// Copyright (c) 2012-2021 Wojciech Figat. All rights reserved. #pragma once #include "Math.h" #include "Engine/Core/Formatting.h" #include "Engine/Core/Templates.h" struct Quaternion; struct Matrix; struct Vector2; struct Vector4; struct Color; class String; struct Int3; struct Int4; /// /// Represents a three dimensional mathematical vector. /// API_STRUCT() struct FLAXENGINE_API Vector3 { DECLARE_SCRIPTING_TYPE_MINIMAL(Vector3); public: union { struct { /// /// The X component. /// API_FIELD() float X; /// /// The Y component. /// API_FIELD() float Y; /// /// The Z component. /// API_FIELD() float Z; }; // Raw values float Raw[3]; }; public: // Vector with all components equal zero (0, 0, 0) static const Vector3 Zero; // Vector with all components equal one (1, 1, 1) static const Vector3 One; // Vector with all components equal half (0.5, 0.5, 0.5) static const Vector3 Half; // The X unit vector (1, 0, 0) static const Vector3 UnitX; // The Y unit vector (0, 1, 0) static const Vector3 UnitY; // The Z unit vector (0, 0, 1) static const Vector3 UnitZ; // A unit vector designating up (0, 1, 0) static const Vector3 Up; // A unit vector designating down (0, -1, 0) static const Vector3 Down; // A unit vector designating a (-1, 0, 0) static const Vector3 Left; // A unit vector designating b (1, 0, 0) static const Vector3 Right; // A unit vector designating forward in a a-handed coordinate system (0, 0, 1) static const Vector3 Forward; // A unit vector designating backward in a a-handed coordinate system (0, 0, -1) static const Vector3 Backward; // A minimum Vector3 static const Vector3 Minimum; // A maximum Vector3 static const Vector3 Maximum; public: /// /// Empty constructor. /// Vector3() { } // Init // @param xyz Value to assign to the all components Vector3(float xyz) : X(xyz) , Y(xyz) , Z(xyz) { } // Init // @param x X component value // @param y Y component value // @param z Z component value Vector3(float x, float y, float z) : X(x) , Y(y) , Z(z) { } /// /// Init /// /// X, Y and Z components in an array explicit Vector3(const float* xyz) : X(xyz[0]) , Y(xyz[1]) , Z(xyz[2]) { } // Init // @param xy Vector2 with X and Y components values // @param z Z component value Vector3(const Vector2& xy, float z); // Init // @param xy Vector3 value explicit Vector3(const Vector2& xy); // Init // @param xy Int22 with X and Y components values // @param z Z component value explicit Vector3(const Int2& xy, float z); // Init // @param xyz Int3 value explicit Vector3(const Int3& xyz); // Init // @param xyzw Int4 value explicit Vector3(const Int4& xyzw); // Init // @param xyz Vector4 value explicit Vector3(const Vector4& xyz); // Init // @param color Color value explicit Vector3(const Color& color); public: String ToString() const; public: // Gets a value indicting whether this instance is normalized bool IsNormalized() const { return Math::IsOne(X * X + Y * Y + Z * Z); } // Gets a value indicting whether this vector is zero bool IsZero() const { return Math::IsZero(X) && Math::IsZero(Y) && Math::IsZero(Z); } // Gets a value indicting whether any vector component is zero bool IsAnyZero() const { return Math::IsZero(X) || Math::IsZero(Y) || Math::IsZero(Z); } // Gets a value indicting whether this vector is one bool IsOne() const { return Math::IsOne(X) && Math::IsOne(Y) && Math::IsOne(Z); } // Calculates length of the vector // @returns Length of the vector float Length() const { return Math::Sqrt(X * X + Y * Y + Z * Z); } // Calculates the squared length of the vector // @returns The squared length of the vector float LengthSquared() const { return X * X + Y * Y + Z * Z; } // Calculates inverted length of the vector (1 / Length()) float InvLength() const { return 1.0f / Length(); } /// /// Calculates a vector with values being absolute values of that vector /// /// Absolute vector Vector3 GetAbsolute() const { return Vector3(Math::Abs(X), Math::Abs(Y), Math::Abs(Z)); } /// /// Calculates a vector with values being opposite to values of that vector /// /// Negative