cgmath/src/vec.rs
2013-04-01 07:35:09 +11:00

1668 lines
43 KiB
Rust

use std::cmp::{FuzzyEq, FUZZY_EPSILON};
use numeric::*;
use numeric::number::Number;
use numeric::number::Number::{zero,one};
/**
* The base generic vector trait.
*
* # Type parameters
*
* * `T` - The type of the components. This is intended to support boolean,
* integer, unsigned integer, and floating point types.
*/
pub trait Vector<T>: Index<uint,T> + Eq {
/**
* Construct the vector from a single value, copying it to each component
*/
fn from_value(value: T) -> Self;
/**
* # Return value
*
* A pointer to the first component of the vector
*/
fn to_ptr(&self) -> *T;
/**
* Get a mutable reference to the component at `i`
*/
fn index_mut(&mut self, i: uint) -> &'self mut T;
/**
* Swap two components of the vector in place
*/
fn swap(&mut self, a: uint, b: uint);
}
/**
* A generic 2-dimensional vector
*/
pub trait Vector2<T>: Vector<T> {
fn new(x: T, y: T) -> Self;
}
/**
* A generic 3-dimensional vector
*/
pub trait Vector3<T>: Vector<T> {
fn new(x: T, y: T, z: T) -> Self;
}
/**
* A generic 4-dimensional vector
*/
pub trait Vector4<T>: Vector<T> {
fn new(x: T, y: T, z: T, w: T) -> Self;
}
/**
* A vector with numeric components
*/
pub trait NumericVector<T>: Vector<T> + Neg<Self> {
/**
* The standard basis vector
*
* # Return value
*
* A vector with each component set to one
*/
fn identity() -> Self;
/**
* The null vector
*
* # Return value
*
* A vector with each component set to zero
*/
fn zero() -> Self;
/**
* # Return value
*
* True if the vector is equal to zero
*/
fn is_zero(&self) -> bool;
/**
* # Return value
*
* The scalar multiplication of the vector and `value`
*/
fn mul_t(&self, value: T) -> Self;
/**
* # Return value
*
* The scalar division of the vector and `value`
*/
fn div_t(&self, value: T) -> Self;
/**
* Component-wise vector addition
*/
fn add_v(&self, other: &Self) -> Self;
/**
* Component-wise vector subtraction
*/
fn sub_v(&self, other: &Self) -> Self;
/**
* Component-wise vector multiplication
*/
fn mul_v(&self, other: &Self) -> Self;
/**
* Component-wise vector division
*/
fn div_v(&self, other: &Self) -> Self;
/**
* # Return value
*
* The dot product of the vector and `other`
*/
fn dot(&self, other: &Self) -> T;
/**
* Negate the vector
*/
fn neg_self(&mut self);
/**
* Multiply the vector by a scalar
*/
fn mul_self_t(&mut self, value: T);
/**
* Divide the vector by a scalar
*/
fn div_self_t(&mut self, value: T);
/**
* Set the vector to the component-wise vector sum
*/
fn add_self_v(&mut self, other: &Self);
/**
* Set the vector to the component-wise vector difference
*/
fn sub_self_v(&mut self, other: &Self);
/**
* Set the vector to the component-wise vector product
*/
fn mul_self_v(&mut self, other: &Self);
/**
* Set the vector to the component-wise vector quotient
*/
fn div_self_v(&mut self, other: &Self);
}
/**
* A 2-dimensional vector with numeric components
*/
pub trait NumericVector2<T>: NumericVector<T> {
fn unit_x() -> Self;
fn unit_y() -> Self;
/**
* # Return value
*
* The perp dot product of the vector and `other`
*/
fn perp_dot(&self, other: &Self) -> T;
}
/**
* A 3-dimensional vector with numeric components
*/
pub trait NumericVector3<T>: NumericVector<T> {
fn unit_x() -> Self;
fn unit_y() -> Self;
fn unit_z() -> Self;
/**
* # Return value
*
* The cross product of the vector and `other`
*/
fn cross(&self, other: &Self) -> Self;
/**
* Set to the cross product of the vector and `other`
*/
fn cross_self(&mut self, other: &Self);
}
/**
* A 4-dimensional vector with numeric components
*/
pub trait NumericVector4<T>: NumericVector<T> {
fn unit_x() -> Self;
fn unit_y() -> Self;
fn unit_z() -> Self;
fn unit_w() -> Self;
}
pub trait ToHomogeneous<H> {
/**
* Convert to a homogenous coordinate
*/
fn to_homogeneous(&self) -> H;
}
/**
* A Euclidean (or Affine) vector
*
* # Type parameters
*
* * `T` - The type of the components. This should be a floating point type.
*/
pub trait EuclideanVector<T>: NumericVector<T> {
/**
* # Return value
*
* The squared length of the vector. This is useful for comparisons where
* the exact length does not need to be calculated.
*/
fn length2(&self) -> T;
/**
* # Return value
*
* The length of the vector
*
* # Performance notes
*
* For instances where the exact length of the vector does not need to be
* known, for example for quaternion-quaternion length comparisons,
* it is advisable to use the `length2` method instead.
*/
fn length(&self) -> T;
/**
* # Return value
*
* The squared distance between the vector and `other`.
*/
fn distance2(&self, other: &Self) -> T;
/**
* # Return value
*
* The distance between the vector and `other`
*/
fn distance(&self, other: &Self) -> T;
/**
* # Return value
*
* The angle between the vector and `other` in radians
*/
fn angle(&self, other: &Self) -> T;
/**
* # Return value
*
* The normalized vector
*/
fn normalize(&self) -> Self;
/**
* Set the length of the vector whilst preserving the direction
*/
fn normalize_to(&self, length: T) -> Self;
/**
* Linearly intoperlate between the vector and `other`
*
* # Return value
*
* The intoperlated vector
*/
fn lerp(&self, other: &Self, amount: T) -> Self;
/**
* Normalize the vector
*/
fn normalize_self(&mut self);
/**
* Set the vector to a specified length whilst preserving the direction
*/
fn normalize_self_to(&mut self, length: T);
/**
* Linearly intoperlate the vector towards `other`
*/
fn lerp_self(&mut self, other: &Self, amount: T);
}
/**
* Component-wise vector comparison methods
*
* The methods contained in this trait correspond to the relational functions
* mentioned in Section 8.7 of the [GLSL 4.30.6 specification]
* (http://www.opengl.org/registry/doc/GLSLangSpec.4.30.6.pdf).
