269 lines
9.5 KiB
C++
269 lines
9.5 KiB
C++
/**
|
|
* PANDA 3D SOFTWARE
|
|
* Copyright (c) Carnegie Mellon University. All rights reserved.
|
|
*
|
|
* All use of this software is subject to the terms of the revised BSD
|
|
* license. You should have received a copy of this license along
|
|
* with this source code in a file named "LICENSE."
|
|
*
|
|
* @file lvecBase4_src.h
|
|
* @author drose
|
|
* @date 2000-03-08
|
|
*/
|
|
|
|
class FLOATNAME(LVecBase2);
|
|
class FLOATNAME(LVecBase3);
|
|
class FLOATNAME(LPoint3);
|
|
class FLOATNAME(LVector3);
|
|
class FLOATNAME(UnalignedLVecBase4);
|
|
|
|
/**
|
|
* This is the base class for all three-component vectors and points.
|
|
*/
|
|
class EXPCL_PANDA_LINMATH ALIGN_LINMATH FLOATNAME(LVecBase4) {
|
|
PUBLISHED:
|
|
typedef FLOATTYPE numeric_type;
|
|
typedef const FLOATTYPE *iterator;
|
|
typedef const FLOATTYPE *const_iterator;
|
|
|
|
enum {
|
|
num_components = 4,
|
|
|
|
#ifdef FLOATTYPE_IS_INT
|
|
is_int = 1
|
|
#else
|
|
is_int = 0
|
|
#endif
|
|
};
|
|
|
|
INLINE_LINMATH FLOATNAME(LVecBase4)() = default;
|
|
INLINE_LINMATH FLOATNAME(LVecBase4)(FLOATTYPE fill_value);
|
|
INLINE_LINMATH FLOATNAME(LVecBase4)(FLOATTYPE x, FLOATTYPE y, FLOATTYPE z, FLOATTYPE w);
|
|
INLINE_LINMATH FLOATNAME(LVecBase4)(const FLOATNAME(UnalignedLVecBase4) ©);
|
|
INLINE_LINMATH FLOATNAME(LVecBase4)(const FLOATNAME(LVecBase3) ©, FLOATTYPE w);
|
|
INLINE_LINMATH FLOATNAME(LVecBase4)(const FLOATNAME(LPoint3) &point);
|
|
INLINE_LINMATH FLOATNAME(LVecBase4)(const FLOATNAME(LVector3) &vector);
|
|
ALLOC_DELETED_CHAIN(FLOATNAME(LVecBase4));
|
|
|
|
#ifdef CPPPARSER
|
|
FLOATNAME(LVecBase4) &operator = (const FLOATNAME(LVecBase4) ©) = default;
|
|
FLOATNAME(LVecBase4) &operator = (FLOATTYPE fill_value) = default;
|
|
#endif
|
|
|
|
INLINE_LINMATH static const FLOATNAME(LVecBase4) &zero();
|
|
INLINE_LINMATH static const FLOATNAME(LVecBase4) &unit_x();
|
|
INLINE_LINMATH static const FLOATNAME(LVecBase4) &unit_y();
|
|
INLINE_LINMATH static const FLOATNAME(LVecBase4) &unit_z();
|
|
INLINE_LINMATH static const FLOATNAME(LVecBase4) &unit_w();
|
|
|
|
EXTENSION(INLINE_LINMATH PyObject *__reduce__(PyObject *self) const);
|
|
EXTENSION(INLINE_LINMATH PyObject *__getattr__(PyObject *self, const std::string &attr_name) const);
|
|
EXTENSION(INLINE_LINMATH int __setattr__(PyObject *self, const std::string &attr_name, PyObject *assign));
|
|
|
|
INLINE_LINMATH FLOATTYPE operator [](int i) const;
|
|
INLINE_LINMATH FLOATTYPE &operator [](int i);
|
|
constexpr static int size() { return 4; }
|
|
|
|
INLINE_LINMATH bool is_nan() const;
|
|
|
|
INLINE_LINMATH FLOATTYPE get_cell(int i) const;
|
|
INLINE_LINMATH void set_cell(int i, FLOATTYPE value);
|
|
|
|
INLINE_LINMATH FLOATTYPE get_x() const;
|
|
INLINE_LINMATH FLOATTYPE get_y() const;
|
|
INLINE_LINMATH FLOATTYPE get_z() const;
|
|
INLINE_LINMATH FLOATTYPE get_w() const;
|
|
|
|
INLINE_LINMATH FLOATNAME(LVecBase3) get_xyz() const;
|
|
INLINE_LINMATH FLOATNAME(LVecBase2) get_xy() const;
|
|
|
|
INLINE_LINMATH void set_x(FLOATTYPE value);
|
|
INLINE_LINMATH void set_y(FLOATTYPE value);
|
|
INLINE_LINMATH void set_z(FLOATTYPE value);
|
|
INLINE_LINMATH void set_w(FLOATTYPE value);
|
|
|
|
PUBLISHED:
|
|
MAKE_PROPERTY(x, get_x, set_x);
|
|
MAKE_PROPERTY(y, get_y, set_y);
|
|
MAKE_PROPERTY(z, get_z, set_z);
|
|
|
|
MAKE_PROPERTY(xyz, get_xyz);
|
|
MAKE_PROPERTY(xy, get_xy);
|
|
|
|
// These next functions add to an existing value. i.e.
