use std::{cmp, ops, ops::Shl};
#[derive(Clone, Debug, Copy)]
pub struct FPNum {
is_negative: bool,
value: u64,
}
impl FPNum {
#[inline]
pub fn new(numerator: i32, denominator: u32) -> Self {
FPNum::from(numerator) / denominator
}
#[inline]
pub fn signum(&self) -> i8 {
if self.is_negative {
-1
} else {
1
}
}
#[inline]
pub const fn is_negative(&self) -> bool {
self.is_negative
}
#[inline]
pub const fn is_positive(&self) -> bool {
!self.is_negative
}
#[inline]
pub const fn is_zero(&self) -> bool {
self.value == 0
}
#[inline]
pub const fn abs(&self) -> Self {
Self {
is_negative: false,
value: self.value,
}
}
#[inline]
pub fn round(&self) -> i32 {
if self.is_negative {
-((self.value >> 32) as i32)
} else {
(self.value >> 32) as i32
}
}
#[inline]
pub const fn abs_round(&self) -> u32 {
(self.value >> 32) as u32
}
#[inline]
pub fn sqr(&self) -> Self {
Self {
is_negative: false,
value: ((self.value as u128).pow(2) >> 32) as u64,
}
}
pub fn sqrt(&self) -> Self {
debug_assert!(!self.is_negative);
let mut t: u64 = 0x4000000000000000;
let mut r: u64 = 0;
let mut q = self.value;
for _ in 0..32 {
let s = r + t;
r >>= 1;
if s <= q {
q -= s;
r += t;
}
t >>= 2;
}
Self {
is_negative: false,
value: r << 16,
}
}
#[inline]
pub const fn with_sign(&self, is_negative: bool) -> FPNum {
FPNum {
is_negative,
..*self
}
}
#[inline]
pub const fn with_sign_as(self, other: FPNum) -> FPNum {
self.with_sign(other.is_negative)
}
#[inline]
pub const fn point(self) -> FPPoint {
FPPoint::new(self, self)
}
}
impl From<i32> for FPNum {
#[inline]
fn from(n: i32) -> Self {
FPNum {
is_negative: n < 0,
value: (n.abs() as u64) << 32,
}
}
}
impl From<u32> for FPNum {
#[inline]
fn from(n: u32) -> Self {
Self {
is_negative: false,
value: (n as u64) << 32,
}
}
}
impl From<FPNum> for f64 {
#[inline]
fn from(n: FPNum) -> Self {
if n.is_negative {
n.value as f64 / (-0x10000000 as f64)
} else {
n.value as f64 / 0x10000000 as f64
}
}
}
impl PartialEq for FPNum {
#[inline]
fn eq(&self, other: &Self) -> bool {
self.value == other.value && (self.is_negative == other.is_negative || self.value == 0)
}
}
impl Eq for FPNum {}
impl PartialOrd for FPNum {
#[inline]
fn partial_cmp(&self, rhs: &Self) -> std::option::Option<std::cmp::Ordering> {
Some(self.cmp(rhs))
}
}
impl Ord for FPNum {
#[inline]
fn cmp(&self, rhs: &Self) -> cmp::Ordering {
#[inline]
fn extend(n: &FPNum) -> i128 {
if n.is_negative {
-(n.value as i128)
} else {
n.value as i128
}
}
extend(self).cmp(&(extend(rhs)))
}
}
impl ops::Add for FPNum {
type Output = Self;
#[inline]
fn add(self, rhs: Self) -> Self {
if self.is_negative == rhs.is_negative {
Self {
is_negative: self.is_negative,
value: self.value + rhs.value,
}
} else if self.value > rhs.value {
Self {
is_negative: self.is_negative,
value: self.value - rhs.value,
}
} else {
Self {
is_negative: rhs.is_negative,
value: rhs.value - self.value,