*/
pub trait OrdinalVector<T, BoolVec>: Vector<T> {
/**
* Component-wise compare of `self < other`
*/
fn less_than(&self, other: &Self) -> BoolVec;
/**
* Component-wise compare of `self <= other`
*/
fn less_than_equal(&self, other: &Self) -> BoolVec;
/**
* Component-wise compare of `self > other`
*/
fn greater_than(&self, other: &Self) -> BoolVec;
/**
* Component-wise compare of `self >= other`
*/
fn greater_than_equal(&self, other: &Self) -> BoolVec;
}
/**
* Component-wise equality comparison methods
*
* The methods contained in this trait correspond to the relational functions
* mentioned in Section 8.7 of the [GLSL 4.30.6 specification]
* (http://www.opengl.org/registry/doc/GLSLangSpec.4.30.6.pdf).
*/
pub trait EquableVector<T, BoolVec>: Vector<T> {
/**
* Component-wise compare of `self == other`
*/
fn equal(&self, other: &Self) -> BoolVec;
/**
* Component-wise compare of `self != other`
*/
fn not_equal(&self, other: &Self) -> BoolVec;
}
/**
* A vector with boolean components
*
* The methods contained in this trait correspond to the relational functions
* mentioned in Section 8.7 of the [GLSL 4.30.6 specification]
* (http://www.opengl.org/registry/doc/GLSLangSpec.4.30.6.pdf).
*/
pub trait BooleanVector: Vector<bool> {
/**
* # Return value
*
* `true` if of any component is `true`
*/
fn any(&self) -> bool;
/**
* # Return value
*
* `true` only if all components are `true`
*/
fn all(&self) -> bool;
/**
* # Return value
*
* the component-wise logical complement
*/
fn not(&self) -> Self;
}
pub trait TrigVec<T>: Vector<T> {
fn radians(&self) -> Self;
fn degrees(&self) -> Self;
// Triganometric functions
fn sin(&self) -> Self;
fn cos(&self) -> Self;
fn tan(&self) -> Self;
// Inverse triganometric functions
fn asin(&self) -> Self;
fn acos(&self) -> Self;
fn atan(&self) -> Self;
fn atan2(&self, other: Self) -> Self;
// Hyperbolic triganometric functions
fn sinh(&self) -> Self;
fn cosh(&self) -> Self;
fn tanh(&self) -> Self;
// fn asinh() -> Self;
// fn acosh() -> Self;
// fn atanh() -> Self;
}
pub trait ExpVec<T>: Vector<T> {
// Exponential functions
fn pow_t(&self, n: Self) -> Self;
fn pow_v(&self, n: T) -> Self;
fn exp(&self) -> Self;
fn exp2(&self) -> Self;
fn ln(&self) -> Self;
fn ln2(&self) -> Self;
fn sqrt(&self) -> Self;
fn inv_sqrt(&self) -> Self;
}
pub trait ApproxVec<T>: Vector<T> {
// Whole-number approximation functions
fn floor(&self) -> Self;
fn trunc(&self) -> Self;
fn round(&self) -> Self;
// fn round_even(&self) -> Self;
fn ceil(&self) -> Self;
fn fract(&self) -> Self;
}
pub trait SignedVec<T,BV>: Vector<T> {
fn is_positive(&self) -> BV;
fn is_negative(&self) -> BV;
fn is_nonpositive(&self) -> BV;
fn is_nonnegative(&self) -> BV;
fn abs(&self) -> Self;
fn sign(&self) -> Self;
fn copysign(&self, other: Self) -> Self;
}
pub trait ExtentVec<T>: Vector<T> {
fn min_v(&self, other: &Self) -> Self;
fn max_v(&self, other: &Self) -> Self;
fn clamp_v(&self, mn: &Self, mx: &Self) -> Self;
fn min_t(&self, other: T) -> Self;
fn max_t(&self, other: T) -> Self;
fn clamp_t(&self, mn: T, mx: T) -> Self;
}
pub trait MixVec<T>: Vector<T> {
// Functions for blending numbers together
fn mix(&self, other: Self, value: Self) -> Self;
fn smooth_step(&self, edge0: Self, edge1: Self) -> Self;
fn step(&self, edge: Self) -> Self;
}
// Utility macros
macro_rules! zip_vec2(
($a:ident[] $method:ident $b:ident[]) => (
Vector2::new($a[0].$method(&($b[0])),
$a[1].$method(&($b[1])))
);
($a:ident[] $method:ident $b:ident) => (
Vector2::new($a[0].$method(&($b)),
$a[1].$method(&($b)))
);
)
macro_rules! zip_vec3(
($a:ident[] $method:ident $b:ident[]) => (
Vector3::new($a[0].$method(&($b[0])),
$a[1].$method(&($b[1])),
$a[2].$method(&($b[2])))
);
($a:ident[] $method:ident $b:ident) => (
Vector3::new($a[0].$method(&($b)),
$a[1].$method(&($b)),
$a[2].$method(&($b)))
);
)
macro_rules! zip_vec4(
($a:ident[] $method:ident $b:ident[]) => (
Vector4::new($a[0].$method(&($b[0])),
$a[1].$method(&($b[1])),
$a[2].$method(&($b[2])),
$a[3].$method(&($b[3])))
);
($a:ident[] $method:ident $b:ident) => (
Vector4::new($a[0].$method(&($b)),
$a[1].$method(&($b)),
$a[2].$method(&($b)),
$a[3].$method(&($b)))
);
)
macro_rules! zip_assign(
($a:ident[] $method:ident $b:ident[] ..2) => ({ $a.index_mut(0).$method($b[0]); $a.index_mut(1).$method($b[1]); });
($a:ident[] $method:ident $b:ident[] ..3) => ({ zip_assign!($a[] $method $b[] ..2); $a.index_mut(2).$method($b[2]); });
($a:ident[] $method:ident $b:ident[] ..4) => ({ zip_assign!($a[] $method $b[] ..3); $a.index_mut(3).$method($b[3]); });
($a:ident[] $method:ident $b:ident ..2) => ({ $a.index_mut(0).$method($b); $a.index_mut(1).$method($b); });
($a:ident[] $method:ident $b:ident ..3) => ({ zip_assign!($a[] $method $b ..2); $a.index_mut(2).$method($b); });
($a:ident[] $method:ident $b:ident ..4) => ({ zip_assign!($a[] $method $b ..3); $a.index_mut(3).$method($b); });
)
/**
* A 2-dimensional vector
*
* # Type parameters
*
* * `T` - The type of the components. This is intended to support boolean,
* integer, unsigned integer, and floating point types.
