mirror of
https://github.com/Jozufozu/Flywheel.git
synced 2024-11-14 06:24:12 +01:00
dd18300b70
- Fix Resources not being closed properly - Change versioning scheme to match Create - Add LICENSE to built jar - Fix mods.toml version sync - Move JOML code to non-src directory - Update Gradle - Organize imports
15356 lines
655 KiB
Java
15356 lines
655 KiB
Java
/*
|
|
* The MIT License
|
|
*
|
|
* Copyright (c) 2015-2021 Richard Greenlees
|
|
*
|
|
* Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
* of this software and associated documentation files (the "Software"), to deal
|
|
* in the Software without restriction, including without limitation the rights
|
|
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
* copies of the Software, and to permit persons to whom the Software is
|
|
* furnished to do so, subject to the following conditions:
|
|
*
|
|
* The above copyright notice and this permission notice shall be included in
|
|
* all copies or substantial portions of the Software.
|
|
*
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
|
* THE SOFTWARE.
|
|
*/
|
|
package com.jozufozu.flywheel.repack.joml;
|
|
|
|
import java.io.Externalizable;
|
|
import java.io.IOException;
|
|
import java.io.ObjectInput;
|
|
import java.io.ObjectOutput;
|
|
import java.nio.ByteBuffer;
|
|
import java.nio.FloatBuffer;
|
|
import java.text.DecimalFormat;
|
|
import java.text.NumberFormat;
|
|
|
|
|
|
/**
|
|
* Contains the definition of a 4x4 matrix of floats, and associated functions to transform
|
|
* it. The matrix is column-major to match OpenGL's interpretation, and it looks like this:
|
|
* <p>
|
|
* m00 m10 m20 m30<br>
|
|
* m01 m11 m21 m31<br>
|
|
* m02 m12 m22 m32<br>
|
|
* m03 m13 m23 m33<br>
|
|
*
|
|
* @author Richard Greenlees
|
|
* @author Kai Burjack
|
|
*/
|
|
public class Matrix4f implements Externalizable, Cloneable, Matrix4fc {
|
|
|
|
private static final long serialVersionUID = 1L;
|
|
|
|
float m00, m01, m02, m03;
|
|
float m10, m11, m12, m13;
|
|
float m20, m21, m22, m23;
|
|
float m30, m31, m32, m33;
|
|
|
|
int properties;
|
|
|
|
/**
|
|
* Create a new {@link Matrix4f} and set it to {@link #identity() identity}.
|
|
*/
|
|
public Matrix4f() {
|
|
this._m00(1.0f)
|
|
._m11(1.0f)
|
|
._m22(1.0f)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_IDENTITY | PROPERTY_AFFINE | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Create a new {@link Matrix4f} by setting its uppper left 3x3 submatrix to the values of the given {@link Matrix3fc}
|
|
* and the rest to identity.
|
|
*
|
|
* @param mat
|
|
* the {@link Matrix3fc}
|
|
*/
|
|
public Matrix4f(Matrix3fc mat) {
|
|
set(mat);
|
|
}
|
|
|
|
/**
|
|
* Create a new {@link Matrix4f} and make it a copy of the given matrix.
|
|
*
|
|
* @param mat
|
|
* the {@link Matrix4fc} to copy the values from
|
|
*/
|
|
public Matrix4f(Matrix4fc mat) {
|
|
set(mat);
|
|
}
|
|
|
|
/**
|
|
* Create a new {@link Matrix4f} and set its upper 4x3 submatrix to the given matrix <code>mat</code>
|
|
* and all other elements to identity.
|
|
*
|
|
* @param mat
|
|
* the {@link Matrix4x3fc} to copy the values from
|
|
*/
|
|
public Matrix4f(Matrix4x3fc mat) {
|
|
set(mat);
|
|
}
|
|
|
|
/**
|
|
* Create a new {@link Matrix4f} and make it a copy of the given matrix.
|
|
* <p>
|
|
* Note that due to the given {@link Matrix4dc} storing values in double-precision and the constructed {@link Matrix4f} storing them
|
|
* in single-precision, there is the possibility of losing precision.
|
|
*
|
|
* @param mat
|
|
* the {@link Matrix4dc} to copy the values from
|
|
*/
|
|
public Matrix4f(Matrix4dc mat) {
|
|
set(mat);
|
|
}
|
|
|
|
/**
|
|
* Create a new 4x4 matrix using the supplied float values.
|
|
* <p>
|
|
* The matrix layout will be:<br><br>
|
|
* m00, m10, m20, m30<br>
|
|
* m01, m11, m21, m31<br>
|
|
* m02, m12, m22, m32<br>
|
|
* m03, m13, m23, m33
|
|
*
|
|
* @param m00
|
|
* the value of m00
|
|
* @param m01
|
|
* the value of m01
|
|
* @param m02
|
|
* the value of m02
|
|
* @param m03
|
|
* the value of m03
|
|
* @param m10
|
|
* the value of m10
|
|
* @param m11
|
|
* the value of m11
|
|
* @param m12
|
|
* the value of m12
|
|
* @param m13
|
|
* the value of m13
|
|
* @param m20
|
|
* the value of m20
|
|
* @param m21
|
|
* the value of m21
|
|
* @param m22
|
|
* the value of m22
|
|
* @param m23
|
|
* the value of m23
|
|
* @param m30
|
|
* the value of m30
|
|
* @param m31
|
|
* the value of m31
|
|
* @param m32
|
|
* the value of m32
|
|
* @param m33
|
|
* the value of m33
|
|
*/
|
|
public Matrix4f(float m00, float m01, float m02, float m03,
|
|
float m10, float m11, float m12, float m13,
|
|
float m20, float m21, float m22, float m23,
|
|
float m30, float m31, float m32, float m33) {
|
|
this._m00(m00)
|
|
._m01(m01)
|
|
._m02(m02)
|
|
._m03(m03)
|
|
._m10(m10)
|
|
._m11(m11)
|
|
._m12(m12)
|
|
._m13(m13)
|
|
._m20(m20)
|
|
._m21(m21)
|
|
._m22(m22)
|
|
._m23(m23)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
.determineProperties();
|
|
}
|
|
|
|
/**
|
|
* Create a new {@link Matrix4f} by reading its 16 float components from the given {@link FloatBuffer}
|
|
* at the buffer's current position.
|
|
* <p>
|
|
* That FloatBuffer is expected to hold the values in column-major order.
|
|
* <p>
|
|
* The buffer's position will not be changed by this method.
|
|
*
|
|
* @param buffer
|
|
* the {@link FloatBuffer} to read the matrix values from
|
|
*/
|
|
public Matrix4f(FloatBuffer buffer) {
|
|
MemUtil.INSTANCE.get(this, buffer.position(), buffer);
|
|
determineProperties();
|
|
}
|
|
|
|
/**
|
|
* Create a new {@link Matrix4f} and initialize its four columns using the supplied vectors.
|
|
*
|
|
* @param col0
|
|
* the first column
|
|
* @param col1
|
|
* the second column
|
|
* @param col2
|
|
* the third column
|
|
* @param col3
|
|
* the fourth column
|
|
*/
|
|
public Matrix4f(Vector4fc col0, Vector4fc col1, Vector4fc col2, Vector4fc col3) {
|
|
set(col0, col1, col2, col3);
|
|
}
|
|
|
|
Matrix4f _properties(int properties) {
|
|
this.properties = properties;
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Assume the given properties about this matrix.
|
|
* <p>
|
|
* Use one or multiple of 0, {@link Matrix4fc#PROPERTY_IDENTITY},
|
|
* {@link Matrix4fc#PROPERTY_TRANSLATION}, {@link Matrix4fc#PROPERTY_AFFINE},
|
|
* {@link Matrix4fc#PROPERTY_PERSPECTIVE}, {@link Matrix4fc#PROPERTY_ORTHONORMAL}.
|
|
*
|
|
* @param properties
|
|
* bitset of the properties to assume about this matrix
|
|
* @return this
|
|
*/
|
|
public Matrix4f assume(int properties) {
|
|
this._properties(properties);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Compute and set the matrix properties returned by {@link #properties()} based
|
|
* on the current matrix element values.
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f determineProperties() {
|
|
int properties = 0;
|
|
if (m03 == 0.0f && m13 == 0.0f) {
|
|
if (m23 == 0.0f && m33 == 1.0f) {
|
|
properties |= PROPERTY_AFFINE;
|
|
if (m00 == 1.0f && m01 == 0.0f && m02 == 0.0f && m10 == 0.0f && m11 == 1.0f && m12 == 0.0f
|
|
&& m20 == 0.0f && m21 == 0.0f && m22 == 1.0f) {
|
|
properties |= PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL;
|
|
if (m30 == 0.0f && m31 == 0.0f && m32 == 0.0f)
|
|
properties |= PROPERTY_IDENTITY;
|
|
}
|
|
/*
|
|
* We do not determine orthogonality, since it would require arbitrary epsilons
|
|
* and is rather expensive (6 dot products) in the worst case.
|
|
*/
|
|
} else if (m01 == 0.0f && m02 == 0.0f && m10 == 0.0f && m12 == 0.0f && m20 == 0.0f && m21 == 0.0f
|
|
&& m30 == 0.0f && m31 == 0.0f && m33 == 0.0f) {
|
|
properties |= PROPERTY_PERSPECTIVE;
|
|
}
|
|
}
|
|
this.properties = properties;
|
|
return this;
|
|
}
|
|
|
|
public int properties() {
|
|
return properties;
|
|
}
|
|
|
|
public float m00() {
|
|
return m00;
|
|
}
|
|
public float m01() {
|
|
return m01;
|
|
}
|
|
public float m02() {
|
|
return m02;
|
|
}
|
|
public float m03() {
|
|
return m03;
|
|
}
|
|
public float m10() {
|
|
return m10;
|
|
}
|
|
public float m11() {
|
|
return m11;
|
|
}
|
|
public float m12() {
|
|
return m12;
|
|
}
|
|
public float m13() {
|
|
return m13;
|
|
}
|
|
public float m20() {
|
|
return m20;
|
|
}
|
|
public float m21() {
|
|
return m21;
|
|
}
|
|
public float m22() {
|
|
return m22;
|
|
}
|
|
public float m23() {
|
|
return m23;
|
|
}
|
|
public float m30() {
|
|
return m30;
|
|
}
|
|
public float m31() {
|
|
return m31;
|
|
}
|
|
public float m32() {
|
|
return m32;
|
|
}
|
|
public float m33() {
|
|
return m33;
|
|
}
|
|
|
|
/**
|
|
* Set the value of the matrix element at column 0 and row 0.
|
|
*
|
|
* @param m00
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m00(float m00) {
|
|
this.m00 = m00;
|
|
properties &= ~PROPERTY_ORTHONORMAL;
|
|
if (m00 != 1.0f)
|
|
properties &= ~(PROPERTY_IDENTITY | PROPERTY_TRANSLATION);
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 0 and row 1.
|
|
*
|
|
* @param m01
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m01(float m01) {
|
|
this.m01 = m01;
|
|
properties &= ~PROPERTY_ORTHONORMAL;
|
|
if (m01 != 0.0f)
|
|
properties &= ~(PROPERTY_IDENTITY | PROPERTY_PERSPECTIVE | PROPERTY_TRANSLATION);
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 0 and row 2.
|
|
*
|
|
* @param m02
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m02(float m02) {
|
|
this.m02 = m02;
|
|
properties &= ~PROPERTY_ORTHONORMAL;
|
|
if (m02 != 0.0f)
|
|
properties &= ~(PROPERTY_IDENTITY | PROPERTY_PERSPECTIVE | PROPERTY_TRANSLATION);
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 0 and row 3.
|
|
*
|
|
* @param m03
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m03(float m03) {
|
|
this.m03 = m03;
|
|
if (m03 != 0.0f)
|
|
properties = 0;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 1 and row 0.
|
|
*
|
|
* @param m10
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m10(float m10) {
|
|
this.m10 = m10;
|
|
properties &= ~PROPERTY_ORTHONORMAL;
|
|
if (m10 != 0.0f)
|
|
properties &= ~(PROPERTY_IDENTITY | PROPERTY_PERSPECTIVE | PROPERTY_TRANSLATION);
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 1 and row 1.
|
|
*
|
|
* @param m11
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m11(float m11) {
|
|
this.m11 = m11;
|
|
properties &= ~PROPERTY_ORTHONORMAL;
|
|
if (m11 != 1.0f)
|
|
properties &= ~(PROPERTY_IDENTITY | PROPERTY_TRANSLATION);
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 1 and row 2.
|
|
*
|
|
* @param m12
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m12(float m12) {
|
|
this.m12 = m12;
|
|
properties &= ~PROPERTY_ORTHONORMAL;
|
|
if (m12 != 0.0f)
|
|
properties &= ~(PROPERTY_IDENTITY | PROPERTY_PERSPECTIVE | PROPERTY_TRANSLATION);
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 1 and row 3.
|
|
*
|
|
* @param m13
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m13(float m13) {
|
|
this.m13 = m13;
|
|
if (m13 != 0.0f)
|
|
properties = 0;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 2 and row 0.
|
|
*
|
|
* @param m20
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m20(float m20) {
|
|
this.m20 = m20;
|
|
properties &= ~PROPERTY_ORTHONORMAL;
|
|
if (m20 != 0.0f)
|
|
properties &= ~(PROPERTY_IDENTITY | PROPERTY_PERSPECTIVE | PROPERTY_TRANSLATION);
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 2 and row 1.
|
|
*
|
|
* @param m21
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m21(float m21) {
|
|
this.m21 = m21;
|
|
properties &= ~PROPERTY_ORTHONORMAL;
|
|
if (m21 != 0.0f)
|
|
properties &= ~(PROPERTY_IDENTITY | PROPERTY_PERSPECTIVE | PROPERTY_TRANSLATION);
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 2 and row 2.
|
|
*
|
|
* @param m22
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m22(float m22) {
|
|
this.m22 = m22;
|
|
properties &= ~PROPERTY_ORTHONORMAL;
|
|
if (m22 != 1.0f)
|
|
properties &= ~(PROPERTY_IDENTITY | PROPERTY_TRANSLATION);
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 2 and row 3.
|
|
*
|
|
* @param m23
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m23(float m23) {
|
|
this.m23 = m23;
|
|
if (m23 != 0.0f)
|
|
properties &= ~(PROPERTY_IDENTITY | PROPERTY_AFFINE | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL);
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 3 and row 0.
|
|
*
|
|
* @param m30
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m30(float m30) {
|
|
this.m30 = m30;
|
|
if (m30 != 0.0f)
|
|
properties &= ~(PROPERTY_IDENTITY | PROPERTY_PERSPECTIVE);
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 3 and row 1.
|
|
*
|
|
* @param m31
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m31(float m31) {
|
|
this.m31 = m31;
|
|
if (m31 != 0.0f)
|
|
properties &= ~(PROPERTY_IDENTITY | PROPERTY_PERSPECTIVE);
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 3 and row 2.
|
|
*
|
|
* @param m32
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m32(float m32) {
|
|
this.m32 = m32;
|
|
if (m32 != 0.0f)
|
|
properties &= ~(PROPERTY_IDENTITY | PROPERTY_PERSPECTIVE);
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 3 and row 3.
|
|
*
|
|
* @param m33
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
public Matrix4f m33(float m33) {
|
|
this.m33 = m33;
|
|
if (m33 != 0.0f)
|
|
properties &= ~(PROPERTY_PERSPECTIVE);
|
|
if (m33 != 1.0f)
|
|
properties &= ~(PROPERTY_IDENTITY | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL | PROPERTY_AFFINE);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set the value of the matrix element at column 0 and row 0 without updating the properties of the matrix.
|
|
*
|
|
* @param m00
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m00(float m00) {
|
|
this.m00 = m00;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 0 and row 1 without updating the properties of the matrix.
|
|
*
|
|
* @param m01
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m01(float m01) {
|
|
this.m01 = m01;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 0 and row 2 without updating the properties of the matrix.
|
|
*
|
|
* @param m02
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m02(float m02) {
|
|
this.m02 = m02;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 0 and row 3 without updating the properties of the matrix.
|
|
*
|
|
* @param m03
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m03(float m03) {
|
|
this.m03 = m03;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 1 and row 0 without updating the properties of the matrix.
|
|
*
|
|
* @param m10
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m10(float m10) {
|
|
this.m10 = m10;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 1 and row 1 without updating the properties of the matrix.
|
|
*
|
|
* @param m11
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m11(float m11) {
|
|
this.m11 = m11;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 1 and row 2 without updating the properties of the matrix.
|
|
*
|
|
* @param m12
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m12(float m12) {
|
|
this.m12 = m12;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 1 and row 3 without updating the properties of the matrix.
|
|
*
|
|
* @param m13
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m13(float m13) {
|
|
this.m13 = m13;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 2 and row 0 without updating the properties of the matrix.
|
|
*
|
|
* @param m20
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m20(float m20) {
|
|
this.m20 = m20;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 2 and row 1 without updating the properties of the matrix.
|
|
*
|
|
* @param m21
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m21(float m21) {
|
|
this.m21 = m21;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 2 and row 2 without updating the properties of the matrix.
|
|
*
|
|
* @param m22
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m22(float m22) {
|
|
this.m22 = m22;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 2 and row 3 without updating the properties of the matrix.
|
|
*
|
|
* @param m23
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m23(float m23) {
|
|
this.m23 = m23;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 3 and row 0 without updating the properties of the matrix.
|
|
*
|
|
* @param m30
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m30(float m30) {
|
|
this.m30 = m30;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 3 and row 1 without updating the properties of the matrix.
|
|
*
|
|
* @param m31
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m31(float m31) {
|
|
this.m31 = m31;
|
|
return this;
|
|
}
|
|
/**
|
|
* Set the value of the matrix element at column 3 and row 2 without updating the properties of the matrix.
|
|
*
|
|
* @param m32
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m32(float m32) {
|
|
this.m32 = m32;
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set the value of the matrix element at column 3 and row 3 without updating the properties of the matrix.
|
|
*
|
|
* @param m33
|
|
* the new value
|
|
* @return this
|
|
*/
|
|
Matrix4f _m33(float m33) {
|
|
this.m33 = m33;
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Reset this matrix to the identity.
|
|
* <p>
|
|
* Please note that if a call to {@link #identity()} is immediately followed by a call to:
|
|
* {@link #translate(float, float, float) translate},
|
|
* {@link #rotate(float, float, float, float) rotate},
|
|
* {@link #scale(float, float, float) scale},
|
|
* {@link #perspective(float, float, float, float) perspective},
|
|
* {@link #frustum(float, float, float, float, float, float) frustum},
|
|
* {@link #ortho(float, float, float, float, float, float) ortho},
|
|
* {@link #ortho2D(float, float, float, float) ortho2D},
|
|
* {@link #lookAt(float, float, float, float, float, float, float, float, float) lookAt},
|
|
* {@link #lookAlong(float, float, float, float, float, float) lookAlong},
|
|
* or any of their overloads, then the call to {@link #identity()} can be omitted and the subsequent call replaced with:
|
|
* {@link #translation(float, float, float) translation},
|
|
* {@link #rotation(float, float, float, float) rotation},
|
|
* {@link #scaling(float, float, float) scaling},
|
|
* {@link #setPerspective(float, float, float, float) setPerspective},
|
|
* {@link #setFrustum(float, float, float, float, float, float) setFrustum},
|
|
* {@link #setOrtho(float, float, float, float, float, float) setOrtho},
|
|
* {@link #setOrtho2D(float, float, float, float) setOrtho2D},
|
|
* {@link #setLookAt(float, float, float, float, float, float, float, float, float) setLookAt},
|
|
* {@link #setLookAlong(float, float, float, float, float, float) setLookAlong},
|
|
* or any of their overloads.
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f identity() {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return this;
|
|
return
|
|
_m00(1.0f).
|
|
_m01(0.0f).
|
|
_m02(0.0f).
|
|
_m03(0.0f).
|
|
_m10(0.0f).
|
|
_m11(1.0f).
|
|
_m12(0.0f).
|
|
_m13(0.0f).
|
|
_m20(0.0f).
|
|
_m21(0.0f).
|
|
_m22(1.0f).
|
|
_m23(0.0f).
|
|
_m30(0.0f).
|
|
_m31(0.0f).
|
|
_m32(0.0f).
|
|
_m33(1.0f).
|
|
_properties(PROPERTY_IDENTITY | PROPERTY_AFFINE | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Store the values of the given matrix <code>m</code> into <code>this</code> matrix.
|
|
*
|
|
* @see #Matrix4f(Matrix4fc)
|
|
* @see #get(Matrix4f)
|
|
*
|
|
* @param m
|
|
* the matrix to copy the values from
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(Matrix4fc m) {
|
|
return
|
|
_m00(m.m00()).
|
|
_m01(m.m01()).
|
|
_m02(m.m02()).
|
|
_m03(m.m03()).
|
|
_m10(m.m10()).
|
|
_m11(m.m11()).
|
|
_m12(m.m12()).
|
|
_m13(m.m13()).
|
|
_m20(m.m20()).
|
|
_m21(m.m21()).
|
|
_m22(m.m22()).
|
|
_m23(m.m23()).
|
|
_m30(m.m30()).
|
|
_m31(m.m31()).
|
|
_m32(m.m32()).
|
|
_m33(m.m33()).
|
|
_properties(m.properties());
|
|
}
|
|
|
|
/**
|
|
* Store the values of the transpose of the given matrix <code>m</code> into <code>this</code> matrix.
|
|
*
|
|
* @param m
|
|
* the matrix to copy the transposed values from
|
|
* @return this
|
|
*/
|
|
public Matrix4f setTransposed(Matrix4fc m) {
|
|
if ((m.properties() & PROPERTY_IDENTITY) != 0)
|
|
return this.identity();
|
|
return setTransposedInternal(m);
|
|
}
|
|
private Matrix4f setTransposedInternal(Matrix4fc m) {
|
|
float nm10 = m.m01(), nm12 = m.m21(), nm13 = m.m31();
|
|
float nm20 = m.m02(), nm21 = m.m12(), nm30 = m.m03();
|
|
float nm31 = m.m13(), nm32 = m.m23();
|
|
return this
|
|
._m00(m.m00())._m01(m.m10())._m02(m.m20())._m03(m.m30())
|
|
._m10(nm10)._m11(m.m11())._m12(nm12)._m13(nm13)
|
|
._m20(nm20)._m21(nm21)._m22(m.m22())._m23(m.m32())
|
|
._m30(nm30)._m31(nm31)._m32(nm32)._m33(m.m33())
|
|
._properties(m.properties() & PROPERTY_IDENTITY);
|
|
}
|
|
|
|
/**
|
|
* Store the values of the given matrix <code>m</code> into <code>this</code> matrix
|
|
* and set the other matrix elements to identity.
|
|
*
|
|
* @see #Matrix4f(Matrix4x3fc)
|
|
*
|
|
* @param m
|
|
* the matrix to copy the values from
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(Matrix4x3fc m) {
|
|
return
|
|
_m00(m.m00()).
|
|
_m01(m.m01()).
|
|
_m02(m.m02()).
|
|
_m03(0.0f).
|
|
_m10(m.m10()).
|
|
_m11(m.m11()).
|
|
_m12(m.m12()).
|
|
_m13(0.0f).
|
|
_m20(m.m20()).
|
|
_m21(m.m21()).
|
|
_m22(m.m22()).
|
|
_m23(0.0f).
|
|
_m30(m.m30()).
|
|
_m31(m.m31()).
|
|
_m32(m.m32()).
|
|
_m33(1.0f).
|
|
_properties(m.properties() | PROPERTY_AFFINE);
|
|
}
|
|
|
|
/**
|
|
* Store the values of the given matrix <code>m</code> into <code>this</code> matrix.
|
|
* <p>
|
|
* Note that due to the given matrix <code>m</code> storing values in double-precision and <code>this</code> matrix storing
|
|
* them in single-precision, there is the possibility to lose precision.
|
|
*
|
|
* @see #Matrix4f(Matrix4dc)
|
|
* @see #get(Matrix4d)
|
|
*
|
|
* @param m
|
|
* the matrix to copy the values from
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(Matrix4dc m) {
|
|
return this
|
|
._m00((float) m.m00())
|
|
._m01((float) m.m01())
|
|
._m02((float) m.m02())
|
|
._m03((float) m.m03())
|
|
._m10((float) m.m10())
|
|
._m11((float) m.m11())
|
|
._m12((float) m.m12())
|
|
._m13((float) m.m13())
|
|
._m20((float) m.m20())
|
|
._m21((float) m.m21())
|
|
._m22((float) m.m22())
|
|
._m23((float) m.m23())
|
|
._m30((float) m.m30())
|
|
._m31((float) m.m31())
|
|
._m32((float) m.m32())
|
|
._m33((float) m.m33())
|
|
._properties(m.properties());
|
|
}
|
|
|
|
/**
|
|
* Set the upper left 3x3 submatrix of this {@link Matrix4f} to the given {@link Matrix3fc}
|
|
* and the rest to identity.
|
|
*
|
|
* @see #Matrix4f(Matrix3fc)
|
|
*
|
|
* @param mat
|
|
* the {@link Matrix3fc}
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(Matrix3fc mat) {
|
|
return this
|
|
._m00(mat.m00())
|
|
._m01(mat.m01())
|
|
._m02(mat.m02())
|
|
._m03(0.0f)
|
|
._m10(mat.m10())
|
|
._m11(mat.m11())
|
|
._m12(mat.m12())
|
|
._m13(0.0f)
|
|
._m20(mat.m20())
|
|
._m21(mat.m21())
|
|
._m22(mat.m22())
|
|
._m23(0.0f)
|
|
._m30(0.0f)
|
|
._m31(0.0f)
|
|
._m32(0.0f)
|
|
._m33(1.0f).
|
|
_properties(PROPERTY_AFFINE);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be equivalent to the rotation specified by the given {@link AxisAngle4f}.
|
|
*
|
|
* @param axisAngle
|
|
* the {@link AxisAngle4f}
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(AxisAngle4f axisAngle) {
|
|
float x = axisAngle.x;
|
|
float y = axisAngle.y;
|
|
float z = axisAngle.z;
|
|
float angle = axisAngle.angle;
|
|
double n = Math.sqrt(x*x + y*y + z*z);
|
|
n = 1/n;
|
|
x *= n;
|
|
y *= n;
|
|
z *= n;
|
|
float s = Math.sin(angle);
|
|
float c = Math.cosFromSin(s, angle);
|
|
float omc = 1.0f - c;
|
|
this._m00((float)(c + x*x*omc))
|
|
._m11((float)(c + y*y*omc))
|
|
._m22((float)(c + z*z*omc));
|
|
float tmp1 = x*y*omc;
|
|
float tmp2 = z*s;
|
|
this._m10((float)(tmp1 - tmp2))
|
|
._m01((float)(tmp1 + tmp2));
|
|
tmp1 = x*z*omc;
|
|
tmp2 = y*s;
|
|
this._m20((float)(tmp1 + tmp2))
|
|
._m02((float)(tmp1 - tmp2));
|
|
tmp1 = y*z*omc;
|
|
tmp2 = x*s;
|
|
return this
|
|
._m21((float)(tmp1 - tmp2))
|
|
._m12((float)(tmp1 + tmp2))
|
|
._m03(0.0f)
|
|
._m13(0.0f)
|
|
._m23(0.0f)
|
|
._m30(0.0f)
|
|
._m31(0.0f)
|
|
._m32(0.0f)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be equivalent to the rotation specified by the given {@link AxisAngle4d}.
|
|
*
|
|
* @param axisAngle
|
|
* the {@link AxisAngle4d}
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(AxisAngle4d axisAngle) {
|
|
double x = axisAngle.x;
|
|
double y = axisAngle.y;
|
|
double z = axisAngle.z;
|
|
double angle = axisAngle.angle;
|
|
double n = Math.sqrt(x*x + y*y + z*z);
|
|
n = 1/n;
|
|
x *= n;
|
|
y *= n;
|
|
z *= n;
|
|
double s = Math.sin(angle);
|
|
double c = Math.cosFromSin(s, angle);
|
|
double omc = 1.0 - c;
|
|
this._m00((float)(c + x*x*omc))
|
|
._m11((float)(c + y*y*omc))
|
|
._m22((float)(c + z*z*omc));
|
|
double tmp1 = x*y*omc;
|
|
double tmp2 = z*s;
|
|
this._m10((float)(tmp1 - tmp2))
|
|
._m01((float)(tmp1 + tmp2));
|
|
tmp1 = x*z*omc;
|
|
tmp2 = y*s;
|
|
this._m20((float)(tmp1 + tmp2))
|
|
._m02((float)(tmp1 - tmp2));
|
|
tmp1 = y*z*omc;
|
|
tmp2 = x*s;
|
|
return this
|
|
._m21((float)(tmp1 - tmp2))
|
|
._m12((float)(tmp1 + tmp2))
|
|
._m03(0.0f)
|
|
._m13(0.0f)
|
|
._m23(0.0f)
|
|
._m30(0.0f)
|
|
._m31(0.0f)
|
|
._m32(0.0f)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be equivalent to the rotation specified by the given {@link Quaternionfc}.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>rotation(q)</code>
|
|
* <p>
|
|
* Reference: <a href="http://www.euclideanspace.com/maths/geometry/rotations/conversions/quaternionToMatrix/">http://www.euclideanspace.com/</a>
|
|
*
|
|
* @see #rotation(Quaternionfc)
|
|
*
|
|
* @param q
|
|
* the {@link Quaternionfc}
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(Quaternionfc q) {
|
|
return rotation(q);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be equivalent to the rotation specified by the given {@link Quaterniondc}.
|
|
* <p>
|
|
* Reference: <a href="http://www.euclideanspace.com/maths/geometry/rotations/conversions/quaternionToMatrix/">http://www.euclideanspace.com/</a>
|
|
*
|
|
* @param q
|
|
* the {@link Quaterniondc}
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(Quaterniondc q) {
|
|
double w2 = q.w() * q.w();
|
|
double x2 = q.x() * q.x();
|
|
double y2 = q.y() * q.y();
|
|
double z2 = q.z() * q.z();
|
|
double zw = q.z() * q.w();
|
|
double xy = q.x() * q.y();
|
|
double xz = q.x() * q.z();
|
|
double yw = q.y() * q.w();
|
|
double yz = q.y() * q.z();
|
|
double xw = q.x() * q.w();
|
|
return
|
|
_m00((float) (w2 + x2 - z2 - y2)).
|
|
_m01((float) (xy + zw + zw + xy)).
|
|
_m02((float) (xz - yw + xz - yw)).
|
|
_m03(0.0f).
|
|
_m10((float) (-zw + xy - zw + xy)).
|
|
_m11((float) (y2 - z2 + w2 - x2)).
|
|
_m12((float) (yz + yz + xw + xw)).
|
|
_m13(0.0f).
|
|
_m20((float) (yw + xz + xz + yw)).
|
|
_m21((float) (yz + yz - xw - xw)).
|
|
_m22((float) (z2 - y2 - x2 + w2)).
|
|
_m30(0.0f).
|
|
_m31(0.0f).
|
|
_m32(0.0f).
|
|
_m33(1.0f).
|
|
_properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Set the upper left 3x3 submatrix of this {@link Matrix4f} to that of the given {@link Matrix4f}
|
|
* and don't change the other elements.
|
|
*
|
|
* @param mat
|
|
* the {@link Matrix4f}
|
|
* @return this
|
|
*/
|
|
public Matrix4f set3x3(Matrix4f mat) {
|
|
MemUtil.INSTANCE.copy3x3(mat, this);
|
|
return _properties(properties & mat.properties & ~(PROPERTY_PERSPECTIVE));
|
|
}
|
|
|
|
|
|
/**
|
|
* Set the upper 4x3 submatrix of this {@link Matrix4f} to the given {@link Matrix4x3fc}
|
|
* and don't change the other elements.
|
|
*
|
|
* @see Matrix4x3f#get(Matrix4f)
|
|
*
|
|
* @param mat
|
|
* the {@link Matrix4x3fc}
|
|
* @return this
|
|
*/
|
|
public Matrix4f set4x3(Matrix4x3fc mat) {
|
|
return
|
|
_m00(mat.m00()).
|
|
_m01(mat.m01()).
|
|
_m02(mat.m02()).
|
|
_m10(mat.m10()).
|
|
_m11(mat.m11()).
|
|
_m12(mat.m12()).
|
|
_m20(mat.m20()).
|
|
_m21(mat.m21()).
|
|
_m22(mat.m22()).
|
|
_m30(mat.m30()).
|
|
_m31(mat.m31()).
|
|
_m32(mat.m32()).
|
|
_properties(properties & mat.properties() & ~(PROPERTY_PERSPECTIVE));
|
|
}
|
|
|
|
/**
|
|
* Set the upper 4x3 submatrix of this {@link Matrix4f} to the upper 4x3 submatrix of the given {@link Matrix4f}
|
|
* and don't change the other elements.
|
|
*
|
|
* @param mat
|
|
* the {@link Matrix4f}
|
|
* @return this
|
|
*/
|
|
public Matrix4f set4x3(Matrix4f mat) {
|
|
MemUtil.INSTANCE.copy4x3(mat, this);
|
|
return _properties(properties & mat.properties & ~(PROPERTY_PERSPECTIVE));
|
|
}
|
|
|
|
/**
|
|
* Multiply this matrix by the supplied <code>right</code> matrix and store the result in <code>this</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the <code>right</code> matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* transformation of the right matrix will be applied first!
|
|
*
|
|
* @param right
|
|
* the right operand of the matrix multiplication
|
|
* @return this
|
|
*/
|
|
public Matrix4f mul(Matrix4fc right) {
|
|
return mul(right, this);
|
|
}
|
|
|
|
public Matrix4f mul(Matrix4fc right, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.set(right);
|
|
else if ((right.properties() & PROPERTY_IDENTITY) != 0)
|
|
return dest.set(this);
|
|
else if ((properties & PROPERTY_TRANSLATION) != 0 && (right.properties() & PROPERTY_AFFINE) != 0)
|
|
return mulTranslationAffine(right, dest);
|
|
else if ((properties & PROPERTY_AFFINE) != 0 && (right.properties() & PROPERTY_AFFINE) != 0)
|
|
return mulAffine(right, dest);
|
|
else if ((properties & PROPERTY_PERSPECTIVE) != 0 && (right.properties() & PROPERTY_AFFINE) != 0)
|
|
return mulPerspectiveAffine(right, dest);
|
|
else if ((right.properties() & PROPERTY_AFFINE) != 0)
|
|
return mulAffineR(right, dest);
|
|
return mul0(right, dest);
|
|
}
|
|
|
|
/**
|
|
* Multiply this matrix by the supplied <code>right</code> matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the <code>right</code> matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* transformation of the right matrix will be applied first!
|
|
* <p>
|
|
* This method neither assumes nor checks for any matrix properties of <code>this</code> or <code>right</code>
|
|
* and will always perform a complete 4x4 matrix multiplication. This method should only be used whenever the
|
|
* multiplied matrices do not have any properties for which there are optimized multiplication methods available.
|
|
*
|
|
* @param right
|
|
* the right operand of the matrix multiplication
|
|
* @return this
|
|
*/
|
|
public Matrix4f mul0(Matrix4fc right) {
|
|
return mul0(right, this);
|
|
}
|
|
|
|
public Matrix4f mul0(Matrix4fc right, Matrix4f dest) {
|
|
float nm00 = Math.fma(m00, right.m00(), Math.fma(m10, right.m01(), Math.fma(m20, right.m02(), m30 * right.m03())));
|
|
float nm01 = Math.fma(m01, right.m00(), Math.fma(m11, right.m01(), Math.fma(m21, right.m02(), m31 * right.m03())));
|
|
float nm02 = Math.fma(m02, right.m00(), Math.fma(m12, right.m01(), Math.fma(m22, right.m02(), m32 * right.m03())));
|
|
float nm03 = Math.fma(m03, right.m00(), Math.fma(m13, right.m01(), Math.fma(m23, right.m02(), m33 * right.m03())));
|
|
float nm10 = Math.fma(m00, right.m10(), Math.fma(m10, right.m11(), Math.fma(m20, right.m12(), m30 * right.m13())));
|
|
float nm11 = Math.fma(m01, right.m10(), Math.fma(m11, right.m11(), Math.fma(m21, right.m12(), m31 * right.m13())));
|
|
float nm12 = Math.fma(m02, right.m10(), Math.fma(m12, right.m11(), Math.fma(m22, right.m12(), m32 * right.m13())));
|
|
float nm13 = Math.fma(m03, right.m10(), Math.fma(m13, right.m11(), Math.fma(m23, right.m12(), m33 * right.m13())));
|
|
float nm20 = Math.fma(m00, right.m20(), Math.fma(m10, right.m21(), Math.fma(m20, right.m22(), m30 * right.m23())));
|
|
float nm21 = Math.fma(m01, right.m20(), Math.fma(m11, right.m21(), Math.fma(m21, right.m22(), m31 * right.m23())));
|
|
float nm22 = Math.fma(m02, right.m20(), Math.fma(m12, right.m21(), Math.fma(m22, right.m22(), m32 * right.m23())));
|
|
float nm23 = Math.fma(m03, right.m20(), Math.fma(m13, right.m21(), Math.fma(m23, right.m22(), m33 * right.m23())));
|
|
float nm30 = Math.fma(m00, right.m30(), Math.fma(m10, right.m31(), Math.fma(m20, right.m32(), m30 * right.m33())));
|
|
float nm31 = Math.fma(m01, right.m30(), Math.fma(m11, right.m31(), Math.fma(m21, right.m32(), m31 * right.m33())));
|
|
float nm32 = Math.fma(m02, right.m30(), Math.fma(m12, right.m31(), Math.fma(m22, right.m32(), m32 * right.m33())));
|
|
float nm33 = Math.fma(m03, right.m30(), Math.fma(m13, right.m31(), Math.fma(m23, right.m32(), m33 * right.m33())));
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(nm33)
|
|
._properties(0);
|
|
}
|
|
|
|
/**
|
|
* Multiply this matrix by the matrix with the supplied elements.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the <code>right</code> matrix whose
|
|
* elements are supplied via the parameters, then the new matrix will be <code>M * R</code>.
|
|
* So when transforming a vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* transformation of the right matrix will be applied first!
|
|
*
|
|
* @param r00
|
|
* the m00 element of the right matrix
|
|
* @param r01
|
|
* the m01 element of the right matrix
|
|
* @param r02
|
|
* the m02 element of the right matrix
|
|
* @param r03
|
|
* the m03 element of the right matrix
|
|
* @param r10
|
|
* the m10 element of the right matrix
|
|
* @param r11
|
|
* the m11 element of the right matrix
|
|
* @param r12
|
|
* the m12 element of the right matrix
|
|
* @param r13
|
|
* the m13 element of the right matrix
|
|
* @param r20
|
|
* the m20 element of the right matrix
|
|
* @param r21
|
|
* the m21 element of the right matrix
|
|
* @param r22
|
|
* the m22 element of the right matrix
|
|
* @param r23
|
|
* the m23 element of the right matrix
|
|
* @param r30
|
|
* the m30 element of the right matrix
|
|
* @param r31
|
|
* the m31 element of the right matrix
|
|
* @param r32
|
|
* the m32 element of the right matrix
|
|
* @param r33
|
|
* the m33 element of the right matrix
|
|
* @return this
|
|
*/
|
|
public Matrix4f mul(
|
|
float r00, float r01, float r02, float r03,
|
|
float r10, float r11, float r12, float r13,
|
|
float r20, float r21, float r22, float r23,
|
|
float r30, float r31, float r32, float r33) {
|
|
return mul(r00, r01, r02, r03, r10, r11, r12, r13, r20, r21, r22, r23, r30, r31, r32, r33, this);
|
|
}
|
|
|
|
public Matrix4f mul(
|
|
float r00, float r01, float r02, float r03,
|
|
float r10, float r11, float r12, float r13,
|
|
float r20, float r21, float r22, float r23,
|
|
float r30, float r31, float r32, float r33, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.set(r00, r01, r02, r03, r10, r11, r12, r13, r20, r21, r22, r23, r30, r31, r32, r33);
|
|
else if ((properties & PROPERTY_AFFINE) != 0)
|
|
return mulAffineL(r00, r01, r02, r03, r10, r11, r12, r13, r20, r21, r22, r23, r30, r31, r32, r33, dest);
|
|
return mulGeneric(r00, r01, r02, r03, r10, r11, r12, r13, r20, r21, r22, r23, r30, r31, r32, r33, dest);
|
|
}
|
|
private Matrix4f mulAffineL(
|
|
float r00, float r01, float r02, float r03,
|
|
float r10, float r11, float r12, float r13,
|
|
float r20, float r21, float r22, float r23,
|
|
float r30, float r31, float r32, float r33, Matrix4f dest) {
|
|
float nm00 = Math.fma(m00, r00, Math.fma(m10, r01, Math.fma(m20, r02, m30 * r03)));
|
|
float nm01 = Math.fma(m01, r00, Math.fma(m11, r01, Math.fma(m21, r02, m31 * r03)));
|
|
float nm02 = Math.fma(m02, r00, Math.fma(m12, r01, Math.fma(m22, r02, m32 * r03)));
|
|
float nm03 = r03;
|
|
float nm10 = Math.fma(m00, r10, Math.fma(m10, r11, Math.fma(m20, r12, m30 * r13)));
|
|
float nm11 = Math.fma(m01, r10, Math.fma(m11, r11, Math.fma(m21, r12, m31 * r13)));
|
|
float nm12 = Math.fma(m02, r10, Math.fma(m12, r11, Math.fma(m22, r12, m32 * r13)));
|
|
float nm13 = r13;
|
|
float nm20 = Math.fma(m00, r20, Math.fma(m10, r21, Math.fma(m20, r22, m30 * r23)));
|
|
float nm21 = Math.fma(m01, r20, Math.fma(m11, r21, Math.fma(m21, r22, m31 * r23)));
|
|
float nm22 = Math.fma(m02, r20, Math.fma(m12, r21, Math.fma(m22, r22, m32 * r23)));
|
|
float nm23 = r23;
|
|
float nm30 = Math.fma(m00, r30, Math.fma(m10, r31, Math.fma(m20, r32, m30 * r33)));
|
|
float nm31 = Math.fma(m01, r30, Math.fma(m11, r31, Math.fma(m21, r32, m31 * r33)));
|
|
float nm32 = Math.fma(m02, r30, Math.fma(m12, r31, Math.fma(m22, r32, m32 * r33)));
|
|
float nm33 = r33;
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(nm33)
|
|
._properties(PROPERTY_AFFINE);
|
|
}
|
|
private Matrix4f mulGeneric(
|
|
float r00, float r01, float r02, float r03,
|
|
float r10, float r11, float r12, float r13,
|
|
float r20, float r21, float r22, float r23,
|
|
float r30, float r31, float r32, float r33, Matrix4f dest) {
|
|
float nm00 = Math.fma(m00, r00, Math.fma(m10, r01, Math.fma(m20, r02, m30 * r03)));
|
|
float nm01 = Math.fma(m01, r00, Math.fma(m11, r01, Math.fma(m21, r02, m31 * r03)));
|
|
float nm02 = Math.fma(m02, r00, Math.fma(m12, r01, Math.fma(m22, r02, m32 * r03)));
|
|
float nm03 = Math.fma(m03, r00, Math.fma(m13, r01, Math.fma(m23, r02, m33 * r03)));
|
|
float nm10 = Math.fma(m00, r10, Math.fma(m10, r11, Math.fma(m20, r12, m30 * r13)));
|
|
float nm11 = Math.fma(m01, r10, Math.fma(m11, r11, Math.fma(m21, r12, m31 * r13)));
|
|
float nm12 = Math.fma(m02, r10, Math.fma(m12, r11, Math.fma(m22, r12, m32 * r13)));
|
|
float nm13 = Math.fma(m03, r10, Math.fma(m13, r11, Math.fma(m23, r12, m33 * r13)));
|
|
float nm20 = Math.fma(m00, r20, Math.fma(m10, r21, Math.fma(m20, r22, m30 * r23)));
|
|
float nm21 = Math.fma(m01, r20, Math.fma(m11, r21, Math.fma(m21, r22, m31 * r23)));
|
|
float nm22 = Math.fma(m02, r20, Math.fma(m12, r21, Math.fma(m22, r22, m32 * r23)));
|
|
float nm23 = Math.fma(m03, r20, Math.fma(m13, r21, Math.fma(m23, r22, m33 * r23)));
|
|
float nm30 = Math.fma(m00, r30, Math.fma(m10, r31, Math.fma(m20, r32, m30 * r33)));
|
|
float nm31 = Math.fma(m01, r30, Math.fma(m11, r31, Math.fma(m21, r32, m31 * r33)));
|
|
float nm32 = Math.fma(m02, r30, Math.fma(m12, r31, Math.fma(m22, r32, m32 * r33)));
|
|
float nm33 = Math.fma(m03, r30, Math.fma(m13, r31, Math.fma(m23, r32, m33 * r33)));
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(nm33)
|
|
._properties(0);
|
|
}
|
|
|
|
/**
|
|
* Multiply this matrix by the 3x3 matrix with the supplied elements expanded to a 4x4 matrix with
|
|
* all other matrix elements set to identity.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the <code>right</code> matrix whose
|
|
* elements are supplied via the parameters, then the new matrix will be <code>M * R</code>.