}
}
}
}
impl ops::Sub for FPNum {
type Output = Self;
#[inline]
fn sub(self, rhs: Self) -> Self {
if self.is_negative == rhs.is_negative {
if self.value > rhs.value {
Self {
is_negative: self.is_negative,
value: self.value - rhs.value,
}
} else {
Self {
is_negative: !rhs.is_negative,
value: rhs.value - self.value,
}
}
} else {
Self {
is_negative: self.is_negative,
value: self.value + rhs.value,
}
}
}
}
impl ops::Neg for FPNum {
type Output = Self;
#[inline]
fn neg(self) -> Self {
Self {
is_negative: !self.is_negative,
value: self.value,
}
}
}
impl ops::Mul for FPNum {
type Output = Self;
#[inline]
fn mul(self, rhs: Self) -> Self {
Self {
is_negative: self.is_negative ^ rhs.is_negative,
value: ((self.value as u128 * rhs.value as u128) >> 32) as u64,
}
}
}
impl ops::Mul<i32> for FPNum {
type Output = Self;
#[inline]
fn mul(self, rhs: i32) -> Self {
Self {
is_negative: self.is_negative ^ (rhs < 0),
value: self.value * rhs.abs() as u64,
}
}
}
impl ops::Div for FPNum {
type Output = Self;
#[inline]
fn div(self, rhs: Self) -> Self {
Self {
is_negative: self.is_negative ^ rhs.is_negative,
value: (((self.value as u128) << 32) / rhs.value as u128) as u64,
}
}
}
impl ops::Div<i32> for FPNum {
type Output = Self;
#[inline]
fn div(self, rhs: i32) -> Self {
Self {
is_negative: self.is_negative ^ (rhs < 0),
value: self.value / rhs.abs() as u64,
}
}
}
impl ops::Div<u32> for FPNum {
type Output = Self;
#[inline]
fn div(self, rhs: u32) -> Self {
Self {
is_negative: self.is_negative,
value: self.value / rhs as u64,
}
}
}
#[macro_export]
macro_rules! fp {
(-$n: literal / $d: literal) => {
FPNum::new(-$n, $d)
};
($n: literal / $d: literal) => {
FPNum::new($n, $d)
};
(-$n: literal) => {
FPNum::from(-$n)
};
($n: literal) => {
FPNum::from($n)
};
}
const LINEARIZE_TRESHOLD: u64 = 0x1_0000;
#[derive(Clone, Copy, Debug)]
pub struct FPPoint {
x_is_negative: bool,
y_is_negative: bool,
x_value: u64,
y_value: u64,
}
impl FPPoint {
#[inline]
pub const fn new(x: FPNum, y: FPNum) -> Self {
Self {
x_is_negative: x.is_negative,
y_is_negative: y.is_negative,
x_value: x.value,
y_value: y.value,
}
}
#[inline]
pub fn zero() -> FPPoint {
FPPoint::new(fp!(0), fp!(0))
}
#[inline]
pub fn unit_x() -> FPPoint {
FPPoint::new(fp!(1), fp!(0))
}
#[inline]
pub fn unit_y() -> FPPoint {
FPPoint::new(fp!(0), fp!(1))
}
#[inline]
pub const fn x(&self) -> FPNum {
FPNum {
is_negative: self.x_is_negative,
value: self.x_value,
}
}
#[inline]
pub const fn y(&self) -> FPNum {
FPNum {
is_negative: self.y_is_negative,
value: self.y_value,
}
}
#[inline]
pub fn is_zero(&self) -> bool {
self.x().is_zero() && self.y().is_zero()
}
#[inline]
pub fn max_norm(&self) -> FPNum {
std::cmp::max(self.x().abs(), self.y().abs())
}
#[inline]
pub fn sqr_distance(&self) -> FPNum {
self.x().sqr() + self.y().sqr()
}
#[inline]
pub fn distance(&self) -> FPNum {