*
* # Fields
*
* * `x` - the first component of the vector
* * `y` - the second component of the vector
*/
#[deriving(Eq)]
pub struct Vec2<T> { x: T, y: T }
impl<T:Copy + Eq> Vector<T> for Vec2<T> {
#[inline(always)]
fn from_value(value: T) -> Vec2<T> {
Vector2::new(value, value)
}
#[inline(always)]
fn to_ptr(&self) -> *T {
unsafe { cast::transmute(self) }
}
#[inline(always)]
fn index_mut(&mut self, i: uint) -> &'self mut T {
match i {
0 => &mut self.x,
1 => &mut self.y,
_ => fail!(fmt!("index out of bounds: expected an index from 0 to 1, but found %u", i))
}
}
#[inline(always)]
fn swap(&mut self, a: uint, b: uint) {
*self.index_mut(a) <-> *self.index_mut(b);
}
}
impl<T> Vector2<T> for Vec2<T> {
#[inline(always)]
fn new(x: T, y: T ) -> Vec2<T> {
Vec2 { x: x, y: y }
}
}
impl<T:Copy + Eq> Index<uint, T> for Vec2<T> {
#[inline(always)]
fn index(&self, i: &uint) -> T {
unsafe { do vec::raw::buf_as_slice(self.to_ptr(), 2) |slice| { slice[*i] } }
}
}
impl<T:Copy + Number + Add<T,T> + Sub<T,T> + Mul<T,T> + Div<T,T> + Neg<T>> NumericVector<T> for Vec2<T> {
#[inline(always)]
fn identity() -> Vec2<T> {
Vector2::new(one::<T>(), one::<T>())
}
#[inline(always)]
fn zero() -> Vec2<T> {
Vector2::new(zero::<T>(), zero::<T>())
}
#[inline(always)]
fn is_zero(&self) -> bool {
self[0] == zero() &&
self[1] == zero()
}
#[inline(always)]
fn mul_t(&self, value: T) -> Vec2<T> {
zip_vec2!(self[] mul value)
}
#[inline(always)]
fn div_t(&self, value: T) -> Vec2<T> {
zip_vec2!(self[] div value)
}
#[inline(always)]
fn add_v(&self, other: &Vec2<T>) -> Vec2<T> {
zip_vec2!(self[] add other[])
}
#[inline(always)]
fn sub_v(&self, other: &Vec2<T>) -> Vec2<T> {
zip_vec2!(self[] sub other[])
}
#[inline(always)]
fn mul_v(&self, other: &Vec2<T>) -> Vec2<T> {
zip_vec2!(self[] mul other[])
}
#[inline(always)]
fn div_v(&self, other: &Vec2<T>) -> Vec2<T> {
zip_vec2!(self[] div other[])
}
#[inline(always)]
fn dot(&self, other: &Vec2<T>) -> T {
self[0] * other[0] +
self[1] * other[1]
}
#[inline(always)]
fn neg_self(&mut self) {
*self.index_mut(0) = -self[0];
*self.index_mut(1) = -self[1];
}
#[inline(always)]
fn mul_self_t(&mut self, value: T) {
zip_assign!(self[] mul_assign value ..2);
}
#[inline(always)]
fn div_self_t(&mut self, value: T) {
zip_assign!(self[] div_assign value ..2);
}
#[inline(always)]
fn add_self_v(&mut self, other: &Vec2<T>) {
zip_assign!(self[] add_assign other[] ..2);
}
#[inline(always)]
fn sub_self_v(&mut self, other: &Vec2<T>) {
zip_assign!(self[] sub_assign other[] ..2);
}
#[inline(always)]
fn mul_self_v(&mut self, other: &Vec2<T>) {
zip_assign!(self[] mul_assign other[] ..2);
}
#[inline(always)]
fn div_self_v(&mut self, other: &Vec2<T>) {
zip_assign!(self[] div_assign other[] ..2);
}
}
impl<T:Copy + Number + Add<T,T> + Sub<T,T> + Mul<T,T> + Div<T,T> + Neg<T>> Neg<Vec2<T>> for Vec2<T> {
#[inline(always)]
fn neg(&self) -> Vec2<T> {
Vector2::new(-self[0], -self[1])
}
}
impl<T:Copy + Number + Add<T,T> + Sub<T,T> + Mul<T,T> + Div<T,T> + Neg<T>> NumericVector2<T> for Vec2<T> {
#[inline(always)]
fn unit_x() -> Vec2<T> {
Vector2::new(one::<T>(), zero::<T>())
}
#[inline(always)]
fn unit_y() -> Vec2<T> {
Vector2::new(zero::<T>(), one::<T>())
}
#[inline(always)]
fn perp_dot(&self, other: &Vec2<T>) ->T {
(self[0] * other[1]) - (self[1] * other[0])
}
}
impl<T:Copy + Number> ToHomogeneous<Vec3<T>> for Vec2<T> {
#[inline(always)]
fn to_homogeneous(&self) -> Vec3<T> {
Vector3::new(self.x, self.y, zero())
}
}
impl<T:Copy + Float + Add<T,T> + Sub<T,T> + Mul<T,T> + Div<T,T> + Neg<T>> EuclideanVector<T> for Vec2<T> {
#[inline(always)]
fn length2(&self) -> T {
self.dot(self)
}
#[inline(always)]
fn length(&self) -> T {
self.length2().sqrt()
}
#[inline(always)]
fn distance2(&self, other: &Vec2<T>) -> T {
other.sub_v(self).length2()
}
#[inline(always)]
fn distance(&self, other: &Vec2<T>) -> T {
other.distance2(self).sqrt()
}
#[inline(always)]
fn angle(&self, other: &Vec2<T>) -> T {
atan2(self.perp_dot(other), self.dot(other))
}
#[inline(always)]
fn normalize(&self) -> Vec2<T> {
self.mul_t(one::<T>()/self.length())
}
#[inline(always)]
fn normalize_to(&self, length: T) -> Vec2<T> {
self.mul_t(length / self.length())
}
#[inline(always)]
fn lerp(&self, other: &Vec2<T>, amount: T) -> Vec2<T> {