|
|
* So when transforming a vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* transformation of the right matrix will be applied first!
|
|
*
|
|
* @param r00
|
|
* the m00 element of the right matrix
|
|
* @param r01
|
|
* the m01 element of the right matrix
|
|
* @param r02
|
|
* the m02 element of the right matrix
|
|
* @param r10
|
|
* the m10 element of the right matrix
|
|
* @param r11
|
|
* the m11 element of the right matrix
|
|
* @param r12
|
|
* the m12 element of the right matrix
|
|
* @param r20
|
|
* the m20 element of the right matrix
|
|
* @param r21
|
|
* the m21 element of the right matrix
|
|
* @param r22
|
|
* the m22 element of the right matrix
|
|
* @return this
|
|
*/
|
|
public Matrix4f mul3x3(
|
|
float r00, float r01, float r02,
|
|
float r10, float r11, float r12,
|
|
float r20, float r21, float r22) {
|
|
return mul3x3(r00, r01, r02, r10, r11, r12, r20, r21, r22, this);
|
|
}
|
|
public Matrix4f mul3x3(
|
|
float r00, float r01, float r02,
|
|
float r10, float r11, float r12,
|
|
float r20, float r21, float r22, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.set(r00, r01, r02, 0, r10, r11, r12, 0, r20, r21, r22, 0, 0, 0, 0, 1);
|
|
return mulGeneric3x3(r00, r01, r02, r10, r11, r12, r20, r21, r22, dest);
|
|
}
|
|
private Matrix4f mulGeneric3x3(
|
|
float r00, float r01, float r02,
|
|
float r10, float r11, float r12,
|
|
float r20, float r21, float r22, Matrix4f dest) {
|
|
float nm00 = Math.fma(m00, r00, Math.fma(m10, r01, m20 * r02));
|
|
float nm01 = Math.fma(m01, r00, Math.fma(m11, r01, m21 * r02));
|
|
float nm02 = Math.fma(m02, r00, Math.fma(m12, r01, m22 * r02));
|
|
float nm03 = Math.fma(m03, r00, Math.fma(m13, r01, m23 * r02));
|
|
float nm10 = Math.fma(m00, r10, Math.fma(m10, r11, m20 * r12));
|
|
float nm11 = Math.fma(m01, r10, Math.fma(m11, r11, m21 * r12));
|
|
float nm12 = Math.fma(m02, r10, Math.fma(m12, r11, m22 * r12));
|
|
float nm13 = Math.fma(m03, r10, Math.fma(m13, r11, m23 * r12));
|
|
float nm20 = Math.fma(m00, r20, Math.fma(m10, r21, m20 * r22));
|
|
float nm21 = Math.fma(m01, r20, Math.fma(m11, r21, m21 * r22));
|
|
float nm22 = Math.fma(m02, r20, Math.fma(m12, r21, m22 * r22));
|
|
float nm23 = Math.fma(m03, r20, Math.fma(m13, r21, m23 * r22));
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(this.properties & PROPERTY_AFFINE);
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply this matrix by the supplied <code>left</code> matrix and store the result in <code>this</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the <code>left</code> matrix,
|
|
* then the new matrix will be <code>L * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>L * M * v</code>, the
|
|
* transformation of <code>this</code> matrix will be applied first!
|
|
*
|
|
* @param left
|
|
* the left operand of the matrix multiplication
|
|
* @return this
|
|
*/
|
|
public Matrix4f mulLocal(Matrix4fc left) {
|
|
return mulLocal(left, this);
|
|
}
|
|
|
|
public Matrix4f mulLocal(Matrix4fc left, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.set(left);
|
|
else if ((left.properties() & PROPERTY_IDENTITY) != 0)
|
|
return dest.set(this);
|
|
else if ((properties & PROPERTY_AFFINE) != 0 && (left.properties() & PROPERTY_AFFINE) != 0)
|
|
return mulLocalAffine(left, dest);
|
|
return mulLocalGeneric(left, dest);
|
|
}
|
|
private Matrix4f mulLocalGeneric(Matrix4fc left, Matrix4f dest) {
|
|
float nm00 = Math.fma(left.m00(), m00, Math.fma(left.m10(), m01, Math.fma(left.m20(), m02, left.m30() * m03)));
|
|
float nm01 = Math.fma(left.m01(), m00, Math.fma(left.m11(), m01, Math.fma(left.m21(), m02, left.m31() * m03)));
|
|
float nm02 = Math.fma(left.m02(), m00, Math.fma(left.m12(), m01, Math.fma(left.m22(), m02, left.m32() * m03)));
|
|
float nm03 = Math.fma(left.m03(), m00, Math.fma(left.m13(), m01, Math.fma(left.m23(), m02, left.m33() * m03)));
|
|
float nm10 = Math.fma(left.m00(), m10, Math.fma(left.m10(), m11, Math.fma(left.m20(), m12, left.m30() * m13)));
|
|
float nm11 = Math.fma(left.m01(), m10, Math.fma(left.m11(), m11, Math.fma(left.m21(), m12, left.m31() * m13)));
|
|
float nm12 = Math.fma(left.m02(), m10, Math.fma(left.m12(), m11, Math.fma(left.m22(), m12, left.m32() * m13)));
|
|
float nm13 = Math.fma(left.m03(), m10, Math.fma(left.m13(), m11, Math.fma(left.m23(), m12, left.m33() * m13)));
|
|
float nm20 = Math.fma(left.m00(), m20, Math.fma(left.m10(), m21, Math.fma(left.m20(), m22, left.m30() * m23)));
|
|
float nm21 = Math.fma(left.m01(), m20, Math.fma(left.m11(), m21, Math.fma(left.m21(), m22, left.m31() * m23)));
|
|
float nm22 = Math.fma(left.m02(), m20, Math.fma(left.m12(), m21, Math.fma(left.m22(), m22, left.m32() * m23)));
|
|
float nm23 = Math.fma(left.m03(), m20, Math.fma(left.m13(), m21, Math.fma(left.m23(), m22, left.m33() * m23)));
|
|
float nm30 = Math.fma(left.m00(), m30, Math.fma(left.m10(), m31, Math.fma(left.m20(), m32, left.m30() * m33)));
|
|
float nm31 = Math.fma(left.m01(), m30, Math.fma(left.m11(), m31, Math.fma(left.m21(), m32, left.m31() * m33)));
|
|
float nm32 = Math.fma(left.m02(), m30, Math.fma(left.m12(), m31, Math.fma(left.m22(), m32, left.m32() * m33)));
|
|
float nm33 = Math.fma(left.m03(), m30, Math.fma(left.m13(), m31, Math.fma(left.m23(), m32, left.m33() * m33)));
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(nm33)
|
|
._properties(0);
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply this matrix by the supplied <code>left</code> matrix, both of which are assumed to be {@link #isAffine() affine}, and store the result in <code>this</code>.
|
|
* <p>
|
|
* This method assumes that <code>this</code> matrix and the given <code>left</code> matrix both represent an {@link #isAffine() affine} transformation
|
|
* (i.e. their last rows are equal to <code>(0, 0, 0, 1)</code>)
|
|
* and can be used to speed up matrix multiplication if the matrices only represent affine transformations, such as translation, rotation, scaling and shearing (in any combination).
|
|
* <p>
|
|
* This method will not modify either the last row of <code>this</code> or the last row of <code>left</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the <code>left</code> matrix,
|
|
* then the new matrix will be <code>L * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>L * M * v</code>, the
|
|
* transformation of <code>this</code> matrix will be applied first!
|
|
*
|
|
* @param left
|
|
* the left operand of the matrix multiplication (the last row is assumed to be <code>(0, 0, 0, 1)</code>)
|
|
* @return this
|
|
*/
|
|
public Matrix4f mulLocalAffine(Matrix4fc left) {
|
|
return mulLocalAffine(left, this);
|
|
}
|
|
|
|
public Matrix4f mulLocalAffine(Matrix4fc left, Matrix4f dest) {
|
|
float nm00 = left.m00() * m00 + left.m10() * m01 + left.m20() * m02;
|
|
float nm01 = left.m01() * m00 + left.m11() * m01 + left.m21() * m02;
|
|
float nm02 = left.m02() * m00 + left.m12() * m01 + left.m22() * m02;
|
|
float nm03 = left.m03();
|
|
float nm10 = left.m00() * m10 + left.m10() * m11 + left.m20() * m12;
|
|
float nm11 = left.m01() * m10 + left.m11() * m11 + left.m21() * m12;
|
|
float nm12 = left.m02() * m10 + left.m12() * m11 + left.m22() * m12;
|
|
float nm13 = left.m13();
|
|
float nm20 = left.m00() * m20 + left.m10() * m21 + left.m20() * m22;
|
|
float nm21 = left.m01() * m20 + left.m11() * m21 + left.m21() * m22;
|
|
float nm22 = left.m02() * m20 + left.m12() * m21 + left.m22() * m22;
|
|
float nm23 = left.m23();
|
|
float nm30 = left.m00() * m30 + left.m10() * m31 + left.m20() * m32 + left.m30();
|
|
float nm31 = left.m01() * m30 + left.m11() * m31 + left.m21() * m32 + left.m31();
|
|
float nm32 = left.m02() * m30 + left.m12() * m31 + left.m22() * m32 + left.m32();
|
|
float nm33 = left.m33();
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(nm33)
|
|
._properties(PROPERTY_AFFINE | (this.properties() & left.properties() & PROPERTY_ORTHONORMAL));
|
|
}
|
|
|
|
/**
|
|
* Multiply this matrix by the supplied <code>right</code> matrix and store the result in <code>this</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the <code>right</code> matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* transformation of the right matrix will be applied first!
|
|
*
|
|
* @param right
|
|
* the right operand of the matrix multiplication
|
|
* @return this
|
|
*/
|
|
public Matrix4f mul(Matrix4x3fc right) {
|
|
return mul(right, this);
|
|
}
|
|
|
|
public Matrix4f mul(Matrix4x3fc right, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.set(right);
|
|
else if ((right.properties() & PROPERTY_IDENTITY) != 0)
|
|
return dest.set(this);
|
|
else if ((properties & PROPERTY_TRANSLATION) != 0)
|
|
return mulTranslation(right, dest);
|
|
else if ((properties & PROPERTY_AFFINE) != 0)
|
|
return mulAffine(right, dest);
|
|
else if ((properties & PROPERTY_PERSPECTIVE) != 0)
|
|
return mulPerspectiveAffine(right, dest);
|
|
return mulGeneric(right, dest);
|
|
}
|
|
private Matrix4f mulTranslation(Matrix4x3fc right, Matrix4f dest) {
|
|
return dest
|
|
._m00(right.m00())
|
|
._m01(right.m01())
|
|
._m02(right.m02())
|
|
._m03(m03)
|
|
._m10(right.m10())
|
|
._m11(right.m11())
|
|
._m12(right.m12())
|
|
._m13(m13)
|
|
._m20(right.m20())
|
|
._m21(right.m21())
|
|
._m22(right.m22())
|
|
._m23(m23)
|
|
._m30(right.m30() + m30)
|
|
._m31(right.m31() + m31)
|
|
._m32(right.m32() + m32)
|
|
._m33(m33)
|
|
._properties(PROPERTY_AFFINE | (right.properties() & PROPERTY_ORTHONORMAL));
|
|
}
|
|
private Matrix4f mulAffine(Matrix4x3fc right, Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
float m20 = this.m20, m21 = this.m21, m22 = this.m22;
|
|
float rm00 = right.m00(), rm01 = right.m01(), rm02 = right.m02();
|
|
float rm10 = right.m10(), rm11 = right.m11(), rm12 = right.m12();
|
|
float rm20 = right.m20(), rm21 = right.m21(), rm22 = right.m22();
|
|
float rm30 = right.m30(), rm31 = right.m31(), rm32 = right.m32();
|
|
return dest
|
|
._m00(Math.fma(m00, rm00, Math.fma(m10, rm01, m20 * rm02)))
|
|
._m01(Math.fma(m01, rm00, Math.fma(m11, rm01, m21 * rm02)))
|
|
._m02(Math.fma(m02, rm00, Math.fma(m12, rm01, m22 * rm02)))
|
|
._m03(m03)
|
|
._m10(Math.fma(m00, rm10, Math.fma(m10, rm11, m20 * rm12)))
|
|
._m11(Math.fma(m01, rm10, Math.fma(m11, rm11, m21 * rm12)))
|
|
._m12(Math.fma(m02, rm10, Math.fma(m12, rm11, m22 * rm12)))
|
|
._m13(m13)
|
|
._m20(Math.fma(m00, rm20, Math.fma(m10, rm21, m20 * rm22)))
|
|
._m21(Math.fma(m01, rm20, Math.fma(m11, rm21, m21 * rm22)))
|
|
._m22(Math.fma(m02, rm20, Math.fma(m12, rm21, m22 * rm22)))
|
|
._m23(m23)
|
|
._m30(Math.fma(m00, rm30, Math.fma(m10, rm31, Math.fma(m20, rm32, m30))))
|
|
._m31(Math.fma(m01, rm30, Math.fma(m11, rm31, Math.fma(m21, rm32, m31))))
|
|
._m32(Math.fma(m02, rm30, Math.fma(m12, rm31, Math.fma(m22, rm32, m32))))
|
|
._m33(m33)
|
|
._properties(PROPERTY_AFFINE | (this.properties & right.properties() & PROPERTY_ORTHONORMAL));
|
|
}
|
|
private Matrix4f mulGeneric(Matrix4x3fc right, Matrix4f dest) {
|
|
float nm00 = Math.fma(m00, right.m00(), Math.fma(m10, right.m01(), m20 * right.m02()));
|
|
float nm01 = Math.fma(m01, right.m00(), Math.fma(m11, right.m01(), m21 * right.m02()));
|
|
float nm02 = Math.fma(m02, right.m00(), Math.fma(m12, right.m01(), m22 * right.m02()));
|
|
float nm03 = Math.fma(m03, right.m00(), Math.fma(m13, right.m01(), m23 * right.m02()));
|
|
float nm10 = Math.fma(m00, right.m10(), Math.fma(m10, right.m11(), m20 * right.m12()));
|
|
float nm11 = Math.fma(m01, right.m10(), Math.fma(m11, right.m11(), m21 * right.m12()));
|
|
float nm12 = Math.fma(m02, right.m10(), Math.fma(m12, right.m11(), m22 * right.m12()));
|
|
float nm13 = Math.fma(m03, right.m10(), Math.fma(m13, right.m11(), m23 * right.m12()));
|
|
float nm20 = Math.fma(m00, right.m20(), Math.fma(m10, right.m21(), m20 * right.m22()));
|
|
float nm21 = Math.fma(m01, right.m20(), Math.fma(m11, right.m21(), m21 * right.m22()));
|
|
float nm22 = Math.fma(m02, right.m20(), Math.fma(m12, right.m21(), m22 * right.m22()));
|
|
float nm23 = Math.fma(m03, right.m20(), Math.fma(m13, right.m21(), m23 * right.m22()));
|
|
float nm30 = Math.fma(m00, right.m30(), Math.fma(m10, right.m31(), Math.fma(m20, right.m32(), m30)));
|
|
float nm31 = Math.fma(m01, right.m30(), Math.fma(m11, right.m31(), Math.fma(m21, right.m32(), m31)));
|
|
float nm32 = Math.fma(m02, right.m30(), Math.fma(m12, right.m31(), Math.fma(m22, right.m32(), m32)));
|
|
float nm33 = Math.fma(m03, right.m30(), Math.fma(m13, right.m31(), Math.fma(m23, right.m32(), m33)));
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(nm33)
|
|
._properties(properties & ~(PROPERTY_IDENTITY | PROPERTY_PERSPECTIVE | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
}
|
|
|
|
/**
|
|
* Multiply this matrix by the supplied <code>right</code> matrix and store the result in <code>this</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the <code>right</code> matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* transformation of the right matrix will be applied first!
|
|
*
|
|
* @param right
|
|
* the right operand of the matrix multiplication
|
|
* @return this
|
|
*/
|
|
public Matrix4f mul(Matrix3x2fc right) {
|
|
return mul(right, this);
|
|
}
|
|
|
|
public Matrix4f mul(Matrix3x2fc right, Matrix4f dest) {
|
|
float nm00 = m00 * right.m00() + m10 * right.m01();
|
|
float nm01 = m01 * right.m00() + m11 * right.m01();
|
|
float nm02 = m02 * right.m00() + m12 * right.m01();
|
|
float nm03 = m03 * right.m00() + m13 * right.m01();
|
|
float nm10 = m00 * right.m10() + m10 * right.m11();
|
|
float nm11 = m01 * right.m10() + m11 * right.m11();
|
|
float nm12 = m02 * right.m10() + m12 * right.m11();
|
|
float nm13 = m03 * right.m10() + m13 * right.m11();
|
|
float nm30 = m00 * right.m20() + m10 * right.m21() + m30;
|
|
float nm31 = m01 * right.m20() + m11 * right.m21() + m31;
|
|
float nm32 = m02 * right.m20() + m12 * right.m21() + m32;
|
|
float nm33 = m03 * right.m20() + m13 * right.m21() + m33;
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m20(m20)
|
|
._m21(m21)
|
|
._m22(m22)
|
|
._m23(m23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(nm33)
|
|
._properties(properties & ~(PROPERTY_IDENTITY | PROPERTY_PERSPECTIVE | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
}
|
|
|
|
/**
|
|
* Multiply <code>this</code> symmetric perspective projection matrix by the supplied {@link #isAffine() affine} <code>view</code> matrix.
|
|
* <p>
|
|
* If <code>P</code> is <code>this</code> matrix and <code>V</code> the <code>view</code> matrix,
|
|
* then the new matrix will be <code>P * V</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>P * V * v</code>, the
|
|
* transformation of the <code>view</code> matrix will be applied first!
|
|
*
|
|
* @param view
|
|
* the {@link #isAffine() affine} matrix to multiply <code>this</code> symmetric perspective projection matrix by
|
|
* @return this
|
|
*/
|
|
public Matrix4f mulPerspectiveAffine(Matrix4fc view) {
|
|
return mulPerspectiveAffine(view, this);
|
|
}
|
|
|
|
public Matrix4f mulPerspectiveAffine(Matrix4fc view, Matrix4f dest) {
|
|
float nm00 = m00 * view.m00(), nm01 = m11 * view.m01(), nm02 = m22 * view.m02(), nm03 = m23 * view.m02();
|
|
float nm10 = m00 * view.m10(), nm11 = m11 * view.m11(), nm12 = m22 * view.m12(), nm13 = m23 * view.m12();
|
|
float nm20 = m00 * view.m20(), nm21 = m11 * view.m21(), nm22 = m22 * view.m22(), nm23 = m23 * view.m22();
|
|
float nm30 = m00 * view.m30(), nm31 = m11 * view.m31(), nm32 = m22 * view.m32() + m32, nm33 = m23 * view.m32();
|
|
return dest
|
|
._m00(nm00)._m01(nm01)._m02(nm02)._m03(nm03)
|
|
._m10(nm10)._m11(nm11)._m12(nm12)._m13(nm13)
|
|
._m20(nm20)._m21(nm21)._m22(nm22)._m23(nm23)
|
|
._m30(nm30)._m31(nm31)._m32(nm32)._m33(nm33)
|
|
._properties(0);
|
|
}
|
|
|
|
/**
|
|
* Multiply <code>this</code> symmetric perspective projection matrix by the supplied <code>view</code> matrix.
|
|
* <p>
|
|
* If <code>P</code> is <code>this</code> matrix and <code>V</code> the <code>view</code> matrix,
|
|
* then the new matrix will be <code>P * V</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>P * V * v</code>, the
|
|
* transformation of the <code>view</code> matrix will be applied first!
|
|
*
|
|
* @param view
|
|
* the matrix to multiply <code>this</code> symmetric perspective projection matrix by
|
|
* @return this
|
|
*/
|
|
public Matrix4f mulPerspectiveAffine(Matrix4x3fc view) {
|
|
return mulPerspectiveAffine(view, this);
|
|
}
|
|
|
|
public Matrix4f mulPerspectiveAffine(Matrix4x3fc view, Matrix4f dest) {
|
|
float lm00 = m00, lm11 = m11, lm22 = m22, lm23 = m23;
|
|
return dest.
|
|
_m00(lm00 * view.m00())._m01(lm11 * view.m01())._m02(lm22 * view.m02())._m03(lm23 * view.m02()).
|
|
_m10(lm00 * view.m10())._m11(lm11 * view.m11())._m12(lm22 * view.m12())._m13(lm23 * view.m12()).
|
|
_m20(lm00 * view.m20())._m21(lm11 * view.m21())._m22(lm22 * view.m22())._m23(lm23 * view.m22()).
|
|
_m30(lm00 * view.m30())._m31(lm11 * view.m31())._m32(lm22 * view.m32() + m32)._m33(lm23 * view.m32()).
|
|
_properties(0);
|
|
}
|
|
|
|
/**
|
|
* Multiply this matrix by the supplied <code>right</code> matrix, which is assumed to be {@link #isAffine() affine}, and store the result in <code>this</code>.
|
|
* <p>
|
|
* This method assumes that the given <code>right</code> matrix represents an {@link #isAffine() affine} transformation (i.e. its last row is equal to <code>(0, 0, 0, 1)</code>)
|
|
* and can be used to speed up matrix multiplication if the matrix only represents affine transformations, such as translation, rotation, scaling and shearing (in any combination).
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the <code>right</code> matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* transformation of the right matrix will be applied first!
|
|
*
|
|
* @param right
|
|
* the right operand of the matrix multiplication (the last row is assumed to be <code>(0, 0, 0, 1)</code>)
|
|
* @return this
|
|
*/
|
|
public Matrix4f mulAffineR(Matrix4fc right) {
|
|
return mulAffineR(right, this);
|
|
}
|
|
|
|
public Matrix4f mulAffineR(Matrix4fc right, Matrix4f dest) {
|
|
float nm00 = Math.fma(m00, right.m00(), Math.fma(m10, right.m01(), m20 * right.m02()));
|
|
float nm01 = Math.fma(m01, right.m00(), Math.fma(m11, right.m01(), m21 * right.m02()));
|
|
float nm02 = Math.fma(m02, right.m00(), Math.fma(m12, right.m01(), m22 * right.m02()));
|
|
float nm03 = Math.fma(m03, right.m00(), Math.fma(m13, right.m01(), m23 * right.m02()));
|
|
float nm10 = Math.fma(m00, right.m10(), Math.fma(m10, right.m11(), m20 * right.m12()));
|
|
float nm11 = Math.fma(m01, right.m10(), Math.fma(m11, right.m11(), m21 * right.m12()));
|
|
float nm12 = Math.fma(m02, right.m10(), Math.fma(m12, right.m11(), m22 * right.m12()));
|
|
float nm13 = Math.fma(m03, right.m10(), Math.fma(m13, right.m11(), m23 * right.m12()));
|
|
float nm20 = Math.fma(m00, right.m20(), Math.fma(m10, right.m21(), m20 * right.m22()));
|
|
float nm21 = Math.fma(m01, right.m20(), Math.fma(m11, right.m21(), m21 * right.m22()));
|
|
float nm22 = Math.fma(m02, right.m20(), Math.fma(m12, right.m21(), m22 * right.m22()));
|
|
float nm23 = Math.fma(m03, right.m20(), Math.fma(m13, right.m21(), m23 * right.m22()));
|
|
float nm30 = Math.fma(m00, right.m30(), Math.fma(m10, right.m31(), Math.fma(m20, right.m32(), m30)));
|
|
float nm31 = Math.fma(m01, right.m30(), Math.fma(m11, right.m31(), Math.fma(m21, right.m32(), m31)));
|
|
float nm32 = Math.fma(m02, right.m30(), Math.fma(m12, right.m31(), Math.fma(m22, right.m32(), m32)));
|
|
float nm33 = Math.fma(m03, right.m30(), Math.fma(m13, right.m31(), Math.fma(m23, right.m32(), m33)));
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(nm33)
|
|
._properties(properties & ~(PROPERTY_IDENTITY | PROPERTY_PERSPECTIVE | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
}
|
|
|
|
/**
|
|
* Multiply this matrix by the supplied <code>right</code> matrix, both of which are assumed to be {@link #isAffine() affine}, and store the result in <code>this</code>.
|
|
* <p>
|
|
* This method assumes that <code>this</code> matrix and the given <code>right</code> matrix both represent an {@link #isAffine() affine} transformation
|
|
* (i.e. their last rows are equal to <code>(0, 0, 0, 1)</code>)
|
|
* and can be used to speed up matrix multiplication if the matrices only represent affine transformations, such as translation, rotation, scaling and shearing (in any combination).
|
|
* <p>
|
|
* This method will not modify either the last row of <code>this</code> or the last row of <code>right</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the <code>right</code> matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* transformation of the right matrix will be applied first!
|
|
*
|
|
* @param right
|
|
* the right operand of the matrix multiplication (the last row is assumed to be <code>(0, 0, 0, 1)</code>)
|
|
* @return this
|
|
*/
|
|
public Matrix4f mulAffine(Matrix4fc right) {
|
|
return mulAffine(right, this);
|
|
}
|
|
|
|
public Matrix4f mulAffine(Matrix4fc right, Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
float m20 = this.m20, m21 = this.m21, m22 = this.m22;
|
|
float rm00 = right.m00(), rm01 = right.m01(), rm02 = right.m02();
|
|
float rm10 = right.m10(), rm11 = right.m11(), rm12 = right.m12();
|
|
float rm20 = right.m20(), rm21 = right.m21(), rm22 = right.m22();
|
|
float rm30 = right.m30(), rm31 = right.m31(), rm32 = right.m32();
|
|
return dest
|
|
._m00(Math.fma(m00, rm00, Math.fma(m10, rm01, m20 * rm02)))
|
|
._m01(Math.fma(m01, rm00, Math.fma(m11, rm01, m21 * rm02)))
|
|
._m02(Math.fma(m02, rm00, Math.fma(m12, rm01, m22 * rm02)))
|
|
._m03(m03)
|
|
._m10(Math.fma(m00, rm10, Math.fma(m10, rm11, m20 * rm12)))
|
|
._m11(Math.fma(m01, rm10, Math.fma(m11, rm11, m21 * rm12)))
|
|
._m12(Math.fma(m02, rm10, Math.fma(m12, rm11, m22 * rm12)))
|
|
._m13(m13)
|
|
._m20(Math.fma(m00, rm20, Math.fma(m10, rm21, m20 * rm22)))
|
|
._m21(Math.fma(m01, rm20, Math.fma(m11, rm21, m21 * rm22)))
|
|
._m22(Math.fma(m02, rm20, Math.fma(m12, rm21, m22 * rm22)))
|
|
._m23(m23)
|
|
._m30(Math.fma(m00, rm30, Math.fma(m10, rm31, Math.fma(m20, rm32, m30))))
|
|
._m31(Math.fma(m01, rm30, Math.fma(m11, rm31, Math.fma(m21, rm32, m31))))
|
|
._m32(Math.fma(m02, rm30, Math.fma(m12, rm31, Math.fma(m22, rm32, m32))))
|
|
._m33(m33)
|
|
._properties(PROPERTY_AFFINE | (this.properties & right.properties() & PROPERTY_ORTHONORMAL));
|
|
}
|
|
|
|
public Matrix4f mulTranslationAffine(Matrix4fc right, Matrix4f dest) {
|
|
return dest
|
|
._m00(right.m00())
|
|
._m01(right.m01())
|
|
._m02(right.m02())
|
|
._m03(m03)
|
|
._m10(right.m10())
|
|
._m11(right.m11())
|
|
._m12(right.m12())
|
|
._m13(m13)
|
|
._m20(right.m20())
|
|
._m21(right.m21())
|
|
._m22(right.m22())
|
|
._m23(m23)
|
|
._m30(right.m30() + m30)
|
|
._m31(right.m31() + m31)
|
|
._m32(right.m32() + m32)
|
|
._m33(m33)
|
|
._properties(PROPERTY_AFFINE | (right.properties() & PROPERTY_ORTHONORMAL));
|
|
}
|
|
|
|
/**
|
|
* Multiply <code>this</code> orthographic projection matrix by the supplied {@link #isAffine() affine} <code>view</code> matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>V</code> the <code>view</code> matrix,
|
|
* then the new matrix will be <code>M * V</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * V * v</code>, the
|
|
* transformation of the <code>view</code> matrix will be applied first!
|
|
*
|
|
* @param view
|
|
* the affine matrix which to multiply <code>this</code> with
|
|
* @return this
|
|
*/
|
|
public Matrix4f mulOrthoAffine(Matrix4fc view) {
|
|
return mulOrthoAffine(view, this);
|
|
}
|
|
|
|
public Matrix4f mulOrthoAffine(Matrix4fc view, Matrix4f dest) {
|
|
float nm00 = m00 * view.m00();
|
|
float nm01 = m11 * view.m01();
|
|
float nm02 = m22 * view.m02();
|
|
float nm10 = m00 * view.m10();
|
|
float nm11 = m11 * view.m11();
|
|
float nm12 = m22 * view.m12();
|
|
float nm20 = m00 * view.m20();
|
|
float nm21 = m11 * view.m21();
|
|
float nm22 = m22 * view.m22();
|
|
float nm30 = m00 * view.m30() + m30;
|
|
float nm31 = m11 * view.m31() + m31;
|
|
float nm32 = m22 * view.m32() + m32;
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(0.0f)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(0.0f)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(0.0f)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE);
|
|
}
|
|
|
|
/**
|
|
* Component-wise add the upper 4x3 submatrices of <code>this</code> and <code>other</code>
|
|
* by first multiplying each component of <code>other</code>'s 4x3 submatrix by <code>otherFactor</code> and
|
|
* adding that result to <code>this</code>.
|
|
* <p>
|
|
* The matrix <code>other</code> will not be changed.
|
|
*
|
|
* @param other
|
|
* the other matrix
|
|
* @param otherFactor
|
|
* the factor to multiply each of the other matrix's 4x3 components
|
|
* @return this
|
|
*/
|
|
public Matrix4f fma4x3(Matrix4fc other, float otherFactor) {
|
|
return fma4x3(other, otherFactor, this);
|
|
}
|
|
|
|
public Matrix4f fma4x3(Matrix4fc other, float otherFactor, Matrix4f dest) {
|
|
dest._m00(Math.fma(other.m00(), otherFactor, m00))
|
|
._m01(Math.fma(other.m01(), otherFactor, m01))
|
|
._m02(Math.fma(other.m02(), otherFactor, m02))
|
|
._m03(m03)
|
|
._m10(Math.fma(other.m10(), otherFactor, m10))
|
|
._m11(Math.fma(other.m11(), otherFactor, m11))
|
|
._m12(Math.fma(other.m12(), otherFactor, m12))
|
|
._m13(m13)
|
|
._m20(Math.fma(other.m20(), otherFactor, m20))
|
|
._m21(Math.fma(other.m21(), otherFactor, m21))
|
|
._m22(Math.fma(other.m22(), otherFactor, m22))
|
|
._m23(m23)
|
|
._m30(Math.fma(other.m30(), otherFactor, m30))
|
|
._m31(Math.fma(other.m31(), otherFactor, m31))
|
|
._m32(Math.fma(other.m32(), otherFactor, m32))
|
|
._m33(m33)
|
|
._properties(0);
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Component-wise add <code>this</code> and <code>other</code>.
|
|
*
|
|
* @param other
|
|
* the other addend
|
|
* @return this
|
|
*/
|
|
public Matrix4f add(Matrix4fc other) {
|
|
return add(other, this);
|
|
}
|
|
|
|
public Matrix4f add(Matrix4fc other, Matrix4f dest) {
|
|
dest._m00(m00 + other.m00())
|
|
._m01(m01 + other.m01())
|
|
._m02(m02 + other.m02())
|
|
._m03(m03 + other.m03())
|
|
._m10(m10 + other.m10())
|
|
._m11(m11 + other.m11())
|
|
._m12(m12 + other.m12())
|
|
._m13(m13 + other.m13())
|
|
._m20(m20 + other.m20())
|
|
._m21(m21 + other.m21())
|
|
._m22(m22 + other.m22())
|
|
._m23(m23 + other.m23())
|
|
._m30(m30 + other.m30())
|
|
._m31(m31 + other.m31())
|
|
._m32(m32 + other.m32())
|
|
._m33(m33 + other.m33())
|
|
._properties(0);
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Component-wise subtract <code>subtrahend</code> from <code>this</code>.
|
|
*
|
|
* @param subtrahend
|
|
* the subtrahend
|
|
* @return this
|
|
*/
|
|
public Matrix4f sub(Matrix4fc subtrahend) {
|
|
return sub(subtrahend, this);
|
|
}
|
|
|
|
public Matrix4f sub(Matrix4fc subtrahend, Matrix4f dest) {
|
|
dest._m00(m00 - subtrahend.m00())
|
|
._m01(m01 - subtrahend.m01())
|
|
._m02(m02 - subtrahend.m02())
|
|
._m03(m03 - subtrahend.m03())
|
|
._m10(m10 - subtrahend.m10())
|
|
._m11(m11 - subtrahend.m11())
|
|
._m12(m12 - subtrahend.m12())
|
|
._m13(m13 - subtrahend.m13())
|
|
._m20(m20 - subtrahend.m20())
|
|
._m21(m21 - subtrahend.m21())
|
|
._m22(m22 - subtrahend.m22())
|
|
._m23(m23 - subtrahend.m23())
|
|
._m30(m30 - subtrahend.m30())
|
|
._m31(m31 - subtrahend.m31())
|
|
._m32(m32 - subtrahend.m32())
|
|
._m33(m33 - subtrahend.m33())
|
|
._properties(0);
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Component-wise multiply <code>this</code> by <code>other</code>.
|
|
*
|
|
* @param other
|
|
* the other matrix
|
|
* @return this
|
|
*/
|
|
public Matrix4f mulComponentWise(Matrix4fc other) {
|
|
return mulComponentWise(other, this);
|
|
}
|
|
|
|
public Matrix4f mulComponentWise(Matrix4fc other, Matrix4f dest) {
|
|
dest._m00(m00 * other.m00())
|
|
._m01(m01 * other.m01())
|
|
._m02(m02 * other.m02())
|
|
._m03(m03 * other.m03())
|
|
._m10(m10 * other.m10())
|
|
._m11(m11 * other.m11())
|
|
._m12(m12 * other.m12())
|
|
._m13(m13 * other.m13())
|
|
._m20(m20 * other.m20())
|
|
._m21(m21 * other.m21())
|
|
._m22(m22 * other.m22())
|
|
._m23(m23 * other.m23())
|
|
._m30(m30 * other.m30())
|
|
._m31(m31 * other.m31())
|
|
._m32(m32 * other.m32())
|
|
._m33(m33 * other.m33())
|
|
._properties(0);
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Component-wise add the upper 4x3 submatrices of <code>this</code> and <code>other</code>.
|
|
*
|
|
* @param other
|
|
* the other addend
|
|
* @return this
|
|
*/
|
|
public Matrix4f add4x3(Matrix4fc other) {
|
|
return add4x3(other, this);
|
|
}
|
|
|
|
public Matrix4f add4x3(Matrix4fc other, Matrix4f dest) {
|
|
dest._m00(m00 + other.m00())
|
|
._m01(m01 + other.m01())
|
|
._m02(m02 + other.m02())
|
|
._m03(m03)
|
|
._m10(m10 + other.m10())
|
|
._m11(m11 + other.m11())
|
|
._m12(m12 + other.m12())
|
|
._m13(m13)
|
|
._m20(m20 + other.m20())
|
|
._m21(m21 + other.m21())
|
|
._m22(m22 + other.m22())
|
|
._m23(m23)
|
|
._m30(m30 + other.m30())
|
|
._m31(m31 + other.m31())
|
|
._m32(m32 + other.m32())
|
|
._m33(m33)
|
|
._properties(0);
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Component-wise subtract the upper 4x3 submatrices of <code>subtrahend</code> from <code>this</code>.
|
|
*
|
|
* @param subtrahend
|
|
* the subtrahend
|
|
* @return this
|
|
*/
|
|
public Matrix4f sub4x3(Matrix4f subtrahend) {
|
|
return sub4x3(subtrahend, this);
|
|
}
|
|
|
|
public Matrix4f sub4x3(Matrix4fc subtrahend, Matrix4f dest) {
|
|
dest._m00(m00 - subtrahend.m00())
|
|
._m01(m01 - subtrahend.m01())
|
|
._m02(m02 - subtrahend.m02())
|
|
._m03(m03)
|
|
._m10(m10 - subtrahend.m10())
|
|
._m11(m11 - subtrahend.m11())
|
|
._m12(m12 - subtrahend.m12())
|
|
._m13(m13)
|
|
._m20(m20 - subtrahend.m20())
|
|
._m21(m21 - subtrahend.m21())
|
|
._m22(m22 - subtrahend.m22())
|
|
._m23(m23)
|
|
._m30(m30 - subtrahend.m30())
|
|
._m31(m31 - subtrahend.m31())
|
|
._m32(m32 - subtrahend.m32())
|
|
._m33(m33)
|
|
._properties(0);
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Component-wise multiply the upper 4x3 submatrices of <code>this</code> by <code>other</code>.
