196 lines
4.9 KiB
C++
196 lines
4.9 KiB
C++
#ifndef FASTMATH_H
|
|
#define FASTMATH_H
|
|
|
|
#include <cmath>
|
|
#include <iostream>
|
|
#include <sstream>
|
|
#include <stdint.h>
|
|
#include <stdlib.h>
|
|
#include <vector>
|
|
#define SHIFT_AMOUNT 16
|
|
#define HALF_SHIFT (SHIFT_AMOUNT / 2)
|
|
|
|
#define SHIFT_MASK ((1 << SHIFT_AMOUNT) - 1)
|
|
#define TO_FLOAT(x) \
|
|
(((float)(x >> SHIFT_AMOUNT)) + \
|
|
((double)(x & SHIFT_MASK) / (1 << SHIFT_AMOUNT)))
|
|
#define TO_INT(x) ((int32_t)(x * (1 << SHIFT_AMOUNT)))
|
|
#define MUL_F(a, b) (((a) >> HALF_SHIFT) * ((b) >> HALF_SHIFT))
|
|
#define DIV_F(a, b) ((((a) << HALF_SHIFT) / (b)) << HALF_SHIFT)
|
|
|
|
struct decimal {
|
|
|
|
const int32_t i;
|
|
|
|
decimal() : i(0) {}
|
|
decimal(float i) : i(TO_INT(i)) {}
|
|
decimal(double i) : i(TO_INT(i)) {}
|
|
decimal(int32_t i) : i(i) {}
|
|
|
|
friend std::ostream &operator<<(std::ostream &os, const decimal &d) {
|
|
return (os << TO_FLOAT(d.i));
|
|
}
|
|
|
|
friend decimal operator+(const decimal &d1, const decimal &d2) {
|
|
return {d1.i + d2.i};
|
|
}
|
|
decimal &operator+=(const decimal &d) { return (*this) = {i + d.i}; }
|
|
|
|
friend decimal operator-(const decimal &d1, const decimal &d2) {
|
|
return {d1.i - d2.i};
|
|
}
|
|
friend decimal operator-(const decimal &d) { return {-d.i}; }
|
|
|
|
friend decimal operator*(const decimal &d1, const decimal &d2) {
|
|
return {MUL_F(d1.i, d2.i)};
|
|
}
|
|
|
|
friend decimal operator/(const decimal &d1, const decimal &d2) {
|
|
return {DIV_F(d1.i, d2.i)};
|
|
}
|
|
|
|
friend bool operator<(const decimal &d1, const decimal &d2) {
|
|
return d1.i < d2.i;
|
|
}
|
|
|
|
friend bool operator>(const decimal &d1, const decimal &d2) {
|
|
return d1.i > d2.i;
|
|
}
|
|
|
|
friend bool operator==(const decimal &d1, const decimal &d2) {
|
|
return d1.i == d2.i;
|
|
}
|
|
|
|
decimal &operator=(decimal const &in) {
|
|
if (this != &in) {
|
|
std::destroy_at(this);
|
|
std::construct_at(this, in);
|
|
}
|
|
return *this;
|
|
}
|
|
|
|
decimal sqrt() { return {((int32_t)sqrtf(i)) << HALF_SHIFT}; }
|
|
|
|
float to_float() { return TO_FLOAT(i); }
|
|
};
|
|
|
|
template <int n, class Dev> struct vec {
|
|
|
|
vec(decimal newV[n]) {
|
|
for (int i = 0; i < n; i++) {
|
|
v[i] = newV[i];
|
|
}
|
|
}
|
|
|
|
vec(std::vector<decimal> newV) {
|
|
for (int i = 0; i < n; i++) {
|
|
v[i] = newV[i];
|
|
}
|
|
}
|
|
|
|
vec() : v{} {}
|
|
|
|
friend Dev operator+(const vec<n, Dev> &v1, const vec<n, Dev> &v2) {
|
|
Dev newV = {};
|
|
|
|
for (int i = 0; i < n; i++) {
|
|
newV.v[i] = v1.v[i] + v2.v[i];
|
|
}
|
|
return newV;
|
|
}
|
|
|
|
friend Dev operator-(const vec<n, Dev> &v1, const vec<n, Dev> &v2) {
|
|
Dev newV = {};
|
|
|
|
for (int i = 0; i < n; i++) {
|
|
newV.v[i] = v1.v[i] - v2.v[i];
|
|
}
|
|
return static_cast<Dev>(newV);
|
|
}
|
|
|
|
friend std::ostream &operator<<(std::ostream &os, const vec<n, Dev> &v) {
|
|
os << "(" << v.v[0];
|
|
for (int i = 1; i < n; i++) {
|
|
os << ", " << v.v[i];
|
|
}
|
|
return (os << ")" << std::endl);
|
|
}
|
|
|
|
Dev operator-() {
|
|
Dev newV = {};
|
|
for (int i = 0; i < n; i++) {
|
|
newV.v[i] = -v[i];
|
|
}
|
|
return newV;
|
|
}
|
|
|
|
friend Dev operator*(const vec<n, Dev> &v, const decimal &d) {
|
|
int32_t f = d.i >> HALF_SHIFT;
|
|
|
|
Dev newV = {};
|
|
for (int i = 0; i < n; i++) {
|
|
newV.v[i] = (v.v[i].i >> HALF_SHIFT) * f;
|
|
}
|
|
return newV;
|
|
}
|
|
|
|
friend Dev operator*(const decimal &d, const vec<n, Dev> &v) {
|
|
return v * d;
|
|
}
|
|
|
|
decimal operator*(const vec<n, Dev> &vec) {
|
|
decimal res;
|
|
for (int i = 0; i < n; i++) {
|
|
res += vec.v[i] * v[i];
|
|
}
|
|
return res;
|
|
}
|
|
|
|
friend bool operator==(const vec<n, Dev> &v1, const vec<n, Dev> &v2) {
|
|
bool res = true;
|
|
for (int i = 0; i < n; i++) {
|
|
res &= v1.v[i] == v2.v[i];
|
|
}
|
|
return res;
|
|
}
|
|
|
|
decimal len_sq() { return *this * *this; }
|
|
|
|
decimal len() { return this->len_sq().sqrt(); }
|
|
|
|
Dev normalize() {
|
|
decimal f = decimal(1.0) / this->len();
|
|
return (*this * f);
|
|
}
|
|
|
|
protected:
|
|
decimal v[n];
|
|
};
|
|
|
|
struct vec3 : public vec<3, vec3> {
|
|
|
|
vec3() : vec<3, vec3>() {}
|
|
|
|
vec3(float x, float y, float z)
|
|
: vec<3, vec3>({decimal(x), decimal(y), decimal(z)}) {}
|
|
|
|
vec3(double x, double y, double z)
|
|
: vec<3, vec3>({decimal(x), decimal(y), decimal(z)}) {}
|
|
|
|
vec3(int32_t x, int32_t y, int32_t z)
|
|
: vec<3, vec3>({decimal(x), decimal(y), decimal(z)}) {}
|
|
|
|
vec3(decimal x, decimal y, decimal z) : vec<3, vec3>({x, y, z}) {}
|
|
|
|
decimal x() const { return v[0]; }
|
|
decimal y() const { return v[1]; }
|
|
decimal z() const { return v[2]; }
|
|
|
|
vec3 cross(const vec3 &v) {
|
|
return vec3((y() * v.z()) - (z() * v.y()),
|
|
(z() * v.x()) - (x() * v.z()),
|
|
(x() * v.y()) - (y() * v.x()));
|
|
}
|
|
};
|
|
#endif
|