /// 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) -> bool32x4 {
        self.x.is_finite() & self.y.is_finite() & self.z.is_finite() & self.w.is_finite()
    }

    /// Returns `true` if any elements are `NaN`.
    #[inline]
    pub fn is_nan(self) -> bool32x4 {
        self.x.is_nan() | self.y.is_nan() | self.z.is_nan() | self.w.is_nan()
    }

    /// 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) -> BVec4x4 {
        BVec4x4::new(
            self.x.is_nan(),
            self.y.is_nan(),
            self.z.is_nan(),
            self.w.is_nan(),
        )
    }

    /// 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 {
            x: self.x.roundf(),
            y: self.y.roundf(),
            z: self.z.roundf(),
            w: self.w.roundf(),
        }
    }

    /// 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 {
            x: self.x.floorf(),
            y: self.y.floorf(),
            z: self.z.floorf(),
            w: self.w.floorf(),
        }
    }

    /// 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 {
            x: self.x.ceilf(),
            y: self.y.ceilf(),
            z: self.z.ceilf(),
            w: self.w.ceilf(),
        }
    }

    /// 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(), self.w.expf())
    }

    /// Returns a vector containing each element of `self` raised to the power of `n`.
    #[inline]
    pub fn powf(self, n: f32x4) -> Self {
        Self::new(
            self.x.powff(n),
            self.y.powff(n),
            self.z.powff(n),
            self.w.powff(n),
        )
    }

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

    /// Computes the length of `self`.
    #[doc(alias = "magnitude")]
    #[inline]
    pub fn length(self) -> f32x4 {
        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) -> f32x4 {
        self.dot(self)
    }

    /// Computes `1.0 / length()`.
    ///
    /// For valid results, `self` must _not_ be of length zero.
    #[inline]
    pub fn length_recip(self) -> f32x4 {
        self.length().recip()
    }

    /// Computes the Euclidean distance between two points in space.
    #[inline]
    pub fn distance(self, rhs: Self) -> f32x4 {
        (self - rhs).length()
    }

    /// Compute the squared euclidean distance between two points in space.
    #[inline]
    pub fn distance_squared(self, rhs: Self) -> f32x4 {
        (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.
    ///
    /// # 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 = self.length();
        glam_assert!(length.gt(f32x4::ZERO), "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.mul(length.recip());
        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, f32x4) {
        #[allow(clippy::let_and_return)]
        let length = self.length();
        glam_assert!(length.gt(f32x4::ZERO), "Trying to normalize {:?}", self);
        glam_assert!(self.is_finite(), "Trying to normalize {:?}", self);
        glam_assert!(length.is_finite(), "Trying to normalize {:?}", self);
        let normalized = self.mul(length.recip());
        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) -> UnitVec4x4 {
        self.normalize().as_unit_vec4x4_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) -> (UnitVec4x4, f32x4) {
        let res = self.normalize_and_length();
        (res.0.as_unit_vec4x4_unchecked(), res.1)
    }

    /// 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.
    ///
    #[must_use]
    #[inline]
    pub fn normalize_or_zero(self) -> Self {
        let rcp = self.length_recip();
        let cond = rcp.is_finite() & rcp.gt(f32x4::ZERO);
        Self::lane_select(cond, self * rcp, Self::ZERO)
    }

    /// Returns whether `self` is length `1.0` or not.
    ///
    /// Uses a precision threshold of `1e-6`.
    #[inline]
    pub fn is_normalized(self) -> bool32x4 {
        // TODO: do something with epsilon
        (self.length_squared() - f32x4::ONE)
            .absf()
            .le(f32x4::from(1e-4))
    }

    /// Select lanes from two SOA vector to compose a new SOA vector.
    #[inline]
    pub fn lane_select(cond: bool32x4, if_true: Self, if_false: Self) -> Self {
        Self {
            x: if_true.x.select(cond, if_false.x),
            y: if_true.y.select(cond, if_false.y),
            z: if_true.z.select(cond, if_false.z),
            w: if_true.w.select(cond, if_false.w),
        }
    }

    /// Create a SOA vector with all lanes set to `v`.
    #[inline]
    pub fn splat_soa(v: Vec4) -> Self {
        Self {
            x: f32x4::splat(v.x),
            y: f32x4::splat(v.y),
            z: f32x4::splat(v.z),
            w: f32x4::splat(v.w),
        }
    }

    /// Extract a single lane from a SOA vector.
    ///
    /// # Panics
    ///
    /// Panics if `i` is larger than 3.
    #[inline]
    pub fn extract_lane(&self, i: usize) -> Vec4 {
        Vec4::new(
            self.x.extract(i),
            self.y.extract(i),
            self.z.extract(i),
            self.w.extract(i),
        )
    }