|
|
// foo.set_x(foo.get_x() + value) These are useful to reduce overhead in
|
|
// scripting languages:
|
|
INLINE_LINMATH void add_to_cell(int i, FLOATTYPE value);
|
|
INLINE_LINMATH void add_x(FLOATTYPE value);
|
|
INLINE_LINMATH void add_y(FLOATTYPE value);
|
|
INLINE_LINMATH void add_z(FLOATTYPE value);
|
|
INLINE_LINMATH void add_w(FLOATTYPE value);
|
|
|
|
INLINE_LINMATH const FLOATTYPE *get_data() const;
|
|
constexpr static int get_num_components() { return 4; }
|
|
INLINE_LINMATH void extract_data(float*){};
|
|
|
|
public:
|
|
INLINE_LINMATH iterator begin();
|
|
INLINE_LINMATH iterator end();
|
|
|
|
INLINE_LINMATH const_iterator begin() const;
|
|
INLINE_LINMATH const_iterator end() const;
|
|
|
|
PUBLISHED:
|
|
INLINE_LINMATH void fill(FLOATTYPE fill_value);
|
|
INLINE_LINMATH void set(FLOATTYPE x, FLOATTYPE y, FLOATTYPE z, FLOATTYPE w);
|
|
|
|
INLINE_LINMATH FLOATTYPE dot(const FLOATNAME(LVecBase4) &other) const;
|
|
INLINE_LINMATH FLOATTYPE length_squared() const;
|
|
|
|
#ifndef FLOATTYPE_IS_INT
|
|
INLINE_LINMATH FLOATTYPE length() const;
|
|
INLINE_LINMATH bool normalize();
|
|
INLINE_LINMATH FLOATNAME(LVecBase4) normalized() const;
|
|
INLINE_LINMATH FLOATNAME(LVecBase4) project(const FLOATNAME(LVecBase4) &onto) const;
|
|
#endif
|
|
|
|
INLINE_LINMATH bool operator < (const FLOATNAME(LVecBase4) &other) const;
|
|
INLINE_LINMATH bool operator == (const FLOATNAME(LVecBase4) &other) const;
|
|
INLINE_LINMATH bool operator != (const FLOATNAME(LVecBase4) &other) const;
|
|
|
|
INLINE_LINMATH int compare_to(const FLOATNAME(LVecBase4) &other) const;
|
|
INLINE_LINMATH size_t get_hash() const;
|
|
INLINE_LINMATH size_t add_hash(size_t hash) const;
|
|
INLINE_LINMATH void generate_hash(ChecksumHashGenerator &hashgen) const;
|
|
|
|
#ifndef FLOATTYPE_IS_INT
|
|
INLINE_LINMATH int compare_to(const FLOATNAME(LVecBase4) &other,
|
|
FLOATTYPE threshold) const;
|
|
INLINE_LINMATH size_t get_hash(FLOATTYPE threshold) const;
|
|
INLINE_LINMATH size_t add_hash(size_t hash, FLOATTYPE threshold) const;
|
|
INLINE_LINMATH void generate_hash(ChecksumHashGenerator &hashgen,
|
|
FLOATTYPE threshold) const;
|
|
#endif
|
|
|
|
INLINE_LINMATH FLOATNAME(LVecBase4) operator - () const;
|
|
|
|
INLINE_LINMATH FLOATNAME(LVecBase4)
|
|
operator + (const FLOATNAME(LVecBase4) &other) const;
|
|
INLINE_LINMATH FLOATNAME(LVecBase4)
|
|
operator - (const FLOATNAME(LVecBase4) &other) const;
|
|
|
|
INLINE_LINMATH FLOATNAME(LVecBase4) operator * (FLOATTYPE scalar) const;
|
|
INLINE_LINMATH FLOATNAME(LVecBase4) operator / (FLOATTYPE scalar) const;
|
|
|
|
INLINE_LINMATH void operator += (const FLOATNAME(LVecBase4) &other);
|
|
INLINE_LINMATH void operator -= (const FLOATNAME(LVecBase4) &other);
|
|
|
|
INLINE_LINMATH void operator *= (FLOATTYPE scalar);
|
|
INLINE_LINMATH void operator /= (FLOATTYPE scalar);
|
|
|
|
INLINE_LINMATH void componentwise_mult(const FLOATNAME(LVecBase4) &other);
|
|
|
|
EXTENSION(INLINE_LINMATH FLOATNAME(LVecBase4) __pow__(FLOATTYPE exponent) const);
|
|
EXTENSION(INLINE_LINMATH PyObject *__ipow__(PyObject *self, FLOATTYPE exponent));
|
|
|
|
INLINE_LINMATH FLOATNAME(LVecBase4) fmax(const FLOATNAME(LVecBase4) &other) const;
|
|
INLINE_LINMATH FLOATNAME(LVecBase4) fmin(const FLOATNAME(LVecBase4) &other) const;
|
|
|
|
INLINE_LINMATH bool almost_equal(const FLOATNAME(LVecBase4) &other,
|
|
FLOATTYPE threshold) const;
|
|
INLINE_LINMATH bool almost_equal(const FLOATNAME(LVecBase4) &other) const;
|
|
|
|
INLINE_LINMATH void output(std::ostream &out) const;
|
|
EXTENSION(INLINE_LINMATH std::string __repr__() const);
|
|
|
|
INLINE_LINMATH void write_datagram_fixed(Datagram &destination) const;
|
|
INLINE_LINMATH void read_datagram_fixed(DatagramIterator &source);
|
|
INLINE_LINMATH void write_datagram(Datagram &destination) const;
|
|
INLINE_LINMATH void read_datagram(DatagramIterator &source);
|
|
|
|
public:
|
|
// The underlying implementation is via the Eigen library, if available.
