mirror of https://github.com/zeldaret/tp.git
509 lines
11 KiB
C++
509 lines
11 KiB
C++
#ifndef JGEOMETRY_H
|
|
#define JGEOMETRY_H
|
|
|
|
#include "dolphin/mtx.h"
|
|
#include "math.h"
|
|
#include "JSystem/JMath/JMath.h"
|
|
|
|
namespace JGeometry {
|
|
|
|
template<typename T>
|
|
struct TUtil {
|
|
static inline T clamp(T v, T min, T max) {
|
|
if (v < min) {
|
|
return min;
|
|
}
|
|
if (v > max) {
|
|
return max;
|
|
}
|
|
return v;
|
|
}
|
|
};
|
|
|
|
template<>
|
|
struct TUtil<f32> {
|
|
static inline f32 clamp(f32 v, f32 min, f32 max) {
|
|
if (v < min) {
|
|
return min;
|
|
}
|
|
if (v > max) {
|
|
return max;
|
|
}
|
|
return v;
|
|
}
|
|
static inline f32 epsilon() { return 32.0f * FLT_EPSILON; }
|
|
static inline f32 PI() { return 3.1415927f; }
|
|
static inline f32 inv_sqrt(f32 x) {
|
|
if (x <= 0.0f) {
|
|
return x;
|
|
}
|
|
f32 root = __frsqrte(x);
|
|
root = 0.5f * root * (3.0f - x * (root * root));
|
|
return root;
|
|
}
|
|
};
|
|
|
|
template<>
|
|
struct TUtil<double> {
|
|
static inline double epsilon() { return 32.0f * FLT_EPSILON; }
|
|
static inline double one() { return 1.0; }
|
|
static inline double atan2(double x, double y) { return ::atan2(x, y); }
|
|
static inline double asin(double x) { return ::asin(x); }
|
|
static inline double halfPI() { return 1.5707963267948966; }
|
|
};
|
|
|
|
template <typename T>
|
|
struct TVec3 {
|
|
T x;
|
|
T y;
|
|
T z;
|
|
|
|
void set(const TVec3& other) {
|
|
x = other.x;
|
|
y = other.y;
|
|
z = other.z;
|
|
}
|
|
};
|
|
|
|
template <>
|
|
struct TVec3<double> {
|
|
double x, y, z;
|
|
|
|
void set(double x_, double y_, double z_) {
|
|
x = x_;
|
|
y = y_;
|
|
z = z_;
|
|
}
|
|
|
|
inline TVec3<double>& operator*=(double b) {
|
|
scale(b);
|
|
return *this;
|
|
}
|
|
|
|
void scale(double b) {
|
|
x *= b;
|
|
y *= b;
|
|
z *= b;
|
|
}
|
|
};
|
|
|
|
template <>
|
|
struct TVec3<s16> {
|
|
s16 x, y, z;
|
|
|
|
TVec3() {}
|
|
|
|
TVec3(s16 x, s16 y, s16 z) {
|
|
set(x, y, z);
|
|
}
|
|
|
|
TVec3& operator=(const TVec3& b) {
|
|
// Force copies to use lwz/lha
|
|
*((s32*)this) = *((s32*)&b);
|
|
z = b.z;
|
|
return *this;
|
|
}
|
|
|
|
void set(s16 x_, s16 y_, s16 z_) {
|
|
x = x_;
|
|
y = y_;
|
|
z = z_;
|
|
}
|
|
};
|
|
|
|
inline void setTVec3f(const f32* vec_a, f32* vec_b) {
|
|
const register f32* v_a = vec_a;
|
|
register f32* v_b = vec_b;
|
|
|
|
register f32 a_x;
|
|
register f32 b_x;
|
|
|
|
#ifdef __MWERKS__
|
|
asm {
|
|
psq_l a_x, 0(v_a), 0, 0
|
|
lfs b_x, 8(v_a)
|
|
psq_st a_x, 0(v_b), 0, 0
|
|
stfs b_x, 8(v_b)
|
|
};
|
|
#endif
|
|
}
|
|
|
|
// Until we figure out TVec3 ctors
|
|
inline void setTVec3f(const Vec& vec_a, Vec& vec_b) {
|
|
setTVec3f(&vec_a.x, &vec_b.x);
|
|
}
|
|
|
|
inline float fsqrt_step(float mag) {
|
|