self.add_v(&other.sub_v(self).mul_t(amount))
}
#[inline(always)]
fn normalize_self(&mut self) {
let n = one::<T>() / self.length();
self.mul_self_t(n);
}
#[inline(always)]
fn normalize_self_to(&mut self, length: T) {
let n = length / self.length();
self.mul_self_t(n);
}
fn lerp_self(&mut self, other: &Vec2<T>, amount: T) {
let v = other.sub_v(self).mul_t(amount);
self.add_self_v(&v);
}
}
impl<T:Copy + Float + FuzzyEq<T>> FuzzyEq<T> for Vec2<T> {
#[inline(always)]
fn fuzzy_eq(&self, other: &Vec2<T>) -> bool {
self.fuzzy_eq_eps(other, &Number::from(FUZZY_EPSILON))
}
#[inline(always)]
fn fuzzy_eq_eps(&self, other: &Vec2<T>, epsilon: &T) -> bool {
self[0].fuzzy_eq_eps(&other[0], epsilon) &&
self[1].fuzzy_eq_eps(&other[1], epsilon)
}
}
impl<T:Copy + Ord + Eq> OrdinalVector<T, Vec2<bool>> for Vec2<T> {
#[inline(always)]
fn less_than(&self, other: &Vec2<T>) -> Vec2<bool> {
zip_vec2!(self[] lt other[])
}
#[inline(always)]
fn less_than_equal(&self, other: &Vec2<T>) -> Vec2<bool> {
zip_vec2!(self[] le other[])
}
#[inline(always)]
fn greater_than(&self, other: &Vec2<T>) -> Vec2<bool> {
zip_vec2!(self[] gt other[])
}
#[inline(always)]
fn greater_than_equal(&self, other: &Vec2<T>) -> Vec2<bool> {
zip_vec2!(self[] ge other[])
}
}
impl<T:Copy + Eq> EquableVector<T, Vec2<bool>> for Vec2<T> {
#[inline(always)]
fn equal(&self, other: &Vec2<T>) -> Vec2<bool> {
zip_vec2!(self[] eq other[])
}
#[inline(always)]
fn not_equal(&self, other: &Vec2<T>) -> Vec2<bool> {
zip_vec2!(self[] ne other[])
}
}
impl BooleanVector for Vec2<bool> {
#[inline(always)]
fn any(&self) -> bool {
self[0] || self[1]
}
#[inline(always)]
fn all(&self) -> bool {
self[0] && self[1]
}
#[inline(always)]
fn not(&self) -> Vec2<bool> {
Vector2::new(!self[0], !self[1])
}
}
macro_rules! vec2_type(
($name:ident <bool>) => (
pub impl $name {
#[inline(always)] fn new(x: bool, y: bool) -> $name { Vector2::new(x, y) }
#[inline(always)] fn from_value(v: bool) -> $name { Vector::from_value(v) }
#[inline(always)] fn dim() -> uint { 2 }
#[inline(always)] fn size_of() -> uint { sys::size_of::<$name>() }
}
);
($name:ident <$T:ty>) => (
pub impl $name {
#[inline(always)] fn new(x: $T, y: $T) -> $name { Vector2::new(x, y) }
#[inline(always)] fn from_value(v: $T) -> $name { Vector::from_value(v) }
#[inline(always)] fn identity() -> $name { NumericVector::identity() }
#[inline(always)] fn zero() -> $name { NumericVector::zero() }
#[inline(always)] fn unit_x() -> $name { NumericVector2::unit_x() }
#[inline(always)] fn unit_y() -> $name { NumericVector2::unit_y() }
#[inline(always)] fn dim() -> uint { 2 }
#[inline(always)] fn size_of() -> uint { sys::size_of::<$name>() }
}
);
)
// GLSL-style type aliases, corresponding to Section 4.1.5 of the [GLSL 4.30.6 specification]
// (http://www.opengl.org/registry/doc/GLSLangSpec.4.30.6.pdf).
// a two-component single-precision floating-point vector
pub type vec2 = Vec2<f32>;
// a two-component double-precision floating-point vector
pub type dvec2 = Vec2<f64>;
// a two-component Boolean vector
pub type bvec2 = Vec2<bool>;
// a two-component signed integer vector
pub type ivec2 = Vec2<i32>;
// a two-component unsigned integer vector
pub type uvec2 = Vec2<u32>;
vec2_type!(vec2<f32>)
vec2_type!(dvec2<f64>)
vec2_type!(bvec2<bool>)
vec2_type!(ivec2<i32>)
vec2_type!(uvec2<u32>)
// Rust-style type aliases
pub type Vec2f = Vec2<float>;
pub type Vec2f32 = Vec2<f32>;
pub type Vec2f64 = Vec2<f64>;
pub type Vec2i = Vec2<int>;
pub type Vec2i8 = Vec2<i8>;
pub type Vec2i16 = Vec2<i16>;
pub type Vec2i32 = Vec2<i32>;
pub type Vec2i64 = Vec2<i64>;
pub type Vec2u = Vec2<uint>;
pub type Vec2u8 = Vec2<u8>;
pub type Vec2u16 = Vec2<u16>;
pub type Vec2u32 = Vec2<u32>;
pub type Vec2u64 = Vec2<u64>;
pub type Vec2b = Vec2<bool>;
vec2_type!(Vec2f<float>)
vec2_type!(Vec2f32<f32>)
vec2_type!(Vec2f64<f64>)
vec2_type!(Vec2i<int>)
vec2_type!(Vec2i8<i8>)
vec2_type!(Vec2i16<i16>)
vec2_type!(Vec2i32<i32>)
vec2_type!(Vec2i64<i64>)
vec2_type!(Vec2u<uint>)
vec2_type!(Vec2u8<u8>)
vec2_type!(Vec2u16<u16>)
vec2_type!(Vec2u32<u32>)
vec2_type!(Vec2u64<u64>)
vec2_type!(Vec2b<bool>)
/**
* A 3-dimensional vector
*
* # Type parameters
*
* * `T` - The type of the components. This is intended to support boolean,
* integer, unsigned integer, and floating point types.