vector Vector3 GetNegative() const { return Vector3(-X, -Y, -Z); } /// /// Calculates a normalized vector that has length equal to 1. /// /// The normalized vector. Vector3 GetNormalized() const { const float rcp = 1.0f / Length(); return Vector3(X * rcp, Y * rcp, Z * rcp); } /// /// Returns average arithmetic of all the components /// /// Average arithmetic of all the components float AverageArithmetic() const { return (X + Y + Z) * 0.333333334f; } /// /// Gets sum of all vector components values /// /// Sum of X,Y and Z float SumValues() const { return X + Y + Z; } /// /// Returns minimum value of all the components /// /// Minimum value float MinValue() const { return Math::Min(X, Y, Z); } /// /// Returns maximum value of all the components /// /// Maximum value float MaxValue() const { return Math::Max(X, Y, Z); } /// /// Returns true if vector has one or more components is not a number (NaN) /// /// True if one or more components is not a number (NaN) bool IsNaN() const { return isnan(X) || isnan(Y) || isnan(Z); } /// /// Returns true if vector has one or more components equal to +/- infinity /// /// True if one or more components equal to +/- infinity bool IsInfinity() const { return isinf(X) || isinf(Y) || isinf(Z); } /// /// Returns true if vector has one or more components equal to +/- infinity or NaN /// /// True if one or more components equal to +/- infinity or NaN bool IsNanOrInfinity() const { return IsInfinity() || IsNaN(); } public: /// /// Performs vector normalization (scales vector up to unit length) /// void Normalize() { const float length = Length(); if (!Math::IsZero(length)) { const float inv = 1.0f / length; X *= inv; Y *= inv; Z *= inv; } } /// /// Performs fast vector normalization (scales vector up to unit length) /// void NormalizeFast() { const float inv = 1.0f / Length(); X *= inv; Y *= inv; Z *= inv; } /// /// Sets all vector components to the absolute values /// void Absolute() { X = Math::Abs(X); Y = Math::Abs(Y); Z = Math::Abs(Z); } /// /// Negates all components of that vector /// void Negate() { X = -X; Y = -Y; Z = -Z; } /// /// When this vector contains Euler angles (degrees), ensure that angles are between +/-180 /// void UnwindEuler(); public: // Arithmetic operators with Vector3 Vector3 operator+(const Vector3& b) const { return Add(*this, b); } Vector3 operator-(const Vector3& b) const { return Subtract(*this, b); } Vector3 operator*(const Vector3& b) const { return Multiply(*this, b); } Vector3 operator/(const Vector3& b) const { return Divide(*this, b); } Vector3 operator-() const { return Vector3(-X, -Y, -Z); } Vector3 operator^(const Vector3& b) const { return Cross(*this, b); } float operator|(const Vector3& b) const { return Dot(*this, b); } // op= operators with Vector3 Vector3& operator+=(const Vector3& b) { *this = Add(*this, b); return *this; } Vector3& operator-=(const Vector3& b) { *this = Subtract(*this, b); return *this; } Vector3& operator*=(const Vector3& b) { *this = Multiply(*this, b); return *this; } Vector3& operator/=(const Vector3& b) { *this = Divide(*this, b); return *this; } // Arithmetic operators with float Vector3 operator+(float b) const { return Add(*this, b); } Vector3 operator-(float b) const { return Subtract(*this, b); } Vector3 operator*(float b) const { return Multiply(*this, b); } Vector3 operator/(float b) const { return Divide(*this, b); } // op= operators with float Vector3& operator+=(float b) { *this = Add(*this, b); return *this; } Vector3& operator-=(float b) { *this = Subtract(*this, b); return *this; } Vector3& operator*=(float b) { *this = Multiply(*this, b); return *this; } Vector3& operator/=(float b) { *this = Divide(*this, b); return *this; } // Comparison operators bool operator==(const Vector3& b) const { return X == b.X && Y == b.Y && Z == b.Z; } bool operator!