|
|
*
|
|
* @param other
|
|
* the other matrix
|
|
* @return this
|
|
*/
|
|
public Matrix4f mul4x3ComponentWise(Matrix4fc other) {
|
|
return mul4x3ComponentWise(other, this);
|
|
}
|
|
|
|
public Matrix4f mul4x3ComponentWise(Matrix4fc other, Matrix4f dest) {
|
|
dest._m00(m00 * other.m00())
|
|
._m01(m01 * other.m01())
|
|
._m02(m02 * other.m02())
|
|
._m03(m03)
|
|
._m10(m10 * other.m10())
|
|
._m11(m11 * other.m11())
|
|
._m12(m12 * other.m12())
|
|
._m13(m13)
|
|
._m20(m20 * other.m20())
|
|
._m21(m21 * other.m21())
|
|
._m22(m22 * other.m22())
|
|
._m23(m23)
|
|
._m30(m30 * other.m30())
|
|
._m31(m31 * other.m31())
|
|
._m32(m32 * other.m32())
|
|
._m33(m33)
|
|
._properties(0);
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Set the values within this matrix to the supplied float values. The matrix will look like this:<br><br>
|
|
*
|
|
* m00, m10, m20, m30<br>
|
|
* m01, m11, m21, m31<br>
|
|
* m02, m12, m22, m32<br>
|
|
* m03, m13, m23, m33
|
|
*
|
|
* @param m00
|
|
* the new value of m00
|
|
* @param m01
|
|
* the new value of m01
|
|
* @param m02
|
|
* the new value of m02
|
|
* @param m03
|
|
* the new value of m03
|
|
* @param m10
|
|
* the new value of m10
|
|
* @param m11
|
|
* the new value of m11
|
|
* @param m12
|
|
* the new value of m12
|
|
* @param m13
|
|
* the new value of m13
|
|
* @param m20
|
|
* the new value of m20
|
|
* @param m21
|
|
* the new value of m21
|
|
* @param m22
|
|
* the new value of m22
|
|
* @param m23
|
|
* the new value of m23
|
|
* @param m30
|
|
* the new value of m30
|
|
* @param m31
|
|
* the new value of m31
|
|
* @param m32
|
|
* the new value of m32
|
|
* @param m33
|
|
* the new value of m33
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(float m00, float m01, float m02, float m03,
|
|
float m10, float m11, float m12, float m13,
|
|
float m20, float m21, float m22, float m23,
|
|
float m30, float m31, float m32, float m33) {
|
|
return this
|
|
._m00(m00)
|
|
._m10(m10)
|
|
._m20(m20)
|
|
._m30(m30)
|
|
._m01(m01)
|
|
._m11(m11)
|
|
._m21(m21)
|
|
._m31(m31)
|
|
._m02(m02)
|
|
._m12(m12)
|
|
._m22(m22)
|
|
._m32(m32)
|
|
._m03(m03)
|
|
._m13(m13)
|
|
._m23(m23)
|
|
._m33(m33)
|
|
.determineProperties();
|
|
}
|
|
|
|
/**
|
|
* Set the values in the matrix using a float array that contains the matrix elements in column-major order.
|
|
* <p>
|
|
* The results will look like this:<br><br>
|
|
*
|
|
* 0, 4, 8, 12<br>
|
|
* 1, 5, 9, 13<br>
|
|
* 2, 6, 10, 14<br>
|
|
* 3, 7, 11, 15<br>
|
|
*
|
|
* @see #set(float[])
|
|
*
|
|
* @param m
|
|
* the array to read the matrix values from
|
|
* @param off
|
|
* the offset into the array
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(float m[], int off) {
|
|
MemUtil.INSTANCE.copy(m, off, this);
|
|
return determineProperties();
|
|
}
|
|
|
|
/**
|
|
* Set the values in the matrix using a float array that contains the matrix elements in column-major order.
|
|
* <p>
|
|
* The results will look like this:<br><br>
|
|
*
|
|
* 0, 4, 8, 12<br>
|
|
* 1, 5, 9, 13<br>
|
|
* 2, 6, 10, 14<br>
|
|
* 3, 7, 11, 15<br>
|
|
*
|
|
* @see #set(float[], int)
|
|
*
|
|
* @param m
|
|
* the array to read the matrix values from
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(float m[]) {
|
|
return set(m, 0);
|
|
}
|
|
|
|
/**
|
|
* Set the values in the matrix using a float array that contains the matrix elements in row-major order.
|
|
* <p>
|
|
* The results will look like this:<br><br>
|
|
*
|
|
* 0, 1, 2, 3<br>
|
|
* 4, 5, 6, 7<br>
|
|
* 8, 9, 10, 11<br>
|
|
* 12, 13, 14, 15<br>
|
|
*
|
|
* @see #setTransposed(float[])
|
|
*
|
|
* @param m
|
|
* the array to read the matrix values from
|
|
* @param off
|
|
* the offset into the array
|
|
* @return this
|
|
*/
|
|
public Matrix4f setTransposed(float m[], int off) {
|
|
MemUtil.INSTANCE.copyTransposed(m, off, this);
|
|
return determineProperties();
|
|
}
|
|
|
|
/**
|
|
* Set the values in the matrix using a float array that contains the matrix elements in row-major order.
|
|
* <p>
|
|
* The results will look like this:<br><br>
|
|
*
|
|
* 0, 1, 2, 3<br>
|
|
* 4, 5, 6, 7<br>
|
|
* 8, 9, 10, 11<br>
|
|
* 12, 13, 14, 15<br>
|
|
*
|
|
* @see #setTransposed(float[], int)
|
|
*
|
|
* @param m
|
|
* the array to read the matrix values from
|
|
* @return this
|
|
*/
|
|
public Matrix4f setTransposed(float m[]) {
|
|
return setTransposed(m, 0);
|
|
}
|
|
|
|
/**
|
|
* Set the values of this matrix by reading 16 float values from the given {@link FloatBuffer} in column-major order,
|
|
* starting at its current position.
|
|
* <p>
|
|
* The FloatBuffer is expected to contain the values in column-major order.
|
|
* <p>
|
|
* The position of the FloatBuffer will not be changed by this method.
|
|
*
|
|
* @param buffer
|
|
* the FloatBuffer to read the matrix values from in column-major order
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(FloatBuffer buffer) {
|
|
MemUtil.INSTANCE.get(this, buffer.position(), buffer);
|
|
return determineProperties();
|
|
}
|
|
|
|
/**
|
|
* Set the values of this matrix by reading 16 float values from the given {@link ByteBuffer} in column-major order,
|
|
* starting at its current position.
|
|
* <p>
|
|
* The ByteBuffer is expected to contain the values in column-major order.
|
|
* <p>
|
|
* The position of the ByteBuffer will not be changed by this method.
|
|
*
|
|
* @param buffer
|
|
* the ByteBuffer to read the matrix values from in column-major order
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(ByteBuffer buffer) {
|
|
MemUtil.INSTANCE.get(this, buffer.position(), buffer);
|
|
return determineProperties();
|
|
}
|
|
|
|
/**
|
|
* Set the values of this matrix by reading 16 float values from the given {@link FloatBuffer} in column-major order,
|
|
* starting at the specified absolute buffer position/index.
|
|
* <p>
|
|
* The FloatBuffer is expected to contain the values in column-major order.
|
|
* <p>
|
|
* The position of the FloatBuffer will not be changed by this method.
|
|
*
|
|
* @param index
|
|
* the absolute position into the FloatBuffer
|
|
* @param buffer
|
|
* the FloatBuffer to read the matrix values from in column-major order
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(int index, FloatBuffer buffer) {
|
|
MemUtil.INSTANCE.get(this, index, buffer);
|
|
return determineProperties();
|
|
}
|
|
|
|
/**
|
|
* Set the values of this matrix by reading 16 float values from the given {@link ByteBuffer} in column-major order,
|
|
* starting at the specified absolute buffer position/index.
|
|
* <p>
|
|
* The ByteBuffer is expected to contain the values in column-major order.
|
|
* <p>
|
|
* The position of the ByteBuffer will not be changed by this method.
|
|
*
|
|
* @param index
|
|
* the absolute position into the ByteBuffer
|
|
* @param buffer
|
|
* the ByteBuffer to read the matrix values from in column-major order
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(int index, ByteBuffer buffer) {
|
|
MemUtil.INSTANCE.get(this, index, buffer);
|
|
return determineProperties();
|
|
}
|
|
|
|
/**
|
|
* Set the values of this matrix by reading 16 float values from the given {@link FloatBuffer} in row-major order,
|
|
* starting at its current position.
|
|
* <p>
|
|
* The FloatBuffer is expected to contain the values in row-major order.
|
|
* <p>
|
|
* The position of the FloatBuffer will not be changed by this method.
|
|
*
|
|
* @param buffer
|
|
* the FloatBuffer to read the matrix values from in row-major order
|
|
* @return this
|
|
*/
|
|
public Matrix4f setTransposed(FloatBuffer buffer) {
|
|
MemUtil.INSTANCE.getTransposed(this, buffer.position(), buffer);
|
|
return determineProperties();
|
|
}
|
|
|
|
/**
|
|
* Set the values of this matrix by reading 16 float values from the given {@link ByteBuffer} in row-major order,
|
|
* starting at its current position.
|
|
* <p>
|
|
* The ByteBuffer is expected to contain the values in row-major order.
|
|
* <p>
|
|
* The position of the ByteBuffer will not be changed by this method.
|
|
*
|
|
* @param buffer
|
|
* the ByteBuffer to read the matrix values from in row-major order
|
|
* @return this
|
|
*/
|
|
public Matrix4f setTransposed(ByteBuffer buffer) {
|
|
MemUtil.INSTANCE.getTransposed(this, buffer.position(), buffer);
|
|
return determineProperties();
|
|
}
|
|
|
|
/**
|
|
* Set the values of this matrix by reading 16 float values from off-heap memory in column-major order,
|
|
* starting at the given address.
|
|
* <p>
|
|
* This method will throw an {@link UnsupportedOperationException} when JOML is used with `-Djoml.nounsafe`.
|
|
* <p>
|
|
* <em>This method is unsafe as it can result in a crash of the JVM process when the specified address range does not belong to this process.</em>
|
|
*
|
|
* @param address
|
|
* the off-heap memory address to read the matrix values from in column-major order
|
|
* @return this
|
|
*/
|
|
public Matrix4f setFromAddress(long address) {
|
|
if (Options.NO_UNSAFE)
|
|
throw new UnsupportedOperationException("Not supported when using joml.nounsafe");
|
|
MemUtil.MemUtilUnsafe.get(this, address);
|
|
return determineProperties();
|
|
}
|
|
|
|
/**
|
|
* Set the values of this matrix by reading 16 float values from off-heap memory in row-major order,
|
|
* starting at the given address.
|
|
* <p>
|
|
* This method will throw an {@link UnsupportedOperationException} when JOML is used with `-Djoml.nounsafe`.
|
|
* <p>
|
|
* <em>This method is unsafe as it can result in a crash of the JVM process when the specified address range does not belong to this process.</em>
|
|
*
|
|
* @param address
|
|
* the off-heap memory address to read the matrix values from in row-major order
|
|
* @return this
|
|
*/
|
|
public Matrix4f setTransposedFromAddress(long address) {
|
|
if (Options.NO_UNSAFE)
|
|
throw new UnsupportedOperationException("Not supported when using joml.nounsafe");
|
|
MemUtil.MemUtilUnsafe.getTransposed(this, address);
|
|
return determineProperties();
|
|
}
|
|
|
|
/**
|
|
* Set the four columns of this matrix to the supplied vectors, respectively.
|
|
*
|
|
* @param col0
|
|
* the first column
|
|
* @param col1
|
|
* the second column
|
|
* @param col2
|
|
* the third column
|
|
* @param col3
|
|
* the fourth column
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(Vector4fc col0, Vector4fc col1, Vector4fc col2, Vector4fc col3) {
|
|
return
|
|
_m00(col0.x()).
|
|
_m01(col0.y()).
|
|
_m02(col0.z()).
|
|
_m03(col0.w()).
|
|
_m10(col1.x()).
|
|
_m11(col1.y()).
|
|
_m12(col1.z()).
|
|
_m13(col1.w()).
|
|
_m20(col2.x()).
|
|
_m21(col2.y()).
|
|
_m22(col2.z()).
|
|
_m23(col2.w()).
|
|
_m30(col3.x()).
|
|
_m31(col3.y()).
|
|
_m32(col3.z()).
|
|
_m33(col3.w()).
|
|
determineProperties();
|
|
}
|
|
|
|
public float determinant() {
|
|
if ((properties & PROPERTY_AFFINE) != 0)
|
|
return determinantAffine();
|
|
return (m00 * m11 - m01 * m10) * (m22 * m33 - m23 * m32)
|
|
+ (m02 * m10 - m00 * m12) * (m21 * m33 - m23 * m31)
|
|
+ (m00 * m13 - m03 * m10) * (m21 * m32 - m22 * m31)
|
|
+ (m01 * m12 - m02 * m11) * (m20 * m33 - m23 * m30)
|
|
+ (m03 * m11 - m01 * m13) * (m20 * m32 - m22 * m30)
|
|
+ (m02 * m13 - m03 * m12) * (m20 * m31 - m21 * m30);
|
|
}
|
|
|
|
public float determinant3x3() {
|
|
return (m00 * m11 - m01 * m10) * m22
|
|
+ (m02 * m10 - m00 * m12) * m21
|
|
+ (m01 * m12 - m02 * m11) * m20;
|
|
}
|
|
|
|
public float determinantAffine() {
|
|
return (m00 * m11 - m01 * m10) * m22
|
|
+ (m02 * m10 - m00 * m12) * m21
|
|
+ (m01 * m12 - m02 * m11) * m20;
|
|
}
|
|
|
|
public Matrix4f invert(Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0) {
|
|
return dest.identity();
|
|
} else if ((properties & PROPERTY_TRANSLATION) != 0)
|
|
return invertTranslation(dest);
|
|
else if ((properties & PROPERTY_ORTHONORMAL) != 0)
|
|
return invertOrthonormal(dest);
|
|
else if ((properties & PROPERTY_AFFINE) != 0)
|
|
return invertAffine(dest);
|
|
else if ((properties & PROPERTY_PERSPECTIVE) != 0)
|
|
return invertPerspective(dest);
|
|
return invertGeneric(dest);
|
|
}
|
|
private Matrix4f invertTranslation(Matrix4f dest) {
|
|
if (dest != this)
|
|
dest.set(this);
|
|
return dest._m30(-m30)._m31(-m31)._m32(-m32)._properties(PROPERTY_AFFINE | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL);
|
|
}
|
|
private Matrix4f invertOrthonormal(Matrix4f dest) {
|
|
float nm30 = -(m00 * m30 + m01 * m31 + m02 * m32);
|
|
float nm31 = -(m10 * m30 + m11 * m31 + m12 * m32);
|
|
float nm32 = -(m20 * m30 + m21 * m31 + m22 * m32);
|
|
float m01 = this.m01;
|
|
float m02 = this.m02;
|
|
float m12 = this.m12;
|
|
return dest
|
|
._m00(m00)
|
|
._m01(m10)
|
|
._m02(m20)
|
|
._m03(0.0f)
|
|
._m10(m01)
|
|
._m11(m11)
|
|
._m12(m21)
|
|
._m13(0.0f)
|
|
._m20(m02)
|
|
._m21(m12)
|
|
._m22(m22)
|
|
._m23(0.0f)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
}
|
|
private Matrix4f invertGeneric(Matrix4f dest) {
|
|
if (this != dest)
|
|
return invertGenericNonThis(dest);
|
|
return invertGenericThis(dest);
|
|
}
|
|
private Matrix4f invertGenericNonThis(Matrix4f dest) {
|
|
float a = m00 * m11 - m01 * m10;
|
|
float b = m00 * m12 - m02 * m10;
|
|
float c = m00 * m13 - m03 * m10;
|
|
float d = m01 * m12 - m02 * m11;
|
|
float e = m01 * m13 - m03 * m11;
|
|
float f = m02 * m13 - m03 * m12;
|
|
float g = m20 * m31 - m21 * m30;
|
|
float h = m20 * m32 - m22 * m30;
|
|
float i = m20 * m33 - m23 * m30;
|
|
float j = m21 * m32 - m22 * m31;
|
|
float k = m21 * m33 - m23 * m31;
|
|
float l = m22 * m33 - m23 * m32;
|
|
float det = a * l - b * k + c * j + d * i - e * h + f * g;
|
|
det = 1.0f / det;
|
|
return dest
|
|
._m00(Math.fma( m11, l, Math.fma(-m12, k, m13 * j)) * det)
|
|
._m01(Math.fma(-m01, l, Math.fma( m02, k, -m03 * j)) * det)
|
|
._m02(Math.fma( m31, f, Math.fma(-m32, e, m33 * d)) * det)
|
|
._m03(Math.fma(-m21, f, Math.fma( m22, e, -m23 * d)) * det)
|
|
._m10(Math.fma(-m10, l, Math.fma( m12, i, -m13 * h)) * det)
|
|
._m11(Math.fma( m00, l, Math.fma(-m02, i, m03 * h)) * det)
|
|
._m12(Math.fma(-m30, f, Math.fma( m32, c, -m33 * b)) * det)
|
|
._m13(Math.fma( m20, f, Math.fma(-m22, c, m23 * b)) * det)
|
|
._m20(Math.fma( m10, k, Math.fma(-m11, i, m13 * g)) * det)
|
|
._m21(Math.fma(-m00, k, Math.fma( m01, i, -m03 * g)) * det)
|
|
._m22(Math.fma( m30, e, Math.fma(-m31, c, m33 * a)) * det)
|
|
._m23(Math.fma(-m20, e, Math.fma( m21, c, -m23 * a)) * det)
|
|
._m30(Math.fma(-m10, j, Math.fma( m11, h, -m12 * g)) * det)
|
|
._m31(Math.fma( m00, j, Math.fma(-m01, h, m02 * g)) * det)
|
|
._m32(Math.fma(-m30, d, Math.fma( m31, b, -m32 * a)) * det)
|
|
._m33(Math.fma( m20, d, Math.fma(-m21, b, m22 * a)) * det)
|
|
._properties(0);
|
|
}
|
|
private Matrix4f invertGenericThis(Matrix4f dest) {
|
|
float a = m00 * m11 - m01 * m10;
|
|
float b = m00 * m12 - m02 * m10;
|
|
float c = m00 * m13 - m03 * m10;
|
|
float d = m01 * m12 - m02 * m11;
|
|
float e = m01 * m13 - m03 * m11;
|
|
float f = m02 * m13 - m03 * m12;
|
|
float g = m20 * m31 - m21 * m30;
|
|
float h = m20 * m32 - m22 * m30;
|
|
float i = m20 * m33 - m23 * m30;
|
|
float j = m21 * m32 - m22 * m31;
|
|
float k = m21 * m33 - m23 * m31;
|
|
float l = m22 * m33 - m23 * m32;
|
|
float det = a * l - b * k + c * j + d * i - e * h + f * g;
|
|
det = 1.0f / det;
|
|
float nm00 = Math.fma( m11, l, Math.fma(-m12, k, m13 * j)) * det;
|
|
float nm01 = Math.fma(-m01, l, Math.fma( m02, k, -m03 * j)) * det;
|
|
float nm02 = Math.fma( m31, f, Math.fma(-m32, e, m33 * d)) * det;
|
|
float nm03 = Math.fma(-m21, f, Math.fma( m22, e, -m23 * d)) * det;
|
|
float nm10 = Math.fma(-m10, l, Math.fma( m12, i, -m13 * h)) * det;
|
|
float nm11 = Math.fma( m00, l, Math.fma(-m02, i, m03 * h)) * det;
|
|
float nm12 = Math.fma(-m30, f, Math.fma( m32, c, -m33 * b)) * det;
|
|
float nm13 = Math.fma( m20, f, Math.fma(-m22, c, m23 * b)) * det;
|
|
float nm20 = Math.fma( m10, k, Math.fma(-m11, i, m13 * g)) * det;
|
|
float nm21 = Math.fma(-m00, k, Math.fma( m01, i, -m03 * g)) * det;
|
|
float nm22 = Math.fma( m30, e, Math.fma(-m31, c, m33 * a)) * det;
|
|
float nm23 = Math.fma(-m20, e, Math.fma( m21, c, -m23 * a)) * det;
|
|
float nm30 = Math.fma(-m10, j, Math.fma( m11, h, -m12 * g)) * det;
|
|
float nm31 = Math.fma( m00, j, Math.fma(-m01, h, m02 * g)) * det;
|
|
float nm32 = Math.fma(-m30, d, Math.fma( m31, b, -m32 * a)) * det;
|
|
float nm33 = Math.fma( m20, d, Math.fma(-m21, b, m22 * a)) * det;
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(nm33)
|
|
._properties(0);
|
|
}
|
|
|
|
/**
|
|
* Invert this matrix.
|
|
* <p>
|
|
* If <code>this</code> matrix represents an {@link #isAffine() affine} transformation, such as translation, rotation, scaling and shearing,
|
|
* and thus its last row is equal to <code>(0, 0, 0, 1)</code>, then {@link #invertAffine()} can be used instead of this method.
|
|
*
|
|
* @see #invertAffine()
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f invert() {
|
|
return invert(this);
|
|
}
|
|
|
|
/**
|
|
* If <code>this</code> is a perspective projection matrix obtained via one of the {@link #perspective(float, float, float, float) perspective()} methods
|
|
* or via {@link #setPerspective(float, float, float, float) setPerspective()}, that is, if <code>this</code> is a symmetrical perspective frustum transformation,
|
|
* then this method builds the inverse of <code>this</code> and stores it into the given <code>dest</code>.
|
|
* <p>
|
|
* This method can be used to quickly obtain the inverse of a perspective projection matrix when being obtained via {@link #perspective(float, float, float, float) perspective()}.
|
|
*
|
|
* @see #perspective(float, float, float, float)
|
|
*
|
|
* @param dest
|
|
* will hold the inverse of <code>this</code>
|
|
* @return dest
|
|
*/
|
|
public Matrix4f invertPerspective(Matrix4f dest) {
|
|
float a = 1.0f / (m00 * m11);
|
|
float l = -1.0f / (m23 * m32);
|
|
return dest
|
|
.set(m11 * a, 0, 0, 0,
|
|
0, m00 * a, 0, 0,
|
|
0, 0, 0, -m23 * l,
|
|
0, 0, -m32 * l, m22 * l)
|
|
._properties(0);
|
|
}
|
|
|
|
/**
|
|
* If <code>this</code> is a perspective projection matrix obtained via one of the {@link #perspective(float, float, float, float) perspective()} methods
|
|
* or via {@link #setPerspective(float, float, float, float) setPerspective()}, that is, if <code>this</code> is a symmetrical perspective frustum transformation,
|
|
* then this method builds the inverse of <code>this</code>.
|
|
* <p>
|
|
* This method can be used to quickly obtain the inverse of a perspective projection matrix when being obtained via {@link #perspective(float, float, float, float) perspective()}.
|
|
*
|
|
* @see #perspective(float, float, float, float)
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f invertPerspective() {
|
|
return invertPerspective(this);
|
|
}
|
|
|
|
/**
|
|
* If <code>this</code> is an arbitrary perspective projection matrix obtained via one of the {@link #frustum(float, float, float, float, float, float) frustum()} methods
|
|
* or via {@link #setFrustum(float, float, float, float, float, float) setFrustum()},
|
|
* then this method builds the inverse of <code>this</code> and stores it into the given <code>dest</code>.
|
|
* <p>
|
|
* This method can be used to quickly obtain the inverse of a perspective projection matrix.
|
|
* <p>
|
|
* If this matrix represents a symmetric perspective frustum transformation, as obtained via {@link #perspective(float, float, float, float) perspective()}, then
|
|
* {@link #invertPerspective(Matrix4f)} should be used instead.
|
|
*
|
|
* @see #frustum(float, float, float, float, float, float)
|
|
* @see #invertPerspective(Matrix4f)
|
|
*
|
|
* @param dest
|
|
* will hold the inverse of <code>this</code>
|
|
* @return dest
|
|
*/
|
|
public Matrix4f invertFrustum(Matrix4f dest) {
|
|
float invM00 = 1.0f / m00;
|
|
float invM11 = 1.0f / m11;
|
|
float invM23 = 1.0f / m23;
|
|
float invM32 = 1.0f / m32;
|
|
return dest
|
|
.set(invM00, 0, 0, 0,
|
|
0, invM11, 0, 0,
|
|
0, 0, 0, invM32,
|
|
-m20 * invM00 * invM23, -m21 * invM11 * invM23, invM23, -m22 * invM23 * invM32);
|
|
}
|
|
|
|
/**
|
|
* If <code>this</code> is an arbitrary perspective projection matrix obtained via one of the {@link #frustum(float, float, float, float, float, float) frustum()} methods
|
|
* or via {@link #setFrustum(float, float, float, float, float, float) setFrustum()},
|
|
* then this method builds the inverse of <code>this</code>.
|
|
* <p>
|
|
* This method can be used to quickly obtain the inverse of a perspective projection matrix.
|
|
* <p>
|
|
* If this matrix represents a symmetric perspective frustum transformation, as obtained via {@link #perspective(float, float, float, float) perspective()}, then
|
|
* {@link #invertPerspective()} should be used instead.
|
|
*
|
|
* @see #frustum(float, float, float, float, float, float)
|
|
* @see #invertPerspective()
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f invertFrustum() {
|
|
return invertFrustum(this);
|
|
}
|
|
|
|
public Matrix4f invertOrtho(Matrix4f dest) {
|
|
float invM00 = 1.0f / m00;
|
|
float invM11 = 1.0f / m11;
|
|
float invM22 = 1.0f / m22;
|
|
return dest
|
|
.set(invM00, 0, 0, 0,
|
|
0, invM11, 0, 0,
|
|
0, 0, invM22, 0,
|
|
-m30 * invM00, -m31 * invM11, -m32 * invM22, 1)
|
|
._properties(PROPERTY_AFFINE | (this.properties & PROPERTY_ORTHONORMAL));
|
|
}
|
|
|
|
/**
|
|
* Invert <code>this</code> orthographic projection matrix.
|
|
* <p>
|
|
* This method can be used to quickly obtain the inverse of an orthographic projection matrix.
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f invertOrtho() {
|
|
return invertOrtho(this);
|
|
}
|
|
|
|
/**
|
|
* If <code>this</code> is a perspective projection matrix obtained via one of the {@link #perspective(float, float, float, float) perspective()} methods
|
|
* or via {@link #setPerspective(float, float, float, float) setPerspective()}, that is, if <code>this</code> is a symmetrical perspective frustum transformation
|
|
* and the given <code>view</code> matrix is {@link #isAffine() affine} and has unit scaling (for example by being obtained via {@link #lookAt(float, float, float, float, float, float, float, float, float) lookAt()}),
|
|
* then this method builds the inverse of <code>this * view</code> and stores it into the given <code>dest</code>.
|
|
* <p>
|
|
* This method can be used to quickly obtain the inverse of the combination of the view and projection matrices, when both were obtained
|
|
* via the common methods {@link #perspective(float, float, float, float) perspective()} and {@link #lookAt(float, float, float, float, float, float, float, float, float) lookAt()} or
|
|
* other methods, that build affine matrices, such as {@link #translate(float, float, float) translate} and {@link #rotate(float, float, float, float)}, except for {@link #scale(float, float, float) scale()}.
|
|
* <p>
|
|
* For the special cases of the matrices <code>this</code> and <code>view</code> mentioned above, this method is equivalent to the following code:
|
|
* <pre>
|
|
* dest.set(this).mul(view).invert();
|
|
* </pre>
|
|
*
|
|
* @param view
|
|
* the view transformation (must be {@link #isAffine() affine} and have unit scaling)
|
|
* @param dest
|
|
* will hold the inverse of <code>this * view</code>
|
|
* @return dest
|
|
*/
|
|
public Matrix4f invertPerspectiveView(Matrix4fc view, Matrix4f dest) {
|
|
float a = 1.0f / (m00 * m11);
|
|
float l = -1.0f / (m23 * m32);
|
|
float pm00 = m11 * a;
|
|
float pm11 = m00 * a;
|
|
float pm23 = -m23 * l;
|
|
float pm32 = -m32 * l;
|
|
float pm33 = m22 * l;
|
|
float vm30 = -view.m00() * view.m30() - view.m01() * view.m31() - view.m02() * view.m32();
|
|
float vm31 = -view.m10() * view.m30() - view.m11() * view.m31() - view.m12() * view.m32();
|
|
float vm32 = -view.m20() * view.m30() - view.m21() * view.m31() - view.m22() * view.m32();
|
|
float nm10 = view.m01() * pm11;
|
|
float nm30 = view.m02() * pm32 + vm30 * pm33;
|
|
float nm31 = view.m12() * pm32 + vm31 * pm33;
|
|
float nm32 = view.m22() * pm32 + vm32 * pm33;
|
|
return dest
|
|
._m00(view.m00() * pm00)
|
|
._m01(view.m10() * pm00)
|
|
._m02(view.m20() * pm00)
|
|
._m03(0.0f)
|
|
._m10(nm10)
|
|
._m11(view.m11() * pm11)
|
|
._m12(view.m21() * pm11)
|
|
._m13(0.0f)
|
|
._m20(vm30 * pm23)
|
|
._m21(vm31 * pm23)
|
|
._m22(vm32 * pm23)
|
|
._m23(pm23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(pm33)
|
|
._properties(0);
|
|
}
|
|
|
|
/**
|
|
* If <code>this</code> is a perspective projection matrix obtained via one of the {@link #perspective(float, float, float, float) perspective()} methods
|
|
* or via {@link #setPerspective(float, float, float, float) setPerspective()}, that is, if <code>this</code> is a symmetrical perspective frustum transformation
|
|
* and the given <code>view</code> matrix has unit scaling,
|
|
* then this method builds the inverse of <code>this * view</code> and stores it into the given <code>dest</code>.
|
|
* <p>
|
|
* This method can be used to quickly obtain the inverse of the combination of the view and projection matrices, when both were obtained
|
|
* via the common methods {@link #perspective(float, float, float, float) perspective()} and {@link #lookAt(float, float, float, float, float, float, float, float, float) lookAt()} or
|
|
* other methods, that build affine matrices, such as {@link #translate(float, float, float) translate} and {@link #rotate(float, float, float, float)}, except for {@link #scale(float, float, float) scale()}.
|
|
* <p>
|
|
* For the special cases of the matrices <code>this</code> and <code>view</code> mentioned above, this method is equivalent to the following code:
|
|
* <pre>
|
|
* dest.set(this).mul(view).invert();
|
|
* </pre>
|
|
*
|
|
* @param view
|
|
* the view transformation (must have unit scaling)
|
|
* @param dest
|
|
* will hold the inverse of <code>this * view</code>
|
|
* @return dest
|
|
*/
|
|
public Matrix4f invertPerspectiveView(Matrix4x3fc view, Matrix4f dest) {
|
|
float a = 1.0f / (m00 * m11);
|
|
float l = -1.0f / (m23 * m32);
|
|
float pm00 = m11 * a;
|
|
float pm11 = m00 * a;
|
|
float pm23 = -m23 * l;
|
|
float pm32 = -m32 * l;
|
|
float pm33 = m22 * l;
|
|
float vm30 = -view.m00() * view.m30() - view.m01() * view.m31() - view.m02() * view.m32();
|
|
float vm31 = -view.m10() * view.m30() - view.m11() * view.m31() - view.m12() * view.m32();
|
|
float vm32 = -view.m20() * view.m30() - view.m21() * view.m31() - view.m22() * view.m32();
|
|
return dest
|
|
._m00(view.m00() * pm00)
|
|
._m01(view.m10() * pm00)
|
|
._m02(view.m20() * pm00)
|
|
._m03(0.0f)
|
|
._m10(view.m01() * pm11)
|
|
._m11(view.m11() * pm11)
|
|
._m12(view.m21() * pm11)
|
|
._m13(0.0f)
|
|
._m20(vm30 * pm23)
|
|
._m21(vm31 * pm23)
|
|
._m22(vm32 * pm23)
|
|
._m23(pm23)
|
|
._m30(view.m02() * pm32 + vm30 * pm33)
|
|
._m31(view.m12() * pm32 + vm31 * pm33)
|
|
._m32(view.m22() * pm32 + vm32 * pm33)
|
|
._m33(pm33)
|
|
._properties(0);
|
|
}
|
|
|
|
public Matrix4f invertAffine(Matrix4f dest) {
|
|
float m11m00 = m00 * m11, m10m01 = m01 * m10, m10m02 = m02 * m10;
|
|
float m12m00 = m00 * m12, m12m01 = m01 * m12, m11m02 = m02 * m11;
|
|
float det = (m11m00 - m10m01) * m22 + (m10m02 - m12m00) * m21 + (m12m01 - m11m02) * m20;
|
|
float s = 1.0f / det;
|
|
float m10m22 = m10 * m22, m10m21 = m10 * m21, m11m22 = m11 * m22;
|
|
float m11m20 = m11 * m20, m12m21 = m12 * m21, m12m20 = m12 * m20;
|
|
float m20m02 = m20 * m02, m20m01 = m20 * m01, m21m02 = m21 * m02;
|
|
float m21m00 = m21 * m00, m22m01 = m22 * m01, m22m00 = m22 * m00;
|
|
float nm31 = (m20m02 * m31 - m20m01 * m32 + m21m00 * m32 - m21m02 * m30 + m22m01 * m30 - m22m00 * m31) * s;
|
|
float nm32 = (m11m02 * m30 - m12m01 * m30 + m12m00 * m31 - m10m02 * m31 + m10m01 * m32 - m11m00 * m32) * s;
|
|
return dest
|
|
._m00((m11m22 - m12m21) * s)
|
|
._m01((m21m02 - m22m01) * s)
|
|
._m02((m12m01 - m11m02) * s)
|
|
._m03(0.0f)
|
|
._m10((m12m20 - m10m22) * s)
|
|
._m11((m22m00 - m20m02) * s)
|
|
._m12((m10m02 - m12m00) * s)
|
|
._m13(0.0f)
|
|
._m20((m10m21 - m11m20) * s)
|
|
._m21((m20m01 - m21m00) * s)
|
|
._m22((m11m00 - m10m01) * s)
|
|
._m23(0.0f)
|
|
._m30((m10m22 * m31 - m10m21 * m32 + m11m20 * m32 - m11m22 * m30 + m12m21 * m30 - m12m20 * m31) * s)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE);
|
|
}
|
|
|
|
/**
|
|
* Invert this matrix by assuming that it is an {@link #isAffine() affine} transformation (i.e. its last row is equal to <code>(0, 0, 0, 1)</code>).
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f invertAffine() {
|
|
return invertAffine(this);
|
|
}
|
|
|
|
public Matrix4f transpose(Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.identity();
|
|
else if (this != dest)
|
|
return transposeNonThisGeneric(dest);
|
|
return transposeThisGeneric(dest);
|
|
}
|
|
private Matrix4f transposeNonThisGeneric(Matrix4f dest) {
|
|
return dest
|
|
._m00(m00)
|
|
._m01(m10)
|
|
._m02(m20)
|
|
._m03(m30)
|
|
._m10(m01)
|
|
._m11(m11)
|
|
._m12(m21)
|
|
._m13(m31)
|
|
._m20(m02)
|
|
._m21(m12)
|
|
._m22(m22)
|
|
._m23(m32)
|
|
._m30(m03)
|
|
._m31(m13)
|
|
._m32(m23)
|
|
._m33(m33)
|
|
._properties(0);
|
|
}
|
|
private Matrix4f transposeThisGeneric(Matrix4f dest) {
|
|
float nm10 = m01;
|
|
float nm20 = m02;
|
|
float nm21 = m12;
|
|
float nm30 = m03;
|
|
float nm31 = m13;
|
|
float nm32 = m23;
|
|
return dest
|
|
._m01(m10)
|
|
._m02(m20)
|
|
._m03(m30)
|
|
._m10(nm10)
|
|
._m12(m21)
|
|
._m13(m31)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m23(m32)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._properties(0);
|
|
}
|
|
|
|
/**
|
|
* Transpose only the upper left 3x3 submatrix of this matrix.
|
|
* <p>
|
|
* All other matrix elements are left unchanged.
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f transpose3x3() {
|
|
return transpose3x3(this);
|
|
}
|
|
|
|
public Matrix4f transpose3x3(Matrix4f dest) {
|
|
float nm10 = m01, nm20 = m02, nm21 = m12;
|
|
return dest
|
|
._m00(m00)
|
|
._m01(m10)
|
|
._m02(m20)
|
|
._m10(nm10)
|
|
._m11(m11)
|
|
._m12(m21)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(m22)
|
|
._properties(this.properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
public Matrix3f transpose3x3(Matrix3f dest) {
|
|
return dest
|
|
._m00(m00)
|
|
._m01(m10)
|
|
._m02(m20)
|
|
._m10(m01)
|
|
._m11(m11)
|
|
._m12(m21)
|
|
._m20(m02)
|
|
._m21(m12)
|
|
._m22(m22);
|
|
}
|
|
|
|
/**
|
|
* Transpose this matrix.
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f transpose() {
|
|
return transpose(this);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a simple translation matrix.
|
|
* <p>
|
|
* The resulting matrix can be multiplied against another transformation
|
|
* matrix to obtain an additional translation.
|
|
* <p>
|
|
* In order to post-multiply a translation transformation directly to a
|
|
* matrix, use {@link #translate(float, float, float) translate()} instead.
|
|
*
|
|
* @see #translate(float, float, float)
|
|
*
|
|
* @param x
|
|
* the offset to translate in x
|
|
* @param y
|
|
* the offset to translate in y
|
|
* @param z
|
|
* the offset to translate in z
|
|
* @return this
|
|
*/
|
|
public Matrix4f translation(float x, float y, float z) {
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
return this._m30(x)._m31(y)._m32(z)._properties(PROPERTY_AFFINE | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a simple translation matrix.
|
|
* <p>
|
|
* The resulting matrix can be multiplied against another transformation
|
|
* matrix to obtain an additional translation.
|
|
* <p>
|
|
* In order to post-multiply a translation transformation directly to a
|
|
* matrix, use {@link #translate(Vector3fc) translate()} instead.
|
|
*
|
|
* @see #translate(float, float, float)
|
|
*
|
|
* @param offset
|
|
* the offsets in x, y and z to translate
|
|
* @return this
|
|
*/
|
|
public Matrix4f translation(Vector3fc offset) {
|
|
return translation(offset.x(), offset.y(), offset.z());
|
|
}
|
|
|
|
/**
|
|
* Set only the translation components <code>(m30, m31, m32)</code> of this matrix to the given values <code>(x, y, z)</code>.
|
|
* <p>
|
|
* Note that this will only work properly for orthogonal matrices (without any perspective).
|
|
* <p>
|
|
* To build a translation matrix instead, use {@link #translation(float, float, float)}.
|
|
* To apply a translation, use {@link #translate(float, float, float)}.
|
|
*
|
|
* @see #translation(float, float, float)
|
|
* @see #translate(float, float, float)
|
|
*
|
|
* @param x
|
|
* the offset to translate in x
|
|
* @param y
|
|
* the offset to translate in y
|
|
* @param z
|
|
* the offset to translate in z
|
|
* @return this
|
|
*/
|
|
public Matrix4f setTranslation(float x, float y, float z) {
|
|
return this._m30(x)._m31(y)._m32(z)._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY));
|
|
}
|
|
|
|
/**
|
|
* Set only the translation components <code>(m30, m31, m32)</code> of this matrix to the values <code>(xyz.x, xyz.y, xyz.z)</code>.
|
|
* <p>
|
|
* Note that this will only work properly for orthogonal matrices (without any perspective).
|
|
* <p>
|
|
* To build a translation matrix instead, use {@link #translation(Vector3fc)}.
|
|
* To apply a translation, use {@link #translate(Vector3fc)}.
|
|
*
|
|
* @see #translation(Vector3fc)
|
|
* @see #translate(Vector3fc)
|
|
*
|
|
* @param xyz
|
|
* the units to translate in <code>(x, y, z)</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f setTranslation(Vector3fc xyz) {
|
|
return setTranslation(xyz.x(), xyz.y(), xyz.z());
|
|
}
|
|
|
|
public Vector3f getTranslation(Vector3f dest) {
|
|
dest.x = m30;
|
|
dest.y = m31;
|
|
dest.z = m32;
|
|
return dest;
|
|
}
|
|
|
|
public Vector3f getScale(Vector3f dest) {
|
|
dest.x = Math.sqrt(m00 * m00 + m01 * m01 + m02 * m02);
|
|
dest.y = Math.sqrt(m10 * m10 + m11 * m11 + m12 * m12);
|
|
dest.z = Math.sqrt(m20 * m20 + m21 * m21 + m22 * m22);
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Return a string representation of this matrix.
|
|
* <p>
|
|
* This method creates a new {@link DecimalFormat} on every invocation with the format string "<code>0.000E0;-</code>".
|
|
*
|
|
* @return the string representation
|
|
*/
|
|
public String toString() {
|
|
String str = toString(Options.NUMBER_FORMAT);
|
|
StringBuffer res = new StringBuffer();
|
|
int eIndex = Integer.MIN_VALUE;
|
|
for (int i = 0; i < str.length(); i++) {
|
|
char c = str.charAt(i);
|
|
if (c == 'E') {
|
|
eIndex = i;
|
|
} else if (c == ' ' && eIndex == i - 1) {
|
|
// workaround Java 1.4 DecimalFormat bug
|
|
res.append('+');
|
|
continue;
|
|
} else if (Character.isDigit(c) && eIndex == i - 1) {
|
|
res.append('+');
|
|
}
|
|
res.append(c);
|
|
}
|
|
return res.toString();
|
|
}
|
|
|
|
/**
|
|
* Return a string representation of this matrix by formatting the matrix elements with the given {@link NumberFormat}.
|
|
*
|
|
* @param formatter
|
|
* the {@link NumberFormat} used to format the matrix values with
|
|
* @return the string representation
|
|
*/
|
|
public String toString(NumberFormat formatter) {
|
|
return Runtime.format(m00, formatter) + " " + Runtime.format(m10, formatter) + " " + Runtime.format(m20, formatter) + " " + Runtime.format(m30, formatter) + "\n"
|
|
+ Runtime.format(m01, formatter) + " " + Runtime.format(m11, formatter) + " " + Runtime.format(m21, formatter) + " " + Runtime.format(m31, formatter) + "\n"
|
|
+ Runtime.format(m02, formatter) + " " + Runtime.format(m12, formatter) + " " + Runtime.format(m22, formatter) + " " + Runtime.format(m32, formatter) + "\n"
|
|
+ Runtime.format(m03, formatter) + " " + Runtime.format(m13, formatter) + " " + Runtime.format(m23, formatter) + " " + Runtime.format(m33, formatter) + "\n";
|
|
}
|
|
|
|
/**
|
|
* Get the current values of <code>this</code> matrix and store them into
|
|
* <code>dest</code>.
|
|
* <p>
|
|
* This is the reverse method of {@link #set(Matrix4fc)} and allows to obtain
|
|
* intermediate calculation results when chaining multiple transformations.
|
|
*
|
|
* @see #set(Matrix4fc)
|
|
*
|
|
* @param dest
|
|
* the destination matrix
|
|
* @return the passed in destination
|
|
*/
|
|
public Matrix4f get(Matrix4f dest) {
|
|
return dest.set(this);
|
|
}
|
|
|
|
public Matrix4x3f get4x3(Matrix4x3f dest) {
|
|
return dest.set(this);
|
|
}
|
|
|
|
/**
|
|
* Get the current values of <code>this</code> matrix and store them into
|
|
* <code>dest</code>.
|
|
* <p>
|
|
* This is the reverse method of {@link #set(Matrix4dc)} and allows to obtain
|
|
* intermediate calculation results when chaining multiple transformations.
|
|
*
|
|
* @see #set(Matrix4dc)
|
|
*
|
|
* @param dest
|
|
* the destination matrix
|
|
* @return the passed in destination
|
|
*/
|
|
public Matrix4d get(Matrix4d dest) {
|
|
return dest.set(this);
|
|
}
|
|
|
|
public Matrix3f get3x3(Matrix3f dest) {
|
|
return dest.set(this);
|
|
}
|
|
|
|
public Matrix3d get3x3(Matrix3d dest) {
|
|
return dest.set(this);
|
|
}
|
|
|
|
public AxisAngle4f getRotation(AxisAngle4f dest) {
|
|
return dest.set(this);
|
|
}
|
|
|
|
public AxisAngle4d getRotation(AxisAngle4d dest) {
|
|
return dest.set(this);
|
|
}
|
|
|
|
public Quaternionf getUnnormalizedRotation(Quaternionf dest) {
|
|
return dest.setFromUnnormalized(this);
|
|
}
|
|
|
|
public Quaternionf getNormalizedRotation(Quaternionf dest) {
|
|
return dest.setFromNormalized(this);
|
|
}
|
|
|
|
public Quaterniond getUnnormalizedRotation(Quaterniond dest) {
|
|
return dest.setFromUnnormalized(this);
|
|
}
|
|
|
|
public Quaterniond getNormalizedRotation(Quaterniond dest) {
|
|
return dest.setFromNormalized(this);
|
|
}
|
|
|
|
|
|
public FloatBuffer get(FloatBuffer buffer) {
|
|
MemUtil.INSTANCE.put(this, buffer.position(), buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public FloatBuffer get(int index, FloatBuffer buffer) {
|
|
MemUtil.INSTANCE.put(this, index, buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public ByteBuffer get(ByteBuffer buffer) {
|
|
MemUtil.INSTANCE.put(this, buffer.position(), buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public ByteBuffer get(int index, ByteBuffer buffer) {
|
|
MemUtil.INSTANCE.put(this, index, buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public FloatBuffer get4x3(FloatBuffer buffer) {
|
|
MemUtil.INSTANCE.put4x3(this, buffer.position(), buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public FloatBuffer get4x3(int index, FloatBuffer buffer) {
|
|
MemUtil.INSTANCE.put4x3(this, index, buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public ByteBuffer get4x3(ByteBuffer buffer) {
|
|
MemUtil.INSTANCE.put4x3(this, buffer.position(), buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public ByteBuffer get4x3(int index, ByteBuffer buffer) {
|
|
MemUtil.INSTANCE.put4x3(this, index, buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public FloatBuffer get3x4(FloatBuffer buffer) {
|
|
MemUtil.INSTANCE.put3x4(this, buffer.position(), buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public FloatBuffer get3x4(int index, FloatBuffer buffer) {
|
|
MemUtil.INSTANCE.put3x4(this, index, buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public ByteBuffer get3x4(ByteBuffer buffer) {
|
|
MemUtil.INSTANCE.put3x4(this, buffer.position(), buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public ByteBuffer get3x4(int index, ByteBuffer buffer) {
|
|
MemUtil.INSTANCE.put3x4(this, index, buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public FloatBuffer getTransposed(FloatBuffer buffer) {
|
|
MemUtil.INSTANCE.putTransposed(this, buffer.position(), buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public FloatBuffer getTransposed(int index, FloatBuffer buffer) {
|
|
MemUtil.INSTANCE.putTransposed(this, index, buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public ByteBuffer getTransposed(ByteBuffer buffer) {
|
|
MemUtil.INSTANCE.putTransposed(this, buffer.position(), buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public ByteBuffer getTransposed(int index, ByteBuffer buffer) {
|
|
MemUtil.INSTANCE.putTransposed(this, index, buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public FloatBuffer get4x3Transposed(FloatBuffer buffer) {
|
|
MemUtil.INSTANCE.put4x3Transposed(this, buffer.position(), buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public FloatBuffer get4x3Transposed(int index, FloatBuffer buffer) {
|
|
MemUtil.INSTANCE.put4x3Transposed(this, index, buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public ByteBuffer get4x3Transposed(ByteBuffer buffer) {
|
|
MemUtil.INSTANCE.put4x3Transposed(this, buffer.position(), buffer);
|
|
return buffer;
|
|
}
|
|
|
|
public ByteBuffer get4x3Transposed(int index, ByteBuffer buffer) {
|
|
MemUtil.INSTANCE.put4x3Transposed(this, index, buffer);
|
|
return buffer;
|
|
}
|
|
public Matrix4fc getToAddress(long address) {
|
|
if (Options.NO_UNSAFE)
|
|
throw new UnsupportedOperationException("Not supported when using joml.nounsafe");
|
|
MemUtil.MemUtilUnsafe.put(this, address);
|
|
return this;
|
|
}
|
|
|
|
public float[] get(float[] arr, int offset) {
|
|
MemUtil.INSTANCE.copy(this, arr, offset);
|
|
return arr;
|
|
}
|
|
|
|
public float[] get(float[] arr) {
|
|
MemUtil.INSTANCE.copy(this, arr, 0);
|
|
return arr;
|
|
}
|
|
|
|
/**
|
|
* Set all the values within this matrix to <code>0</code>.