let r = self.x_value | self.y_value;
if r < LINEARIZE_TRESHOLD {
FPNum::from(r as u32)
} else {
let mut sqr: u128 = (self.x_value as u128).pow(2) + (self.y_value as u128).pow(2);
let mut t: u128 = 0x40000000_00000000_00000000_00000000;
let mut r: u128 = 0;
for _ in 0..64 {
let s = r + t;
r >>= 1;
if s <= sqr {
sqr -= s;
r += t;
}
t >>= 2;
}
FPNum {
is_negative: false,
value: r as u64,
}
}
}
#[inline]
pub fn is_in_range(&self, radius: FPNum) -> bool {
self.max_norm() < radius && self.sqr_distance() < radius.sqr()
}
#[inline]
pub fn dot(&self, other: &FPPoint) -> FPNum {
self.x() * other.x() + self.y() * other.y()
}
}
impl PartialEq for FPPoint {
#[inline]
fn eq(&self, other: &Self) -> bool {
self.x() == other.x() && self.y() == other.y()
}
}
impl Eq for FPPoint {}
impl ops::Neg for FPPoint {
type Output = Self;
#[inline]
fn neg(self) -> Self {
Self::new(-self.x(), -self.y())
}
}
macro_rules! bin_op_impl {
($op: ty, $name: tt) => {
impl $op for FPPoint {
type Output = Self;
#[inline]
fn $name(self, rhs: Self) -> Self {
Self::new(self.x().$name(rhs.x()), self.y().$name(rhs.y()))
}
}
};
}
macro_rules! right_scalar_bin_op_impl {
($($op: tt)::+, $name: tt) => {
impl $($op)::+<FPNum> for FPPoint {
type Output = Self;
#[inline]
fn $name(self, rhs: FPNum) -> Self {
Self::new(self.x().$name(rhs),
self.y().$name(rhs))
}
}
};
($($op: tt)::+<$arg: tt>, $name: tt) => {
impl $($op)::+<$arg> for FPPoint {
type Output = Self;
#[inline]
fn $name(self, rhs: $arg) -> Self {
Self::new(self.x().$name(rhs),
self.y().$name(rhs))
}
}
}
}
macro_rules! left_scalar_bin_op_impl {
($($op: tt)::+, $name: tt) => {
impl $($op)::+<FPPoint> for FPNum {
type Output = FPPoint;
#[inline]
fn $name(self, rhs: FPPoint) -> Self::Output {
Self::Output::new(self.$name(rhs.x()),
self.$name(rhs.y()))
}
}
}
}
bin_op_impl!(ops::Add, add);
bin_op_impl!(ops::Sub, sub);
bin_op_impl!(ops::Mul, mul);
bin_op_impl!(ops::Div, div);
right_scalar_bin_op_impl!(ops::Add, add);
right_scalar_bin_op_impl!(ops::Mul, mul);
right_scalar_bin_op_impl!(ops::Sub, sub);
right_scalar_bin_op_impl!(ops::Div, div);
right_scalar_bin_op_impl!(ops::Div<u32>, div);
left_scalar_bin_op_impl!(ops::Mul, mul);
macro_rules! bin_assign_op_impl {
($typ: tt, $($op: tt)::+, $name: tt, $delegate: tt) => {
bin_assign_op_impl!($typ, $($op)::+<$typ>, $name, $delegate);
};
($typ: tt, $($op: tt)::+<$arg: tt>, $name: tt, $delegate: tt) => {
impl $($op)::+<$arg> for $typ {
#[inline]
fn $name(&mut self, rhs: $arg) {
*self = *self $delegate rhs;
}
}
}
}
bin_assign_op_impl!(FPNum, ops::AddAssign, add_assign, +);
bin_assign_op_impl!(FPNum, ops::SubAssign, sub_assign, -);
bin_assign_op_impl!(FPNum, ops::MulAssign, mul_assign, *);
bin_assign_op_impl!(FPNum, ops::DivAssign, div_assign, /);
bin_assign_op_impl!(FPNum, ops::MulAssign<i32>, mul_assign, *);
bin_assign_op_impl!(FPNum, ops::DivAssign<i32>, div_assign, /);