=(const Vector3& b) const { return X != b.X || Y != b.Y || Z != b.Z; } bool operator>(const Vector3& b) const { return X > b.X && Y > b.Y && Z > b.Z; } bool operator>=(const Vector3& b) const { return X >= b.X && Y >= b.Y && Z >= b.Z; } bool operator<(const Vector3& b) const { return X < b.X && Y < b.Y && Z < b.Z; } bool operator<=(const Vector3& b) const { return X <= b.X && Y <= b.Y && Z <= b.Z; } public: static bool NearEqual(const Vector3& a, const Vector3& b) { return Math::NearEqual(a.X, b.X) && Math::NearEqual(a.Y, b.Y) && Math::NearEqual(a.Z, b.Z); } static bool NearEqual(const Vector3& a, const Vector3& b, float epsilon) { return Math::NearEqual(a.X, b.X, epsilon) && Math::NearEqual(a.Y, b.Y, epsilon) && Math::NearEqual(a.Z, b.Z, epsilon); } public: static void Add(const Vector3& a, const Vector3& b, Vector3& result) { result.X = a.X + b.X; result.Y = a.Y + b.Y; result.Z = a.Z + b.Z; } static Vector3 Add(const Vector3& a, const Vector3& b) { Vector3 result; Add(a, b, result); return result; } static void Subtract(const Vector3& a, const Vector3& b, Vector3& result) { result.X = a.X - b.X; result.Y = a.Y - b.Y; result.Z = a.Z - b.Z; } static Vector3 Subtract(const Vector3& a, const Vector3& b) { Vector3 result; Subtract(a, b, result); return result; } static Vector3 Multiply(const Vector3& a, const Vector3& b) { return Vector3(a.X * b.X, a.Y * b.Y, a.Z * b.Z); } static void Multiply(const Vector3& a, const Vector3& b, Vector3& result) { result = Vector3(a.X * b.X, a.Y * b.Y, a.Z * b.Z); } static Vector3 Multiply(const Vector3& a, float b) { return Vector3(a.X * b, a.Y * b, a.Z * b); } static Vector3 Divide(const Vector3& a, const Vector3& b) { return Vector3(a.X / b.X, a.Y / b.Y, a.Z / b.Z); } static void Divide(const Vector3& a, const Vector3& b, Vector3& result) { result = Vector3(a.X / b.X, a.Y / b.Y, a.Z / b.Z); } static Vector3 Divide(const Vector3& a, float b) { return Vector3(a.X / b, a.Y / b, a.Z / b); } static Vector3 Floor(const Vector3& v); static Vector3 Frac(const Vector3& v); static float ScalarProduct(const Vector3& a, const Vector3& b) { return a.X * b.X + a.Y * b.Y + a.Z * b.Z; } public: // Restricts a value to be within a specified range // @param value The value to clamp // @param min The minimum value, // @param max The maximum value // @returns Clamped value static Vector3 Clamp(const Vector3& value, const Vector3& min, const Vector3& max); // Restricts a value to be within a specified range // @param value The value to clamp // @param min The minimum value, // @param max The maximum value // @param result When the method completes, contains the clamped value static void Clamp(const Vector3& value, const Vector3& min, const Vector3& max, Vector3& result); // Calculates the distance between two vectors // @param value1 The first vector // @param value2 The second vector // @returns The distance between the two vectors static float Distance(const Vector3& value1, const Vector3& value2); // Calculates the squared distance between two vectors // @param value1 The first vector // @param value2 The second vector // @returns The squared distance between the two vectors static float DistanceSquared(const Vector3& value1, const Vector3& value2); // Performs vector normalization (scales vector up to unit length) // @param inout Input vector to normalize // @returns Output vector that is normalized (has unit length) static Vector3 Normalize(const Vector3& input); // Performs vector normalization (scales vector up to unit length). This is a faster version that does not performs check for length equal 0 (it assumes that input vector is not empty). // @param inout Input vector to normalize (cannot be zero). // @returns Output vector that is normalized (has unit length) static Vector3 NormalizeFast(const Vector3& input) { const float inv = 1.0f / input.Length(); return Vector3(input.X * inv, input.Y * inv, input.Z * inv); } // Performs vector normalization (scales vector up to unit length) // @param inout Input vector to normalize // @param output Output vector that is normalized (has unit length) static void Normalize(const Vector3& input, Vector3& result); // dot product with another vector static float Dot(const Vector3& a, const