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f zero() {
|
|
MemUtil.INSTANCE.zero(this);
|
|
return _properties(0);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a simple scale matrix, which scales all axes uniformly by the given factor.
|
|
* <p>
|
|
* The resulting matrix can be multiplied against another transformation
|
|
* matrix to obtain an additional scaling.
|
|
* <p>
|
|
* In order to post-multiply a scaling transformation directly to a
|
|
* matrix, use {@link #scale(float) scale()} instead.
|
|
*
|
|
* @see #scale(float)
|
|
*
|
|
* @param factor
|
|
* the scale factor in x, y and z
|
|
* @return this
|
|
*/
|
|
public Matrix4f scaling(float factor) {
|
|
return scaling(factor, factor, factor);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a simple scale matrix.
|
|
* <p>
|
|
* The resulting matrix can be multiplied against another transformation
|
|
* matrix to obtain an additional scaling.
|
|
* <p>
|
|
* In order to post-multiply a scaling transformation directly to a
|
|
* matrix, use {@link #scale(float, float, float) scale()} instead.
|
|
*
|
|
* @see #scale(float, float, float)
|
|
*
|
|
* @param x
|
|
* the scale in x
|
|
* @param y
|
|
* the scale in y
|
|
* @param z
|
|
* the scale in z
|
|
* @return this
|
|
*/
|
|
public Matrix4f scaling(float x, float y, float z) {
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
boolean one = Math.absEqualsOne(x) && Math.absEqualsOne(y) && Math.absEqualsOne(z);
|
|
return this
|
|
._m00(x)
|
|
._m11(y)
|
|
._m22(z)
|
|
._properties(PROPERTY_AFFINE | (one ? PROPERTY_ORTHONORMAL : 0));
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a simple scale matrix which scales the base axes by <code>xyz.x</code>, <code>xyz.y</code> and <code>xyz.z</code> respectively.
|
|
* <p>
|
|
* The resulting matrix can be multiplied against another transformation
|
|
* matrix to obtain an additional scaling.
|
|
* <p>
|
|
* In order to post-multiply a scaling transformation directly to a
|
|
* matrix use {@link #scale(Vector3fc) scale()} instead.
|
|
*
|
|
* @see #scale(Vector3fc)
|
|
*
|
|
* @param xyz
|
|
* the scale in x, y and z respectively
|
|
* @return this
|
|
*/
|
|
public Matrix4f scaling(Vector3fc xyz) {
|
|
return scaling(xyz.x(), xyz.y(), xyz.z());
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a rotation matrix which rotates the given radians about a given axis.
|
|
* <p>
|
|
* The axis described by the <code>axis</code> vector needs to be a unit vector.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* The resulting matrix can be multiplied against another transformation
|
|
* matrix to obtain an additional rotation.
|
|
* <p>
|
|
* In order to post-multiply a rotation transformation directly to a
|
|
* matrix, use {@link #rotate(float, Vector3fc) rotate()} instead.
|
|
*
|
|
* @see #rotate(float, Vector3fc)
|
|
*
|
|
* @param angle
|
|
* the angle in radians
|
|
* @param axis
|
|
* the axis to rotate about (needs to be {@link Vector3f#normalize() normalized})
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotation(float angle, Vector3fc axis) {
|
|
return rotation(angle, axis.x(), axis.y(), axis.z());
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a rotation transformation using the given {@link AxisAngle4f}.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* The resulting matrix can be multiplied against another transformation
|
|
* matrix to obtain an additional rotation.
|
|
* <p>
|
|
* In order to apply the rotation transformation to an existing transformation,
|
|
* use {@link #rotate(AxisAngle4f) rotate()} instead.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotate(AxisAngle4f)
|
|
*
|
|
* @param axisAngle
|
|
* the {@link AxisAngle4f} (needs to be {@link AxisAngle4f#normalize() normalized})
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotation(AxisAngle4f axisAngle) {
|
|
return rotation(axisAngle.angle, axisAngle.x, axisAngle.y, axisAngle.z);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a rotation matrix which rotates the given radians about a given axis.
|
|
* <p>
|
|
* The axis described by the three components needs to be a unit vector.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* The resulting matrix can be multiplied against another transformation
|
|
* matrix to obtain an additional rotation.
|
|
* <p>
|
|
* In order to apply the rotation transformation to an existing transformation,
|
|
* use {@link #rotate(float, float, float, float) rotate()} instead.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Rotation_matrix_from_axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotate(float, float, float, float)
|
|
*
|
|
* @param angle
|
|
* the angle in radians
|
|
* @param x
|
|
* the x-component of the rotation axis
|
|
* @param y
|
|
* the y-component of the rotation axis
|
|
* @param z
|
|
* the z-component of the rotation axis
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotation(float angle, float x, float y, float z) {
|
|
if (y == 0.0f && z == 0.0f && Math.absEqualsOne(x))
|
|
return rotationX(x * angle);
|
|
else if (x == 0.0f && z == 0.0f && Math.absEqualsOne(y))
|
|
return rotationY(y * angle);
|
|
else if (x == 0.0f && y == 0.0f && Math.absEqualsOne(z))
|
|
return rotationZ(z * angle);
|
|
return rotationInternal(angle, x, y, z);
|
|
}
|
|
private Matrix4f rotationInternal(float angle, float x, float y, float z) {
|
|
float sin = Math.sin(angle), cos = Math.cosFromSin(sin, angle);
|
|
float C = 1.0f - cos, xy = x * y, xz = x * z, yz = y * z;
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
return this
|
|
._m00(cos + x * x * C)
|
|
._m10(xy * C - z * sin)
|
|
._m20(xz * C + y * sin)
|
|
._m01(xy * C + z * sin)
|
|
._m11(cos + y * y * C)
|
|
._m21(yz * C - x * sin)
|
|
._m02(xz * C - y * sin)
|
|
._m12(yz * C + x * sin)
|
|
._m22(cos + z * z * C)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a rotation transformation about the X axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Basic_rotations">http://en.wikipedia.org</a>
|
|
*
|
|
* @param ang
|
|
* the angle in radians
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotationX(float ang) {
|
|
float sin = Math.sin(ang), cos = Math.cosFromSin(sin, ang);
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
this._m11(cos)._m12(sin)._m21(-sin)._m22(cos)._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a rotation transformation about the Y axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Basic_rotations">http://en.wikipedia.org</a>
|
|
*
|
|
* @param ang
|
|
* the angle in radians
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotationY(float ang) {
|
|
float sin = Math.sin(ang), cos = Math.cosFromSin(sin, ang);
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
this._m00(cos)._m02(-sin)._m20(sin)._m22(cos)._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a rotation transformation about the Z axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Basic_rotations">http://en.wikipedia.org</a>
|
|
*
|
|
* @param ang
|
|
* the angle in radians
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotationZ(float ang) {
|
|
float sin = Math.sin(ang), cos = Math.cosFromSin(sin, ang);
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
return this._m00(cos)._m01(sin)._m10(-sin)._m11(cos)._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a rotation transformation about the Z axis to align the local <code>+X</code> towards <code>(dirX, dirY)</code>.
|
|
* <p>
|
|
* The vector <code>(dirX, dirY)</code> must be a unit vector.
|
|
*
|
|
* @param dirX
|
|
* the x component of the normalized direction
|
|
* @param dirY
|
|
* the y component of the normalized direction
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotationTowardsXY(float dirX, float dirY) {
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
return this._m00(dirY)._m01(dirX)._m10(-dirX)._m11(dirY)._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a rotation of <code>angleX</code> radians about the X axis, followed by a rotation
|
|
* of <code>angleY</code> radians about the Y axis and followed by a rotation of <code>angleZ</code> radians about the Z axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>rotationX(angleX).rotateY(angleY).rotateZ(angleZ)</code>
|
|
*
|
|
* @param angleX
|
|
* the angle to rotate about X
|
|
* @param angleY
|
|
* the angle to rotate about Y
|
|
* @param angleZ
|
|
* the angle to rotate about Z
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotationXYZ(float angleX, float angleY, float angleZ) {
|
|
float sinX = Math.sin(angleX);
|
|
float cosX = Math.cosFromSin(sinX, angleX);
|
|
float sinY = Math.sin(angleY);
|
|
float cosY = Math.cosFromSin(sinY, angleY);
|
|
float sinZ = Math.sin(angleZ);
|
|
float cosZ = Math.cosFromSin(sinZ, angleZ);
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
float nm01 = -sinX * -sinY, nm02 = cosX * -sinY;
|
|
return this
|
|
._m20(sinY)
|
|
._m21(-sinX * cosY)
|
|
._m22(cosX * cosY)
|
|
._m00(cosY * cosZ)
|
|
._m01(nm01 * cosZ + cosX * sinZ)
|
|
._m02(nm02 * cosZ + sinX * sinZ)
|
|
._m10(cosY * -sinZ)
|
|
._m11(nm01 * -sinZ + cosX * cosZ)
|
|
._m12(nm02 * -sinZ + sinX * cosZ)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a rotation of <code>angleZ</code> radians about the Z axis, followed by a rotation
|
|
* of <code>angleY</code> radians about the Y axis and followed by a rotation of <code>angleX</code> radians about the X axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>rotationZ(angleZ).rotateY(angleY).rotateX(angleX)</code>
|
|
*
|
|
* @param angleZ
|
|
* the angle to rotate about Z
|
|
* @param angleY
|
|
* the angle to rotate about Y
|
|
* @param angleX
|
|
* the angle to rotate about X
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotationZYX(float angleZ, float angleY, float angleX) {
|
|
float sinX = Math.sin(angleX);
|
|
float cosX = Math.cosFromSin(sinX, angleX);
|
|
float sinY = Math.sin(angleY);
|
|
float cosY = Math.cosFromSin(sinY, angleY);
|
|
float sinZ = Math.sin(angleZ);
|
|
float cosZ = Math.cosFromSin(sinZ, angleZ);
|
|
float nm20 = cosZ * sinY;
|
|
float nm21 = sinZ * sinY;
|
|
return this
|
|
._m00(cosZ * cosY)
|
|
._m01(sinZ * cosY)
|
|
._m02(-sinY)
|
|
._m03(0.0f)
|
|
._m10(-sinZ * cosX + nm20 * sinX)
|
|
._m11(cosZ * cosX + nm21 * sinX)
|
|
._m12(cosY * sinX)
|
|
._m13(0.0f)
|
|
._m20(-sinZ * -sinX + nm20 * cosX)
|
|
._m21(cosZ * -sinX + nm21 * cosX)
|
|
._m22(cosY * cosX)
|
|
._m23(0.0f)
|
|
._m30(0.0f)
|
|
._m31(0.0f)
|
|
._m32(0.0f)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a rotation of <code>angleY</code> radians about the Y axis, followed by a rotation
|
|
* of <code>angleX</code> radians about the X axis and followed by a rotation of <code>angleZ</code> radians about the Z axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>rotationY(angleY).rotateX(angleX).rotateZ(angleZ)</code>
|
|
*
|
|
* @param angleY
|
|
* the angle to rotate about Y
|
|
* @param angleX
|
|
* the angle to rotate about X
|
|
* @param angleZ
|
|
* the angle to rotate about Z
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotationYXZ(float angleY, float angleX, float angleZ) {
|
|
float sinX = Math.sin(angleX);
|
|
float cosX = Math.cosFromSin(sinX, angleX);
|
|
float sinY = Math.sin(angleY);
|
|
float cosY = Math.cosFromSin(sinY, angleY);
|
|
float sinZ = Math.sin(angleZ);
|
|
float cosZ = Math.cosFromSin(sinZ, angleZ);
|
|
float nm10 = sinY * sinX, nm12 = cosY * sinX;
|
|
return this
|
|
._m20(sinY * cosX)
|
|
._m21(-sinX)
|
|
._m22(cosY * cosX)
|
|
._m23(0.0f)
|
|
._m00(cosY * cosZ + nm10 * sinZ)
|
|
._m01(cosX * sinZ)
|
|
._m02(-sinY * cosZ + nm12 * sinZ)
|
|
._m03(0.0f)
|
|
._m10(cosY * -sinZ + nm10 * cosZ)
|
|
._m11(cosX * cosZ)
|
|
._m12(-sinY * -sinZ + nm12 * cosZ)
|
|
._m13(0.0f)
|
|
._m30(0.0f)
|
|
._m31(0.0f)
|
|
._m32(0.0f)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Set only the upper left 3x3 submatrix of this matrix to a rotation of <code>angleX</code> radians about the X axis, followed by a rotation
|
|
* of <code>angleY</code> radians about the Y axis and followed by a rotation of <code>angleZ</code> radians about the Z axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
*
|
|
* @param angleX
|
|
* the angle to rotate about X
|
|
* @param angleY
|
|
* the angle to rotate about Y
|
|
* @param angleZ
|
|
* the angle to rotate about Z
|
|
* @return this
|
|
*/
|
|
public Matrix4f setRotationXYZ(float angleX, float angleY, float angleZ) {
|
|
float sinX = Math.sin(angleX);
|
|
float cosX = Math.cosFromSin(sinX, angleX);
|
|
float sinY = Math.sin(angleY);
|
|
float cosY = Math.cosFromSin(sinY, angleY);
|
|
float sinZ = Math.sin(angleZ);
|
|
float cosZ = Math.cosFromSin(sinZ, angleZ);
|
|
float nm01 = -sinX * -sinY;
|
|
float nm02 = cosX * -sinY;
|
|
return this
|
|
._m20(sinY)
|
|
._m21(-sinX * cosY)
|
|
._m22(cosX * cosY)
|
|
._m00(cosY * cosZ)
|
|
._m01(nm01 * cosZ + cosX * sinZ)
|
|
._m02(nm02 * cosZ + sinX * sinZ)
|
|
._m10(cosY * -sinZ)
|
|
._m11(nm01 * -sinZ + cosX * cosZ)
|
|
._m12(nm02 * -sinZ + sinX * cosZ)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Set only the upper left 3x3 submatrix of this matrix to a rotation of <code>angleZ</code> radians about the Z axis, followed by a rotation
|
|
* of <code>angleY</code> radians about the Y axis and followed by a rotation of <code>angleX</code> radians about the X axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
*
|
|
* @param angleZ
|
|
* the angle to rotate about Z
|
|
* @param angleY
|
|
* the angle to rotate about Y
|
|
* @param angleX
|
|
* the angle to rotate about X
|
|
* @return this
|
|
*/
|
|
public Matrix4f setRotationZYX(float angleZ, float angleY, float angleX) {
|
|
float sinX = Math.sin(angleX);
|
|
float cosX = Math.cosFromSin(sinX, angleX);
|
|
float sinY = Math.sin(angleY);
|
|
float cosY = Math.cosFromSin(sinY, angleY);
|
|
float sinZ = Math.sin(angleZ);
|
|
float cosZ = Math.cosFromSin(sinZ, angleZ);
|
|
float nm20 = cosZ * sinY, nm21 = sinZ * sinY;
|
|
return this
|
|
._m00(cosZ * cosY)
|
|
._m01(sinZ * cosY)
|
|
._m02(-sinY)
|
|
._m10(-sinZ * cosX + nm20 * sinX)
|
|
._m11(cosZ * cosX + nm21 * sinX)
|
|
._m12(cosY * sinX)
|
|
._m20(-sinZ * -sinX + nm20 * cosX)
|
|
._m21(cosZ * -sinX + nm21 * cosX)
|
|
._m22(cosY * cosX)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Set only the upper left 3x3 submatrix of this matrix to a rotation of <code>angleY</code> radians about the Y axis, followed by a rotation
|
|
* of <code>angleX</code> radians about the X axis and followed by a rotation of <code>angleZ</code> radians about the Z axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
*
|
|
* @param angleY
|
|
* the angle to rotate about Y
|
|
* @param angleX
|
|
* the angle to rotate about X
|
|
* @param angleZ
|
|
* the angle to rotate about Z
|
|
* @return this
|
|
*/
|
|
public Matrix4f setRotationYXZ(float angleY, float angleX, float angleZ) {
|
|
float sinX = Math.sin(angleX);
|
|
float cosX = Math.cosFromSin(sinX, angleX);
|
|
float sinY = Math.sin(angleY);
|
|
float cosY = Math.cosFromSin(sinY, angleY);
|
|
float sinZ = Math.sin(angleZ);
|
|
float cosZ = Math.cosFromSin(sinZ, angleZ);
|
|
float nm10 = sinY * sinX, nm12 = cosY * sinX;
|
|
return this
|
|
._m20(sinY * cosX)
|
|
._m21(-sinX)
|
|
._m22(cosY * cosX)
|
|
._m00(cosY * cosZ + nm10 * sinZ)
|
|
._m01(cosX * sinZ)
|
|
._m02(-sinY * cosZ + nm12 * sinZ)
|
|
._m10(cosY * -sinZ + nm10 * cosZ)
|
|
._m11(cosX * cosZ)
|
|
._m12(-sinY * -sinZ + nm12 * cosZ)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to the rotation transformation of the given {@link Quaternionfc}.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* The resulting matrix can be multiplied against another transformation
|
|
* matrix to obtain an additional rotation.
|
|
* <p>
|
|
* In order to apply the rotation transformation to an existing transformation,
|
|
* use {@link #rotate(Quaternionfc) rotate()} instead.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Quaternion">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotate(Quaternionfc)
|
|
*
|
|
* @param quat
|
|
* the {@link Quaternionfc}
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotation(Quaternionfc quat) {
|
|
float w2 = quat.w() * quat.w();
|
|
float x2 = quat.x() * quat.x();
|
|
float y2 = quat.y() * quat.y();
|
|
float z2 = quat.z() * quat.z();
|
|
float zw = quat.z() * quat.w(), dzw = zw + zw;
|
|
float xy = quat.x() * quat.y(), dxy = xy + xy;
|
|
float xz = quat.x() * quat.z(), dxz = xz + xz;
|
|
float yw = quat.y() * quat.w(), dyw = yw + yw;
|
|
float yz = quat.y() * quat.z(), dyz = yz + yz;
|
|
float xw = quat.x() * quat.w(), dxw = xw + xw;
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
return this
|
|
._m00(w2 + x2 - z2 - y2)
|
|
._m01(dxy + dzw)
|
|
._m02(dxz - dyw)
|
|
._m10(-dzw + dxy)
|
|
._m11(y2 - z2 + w2 - x2)
|
|
._m12(dyz + dxw)
|
|
._m20(dyw + dxz)
|
|
._m21(dyz - dxw)
|
|
._m22(z2 - y2 - x2 + w2)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Set <code>this</code> matrix to <code>T * R * S</code>, where <code>T</code> is a translation by the given <code>(tx, ty, tz)</code>,
|
|
* <code>R</code> is a rotation transformation specified by the quaternion <code>(qx, qy, qz, qw)</code>, and <code>S</code> is a scaling transformation
|
|
* which scales the three axes x, y and z by <code>(sx, sy, sz)</code>.
|
|
* <p>
|
|
* When transforming a vector by the resulting matrix the scaling transformation will be applied first, then the rotation and
|
|
* at last the translation.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translation(tx, ty, tz).rotate(quat).scale(sx, sy, sz)</code>
|
|
*
|
|
* @see #translation(float, float, float)
|
|
* @see #rotate(Quaternionfc)
|
|
* @see #scale(float, float, float)
|
|
*
|
|
* @param tx
|
|
* the number of units by which to translate the x-component
|
|
* @param ty
|
|
* the number of units by which to translate the y-component
|
|
* @param tz
|
|
* the number of units by which to translate the z-component
|
|
* @param qx
|
|
* the x-coordinate of the vector part of the quaternion
|
|
* @param qy
|
|
* the y-coordinate of the vector part of the quaternion
|
|
* @param qz
|
|
* the z-coordinate of the vector part of the quaternion
|
|
* @param qw
|
|
* the scalar part of the quaternion
|
|
* @param sx
|
|
* the scaling factor for the x-axis
|
|
* @param sy
|
|
* the scaling factor for the y-axis
|
|
* @param sz
|
|
* the scaling factor for the z-axis
|
|
* @return this
|
|
*/
|
|
public Matrix4f translationRotateScale(float tx, float ty, float tz,
|
|
float qx, float qy, float qz, float qw,
|
|
float sx, float sy, float sz) {
|
|
float dqx = qx + qx;
|
|
float dqy = qy + qy;
|
|
float dqz = qz + qz;
|
|
float q00 = dqx * qx;
|
|
float q11 = dqy * qy;
|
|
float q22 = dqz * qz;
|
|
float q01 = dqx * qy;
|
|
float q02 = dqx * qz;
|
|
float q03 = dqx * qw;
|
|
float q12 = dqy * qz;
|
|
float q13 = dqy * qw;
|
|
float q23 = dqz * qw;
|
|
boolean one = Math.absEqualsOne(sx) && Math.absEqualsOne(sy) && Math.absEqualsOne(sz);
|
|
return this
|
|
._m00(sx - (q11 + q22) * sx)
|
|
._m01((q01 + q23) * sx)
|
|
._m02((q02 - q13) * sx)
|
|
._m03(0.0f)
|
|
._m10((q01 - q23) * sy)
|
|
._m11(sy - (q22 + q00) * sy)
|
|
._m12((q12 + q03) * sy)
|
|
._m13(0.0f)
|
|
._m20((q02 + q13) * sz)
|
|
._m21((q12 - q03) * sz)
|
|
._m22(sz - (q11 + q00) * sz)
|
|
._m23(0.0f)
|
|
._m30(tx)
|
|
._m31(ty)
|
|
._m32(tz)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | (one ? PROPERTY_ORTHONORMAL : 0));
|
|
}
|
|
|
|
/**
|
|
* Set <code>this</code> matrix to <code>T * R * S</code>, where <code>T</code> is the given <code>translation</code>,
|
|
* <code>R</code> is a rotation transformation specified by the given quaternion, and <code>S</code> is a scaling transformation
|
|
* which scales the axes by <code>scale</code>.
|
|
* <p>
|
|
* When transforming a vector by the resulting matrix the scaling transformation will be applied first, then the rotation and
|
|
* at last the translation.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translation(translation).rotate(quat).scale(scale)</code>
|
|
*
|
|
* @see #translation(Vector3fc)
|
|
* @see #rotate(Quaternionfc)
|
|
* @see #scale(Vector3fc)
|
|
*
|
|
* @param translation
|
|
* the translation
|
|
* @param quat
|
|
* the quaternion representing a rotation
|
|
* @param scale
|
|
* the scaling factors
|
|
* @return this
|
|
*/
|
|
public Matrix4f translationRotateScale(Vector3fc translation,
|
|
Quaternionfc quat,
|
|
Vector3fc scale) {
|
|
return translationRotateScale(translation.x(), translation.y(), translation.z(), quat.x(), quat.y(), quat.z(), quat.w(), scale.x(), scale.y(), scale.z());
|
|
}
|
|
|
|
/**
|
|
* Set <code>this</code> matrix to <code>T * R * S</code>, where <code>T</code> is a translation by the given <code>(tx, ty, tz)</code>,
|
|
* <code>R</code> is a rotation transformation specified by the quaternion <code>(qx, qy, qz, qw)</code>, and <code>S</code> is a scaling transformation
|
|
* which scales all three axes by <code>scale</code>.
|
|
* <p>
|
|
* When transforming a vector by the resulting matrix the scaling transformation will be applied first, then the rotation and
|
|
* at last the translation.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translation(tx, ty, tz).rotate(quat).scale(scale)</code>
|
|
*
|
|
* @see #translation(float, float, float)
|
|
* @see #rotate(Quaternionfc)
|
|
* @see #scale(float)
|
|
*
|
|
* @param tx
|
|
* the number of units by which to translate the x-component
|
|
* @param ty
|
|
* the number of units by which to translate the y-component
|
|
* @param tz
|
|
* the number of units by which to translate the z-component
|
|
* @param qx
|
|
* the x-coordinate of the vector part of the quaternion
|
|
* @param qy
|
|
* the y-coordinate of the vector part of the quaternion
|
|
* @param qz
|
|
* the z-coordinate of the vector part of the quaternion
|
|
* @param qw
|
|
* the scalar part of the quaternion
|
|
* @param scale
|
|
* the scaling factor for all three axes
|
|
* @return this
|
|
*/
|
|
public Matrix4f translationRotateScale(float tx, float ty, float tz,
|
|
float qx, float qy, float qz, float qw,
|
|
float scale) {
|
|
return translationRotateScale(tx, ty, tz, qx, qy, qz, qw, scale, scale, scale);
|
|
}
|
|
|
|
/**
|
|
* Set <code>this</code> matrix to <code>T * R * S</code>, where <code>T</code> is the given <code>translation</code>,
|
|
* <code>R</code> is a rotation transformation specified by the given quaternion, and <code>S</code> is a scaling transformation
|
|
* which scales all three axes by <code>scale</code>.
|
|
* <p>
|
|
* When transforming a vector by the resulting matrix the scaling transformation will be applied first, then the rotation and
|
|
* at last the translation.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translation(translation).rotate(quat).scale(scale)</code>
|
|
*
|
|
* @see #translation(Vector3fc)
|
|
* @see #rotate(Quaternionfc)
|
|
* @see #scale(float)
|
|
*
|
|
* @param translation
|
|
* the translation
|
|
* @param quat
|
|
* the quaternion representing a rotation
|
|
* @param scale
|
|
* the scaling factors
|
|
* @return this
|
|
*/
|
|
public Matrix4f translationRotateScale(Vector3fc translation,
|
|
Quaternionfc quat,
|
|
float scale) {
|
|
return translationRotateScale(translation.x(), translation.y(), translation.z(), quat.x(), quat.y(), quat.z(), quat.w(), scale, scale, scale);
|
|
}
|
|
|
|
/**
|
|
* Set <code>this</code> matrix to <code>(T * R * S)<sup>-1</sup></code>, where <code>T</code> is a translation by the given <code>(tx, ty, tz)</code>,
|
|
* <code>R</code> is a rotation transformation specified by the quaternion <code>(qx, qy, qz, qw)</code>, and <code>S</code> is a scaling transformation
|
|
* which scales the three axes x, y and z by <code>(sx, sy, sz)</code>.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translationRotateScale(...).invert()</code>
|
|
*
|
|
* @see #translationRotateScale(float, float, float, float, float, float, float, float, float, float)
|
|
* @see #invert()
|
|
*
|
|
* @param tx
|
|
* the number of units by which to translate the x-component
|
|
* @param ty
|
|
* the number of units by which to translate the y-component
|
|
* @param tz
|
|
* the number of units by which to translate the z-component
|
|
* @param qx
|
|
* the x-coordinate of the vector part of the quaternion
|
|
* @param qy
|
|
* the y-coordinate of the vector part of the quaternion
|
|
* @param qz
|
|
* the z-coordinate of the vector part of the quaternion
|
|
* @param qw
|
|
* the scalar part of the quaternion
|
|
* @param sx
|
|
* the scaling factor for the x-axis
|
|
* @param sy
|
|
* the scaling factor for the y-axis
|
|
* @param sz
|
|
* the scaling factor for the z-axis
|
|
* @return this
|
|
*/
|
|
public Matrix4f translationRotateScaleInvert(float tx, float ty, float tz,
|
|
float qx, float qy, float qz, float qw,
|
|
float sx, float sy, float sz) {
|
|
boolean one = Math.absEqualsOne(sx) && Math.absEqualsOne(sy) && Math.absEqualsOne(sz);
|
|
if (one)
|
|
return translationRotateScale(tx, ty, tz, qx, qy, qz, qw, sx, sy, sz).invertOrthonormal(this);
|
|
float nqx = -qx, nqy = -qy, nqz = -qz;
|
|
float dqx = nqx + nqx;
|
|
float dqy = nqy + nqy;
|
|
float dqz = nqz + nqz;
|
|
float q00 = dqx * nqx;
|
|
float q11 = dqy * nqy;
|
|
float q22 = dqz * nqz;
|
|
float q01 = dqx * nqy;
|
|
float q02 = dqx * nqz;
|
|
float q03 = dqx * qw;
|
|
float q12 = dqy * nqz;
|
|
float q13 = dqy * qw;
|
|
float q23 = dqz * qw;
|
|
float isx = 1/sx, isy = 1/sy, isz = 1/sz;
|
|
return this
|
|
._m00(isx * (1.0f - q11 - q22))
|
|
._m01(isy * (q01 + q23))
|
|
._m02(isz * (q02 - q13))
|
|
._m03(0.0f)
|
|
._m10(isx * (q01 - q23))
|
|
._m11(isy * (1.0f - q22 - q00))
|
|
._m12(isz * (q12 + q03))
|
|
._m13(0.0f)
|
|
._m20(isx * (q02 + q13))
|
|
._m21(isy * (q12 - q03))
|
|
._m22(isz * (1.0f - q11 - q00))
|
|
._m23(0.0f)
|
|
._m30(-m00 * tx - m10 * ty - m20 * tz)
|
|
._m31(-m01 * tx - m11 * ty - m21 * tz)
|
|
._m32(-m02 * tx - m12 * ty - m22 * tz)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE);
|
|
}
|
|
|
|
/**
|
|
* Set <code>this</code> matrix to <code>(T * R * S)<sup>-1</sup></code>, where <code>T</code> is the given <code>translation</code>,
|
|
* <code>R</code> is a rotation transformation specified by the given quaternion, and <code>S</code> is a scaling transformation
|
|
* which scales the axes by <code>scale</code>.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translationRotateScale(...).invert()</code>
|
|
*
|
|
* @see #translationRotateScale(Vector3fc, Quaternionfc, Vector3fc)
|
|
* @see #invert()
|
|
*
|
|
* @param translation
|
|
* the translation
|
|
* @param quat
|
|
* the quaternion representing a rotation
|
|
* @param scale
|
|
* the scaling factors
|
|
* @return this
|
|
*/
|
|
public Matrix4f translationRotateScaleInvert(Vector3fc translation,
|
|
Quaternionfc quat,
|
|
Vector3fc scale) {
|
|
return translationRotateScaleInvert(translation.x(), translation.y(), translation.z(), quat.x(), quat.y(), quat.z(), quat.w(), scale.x(), scale.y(), scale.z());
|
|
}
|
|
|
|
/**
|
|
* Set <code>this</code> matrix to <code>(T * R * S)<sup>-1</sup></code>, where <code>T</code> is the given <code>translation</code>,
|
|
* <code>R</code> is a rotation transformation specified by the given quaternion, and <code>S</code> is a scaling transformation
|
|
* which scales all three axes by <code>scale</code>.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translationRotateScale(...).invert()</code>
|
|
*
|
|
* @see #translationRotateScale(Vector3fc, Quaternionfc, float)
|
|
* @see #invert()
|
|
*
|
|
* @param translation
|
|
* the translation
|
|
* @param quat
|
|
* the quaternion representing a rotation
|
|
* @param scale
|
|
* the scaling factors
|
|
* @return this
|
|
*/
|
|
public Matrix4f translationRotateScaleInvert(Vector3fc translation,
|
|
Quaternionfc quat,
|
|
float scale) {
|
|
return translationRotateScaleInvert(translation.x(), translation.y(), translation.z(), quat.x(), quat.y(), quat.z(), quat.w(), scale, scale, scale);
|
|
}
|
|
|
|
/**
|
|
* Set <code>this</code> matrix to <code>T * R * S * M</code>, where <code>T</code> is a translation by the given <code>(tx, ty, tz)</code>,
|
|
* <code>R</code> is a rotation - and possibly scaling - transformation specified by the quaternion <code>(qx, qy, qz, qw)</code>, <code>S</code> is a scaling transformation
|
|
* which scales the three axes x, y and z by <code>(sx, sy, sz)</code> and <code>M</code> is an {@link #isAffine() affine} matrix.
|
|
* <p>
|
|
* When transforming a vector by the resulting matrix the transformation described by <code>M</code> will be applied first, then the scaling, then rotation and
|
|
* at last the translation.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translation(tx, ty, tz).rotate(quat).scale(sx, sy, sz).mulAffine(m)</code>
|
|
*
|
|
* @see #translation(float, float, float)
|
|
* @see #rotate(Quaternionfc)
|
|
* @see #scale(float, float, float)
|
|
* @see #mulAffine(Matrix4fc)
|
|
*
|
|
* @param tx
|
|
* the number of units by which to translate the x-component
|
|
* @param ty
|
|
* the number of units by which to translate the y-component
|
|
* @param tz
|
|
* the number of units by which to translate the z-component
|
|
* @param qx
|
|
* the x-coordinate of the vector part of the quaternion
|
|
* @param qy
|
|
* the y-coordinate of the vector part of the quaternion
|
|
* @param qz
|
|
* the z-coordinate of the vector part of the quaternion
|
|
* @param qw
|
|
* the scalar part of the quaternion
|
|
* @param sx
|
|
* the scaling factor for the x-axis
|
|
* @param sy
|
|
* the scaling factor for the y-axis
|
|
* @param sz
|
|
* the scaling factor for the z-axis
|
|
* @param m
|
|
* the {@link #isAffine() affine} matrix to multiply by
|
|
* @return this
|
|
*/
|
|
public Matrix4f translationRotateScaleMulAffine(float tx, float ty, float tz,
|
|
float qx, float qy, float qz, float qw,
|
|
float sx, float sy, float sz,
|
|
Matrix4f m) {
|
|
float w2 = qw * qw;
|
|
float x2 = qx * qx;
|
|
float y2 = qy * qy;
|
|
float z2 = qz * qz;
|
|
float zw = qz * qw;
|
|
float xy = qx * qy;
|
|
float xz = qx * qz;
|
|
float yw = qy * qw;
|
|
float yz = qy * qz;
|
|
float xw = qx * qw;
|
|
float nm00 = w2 + x2 - z2 - y2;
|
|
float nm01 = xy + zw + zw + xy;
|
|
float nm02 = xz - yw + xz - yw;
|
|
float nm10 = -zw + xy - zw + xy;
|
|
float nm11 = y2 - z2 + w2 - x2;
|
|
float nm12 = yz + yz + xw + xw;
|
|
float nm20 = yw + xz + xz + yw;
|
|
float nm21 = yz + yz - xw - xw;
|
|
float nm22 = z2 - y2 - x2 + w2;
|
|
float m00 = nm00 * m.m00 + nm10 * m.m01 + nm20 * m.m02;
|
|
float m01 = nm01 * m.m00 + nm11 * m.m01 + nm21 * m.m02;
|
|
this._m02(nm02 * m.m00 + nm12 * m.m01 + nm22 * m.m02)
|
|
._m00(m00)
|
|
._m01(m01)
|
|
._m03(0.0f);
|
|
float m10 = nm00 * m.m10 + nm10 * m.m11 + nm20 * m.m12;
|
|
float m11 = nm01 * m.m10 + nm11 * m.m11 + nm21 * m.m12;
|
|
this._m12(nm02 * m.m10 + nm12 * m.m11 + nm22 * m.m12)
|
|
._m10(m10)
|
|
._m11(m11)
|
|
._m13(0.0f);
|
|
float m20 = nm00 * m.m20 + nm10 * m.m21 + nm20 * m.m22;
|
|
float m21 = nm01 * m.m20 + nm11 * m.m21 + nm21 * m.m22;
|
|
this._m22(nm02 * m.m20 + nm12 * m.m21 + nm22 * m.m22)
|
|
._m20(m20)
|
|
._m21(m21)
|
|
._m23(0.0f);
|
|
float m30 = nm00 * m.m30 + nm10 * m.m31 + nm20 * m.m32 + tx;
|
|
float m31 = nm01 * m.m30 + nm11 * m.m31 + nm21 * m.m32 + ty;
|
|
this._m32(nm02 * m.m30 + nm12 * m.m31 + nm22 * m.m32 + tz)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m33(1.0f);
|
|
boolean one = Math.absEqualsOne(sx) && Math.absEqualsOne(sy) && Math.absEqualsOne(sz);
|
|
return _properties(PROPERTY_AFFINE | (one && (m.properties & PROPERTY_ORTHONORMAL) != 0 ? PROPERTY_ORTHONORMAL : 0));
|
|
}
|
|
|
|
/**
|
|
* Set <code>this</code> matrix to <code>T * R * S * M</code>, where <code>T</code> is the given <code>translation</code>,
|
|
* <code>R</code> is a rotation - and possibly scaling - transformation specified by the given quaternion, <code>S</code> is a scaling transformation
|
|
* which scales the axes by <code>scale</code> and <code>M</code> is an {@link #isAffine() affine} matrix.