    /// Replace a single lane in a SOA vector.
    ///
    /// # Panics
    ///
    /// Panics if `i` is larger than 3.
    #[inline]
    pub fn replace_lane(&mut self, i: usize, v: Vec4) {
        self.x.replace(i, v.x);
        self.y.replace(i, v.y);
        self.z.replace(i, v.z);
        self.w.replace(i, v.w);
    }

    /// Split a SOA vector into 4 vector.
    #[inline]
    pub fn split_soa(self) -> [Vec4; 4] {
        [
            self.extract_lane(0),
            self.extract_lane(1),
            self.extract_lane(2),
            self.extract_lane(3),
        ]
    }

    /// Compose a SOA vector from 4 vector.
    #[inline]
    pub fn compose_soa(a: &[Vec4; 4]) -> Vec4x4 {
        let mut v = Self::default();
        v.replace_lane(0, a[0]);
        v.replace_lane(1, a[1]);
        v.replace_lane(2, a[2]);
        v.replace_lane(3, a[3]);
        v
    }

    /// 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: UnitVec4x4) -> Self {
        rhs.as_vec4x4() * self.dot(rhs.as_vec4x4())
    }

    /// Returns the vector rejection of `self` from `rhs`.
    #[must_use]
    #[inline]
    pub fn reject_from_normalized(self, rhs: UnitVec4x4) -> 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: f32x4) -> 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: f32x4) -> bool32x4 {
        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: f32x4, max: f32x4) -> Self {
        glam_assert!(min.le(max));
        let length_sq = self.length_squared();
        let cond1 = length_sq.ge(min * min);
        let cond2 = length_sq.lt(max * max);
        let res1 = self * (length_sq.sqrtf().recip() * min);
        let res2 = self * (length_sq.sqrtf().recip() * max);
        let res3 = self;
        let x = res3.x.select(cond1, res1.x).select(cond2, res2.x);
        let y = res3.y.select(cond1, res1.y).select(cond2, res2.y);
        let z = res3.z.select(cond1, res1.z).select(cond2, res2.z);
        let w = res3.w.select(cond1, res1.w).select(cond2, res2.w);
        Self { x, y, z, w }
    }

    /// Returns a vector with a length no more than `max`
    #[inline]
    pub fn clamp_length_max(self, max: f32x4) -> Self {
        let length_sq = self.length_squared();
        let cond = length_sq.gt(max * max);
        let res1 = self * (length_sq.sqrtf().recip() * max);
        let res2 = self;
        let x = res1.x.select(cond, res2.x);
        let y = res1.y.select(cond, res2.y);
        let z = res1.z.select(cond, res2.z);
        let w = res1.w.select(cond, res2.w);
        Self { x, y, z, w }
    }

    /// Returns a vector with a length no less than `min`
    #[inline]
    pub fn clamp_length_min(self, min: f32x4) -> Self {
        let length_sq = self.length_squared();
        let cond = length_sq.lt(min * min);
        let res1 = self * (length_sq.sqrtf().recip() * min);
        let res2 = self;
        let x = res1.x.select(cond, res2.x);
        let y = res1.y.select(cond, res2.y);
        let z = res1.z.select(cond, res2.z);
        let w = res1.w.select(cond, res2.w);
        Self { x, y, z, w }
    }

    /// 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),
            self.w.mul_addf(a.w, b.w),
        )
    }
}

impl Default for Vec4x4 {
    #[inline]
    fn default() -> Self {
        Self::ZERO
    }
}

impl_op_ex!(/ |a: &Vec4x4, b: &Vec4x4| -> Vec4x4 {
        Vec4x4 {
                x: a.x.div(b.x),
                y: a.y.div(b.y),
                z: a.z.div(b.z),
                w: a.w.div(b.w),
        }
});

impl_op_ex!(/= |a: &mut Vec4x4, b: &Vec4x4| {
            a.x.div_assign(b.x);
            a.y.div_assign(b.y);
            a.z.div_assign(b.z);
            a.w.div_assign(b.w);
});

impl_op_ex!(/ |a: &Vec4x4, b: &f32x4| -> Vec4x4 {
        Vec4x4 {
                x: a.x.div(b),
                y: a.y.div(b),
                z: a.z.div(b),
                w: a.w.div(b),
        }
});

impl_op_ex!(/= |a: &mut Vec4x4, b: &f32x4| {
            a.x.div_assign(b);
            a.y.div_assign(b);
            a.z.div_assign(b);
            a.w.div_assign(b);
});