Vector3& b) { return a.X * b.X + a.Y * b.Y + a.Z * b.Z; } // Calculates the cross product of two vectors // @param a First source vector // @param b Second source vector // @param result When the method completes, contains the cross product of the two vectors static void Cross(const Vector3& a, const Vector3& b, Vector3& result) { result = Vector3( a.Y * b.Z - a.Z * b.Y, a.Z * b.X - a.X * b.Z, a.X * b.Y - a.Y * b.X); } // Calculates the cross product of two vectors // @param a First source vector // @param b Second source vector // @returns Cross product of the two vectors static Vector3 Cross(const Vector3& a, const Vector3& b) { return Vector3( a.Y * b.Z - a.Z * b.Y, a.Z * b.X - a.X * b.Z, a.X * b.Y - a.Y * b.X); } // Performs a linear interpolation between two vectors // @param start Start vector // @param end End vector // @param amount Value between 0 and 1 indicating the weight of end // @param result When the method completes, contains the linear interpolation of the two vectors static void Lerp(const Vector3& start, const Vector3& end, float amount, Vector3& result) { result.X = Math::Lerp(start.X, end.X, amount); result.Y = Math::Lerp(start.Y, end.Y, amount); result.Z = Math::Lerp(start.Z, end.Z, amount); } // // Performs a linear interpolation between two vectors. // // @param start Start vector, // @param end End vector, // @param amount Value between 0 and 1 indicating the weight of @paramref end"/>, // @returns The linear interpolation of the two vectors static Vector3 Lerp(const Vector3& start, const Vector3& end, float amount) { Vector3 result; Lerp(start, end, amount, result); return result; } // Performs a cubic interpolation between two vectors // @param start Start vector // @param end End vector // @param amount Value between 0 and 1 indicating the weight of end // @param result When the method completes, contains the cubic interpolation of the two vectors static void SmoothStep(const Vector3& start, const Vector3& end, float amount, Vector3& result) { amount = Math::SmoothStep(amount); Lerp(start, end, amount, result); } // Performs a Hermite spline interpolation. // @param value1 First source position vector // @param tangent1 First source tangent vector // @param value2 Second source position vector // @param tangent2 Second source tangent vector // @param amount Weighting factor, // @param result When the method completes, contains the result of the Hermite spline interpolation, static void Hermite(const Vector3& value1, const Vector3& tangent1, const Vector3& value2, const Vector3& tangent2, float amount, Vector3& result); // Returns the reflection of a vector off a surface that has the specified normal // @param vector The source vector // @param normal Normal of the surface // @param result When the method completes, contains the reflected vector static void Reflect(const Vector3& vector, const Vector3& normal, Vector3& result); // Transforms a 3D vector by the given Quaternion rotation // @param vector The vector to rotate // @param rotation The Quaternion rotation to apply // @param result When the method completes, contains the transformed Vector4 static void Transform(const Vector3& vector, const Quaternion& rotation, Vector3& result); // Transforms a 3D vector by the given Quaternion rotation // @param vector The vector to rotate // @param rotation The Quaternion rotation to apply // @returns The transformed Vector4 static Vector3 Transform(const Vector3& vector, const Quaternion& rotation); // Transforms a 3D vector by the given matrix // @param vector The source vector // @param transform The transformation matrix // @param result When the method completes, contains the transformed Vector3 static void Transform(const Vector3& vector, const Matrix& transform, Vector3& result); // Transforms a 3D vectors by the given matrix // @param vectors The source vectors // @param transform The transformation matrix // @param results When the method completes, contains the transformed Vector3s // @param vectorsCount Amount of vectors