|
|
* <p>
|
|
* When transforming a vector by the resulting matrix the transformation described by <code>M</code> will be applied first, then the scaling, then rotation and
|
|
* at last the translation.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translation(translation).rotate(quat).scale(scale).mulAffine(m)</code>
|
|
*
|
|
* @see #translation(Vector3fc)
|
|
* @see #rotate(Quaternionfc)
|
|
* @see #mulAffine(Matrix4fc)
|
|
*
|
|
* @param translation
|
|
* the translation
|
|
* @param quat
|
|
* the quaternion representing a rotation
|
|
* @param scale
|
|
* the scaling factors
|
|
* @param m
|
|
* the {@link #isAffine() affine} matrix to multiply by
|
|
* @return this
|
|
*/
|
|
public Matrix4f translationRotateScaleMulAffine(Vector3fc translation,
|
|
Quaternionfc quat,
|
|
Vector3fc scale,
|
|
Matrix4f m) {
|
|
return translationRotateScaleMulAffine(translation.x(), translation.y(), translation.z(), quat.x(), quat.y(), quat.z(), quat.w(), scale.x(), scale.y(), scale.z(), m);
|
|
}
|
|
|
|
/**
|
|
* Set <code>this</code> matrix to <code>T * R</code>, where <code>T</code> is a translation by the given <code>(tx, ty, tz)</code> and
|
|
* <code>R</code> is a rotation - and possibly scaling - transformation specified by the quaternion <code>(qx, qy, qz, qw)</code>.
|
|
* <p>
|
|
* When transforming a vector by the resulting matrix the rotation - and possibly scaling - transformation will be applied first and then the translation.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translation(tx, ty, tz).rotate(quat)</code>
|
|
*
|
|
* @see #translation(float, float, float)
|
|
* @see #rotate(Quaternionfc)
|
|
*
|
|
* @param tx
|
|
* the number of units by which to translate the x-component
|
|
* @param ty
|
|
* the number of units by which to translate the y-component
|
|
* @param tz
|
|
* the number of units by which to translate the z-component
|
|
* @param qx
|
|
* the x-coordinate of the vector part of the quaternion
|
|
* @param qy
|
|
* the y-coordinate of the vector part of the quaternion
|
|
* @param qz
|
|
* the z-coordinate of the vector part of the quaternion
|
|
* @param qw
|
|
* the scalar part of the quaternion
|
|
* @return this
|
|
*/
|
|
public Matrix4f translationRotate(float tx, float ty, float tz, float qx, float qy, float qz, float qw) {
|
|
float w2 = qw * qw;
|
|
float x2 = qx * qx;
|
|
float y2 = qy * qy;
|
|
float z2 = qz * qz;
|
|
float zw = qz * qw;
|
|
float xy = qx * qy;
|
|
float xz = qx * qz;
|
|
float yw = qy * qw;
|
|
float yz = qy * qz;
|
|
float xw = qx * qw;
|
|
return this
|
|
._m00(w2 + x2 - z2 - y2)
|
|
._m01(xy + zw + zw + xy)
|
|
._m02(xz - yw + xz - yw)
|
|
._m10(-zw + xy - zw + xy)
|
|
._m11(y2 - z2 + w2 - x2)
|
|
._m12(yz + yz + xw + xw)
|
|
._m20(yw + xz + xz + yw)
|
|
._m21(yz + yz - xw - xw)
|
|
._m22(z2 - y2 - x2 + w2)
|
|
._m30(tx)
|
|
._m31(ty)
|
|
._m32(tz)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Set <code>this</code> matrix to <code>T * R</code>, where <code>T</code> is a translation by the given <code>(tx, ty, tz)</code> and
|
|
* <code>R</code> is a rotation - and possibly scaling - transformation specified by the given quaternion.
|
|
* <p>
|
|
* When transforming a vector by the resulting matrix the rotation - and possibly scaling - transformation will be applied first and then the translation.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translation(tx, ty, tz).rotate(quat)</code>
|
|
*
|
|
* @see #translation(float, float, float)
|
|
* @see #rotate(Quaternionfc)
|
|
*
|
|
* @param tx
|
|
* the number of units by which to translate the x-component
|
|
* @param ty
|
|
* the number of units by which to translate the y-component
|
|
* @param tz
|
|
* the number of units by which to translate the z-component
|
|
* @param quat
|
|
* the quaternion representing a rotation
|
|
* @return this
|
|
*/
|
|
public Matrix4f translationRotate(float tx, float ty, float tz, Quaternionfc quat) {
|
|
return translationRotate(tx, ty, tz, quat.x(), quat.y(), quat.z(), quat.w());
|
|
}
|
|
|
|
/**
|
|
* Set the upper left 3x3 submatrix of this {@link Matrix4f} to the given {@link Matrix3fc} and don't change the other elements.
|
|
*
|
|
* @param mat
|
|
* the 3x3 matrix
|
|
* @return this
|
|
*/
|
|
public Matrix4f set3x3(Matrix3fc mat) {
|
|
return
|
|
set3x3Matrix3fc(mat).
|
|
_properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
}
|
|
private Matrix4f set3x3Matrix3fc(Matrix3fc mat) {
|
|
return this
|
|
._m00(mat.m00())
|
|
._m01(mat.m01())
|
|
._m02(mat.m02())
|
|
._m10(mat.m10())
|
|
._m11(mat.m11())
|
|
._m12(mat.m12())
|
|
._m20(mat.m20())
|
|
._m21(mat.m21())
|
|
._m22(mat.m22());
|
|
}
|
|
|
|
public Vector4f transform(Vector4f v) {
|
|
return v.mul(this);
|
|
}
|
|
|
|
public Vector4f transform(Vector4fc v, Vector4f dest) {
|
|
return v.mul(this, dest);
|
|
}
|
|
|
|
public Vector4f transform(float x, float y, float z, float w, Vector4f dest) {
|
|
return dest.set(x, y, z, w).mul(this);
|
|
}
|
|
|
|
public Vector4f transformTranspose(Vector4f v) {
|
|
return v.mulTranspose(this);
|
|
}
|
|
public Vector4f transformTranspose(Vector4fc v, Vector4f dest) {
|
|
return v.mulTranspose(this, dest);
|
|
}
|
|
public Vector4f transformTranspose(float x, float y, float z, float w, Vector4f dest) {
|
|
return dest.set(x, y, z, w).mulTranspose(this);
|
|
}
|
|
|
|
public Vector4f transformProject(Vector4f v) {
|
|
return v.mulProject(this);
|
|
}
|
|
|
|
public Vector4f transformProject(Vector4fc v, Vector4f dest) {
|
|
return v.mulProject(this, dest);
|
|
}
|
|
|
|
public Vector4f transformProject(float x, float y, float z, float w, Vector4f dest) {
|
|
return dest.set(x, y, z, w).mulProject(this);
|
|
}
|
|
|
|
public Vector3f transformProject(Vector4fc v, Vector3f dest) {
|
|
return v.mulProject(this, dest);
|
|
}
|
|
|
|
public Vector3f transformProject(float x, float y, float z, float w, Vector3f dest) {
|
|
return dest.set(x, y, z).mulProject(this, w, dest);
|
|
}
|
|
|
|
public Vector3f transformProject(Vector3f v) {
|
|
return v.mulProject(this);
|
|
}
|
|
|
|
public Vector3f transformProject(Vector3fc v, Vector3f dest) {
|
|
return v.mulProject(this, dest);
|
|
}
|
|
|
|
public Vector3f transformProject(float x, float y, float z, Vector3f dest) {
|
|
return dest.set(x, y, z).mulProject(this);
|
|
}
|
|
|
|
public Vector3f transformPosition(Vector3f v) {
|
|
return v.mulPosition(this);
|
|
}
|
|
|
|
public Vector3f transformPosition(Vector3fc v, Vector3f dest) {
|
|
return transformPosition(v.x(), v.y(), v.z(), dest);
|
|
}
|
|
|
|
public Vector3f transformPosition(float x, float y, float z, Vector3f dest) {
|
|
return dest.set(x, y, z).mulPosition(this);
|
|
}
|
|
|
|
public Vector3f transformDirection(Vector3f v) {
|
|
return transformDirection(v.x, v.y, v.z, v);
|
|
}
|
|
|
|
public Vector3f transformDirection(Vector3fc v, Vector3f dest) {
|
|
return transformDirection(v.x(), v.y(), v.z(), dest);
|
|
}
|
|
|
|
public Vector3f transformDirection(float x, float y, float z, Vector3f dest) {
|
|
return dest.set(x, y, z).mulDirection(this);
|
|
}
|
|
|
|
public Vector4f transformAffine(Vector4f v) {
|
|
return v.mulAffine(this, v);
|
|
}
|
|
|
|
public Vector4f transformAffine(Vector4fc v, Vector4f dest) {
|
|
return transformAffine(v.x(), v.y(), v.z(), v.w(), dest);
|
|
}
|
|
|
|
public Vector4f transformAffine(float x, float y, float z, float w, Vector4f dest) {
|
|
return dest.set(x, y, z, w).mulAffine(this, dest);
|
|
}
|
|
|
|
public Matrix4f scale(Vector3fc xyz, Matrix4f dest) {
|
|
return scale(xyz.x(), xyz.y(), xyz.z(), dest);
|
|
}
|
|
|
|
/**
|
|
* Apply scaling to this matrix by scaling the base axes by the given <code>xyz.x</code>,
|
|
* <code>xyz.y</code> and <code>xyz.z</code> factors, respectively.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>S</code> the scaling matrix,
|
|
* then the new matrix will be <code>M * S</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * S * v</code>, the
|
|
* scaling will be applied first!
|
|
*
|
|
* @param xyz
|
|
* the factors of the x, y and z component, respectively
|
|
* @return this
|
|
*/
|
|
public Matrix4f scale(Vector3fc xyz) {
|
|
return scale(xyz.x(), xyz.y(), xyz.z(), this);
|
|
}
|
|
|
|
public Matrix4f scale(float xyz, Matrix4f dest) {
|
|
return scale(xyz, xyz, xyz, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply scaling to this matrix by uniformly scaling all base axes by the given <code>xyz</code> factor.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>S</code> the scaling matrix,
|
|
* then the new matrix will be <code>M * S</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * S * v</code>, the
|
|
* scaling will be applied first!
|
|
* <p>
|
|
* Individual scaling of all three axes can be applied using {@link #scale(float, float, float)}.
|
|
*
|
|
* @see #scale(float, float, float)
|
|
*
|
|
* @param xyz
|
|
* the factor for all components
|
|
* @return this
|
|
*/
|
|
public Matrix4f scale(float xyz) {
|
|
return scale(xyz, xyz, xyz);
|
|
}
|
|
|
|
public Matrix4f scaleXY(float x, float y, Matrix4f dest) {
|
|
return scale(x, y, 1.0f, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply scaling to this matrix by scaling the X axis by <code>x</code> and the Y axis by <code>y</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>S</code> the scaling matrix,
|
|
* then the new matrix will be <code>M * S</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * S * v</code>, the
|
|
* scaling will be applied first!
|
|
*
|
|
* @param x
|
|
* the factor of the x component
|
|
* @param y
|
|
* the factor of the y component
|
|
* @return this
|
|
*/
|
|
public Matrix4f scaleXY(float x, float y) {
|
|
return scale(x, y, 1.0f);
|
|
}
|
|
|
|
public Matrix4f scale(float x, float y, float z, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.scaling(x, y, z);
|
|
return scaleGeneric(x, y, z, dest);
|
|
}
|
|
private Matrix4f scaleGeneric(float x, float y, float z, Matrix4f dest) {
|
|
boolean one = Math.absEqualsOne(x) && Math.absEqualsOne(y) && Math.absEqualsOne(z);
|
|
return dest
|
|
._m00(m00 * x)
|
|
._m01(m01 * x)
|
|
._m02(m02 * x)
|
|
._m03(m03 * x)
|
|
._m10(m10 * y)
|
|
._m11(m11 * y)
|
|
._m12(m12 * y)
|
|
._m13(m13 * y)
|
|
._m20(m20 * z)
|
|
._m21(m21 * z)
|
|
._m22(m22 * z)
|
|
._m23(m23 * z)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION
|
|
| (one ? 0 : PROPERTY_ORTHONORMAL)));
|
|
}
|
|
|
|
/**
|
|
* Apply scaling to this matrix by scaling the base axes by the given sx,
|
|
* sy and sz factors.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>S</code> the scaling matrix,
|
|
* then the new matrix will be <code>M * S</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * S * v</code>, the
|
|
* scaling will be applied first!
|
|
*
|
|
* @param x
|
|
* the factor of the x component
|
|
* @param y
|
|
* the factor of the y component
|
|
* @param z
|
|
* the factor of the z component
|
|
* @return this
|
|
*/
|
|
public Matrix4f scale(float x, float y, float z) {
|
|
return scale(x, y, z, this);
|
|
}
|
|
|
|
public Matrix4f scaleAround(float sx, float sy, float sz, float ox, float oy, float oz, Matrix4f dest) {
|
|
float nm30 = m00 * ox + m10 * oy + m20 * oz + m30;
|
|
float nm31 = m01 * ox + m11 * oy + m21 * oz + m31;
|
|
float nm32 = m02 * ox + m12 * oy + m22 * oz + m32;
|
|
float nm33 = m03 * ox + m13 * oy + m23 * oz + m33;
|
|
boolean one = Math.absEqualsOne(sx) && Math.absEqualsOne(sy) && Math.absEqualsOne(sz);
|
|
return dest
|
|
._m00(m00 * sx)
|
|
._m01(m01 * sx)
|
|
._m02(m02 * sx)
|
|
._m03(m03 * sx)
|
|
._m10(m10 * sy)
|
|
._m11(m11 * sy)
|
|
._m12(m12 * sy)
|
|
._m13(m13 * sy)
|
|
._m20(m20 * sz)
|
|
._m21(m21 * sz)
|
|
._m22(m22 * sz)
|
|
._m23(m23 * sz)
|
|
._m30(-dest.m00 * ox - dest.m10 * oy - dest.m20 * oz + nm30)
|
|
._m31(-dest.m01 * ox - dest.m11 * oy - dest.m21 * oz + nm31)
|
|
._m32(-dest.m02 * ox - dest.m12 * oy - dest.m22 * oz + nm32)
|
|
._m33(-dest.m03 * ox - dest.m13 * oy - dest.m23 * oz + nm33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION
|
|
| (one ? 0 : PROPERTY_ORTHONORMAL)));
|
|
}
|
|
|
|
/**
|
|
* Apply scaling to this matrix by scaling the base axes by the given sx,
|
|
* sy and sz factors while using <code>(ox, oy, oz)</code> as the scaling origin.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>S</code> the scaling matrix,
|
|
* then the new matrix will be <code>M * S</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * S * v</code>, the
|
|
* scaling will be applied first!
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translate(ox, oy, oz).scale(sx, sy, sz).translate(-ox, -oy, -oz)</code>
|
|
*
|
|
* @param sx
|
|
* the scaling factor of the x component
|
|
* @param sy
|
|
* the scaling factor of the y component
|
|
* @param sz
|
|
* the scaling factor of the z component
|
|
* @param ox
|
|
* the x coordinate of the scaling origin
|
|
* @param oy
|
|
* the y coordinate of the scaling origin
|
|
* @param oz
|
|
* the z coordinate of the scaling origin
|
|
* @return this
|
|
*/
|
|
public Matrix4f scaleAround(float sx, float sy, float sz, float ox, float oy, float oz) {
|
|
return scaleAround(sx, sy, sz, ox, oy, oz, this);
|
|
}
|
|
|
|
/**
|
|
* Apply scaling to this matrix by scaling all three base axes by the given <code>factor</code>
|
|
* while using <code>(ox, oy, oz)</code> as the scaling origin.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>S</code> the scaling matrix,
|
|
* then the new matrix will be <code>M * S</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * S * v</code>, the
|
|
* scaling will be applied first!
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translate(ox, oy, oz).scale(factor).translate(-ox, -oy, -oz)</code>
|
|
*
|
|
* @param factor
|
|
* the scaling factor for all three axes
|
|
* @param ox
|
|
* the x coordinate of the scaling origin
|
|
* @param oy
|
|
* the y coordinate of the scaling origin
|
|
* @param oz
|
|
* the z coordinate of the scaling origin
|
|
* @return this
|
|
*/
|
|
public Matrix4f scaleAround(float factor, float ox, float oy, float oz) {
|
|
return scaleAround(factor, factor, factor, ox, oy, oz, this);
|
|
}
|
|
|
|
public Matrix4f scaleAround(float factor, float ox, float oy, float oz, Matrix4f dest) {
|
|
return scaleAround(factor, factor, factor, ox, oy, oz, dest);
|
|
}
|
|
|
|
public Matrix4f scaleLocal(float x, float y, float z, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.scaling(x, y, z);
|
|
return scaleLocalGeneric(x, y, z, dest);
|
|
}
|
|
private Matrix4f scaleLocalGeneric(float x, float y, float z, Matrix4f dest) {
|
|
float nm00 = x * m00;
|
|
float nm01 = y * m01;
|
|
float nm02 = z * m02;
|
|
float nm10 = x * m10;
|
|
float nm11 = y * m11;
|
|
float nm12 = z * m12;
|
|
float nm20 = x * m20;
|
|
float nm21 = y * m21;
|
|
float nm22 = z * m22;
|
|
float nm30 = x * m30;
|
|
float nm31 = y * m31;
|
|
float nm32 = z * m32;
|
|
boolean one = Math.absEqualsOne(x) && Math.absEqualsOne(y) && Math.absEqualsOne(z);
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(m03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(m13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(m23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION
|
|
| (one ? 0 : PROPERTY_ORTHONORMAL)));
|
|
}
|
|
|
|
public Matrix4f scaleLocal(float xyz, Matrix4f dest) {
|
|
return scaleLocal(xyz, xyz, xyz, dest);
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply scaling to this matrix by scaling the base axes by the given xyz factor.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>S</code> the scaling matrix,
|
|
* then the new matrix will be <code>S * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>S * M * v</code>, the
|
|
* scaling will be applied last!
|
|
*
|
|
* @param xyz
|
|
* the factor of the x, y and z component
|
|
* @return this
|
|
*/
|
|
public Matrix4f scaleLocal(float xyz) {
|
|
return scaleLocal(xyz, this);
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply scaling to this matrix by scaling the base axes by the given x,
|
|
* y and z factors.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>S</code> the scaling matrix,
|
|
* then the new matrix will be <code>S * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>S * M * v</code>, the
|
|
* scaling will be applied last!
|
|
*
|
|
* @param x
|
|
* the factor of the x component
|
|
* @param y
|
|
* the factor of the y component
|
|
* @param z
|
|
* the factor of the z component
|
|
* @return this
|
|
*/
|
|
public Matrix4f scaleLocal(float x, float y, float z) {
|
|
return scaleLocal(x, y, z, this);
|
|
}
|
|
|
|
public Matrix4f scaleAroundLocal(float sx, float sy, float sz, float ox, float oy, float oz, Matrix4f dest) {
|
|
boolean one = Math.absEqualsOne(sx) && Math.absEqualsOne(sy) && Math.absEqualsOne(sz);
|
|
return dest
|
|
._m00(sx * (m00 - ox * m03) + ox * m03)
|
|
._m01(sy * (m01 - oy * m03) + oy * m03)
|
|
._m02(sz * (m02 - oz * m03) + oz * m03)
|
|
._m03(m03)
|
|
._m10(sx * (m10 - ox * m13) + ox * m13)
|
|
._m11(sy * (m11 - oy * m13) + oy * m13)
|
|
._m12(sz * (m12 - oz * m13) + oz * m13)
|
|
._m13(m13)
|
|
._m20(sx * (m20 - ox * m23) + ox * m23)
|
|
._m21(sy * (m21 - oy * m23) + oy * m23)
|
|
._m22(sz * (m22 - oz * m23) + oz * m23)
|
|
._m23(m23)
|
|
._m30(sx * (m30 - ox * m33) + ox * m33)
|
|
._m31(sy * (m31 - oy * m33) + oy * m33)
|
|
._m32(sz * (m32 - oz * m33) + oz * m33)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION
|
|
| (one ? 0 : PROPERTY_ORTHONORMAL)));
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply scaling to this matrix by scaling the base axes by the given sx,
|
|
* sy and sz factors while using <code>(ox, oy, oz)</code> as the scaling origin.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>S</code> the scaling matrix,
|
|
* then the new matrix will be <code>S * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>S * M * v</code>, the
|
|
* scaling will be applied last!
|
|
* <p>
|
|
* This method is equivalent to calling: <code>new Matrix4f().translate(ox, oy, oz).scale(sx, sy, sz).translate(-ox, -oy, -oz).mul(this, this)</code>
|
|
*
|
|
* @param sx
|
|
* the scaling factor of the x component
|
|
* @param sy
|
|
* the scaling factor of the y component
|
|
* @param sz
|
|
* the scaling factor of the z component
|
|
* @param ox
|
|
* the x coordinate of the scaling origin
|
|
* @param oy
|
|
* the y coordinate of the scaling origin
|
|
* @param oz
|
|
* the z coordinate of the scaling origin
|
|
* @return this
|
|
*/
|
|
public Matrix4f scaleAroundLocal(float sx, float sy, float sz, float ox, float oy, float oz) {
|
|
return scaleAroundLocal(sx, sy, sz, ox, oy, oz, this);
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply scaling to this matrix by scaling all three base axes by the given <code>factor</code>
|
|
* while using <code>(ox, oy, oz)</code> as the scaling origin.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>S</code> the scaling matrix,
|
|
* then the new matrix will be <code>S * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>S * M * v</code>, the
|
|
* scaling will be applied last!
|
|
* <p>
|
|
* This method is equivalent to calling: <code>new Matrix4f().translate(ox, oy, oz).scale(factor).translate(-ox, -oy, -oz).mul(this, this)</code>
|
|
*
|
|
* @param factor
|
|
* the scaling factor for all three axes
|
|
* @param ox
|
|
* the x coordinate of the scaling origin
|
|
* @param oy
|
|
* the y coordinate of the scaling origin
|
|
* @param oz
|
|
* the z coordinate of the scaling origin
|
|
* @return this
|
|
*/
|
|
public Matrix4f scaleAroundLocal(float factor, float ox, float oy, float oz) {
|
|
return scaleAroundLocal(factor, factor, factor, ox, oy, oz, this);
|
|
}
|
|
|
|
public Matrix4f scaleAroundLocal(float factor, float ox, float oy, float oz, Matrix4f dest) {
|
|
return scaleAroundLocal(factor, factor, factor, ox, oy, oz, dest);
|
|
}
|
|
|
|
public Matrix4f rotateX(float ang, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.rotationX(ang);
|
|
else if ((properties & PROPERTY_TRANSLATION) != 0) {
|
|
float x = m30, y = m31, z = m32;
|
|
return dest.rotationX(ang).setTranslation(x, y, z);
|
|
}
|
|
return rotateXInternal(ang, dest);
|
|
}
|
|
private Matrix4f rotateXInternal(float ang, Matrix4f dest) {
|
|
float sin = Math.sin(ang), cos = Math.cosFromSin(sin, ang);
|
|
float lm10 = m10, lm11 = m11, lm12 = m12, lm13 = m13, lm20 = m20, lm21 = m21, lm22 = m22, lm23 = m23;
|
|
return dest
|
|
._m20(Math.fma(lm10, -sin, lm20 * cos))
|
|
._m21(Math.fma(lm11, -sin, lm21 * cos))
|
|
._m22(Math.fma(lm12, -sin, lm22 * cos))
|
|
._m23(Math.fma(lm13, -sin, lm23 * cos))
|
|
._m10(Math.fma(lm10, cos, lm20 * sin))
|
|
._m11(Math.fma(lm11, cos, lm21 * sin))
|
|
._m12(Math.fma(lm12, cos, lm22 * sin))
|
|
._m13(Math.fma(lm13, cos, lm23 * sin))
|
|
._m00(m00)
|
|
._m01(m01)
|
|
._m02(m02)
|
|
._m03(m03)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply rotation about the X axis to this matrix by rotating the given amount of radians.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Basic_rotations">http://en.wikipedia.org</a>
|
|
*
|
|
* @param ang
|
|
* the angle in radians
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateX(float ang) {
|
|
return rotateX(ang, this);
|
|
}
|
|
|
|
public Matrix4f rotateY(float ang, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.rotationY(ang);
|
|
else if ((properties & PROPERTY_TRANSLATION) != 0) {
|
|
float x = m30, y = m31, z = m32;
|
|
return dest.rotationY(ang).setTranslation(x, y, z);
|
|
}
|
|
return rotateYInternal(ang, dest);
|
|
}
|
|
private Matrix4f rotateYInternal(float ang, Matrix4f dest) {
|
|
float sin = Math.sin(ang);
|
|
float cos = Math.cosFromSin(sin, ang);
|
|
// add temporaries for dependent values
|
|
float nm00 = Math.fma(m00, cos, m20 * -sin);
|
|
float nm01 = Math.fma(m01, cos, m21 * -sin);
|
|
float nm02 = Math.fma(m02, cos, m22 * -sin);
|
|
float nm03 = Math.fma(m03, cos, m23 * -sin);
|
|
// set non-dependent values directly
|
|
return dest
|
|
._m20(Math.fma(m00, sin, m20 * cos))
|
|
._m21(Math.fma(m01, sin, m21 * cos))
|
|
._m22(Math.fma(m02, sin, m22 * cos))
|
|
._m23(Math.fma(m03, sin, m23 * cos))
|
|
// set other values
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(m10)
|
|
._m11(m11)
|
|
._m12(m12)
|
|
._m13(m13)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply rotation about the Y axis to this matrix by rotating the given amount of radians.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Basic_rotations">http://en.wikipedia.org</a>
|
|
*
|
|
* @param ang
|
|
* the angle in radians
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateY(float ang) {
|
|
return rotateY(ang, this);
|
|
}
|
|
|
|
public Matrix4f rotateZ(float ang, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.rotationZ(ang);
|
|
else if ((properties & PROPERTY_TRANSLATION) != 0) {
|
|
float x = m30, y = m31, z = m32;
|
|
return dest.rotationZ(ang).setTranslation(x, y, z);
|
|
}
|
|
return rotateZInternal(ang, dest);
|
|
}
|
|
private Matrix4f rotateZInternal(float ang, Matrix4f dest) {
|
|
float sin = Math.sin(ang);
|
|
float cos = Math.cosFromSin(sin, ang);
|
|
return rotateTowardsXY(sin, cos, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply rotation about the Z axis to this matrix by rotating the given amount of radians.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Basic_rotations">http://en.wikipedia.org</a>
|
|
*
|
|
* @param ang
|
|
* the angle in radians
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateZ(float ang) {
|
|
return rotateZ(ang, this);
|
|
}
|
|
|
|
/**
|
|
* Apply rotation about the Z axis to align the local <code>+X</code> towards <code>(dirX, dirY)</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* The vector <code>(dirX, dirY)</code> must be a unit vector.
|
|
*
|
|
* @param dirX
|
|
* the x component of the normalized direction
|
|
* @param dirY
|
|
* the y component of the normalized direction
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateTowardsXY(float dirX, float dirY) {
|
|
return rotateTowardsXY(dirX, dirY, this);
|
|
}
|
|
|
|
public Matrix4f rotateTowardsXY(float dirX, float dirY, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.rotationTowardsXY(dirX, dirY);
|
|
float nm00 = Math.fma(m00, dirY, m10 * dirX);
|
|
float nm01 = Math.fma(m01, dirY, m11 * dirX);
|
|
float nm02 = Math.fma(m02, dirY, m12 * dirX);
|
|
float nm03 = Math.fma(m03, dirY, m13 * dirX);
|
|
return dest
|
|
._m10(Math.fma(m00, -dirX, m10 * dirY))
|
|
._m11(Math.fma(m01, -dirX, m11 * dirY))
|
|
._m12(Math.fma(m02, -dirX, m12 * dirY))
|
|
._m13(Math.fma(m03, -dirX, m13 * dirY))
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m20(m20)
|
|
._m21(m21)
|
|
._m22(m22)
|
|
._m23(m23)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply rotation of <code>angles.x</code> radians about the X axis, followed by a rotation of <code>angles.y</code> radians about the Y axis and
|
|
* followed by a rotation of <code>angles.z</code> radians about the Z axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* This method is equivalent to calling: <code>rotateX(angles.x()).rotateY(angles.y()).rotateZ(angles.z())</code>
|
|
*
|
|
* @param angles
|
|
* the Euler angles
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateXYZ(Vector3fc angles) {
|
|
return rotateXYZ(angles.x(), angles.y(), angles.z());
|
|
}
|
|
|
|
/**
|
|
* Apply rotation of <code>angleX</code> radians about the X axis, followed by a rotation of <code>angleY</code> radians about the Y axis and
|
|
* followed by a rotation of <code>angleZ</code> radians about the Z axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* This method is equivalent to calling: <code>rotateX(angleX).rotateY(angleY).rotateZ(angleZ)</code>
|
|
*
|
|
* @param angleX
|
|
* the angle to rotate about X
|
|
* @param angleY
|
|
* the angle to rotate about Y
|
|
* @param angleZ
|
|
* the angle to rotate about Z
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateXYZ(float angleX, float angleY, float angleZ) {
|
|
return rotateXYZ(angleX, angleY, angleZ, this);
|
|
}
|
|
|
|
public Matrix4f rotateXYZ(float angleX, float angleY, float angleZ, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.rotationXYZ(angleX, angleY, angleZ);
|
|
else if ((properties & PROPERTY_TRANSLATION) != 0) {
|
|
float tx = m30, ty = m31, tz = m32;
|
|
return dest.rotationXYZ(angleX, angleY, angleZ).setTranslation(tx, ty, tz);
|
|
} else if ((properties & PROPERTY_AFFINE) != 0)
|
|
return dest.rotateAffineXYZ(angleX, angleY, angleZ);
|
|
return rotateXYZInternal(angleX, angleY, angleZ, dest);
|
|
}
|
|
private Matrix4f rotateXYZInternal(float angleX, float angleY, float angleZ, Matrix4f dest) {
|
|
float sinX = Math.sin(angleX);
|
|
float cosX = Math.cosFromSin(sinX, angleX);
|
|
float sinY = Math.sin(angleY);
|
|
float cosY = Math.cosFromSin(sinY, angleY);
|
|
float sinZ = Math.sin(angleZ);
|
|
float cosZ = Math.cosFromSin(sinZ, angleZ);
|
|
float m_sinX = -sinX;
|
|
float m_sinY = -sinY;
|
|
float m_sinZ = -sinZ;
|
|
|
|
// rotateX
|
|
float nm10 = Math.fma(m10, cosX, m20 * sinX);
|
|
float nm11 = Math.fma(m11, cosX, m21 * sinX);
|
|
float nm12 = Math.fma(m12, cosX, m22 * sinX);
|
|
float nm13 = Math.fma(m13, cosX, m23 * sinX);
|
|
float nm20 = Math.fma(m10, m_sinX, m20 * cosX);
|
|
float nm21 = Math.fma(m11, m_sinX, m21 * cosX);
|
|
float nm22 = Math.fma(m12, m_sinX, m22 * cosX);
|
|
float nm23 = Math.fma(m13, m_sinX, m23 * cosX);
|
|
// rotateY
|
|
float nm00 = Math.fma(m00, cosY, nm20 * m_sinY);
|
|
float nm01 = Math.fma(m01, cosY, nm21 * m_sinY);
|
|
float nm02 = Math.fma(m02, cosY, nm22 * m_sinY);
|
|
float nm03 = Math.fma(m03, cosY, nm23 * m_sinY);
|
|
return dest
|
|
._m20(Math.fma(m00, sinY, nm20 * cosY))
|
|
._m21(Math.fma(m01, sinY, nm21 * cosY))
|
|
._m22(Math.fma(m02, sinY, nm22 * cosY))
|
|
._m23(Math.fma(m03, sinY, nm23 * cosY))
|
|
// rotateZ
|
|
._m00(Math.fma(nm00, cosZ, nm10 * sinZ))
|
|
._m01(Math.fma(nm01, cosZ, nm11 * sinZ))
|
|
._m02(Math.fma(nm02, cosZ, nm12 * sinZ))
|
|
._m03(Math.fma(nm03, cosZ, nm13 * sinZ))
|
|
._m10(Math.fma(nm00, m_sinZ, nm10 * cosZ))
|
|
._m11(Math.fma(nm01, m_sinZ, nm11 * cosZ))
|
|
._m12(Math.fma(nm02, m_sinZ, nm12 * cosZ))
|
|
._m13(Math.fma(nm03, m_sinZ, nm13 * cosZ))
|
|
// copy last column from 'this'
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply rotation of <code>angleX</code> radians about the X axis, followed by a rotation of <code>angleY</code> radians about the Y axis and
|
|
* followed by a rotation of <code>angleZ</code> radians about the Z axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method assumes that <code>this</code> matrix represents an {@link #isAffine() affine} transformation (i.e. its last row is equal to <code>(0, 0, 0, 1)</code>)
|
|
* and can be used to speed up matrix multiplication if the matrix only represents affine transformations, such as translation, rotation, scaling and shearing (in any combination).
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* This method is equivalent to calling: <code>rotateX(angleX).rotateY(angleY).rotateZ(angleZ)</code>
|
|
*
|
|
* @param angleX
|
|
* the angle to rotate about X
|
|
* @param angleY
|
|
* the angle to rotate about Y
|
|
* @param angleZ
|
|
* the angle to rotate about Z
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateAffineXYZ(float angleX, float angleY, float angleZ) {
|
|
return rotateAffineXYZ(angleX, angleY, angleZ, this);
|
|
}
|
|
|
|
public Matrix4f rotateAffineXYZ(float angleX, float angleY, float angleZ, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.rotationXYZ(angleX, angleY, angleZ);
|
|
else if ((properties & PROPERTY_TRANSLATION) != 0) {
|
|
float tx = m30, ty = m31, tz = m32;
|
|
return dest.rotationXYZ(angleX, angleY, angleZ).setTranslation(tx, ty, tz);
|
|
}
|
|
return rotateAffineXYZInternal(angleX, angleY, angleZ, dest);
|
|
}
|
|
private Matrix4f rotateAffineXYZInternal(float angleX, float angleY, float angleZ, Matrix4f dest) {
|
|
float sinX = Math.sin(angleX);
|
|
float cosX = Math.cosFromSin(sinX, angleX);
|
|
float sinY = Math.sin(angleY);
|
|
float cosY = Math.cosFromSin(sinY, angleY);
|
|
float sinZ = Math.sin(angleZ);
|
|
float cosZ = Math.cosFromSin(sinZ, angleZ);
|
|
float m_sinX = -sinX;
|
|
float m_sinY = -sinY;
|
|
float m_sinZ = -sinZ;
|
|
|
|
// rotateX
|
|
float nm10 = Math.fma(m10, cosX, m20 * sinX);
|
|
float nm11 = Math.fma(m11, cosX, m21 * sinX);
|
|
float nm12 = Math.fma(m12, cosX, m22 * sinX);
|
|
float nm20 = Math.fma(m10, m_sinX, m20 * cosX);
|
|
float nm21 = Math.fma(m11, m_sinX, m21 * cosX);
|
|
float nm22 = Math.fma(m12, m_sinX, m22 * cosX);
|
|
// rotateY
|
|
float nm00 = Math.fma(m00, cosY, nm20 * m_sinY);
|
|
float nm01 = Math.fma(m01, cosY, nm21 * m_sinY);
|
|
float nm02 = Math.fma(m02, cosY, nm22 * m_sinY);
|
|
return dest
|
|
._m20(Math.fma(m00, sinY, nm20 * cosY))
|
|
._m21(Math.fma(m01, sinY, nm21 * cosY))
|
|
._m22(Math.fma(m02, sinY, nm22 * cosY))
|
|
._m23(0.0f)
|
|
// rotateZ
|
|
._m00(Math.fma(nm00, cosZ, nm10 * sinZ))
|
|
._m01(Math.fma(nm01, cosZ, nm11 * sinZ))
|
|
._m02(Math.fma(nm02, cosZ, nm12 * sinZ))
|
|
._m03(0.0f)
|
|
._m10(Math.fma(nm00, m_sinZ, nm10 * cosZ))
|
|
._m11(Math.fma(nm01, m_sinZ, nm11 * cosZ))
|
|
._m12(Math.fma(nm02, m_sinZ, nm12 * cosZ))
|
|
._m13(0.0f)
|
|
// copy last column from 'this'
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply rotation of <code>angles.z</code> radians about the Z axis, followed by a rotation of <code>angles.y</code> radians about the Y axis and
|
|
* followed by a rotation of <code>angles.x</code> radians about the X axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* This method is equivalent to calling: <code>rotateZ(angles.z).rotateY(angles.y).rotateX(angles.x)</code>
|
|
*
|
|
* @param angles
|
|
* the Euler angles
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateZYX(Vector3f angles) {
|
|
return rotateZYX(angles.z, angles.y, angles.x);
|
|
}
|
|
|
|
/**
|
|
* Apply rotation of <code>angleZ</code> radians about the Z axis, followed by a rotation of <code>angleY</code> radians about the Y axis and
|
|
* followed by a rotation of <code>angleX</code> radians about the X axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* This method is equivalent to calling: <code>rotateZ(angleZ).rotateY(angleY).rotateX(angleX)</code>
|
|
*
|
|
* @param angleZ
|
|
* the angle to rotate about Z
|
|
* @param angleY
|
|
* the angle to rotate about Y
|
|
* @param angleX
|
|
* the angle to rotate about X
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateZYX(float angleZ, float angleY, float angleX) {
|
|
return rotateZYX(angleZ, angleY, angleX, this);
|
|
}
|
|
|
|
public Matrix4f rotateZYX(float angleZ, float angleY, float angleX, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.rotationZYX(angleZ, angleY, angleX);
|
|
else if ((properties & PROPERTY_TRANSLATION) != 0) {
|
|
float tx = m30, ty = m31, tz = m32;
|
|
return dest.rotationZYX(angleZ, angleY, angleX).setTranslation(tx, ty, tz);
|
|
} else if ((properties & PROPERTY_AFFINE) != 0)
|
|
return dest.rotateAffineZYX(angleZ, angleY, angleX);
|
|
return rotateZYXInternal(angleZ, angleY, angleX, dest);
|
|
}
|
|
private Matrix4f rotateZYXInternal(float angleZ, float angleY, float angleX, Matrix4f dest) {
|
|
float sinX = Math.sin(angleX);
|
|
float cosX = Math.cosFromSin(sinX, angleX);
|
|
float sinY = Math.sin(angleY);
|
|
float cosY = Math.cosFromSin(sinY, angleY);
|
|
float sinZ = Math.sin(angleZ);
|
|
float cosZ = Math.cosFromSin(sinZ, angleZ);
|
|
float m_sinZ = -sinZ;
|
|
float m_sinY = -sinY;
|
|
float m_sinX = -sinX;
|
|
|
|
// rotateZ
|
|
float nm00 = m00 * cosZ + m10 * sinZ;
|
|
float nm01 = m01 * cosZ + m11 * sinZ;
|
|
float nm02 = m02 * cosZ + m12 * sinZ;
|
|
float nm03 = m03 * cosZ + m13 * sinZ;
|
|
float nm10 = m00 * m_sinZ + m10 * cosZ;
|
|
float nm11 = m01 * m_sinZ + m11 * cosZ;
|
|
float nm12 = m02 * m_sinZ + m12 * cosZ;
|
|
float nm13 = m03 * m_sinZ + m13 * cosZ;
|
|
// rotateY
|
|
float nm20 = nm00 * sinY + m20 * cosY;
|
|
float nm21 = nm01 * sinY + m21 * cosY;
|
|
float nm22 = nm02 * sinY + m22 * cosY;
|
|
float nm23 = nm03 * sinY + m23 * cosY;
|
|
return dest
|
|
._m00(nm00 * cosY + m20 * m_sinY)
|
|
._m01(nm01 * cosY + m21 * m_sinY)
|
|
._m02(nm02 * cosY + m22 * m_sinY)
|
|
._m03(nm03 * cosY + m23 * m_sinY)
|
|
._m10(nm10 * cosX + nm20 * sinX)
|
|
._m11(nm11 * cosX + nm21 * sinX)
|
|
._m12(nm12 * cosX + nm22 * sinX)
|
|
._m13(nm13 * cosX + nm23 * sinX)
|
|
._m20(nm10 * m_sinX + nm20 * cosX)
|
|
._m21(nm11 * m_sinX + nm21 * cosX)
|
|
._m22(nm12 * m_sinX + nm22 * cosX)
|
|
._m23(nm13 * m_sinX + nm23 * cosX)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply rotation of <code>angleZ</code> radians about the Z axis, followed by a rotation of <code>angleY</code> radians about the Y axis and
|
|
* followed by a rotation of <code>angleX</code> radians about the X axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method assumes that <code>this</code> matrix represents an {@link #isAffine() affine} transformation (i.e. its last row is equal to <code>(0, 0, 0, 1)</code>)
|
|
* and can be used to speed up matrix multiplication if the matrix only represents affine transformations, such as translation, rotation, scaling and shearing (in any combination).