*
* # Fields
*
* * `x` - the first component of the vector
* * `y` - the second component of the vector
* * `z` - the third component of the vector
*/
#[deriving(Eq)]
pub struct Vec3<T> { x: T, y: T, z: T }
impl<T:Copy + Eq> Vector<T> for Vec3<T> {
#[inline(always)]
fn from_value(value: T) -> Vec3<T> {
Vector3::new(value, value, value)
}
#[inline(always)]
fn to_ptr(&self) -> *T {
unsafe { cast::transmute(self) }
}
#[inline(always)]
fn index_mut(&mut self, i: uint) -> &'self mut T {
match i {
0 => &mut self.x,
1 => &mut self.y,
2 => &mut self.z,
_ => fail!(fmt!("index out of bounds: expected an index from 0 to 2, but found %u", i))
}
}
#[inline(always)]
fn swap(&mut self, a: uint, b: uint) {
*self.index_mut(a) <-> *self.index_mut(b);
}
}
impl<T> Vector3<T> for Vec3<T> {
#[inline(always)]
fn new(x: T, y: T, z: T) -> Vec3<T> {
Vec3 { x: x, y: y, z: z }
}
}
impl<T:Copy + Eq> Index<uint, T> for Vec3<T> {
#[inline(always)]
fn index(&self, i: &uint) -> T {
unsafe { do vec::raw::buf_as_slice(self.to_ptr(), 3) |slice| { slice[*i] } }
}
}
impl<T:Copy + Number + Add<T,T> + Sub<T,T> + Mul<T,T> + Div<T,T> + Neg<T>> NumericVector<T> for Vec3<T> {
#[inline(always)]
fn identity() -> Vec3<T> {
Vector3::new(one::<T>(), one::<T>(), one::<T>())
}
#[inline(always)]
fn zero() -> Vec3<T> {
Vector3::new(zero::<T>(), zero::<T>(), zero::<T>())
}
#[inline(always)]
fn is_zero(&self) -> bool {
self[0] == zero() &&
self[1] == zero() &&
self[2] == zero()
}
#[inline(always)]
fn mul_t(&self, value: T) -> Vec3<T> {
zip_vec3!(self[] mul value)
}
#[inline(always)]
fn div_t(&self, value: T) -> Vec3<T> {
zip_vec3!(self[] div value)
}
#[inline(always)]
fn add_v(&self, other: &Vec3<T>) -> Vec3<T> {
zip_vec3!(self[] add other[])
}
#[inline(always)]
fn sub_v(&self, other: &Vec3<T>) -> Vec3<T> {
zip_vec3!(self[] sub other[])
}
#[inline(always)]
fn mul_v(&self, other: &Vec3<T>) -> Vec3<T> {
zip_vec3!(self[] mul other[])
}
#[inline(always)]
fn div_v(&self, other: &Vec3<T>) -> Vec3<T> {
zip_vec3!(self[] div other[])
}
#[inline(always)]
fn dot(&self, other: &Vec3<T>) -> T {
self[0] * other[0] +
self[1] * other[1] +
self[2] * other[2]
}
#[inline(always)]
fn neg_self(&mut self) {
*self.index_mut(0) = -self[0];
*self.index_mut(1) = -self[1];
*self.index_mut(2) = -self[2];
}
#[inline(always)]
fn mul_self_t(&mut self, value: T) {
zip_assign!(self[] mul_assign value ..3);
}
#[inline(always)]
fn div_self_t(&mut self, value: T) {
zip_assign!(self[] div_assign value ..3);
}
#[inline(always)]
fn add_self_v(&mut self, other: &Vec3<T>) {
zip_assign!(self[] add_assign other[] ..3);
}
#[inline(always)]
fn sub_self_v(&mut self, other: &Vec3<T>) {
zip_assign!(self[] sub_assign other[] ..3);
}
#[inline(always)]
fn mul_self_v(&mut self, other: &Vec3<T>) {
zip_assign!(self[] mul_assign other[] ..3);
}
#[inline(always)]
fn div_self_v(&mut self, other: &Vec3<T>) {
zip_assign!(self[] div_assign other[] ..3);
}
}
impl<T:Copy + Number + Add<T,T> + Sub<T,T> + Mul<T,T> + Div<T,T> + Neg<T>> Neg<Vec3<T>> for Vec3<T> {
#[inline(always)]
fn neg(&self) -> Vec3<T> {
Vector3::new(-self[0], -self[1], -self[2])
}
}
impl<T:Copy + Number + Add<T,T> + Sub<T,T> + Mul<T,T> + Div<T,T> + Neg<T>> NumericVector3<T> for Vec3<T> {
#[inline(always)]
fn unit_x() -> Vec3<T> {
Vector3::new(one::<T>(), zero::<T>(), zero::<T>())
}
#[inline(always)]
fn unit_y() -> Vec3<T> {
Vector3::new(zero::<T>(), one::<T>(), zero::<T>())
}
#[inline(always)]
fn unit_z() -> Vec3<T> {
Vector3::new(zero::<T>(), zero::<T>(), one::<T>())
}
#[inline(always)]
fn cross(&self, other: &Vec3<T>) -> Vec3<T> {
Vector3::new((self[1] * other[2]) - (self[2] * other[1]),
(self[2] * other[0]) - (self[0] * other[2]),
(self[0] * other[1]) - (self[1] * other[0]))
}
#[inline(always)]
fn cross_self(&mut self, other: &Vec3<T>) {
*self = self.cross(other);
}
}
impl<T:Copy + Number + Add<T,T> + Sub<T,T> + Mul<T,T> + Div<T,T> + Neg<T>> ToHomogeneous<Vec4<T>> for Vec3<T> {
#[inline(always)]
fn to_homogeneous(&self) -> Vec4<T> {
Vector4::new(self.x, self.y, self.z, zero())
}
}
impl<T:Copy + Float + Add<T,T> + Sub<T,T> + Mul<T,T> + Div<T,T> + Neg<T>> EuclideanVector<T> for Vec3<T> {
#[inline(always)]
fn length2(&self) -> T {
self.dot(self)
}
#[inline(always)]
fn length(&self) -> T {
self.length2().sqrt()
}
#[inline(always)]
fn distance2(&self, other: &Vec3<T>) -> T {
other.sub_v(self).length2()
}
#[inline(always)]
fn distance(&self, other: &Vec3<T>) -> T {
other.distance2(self).sqrt()
}
#[inline(always)]
fn angle(&self, other: &Vec3<T>) -> T {
atan2(self.cross(other).length(), self.dot(other))
}
#[inline(always)]
fn normalize(&self) -> Vec3<T> {
self.mul_t(one::<T>()/self.length())
