fixings for physics
This commit is contained in:
parent
fa8ea26ddc
commit
bf6f37fa00
@ -234,7 +234,7 @@ impl Ratio64Vec2{
|
|||||||
Self{x,y}
|
Self{x,y}
|
||||||
}
|
}
|
||||||
#[inline]
|
#[inline]
|
||||||
pub fn mul_int(self,rhs:glam::I64Vec2)->glam::I64Vec2{
|
pub fn mul_int(&self,rhs:glam::I64Vec2)->glam::I64Vec2{
|
||||||
glam::i64vec2(
|
glam::i64vec2(
|
||||||
self.x.mul_int(rhs.x),
|
self.x.mul_int(rhs.x),
|
||||||
self.y.mul_int(rhs.y),
|
self.y.mul_int(rhs.y),
|
||||||
@ -654,6 +654,17 @@ impl std::ops::SubAssign<Planar64Vec3> for Planar64Vec3{
|
|||||||
*self=*self-rhs
|
*self=*self-rhs
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
impl std::ops::Mul<Planar64Vec3> for Planar64Vec3{
|
||||||
|
type Output=Planar64Vec3;
|
||||||
|
#[inline]
|
||||||
|
fn mul(self, rhs: Planar64Vec3) -> Self::Output {
|
||||||
|
Planar64Vec3(glam::i64vec3(
|
||||||
|
(((self.0.x as i128)*(rhs.0.x as i128))>>64) as i64,
|
||||||
|
(((self.0.y as i128)*(rhs.0.y as i128))>>64) as i64,
|
||||||
|
(((self.0.z as i128)*(rhs.0.z as i128))>>64) as i64
|
||||||
|
))
|
||||||
|
}
|
||||||
|
}
|
||||||
impl std::ops::Mul<Planar64> for Planar64Vec3{
|
impl std::ops::Mul<Planar64> for Planar64Vec3{
|
||||||
type Output=Planar64Vec3;
|
type Output=Planar64Vec3;
|
||||||
#[inline]
|
#[inline]
|
||||||
|
Loading…
Reference in New Issue
Block a user