|
|
|
|
// Unlike LVecBase2 and LVecBase3, we fully align LVecBase4 to 16-byte
|
|
// boundaries, to take advantage of SSE2 optimizations when available.
|
|
// Sometimes this alignment requirement is inconvenient, so we also provide
|
|
// UnalignedLVecBase4, below.
|
|
typedef LINMATH_MATRIX(FLOATTYPE, 1, 4) EVector4;
|
|
EVector4 _v;
|
|
|
|
INLINE_LINMATH FLOATNAME(LVecBase4)(const EVector4 &v) : _v(v) { }
|
|
|
|
private:
|
|
static const FLOATNAME(LVecBase4) _zero;
|
|
static const FLOATNAME(LVecBase4) _unit_x;
|
|
static const FLOATNAME(LVecBase4) _unit_y;
|
|
static const FLOATNAME(LVecBase4) _unit_z;
|
|
static const FLOATNAME(LVecBase4) _unit_w;
|
|
|
|
public:
|
|
static TypeHandle get_class_type() {
|
|
return _type_handle;
|
|
}
|
|
static void init_type();
|
|
|
|
private:
|
|
static TypeHandle _type_handle;
|
|
};
|
|
|
|
/**
|
|
* This is an "unaligned" LVecBase4. It has no functionality other than to
|
|
* store numbers, and it will pack them in as tightly as possible, avoiding
|
|
* any SSE2 alignment requirements shared by the primary LVecBase4 class.
|
|
*
|
|
* Use it only when you need to pack numbers tightly without respect to
|
|
* alignment, and then copy it to a proper LVecBase4 to get actual use from
|
|
* it.
|
|
*/
|
|
class EXPCL_PANDA_LINMATH FLOATNAME(UnalignedLVecBase4) {
|
|
PUBLISHED:
|
|
enum {
|
|
num_components = 4,
|
|
|
|
#ifdef FLOATTYPE_IS_INT
|
|
is_int = 1
|
|
#else
|
|
is_int = 0
|
|
#endif
|
|
};
|
|
|
|
INLINE_LINMATH FLOATNAME(UnalignedLVecBase4)() = default;
|
|
INLINE_LINMATH FLOATNAME(UnalignedLVecBase4)(const FLOATNAME(LVecBase4) ©);
|
|
INLINE_LINMATH FLOATNAME(UnalignedLVecBase4)(FLOATTYPE fill_value);
|
|
INLINE_LINMATH FLOATNAME(UnalignedLVecBase4)(FLOATTYPE x, FLOATTYPE y, FLOATTYPE z, FLOATTYPE w);
|
|
|
|
INLINE_LINMATH void fill(FLOATTYPE fill_value);
|
|
INLINE_LINMATH void set(FLOATTYPE x, FLOATTYPE y, FLOATTYPE z, FLOATTYPE w);
|
|
|
|
INLINE_LINMATH FLOATTYPE operator [](int i) const;
|
|
INLINE_LINMATH FLOATTYPE &operator [](int i);
|
|
constexpr static int size() { return 4; }
|
|
|
|
INLINE_LINMATH const FLOATTYPE *get_data() const;
|
|
constexpr static int get_num_components() { return 4; }
|
|
|
|
INLINE_LINMATH bool operator == (const FLOATNAME(UnalignedLVecBase4) &other) const;
|
|
INLINE_LINMATH bool operator != (const FLOATNAME(UnalignedLVecBase4) &other) const;
|
|
|
|
public:
|
|
typedef FLOATTYPE numeric_type;
|
|
typedef UNALIGNED_LINMATH_MATRIX(FLOATTYPE, 1, 4) UVector4;
|
|
UVector4 _v;
|
|
|
|
public:
|
|
static TypeHandle get_class_type() {
|
|
return _type_handle;
|
|
}
|
|
static void init_type();
|
|
|
|
private:
|
|
static TypeHandle _type_handle;
|
|
};
|
|
|
|
INLINE std::ostream &operator << (std::ostream &out, const FLOATNAME(LVecBase4) &vec) {
|
|
vec.output(out);
|
|
return out;
|
|
}
|
|
|
|
#include "lvecBase4_src.I"
|