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
*
|
|
* @param angleZ
|
|
* the angle to rotate about Z
|
|
* @param angleY
|
|
* the angle to rotate about Y
|
|
* @param angleX
|
|
* the angle to rotate about X
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateAffineZYX(float angleZ, float angleY, float angleX) {
|
|
return rotateAffineZYX(angleZ, angleY, angleX, this);
|
|
}
|
|
|
|
public Matrix4f rotateAffineZYX(float angleZ, float angleY, float angleX, Matrix4f dest) {
|
|
float sinX = Math.sin(angleX);
|
|
float cosX = Math.cosFromSin(sinX, angleX);
|
|
float sinY = Math.sin(angleY);
|
|
float cosY = Math.cosFromSin(sinY, angleY);
|
|
float sinZ = Math.sin(angleZ);
|
|
float cosZ = Math.cosFromSin(sinZ, angleZ);
|
|
float m_sinZ = -sinZ;
|
|
float m_sinY = -sinY;
|
|
float m_sinX = -sinX;
|
|
|
|
// rotateZ
|
|
float nm00 = m00 * cosZ + m10 * sinZ;
|
|
float nm01 = m01 * cosZ + m11 * sinZ;
|
|
float nm02 = m02 * cosZ + m12 * sinZ;
|
|
float nm10 = m00 * m_sinZ + m10 * cosZ;
|
|
float nm11 = m01 * m_sinZ + m11 * cosZ;
|
|
float nm12 = m02 * m_sinZ + m12 * cosZ;
|
|
// rotateY
|
|
float nm20 = nm00 * sinY + m20 * cosY;
|
|
float nm21 = nm01 * sinY + m21 * cosY;
|
|
float nm22 = nm02 * sinY + m22 * cosY;
|
|
return dest
|
|
._m00(nm00 * cosY + m20 * m_sinY)
|
|
._m01(nm01 * cosY + m21 * m_sinY)
|
|
._m02(nm02 * cosY + m22 * m_sinY)
|
|
._m03(0.0f)
|
|
._m10(nm10 * cosX + nm20 * sinX)
|
|
._m11(nm11 * cosX + nm21 * sinX)
|
|
._m12(nm12 * cosX + nm22 * sinX)
|
|
._m13(0.0f)
|
|
._m20(nm10 * m_sinX + nm20 * cosX)
|
|
._m21(nm11 * m_sinX + nm21 * cosX)
|
|
._m22(nm12 * m_sinX + nm22 * cosX)
|
|
._m23(0.0f)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply rotation of <code>angles.y</code> radians about the Y axis, followed by a rotation of <code>angles.x</code> radians about the X axis and
|
|
* followed by a rotation of <code>angles.z</code> radians about the Z axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* This method is equivalent to calling: <code>rotateY(angles.y).rotateX(angles.x).rotateZ(angles.z)</code>
|
|
*
|
|
* @param angles
|
|
* the Euler angles
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateYXZ(Vector3f angles) {
|
|
return rotateYXZ(angles.y, angles.x, angles.z);
|
|
}
|
|
|
|
/**
|
|
* Apply rotation of <code>angleY</code> radians about the Y axis, followed by a rotation of <code>angleX</code> radians about the X axis and
|
|
* followed by a rotation of <code>angleZ</code> radians about the Z axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* This method is equivalent to calling: <code>rotateY(angleY).rotateX(angleX).rotateZ(angleZ)</code>
|
|
*
|
|
* @param angleY
|
|
* the angle to rotate about Y
|
|
* @param angleX
|
|
* the angle to rotate about X
|
|
* @param angleZ
|
|
* the angle to rotate about Z
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateYXZ(float angleY, float angleX, float angleZ) {
|
|
return rotateYXZ(angleY, angleX, angleZ, this);
|
|
}
|
|
|
|
public Matrix4f rotateYXZ(float angleY, float angleX, float angleZ, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.rotationYXZ(angleY, angleX, angleZ);
|
|
else if ((properties & PROPERTY_TRANSLATION) != 0) {
|
|
float tx = m30, ty = m31, tz = m32;
|
|
return dest.rotationYXZ(angleY, angleX, angleZ).setTranslation(tx, ty, tz);
|
|
} else if ((properties & PROPERTY_AFFINE) != 0)
|
|
return dest.rotateAffineYXZ(angleY, angleX, angleZ);
|
|
return rotateYXZInternal(angleY, angleX, angleZ, dest);
|
|
}
|
|
private Matrix4f rotateYXZInternal(float angleY, float angleX, float angleZ, Matrix4f dest) {
|
|
float sinX = Math.sin(angleX);
|
|
float cosX = Math.cosFromSin(sinX, angleX);
|
|
float sinY = Math.sin(angleY);
|
|
float cosY = Math.cosFromSin(sinY, angleY);
|
|
float sinZ = Math.sin(angleZ);
|
|
float cosZ = Math.cosFromSin(sinZ, angleZ);
|
|
float m_sinY = -sinY;
|
|
float m_sinX = -sinX;
|
|
float m_sinZ = -sinZ;
|
|
|
|
// rotateY
|
|
float nm20 = m00 * sinY + m20 * cosY;
|
|
float nm21 = m01 * sinY + m21 * cosY;
|
|
float nm22 = m02 * sinY + m22 * cosY;
|
|
float nm23 = m03 * sinY + m23 * cosY;
|
|
float nm00 = m00 * cosY + m20 * m_sinY;
|
|
float nm01 = m01 * cosY + m21 * m_sinY;
|
|
float nm02 = m02 * cosY + m22 * m_sinY;
|
|
float nm03 = m03 * cosY + m23 * m_sinY;
|
|
// rotateX
|
|
float nm10 = m10 * cosX + nm20 * sinX;
|
|
float nm11 = m11 * cosX + nm21 * sinX;
|
|
float nm12 = m12 * cosX + nm22 * sinX;
|
|
float nm13 = m13 * cosX + nm23 * sinX;
|
|
return dest
|
|
._m20(m10 * m_sinX + nm20 * cosX)
|
|
._m21(m11 * m_sinX + nm21 * cosX)
|
|
._m22(m12 * m_sinX + nm22 * cosX)
|
|
._m23(m13 * m_sinX + nm23 * cosX)
|
|
._m00(nm00 * cosZ + nm10 * sinZ)
|
|
._m01(nm01 * cosZ + nm11 * sinZ)
|
|
._m02(nm02 * cosZ + nm12 * sinZ)
|
|
._m03(nm03 * cosZ + nm13 * sinZ)
|
|
._m10(nm00 * m_sinZ + nm10 * cosZ)
|
|
._m11(nm01 * m_sinZ + nm11 * cosZ)
|
|
._m12(nm02 * m_sinZ + nm12 * cosZ)
|
|
._m13(nm03 * m_sinZ + nm13 * cosZ)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply rotation of <code>angleY</code> radians about the Y axis, followed by a rotation of <code>angleX</code> radians about the X axis and
|
|
* followed by a rotation of <code>angleZ</code> radians about the Z axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method assumes that <code>this</code> matrix represents an {@link #isAffine() affine} transformation (i.e. its last row is equal to <code>(0, 0, 0, 1)</code>)
|
|
* and can be used to speed up matrix multiplication if the matrix only represents affine transformations, such as translation, rotation, scaling and shearing (in any combination).
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
*
|
|
* @param angleY
|
|
* the angle to rotate about Y
|
|
* @param angleX
|
|
* the angle to rotate about X
|
|
* @param angleZ
|
|
* the angle to rotate about Z
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateAffineYXZ(float angleY, float angleX, float angleZ) {
|
|
return rotateAffineYXZ(angleY, angleX, angleZ, this);
|
|
}
|
|
|
|
public Matrix4f rotateAffineYXZ(float angleY, float angleX, float angleZ, Matrix4f dest) {
|
|
float sinX = Math.sin(angleX);
|
|
float cosX = Math.cosFromSin(sinX, angleX);
|
|
float sinY = Math.sin(angleY);
|
|
float cosY = Math.cosFromSin(sinY, angleY);
|
|
float sinZ = Math.sin(angleZ);
|
|
float cosZ = Math.cosFromSin(sinZ, angleZ);
|
|
float m_sinY = -sinY;
|
|
float m_sinX = -sinX;
|
|
float m_sinZ = -sinZ;
|
|
|
|
// rotateY
|
|
float nm20 = m00 * sinY + m20 * cosY;
|
|
float nm21 = m01 * sinY + m21 * cosY;
|
|
float nm22 = m02 * sinY + m22 * cosY;
|
|
float nm00 = m00 * cosY + m20 * m_sinY;
|
|
float nm01 = m01 * cosY + m21 * m_sinY;
|
|
float nm02 = m02 * cosY + m22 * m_sinY;
|
|
// rotateX
|
|
float nm10 = m10 * cosX + nm20 * sinX;
|
|
float nm11 = m11 * cosX + nm21 * sinX;
|
|
float nm12 = m12 * cosX + nm22 * sinX;
|
|
return dest
|
|
._m20(m10 * m_sinX + nm20 * cosX)
|
|
._m21(m11 * m_sinX + nm21 * cosX)
|
|
._m22(m12 * m_sinX + nm22 * cosX)
|
|
._m23(0.0f)
|
|
._m00(nm00 * cosZ + nm10 * sinZ)
|
|
._m01(nm01 * cosZ + nm11 * sinZ)
|
|
._m02(nm02 * cosZ + nm12 * sinZ)
|
|
._m03(0.0f)
|
|
._m10(nm00 * m_sinZ + nm10 * cosZ)
|
|
._m11(nm01 * m_sinZ + nm11 * cosZ)
|
|
._m12(nm02 * m_sinZ + nm12 * cosZ)
|
|
._m13(0.0f)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply rotation to this matrix by rotating the given amount of radians
|
|
* about the specified <code>(x, y, z)</code> axis and store the result in <code>dest</code>.
|
|
* <p>
|
|
* The axis described by the three components needs to be a unit vector.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation matrix without post-multiplying the rotation
|
|
* transformation, use {@link #rotation(float, float, float, float) rotation()}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Rotation_matrix_from_axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotation(float, float, float, float)
|
|
*
|
|
* @param ang
|
|
* the angle in radians
|
|
* @param x
|
|
* the x component of the axis
|
|
* @param y
|
|
* the y component of the axis
|
|
* @param z
|
|
* the z component of the axis
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotate(float ang, float x, float y, float z, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.rotation(ang, x, y, z);
|
|
else if ((properties & PROPERTY_TRANSLATION) != 0)
|
|
return rotateTranslation(ang, x, y, z, dest);
|
|
else if ((properties & PROPERTY_AFFINE) != 0)
|
|
return rotateAffine(ang, x, y, z, dest);
|
|
return rotateGeneric(ang, x, y, z, dest);
|
|
}
|
|
private Matrix4f rotateGeneric(float ang, float x, float y, float z, Matrix4f dest) {
|
|
if (y == 0.0f && z == 0.0f && Math.absEqualsOne(x))
|
|
return rotateX(x * ang, dest);
|
|
else if (x == 0.0f && z == 0.0f && Math.absEqualsOne(y))
|
|
return rotateY(y * ang, dest);
|
|
else if (x == 0.0f && y == 0.0f && Math.absEqualsOne(z))
|
|
return rotateZ(z * ang, dest);
|
|
return rotateGenericInternal(ang, x, y, z, dest);
|
|
}
|
|
private Matrix4f rotateGenericInternal(float ang, float x, float y, float z, Matrix4f dest) {
|
|
float s = Math.sin(ang);
|
|
float c = Math.cosFromSin(s, ang);
|
|
float C = 1.0f - c;
|
|
float xx = x * x, xy = x * y, xz = x * z;
|
|
float yy = y * y, yz = y * z;
|
|
float zz = z * z;
|
|
float rm00 = xx * C + c;
|
|
float rm01 = xy * C + z * s;
|
|
float rm02 = xz * C - y * s;
|
|
float rm10 = xy * C - z * s;
|
|
float rm11 = yy * C + c;
|
|
float rm12 = yz * C + x * s;
|
|
float rm20 = xz * C + y * s;
|
|
float rm21 = yz * C - x * s;
|
|
float rm22 = zz * C + c;
|
|
float nm00 = m00 * rm00 + m10 * rm01 + m20 * rm02;
|
|
float nm01 = m01 * rm00 + m11 * rm01 + m21 * rm02;
|
|
float nm02 = m02 * rm00 + m12 * rm01 + m22 * rm02;
|
|
float nm03 = m03 * rm00 + m13 * rm01 + m23 * rm02;
|
|
float nm10 = m00 * rm10 + m10 * rm11 + m20 * rm12;
|
|
float nm11 = m01 * rm10 + m11 * rm11 + m21 * rm12;
|
|
float nm12 = m02 * rm10 + m12 * rm11 + m22 * rm12;
|
|
float nm13 = m03 * rm10 + m13 * rm11 + m23 * rm12;
|
|
return dest
|
|
._m20(m00 * rm20 + m10 * rm21 + m20 * rm22)
|
|
._m21(m01 * rm20 + m11 * rm21 + m21 * rm22)
|
|
._m22(m02 * rm20 + m12 * rm21 + m22 * rm22)
|
|
._m23(m03 * rm20 + m13 * rm21 + m23 * rm22)
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply rotation to this matrix by rotating the given amount of radians
|
|
* about the specified <code>(x, y, z)</code> axis.
|
|
* <p>
|
|
* The axis described by the three components needs to be a unit vector.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation matrix without post-multiplying the rotation
|
|
* transformation, use {@link #rotation(float, float, float, float) rotation()}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Rotation_matrix_from_axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotation(float, float, float, float)
|
|
*
|
|
* @param ang
|
|
* the angle in radians
|
|
* @param x
|
|
* the x component of the axis
|
|
* @param y
|
|
* the y component of the axis
|
|
* @param z
|
|
* the z component of the axis
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotate(float ang, float x, float y, float z) {
|
|
return rotate(ang, x, y, z, this);
|
|
}
|
|
|
|
/**
|
|
* Apply rotation to this matrix, which is assumed to only contain a translation, by rotating the given amount of radians
|
|
* about the specified <code>(x, y, z)</code> axis and store the result in <code>dest</code>.
|
|
* <p>
|
|
* This method assumes <code>this</code> to only contain a translation.
|
|
* <p>
|
|
* The axis described by the three components needs to be a unit vector.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation matrix without post-multiplying the rotation
|
|
* transformation, use {@link #rotation(float, float, float, float) rotation()}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Rotation_matrix_from_axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotation(float, float, float, float)
|
|
*
|
|
* @param ang
|
|
* the angle in radians
|
|
* @param x
|
|
* the x component of the axis
|
|
* @param y
|
|
* the y component of the axis
|
|
* @param z
|
|
* the z component of the axis
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotateTranslation(float ang, float x, float y, float z, Matrix4f dest) {
|
|
float tx = m30, ty = m31, tz = m32;
|
|
if (y == 0.0f && z == 0.0f && Math.absEqualsOne(x))
|
|
return dest.rotationX(x * ang).setTranslation(tx, ty, tz);
|
|
else if (x == 0.0f && z == 0.0f && Math.absEqualsOne(y))
|
|
return dest.rotationY(y * ang).setTranslation(tx, ty, tz);
|
|
else if (x == 0.0f && y == 0.0f && Math.absEqualsOne(z))
|
|
return dest.rotationZ(z * ang).setTranslation(tx, ty, tz);
|
|
return rotateTranslationInternal(ang, x, y, z, dest);
|
|
}
|
|
private Matrix4f rotateTranslationInternal(float ang, float x, float y, float z, Matrix4f dest) {
|
|
float s = Math.sin(ang);
|
|
float c = Math.cosFromSin(s, ang);
|
|
float C = 1.0f - c;
|
|
float xx = x * x, xy = x * y, xz = x * z;
|
|
float yy = y * y, yz = y * z;
|
|
float zz = z * z;
|
|
float rm00 = xx * C + c;
|
|
float rm01 = xy * C + z * s;
|
|
float rm02 = xz * C - y * s;
|
|
float rm10 = xy * C - z * s;
|
|
float rm11 = yy * C + c;
|
|
float rm12 = yz * C + x * s;
|
|
float rm20 = xz * C + y * s;
|
|
float rm21 = yz * C - x * s;
|
|
float rm22 = zz * C + c;
|
|
return dest
|
|
._m20(rm20)
|
|
._m21(rm21)
|
|
._m22(rm22)
|
|
._m00(rm00)
|
|
._m01(rm01)
|
|
._m02(rm02)
|
|
._m03(0.0f)
|
|
._m10(rm10)
|
|
._m11(rm11)
|
|
._m12(rm12)
|
|
._m13(0.0f)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(1.0f)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply rotation to this {@link #isAffine() affine} matrix by rotating the given amount of radians
|
|
* about the specified <code>(x, y, z)</code> axis and store the result in <code>dest</code>.
|
|
* <p>
|
|
* This method assumes <code>this</code> to be {@link #isAffine() affine}.
|
|
* <p>
|
|
* The axis described by the three components needs to be a unit vector.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation matrix without post-multiplying the rotation
|
|
* transformation, use {@link #rotation(float, float, float, float) rotation()}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Rotation_matrix_from_axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotation(float, float, float, float)
|
|
*
|
|
* @param ang
|
|
* the angle in radians
|
|
* @param x
|
|
* the x component of the axis
|
|
* @param y
|
|
* the y component of the axis
|
|
* @param z
|
|
* the z component of the axis
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotateAffine(float ang, float x, float y, float z, Matrix4f dest) {
|
|
if (y == 0.0f && z == 0.0f && Math.absEqualsOne(x))
|
|
return rotateX(x * ang, dest);
|
|
else if (x == 0.0f && z == 0.0f && Math.absEqualsOne(y))
|
|
return rotateY(y * ang, dest);
|
|
else if (x == 0.0f && y == 0.0f && Math.absEqualsOne(z))
|
|
return rotateZ(z * ang, dest);
|
|
return rotateAffineInternal(ang, x, y, z, dest);
|
|
}
|
|
private Matrix4f rotateAffineInternal(float ang, float x, float y, float z, Matrix4f dest) {
|
|
float s = Math.sin(ang);
|
|
float c = Math.cosFromSin(s, ang);
|
|
float C = 1.0f - c;
|
|
float xx = x * x, xy = x * y, xz = x * z;
|
|
float yy = y * y, yz = y * z;
|
|
float zz = z * z;
|
|
float rm00 = xx * C + c;
|
|
float rm01 = xy * C + z * s;
|
|
float rm02 = xz * C - y * s;
|
|
float rm10 = xy * C - z * s;
|
|
float rm11 = yy * C + c;
|
|
float rm12 = yz * C + x * s;
|
|
float rm20 = xz * C + y * s;
|
|
float rm21 = yz * C - x * s;
|
|
float rm22 = zz * C + c;
|
|
// add temporaries for dependent values
|
|
float nm00 = m00 * rm00 + m10 * rm01 + m20 * rm02;
|
|
float nm01 = m01 * rm00 + m11 * rm01 + m21 * rm02;
|
|
float nm02 = m02 * rm00 + m12 * rm01 + m22 * rm02;
|
|
float nm10 = m00 * rm10 + m10 * rm11 + m20 * rm12;
|
|
float nm11 = m01 * rm10 + m11 * rm11 + m21 * rm12;
|
|
float nm12 = m02 * rm10 + m12 * rm11 + m22 * rm12;
|
|
// set non-dependent values directly
|
|
return dest
|
|
._m20(m00 * rm20 + m10 * rm21 + m20 * rm22)
|
|
._m21(m01 * rm20 + m11 * rm21 + m21 * rm22)
|
|
._m22(m02 * rm20 + m12 * rm21 + m22 * rm22)
|
|
._m23(0.0f)
|
|
// set other values
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(0.0f)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(0.0f)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(1.0f)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply rotation to this {@link #isAffine() affine} matrix by rotating the given amount of radians
|
|
* about the specified <code>(x, y, z)</code> axis.
|
|
* <p>
|
|
* This method assumes <code>this</code> to be {@link #isAffine() affine}.
|
|
* <p>
|
|
* The axis described by the three components needs to be a unit vector.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* rotation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation matrix without post-multiplying the rotation
|
|
* transformation, use {@link #rotation(float, float, float, float) rotation()}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Rotation_matrix_from_axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotation(float, float, float, float)
|
|
*
|
|
* @param ang
|
|
* the angle in radians
|
|
* @param x
|
|
* the x component of the axis
|
|
* @param y
|
|
* the y component of the axis
|
|
* @param z
|
|
* the z component of the axis
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateAffine(float ang, float x, float y, float z) {
|
|
return rotateAffine(ang, x, y, z, this);
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply a rotation to this matrix by rotating the given amount of radians
|
|
* about the specified <code>(x, y, z)</code> axis and store the result in <code>dest</code>.
|
|
* <p>
|
|
* The axis described by the three components needs to be a unit vector.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>R * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>R * M * v</code>, the
|
|
* rotation will be applied last!
|
|
* <p>
|
|
* In order to set the matrix to a rotation matrix without pre-multiplying the rotation
|
|
* transformation, use {@link #rotation(float, float, float, float) rotation()}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Rotation_matrix_from_axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotation(float, float, float, float)
|
|
*
|
|
* @param ang
|
|
* the angle in radians
|
|
* @param x
|
|
* the x component of the axis
|
|
* @param y
|
|
* the y component of the axis
|
|
* @param z
|
|
* the z component of the axis
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotateLocal(float ang, float x, float y, float z, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.rotation(ang, x, y, z);
|
|
return rotateLocalGeneric(ang, x, y, z, dest);
|
|
}
|
|
private Matrix4f rotateLocalGeneric(float ang, float x, float y, float z, Matrix4f dest) {
|
|
if (y == 0.0f && z == 0.0f && Math.absEqualsOne(x))
|
|
return rotateLocalX(x * ang, dest);
|
|
else if (x == 0.0f && z == 0.0f && Math.absEqualsOne(y))
|
|
return rotateLocalY(y * ang, dest);
|
|
else if (x == 0.0f && y == 0.0f && Math.absEqualsOne(z))
|
|
return rotateLocalZ(z * ang, dest);
|
|
return rotateLocalGenericInternal(ang, x, y, z, dest);
|
|
}
|
|
private Matrix4f rotateLocalGenericInternal(float ang, float x, float y, float z, Matrix4f dest) {
|
|
float s = Math.sin(ang);
|
|
float c = Math.cosFromSin(s, ang);
|
|
float C = 1.0f - c;
|
|
float xx = x * x, xy = x * y, xz = x * z;
|
|
float yy = y * y, yz = y * z;
|
|
float zz = z * z;
|
|
float lm00 = xx * C + c;
|
|
float lm01 = xy * C + z * s;
|
|
float lm02 = xz * C - y * s;
|
|
float lm10 = xy * C - z * s;
|
|
float lm11 = yy * C + c;
|
|
float lm12 = yz * C + x * s;
|
|
float lm20 = xz * C + y * s;
|
|
float lm21 = yz * C - x * s;
|
|
float lm22 = zz * C + c;
|
|
float nm00 = lm00 * m00 + lm10 * m01 + lm20 * m02;
|
|
float nm01 = lm01 * m00 + lm11 * m01 + lm21 * m02;
|
|
float nm02 = lm02 * m00 + lm12 * m01 + lm22 * m02;
|
|
float nm10 = lm00 * m10 + lm10 * m11 + lm20 * m12;
|
|
float nm11 = lm01 * m10 + lm11 * m11 + lm21 * m12;
|
|
float nm12 = lm02 * m10 + lm12 * m11 + lm22 * m12;
|
|
float nm20 = lm00 * m20 + lm10 * m21 + lm20 * m22;
|
|
float nm21 = lm01 * m20 + lm11 * m21 + lm21 * m22;
|
|
float nm22 = lm02 * m20 + lm12 * m21 + lm22 * m22;
|
|
float nm30 = lm00 * m30 + lm10 * m31 + lm20 * m32;
|
|
float nm31 = lm01 * m30 + lm11 * m31 + lm21 * m32;
|
|
float nm32 = lm02 * m30 + lm12 * m31 + lm22 * m32;
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(m03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(m13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(m23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply a rotation to this matrix by rotating the given amount of radians
|
|
* about the specified <code>(x, y, z)</code> axis.
|
|
* <p>
|
|
* The axis described by the three components needs to be a unit vector.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>R * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>R * M * v</code>, the
|
|
* rotation will be applied last!
|
|
* <p>
|
|
* In order to set the matrix to a rotation matrix without pre-multiplying the rotation
|
|
* transformation, use {@link #rotation(float, float, float, float) rotation()}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Rotation_matrix_from_axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotation(float, float, float, float)
|
|
*
|
|
* @param ang
|
|
* the angle in radians
|
|
* @param x
|
|
* the x component of the axis
|
|
* @param y
|
|
* the y component of the axis
|
|
* @param z
|
|
* the z component of the axis
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateLocal(float ang, float x, float y, float z) {
|
|
return rotateLocal(ang, x, y, z, this);
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply a rotation around the X axis to this matrix by rotating the given amount of radians
|
|
* about the X axis and store the result in <code>dest</code>.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>R * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>R * M * v</code>, the
|
|
* rotation will be applied last!
|
|
* <p>
|
|
* In order to set the matrix to a rotation matrix without pre-multiplying the rotation
|
|
* transformation, use {@link #rotationX(float) rotationX()}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Rotation_matrix_from_axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotationX(float)
|
|
*
|
|
* @param ang
|
|
* the angle in radians to rotate about the X axis
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotateLocalX(float ang, Matrix4f dest) {
|
|
float sin = Math.sin(ang);
|
|
float cos = Math.cosFromSin(sin, ang);
|
|
float nm02 = sin * m01 + cos * m02;
|
|
float nm12 = sin * m11 + cos * m12;
|
|
float nm22 = sin * m21 + cos * m22;
|
|
float nm32 = sin * m31 + cos * m32;
|
|
return dest
|
|
._m00(m00)
|
|
._m01(cos * m01 - sin * m02)
|
|
._m02(nm02)
|
|
._m03(m03)
|
|
._m10(m10)
|
|
._m11(cos * m11 - sin * m12)
|
|
._m12(nm12)
|
|
._m13(m13)
|
|
._m20(m20)
|
|
._m21(cos * m21 - sin * m22)
|
|
._m22(nm22)
|
|
._m23(m23)
|
|
._m30(m30)
|
|
._m31(cos * m31 - sin * m32)
|
|
._m32(nm32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply a rotation to this matrix by rotating the given amount of radians about the X axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>R * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>R * M * v</code>, the
|
|
* rotation will be applied last!
|
|
* <p>
|
|
* In order to set the matrix to a rotation matrix without pre-multiplying the rotation
|
|
* transformation, use {@link #rotationX(float) rotationX()}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Rotation_matrix_from_axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotationX(float)
|
|
*
|
|
* @param ang
|
|
* the angle in radians to rotate about the X axis
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateLocalX(float ang) {
|
|
return rotateLocalX(ang, this);
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply a rotation around the Y axis to this matrix by rotating the given amount of radians
|
|
* about the Y axis and store the result in <code>dest</code>.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>R * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>R * M * v</code>, the
|
|
* rotation will be applied last!
|
|
* <p>
|
|
* In order to set the matrix to a rotation matrix without pre-multiplying the rotation
|
|
* transformation, use {@link #rotationY(float) rotationY()}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Rotation_matrix_from_axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotationY(float)
|
|
*
|
|
* @param ang
|
|
* the angle in radians to rotate about the Y axis
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotateLocalY(float ang, Matrix4f dest) {
|
|
float sin = Math.sin(ang);
|
|
float cos = Math.cosFromSin(sin, ang);
|
|
float nm02 = -sin * m00 + cos * m02;
|
|
float nm12 = -sin * m10 + cos * m12;
|
|
float nm22 = -sin * m20 + cos * m22;
|
|
float nm32 = -sin * m30 + cos * m32;
|
|
return dest
|
|
._m00(cos * m00 + sin * m02)
|
|
._m01(m01)
|
|
._m02(nm02)
|
|
._m03(m03)
|
|
._m10(cos * m10 + sin * m12)
|
|
._m11(m11)
|
|
._m12(nm12)
|
|
._m13(m13)
|
|
._m20(cos * m20 + sin * m22)
|
|
._m21(m21)
|
|
._m22(nm22)
|
|
._m23(m23)
|
|
._m30(cos * m30 + sin * m32)
|
|
._m31(m31)
|
|
._m32(nm32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply a rotation to this matrix by rotating the given amount of radians about the Y axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>R * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>R * M * v</code>, the
|
|
* rotation will be applied last!
|
|
* <p>
|
|
* In order to set the matrix to a rotation matrix without pre-multiplying the rotation
|
|
* transformation, use {@link #rotationY(float) rotationY()}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Rotation_matrix_from_axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotationY(float)
|
|
*
|
|
* @param ang
|
|
* the angle in radians to rotate about the Y axis
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateLocalY(float ang) {
|
|
return rotateLocalY(ang, this);
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply a rotation around the Z axis to this matrix by rotating the given amount of radians
|
|
* about the Z axis and store the result in <code>dest</code>.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>R * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>R * M * v</code>, the
|
|
* rotation will be applied last!
|
|
* <p>
|
|
* In order to set the matrix to a rotation matrix without pre-multiplying the rotation
|
|
* transformation, use {@link #rotationZ(float) rotationZ()}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Rotation_matrix_from_axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotationZ(float)
|
|
*
|
|
* @param ang
|
|
* the angle in radians to rotate about the Z axis
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotateLocalZ(float ang, Matrix4f dest) {
|
|
float sin = Math.sin(ang);
|
|
float cos = Math.cosFromSin(sin, ang);
|
|
float nm01 = sin * m00 + cos * m01;
|
|
float nm11 = sin * m10 + cos * m11;
|
|
float nm21 = sin * m20 + cos * m21;
|
|
float nm31 = sin * m30 + cos * m31;
|
|
return dest
|
|
._m00(cos * m00 - sin * m01)
|
|
._m01(nm01)
|
|
._m02(m02)
|
|
._m03(m03)
|
|
._m10(cos * m10 - sin * m11)
|
|
._m11(nm11)
|
|
._m12(m12)
|
|
._m13(m13)
|
|
._m20(cos * m20 - sin * m21)
|
|
._m21(nm21)
|
|
._m22(m22)
|
|
._m23(m23)
|
|
._m30(cos * m30 - sin * m31)
|
|
._m31(nm31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply a rotation to this matrix by rotating the given amount of radians about the Z axis.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the rotation matrix,
|
|
* then the new matrix will be <code>R * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>R * M * v</code>, the
|
|
* rotation will be applied last!
|
|
* <p>
|
|
* In order to set the matrix to a rotation matrix without pre-multiplying the rotation
|
|
* transformation, use {@link #rotationZ(float) rotationY()}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Rotation_matrix_from_axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotationY(float)
|
|
*
|
|
* @param ang
|
|
* the angle in radians to rotate about the Z axis
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateLocalZ(float ang) {
|
|
return rotateLocalZ(ang, this);
|
|
}
|
|
|
|
/**
|
|
* Apply a translation to this matrix by translating by the given number of
|
|
* units in x, y and z.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>T</code> the translation
|
|
* matrix, then the new matrix will be <code>M * T</code>. So when
|
|
* transforming a vector <code>v</code> with the new matrix by using
|
|
* <code>M * T * v</code>, the translation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a translation transformation without post-multiplying
|
|
* it, use {@link #translation(Vector3fc)}.
|
|
*
|
|
* @see #translation(Vector3fc)
|
|
*
|
|
* @param offset
|
|
* the number of units in x, y and z by which to translate
|
|
* @return this
|
|
*/
|
|
public Matrix4f translate(Vector3fc offset) {
|
|
return translate(offset.x(), offset.y(), offset.z());
|
|
}
|
|
|
|
/**
|
|
* Apply a translation to this matrix by translating by the given number of
|
|
* units in x, y and z and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>T</code> the translation
|
|
* matrix, then the new matrix will be <code>M * T</code>. So when
|
|
* transforming a vector <code>v</code> with the new matrix by using
|
|
* <code>M * T * v</code>, the translation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a translation transformation without post-multiplying
|
|
* it, use {@link #translation(Vector3fc)}.
|
|
*
|
|
* @see #translation(Vector3fc)
|
|
*
|
|
* @param offset
|
|
* the number of units in x, y and z by which to translate
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f translate(Vector3fc offset, Matrix4f dest) {
|
|
return translate(offset.x(), offset.y(), offset.z(), dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a translation to this matrix by translating by the given number of
|
|
* units in x, y and z and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>T</code> the translation
|
|
* matrix, then the new matrix will be <code>M * T</code>. So when
|
|
* transforming a vector <code>v</code> with the new matrix by using
|
|
* <code>M * T * v</code>, the translation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a translation transformation without post-multiplying
|
|
* it, use {@link #translation(float, float, float)}.
|
|
*
|
|
* @see #translation(float, float, float)
|
|
*
|
|
* @param x
|
|
* the offset to translate in x
|
|
* @param y
|
|
* the offset to translate in y
|
|
* @param z
|
|
* the offset to translate in z
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f translate(float x, float y, float z, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.translation(x, y, z);
|
|
return translateGeneric(x, y, z, dest);
|
|
}
|
|
private Matrix4f translateGeneric(float x, float y, float z, Matrix4f dest) {
|
|
MemUtil.INSTANCE.copy(this, dest);
|
|
return dest
|
|
._m30(Math.fma(m00, x, Math.fma(m10, y, Math.fma(m20, z, m30))))
|
|
._m31(Math.fma(m01, x, Math.fma(m11, y, Math.fma(m21, z, m31))))
|
|
._m32(Math.fma(m02, x, Math.fma(m12, y, Math.fma(m22, z, m32))))
|
|
._m33(Math.fma(m03, x, Math.fma(m13, y, Math.fma(m23, z, m33))))
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY));
|
|
}
|
|
|
|
/**
|
|
* Apply a translation to this matrix by translating by the given number of
|
|
* units in x, y and z.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>T</code> the translation
|
|
* matrix, then the new matrix will be <code>M * T</code>. So when
|
|
* transforming a vector <code>v</code> with the new matrix by using
|
|
* <code>M * T * v</code>, the translation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a translation transformation without post-multiplying
|
|
* it, use {@link #translation(float, float, float)}.
|
|
*
|
|
* @see #translation(float, float, float)
|
|
*
|
|
* @param x
|
|
* the offset to translate in x
|
|
* @param y
|
|
* the offset to translate in y
|
|
* @param z
|
|
* the offset to translate in z
|
|
* @return this
|
|
*/
|
|
public Matrix4f translate(float x, float y, float z) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return translation(x, y, z);
|
|
return translateGeneric(x, y, z);
|
|
}
|
|
private Matrix4f translateGeneric(float x, float y, float z) {
|
|
return this
|
|
._m30(Math.fma(m00, x, Math.fma(m10, y, Math.fma(m20, z, m30))))
|
|
._m31(Math.fma(m01, x, Math.fma(m11, y, Math.fma(m21, z, m31))))
|
|
._m32(Math.fma(m02, x, Math.fma(m12, y, Math.fma(m22, z, m32))))
|
|
._m33(Math.fma(m03, x, Math.fma(m13, y, Math.fma(m23, z, m33))))
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY));
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply a translation to this matrix by translating by the given number of
|
|
* units in x, y and z.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>T</code> the translation
|
|
* matrix, then the new matrix will be <code>T * M</code>. So when
|
|
* transforming a vector <code>v</code> with the new matrix by using
|
|
* <code>T * M * v</code>, the translation will be applied last!
|
|
* <p>
|
|
* In order to set the matrix to a translation transformation without pre-multiplying
|
|
* it, use {@link #translation(Vector3fc)}.
|
|
*
|
|
* @see #translation(Vector3fc)
|
|
*
|
|
* @param offset
|
|
* the number of units in x, y and z by which to translate
|
|
* @return this
|
|
*/
|
|
public Matrix4f translateLocal(Vector3fc offset) {
|
|
return translateLocal(offset.x(), offset.y(), offset.z());
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply a translation to this matrix by translating by the given number of
|
|
* units in x, y and z and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>T</code> the translation
|
|
* matrix, then the new matrix will be <code>T * M</code>. So when
|
|
* transforming a vector <code>v</code> with the new matrix by using
|
|
* <code>T * M * v</code>, the translation will be applied last!
|
|
* <p>
|
|
* In order to set the matrix to a translation transformation without pre-multiplying
|
|
* it, use {@link #translation(Vector3fc)}.
|
|
*
|
|
* @see #translation(Vector3fc)
|
|
*
|
|
* @param offset
|
|
* the number of units in x, y and z by which to translate
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f translateLocal(Vector3fc offset, Matrix4f dest) {
|
|
return translateLocal(offset.x(), offset.y(), offset.z(), dest);
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply a translation to this matrix by translating by the given number of
|
|
* units in x, y and z and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>T</code> the translation
|
|
* matrix, then the new matrix will be <code>T * M</code>. So when
|
|
* transforming a vector <code>v</code> with the new matrix by using
|
|
* <code>T * M * v</code>, the translation will be applied last!
|
|
* <p>
|
|
* In order to set the matrix to a translation transformation without pre-multiplying
|
|
* it, use {@link #translation(float, float, float)}.
|
|
*
|
|
* @see #translation(float, float, float)
|
|
*
|
|
* @param x
|
|
* the offset to translate in x
|
|
* @param y
|
|
* the offset to translate in y
|
|
* @param z
|
|
* the offset to translate in z
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f translateLocal(float x, float y, float z, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.translation(x, y, z);
|
|
return translateLocalGeneric(x, y, z, dest);
|
|
}
|
|
private Matrix4f translateLocalGeneric(float x, float y, float z, Matrix4f dest) {
|
|
float nm00 = m00 + x * m03;
|
|
float nm01 = m01 + y * m03;
|
|
float nm02 = m02 + z * m03;
|
|
float nm10 = m10 + x * m13;
|
|
float nm11 = m11 + y * m13;
|
|
float nm12 = m12 + z * m13;
|
|
float nm20 = m20 + x * m23;
|
|
float nm21 = m21 + y * m23;
|
|
float nm22 = m22 + z * m23;
|
|
float nm30 = m30 + x * m33;
|
|
float nm31 = m31 + y * m33;
|
|
float nm32 = m32 + z * m33;
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(m03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(m13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(m23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY));
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply a translation to this matrix by translating by the given number of
|
|
* units in x, y and z.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>T</code> the translation
|
|
* matrix, then the new matrix will be <code>T * M</code>. So when
|
|
* transforming a vector <code>v</code> with the new matrix by using
|
|
* <code>T * M * v</code>, the translation will be applied last!
|
|
* <p>
|
|
* In order to set the matrix to a translation transformation without pre-multiplying
|
|
* it, use {@link #translation(float, float, float)}.