to transform static void Transform(const Vector3* vectors, const Matrix& transform, Vector3* results, int32 vectorsCount); // Transforms a 3D vector by the given matrix // @param vector The source vector // @param transform The transformation matrix // @returns Transformed Vector3 static Vector3 Transform(const Vector3& vector, const Matrix& transform); // Transforms a 3D vector by the given matrix // @param vector The source vector // @param transform The transformation matrix // @param result When the method completes, contains the transformed Vector4 static void Transform(const Vector3& vector, const Matrix& transform, Vector4& result); // Performs a coordinate transformation using the given matrix // @param coordinate The coordinate vector to transform // @param transform The transformation matrix // @param result When the method completes, contains the transformed coordinates static void TransformCoordinate(const Vector3& coordinate, const Matrix& transform, Vector3& result); // Performs a normal transformation using the given matrix // @param normal The normal vector to transform // @param transform The transformation matrix // @param result When the method completes, contains the transformed normal static void TransformNormal(const Vector3& normal, const Matrix& transform, Vector3& result); // Returns a vector containing the largest components of the specified vectors // @param a The first source vector // @param b The second source vector // @param result When the method completes, contains an new vector composed of the largest components of the source vectors static Vector3 Max(const Vector3& a, const Vector3& b) { return Vector3(a.X > b.X ? a.X : b.X, a.Y > b.Y ? a.Y : b.Y, a.Z > b.Z ? a.Z : b.Z); } // Returns a vector containing the smallest components of the specified vectors // @param a The first source vector // @param b The second source vector // @param result When the method completes, contains an new vector composed of the smallest components of the source vectors static Vector3 Min(const Vector3& a, const Vector3& b) { return Vector3(a.X < b.X ? a.X : b.X, a.Y < b.Y ? a.Y : b.Y, a.Z < b.Z ? a.Z : b.Z); } // Returns a vector containing the largest components of the specified vectors // @param a The first source vector // @param b The second source vector // @param result When the method completes, contains an new vector composed of the largest components of the source vectors static void Max(const Vector3& a, const Vector3& b, Vector3& result) { result = Vector3(a.X > b.X ? a.X : b.X, a.Y > b.Y ? a.Y : b.Y, a.Z > b.Z ? a.Z : b.Z); } // Returns a vector containing the smallest components of the specified vectors // @param a The first source vector // @param b The second source vector // @param result When the method completes, contains an new vector composed of the smallest components of the source vectors static void Min(const Vector3& a, const Vector3& b, Vector3& result) { result = Vector3(a.X < b.X ? a.X : b.X, a.Y < b.Y ? a.Y : b.Y, a.Z < b.Z ? a.Z : b.Z); } /// /// Projects a vector onto another vector. /// /// The vector to project. /// The projection normal vector. /// The projected vector. static Vector3 Project(const Vector3& vector, const Vector3& onNormal); /// /// Projects a vector onto a plane defined by a normal orthogonal to the plane. /// /// The vector to project. /// The plane normal vector. /// The projected vector. static Vector3 ProjectOnPlane(const Vector3& vector, const Vector3& planeNormal) { return vector - Project(vector, planeNormal); } // Projects a 3D vector from object space into screen space // @param vector The vector to project // @param x The X position of the viewport // @param y The Y position of the viewport // @param width The width of the viewport // @param height The height of the viewport // @param minZ The minimum depth of the viewport // @param maxZ The maximum depth of the viewport // @param worldViewProjection The combined world-view-projection matrix // @param result When the method