impl_op_ex!(/ |a: &f32x4, b: &Vec4x4| -> Vec4x4 {
        Vec4x4 {
                x: a.div(b.x),
                y: a.div(b.y),
                z: a.div(b.z),
                w: a.div(b.w),
        }
});

impl_op_ex!(*|a: &Vec4x4, b: &Vec4x4| -> Vec4x4 {
    Vec4x4 {
        x: a.x.mul(b.x),
        y: a.y.mul(b.y),
        z: a.z.mul(b.z),
        w: a.w.mul(b.w),
    }
});

impl_op_ex!(*= |a: &mut Vec4x4, b: &Vec4x4| {
            a.x.mul_assign(b.x);
            a.y.mul_assign(b.y);
            a.z.mul_assign(b.z);
            a.w.mul_assign(b.w);
});

impl_op_ex!(*|a: &Vec4x4, b: &f32x4| -> Vec4x4 {
    Vec4x4 {
        x: a.x.mul(b),
        y: a.y.mul(b),
        z: a.z.mul(b),
        w: a.w.mul(b),
    }
});

impl_op_ex!(*= |a: &mut Vec4x4, b: &f32x4| {
            a.x.mul_assign(b);
            a.y.mul_assign(b);
            a.z.mul_assign(b);
            a.w.mul_assign(b);
});

impl_op_ex!(*|a: &f32x4, b: &Vec4x4| -> Vec4x4 {
    Vec4x4 {
        x: a.mul(b.x),
        y: a.mul(b.y),
        z: a.mul(b.z),
        w: a.mul(b.w),
    }
});

impl_op_ex!(+ |a: &Vec4x4, b: &Vec4x4| -> Vec4x4 {
        Vec4x4 {
                x: a.x.add(b.x),
                y: a.y.add(b.y),
                z: a.z.add(b.z),
                w: a.w.add(b.w),
        }
});

impl_op_ex!(+= |a: &mut Vec4x4, b: &Vec4x4| {
            a.x.add_assign(b.x);
            a.y.add_assign(b.y);
            a.z.add_assign(b.z);
            a.w.add_assign(b.w);
});

impl_op_ex!(+ |a: &Vec4x4, b: &f32x4| -> Vec4x4 {
        Vec4x4 {
                x: a.x.add(b),
                y: a.y.add(b),
                z: a.z.add(b),
                w: a.w.add(b),
        }
});

impl_op_ex!(+= |a: &mut Vec4x4, b: &f32x4| {
            a.x.add_assign(b);
            a.y.add_assign(b);
            a.z.add_assign(b);
            a.w.add_assign(b);
});

impl_op_ex!(+ |a: &f32x4, b: &Vec4x4| -> Vec4x4 {
        Vec4x4 {
                x: a.add(b.x),
                y: a.add(b.y),
                z: a.add(b.z),
                w: a.add(b.w),
        }
});

impl_op_ex!(-|a: &Vec4x4, b: &Vec4x4| -> Vec4x4 {
    Vec4x4 {
        x: a.x.sub(b.x),
        y: a.y.sub(b.y),
        z: a.z.sub(b.z),
        w: a.w.sub(b.w),
    }
});

impl_op_ex!(-= |a: &mut Vec4x4, b: &Vec4x4| {
            a.x.sub_assign(b.x);
            a.y.sub_assign(b.y);
            a.z.sub_assign(b.z);
            a.w.sub_assign(b.w);
});

impl_op_ex!(-|a: &Vec4x4, b: &f32x4| -> Vec4x4 {
    Vec4x4 {
        x: a.x.sub(b),
        y: a.y.sub(b),
        z: a.z.sub(b),
        w: a.w.sub(b),
    }
});

impl_op_ex!(-= |a: &mut Vec4x4, b: &f32x4| {
            a.x.sub_assign(b);
            a.y.sub_assign(b);
            a.z.sub_assign(b);
            a.w.sub_assign(b);
});

impl_op_ex!(-|a: &f32x4, b: &Vec4x4| -> Vec4x4 {
    Vec4x4 {
        x: a.sub(b.x),
        y: a.sub(b.y),
        z: a.sub(b.z),
        w: a.sub(b.w),
    }
});

impl_op_ex!(% |a: &Vec4x4, b: &Vec4x4| -> Vec4x4 {
        Vec4x4 {
                x: a.x.rem(b.x),
                y: a.y.rem(b.y),
                z: a.z.rem(b.z),
                w: a.w.rem(b.w),
        }
});

impl_op_ex!(%= |a: &mut Vec4x4, b: &Vec4x4| {
            a.x.rem_assign(b.x);
            a.y.rem_assign(b.y);
            a.z.rem_assign(b.z);
            a.w.rem_assign(b.w);
});