|
|
*
|
|
* @see #translation(float, float, float)
|
|
*
|
|
* @param x
|
|
* the offset to translate in x
|
|
* @param y
|
|
* the offset to translate in y
|
|
* @param z
|
|
* the offset to translate in z
|
|
* @return this
|
|
*/
|
|
public Matrix4f translateLocal(float x, float y, float z) {
|
|
return translateLocal(x, y, z, this);
|
|
}
|
|
|
|
public void writeExternal(ObjectOutput out) throws IOException {
|
|
out.writeFloat(m00);
|
|
out.writeFloat(m01);
|
|
out.writeFloat(m02);
|
|
out.writeFloat(m03);
|
|
out.writeFloat(m10);
|
|
out.writeFloat(m11);
|
|
out.writeFloat(m12);
|
|
out.writeFloat(m13);
|
|
out.writeFloat(m20);
|
|
out.writeFloat(m21);
|
|
out.writeFloat(m22);
|
|
out.writeFloat(m23);
|
|
out.writeFloat(m30);
|
|
out.writeFloat(m31);
|
|
out.writeFloat(m32);
|
|
out.writeFloat(m33);
|
|
}
|
|
|
|
public void readExternal(ObjectInput in) throws IOException {
|
|
this._m00(in.readFloat())
|
|
._m01(in.readFloat())
|
|
._m02(in.readFloat())
|
|
._m03(in.readFloat())
|
|
._m10(in.readFloat())
|
|
._m11(in.readFloat())
|
|
._m12(in.readFloat())
|
|
._m13(in.readFloat())
|
|
._m20(in.readFloat())
|
|
._m21(in.readFloat())
|
|
._m22(in.readFloat())
|
|
._m23(in.readFloat())
|
|
._m30(in.readFloat())
|
|
._m31(in.readFloat())
|
|
._m32(in.readFloat())
|
|
._m33(in.readFloat())
|
|
.determineProperties();
|
|
}
|
|
|
|
/**
|
|
* Apply an orthographic projection transformation for a right-handed coordinate system
|
|
* using the given NDC z range to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to an orthographic projection without post-multiplying it,
|
|
* use {@link #setOrtho(float, float, float, float, float, float, boolean) setOrtho()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrtho(float, float, float, float, float, float, boolean)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f ortho(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setOrtho(left, right, bottom, top, zNear, zFar, zZeroToOne);
|
|
return orthoGeneric(left, right, bottom, top, zNear, zFar, zZeroToOne, dest);
|
|
}
|
|
private Matrix4f orthoGeneric(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
// calculate right matrix elements
|
|
float rm00 = 2.0f / (right - left);
|
|
float rm11 = 2.0f / (top - bottom);
|
|
float rm22 = (zZeroToOne ? 1.0f : 2.0f) / (zNear - zFar);
|
|
float rm30 = (left + right) / (left - right);
|
|
float rm31 = (top + bottom) / (bottom - top);
|
|
float rm32 = (zZeroToOne ? zNear : (zFar + zNear)) / (zNear - zFar);
|
|
// perform optimized multiplication
|
|
// compute the last column first, because other columns do not depend on it
|
|
dest._m30(m00 * rm30 + m10 * rm31 + m20 * rm32 + m30)
|
|
._m31(m01 * rm30 + m11 * rm31 + m21 * rm32 + m31)
|
|
._m32(m02 * rm30 + m12 * rm31 + m22 * rm32 + m32)
|
|
._m33(m03 * rm30 + m13 * rm31 + m23 * rm32 + m33)
|
|
._m00(m00 * rm00)
|
|
._m01(m01 * rm00)
|
|
._m02(m02 * rm00)
|
|
._m03(m03 * rm00)
|
|
._m10(m10 * rm11)
|
|
._m11(m11 * rm11)
|
|
._m12(m12 * rm11)
|
|
._m13(m13 * rm11)
|
|
._m20(m20 * rm22)
|
|
._m21(m21 * rm22)
|
|
._m22(m22 * rm22)
|
|
._m23(m23 * rm22)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply an orthographic projection transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to an orthographic projection without post-multiplying it,
|
|
* use {@link #setOrtho(float, float, float, float, float, float) setOrtho()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrtho(float, float, float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f ortho(float left, float right, float bottom, float top, float zNear, float zFar, Matrix4f dest) {
|
|
return ortho(left, right, bottom, top, zNear, zFar, false, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply an orthographic projection transformation for a right-handed coordinate system using the given NDC z range to this matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to an orthographic projection without post-multiplying it,
|
|
* use {@link #setOrtho(float, float, float, float, float, float, boolean) setOrtho()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrtho(float, float, float, float, float, float, boolean)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f ortho(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne) {
|
|
return ortho(left, right, bottom, top, zNear, zFar, zZeroToOne, this);
|
|
}
|
|
|
|
/**
|
|
* Apply an orthographic projection transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to an orthographic projection without post-multiplying it,
|
|
* use {@link #setOrtho(float, float, float, float, float, float) setOrtho()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrtho(float, float, float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @return this
|
|
*/
|
|
public Matrix4f ortho(float left, float right, float bottom, float top, float zNear, float zFar) {
|
|
return ortho(left, right, bottom, top, zNear, zFar, false);
|
|
}
|
|
|
|
/**
|
|
* Apply an orthographic projection transformation for a left-handed coordiante system
|
|
* using the given NDC z range to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to an orthographic projection without post-multiplying it,
|
|
* use {@link #setOrthoLH(float, float, float, float, float, float, boolean) setOrthoLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrthoLH(float, float, float, float, float, float, boolean)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f orthoLH(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setOrthoLH(left, right, bottom, top, zNear, zFar, zZeroToOne);
|
|
return orthoLHGeneric(left, right, bottom, top, zNear, zFar, zZeroToOne, dest);
|
|
}
|
|
private Matrix4f orthoLHGeneric(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
// calculate right matrix elements
|
|
float rm00 = 2.0f / (right - left);
|
|
float rm11 = 2.0f / (top - bottom);
|
|
float rm22 = (zZeroToOne ? 1.0f : 2.0f) / (zFar - zNear);
|
|
float rm30 = (left + right) / (left - right);
|
|
float rm31 = (top + bottom) / (bottom - top);
|
|
float rm32 = (zZeroToOne ? zNear : (zFar + zNear)) / (zNear - zFar);
|
|
|
|
// perform optimized multiplication
|
|
// compute the last column first, because other columns do not depend on it
|
|
dest._m30(m00 * rm30 + m10 * rm31 + m20 * rm32 + m30)
|
|
._m31(m01 * rm30 + m11 * rm31 + m21 * rm32 + m31)
|
|
._m32(m02 * rm30 + m12 * rm31 + m22 * rm32 + m32)
|
|
._m33(m03 * rm30 + m13 * rm31 + m23 * rm32 + m33)
|
|
._m00(m00 * rm00)
|
|
._m01(m01 * rm00)
|
|
._m02(m02 * rm00)
|
|
._m03(m03 * rm00)
|
|
._m10(m10 * rm11)
|
|
._m11(m11 * rm11)
|
|
._m12(m12 * rm11)
|
|
._m13(m13 * rm11)
|
|
._m20(m20 * rm22)
|
|
._m21(m21 * rm22)
|
|
._m22(m22 * rm22)
|
|
._m23(m23 * rm22)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply an orthographic projection transformation for a left-handed coordiante system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to an orthographic projection without post-multiplying it,
|
|
* use {@link #setOrthoLH(float, float, float, float, float, float) setOrthoLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrthoLH(float, float, float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f orthoLH(float left, float right, float bottom, float top, float zNear, float zFar, Matrix4f dest) {
|
|
return orthoLH(left, right, bottom, top, zNear, zFar, false, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply an orthographic projection transformation for a left-handed coordiante system
|
|
* using the given NDC z range to this matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to an orthographic projection without post-multiplying it,
|
|
* use {@link #setOrthoLH(float, float, float, float, float, float, boolean) setOrthoLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrthoLH(float, float, float, float, float, float, boolean)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f orthoLH(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne) {
|
|
return orthoLH(left, right, bottom, top, zNear, zFar, zZeroToOne, this);
|
|
}
|
|
|
|
/**
|
|
* Apply an orthographic projection transformation for a left-handed coordiante system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to an orthographic projection without post-multiplying it,
|
|
* use {@link #setOrthoLH(float, float, float, float, float, float) setOrthoLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrthoLH(float, float, float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @return this
|
|
*/
|
|
public Matrix4f orthoLH(float left, float right, float bottom, float top, float zNear, float zFar) {
|
|
return orthoLH(left, right, bottom, top, zNear, zFar, false);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be an orthographic projection transformation for a right-handed coordinate system
|
|
* using the given NDC z range.
|
|
* <p>
|
|
* In order to apply the orthographic projection to an already existing transformation,
|
|
* use {@link #ortho(float, float, float, float, float, float, boolean) ortho()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #ortho(float, float, float, float, float, float, boolean)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f setOrtho(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne) {
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
this._m00(2.0f / (right - left))
|
|
._m11(2.0f / (top - bottom))
|
|
._m22((zZeroToOne ? 1.0f : 2.0f) / (zNear - zFar))
|
|
._m30((right + left) / (left - right))
|
|
._m31((top + bottom) / (bottom - top))
|
|
._m32((zZeroToOne ? zNear : (zFar + zNear)) / (zNear - zFar))
|
|
._properties(PROPERTY_AFFINE);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be an orthographic projection transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code>.
|
|
* <p>
|
|
* In order to apply the orthographic projection to an already existing transformation,
|
|
* use {@link #ortho(float, float, float, float, float, float) ortho()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #ortho(float, float, float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @return this
|
|
*/
|
|
public Matrix4f setOrtho(float left, float right, float bottom, float top, float zNear, float zFar) {
|
|
return setOrtho(left, right, bottom, top, zNear, zFar, false);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be an orthographic projection transformation for a left-handed coordinate system
|
|
* using the given NDC z range.
|
|
* <p>
|
|
* In order to apply the orthographic projection to an already existing transformation,
|
|
* use {@link #orthoLH(float, float, float, float, float, float, boolean) orthoLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #orthoLH(float, float, float, float, float, float, boolean)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f setOrthoLH(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne) {
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
this._m00(2.0f / (right - left))
|
|
._m11(2.0f / (top - bottom))
|
|
._m22((zZeroToOne ? 1.0f : 2.0f) / (zFar - zNear))
|
|
._m30((right + left) / (left - right))
|
|
._m31((top + bottom) / (bottom - top))
|
|
._m32((zZeroToOne ? zNear : (zFar + zNear)) / (zNear - zFar))
|
|
._properties(PROPERTY_AFFINE);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be an orthographic projection transformation for a left-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code>.
|
|
* <p>
|
|
* In order to apply the orthographic projection to an already existing transformation,
|
|
* use {@link #orthoLH(float, float, float, float, float, float) orthoLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #orthoLH(float, float, float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @return this
|
|
*/
|
|
public Matrix4f setOrthoLH(float left, float right, float bottom, float top, float zNear, float zFar) {
|
|
return setOrthoLH(left, right, bottom, top, zNear, zFar, false);
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric orthographic projection transformation for a right-handed coordinate system
|
|
* using the given NDC z range to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #ortho(float, float, float, float, float, float, boolean, Matrix4f) ortho()} with
|
|
* <code>left=-width/2</code>, <code>right=+width/2</code>, <code>bottom=-height/2</code> and <code>top=+height/2</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a symmetric orthographic projection without post-multiplying it,
|
|
* use {@link #setOrthoSymmetric(float, float, float, float, boolean) setOrthoSymmetric()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrthoSymmetric(float, float, float, float, boolean)
|
|
*
|
|
* @param width
|
|
* the distance between the right and left frustum edges
|
|
* @param height
|
|
* the distance between the top and bottom frustum edges
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param dest
|
|
* will hold the result
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return dest
|
|
*/
|
|
public Matrix4f orthoSymmetric(float width, float height, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setOrthoSymmetric(width, height, zNear, zFar, zZeroToOne);
|
|
return orthoSymmetricGeneric(width, height, zNear, zFar, zZeroToOne, dest);
|
|
}
|
|
private Matrix4f orthoSymmetricGeneric(float width, float height, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
// calculate right matrix elements
|
|
float rm00 = 2.0f / width;
|
|
float rm11 = 2.0f / height;
|
|
float rm22 = (zZeroToOne ? 1.0f : 2.0f) / (zNear - zFar);
|
|
float rm32 = (zZeroToOne ? zNear : (zFar + zNear)) / (zNear - zFar);
|
|
// perform optimized multiplication
|
|
// compute the last column first, because other columns do not depend on it
|
|
dest._m30(m20 * rm32 + m30)
|
|
._m31(m21 * rm32 + m31)
|
|
._m32(m22 * rm32 + m32)
|
|
._m33(m23 * rm32 + m33)
|
|
._m00(m00 * rm00)
|
|
._m01(m01 * rm00)
|
|
._m02(m02 * rm00)
|
|
._m03(m03 * rm00)
|
|
._m10(m10 * rm11)
|
|
._m11(m11 * rm11)
|
|
._m12(m12 * rm11)
|
|
._m13(m13 * rm11)
|
|
._m20(m20 * rm22)
|
|
._m21(m21 * rm22)
|
|
._m22(m22 * rm22)
|
|
._m23(m23 * rm22)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric orthographic projection transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #ortho(float, float, float, float, float, float, Matrix4f) ortho()} with
|
|
* <code>left=-width/2</code>, <code>right=+width/2</code>, <code>bottom=-height/2</code> and <code>top=+height/2</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a symmetric orthographic projection without post-multiplying it,
|
|
* use {@link #setOrthoSymmetric(float, float, float, float) setOrthoSymmetric()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrthoSymmetric(float, float, float, float)
|
|
*
|
|
* @param width
|
|
* the distance between the right and left frustum edges
|
|
* @param height
|
|
* the distance between the top and bottom frustum edges
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f orthoSymmetric(float width, float height, float zNear, float zFar, Matrix4f dest) {
|
|
return orthoSymmetric(width, height, zNear, zFar, false, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric orthographic projection transformation for a right-handed coordinate system
|
|
* using the given NDC z range to this matrix.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #ortho(float, float, float, float, float, float, boolean) ortho()} with
|
|
* <code>left=-width/2</code>, <code>right=+width/2</code>, <code>bottom=-height/2</code> and <code>top=+height/2</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a symmetric orthographic projection without post-multiplying it,
|
|
* use {@link #setOrthoSymmetric(float, float, float, float, boolean) setOrthoSymmetric()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrthoSymmetric(float, float, float, float, boolean)
|
|
*
|
|
* @param width
|
|
* the distance between the right and left frustum edges
|
|
* @param height
|
|
* the distance between the top and bottom frustum edges
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f orthoSymmetric(float width, float height, float zNear, float zFar, boolean zZeroToOne) {
|
|
return orthoSymmetric(width, height, zNear, zFar, zZeroToOne, this);
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric orthographic projection transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #ortho(float, float, float, float, float, float) ortho()} with
|
|
* <code>left=-width/2</code>, <code>right=+width/2</code>, <code>bottom=-height/2</code> and <code>top=+height/2</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a symmetric orthographic projection without post-multiplying it,
|
|
* use {@link #setOrthoSymmetric(float, float, float, float) setOrthoSymmetric()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrthoSymmetric(float, float, float, float)
|
|
*
|
|
* @param width
|
|
* the distance between the right and left frustum edges
|
|
* @param height
|
|
* the distance between the top and bottom frustum edges
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @return this
|
|
*/
|
|
public Matrix4f orthoSymmetric(float width, float height, float zNear, float zFar) {
|
|
return orthoSymmetric(width, height, zNear, zFar, false, this);
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric orthographic projection transformation for a left-handed coordinate system
|
|
* using the given NDC z range to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #orthoLH(float, float, float, float, float, float, boolean, Matrix4f) orthoLH()} with
|
|
* <code>left=-width/2</code>, <code>right=+width/2</code>, <code>bottom=-height/2</code> and <code>top=+height/2</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a symmetric orthographic projection without post-multiplying it,
|
|
* use {@link #setOrthoSymmetricLH(float, float, float, float, boolean) setOrthoSymmetricLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrthoSymmetricLH(float, float, float, float, boolean)
|
|
*
|
|
* @param width
|
|
* the distance between the right and left frustum edges
|
|
* @param height
|
|
* the distance between the top and bottom frustum edges
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param dest
|
|
* will hold the result
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return dest
|
|
*/
|
|
public Matrix4f orthoSymmetricLH(float width, float height, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setOrthoSymmetricLH(width, height, zNear, zFar, zZeroToOne);
|
|
return orthoSymmetricLHGeneric(width, height, zNear, zFar, zZeroToOne, dest);
|
|
}
|
|
private Matrix4f orthoSymmetricLHGeneric(float width, float height, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
// calculate right matrix elements
|
|
float rm00 = 2.0f / width;
|
|
float rm11 = 2.0f / height;
|
|
float rm22 = (zZeroToOne ? 1.0f : 2.0f) / (zFar - zNear);
|
|
float rm32 = (zZeroToOne ? zNear : (zFar + zNear)) / (zNear - zFar);
|
|
// perform optimized multiplication
|
|
// compute the last column first, because other columns do not depend on it
|
|
dest._m30(m20 * rm32 + m30)
|
|
._m31(m21 * rm32 + m31)
|
|
._m32(m22 * rm32 + m32)
|
|
._m33(m23 * rm32 + m33)
|
|
._m00(m00 * rm00)
|
|
._m01(m01 * rm00)
|
|
._m02(m02 * rm00)
|
|
._m03(m03 * rm00)
|
|
._m10(m10 * rm11)
|
|
._m11(m11 * rm11)
|
|
._m12(m12 * rm11)
|
|
._m13(m13 * rm11)
|
|
._m20(m20 * rm22)
|
|
._m21(m21 * rm22)
|
|
._m22(m22 * rm22)
|
|
._m23(m23 * rm22)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric orthographic projection transformation for a left-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #orthoLH(float, float, float, float, float, float, Matrix4f) orthoLH()} with
|
|
* <code>left=-width/2</code>, <code>right=+width/2</code>, <code>bottom=-height/2</code> and <code>top=+height/2</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a symmetric orthographic projection without post-multiplying it,
|
|
* use {@link #setOrthoSymmetricLH(float, float, float, float) setOrthoSymmetricLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrthoSymmetricLH(float, float, float, float)
|
|
*
|
|
* @param width
|
|
* the distance between the right and left frustum edges
|
|
* @param height
|
|
* the distance between the top and bottom frustum edges
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f orthoSymmetricLH(float width, float height, float zNear, float zFar, Matrix4f dest) {
|
|
return orthoSymmetricLH(width, height, zNear, zFar, false, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric orthographic projection transformation for a left-handed coordinate system
|
|
* using the given NDC z range to this matrix.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #orthoLH(float, float, float, float, float, float, boolean) orthoLH()} with
|
|
* <code>left=-width/2</code>, <code>right=+width/2</code>, <code>bottom=-height/2</code> and <code>top=+height/2</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a symmetric orthographic projection without post-multiplying it,
|
|
* use {@link #setOrthoSymmetricLH(float, float, float, float, boolean) setOrthoSymmetricLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrthoSymmetricLH(float, float, float, float, boolean)
|
|
*
|
|
* @param width
|
|
* the distance between the right and left frustum edges
|
|
* @param height
|
|
* the distance between the top and bottom frustum edges
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f orthoSymmetricLH(float width, float height, float zNear, float zFar, boolean zZeroToOne) {
|
|
return orthoSymmetricLH(width, height, zNear, zFar, zZeroToOne, this);
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric orthographic projection transformation for a left-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #orthoLH(float, float, float, float, float, float) orthoLH()} with
|
|
* <code>left=-width/2</code>, <code>right=+width/2</code>, <code>bottom=-height/2</code> and <code>top=+height/2</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a symmetric orthographic projection without post-multiplying it,
|
|
* use {@link #setOrthoSymmetricLH(float, float, float, float) setOrthoSymmetricLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrthoSymmetricLH(float, float, float, float)
|
|
*
|
|
* @param width
|
|
* the distance between the right and left frustum edges
|
|
* @param height
|
|
* the distance between the top and bottom frustum edges
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @return this
|
|
*/
|
|
public Matrix4f orthoSymmetricLH(float width, float height, float zNear, float zFar) {
|
|
return orthoSymmetricLH(width, height, zNear, zFar, false, this);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a symmetric orthographic projection transformation for a right-handed coordinate system using the given NDC z range.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #setOrtho(float, float, float, float, float, float, boolean) setOrtho()} with
|
|
* <code>left=-width/2</code>, <code>right=+width/2</code>, <code>bottom=-height/2</code> and <code>top=+height/2</code>.
|
|
* <p>
|
|
* In order to apply the symmetric orthographic projection to an already existing transformation,
|
|
* use {@link #orthoSymmetric(float, float, float, float, boolean) orthoSymmetric()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #orthoSymmetric(float, float, float, float, boolean)
|
|
*
|
|
* @param width
|
|
* the distance between the right and left frustum edges
|
|
* @param height
|
|
* the distance between the top and bottom frustum edges
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f setOrthoSymmetric(float width, float height, float zNear, float zFar, boolean zZeroToOne) {
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
this._m00(2.0f / width)
|
|
._m11(2.0f / height)
|
|
._m22((zZeroToOne ? 1.0f : 2.0f) / (zNear - zFar))
|
|
._m32((zZeroToOne ? zNear : (zFar + zNear)) / (zNear - zFar))
|
|
._properties(PROPERTY_AFFINE);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a symmetric orthographic projection transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code>.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #setOrtho(float, float, float, float, float, float) setOrtho()} with
|
|
* <code>left=-width/2</code>, <code>right=+width/2</code>, <code>bottom=-height/2</code> and <code>top=+height/2</code>.
|
|
* <p>
|
|
* In order to apply the symmetric orthographic projection to an already existing transformation,
|
|
* use {@link #orthoSymmetric(float, float, float, float) orthoSymmetric()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #orthoSymmetric(float, float, float, float)
|
|
*
|
|
* @param width
|
|
* the distance between the right and left frustum edges
|
|
* @param height
|
|
* the distance between the top and bottom frustum edges
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @return this
|
|
*/
|
|
public Matrix4f setOrthoSymmetric(float width, float height, float zNear, float zFar) {
|
|
return setOrthoSymmetric(width, height, zNear, zFar, false);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a symmetric orthographic projection transformation for a left-handed coordinate system using the given NDC z range.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #setOrtho(float, float, float, float, float, float, boolean) setOrtho()} with
|
|
* <code>left=-width/2</code>, <code>right=+width/2</code>, <code>bottom=-height/2</code> and <code>top=+height/2</code>.
|
|
* <p>
|
|
* In order to apply the symmetric orthographic projection to an already existing transformation,
|
|
* use {@link #orthoSymmetricLH(float, float, float, float, boolean) orthoSymmetricLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #orthoSymmetricLH(float, float, float, float, boolean)
|
|
*
|
|
* @param width
|
|
* the distance between the right and left frustum edges
|
|
* @param height
|
|
* the distance between the top and bottom frustum edges
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f setOrthoSymmetricLH(float width, float height, float zNear, float zFar, boolean zZeroToOne) {
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
this._m00(2.0f / width)
|
|
._m11(2.0f / height)
|
|
._m22((zZeroToOne ? 1.0f : 2.0f) / (zFar - zNear))
|
|
._m32((zZeroToOne ? zNear : (zFar + zNear)) / (zNear - zFar))
|
|
._properties(PROPERTY_AFFINE);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a symmetric orthographic projection transformation for a left-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code>.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #setOrthoLH(float, float, float, float, float, float) setOrthoLH()} with
|
|
* <code>left=-width/2</code>, <code>right=+width/2</code>, <code>bottom=-height/2</code> and <code>top=+height/2</code>.
|
|
* <p>
|
|
* In order to apply the symmetric orthographic projection to an already existing transformation,
|
|
* use {@link #orthoSymmetricLH(float, float, float, float) orthoSymmetricLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #orthoSymmetricLH(float, float, float, float)
|
|
*
|
|
* @param width
|
|
* the distance between the right and left frustum edges
|
|
* @param height
|
|
* the distance between the top and bottom frustum edges
|
|
* @param zNear
|
|
* near clipping plane distance
|
|
* @param zFar
|
|
* far clipping plane distance
|
|
* @return this
|
|
*/
|
|
public Matrix4f setOrthoSymmetricLH(float width, float height, float zNear, float zFar) {
|
|
return setOrthoSymmetricLH(width, height, zNear, zFar, false);
|
|
}
|
|
|
|
/**
|
|
* Apply an orthographic projection transformation for a right-handed coordinate system to this matrix
|
|
* and store the result in <code>dest</code>.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #ortho(float, float, float, float, float, float, Matrix4f) ortho()} with
|
|
* <code>zNear=-1</code> and <code>zFar=+1</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to an orthographic projection without post-multiplying it,
|
|
* use {@link #setOrtho2D(float, float, float, float) setOrtho()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #ortho(float, float, float, float, float, float, Matrix4f)
|
|
* @see #setOrtho2D(float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f ortho2D(float left, float right, float bottom, float top, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setOrtho2D(left, right, bottom, top);
|
|
return ortho2DGeneric(left, right, bottom, top, dest);
|
|
}
|
|
private Matrix4f ortho2DGeneric(float left, float right, float bottom, float top, Matrix4f dest) {
|
|
// calculate right matrix elements
|
|
float rm00 = 2.0f / (right - left);
|
|
float rm11 = 2.0f / (top - bottom);
|
|
float rm30 = (right + left) / (left - right);
|
|
float rm31 = (top + bottom) / (bottom - top);
|
|
// perform optimized multiplication
|
|
// compute the last column first, because other columns do not depend on it
|
|
dest._m30(m00 * rm30 + m10 * rm31 + m30)
|
|
._m31(m01 * rm30 + m11 * rm31 + m31)
|
|
._m32(m02 * rm30 + m12 * rm31 + m32)
|
|
._m33(m03 * rm30 + m13 * rm31 + m33)
|
|
._m00(m00 * rm00)
|
|
._m01(m01 * rm00)
|
|
._m02(m02 * rm00)
|
|
._m03(m03 * rm00)
|
|
._m10(m10 * rm11)
|
|
._m11(m11 * rm11)
|
|
._m12(m12 * rm11)
|
|
._m13(m13 * rm11)
|
|
._m20(-m20)
|
|
._m21(-m21)
|
|
._m22(-m22)
|
|
._m23(-m23)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply an orthographic projection transformation for a right-handed coordinate system to this matrix.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #ortho(float, float, float, float, float, float) ortho()} with
|
|
* <code>zNear=-1</code> and <code>zFar=+1</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to an orthographic projection without post-multiplying it,
|
|
* use {@link #setOrtho2D(float, float, float, float) setOrtho2D()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #ortho(float, float, float, float, float, float)
|
|
* @see #setOrtho2D(float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @return this
|
|
*/
|
|
public Matrix4f ortho2D(float left, float right, float bottom, float top) {
|
|
return ortho2D(left, right, bottom, top, this);
|
|
}
|
|
|
|
/**
|
|
* Apply an orthographic projection transformation for a left-handed coordinate system to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #orthoLH(float, float, float, float, float, float, Matrix4f) orthoLH()} with
|
|
* <code>zNear=-1</code> and <code>zFar=+1</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to an orthographic projection without post-multiplying it,
|
|
* use {@link #setOrtho2DLH(float, float, float, float) setOrthoLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #orthoLH(float, float, float, float, float, float, Matrix4f)
|
|
* @see #setOrtho2DLH(float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f ortho2DLH(float left, float right, float bottom, float top, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setOrtho2DLH(left, right, bottom, top);
|
|
return ortho2DLHGeneric(left, right, bottom, top, dest);
|
|
}
|
|
private Matrix4f ortho2DLHGeneric(float left, float right, float bottom, float top, Matrix4f dest) {
|
|
// calculate right matrix elements
|
|
float rm00 = 2.0f / (right - left);
|
|
float rm11 = 2.0f / (top - bottom);
|
|
float rm30 = (right + left) / (left - right);
|
|
float rm31 = (top + bottom) / (bottom - top);
|
|
|
|
// perform optimized multiplication
|
|
// compute the last column first, because other columns do not depend on it
|
|
dest._m30(m00 * rm30 + m10 * rm31 + m30)
|
|
._m31(m01 * rm30 + m11 * rm31 + m31)
|
|
._m32(m02 * rm30 + m12 * rm31 + m32)
|
|
._m33(m03 * rm30 + m13 * rm31 + m33)
|
|
._m00(m00 * rm00)
|
|
._m01(m01 * rm00)
|
|
._m02(m02 * rm00)
|
|
._m03(m03 * rm00)
|
|
._m10(m10 * rm11)
|
|
._m11(m11 * rm11)
|
|
._m12(m12 * rm11)
|
|
._m13(m13 * rm11)
|
|
._m20(m20)
|
|
._m21(m21)
|
|
._m22(m22)
|
|
._m23(m23)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply an orthographic projection transformation for a left-handed coordinate system to this matrix.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #orthoLH(float, float, float, float, float, float) orthoLH()} with
|
|
* <code>zNear=-1</code> and <code>zFar=+1</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the orthographic projection matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* orthographic projection transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to an orthographic projection without post-multiplying it,
|
|
* use {@link #setOrtho2DLH(float, float, float, float) setOrtho2DLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #orthoLH(float, float, float, float, float, float)
|
|
* @see #setOrtho2DLH(float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @return this
|
|
*/
|
|
public Matrix4f ortho2DLH(float left, float right, float bottom, float top) {
|
|
return ortho2DLH(left, right, bottom, top, this);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be an orthographic projection transformation for a right-handed coordinate system.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #setOrtho(float, float, float, float, float, float) setOrtho()} with
|
|
* <code>zNear=-1</code> and <code>zFar=+1</code>.
|
|
* <p>
|
|
* In order to apply the orthographic projection to an already existing transformation,
|
|
* use {@link #ortho2D(float, float, float, float) ortho2D()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrtho(float, float, float, float, float, float)
|
|
* @see #ortho2D(float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @return this
|
|
*/
|
|
public Matrix4f setOrtho2D(float left, float right, float bottom, float top) {
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
this._m00(2.0f / (right - left))
|
|
._m11(2.0f / (top - bottom))
|
|
._m22(-1.0f)
|
|
._m30((right + left) / (left - right))
|
|
._m31((top + bottom) / (bottom - top))
|
|
._properties(PROPERTY_AFFINE);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be an orthographic projection transformation for a left-handed coordinate system.
|
|
* <p>
|
|
* This method is equivalent to calling {@link #setOrtho(float, float, float, float, float, float) setOrthoLH()} with
|
|
* <code>zNear=-1</code> and <code>zFar=+1</code>.
|
|
* <p>
|
|
* In order to apply the orthographic projection to an already existing transformation,
|
|
* use {@link #ortho2DLH(float, float, float, float) ortho2DLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#ortho">http://www.songho.ca</a>
|
|
*
|
|
* @see #setOrthoLH(float, float, float, float, float, float)
|
|
* @see #ortho2DLH(float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance from the center to the left frustum edge
|
|
* @param right
|
|
* the distance from the center to the right frustum edge
|
|
* @param bottom
|
|
* the distance from the center to the bottom frustum edge
|
|
* @param top
|
|
* the distance from the center to the top frustum edge
|
|
* @return this
|
|
*/
|
|
public Matrix4f setOrtho2DLH(float left, float right, float bottom, float top) {
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
this._m00(2.0f / (right - left))
|
|
._m11(2.0f / (top - bottom))
|
|
._m30((right + left) / (left - right))
|
|
._m31((top + bottom) / (bottom - top))
|
|
._properties(PROPERTY_AFFINE);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Apply a rotation transformation to this matrix to make <code>-z</code> point along <code>dir</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookalong rotation matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>, the
|
|
* lookalong rotation transformation will be applied first!
|
|
* <p>
|
|
* This is equivalent to calling
|
|
* {@link #lookAt(Vector3fc, Vector3fc, Vector3fc) lookAt}
|
|
* with <code>eye = (0, 0, 0)</code> and <code>center = dir</code>.
|
|
* <p>
|
|
* In order to set the matrix to a lookalong transformation without post-multiplying it,
|
|
* use {@link #setLookAlong(Vector3fc, Vector3fc) setLookAlong()}.
|
|
*
|
|
* @see #lookAlong(float, float, float, float, float, float)
|
|
* @see #lookAt(Vector3fc, Vector3fc, Vector3fc)
|
|
* @see #setLookAlong(Vector3fc, Vector3fc)
|
|
*
|
|
* @param dir
|
|
* the direction in space to look along
|
|
* @param up
|
|
* the direction of 'up'
|
|
* @return this
|
|
*/
|
|
public Matrix4f lookAlong(Vector3fc dir, Vector3fc up) {
|
|
return lookAlong(dir.x(), dir.y(), dir.z(), up.x(), up.y(), up.z(), this);
|
|
}
|
|
|
|
/**
|
|
* Apply a rotation transformation to this matrix to make <code>-z</code> point along <code>dir</code>
|
|
* and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookalong rotation matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>, the
|
|
* lookalong rotation transformation will be applied first!
|
|
* <p>
|
|
* This is equivalent to calling
|
|
* {@link #lookAt(Vector3fc, Vector3fc, Vector3fc) lookAt}
|
|
* with <code>eye = (0, 0, 0)</code> and <code>center = dir</code>.
|
|
* <p>
|
|
* In order to set the matrix to a lookalong transformation without post-multiplying it,
|
|
* use {@link #setLookAlong(Vector3fc, Vector3fc) setLookAlong()}.
|
|
*
|
|
* @see #lookAlong(float, float, float, float, float, float)
|
|
* @see #lookAt(Vector3fc, Vector3fc, Vector3fc)
|
|
* @see #setLookAlong(Vector3fc, Vector3fc)
|
|
*
|
|
* @param dir
|
|
* the direction in space to look along
|
|
* @param up
|
|
* the direction of 'up'
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f lookAlong(Vector3fc dir, Vector3fc up, Matrix4f dest) {
|
|
return lookAlong(dir.x(), dir.y(), dir.z(), up.x(), up.y(), up.z(), dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a rotation transformation to this matrix to make <code>-z</code> point along <code>dir</code>
|
|
* and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookalong rotation matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>, the
|
|
* lookalong rotation transformation will be applied first!
|
|
* <p>
|
|
* This is equivalent to calling
|
|
* {@link #lookAt(float, float, float, float, float, float, float, float, float) lookAt()}
|
|
* with <code>eye = (0, 0, 0)</code> and <code>center = dir</code>.
|
|
* <p>
|
|
* In order to set the matrix to a lookalong transformation without post-multiplying it,
|
|
* use {@link #setLookAlong(float, float, float, float, float, float) setLookAlong()}
|
|
*
|
|
* @see #lookAt(float, float, float, float, float, float, float, float, float)
|
|
* @see #setLookAlong(float, float, float, float, float, float)
|
|
*
|
|
* @param dirX
|
|
* the x-coordinate of the direction to look along
|
|
* @param dirY
|
|
* the y-coordinate of the direction to look along
|
|
* @param dirZ
|
|
* the z-coordinate of the direction to look along
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f lookAlong(float dirX, float dirY, float dirZ, float upX, float upY, float upZ, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setLookAlong(dirX, dirY, dirZ, upX, upY, upZ);
|
|
return lookAlongGeneric(dirX, dirY, dirZ, upX, upY, upZ, dest);
|
|
}
|
|
|
|
private Matrix4f lookAlongGeneric(float dirX, float dirY, float dirZ, float upX, float upY, float upZ, Matrix4f dest) {
|
|
// Normalize direction
|
|
float invDirLength = Math.invsqrt(dirX * dirX + dirY * dirY + dirZ * dirZ);
|
|
dirX *= -invDirLength;
|
|
dirY *= -invDirLength;
|
|
dirZ *= -invDirLength;
|
|
// left = up x direction
|
|
float leftX, leftY, leftZ;
|
|
leftX = upY * dirZ - upZ * dirY;
|
|
leftY = upZ * dirX - upX * dirZ;
|
|
leftZ = upX * dirY - upY * dirX;
|
|
// normalize left
|
|
float invLeftLength = Math.invsqrt(leftX * leftX + leftY * leftY + leftZ * leftZ);
|
|
leftX *= invLeftLength;
|
|
leftY *= invLeftLength;
|
|
leftZ *= invLeftLength;
|
|
// up = direction x left
|
|
float upnX = dirY * leftZ - dirZ * leftY;
|
|
float upnY = dirZ * leftX - dirX * leftZ;
|
|
float upnZ = dirX * leftY - dirY * leftX;
|
|
// perform optimized matrix multiplication
|
|
// introduce temporaries for dependent results
|
|
float nm00 = m00 * leftX + m10 * upnX + m20 * dirX;
|
|
float nm01 = m01 * leftX + m11 * upnX + m21 * dirX;
|
|
float nm02 = m02 * leftX + m12 * upnX + m22 * dirX;
|
|
float nm03 = m03 * leftX + m13 * upnX + m23 * dirX;
|
|
float nm10 = m00 * leftY + m10 * upnY + m20 * dirY;
|
|
float nm11 = m01 * leftY + m11 * upnY + m21 * dirY;
|
|
float nm12 = m02 * leftY + m12 * upnY + m22 * dirY;
|
|
float nm13 = m03 * leftY + m13 * upnY + m23 * dirY;
|
|
return dest
|
|
._m20(m00 * leftZ + m10 * upnZ + m20 * dirZ)
|
|
._m21(m01 * leftZ + m11 * upnZ + m21 * dirZ)
|
|
._m22(m02 * leftZ + m12 * upnZ + m22 * dirZ)
|
|
._m23(m03 * leftZ + m13 * upnZ + m23 * dirZ)
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply a rotation transformation to this matrix to make <code>-z</code> point along <code>dir</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookalong rotation matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>, the
|
|
* lookalong rotation transformation will be applied first!
|
|
* <p>
|
|
* This is equivalent to calling
|
|
* {@link #lookAt(float, float, float, float, float, float, float, float, float) lookAt()}
|
|
* with <code>eye = (0, 0, 0)</code> and <code>center = dir</code>.
|
|
* <p>
|
|
* In order to set the matrix to a lookalong transformation without post-multiplying it,
|
|
* use {@link #setLookAlong(float, float, float, float, float, float) setLookAlong()}
|
|
*
|
|
* @see #lookAt(float, float, float, float, float, float, float, float, float)
|
|
* @see #setLookAlong(float, float, float, float, float, float)
|
|
*
|
|
* @param dirX
|
|
* the x-coordinate of the direction to look along
|
|
* @param dirY
|
|
* the y-coordinate of the direction to look along
|
|
* @param dirZ
|
|
* the z-coordinate of the direction to look along
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @return this
|
|
*/
|
|
public Matrix4f lookAlong(float dirX, float dirY, float dirZ,
|
|
float upX, float upY, float upZ) {
|
|
return lookAlong(dirX, dirY, dirZ, upX, upY, upZ, this);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a rotation transformation to make <code>-z</code>
|
|
* point along <code>dir</code>.
|
|
* <p>
|
|
* This is equivalent to calling
|
|
* {@link #setLookAt(Vector3fc, Vector3fc, Vector3fc) setLookAt()}
|
|
* with <code>eye = (0, 0, 0)</code> and <code>center = dir</code>.
|
|
* <p>
|
|
* In order to apply the lookalong transformation to any previous existing transformation,
|
|
* use {@link #lookAlong(Vector3fc, Vector3fc)}.
|
|
*
|
|
* @see #setLookAlong(Vector3fc, Vector3fc)
|
|
* @see #lookAlong(Vector3fc, Vector3fc)
|
|
*
|
|
* @param dir
|
|
* the direction in space to look along
|
|
* @param up
|
|
* the direction of 'up'
|
|
* @return this
|
|
*/
|
|
public Matrix4f setLookAlong(Vector3fc dir, Vector3fc up) {
|
|
return setLookAlong(dir.x(), dir.y(), dir.z(), up.x(), up.y(), up.z());
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a rotation transformation to make <code>-z</code>
|
|
* point along <code>dir</code>.
|
|
* <p>
|
|
* This is equivalent to calling
|
|
* {@link #setLookAt(float, float, float, float, float, float, float, float, float)
|
|
* setLookAt()} with <code>eye = (0, 0, 0)</code> and <code>center = dir</code>.
|
|
* <p>
|
|
* In order to apply the lookalong transformation to any previous existing transformation,
|
|
* use {@link #lookAlong(float, float, float, float, float, float) lookAlong()}
|
|
*
|
|
* @see #setLookAlong(float, float, float, float, float, float)
|
|
* @see #lookAlong(float, float, float, float, float, float)
|
|
*
|
|
* @param dirX
|
|
* the x-coordinate of the direction to look along
|
|
* @param dirY
|
|
* the y-coordinate of the direction to look along
|
|
* @param dirZ
|
|
* the z-coordinate of the direction to look along
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @return this
|
|
*/
|
|
public Matrix4f setLookAlong(float dirX, float dirY, float dirZ,
|
|
float upX, float upY, float upZ) {
|
|
// Normalize direction
|
|
float invDirLength = Math.invsqrt(dirX * dirX + dirY * dirY + dirZ * dirZ);
|
|
dirX *= -invDirLength;
|
|
dirY *= -invDirLength;
|
|
dirZ *= -invDirLength;
|
|
// left = up x direction
|
|
float leftX, leftY, leftZ;
|
|
leftX = upY * dirZ - upZ * dirY;
|
|
leftY = upZ * dirX - upX * dirZ;
|
|
leftZ = upX * dirY - upY * dirX;
|
|
// normalize left
|
|
float invLeftLength = Math.invsqrt(leftX * leftX + leftY * leftY + leftZ * leftZ);
|
|
leftX *= invLeftLength;
|
|
leftY *= invLeftLength;
|
|
leftZ *= invLeftLength;
|
|
this._m00(leftX)
|
|
._m01(dirY * leftZ - dirZ * leftY)
|
|
._m02(dirX)
|
|
._m03(0.0f)
|
|
._m10(leftY)
|
|
._m11(dirZ * leftX - dirX * leftZ)
|
|
._m12(dirY)
|
|
._m13(0.0f)
|
|
._m20(leftZ)
|
|
._m21(dirX * leftY - dirY * leftX)
|
|
._m22(dirZ)
|
|
._m23(0.0f)
|
|
._m30(0.0f)
|
|
._m31(0.0f)
|
|
._m32(0.0f)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a "lookat" transformation for a right-handed coordinate system, that aligns
|
|
* <code>-z</code> with <code>center - eye</code>.
|
|
* <p>
|
|
* In order to not make use of vectors to specify <code>eye</code>, <code>center</code> and <code>up</code> but use primitives,
|
|
* like in the GLU function, use {@link #setLookAt(float, float, float, float, float, float, float, float, float) setLookAt()}
|
|
* instead.
|
|
* <p>
|
|
* In order to apply the lookat transformation to a previous existing transformation,
|
|
* use {@link #lookAt(Vector3fc, Vector3fc, Vector3fc) lookAt()}.
|
|
*
|
|
* @see #setLookAt(float, float, float, float, float, float, float, float, float)
|
|
* @see #lookAt(Vector3fc, Vector3fc, Vector3fc)
|
|
*
|
|
* @param eye
|
|
* the position of the camera
|
|
* @param center
|
|
* the point in space to look at
|
|
* @param up
|
|
* the direction of 'up'
|
|
* @return this
|
|
*/
|
|
public Matrix4f setLookAt(Vector3fc eye, Vector3fc center, Vector3fc up) {
|
|
return setLookAt(eye.x(), eye.y(), eye.z(), center.x(), center.y(), center.z(), up.x(), up.y(), up.z());
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a "lookat" transformation for a right-handed coordinate system,
|
|
* that aligns <code>-z</code> with <code>center - eye</code>.
|
|
* <p>
|
|
* In order to apply the lookat transformation to a previous existing transformation,
|
|
* use {@link #lookAt(float, float, float, float, float, float, float, float, float) lookAt}.