completes, contains the vector in screen space static void Project(const Vector3& vector, float x, float y, float width, float height, float minZ, float maxZ, const Matrix& worldViewProjection, Vector3& result); // Projects a 3D vector from object space into screen space // @param vector The vector to project // @param x The X position of the viewport // @param y The Y position of the viewport // @param width The width of the viewport // @param height The height of the viewport // @param minZ The minimum depth of the viewport // @param maxZ The maximum depth of the viewport // @param worldViewProjection The combined world-view-projection matrix // @returns The vector in screen space static Vector3 Project(const Vector3& vector, float x, float y, float width, float height, float minZ, float maxZ, const Matrix& worldViewProjection) { Vector3 result; Project(vector, x, y, width, height, minZ, maxZ, worldViewProjection, result); return result; } // Projects a 3D vector from screen space into object space // @param vector The vector to project // @param x The X position of the viewport // @param y The Y position of the viewport // @param width The width of the viewport // @param height The height of the viewport // @param minZ The minimum depth of the viewport // @param maxZ The maximum depth of the viewport // @param worldViewProjection The combined world-view-projection matrix // @param result When the method completes, contains the vector in object space static void Unproject(const Vector3& vector, float x, float y, float width, float height, float minZ, float maxZ, const Matrix& worldViewProjection, Vector3& result); // Projects a 3D vector from screen space into object space // @param vector The vector to project // @param x The X position of the viewport // @param y The Y position of the viewport // @param width The width of the viewport // @param height The height of the viewport // @param minZ The minimum depth of the viewport // @param maxZ The maximum depth of the viewport // @param worldViewProjection The combined world-view-projection matrix // @returns The vector in object space static Vector3 Unproject(const Vector3& vector, float x, float y, float width, float height, float minZ, float maxZ, const Matrix& worldViewProjection) { Vector3 result; Unproject(vector, x, y, width, height, minZ, maxZ, worldViewProjection, result); return result; } /// /// Creates an orthonormal basis from a basis with at least two orthogonal vectors. /// /// The X axis. /// The y axis. /// The z axis. static void CreateOrthonormalBasis(Vector3& xAxis, Vector3& yAxis, Vector3& zAxis); /// /// Finds the best arbitrary axis vectors to represent U and V axes of a plane, by using this vector as the normal of the plane. /// /// The reference to first axis. /// The reference to second axis. void FindBestAxisVectors(Vector3& firstAxis, Vector3& secondAxis) const; static Vector3 Round(const Vector3& v) { return Vector3( Math::Round(v.X), Math::Round(v.Y), Math::Round(v.Z) ); } static Vector3 Ceil(const Vector3& v) { return Vector3( Math::Ceil(v.X), Math::Ceil(v.Y), Math::Ceil(v.Z) ); } static Vector3 Abs(const Vector3& v) { return Vector3(Math::Abs(v.X), Math::Abs(v.Y), Math::Abs(v.Z)); } /// /// Calculates the area of the triangle. /// /// The first triangle vertex. /// The second triangle vertex. /// The third triangle vertex. /// The triangle area. static float TriangleArea(const Vector3& v0, const Vector3& v1, const Vector3& v2); }; inline Vector3 operator+(float a, const Vector3& b) { return b + a; } inline Vector3 operator-(float a, const Vector3& b) { return Vector3(a) - b; } inline Vector3 operator*(float a, const Vector3& b) { return b * a; } inline Vector3 operator/(float a, const Vector3& b) { return Vector3(a) / b; } namespace Math { FORCE_INLINE static bool NearEqual(const Vector3& a, const Vector3& b) { return Vector3::NearEqual(a, b); } } template<> struct TIsPODType { enum { Value = true }; }; DEFINE_DEFAULT_FORMATTING(Vector3, "X:{0} Y:{1} Z:{2}", v.X, v.Y, v.Z);