impl_op_ex!(% |a: &Vec4x4, b: &f32x4| -> Vec4x4 {
        Vec4x4 {
                x: a.x.rem(b),
                y: a.y.rem(b),
                z: a.z.rem(b),
                w: a.w.rem(b),
        }
});

impl_op_ex!(%= |a: &mut Vec4x4, b: &f32x4| {
            a.x.rem_assign(b);
            a.y.rem_assign(b);
            a.z.rem_assign(b);
            a.w.rem_assign(b);
});

impl_op_ex!(% |a: &f32x4, b: &Vec4x4| -> Vec4x4 {
        Vec4x4 {
                x: a.rem(b.x),
                y: a.rem(b.y),
                z: a.rem(b.z),
                w: a.rem(b.w),
        }
});

impl<'a> Sum<&'a Self> for Vec4x4 {
    #[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 Vec4x4 {
    #[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 Vec4x4 {
    type Output = Self;
    #[inline]
    fn neg(self) -> Self {
        Self {
            x: self.x.neg(),
            y: self.y.neg(),
            z: self.z.neg(),
            w: self.w.neg(),
        }
    }
}

impl Index<usize> for Vec4x4 {
    type Output = f32x4;
    #[inline]
    fn index(&self, index: usize) -> &Self::Output {
        match index {
            0 => &self.x,
            1 => &self.y,
            2 => &self.z,
            3 => &self.w,
            _ => panic!("index out of bounds"),
        }
    }
}

impl IndexMut<usize> for Vec4x4 {
    #[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,
            3 => &mut self.w,
            _ => panic!("index out of bounds"),
        }
    }
}

#[cfg(not(target_arch = "spirv"))]
impl fmt::Display for Vec4x4 {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        write!(f, "[{}, {}, {}, {}]", self.x, self.y, self.z, self.w)
    }
}

#[cfg(not(target_arch = "spirv"))]
impl fmt::Debug for Vec4x4 {
    fn fmt(&self, fmt: &mut fmt::Formatter<'_>) -> fmt::Result {
        fmt.debug_tuple(stringify!(Vec4x4))
            .field(&self.x)
            .field(&self.y)
            .field(&self.z)
            .field(&self.w)
            .finish()
    }
}

impl From<[f32x4; 4]> for Vec4x4 {
    #[inline]
    fn from(a: [f32x4; 4]) -> Self {
        Self::new(a[0], a[1], a[2], a[3])
    }
}

impl From<Vec4x4> for [f32x4; 4] {
    #[inline]
    fn from(v: Vec4x4) -> Self {
        [v.x, v.y, v.z, v.w]
    }
}

impl From<(f32x4, f32x4, f32x4, f32x4)> for Vec4x4 {
    #[inline]
    fn from(t: (f32x4, f32x4, f32x4, f32x4)) -> Self {
        Self::new(t.0, t.1, t.2, t.3)
    }
}

impl From<Vec4x4> for (f32x4, f32x4, f32x4, f32x4) {
    #[inline]
    fn from(v: Vec4x4) -> Self {
        (v.x, v.y, v.z, v.w)
    }
}

impl From<(Vec3x4, f32x4)> for Vec4x4 {
    #[inline]
    fn from((v, w): (Vec3x4, f32x4)) -> Self {
        Self::new(v.x, v.y, v.z, w)
    }
}

impl From<(f32x4, Vec3x4)> for Vec4x4 {
    #[inline]
    fn from((x, v): (f32x4, Vec3x4)) -> Self {
        Self::new(x, v.x, v.y, v.z)
    }
}

impl From<(Vec2x4, f32x4, f32x4)> for Vec4x4 {
    #[inline]
    fn from((v, z, w): (Vec2x4, f32x4, f32x4)) -> Self {
        Self::new(v.x, v.y, z, w)
    }
}

impl From<(Vec2x4, Vec2x4)> for Vec4x4 {
    #[inline]
    fn from((v, u): (Vec2x4, Vec2x4)) -> Self {
        Self::new(v.x, v.y, u.x, u.y)
    }
}

#[cfg(not(target_arch = "spirv"))]
impl AsRef<[f32x4; 4]> for Vec4x4 {
    #[inline]
    fn as_ref(&self) -> &[f32x4; 4] {
        unsafe { &*(self as *const Vec4x4 as *const [f32x4; 4]) }
    }
}

#[cfg(not(target_arch = "spirv"))]
impl AsMut<[f32x4; 4]> for Vec4x4 {
    #[inline]
    fn as_mut(&mut self) -> &mut [f32x4; 4] {
        unsafe { &mut *(self as *mut Vec4x4 as *mut [f32x4; 4]) }
    }
}