bin_assign_op_impl!(FPNum, ops::DivAssign<u32>, div_assign, /);
bin_assign_op_impl!(FPPoint, ops::AddAssign, add_assign, +);
bin_assign_op_impl!(FPPoint, ops::SubAssign, sub_assign, -);
bin_assign_op_impl!(FPPoint, ops::MulAssign, mul_assign, *);
bin_assign_op_impl!(FPPoint, ops::DivAssign, div_assign, /);
bin_assign_op_impl!(FPPoint, ops::AddAssign<FPNum>, add_assign, +);
bin_assign_op_impl!(FPPoint, ops::SubAssign<FPNum>, sub_assign, -);
bin_assign_op_impl!(FPPoint, ops::MulAssign<FPNum>, mul_assign, *);
bin_assign_op_impl!(FPPoint, ops::DivAssign<FPNum>, div_assign, /);
pub fn distance<T>(x: T, y: T) -> FPNum
where
T: Into<i64> + std::fmt::Debug,
{
let mut sqr: u128 = (x.into().pow(2) as u128).shl(64) + (y.into().pow(2) as u128).shl(64);
let mut t: u128 = 0x40000000_00000000_00000000_00000000;
let mut r: u128 = 0;
for _ in 0..64 {
let s = r + t;
r >>= 1;
if s <= sqr {
sqr -= s;
r += t;
}
t >>= 2;
}
FPNum {
is_negative: false,
value: r as u64,
}
}
/* TODO:
AngleSin
AngleCos
*/
#[cfg(test)]
#[test]
fn basics() {
let n = fp!(15 / 2);
assert!(n.is_positive());
assert!(!n.is_negative());
assert!(!(-n).is_positive());
assert!((-n).is_negative());
assert_eq!(-(-n), n);
assert_eq!((-n).abs(), n);
assert_eq!(-n, fp!(-15 / 2));
assert_eq!(n.round(), 7);
assert_eq!((-n).round(), -7);
}
#[test]
fn zero() {
let z = fp!(0);
let n = fp!(15 / 2);
assert!(z.is_zero());
assert!(z.is_positive());
assert!((-z).is_negative);
assert_eq!(n - n, z);
assert_eq!(-n + n, z);
assert_eq!(n.with_sign_as(-n), -n);
}
#[test]
fn ord() {
let z = fp!(0);
let n1_5 = fp!(3 / 2);
let n2_25 = fp!(9 / 4);
assert!(!(z > z));
assert!(!(z < z));
assert!(n2_25 > n1_5);
assert!(-n2_25 < n1_5);
assert!(-n2_25 < -n1_5);
}
#[test]
fn arith() {
let n1_5 = fp!(3 / 2);
let n2_25 = fp!(9 / 4);
let n_0_15 = fp!(-15 / 100);
assert_eq!(n1_5 + n1_5, fp!(3));
assert_eq!(-n1_5 - n1_5, fp!(-3));
assert_eq!(n1_5 * n1_5, n2_25);
assert_eq!(-n1_5 * -n1_5, n2_25);
assert_eq!(n1_5 * -n1_5, -n2_25);
assert_eq!(-n1_5 * n1_5, -n2_25);
assert_eq!(-n2_25 / -n1_5, n1_5);
assert_eq!(n1_5 / -10, n_0_15);
assert_eq!(n1_5.sqr(), n2_25);
assert_eq!((-n1_5).sqr(), n2_25);
assert_eq!(n2_25.sqrt(), n1_5);
assert_eq!((n1_5 * n1_5 * n1_5.sqr()).sqrt(), n2_25);
let mut m = fp!(1);
m += n1_5;
assert_eq!(m, fp!(5 / 2));
}
#[test]
fn test_distance_high_values() {
assert_eq!(distance(1_000_000i32, 0), fp!(1_000_000));
assert_eq!(
FPPoint::new(fp!(1_000_000), fp!(0)).distance(),
fp!(1_000_000)
);
}
#[test]
fn point() {
let z = FPPoint::zero();
let n = fp!(16 / 9);
let p = FPPoint::new(fp!(1), fp!(-2));
assert_eq!(p.sqr_distance(), fp!(5));
assert_eq!(p + -p, FPPoint::zero());
assert_eq!(p * z, z);
assert_eq!(p.dot(&z), fp!(0));
assert_eq!(n * p, p * n);
assert_eq!(distance(4, 3), fp!(5));
assert_eq!(p * fp!(-3), FPPoint::new(fp!(-3), fp!(6)));
assert_eq!(p.max_norm(), fp!(2));
}