f32 root = __frsqrte(mag);
|
|
return 0.5f * root * (3.0f - mag * (root * root));
|
|
}
|
|
|
|
inline void mulInternal(register const f32* a, register const f32* b, register float* dst) {
|
|
register f32 a_x_y;
|
|
register f32 b_x_y;
|
|
register f32 x_y;
|
|
register f32 za;
|
|
register f32 zb;
|
|
register f32 z;
|
|
#ifdef __MWERKS__
|
|
asm {
|
|
psq_l a_x_y, 0(a), 0, 0
|
|
psq_l b_x_y, 0(b), 0, 0
|
|
ps_mul x_y, a_x_y, b_x_y
|
|
psq_st x_y, 0(dst), 0, 0
|
|
};
|
|
dst[2] = a[2] * b[2];
|
|
#endif
|
|
}
|
|
|
|
template <>
|
|
struct TVec3<f32> : public Vec {
|
|
inline TVec3(const Vec& i_vec) {
|
|
setTVec3f(&i_vec.x, &x);
|
|
}
|
|
|
|
inline TVec3(const TVec3<f32>& i_vec) {
|
|
setTVec3f(&i_vec.x, &x);
|
|
}
|
|
|
|
template<class U>
|
|
TVec3(U x, U y, U z) {
|
|
set(x, y, z);
|
|
}
|
|
|
|
TVec3() {}
|
|
|
|
operator Vec*() { return (Vec*)&x; }
|
|
operator const Vec*() const { return (Vec*)&x; }
|
|
|
|
template<class U>
|
|
void set(const TVec3<U>& other) {
|
|
x = other.x;
|
|
y = other.y;
|
|
z = other.z;
|
|
}
|
|
|
|
void set(const Vec& other) {
|
|
x = other.x;
|
|
y = other.y;
|
|
z = other.z;
|
|
}
|
|
|
|
template<class U>
|
|
void set(U x_, U y_, U z_) {
|
|
x = x_;
|
|
y = y_;
|
|
z = z_;
|
|
}
|
|
|
|
inline void add(const TVec3<f32>& b) {
|
|
JMathInlineVEC::C_VECAdd((Vec*)&x, (Vec*)&b.x, (Vec*)&x);
|
|
}
|
|
|
|
void zero() { x = y = z = 0.0f; }
|
|
|
|
void mul(const TVec3<f32>& a, const TVec3<f32>& b) {
|
|
mulInternal(&a.x, &b.x, &this->x);
|
|
}
|
|
|
|
inline void mul(const TVec3<f32>& a) {
|
|
mul(*this, a);
|
|
}
|
|
|
|
inline TVec3<f32>& operator=(const Vec& b) {
|
|
setTVec3f(&b.x, &this->x);
|
|
return *this;
|
|
}
|
|
|
|
inline TVec3<f32>& operator=(const TVec3<f32>& b) {
|
|
set(b.x, b.y, b.z);
|
|
return *this;
|
|
}
|
|
|
|
inline TVec3<f32>& operator+=(const TVec3<f32>& b) {
|
|
add(b);
|
|
return *this;
|
|
}
|
|
|
|
inline TVec3<f32> operator+(const TVec3<f32>& b) {
|
|
TVec3<f32> a = *this;
|
|
a += b;
|
|
return a;
|
|
}
|
|
|
|
// inline TVec3<f32> operator+(const TVec3<f32>& b) {
|
|
// TVec3<f32> res(*(Vec*)this);
|
|
// res += b;
|
|
// return res;
|
|
// }
|
|
|
|
f32 squared() const {
|
|
return JMathInlineVEC::C_VECSquareMag((Vec*)&x);
|
|
}
|
|
|
|
f32 normalize() {
|
|
f32 sq = squared();
|
|
if (sq <= TUtil<f32>::epsilon()) {
|
|
return 0.0f;
|
|
}
|
|
f32 inv_norm = TUtil<f32>::inv_sqrt(sq);
|
|
scale(inv_norm);
|
|
return inv_norm * sq;
|
|
}
|
|
|
|
void normalize(const TVec3<f32>& other) {
|
|
f32 sq = other.squared();
|
|
if (sq <= TUtil<f32>::epsilon()) {
|
|
zero();
|
|
return;
|
|
}
|
|
f32 norm;
|
|
if (sq <= 0.0f) {
|
|
norm = sq;
|
|
} else {
|
|
norm = fsqrt_step(sq);
|
|
}
|
|
scale(norm, other);
|
|
}
|
|
|
|
f32 length() const {
|
|
return VECMag((Vec*)this);