}
#[inline(always)]
fn normalize_to(&self, length: T) -> Vec3<T> {
self.mul_t(length / self.length())
}
#[inline(always)]
fn lerp(&self, other: &Vec3<T>, amount: T) -> Vec3<T> {
self.add_v(&other.sub_v(self).mul_t(amount))
}
#[inline(always)]
fn normalize_self(&mut self) {
let n = one::<T>() / self.length();
self.mul_self_t(n);
}
#[inline(always)]
fn normalize_self_to(&mut self, length: T) {
let n = length / self.length();
self.mul_self_t(n);
}
fn lerp_self(&mut self, other: &Vec3<T>, amount: T) {
let v = other.sub_v(self).mul_t(amount);
self.add_self_v(&v);
}
}
impl<T:Copy + Float + FuzzyEq<T>> FuzzyEq<T> for Vec3<T> {
#[inline(always)]
fn fuzzy_eq(&self, other: &Vec3<T>) -> bool {
self.fuzzy_eq_eps(other, &Number::from(FUZZY_EPSILON))
}
#[inline(always)]
fn fuzzy_eq_eps(&self, other: &Vec3<T>, epsilon: &T) -> bool {
self[0].fuzzy_eq_eps(&other[0], epsilon) &&
self[1].fuzzy_eq_eps(&other[1], epsilon) &&
self[2].fuzzy_eq_eps(&other[2], epsilon)
}
}
impl<T:Copy + Ord + Eq> OrdinalVector<T, Vec3<bool>> for Vec3<T> {
#[inline(always)]
fn less_than(&self, other: &Vec3<T>) -> Vec3<bool> {
zip_vec3!(self[] lt other[])
}
#[inline(always)]
fn less_than_equal(&self, other: &Vec3<T>) -> Vec3<bool> {
zip_vec3!(self[] le other[])
}
#[inline(always)]
fn greater_than(&self, other: &Vec3<T>) -> Vec3<bool> {
zip_vec3!(self[] gt other[])
}
#[inline(always)]
fn greater_than_equal(&self, other: &Vec3<T>) -> Vec3<bool> {
zip_vec3!(self[] ge other[])
}
}
impl<T:Copy + Eq> EquableVector<T, Vec3<bool>> for Vec3<T> {
#[inline(always)]
fn equal(&self, other: &Vec3<T>) -> Vec3<bool> {
zip_vec3!(self[] eq other[])
}
#[inline(always)]
fn not_equal(&self, other: &Vec3<T>) -> Vec3<bool> {
zip_vec3!(self[] ne other[])
}
}
impl BooleanVector for Vec3<bool> {
#[inline(always)]
fn any(&self) -> bool {
self[0] || self[1] || self[2]
}
#[inline(always)]
fn all(&self) -> bool {
self[0] && self[1] && self[2]
}
#[inline(always)]
fn not(&self) -> Vec3<bool> {
Vector3::new(!self[0], !self[1], !self[2])
}
}
macro_rules! vec3_type(
($name:ident <bool>) => (
pub impl $name {
#[inline(always)] fn new(x: bool, y: bool, z: bool) -> $name { Vector3::new(x, y, z) }
#[inline(always)] fn from_value(v: bool) -> $name { Vector::from_value(v) }
#[inline(always)] fn dim() -> uint { 3 }
#[inline(always)] fn size_of() -> uint { sys::size_of::<$name>() }
}
);
($name:ident <$T:ty>) => (
pub impl $name {
#[inline(always)] fn new(x: $T, y: $T, z: $T) -> $name { Vector3::new(x, y, z) }
#[inline(always)] fn from_value(v: $T) -> $name { Vector::from_value(v) }
#[inline(always)] fn identity() -> $name { NumericVector::identity() }
#[inline(always)] fn zero() -> $name { NumericVector::zero() }
#[inline(always)] fn unit_x() -> $name { NumericVector3::unit_x() }
#[inline(always)] fn unit_y() -> $name { NumericVector3::unit_y() }
#[inline(always)] fn unit_z() -> $name { NumericVector3::unit_z() }
#[inline(always)] fn dim() -> uint { 3 }
#[inline(always)] fn size_of() -> uint { sys::size_of::<$name>() }
}
);
)
// GLSL-style type aliases, corresponding to Section 4.1.5 of the [GLSL 4.30.6 specification]
// (http://www.opengl.org/registry/doc/GLSLangSpec.4.30.6.pdf).
// a three-component single-precision floating-point vector
pub type vec3 = Vec3<f32>;
// a three-component double-precision floating-point vector
pub type dvec3 = Vec3<f64>;
// a three-component Boolean vector
pub type bvec3 = Vec3<bool>;
// a three-component signed integer vector
pub type ivec3 = Vec3<i32>;
// a three-component unsigned integer vector
pub type uvec3 = Vec3<u32>;
vec3_type!(vec3<f32>)
vec3_type!(dvec3<f64>)
vec3_type!(bvec3<bool>)
vec3_type!(ivec3<i32>)
vec3_type!(uvec3<u32>)
// Rust-style type aliases
pub type Vec3f = Vec3<float>;
pub type Vec3f32 = Vec3<f32>;
pub type Vec3f64 = Vec3<f64>;
pub type Vec3i = Vec3<int>;
pub type Vec3i8 = Vec3<i8>;
pub type Vec3i16 = Vec3<i16>;
pub type Vec3i32 = Vec3<i32>;
pub type Vec3i64 = Vec3<i64>;
pub type Vec3u = Vec3<uint>;
pub type Vec3u8 = Vec3<u8>;
pub type Vec3u16 = Vec3<u16>;
pub type Vec3u32 = Vec3<u32>;
pub type Vec3u64 = Vec3<u64>;
pub type Vec3b = Vec3<bool>;
vec3_type!(Vec3f<float>)
vec3_type!(Vec3f32<f32>)
vec3_type!(Vec3f64<f64>)
vec3_type!(Vec3i<int>)
vec3_type!(Vec3i8<i8>)
vec3_type!(Vec3i16<i16>)
vec3_type!(Vec3i32<i32>)
vec3_type!(Vec3i64<i64>)
vec3_type!(Vec3u<uint>)
vec3_type!(Vec3u8<u8>)
vec3_type!(Vec3u16<u16>)
vec3_type!(Vec3u32<u32>)
vec3_type!(Vec3u64<u64>)
vec3_type!(Vec3b<bool>)
/**
* A 4-dimensional vector
*
* # Type parameters
*
* * `T` - The type of the components. This is intended to support boolean,
* integer, unsigned integer, and floating point types.