|
|
}
|
|
|
|
void scale(register f32 sc) {
|
|
register f32 z;
|
|
register f32 x_y;
|
|
register f32* dst = &x;
|
|
register f32 zres;
|
|
#ifdef __MWERKS__
|
|
asm {
|
|
psq_l x_y, 0(dst), 0, 0
|
|
psq_l z, 8(dst), 1, 0
|
|
ps_muls0 x_y, x_y, sc
|
|
psq_st x_y, 0(dst), 0, 0
|
|
ps_muls0 zres, z, sc
|
|
psq_st zres, 8(dst), 1, 0
|
|
};
|
|
#endif
|
|
}
|
|
|
|
void scale(register f32 sc, const TVec3<f32>& other) {
|
|
register const f32* src = &other.x;
|
|
register f32 z;
|
|
register f32 x_y;
|
|
register f32* dst = &x;
|
|
register f32 zres;
|
|
#ifdef __MWERKS__
|
|
asm {
|
|
psq_l x_y, 0(src), 0, 0
|
|
psq_l z, 8(src), 1, 0
|
|
ps_muls0 x_y, x_y, sc
|
|
psq_st x_y, 0(dst), 0, 0
|
|
ps_muls0 zres, z, sc
|
|
psq_st zres, 8(dst), 1, 0
|
|
};
|
|
#endif
|
|
}
|
|
|
|
void scaleAdd(register f32 sc, const TVec3<f32>& a, const TVec3<f32>& b) {
|
|
JMAVECScaleAdd(&a, &b, this, sc);
|
|
}
|
|
|
|
void negateInternal(TVec3<f32>* dst) {
|
|
register f32* rdst = &dst->x;
|
|
const register f32* src = &x;
|
|
register f32 x_y;
|
|
register f32 z;
|
|
#ifdef __MWERKS__
|
|
asm {
|
|
psq_l x_y, 0(src), 0, 0
|
|
ps_neg x_y, x_y
|
|
psq_st x_y, 0(rdst), 0, 0
|
|
lfs z, 8(src)
|
|
fneg z, z
|
|
stfs z, 8(rdst)
|
|
};
|
|
#endif
|
|
}
|
|
|
|
void negate() {
|
|
negateInternal(this);
|
|
}
|
|
|
|
void sub(const TVec3<f32>& b) {
|
|
JMathInlineVEC::C_VECSubtract((Vec*)&x, (Vec*)&b.x, (Vec*)&x);
|
|
}
|
|
|
|
void sub(const TVec3<f32>& a, const TVec3<f32>& b) {
|
|
JMathInlineVEC::C_VECSubtract((Vec*)&a.x, (Vec*)&b.x, (Vec*)&x);
|
|
}
|
|
|
|
bool isZero() const {
|
|
return squared() <= TUtil<f32>::epsilon();
|
|
}
|
|
|
|
void cross(const TVec3<f32>& a, const TVec3<f32>& b) {
|
|
VECCrossProduct(a, b, *this);
|
|
}
|
|
|
|
f32 setLength(f32 len) {
|
|
f32 sq = squared();
|
|
if (sq <= TUtil<f32>::epsilon()) {
|
|
return 0.0f;
|
|
}
|
|
f32 inv_norm = TUtil<f32>::inv_sqrt(sq);
|
|
scale(inv_norm * len);
|
|
return inv_norm * sq;
|
|
}
|
|
|
|
f32 setLength(const TVec3<f32>& other, f32 len) {
|
|
f32 sq = other.squared();
|
|
if (sq <= TUtil<f32>::epsilon()) {
|
|
zero();
|
|
return 0.0f;
|
|
}
|
|
f32 inv_norm = TUtil<f32>::inv_sqrt(sq);
|
|
scale(inv_norm * len, other);
|
|
return inv_norm * sq;
|
|
}
|
|
|
|
f32 dot(const TVec3<f32>& other) const {
|
|
return JMathInlineVEC::C_VECDotProduct(this, &other);
|
|
}
|
|
|
|
void cubic(const TVec3<f32>& param_1, const TVec3<f32>& param_2, const TVec3<f32>& param_3,
|
|
const TVec3<f32>& param_4, f32 param_5) {
|
|
f32 fVar5 = param_5 * param_5;
|
|
f32 fVar6 = fVar5 * param_5;
|
|
f32 fVar8 = 1.0f + (2.0f * fVar6 - 3.0f * fVar5);
|
|
f32 fVar9 = -2.0f * fVar6 + 3.0f * fVar5;
|
|
f32 fVar7 = param_5 + (fVar6 - 2.0f * fVar5);
|
|
f32 fVar4 = fVar6 - fVar5;
|
|
x = fVar8 * param_1.x + fVar9 * param_4.x + fVar7 * param_2.x + fVar4 * param_3.x;
|
|
y = fVar8 * param_1.y + fVar9 * param_4.y + fVar7 * param_2.y + fVar4 * param_3.y;
|
|
z = fVar8 * param_1.z + fVar9 * param_4.z + fVar7 * param_2.z + fVar4 * param_3.z;
|
|
}
|
|
};
|
|
|
|
template <typename T>
|
|
struct TVec2 {
|
|
TVec2() {}
|
|
TVec2(T v) { set(v); }
|
|
TVec2(T x, T y) { set(x, y); }
|
|
|
|
void set(T v) { y = x = v; }
|
|
|
|
void set(T x, T y) {
|
|
this->x = x;
|
|
this->y = y;
|
|
}
|
|
|
|
void set(const TVec2& other) {
|
|
x = other.x;
|
|
y = other.y;
|
|
}
|
|
|
|
void setMin(const TVec2<f32>& min) {
|
|
if (x >= min.x)
|
|
x = min.x;
|
|
if (y >= min.y)
|
|
y = min.y;
|
|
}
|
|
|
|
void setMax(const TVec2<f32>& max) {
|
|
if (x <= max.x)
|
|
x = max.x;
|
|
if (y <= max.y)
|
|
y = max.y;
|
|
}
|
|
|
|
void add(const TVec2<T>& other) {
|
|
x += other.x;
|
|
y += other.y;
|
|
}
|
|
|
|
bool isAbove(const TVec2<T>& other) const {
|
|
return (x >= other.x) && (y >= other.y) ? true : false;
|
|
}
|
|
|
|
f32 dot(const TVec2<T>& other) const {
|
|
return x * other.x + y * other.y;
|
|
}
|
|
|
|
f32 squared() const {
|
|
return dot(*this);
|
|
}
|
|
|
|
f32 length() const {
|
|
f32 sqr = squared();
|
|
if (sqr <= 0.0f) {
|
|
return sqr;
|
|
}
|
|
sqr *= fsqrt_step(sqr);
|
|
return sqr;
|
|
}
|
|
|
|
T x;
|
|
T y;
|
|
};
|
|
|
|
template <class T>
|
|
struct TBox {
|
|
TBox() : i(), f() {}
|
|
TBox(const TBox& other) : i(other.i), f(other.f) {}
|
|
|
|
T i, f;
|
|
};
|
|
|
|
// clang-format off
|
|
template<> struct TBox<TVec2<f32> > {
|
|
f32 getWidth() const { return f.x - i.x; }
|
|
f32 getHeight() const { return f.y - i.y; }
|
|
|
|
bool isValid() const { return f.isAbove(i); }
|
|
|
|
void addPos(f32 x, f32 y) {
|
|
addPos(TVec2<f32>(x, y));
|
|
}
|
|
|
|
void addPos(const TVec2<f32>& pos) {
|
|
i.add(pos);
|
|
f.add(pos);
|
|
}
|
|
|
|
bool intersect(const TBox<TVec2<f32> >& other) {
|
|
i.setMax(other.i);
|
|
f.setMin(other.f);
|
|
return isValid();
|
|
}
|
|
|
|
TVec2<f32> i, f;
|
|
};
|
|
|
|
template <typename T>
|
|
struct TBox2 : TBox<TVec2<T> > {
|
|
TBox2() {}
|
|
TBox2(const TVec2<f32>& i, const TVec2<f32>& f) { set(i, f); }
|
|
TBox2(f32 x0, f32 y0, f32 x1, f32 y1) { set(x0, y0, x1, y1); }
|
|
|
|
void absolute() {
|
|
if (!this->isValid()) {
|
|
TBox2<T> box(*this);
|
|
this->i.setMin(box.i);
|
|
this->i.setMin(box.f);
|
|
this->f.setMax(box.i);
|
|
this->f.setMax(box.f);
|
|
}
|
|
}
|
|
|
|
void set(const TBox2& other) { set(other.i, other.f); }
|
|
void set(const TVec2<f32>& i, const TVec2<f32>& f) { this->i.set(i), this->f.set(f); }
|
|
void set(f32 x0, f32 y0, f32 x1, f32 y1) { this->i.set(x0, y0); this->f.set(x1, y1); }
|
|
};
|
|
|
|
// clang-format on
|
|
|
|
} // namespace JGeometry
|
|
|
|
#endif
|