*
* # Fields
*
* * `x` - the first component of the vector
* * `y` - the second component of the vector
* * `z` - the third component of the vector
* * `w` - the fourth component of the vector
*/
#[deriving(Eq)]
pub struct Vec4<T> { x: T, y: T, z: T, w: T }
impl<T:Copy + Eq> Vector<T> for Vec4<T> {
#[inline(always)]
fn from_value(value: T) -> Vec4<T> {
Vector4::new(value, value, value, value)
}
#[inline(always)]
fn to_ptr(&self) -> *T {
unsafe { cast::transmute(self) }
}
#[inline(always)]
fn index_mut(&mut self, i: uint) -> &'self mut T {
match i {
0 => &mut self.x,
1 => &mut self.y,
2 => &mut self.z,
3 => &mut self.w,
_ => fail!(fmt!("index out of bounds: expected an index from 0 to 3, but found %u", i))
}
}
#[inline(always)]
fn swap(&mut self, a: uint, b: uint) {
*self.index_mut(a) <-> *self.index_mut(b);
}
}
impl<T> Vector4<T> for Vec4<T> {
#[inline(always)]
fn new(x: T, y: T, z: T, w: T) -> Vec4<T> {
Vec4 { x: x, y: y, z: z, w: w }
}
}
impl<T:Copy + Eq> Index<uint, T> for Vec4<T> {
#[inline(always)]
fn index(&self, i: &uint) -> T {
unsafe { do vec::raw::buf_as_slice(self.to_ptr(), 4) |slice| { slice[*i] } }
}
}
impl<T:Copy + Number + Add<T,T> + Sub<T,T> + Mul<T,T> + Div<T,T> + Neg<T>> NumericVector<T> for Vec4<T> {
#[inline(always)]
fn identity() -> Vec4<T> {
Vector4::new(one::<T>(), one::<T>(), one::<T>(), one::<T>())
}
#[inline(always)]
fn zero() -> Vec4<T> {
Vector4::new(zero::<T>(), zero::<T>(), zero::<T>(), zero::<T>())
}
#[inline(always)]
fn is_zero(&self) -> bool {
self[0] == zero() &&
self[1] == zero() &&
self[2] == zero() &&
self[3] == zero()
}
#[inline(always)]
fn mul_t(&self, value: T) -> Vec4<T> {
zip_vec4!(self[] mul value)
}
#[inline(always)]
fn div_t(&self, value: T) -> Vec4<T> {
zip_vec4!(self[] div value)
}
#[inline(always)]
fn add_v(&self, other: &Vec4<T>) -> Vec4<T> {
zip_vec4!(self[] add other[])
}
#[inline(always)]
fn sub_v(&self, other: &Vec4<T>) -> Vec4<T> {
zip_vec4!(self[] sub other[])
}
#[inline(always)]
fn mul_v(&self, other: &Vec4<T>) -> Vec4<T> {
zip_vec4!(self[] mul other[])
}
#[inline(always)]
fn div_v(&self, other: &Vec4<T>) -> Vec4<T> {
zip_vec4!(self[] div other[])
}
#[inline(always)]
fn dot(&self, other: &Vec4<T>) -> T {
self[0] * other[0] +
self[1] * other[1] +
self[2] * other[2] +
self[3] * other[3]
}
#[inline(always)]
fn neg_self(&mut self) {
*self.index_mut(0) = -self[0];
*self.index_mut(1) = -self[1];
*self.index_mut(2) = -self[2];
*self.index_mut(3) = -self[3];
}
#[inline(always)]
fn mul_self_t(&mut self, value: T) {
zip_assign!(self[] mul_assign value ..4);
}
#[inline(always)]
fn div_self_t(&mut self, value: T) {
zip_assign!(self[] div_assign value ..4);
}
#[inline(always)]
fn add_self_v(&mut self, other: &Vec4<T>) {
zip_assign!(self[] add_assign other[] ..4);
}
#[inline(always)]
fn sub_self_v(&mut self, other: &Vec4<T>) {
zip_assign!(self[] sub_assign other[] ..4);
}
#[inline(always)]
fn mul_self_v(&mut self, other: &Vec4<T>) {
zip_assign!(self[] mul_assign other[] ..4);
}
#[inline(always)]
fn div_self_v(&mut self, other: &Vec4<T>) {
zip_assign!(self[] div_assign other[] ..4);
}
}
impl<T:Copy + Number + Add<T,T> + Sub<T,T> + Mul<T,T> + Div<T,T> + Neg<T>> Neg<Vec4<T>> for Vec4<T> {
#[inline(always)]
fn neg(&self) -> Vec4<T> {
Vector4::new(-self[0], -self[1], -self[2], -self[3])
}
}
impl<T:Copy + Number> NumericVector4<T> for Vec4<T> {
#[inline(always)]
fn unit_x() -> Vec4<T> {
Vector4::new(one::<T>(), zero::<T>(), zero::<T>(), zero::<T>())
}
#[inline(always)]
fn unit_y() -> Vec4<T> {
Vector4::new(zero::<T>(), one::<T>(), zero::<T>(), zero::<T>())
}
#[inline(always)]
fn unit_z() -> Vec4<T> {
Vector4::new(zero::<T>(), zero::<T>(), one::<T>(), zero::<T>())
}
#[inline(always)]
fn unit_w() -> Vec4<T> {
Vector4::new(zero::<T>(), zero::<T>(), zero::<T>(), one::<T>())
}
}
impl<T:Copy + Float + Add<T,T> + Sub<T,T> + Mul<T,T> + Div<T,T> + Neg<T>> EuclideanVector<T> for Vec4<T> {
#[inline(always)]
fn length2(&self) -> T {
self.dot(self)
}
#[inline(always)]
fn length(&self) -> T {
self.length2().sqrt()
}
#[inline(always)]
fn distance2(&self, other: &Vec4<T>) -> T {
other.sub_v(self).length2()
}
#[inline(always)]
fn distance(&self, other: &Vec4<T>) -> T {
other.distance2(self).sqrt()
}
#[inline(always)]
fn angle(&self, other: &Vec4<T>) -> T {
acos(self.dot(other) / (self.length() * other.length()))
}
#[inline(always)]
fn normalize(&self) -> Vec4<T> {
self.mul_t(one::<T>()/self.length())
}
#[inline(always)]
fn normalize_to(&self, length: T) -> Vec4<T> {
self.mul_t(length / self.length())
}
#[inline(always)]
fn lerp(&self, other: &Vec4<T>, amount: T) -> Vec4<T> {
self.add_v(&other.sub_v(self).mul_t(amount))
}
#[inline(always)]
fn normalize_self(&mut self) {
let n = one::<T>() / self.length();
self.mul_self_t(n);
}
#[inline(always)]
fn normalize_self_to(&mut self, length: T) {
let n = length / self.length();
self.mul_self_t(n);
}
fn lerp_self(&mut self, other: &Vec4<T>, amount: T) {
let v = other.sub_v(self).mul_t(amount);
self.add_self_v(&v);
}
}
impl<T:Copy + Float + FuzzyEq<T>> FuzzyEq<T> for Vec4<T> {
#[inline(always)]
fn fuzzy_eq(&self, other: &Vec4<T>) -> bool {
self.fuzzy_eq_eps(other, &Number::from(FUZZY_EPSILON))
}
#[inline(always)]
fn fuzzy_eq_eps(&self, other: &Vec4<T>, epsilon: &T) -> bool {
self[0].fuzzy_eq_eps(&other[0], epsilon) &&
self[1].fuzzy_eq_eps(&other[1], epsilon) &&
self[2].fuzzy_eq_eps(&other[2], epsilon) &&
self[3].fuzzy_eq_eps(&other[3], epsilon)
}
}
impl<T:Copy + Ord + Eq> OrdinalVector<T, Vec4<bool>> for Vec4<T> {
#[inline(always)]
fn less_than(&self, other: &Vec4<T>) -> Vec4<bool> {
zip_vec4!(self[] lt other[])
}
#[inline(always)]
fn less_than_equal(&self, other: &Vec4<T>) -> Vec4<bool> {
zip_vec4!(self[] le other[])
}
#[inline(always)]
fn greater_than(&self, other: &Vec4<T>) -> Vec4<bool> {
zip_vec4!(self[] gt other[])
}
#[inline(always)]
fn greater_than_equal(&self, other: &Vec4<T>) -> Vec4<bool> {
zip_vec4!(self[] ge other[])
}
}
impl<T:Copy + Eq> EquableVector<T, Vec4<bool>> for Vec4<T> {
#[inline(always)]
fn equal(&self, other: &Vec4<T>) -> Vec4<bool> {
zip_vec4!(self[] eq other[])
}
#[inline(always)]
fn not_equal(&self, other: &Vec4<T>) -> Vec4<bool> {
zip_vec4!(self[] ne other[])
}
}
impl BooleanVector for Vec4<bool> {
#[inline(always)]
fn any(&self) -> bool {
self[0] || self[1] || self[2] || self[3]
}
#[inline(always)]
fn all(&self) -> bool {
self[0] && self[1] && self[2] && self[3]
}
#[inline(always)]
fn not(&self) -> Vec4<bool> {
Vector4::new(!self[0], !self[1], !self[2], !self[3])
}
}
macro_rules! vec4_type(
($name:ident <bool>) => (
pub impl $name {
#[inline(always)] fn new(x: bool, y: bool, z: bool, w: bool) -> $name { Vector4::new(x, y, z, w) }
#[inline(always)] fn from_value(v: bool) -> $name { Vector::from_value(v) }
#[inline(always)] fn dim() -> uint { 4 }
#[inline(always)] fn size_of() -> uint { sys::size_of::<$name>() }
}
);
($name:ident <$T:ty>) => (
pub impl $name {
#[inline(always)] fn new(x: $T, y: $T, z: $T, w: $T) -> $name { Vector4::new(x, y, z, w) }
#[inline(always)] fn from_value(v: $T) -> $name { Vector::from_value(v) }
#[inline(always)] fn identity() -> $name { NumericVector::identity() }
#[inline(always)] fn zero() -> $name { NumericVector::zero() }
#[inline(always)] fn unit_x() -> $name { NumericVector4::unit_x() }
#[inline(always)] fn unit_y() -> $name { NumericVector4::unit_y() }
#[inline(always)] fn unit_z() -> $name { NumericVector4::unit_z() }
#[inline(always)] fn unit_w() -> $name { NumericVector4::unit_w() }
#[inline(always)] fn dim() -> uint { 4 }
#[inline(always)] fn size_of() -> uint { sys::size_of::<$name>() }
}
);
)
// GLSL-style type aliases, corresponding to Section 4.1.5 of the [GLSL 4.30.6 specification]
// (http://www.opengl.org/registry/doc/GLSLangSpec.4.30.6.pdf).
// a four-component single-precision floating-point vector
pub type vec4 = Vec4<f32>;
// a four-component double-precision floating-point vector
pub type dvec4 = Vec4<f64>;
// a four-component Boolean vector
pub type bvec4 = Vec4<bool>;
// a four-component signed integer vector
pub type ivec4 = Vec4<i32>;
// a four-component unsigned integer vector
pub type uvec4 = Vec4<u32>;
vec4_type!(vec4<f32>)
vec4_type!(dvec4<f64>)
vec4_type!(bvec4<bool>)
vec4_type!(ivec4<i32>)
vec4_type!(uvec4<u32>)
// Rust-style type aliases
pub type Vec4f = Vec4<float>;
pub type Vec4f32 = Vec4<f32>;
pub type Vec4f64 = Vec4<f64>;
pub type Vec4i = Vec4<int>;
pub type Vec4i8 = Vec4<i8>;
pub type Vec4i16 = Vec4<i16>;
pub type Vec4i32 = Vec4<i32>;
pub type Vec4i64 = Vec4<i64>;
pub type Vec4u = Vec4<uint>;
pub type Vec4u8 = Vec4<u8>;
pub type Vec4u16 = Vec4<u16>;
pub type Vec4u32 = Vec4<u32>;
pub type Vec4u64 = Vec4<u64>;
pub type Vec4b = Vec4<bool>;
vec4_type!(Vec4f<float>)
vec4_type!(Vec4f32<f32>)
vec4_type!(Vec4f64<f64>)
vec4_type!(Vec4i<int>)
vec4_type!(Vec4i8<i8>)
vec4_type!(Vec4i16<i16>)
vec4_type!(Vec4i32<i32>)
vec4_type!(Vec4i64<i64>)
vec4_type!(Vec4u<uint>)
vec4_type!(Vec4u8<u8>)
vec4_type!(Vec4u16<u16>)
vec4_type!(Vec4u32<u32>)
vec4_type!(Vec4u64<u64>)
vec4_type!(Vec4b<bool>)