|
|
*
|
|
* @see #setLookAt(Vector3fc, Vector3fc, Vector3fc)
|
|
* @see #lookAt(float, float, float, float, float, float, float, float, float)
|
|
*
|
|
* @param eyeX
|
|
* the x-coordinate of the eye/camera location
|
|
* @param eyeY
|
|
* the y-coordinate of the eye/camera location
|
|
* @param eyeZ
|
|
* the z-coordinate of the eye/camera location
|
|
* @param centerX
|
|
* the x-coordinate of the point to look at
|
|
* @param centerY
|
|
* the y-coordinate of the point to look at
|
|
* @param centerZ
|
|
* the z-coordinate of the point to look at
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @return this
|
|
*/
|
|
public Matrix4f setLookAt(float eyeX, float eyeY, float eyeZ,
|
|
float centerX, float centerY, float centerZ,
|
|
float upX, float upY, float upZ) {
|
|
// Compute direction from position to lookAt
|
|
float dirX, dirY, dirZ;
|
|
dirX = eyeX - centerX;
|
|
dirY = eyeY - centerY;
|
|
dirZ = eyeZ - centerZ;
|
|
// Normalize direction
|
|
float invDirLength = Math.invsqrt(dirX * dirX + dirY * dirY + dirZ * dirZ);
|
|
dirX *= invDirLength;
|
|
dirY *= invDirLength;
|
|
dirZ *= invDirLength;
|
|
// left = up x direction
|
|
float leftX, leftY, leftZ;
|
|
leftX = upY * dirZ - upZ * dirY;
|
|
leftY = upZ * dirX - upX * dirZ;
|
|
leftZ = upX * dirY - upY * dirX;
|
|
// normalize left
|
|
float invLeftLength = Math.invsqrt(leftX * leftX + leftY * leftY + leftZ * leftZ);
|
|
leftX *= invLeftLength;
|
|
leftY *= invLeftLength;
|
|
leftZ *= invLeftLength;
|
|
// up = direction x left
|
|
float upnX = dirY * leftZ - dirZ * leftY;
|
|
float upnY = dirZ * leftX - dirX * leftZ;
|
|
float upnZ = dirX * leftY - dirY * leftX;
|
|
return this
|
|
._m00(leftX)
|
|
._m01(upnX)
|
|
._m02(dirX)
|
|
._m03(0.0f)
|
|
._m10(leftY)
|
|
._m11(upnY)
|
|
._m12(dirY)
|
|
._m13(0.0f)
|
|
._m20(leftZ)
|
|
._m21(upnZ)
|
|
._m22(dirZ)
|
|
._m23(0.0f)
|
|
._m30(-(leftX * eyeX + leftY * eyeY + leftZ * eyeZ))
|
|
._m31(-(upnX * eyeX + upnY * eyeY + upnZ * eyeZ))
|
|
._m32(-(dirX * eyeX + dirY * eyeY + dirZ * eyeZ))
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
}
|
|
|
|
/**
|
|
* Apply a "lookat" transformation to this matrix for a right-handed coordinate system,
|
|
* that aligns <code>-z</code> with <code>center - eye</code> and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookat matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>,
|
|
* the lookat transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a lookat transformation without post-multiplying it,
|
|
* use {@link #setLookAt(Vector3fc, Vector3fc, Vector3fc)}.
|
|
*
|
|
* @see #lookAt(float, float, float, float, float, float, float, float, float)
|
|
* @see #setLookAlong(Vector3fc, Vector3fc)
|
|
*
|
|
* @param eye
|
|
* the position of the camera
|
|
* @param center
|
|
* the point in space to look at
|
|
* @param up
|
|
* the direction of 'up'
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f lookAt(Vector3fc eye, Vector3fc center, Vector3fc up, Matrix4f dest) {
|
|
return lookAt(eye.x(), eye.y(), eye.z(), center.x(), center.y(), center.z(), up.x(), up.y(), up.z(), dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a "lookat" transformation to this matrix for a right-handed coordinate system,
|
|
* that aligns <code>-z</code> with <code>center - eye</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookat matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>,
|
|
* the lookat transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a lookat transformation without post-multiplying it,
|
|
* use {@link #setLookAt(Vector3fc, Vector3fc, Vector3fc)}.
|
|
*
|
|
* @see #lookAt(float, float, float, float, float, float, float, float, float)
|
|
* @see #setLookAlong(Vector3fc, Vector3fc)
|
|
*
|
|
* @param eye
|
|
* the position of the camera
|
|
* @param center
|
|
* the point in space to look at
|
|
* @param up
|
|
* the direction of 'up'
|
|
* @return this
|
|
*/
|
|
public Matrix4f lookAt(Vector3fc eye, Vector3fc center, Vector3fc up) {
|
|
return lookAt(eye.x(), eye.y(), eye.z(), center.x(), center.y(), center.z(), up.x(), up.y(), up.z(), this);
|
|
}
|
|
|
|
/**
|
|
* Apply a "lookat" transformation to this matrix for a right-handed coordinate system,
|
|
* that aligns <code>-z</code> with <code>center - eye</code> and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookat matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>,
|
|
* the lookat transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a lookat transformation without post-multiplying it,
|
|
* use {@link #setLookAt(float, float, float, float, float, float, float, float, float) setLookAt()}.
|
|
*
|
|
* @see #lookAt(Vector3fc, Vector3fc, Vector3fc)
|
|
* @see #setLookAt(float, float, float, float, float, float, float, float, float)
|
|
*
|
|
* @param eyeX
|
|
* the x-coordinate of the eye/camera location
|
|
* @param eyeY
|
|
* the y-coordinate of the eye/camera location
|
|
* @param eyeZ
|
|
* the z-coordinate of the eye/camera location
|
|
* @param centerX
|
|
* the x-coordinate of the point to look at
|
|
* @param centerY
|
|
* the y-coordinate of the point to look at
|
|
* @param centerZ
|
|
* the z-coordinate of the point to look at
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f lookAt(float eyeX, float eyeY, float eyeZ,
|
|
float centerX, float centerY, float centerZ,
|
|
float upX, float upY, float upZ, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setLookAt(eyeX, eyeY, eyeZ, centerX, centerY, centerZ, upX, upY, upZ);
|
|
else if ((properties & PROPERTY_PERSPECTIVE) != 0)
|
|
return lookAtPerspective(eyeX, eyeY, eyeZ, centerX, centerY, centerZ, upX, upY, upZ, dest);
|
|
return lookAtGeneric(eyeX, eyeY, eyeZ, centerX, centerY, centerZ, upX, upY, upZ, dest);
|
|
}
|
|
private Matrix4f lookAtGeneric(float eyeX, float eyeY, float eyeZ,
|
|
float centerX, float centerY, float centerZ,
|
|
float upX, float upY, float upZ, Matrix4f dest) {
|
|
// Compute direction from position to lookAt
|
|
float dirX, dirY, dirZ;
|
|
dirX = eyeX - centerX;
|
|
dirY = eyeY - centerY;
|
|
dirZ = eyeZ - centerZ;
|
|
// Normalize direction
|
|
float invDirLength = Math.invsqrt(dirX * dirX + dirY * dirY + dirZ * dirZ);
|
|
dirX *= invDirLength;
|
|
dirY *= invDirLength;
|
|
dirZ *= invDirLength;
|
|
// left = up x direction
|
|
float leftX, leftY, leftZ;
|
|
leftX = upY * dirZ - upZ * dirY;
|
|
leftY = upZ * dirX - upX * dirZ;
|
|
leftZ = upX * dirY - upY * dirX;
|
|
// normalize left
|
|
float invLeftLength = Math.invsqrt(leftX * leftX + leftY * leftY + leftZ * leftZ);
|
|
leftX *= invLeftLength;
|
|
leftY *= invLeftLength;
|
|
leftZ *= invLeftLength;
|
|
// up = direction x left
|
|
float upnX = dirY * leftZ - dirZ * leftY;
|
|
float upnY = dirZ * leftX - dirX * leftZ;
|
|
float upnZ = dirX * leftY - dirY * leftX;
|
|
|
|
// calculate right matrix elements
|
|
float rm30 = -(leftX * eyeX + leftY * eyeY + leftZ * eyeZ);
|
|
float rm31 = -(upnX * eyeX + upnY * eyeY + upnZ * eyeZ);
|
|
float rm32 = -(dirX * eyeX + dirY * eyeY + dirZ * eyeZ);
|
|
// introduce temporaries for dependent results
|
|
float nm00 = m00 * leftX + m10 * upnX + m20 * dirX;
|
|
float nm01 = m01 * leftX + m11 * upnX + m21 * dirX;
|
|
float nm02 = m02 * leftX + m12 * upnX + m22 * dirX;
|
|
float nm03 = m03 * leftX + m13 * upnX + m23 * dirX;
|
|
float nm10 = m00 * leftY + m10 * upnY + m20 * dirY;
|
|
float nm11 = m01 * leftY + m11 * upnY + m21 * dirY;
|
|
float nm12 = m02 * leftY + m12 * upnY + m22 * dirY;
|
|
float nm13 = m03 * leftY + m13 * upnY + m23 * dirY;
|
|
|
|
// perform optimized matrix multiplication
|
|
// compute last column first, because others do not depend on it
|
|
return dest
|
|
._m30(m00 * rm30 + m10 * rm31 + m20 * rm32 + m30)
|
|
._m31(m01 * rm30 + m11 * rm31 + m21 * rm32 + m31)
|
|
._m32(m02 * rm30 + m12 * rm31 + m22 * rm32 + m32)
|
|
._m33(m03 * rm30 + m13 * rm31 + m23 * rm32 + m33)
|
|
._m20(m00 * leftZ + m10 * upnZ + m20 * dirZ)
|
|
._m21(m01 * leftZ + m11 * upnZ + m21 * dirZ)
|
|
._m22(m02 * leftZ + m12 * upnZ + m22 * dirZ)
|
|
._m23(m03 * leftZ + m13 * upnZ + m23 * dirZ)
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply a "lookat" transformation to this matrix for a right-handed coordinate system,
|
|
* that aligns <code>-z</code> with <code>center - eye</code> and store the result in <code>dest</code>.
|
|
* <p>
|
|
* This method assumes <code>this</code> to be a perspective transformation, obtained via
|
|
* {@link #frustum(float, float, float, float, float, float) frustum()} or {@link #perspective(float, float, float, float) perspective()} or
|
|
* one of their overloads.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookat matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>,
|
|
* the lookat transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a lookat transformation without post-multiplying it,
|
|
* use {@link #setLookAt(float, float, float, float, float, float, float, float, float) setLookAt()}.
|
|
*
|
|
* @see #setLookAt(float, float, float, float, float, float, float, float, float)
|
|
*
|
|
* @param eyeX
|
|
* the x-coordinate of the eye/camera location
|
|
* @param eyeY
|
|
* the y-coordinate of the eye/camera location
|
|
* @param eyeZ
|
|
* the z-coordinate of the eye/camera location
|
|
* @param centerX
|
|
* the x-coordinate of the point to look at
|
|
* @param centerY
|
|
* the y-coordinate of the point to look at
|
|
* @param centerZ
|
|
* the z-coordinate of the point to look at
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f lookAtPerspective(float eyeX, float eyeY, float eyeZ,
|
|
float centerX, float centerY, float centerZ,
|
|
float upX, float upY, float upZ, Matrix4f dest) {
|
|
// Compute direction from position to lookAt
|
|
float dirX, dirY, dirZ;
|
|
dirX = eyeX - centerX;
|
|
dirY = eyeY - centerY;
|
|
dirZ = eyeZ - centerZ;
|
|
// Normalize direction
|
|
float invDirLength = Math.invsqrt(dirX * dirX + dirY * dirY + dirZ * dirZ);
|
|
dirX *= invDirLength;
|
|
dirY *= invDirLength;
|
|
dirZ *= invDirLength;
|
|
// left = up x direction
|
|
float leftX, leftY, leftZ;
|
|
leftX = upY * dirZ - upZ * dirY;
|
|
leftY = upZ * dirX - upX * dirZ;
|
|
leftZ = upX * dirY - upY * dirX;
|
|
// normalize left
|
|
float invLeftLength = Math.invsqrt(leftX * leftX + leftY * leftY + leftZ * leftZ);
|
|
leftX *= invLeftLength;
|
|
leftY *= invLeftLength;
|
|
leftZ *= invLeftLength;
|
|
// up = direction x left
|
|
float upnX = dirY * leftZ - dirZ * leftY;
|
|
float upnY = dirZ * leftX - dirX * leftZ;
|
|
float upnZ = dirX * leftY - dirY * leftX;
|
|
float rm30 = -(leftX * eyeX + leftY * eyeY + leftZ * eyeZ);
|
|
float rm31 = -(upnX * eyeX + upnY * eyeY + upnZ * eyeZ);
|
|
float rm32 = -(dirX * eyeX + dirY * eyeY + dirZ * eyeZ);
|
|
float nm10 = m00 * leftY;
|
|
float nm20 = m00 * leftZ;
|
|
float nm21 = m11 * upnZ;
|
|
float nm30 = m00 * rm30;
|
|
float nm31 = m11 * rm31;
|
|
float nm32 = m22 * rm32 + m32;
|
|
float nm33 = m23 * rm32;
|
|
return dest
|
|
._m00(m00 * leftX)
|
|
._m01(m11 * upnX)
|
|
._m02(m22 * dirX)
|
|
._m03(m23 * dirX)
|
|
._m10(nm10)
|
|
._m11(m11 * upnY)
|
|
._m12(m22 * dirY)
|
|
._m13(m23 * dirY)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(m22 * dirZ)
|
|
._m23(m23 * dirZ)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(nm33)
|
|
._properties(0);
|
|
}
|
|
|
|
/**
|
|
* Apply a "lookat" transformation to this matrix for a right-handed coordinate system,
|
|
* that aligns <code>-z</code> with <code>center - eye</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookat matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>,
|
|
* the lookat transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a lookat transformation without post-multiplying it,
|
|
* use {@link #setLookAt(float, float, float, float, float, float, float, float, float) setLookAt()}.
|
|
*
|
|
* @see #lookAt(Vector3fc, Vector3fc, Vector3fc)
|
|
* @see #setLookAt(float, float, float, float, float, float, float, float, float)
|
|
*
|
|
* @param eyeX
|
|
* the x-coordinate of the eye/camera location
|
|
* @param eyeY
|
|
* the y-coordinate of the eye/camera location
|
|
* @param eyeZ
|
|
* the z-coordinate of the eye/camera location
|
|
* @param centerX
|
|
* the x-coordinate of the point to look at
|
|
* @param centerY
|
|
* the y-coordinate of the point to look at
|
|
* @param centerZ
|
|
* the z-coordinate of the point to look at
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @return this
|
|
*/
|
|
public Matrix4f lookAt(float eyeX, float eyeY, float eyeZ,
|
|
float centerX, float centerY, float centerZ,
|
|
float upX, float upY, float upZ) {
|
|
return lookAt(eyeX, eyeY, eyeZ, centerX, centerY, centerZ, upX, upY, upZ, this);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a "lookat" transformation for a left-handed coordinate system, that aligns
|
|
* <code>+z</code> with <code>center - eye</code>.
|
|
* <p>
|
|
* In order to not make use of vectors to specify <code>eye</code>, <code>center</code> and <code>up</code> but use primitives,
|
|
* like in the GLU function, use {@link #setLookAtLH(float, float, float, float, float, float, float, float, float) setLookAtLH()}
|
|
* instead.
|
|
* <p>
|
|
* In order to apply the lookat transformation to a previous existing transformation,
|
|
* use {@link #lookAtLH(Vector3fc, Vector3fc, Vector3fc) lookAt()}.
|
|
*
|
|
* @see #setLookAtLH(float, float, float, float, float, float, float, float, float)
|
|
* @see #lookAtLH(Vector3fc, Vector3fc, Vector3fc)
|
|
*
|
|
* @param eye
|
|
* the position of the camera
|
|
* @param center
|
|
* the point in space to look at
|
|
* @param up
|
|
* the direction of 'up'
|
|
* @return this
|
|
*/
|
|
public Matrix4f setLookAtLH(Vector3fc eye, Vector3fc center, Vector3fc up) {
|
|
return setLookAtLH(eye.x(), eye.y(), eye.z(), center.x(), center.y(), center.z(), up.x(), up.y(), up.z());
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a "lookat" transformation for a left-handed coordinate system,
|
|
* that aligns <code>+z</code> with <code>center - eye</code>.
|
|
* <p>
|
|
* In order to apply the lookat transformation to a previous existing transformation,
|
|
* use {@link #lookAtLH(float, float, float, float, float, float, float, float, float) lookAtLH}.
|
|
*
|
|
* @see #setLookAtLH(Vector3fc, Vector3fc, Vector3fc)
|
|
* @see #lookAtLH(float, float, float, float, float, float, float, float, float)
|
|
*
|
|
* @param eyeX
|
|
* the x-coordinate of the eye/camera location
|
|
* @param eyeY
|
|
* the y-coordinate of the eye/camera location
|
|
* @param eyeZ
|
|
* the z-coordinate of the eye/camera location
|
|
* @param centerX
|
|
* the x-coordinate of the point to look at
|
|
* @param centerY
|
|
* the y-coordinate of the point to look at
|
|
* @param centerZ
|
|
* the z-coordinate of the point to look at
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @return this
|
|
*/
|
|
public Matrix4f setLookAtLH(float eyeX, float eyeY, float eyeZ,
|
|
float centerX, float centerY, float centerZ,
|
|
float upX, float upY, float upZ) {
|
|
// Compute direction from position to lookAt
|
|
float dirX, dirY, dirZ;
|
|
dirX = centerX - eyeX;
|
|
dirY = centerY - eyeY;
|
|
dirZ = centerZ - eyeZ;
|
|
// Normalize direction
|
|
float invDirLength = Math.invsqrt(dirX * dirX + dirY * dirY + dirZ * dirZ);
|
|
dirX *= invDirLength;
|
|
dirY *= invDirLength;
|
|
dirZ *= invDirLength;
|
|
// left = up x direction
|
|
float leftX, leftY, leftZ;
|
|
leftX = upY * dirZ - upZ * dirY;
|
|
leftY = upZ * dirX - upX * dirZ;
|
|
leftZ = upX * dirY - upY * dirX;
|
|
// normalize left
|
|
float invLeftLength = Math.invsqrt(leftX * leftX + leftY * leftY + leftZ * leftZ);
|
|
leftX *= invLeftLength;
|
|
leftY *= invLeftLength;
|
|
leftZ *= invLeftLength;
|
|
// up = direction x left
|
|
float upnX = dirY * leftZ - dirZ * leftY;
|
|
float upnY = dirZ * leftX - dirX * leftZ;
|
|
float upnZ = dirX * leftY - dirY * leftX;
|
|
this._m00(leftX)
|
|
._m01(upnX)
|
|
._m02(dirX)
|
|
._m03(0.0f)
|
|
._m10(leftY)
|
|
._m11(upnY)
|
|
._m12(dirY)
|
|
._m13(0.0f)
|
|
._m20(leftZ)
|
|
._m21(upnZ)
|
|
._m22(dirZ)
|
|
._m23(0.0f)
|
|
._m30(-(leftX * eyeX + leftY * eyeY + leftZ * eyeZ))
|
|
._m31(-(upnX * eyeX + upnY * eyeY + upnZ * eyeZ))
|
|
._m32(-(dirX * eyeX + dirY * eyeY + dirZ * eyeZ))
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Apply a "lookat" transformation to this matrix for a left-handed coordinate system,
|
|
* that aligns <code>+z</code> with <code>center - eye</code> and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookat matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>,
|
|
* the lookat transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a lookat transformation without post-multiplying it,
|
|
* use {@link #setLookAtLH(Vector3fc, Vector3fc, Vector3fc)}.
|
|
*
|
|
* @see #lookAtLH(float, float, float, float, float, float, float, float, float)
|
|
*
|
|
* @param eye
|
|
* the position of the camera
|
|
* @param center
|
|
* the point in space to look at
|
|
* @param up
|
|
* the direction of 'up'
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f lookAtLH(Vector3fc eye, Vector3fc center, Vector3fc up, Matrix4f dest) {
|
|
return lookAtLH(eye.x(), eye.y(), eye.z(), center.x(), center.y(), center.z(), up.x(), up.y(), up.z(), dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a "lookat" transformation to this matrix for a left-handed coordinate system,
|
|
* that aligns <code>+z</code> with <code>center - eye</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookat matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>,
|
|
* the lookat transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a lookat transformation without post-multiplying it,
|
|
* use {@link #setLookAtLH(Vector3fc, Vector3fc, Vector3fc)}.
|
|
*
|
|
* @see #lookAtLH(float, float, float, float, float, float, float, float, float)
|
|
*
|
|
* @param eye
|
|
* the position of the camera
|
|
* @param center
|
|
* the point in space to look at
|
|
* @param up
|
|
* the direction of 'up'
|
|
* @return this
|
|
*/
|
|
public Matrix4f lookAtLH(Vector3fc eye, Vector3fc center, Vector3fc up) {
|
|
return lookAtLH(eye.x(), eye.y(), eye.z(), center.x(), center.y(), center.z(), up.x(), up.y(), up.z(), this);
|
|
}
|
|
|
|
/**
|
|
* Apply a "lookat" transformation to this matrix for a left-handed coordinate system,
|
|
* that aligns <code>+z</code> with <code>center - eye</code> and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookat matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>,
|
|
* the lookat transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a lookat transformation without post-multiplying it,
|
|
* use {@link #setLookAtLH(float, float, float, float, float, float, float, float, float) setLookAtLH()}.
|
|
*
|
|
* @see #lookAtLH(Vector3fc, Vector3fc, Vector3fc)
|
|
* @see #setLookAtLH(float, float, float, float, float, float, float, float, float)
|
|
*
|
|
* @param eyeX
|
|
* the x-coordinate of the eye/camera location
|
|
* @param eyeY
|
|
* the y-coordinate of the eye/camera location
|
|
* @param eyeZ
|
|
* the z-coordinate of the eye/camera location
|
|
* @param centerX
|
|
* the x-coordinate of the point to look at
|
|
* @param centerY
|
|
* the y-coordinate of the point to look at
|
|
* @param centerZ
|
|
* the z-coordinate of the point to look at
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f lookAtLH(float eyeX, float eyeY, float eyeZ,
|
|
float centerX, float centerY, float centerZ,
|
|
float upX, float upY, float upZ, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setLookAtLH(eyeX, eyeY, eyeZ, centerX, centerY, centerZ, upX, upY, upZ);
|
|
else if ((properties & PROPERTY_PERSPECTIVE) != 0)
|
|
return lookAtPerspectiveLH(eyeX, eyeY, eyeZ, centerX, centerY, centerZ, upX, upY, upZ, dest);
|
|
return lookAtLHGeneric(eyeX, eyeY, eyeZ, centerX, centerY, centerZ, upX, upY, upZ, dest);
|
|
}
|
|
private Matrix4f lookAtLHGeneric(float eyeX, float eyeY, float eyeZ,
|
|
float centerX, float centerY, float centerZ,
|
|
float upX, float upY, float upZ, Matrix4f dest) {
|
|
// Compute direction from position to lookAt
|
|
float dirX, dirY, dirZ;
|
|
dirX = centerX - eyeX;
|
|
dirY = centerY - eyeY;
|
|
dirZ = centerZ - eyeZ;
|
|
// Normalize direction
|
|
float invDirLength = Math.invsqrt(dirX * dirX + dirY * dirY + dirZ * dirZ);
|
|
dirX *= invDirLength;
|
|
dirY *= invDirLength;
|
|
dirZ *= invDirLength;
|
|
// left = up x direction
|
|
float leftX, leftY, leftZ;
|
|
leftX = upY * dirZ - upZ * dirY;
|
|
leftY = upZ * dirX - upX * dirZ;
|
|
leftZ = upX * dirY - upY * dirX;
|
|
// normalize left
|
|
float invLeftLength = Math.invsqrt(leftX * leftX + leftY * leftY + leftZ * leftZ);
|
|
leftX *= invLeftLength;
|
|
leftY *= invLeftLength;
|
|
leftZ *= invLeftLength;
|
|
// up = direction x left
|
|
float upnX = dirY * leftZ - dirZ * leftY;
|
|
float upnY = dirZ * leftX - dirX * leftZ;
|
|
float upnZ = dirX * leftY - dirY * leftX;
|
|
|
|
// calculate right matrix elements
|
|
float rm30 = -(leftX * eyeX + leftY * eyeY + leftZ * eyeZ);
|
|
float rm31 = -(upnX * eyeX + upnY * eyeY + upnZ * eyeZ);
|
|
float rm32 = -(dirX * eyeX + dirY * eyeY + dirZ * eyeZ);
|
|
// introduce temporaries for dependent results
|
|
float nm00 = m00 * leftX + m10 * upnX + m20 * dirX;
|
|
float nm01 = m01 * leftX + m11 * upnX + m21 * dirX;
|
|
float nm02 = m02 * leftX + m12 * upnX + m22 * dirX;
|
|
float nm03 = m03 * leftX + m13 * upnX + m23 * dirX;
|
|
float nm10 = m00 * leftY + m10 * upnY + m20 * dirY;
|
|
float nm11 = m01 * leftY + m11 * upnY + m21 * dirY;
|
|
float nm12 = m02 * leftY + m12 * upnY + m22 * dirY;
|
|
float nm13 = m03 * leftY + m13 * upnY + m23 * dirY;
|
|
|
|
// perform optimized matrix multiplication
|
|
// compute last column first, because others do not depend on it
|
|
return dest
|
|
._m30(m00 * rm30 + m10 * rm31 + m20 * rm32 + m30)
|
|
._m31(m01 * rm30 + m11 * rm31 + m21 * rm32 + m31)
|
|
._m32(m02 * rm30 + m12 * rm31 + m22 * rm32 + m32)
|
|
._m33(m03 * rm30 + m13 * rm31 + m23 * rm32 + m33)
|
|
._m20(m00 * leftZ + m10 * upnZ + m20 * dirZ)
|
|
._m21(m01 * leftZ + m11 * upnZ + m21 * dirZ)
|
|
._m22(m02 * leftZ + m12 * upnZ + m22 * dirZ)
|
|
._m23(m03 * leftZ + m13 * upnZ + m23 * dirZ)
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply a "lookat" transformation to this matrix for a left-handed coordinate system,
|
|
* that aligns <code>+z</code> with <code>center - eye</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookat matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>,
|
|
* the lookat transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a lookat transformation without post-multiplying it,
|
|
* use {@link #setLookAtLH(float, float, float, float, float, float, float, float, float) setLookAtLH()}.
|
|
*
|
|
* @see #lookAtLH(Vector3fc, Vector3fc, Vector3fc)
|
|
* @see #setLookAtLH(float, float, float, float, float, float, float, float, float)
|
|
*
|
|
* @param eyeX
|
|
* the x-coordinate of the eye/camera location
|
|
* @param eyeY
|
|
* the y-coordinate of the eye/camera location
|
|
* @param eyeZ
|
|
* the z-coordinate of the eye/camera location
|
|
* @param centerX
|
|
* the x-coordinate of the point to look at
|
|
* @param centerY
|
|
* the y-coordinate of the point to look at
|
|
* @param centerZ
|
|
* the z-coordinate of the point to look at
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @return this
|
|
*/
|
|
public Matrix4f lookAtLH(float eyeX, float eyeY, float eyeZ,
|
|
float centerX, float centerY, float centerZ,
|
|
float upX, float upY, float upZ) {
|
|
return lookAtLH(eyeX, eyeY, eyeZ, centerX, centerY, centerZ, upX, upY, upZ, this);
|
|
}
|
|
|
|
/**
|
|
* Apply a "lookat" transformation to this matrix for a left-handed coordinate system,
|
|
* that aligns <code>+z</code> with <code>center - eye</code> and store the result in <code>dest</code>.
|
|
* <p>
|
|
* This method assumes <code>this</code> to be a perspective transformation, obtained via
|
|
* {@link #frustumLH(float, float, float, float, float, float) frustumLH()} or {@link #perspectiveLH(float, float, float, float) perspectiveLH()} or
|
|
* one of their overloads.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookat matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>,
|
|
* the lookat transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a lookat transformation without post-multiplying it,
|
|
* use {@link #setLookAtLH(float, float, float, float, float, float, float, float, float) setLookAtLH()}.
|
|
*
|
|
* @see #setLookAtLH(float, float, float, float, float, float, float, float, float)
|
|
*
|
|
* @param eyeX
|
|
* the x-coordinate of the eye/camera location
|
|
* @param eyeY
|
|
* the y-coordinate of the eye/camera location
|
|
* @param eyeZ
|
|
* the z-coordinate of the eye/camera location
|
|
* @param centerX
|
|
* the x-coordinate of the point to look at
|
|
* @param centerY
|
|
* the y-coordinate of the point to look at
|
|
* @param centerZ
|
|
* the z-coordinate of the point to look at
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f lookAtPerspectiveLH(float eyeX, float eyeY, float eyeZ,
|
|
float centerX, float centerY, float centerZ,
|
|
float upX, float upY, float upZ, Matrix4f dest) {
|
|
// Compute direction from position to lookAt
|
|
float dirX, dirY, dirZ;
|
|
dirX = centerX - eyeX;
|
|
dirY = centerY - eyeY;
|
|
dirZ = centerZ - eyeZ;
|
|
// Normalize direction
|
|
float invDirLength = Math.invsqrt(dirX * dirX + dirY * dirY + dirZ * dirZ);
|
|
dirX *= invDirLength;
|
|
dirY *= invDirLength;
|
|
dirZ *= invDirLength;
|
|
// left = up x direction
|
|
float leftX, leftY, leftZ;
|
|
leftX = upY * dirZ - upZ * dirY;
|
|
leftY = upZ * dirX - upX * dirZ;
|
|
leftZ = upX * dirY - upY * dirX;
|
|
// normalize left
|
|
float invLeftLength = Math.invsqrt(leftX * leftX + leftY * leftY + leftZ * leftZ);
|
|
leftX *= invLeftLength;
|
|
leftY *= invLeftLength;
|
|
leftZ *= invLeftLength;
|
|
// up = direction x left
|
|
float upnX = dirY * leftZ - dirZ * leftY;
|
|
float upnY = dirZ * leftX - dirX * leftZ;
|
|
float upnZ = dirX * leftY - dirY * leftX;
|
|
|
|
// calculate right matrix elements
|
|
float rm30 = -(leftX * eyeX + leftY * eyeY + leftZ * eyeZ);
|
|
float rm31 = -(upnX * eyeX + upnY * eyeY + upnZ * eyeZ);
|
|
float rm32 = -(dirX * eyeX + dirY * eyeY + dirZ * eyeZ);
|
|
|
|
float nm00 = m00 * leftX;
|
|
float nm01 = m11 * upnX;
|
|
float nm02 = m22 * dirX;
|
|
float nm03 = m23 * dirX;
|
|
float nm10 = m00 * leftY;
|
|
float nm11 = m11 * upnY;
|
|
float nm12 = m22 * dirY;
|
|
float nm13 = m23 * dirY;
|
|
float nm20 = m00 * leftZ;
|
|
float nm21 = m11 * upnZ;
|
|
float nm22 = m22 * dirZ;
|
|
float nm23 = m23 * dirZ;
|
|
float nm30 = m00 * rm30;
|
|
float nm31 = m11 * rm31;
|
|
float nm32 = m22 * rm32 + m32;
|
|
float nm33 = m23 * rm32;
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(nm33)
|
|
._properties(0);
|
|
}
|
|
|
|
/**
|
|
* This method is equivalent to calling: <code>translate(w-1-2*x, h-1-2*y, 0).scale(w, h, 1)</code>
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>T</code> the created transformation matrix,
|
|
* then the new matrix will be <code>M * T</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * T * v</code>, the
|
|
* created transformation will be applied first!
|
|
*
|
|
* @param x
|
|
* the tile's x coordinate/index (should be in <code>[0..w)</code>)
|
|
* @param y
|
|
* the tile's y coordinate/index (should be in <code>[0..h)</code>)
|
|
* @param w
|
|
* the number of tiles along the x axis
|
|
* @param h
|
|
* the number of tiles along the y axis
|
|
* @return this
|
|
*/
|
|
public Matrix4f tile(int x, int y, int w, int h) {
|
|
return tile(x, y, w, h, this);
|
|
}
|
|
public Matrix4f tile(int x, int y, int w, int h, Matrix4f dest) {
|
|
float tx = w - 1 - (x<<1), ty = h - 1 - (y<<1);
|
|
return dest
|
|
._m30(Math.fma(m00, tx, Math.fma(m10, ty, m30)))
|
|
._m31(Math.fma(m01, tx, Math.fma(m11, ty, m31)))
|
|
._m32(Math.fma(m02, tx, Math.fma(m12, ty, m32)))
|
|
._m33(Math.fma(m03, tx, Math.fma(m13, ty, m33)))
|
|
._m00(m00 * w)
|
|
._m01(m01 * w)
|
|
._m02(m02 * w)
|
|
._m03(m03 * w)
|
|
._m10(m10 * h)
|
|
._m11(m11 * h)
|
|
._m12(m12 * h)
|
|
._m13(m13 * h)
|
|
._m20(m20)
|
|
._m21(m21)
|
|
._m22(m22)
|
|
._m23(m23)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric perspective projection frustum transformation for a right-handed coordinate system
|
|
* using the given NDC z range to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspective(float, float, float, float, boolean) setPerspective}.
|
|
*
|
|
* @see #setPerspective(float, float, float, float, boolean)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param dest
|
|
* will hold the result
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return dest
|
|
*/
|
|
public Matrix4f perspective(float fovy, float aspect, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setPerspective(fovy, aspect, zNear, zFar, zZeroToOne);
|
|
return perspectiveGeneric(fovy, aspect, zNear, zFar, zZeroToOne, dest);
|
|
}
|
|
private Matrix4f perspectiveGeneric(float fovy, float aspect, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
float h = Math.tan(fovy * 0.5f);
|
|
// calculate right matrix elements
|
|
float rm00 = 1.0f / (h * aspect);
|
|
float rm11 = 1.0f / h;
|
|
float rm22;
|
|
float rm32;
|
|
boolean farInf = zFar > 0 && Float.isInfinite(zFar);
|
|
boolean nearInf = zNear > 0 && Float.isInfinite(zNear);
|
|
if (farInf) {
|
|
// See: "Infinite Projection Matrix" (http://www.terathon.com/gdc07_lengyel.pdf)
|
|
float e = 1E-6f;
|
|
rm22 = e - 1.0f;
|
|
rm32 = (e - (zZeroToOne ? 1.0f : 2.0f)) * zNear;
|
|
} else if (nearInf) {
|
|
float e = 1E-6f;
|
|
rm22 = (zZeroToOne ? 0.0f : 1.0f) - e;
|
|
rm32 = ((zZeroToOne ? 1.0f : 2.0f) - e) * zFar;
|
|
} else {
|
|
rm22 = (zZeroToOne ? zFar : zFar + zNear) / (zNear - zFar);
|
|
rm32 = (zZeroToOne ? zFar : zFar + zFar) * zNear / (zNear - zFar);
|
|
}
|
|
// perform optimized matrix multiplication
|
|
float nm20 = m20 * rm22 - m30;
|
|
float nm21 = m21 * rm22 - m31;
|
|
float nm22 = m22 * rm22 - m32;
|
|
float nm23 = m23 * rm22 - m33;
|
|
dest._m00(m00 * rm00)
|
|
._m01(m01 * rm00)
|
|
._m02(m02 * rm00)
|
|
._m03(m03 * rm00)
|
|
._m10(m10 * rm11)
|
|
._m11(m11 * rm11)
|
|
._m12(m12 * rm11)
|
|
._m13(m13 * rm11)
|
|
._m30(m20 * rm32)
|
|
._m31(m21 * rm32)
|
|
._m32(m22 * rm32)
|
|
._m33(m23 * rm32)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._properties(properties & ~(PROPERTY_AFFINE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric perspective projection frustum transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspective(float, float, float, float) setPerspective}.
|
|
*
|
|
* @see #setPerspective(float, float, float, float)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f perspective(float fovy, float aspect, float zNear, float zFar, Matrix4f dest) {
|
|
return perspective(fovy, aspect, zNear, zFar, false, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric perspective projection frustum transformation using for a right-handed coordinate system
|
|
* the given NDC z range to this matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspective(float, float, float, float, boolean) setPerspective}.
|
|
*
|
|
* @see #setPerspective(float, float, float, float, boolean)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f perspective(float fovy, float aspect, float zNear, float zFar, boolean zZeroToOne) {
|
|
return perspective(fovy, aspect, zNear, zFar, zZeroToOne, this);
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric perspective projection frustum transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspective(float, float, float, float) setPerspective}.
|
|
*
|
|
* @see #setPerspective(float, float, float, float)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @return this
|
|
*/
|
|
public Matrix4f perspective(float fovy, float aspect, float zNear, float zFar) {
|
|
return perspective(fovy, aspect, zNear, zFar, this);
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric perspective projection frustum transformation for a right-handed coordinate system
|
|
* using the given NDC z range to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspectiveRect(float, float, float, float, boolean) setPerspectiveRect}.
|
|
*
|
|
* @see #setPerspectiveRect(float, float, float, float, boolean)
|
|
*
|
|
* @param width
|
|
* the width of the near frustum plane
|
|
* @param height
|
|
* the height of the near frustum plane
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param dest
|
|
* will hold the result
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return dest
|
|
*/
|
|
public Matrix4f perspectiveRect(float width, float height, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setPerspectiveRect(width, height, zNear, zFar, zZeroToOne);
|
|
return perspectiveRectGeneric(width, height, zNear, zFar, zZeroToOne, dest);
|
|
}
|
|
private Matrix4f perspectiveRectGeneric(float width, float height, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
float rm00 = (zNear + zNear) / width;
|
|
float rm11 = (zNear + zNear) / height;
|
|
float rm22, rm32;
|
|
boolean farInf = zFar > 0 && Float.isInfinite(zFar);
|
|
boolean nearInf = zNear > 0 && Float.isInfinite(zNear);
|
|
if (farInf) {
|
|
// See: "Infinite Projection Matrix" (http://www.terathon.com/gdc07_lengyel.pdf)
|
|
float e = 1E-6f;
|
|
rm22 = e - 1.0f;
|
|
rm32 = (e - (zZeroToOne ? 1.0f : 2.0f)) * zNear;
|
|
} else if (nearInf) {
|
|
float e = 1E-6f;
|
|
rm22 = (zZeroToOne ? 0.0f : 1.0f) - e;
|
|
rm32 = ((zZeroToOne ? 1.0f : 2.0f) - e) * zFar;
|
|
} else {
|
|
rm22 = (zZeroToOne ? zFar : zFar + zNear) / (zNear - zFar);
|
|
rm32 = (zZeroToOne ? zFar : zFar + zFar) * zNear / (zNear - zFar);
|
|
}
|
|
// perform optimized matrix multiplication
|
|
float nm20 = m20 * rm22 - m30;
|
|
float nm21 = m21 * rm22 - m31;
|
|
float nm22 = m22 * rm22 - m32;
|
|
float nm23 = m23 * rm22 - m33;
|
|
dest._m00(m00 * rm00)
|
|
._m01(m01 * rm00)
|
|
._m02(m02 * rm00)
|
|
._m03(m03 * rm00)
|
|
._m10(m10 * rm11)
|
|
._m11(m11 * rm11)
|
|
._m12(m12 * rm11)
|
|
._m13(m13 * rm11)
|
|
._m30(m20 * rm32)
|
|
._m31(m21 * rm32)
|
|
._m32(m22 * rm32)
|
|
._m33(m23 * rm32)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._properties(properties & ~(PROPERTY_AFFINE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric perspective projection frustum transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspectiveRect(float, float, float, float) setPerspectiveRect}.
|
|
*
|
|
* @see #setPerspectiveRect(float, float, float, float)
|
|
*
|
|
* @param width
|
|
* the width of the near frustum plane
|
|
* @param height
|
|
* the height of the near frustum plane
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f perspectiveRect(float width, float height, float zNear, float zFar, Matrix4f dest) {
|
|
return perspectiveRect(width, height, zNear, zFar, false, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric perspective projection frustum transformation using for a right-handed coordinate system
|
|
* the given NDC z range to this matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspectiveRect(float, float, float, float, boolean) setPerspectiveRect}.
|
|
*
|
|
* @see #setPerspectiveRect(float, float, float, float, boolean)
|
|
*
|
|
* @param width
|
|
* the width of the near frustum plane
|
|
* @param height
|
|
* the height of the near frustum plane
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f perspectiveRect(float width, float height, float zNear, float zFar, boolean zZeroToOne) {
|
|
return perspectiveRect(width, height, zNear, zFar, zZeroToOne, this);
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric perspective projection frustum transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspectiveRect(float, float, float, float) setPerspectiveRect}.
|
|
*
|
|
* @see #setPerspectiveRect(float, float, float, float)
|
|
*
|
|
* @param width
|
|
* the width of the near frustum plane
|
|
* @param height
|
|
* the height of the near frustum plane
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @return this
|
|
*/
|
|
public Matrix4f perspectiveRect(float width, float height, float zNear, float zFar) {
|
|
return perspectiveRect(width, height, zNear, zFar, this);
|
|
}
|
|
|
|
/**
|
|
* Apply an asymmetric off-center perspective projection frustum transformation for a right-handed coordinate system
|
|
* using the given NDC z range to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* The given angles <code>offAngleX</code> and <code>offAngleY</code> are the horizontal and vertical angles between
|
|
* the line of sight and the line given by the center of the near and far frustum planes. So, when <code>offAngleY</code>
|
|
* is just <code>fovy/2</code> then the projection frustum is rotated towards +Y and the bottom frustum plane
|
|
* is parallel to the XZ-plane.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspectiveOffCenter(float, float, float, float, float, float, boolean) setPerspectiveOffCenter}.
|
|
*
|
|
* @see #setPerspectiveOffCenter(float, float, float, float, float, float, boolean)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param offAngleX
|
|
* the horizontal angle between the line of sight and the line crossing the center of the near and far frustum planes
|
|
* @param offAngleY
|
|
* the vertical angle between the line of sight and the line crossing the center of the near and far frustum planes
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param dest
|
|
* will hold the result
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return dest
|
|
*/
|
|
public Matrix4f perspectiveOffCenter(float fovy, float offAngleX, float offAngleY, float aspect, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setPerspectiveOffCenter(fovy, offAngleX, offAngleY, aspect, zNear, zFar, zZeroToOne);
|
|
return perspectiveOffCenterGeneric(fovy, offAngleX, offAngleY, aspect, zNear, zFar, zZeroToOne, dest);
|
|
}
|
|
private Matrix4f perspectiveOffCenterGeneric(float fovy, float offAngleX, float offAngleY, float aspect, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
float h = Math.tan(fovy * 0.5f);
|
|
// calculate right matrix elements
|
|
float xScale = 1.0f / (h * aspect);
|
|
float yScale = 1.0f / h;
|
|
float offX = Math.tan(offAngleX), offY = Math.tan(offAngleY);
|
|
float rm20 = offX * xScale;
|
|
float rm21 = offY * yScale;
|
|
float rm22;
|
|
float rm32;
|
|
boolean farInf = zFar > 0 && Float.isInfinite(zFar);
|
|
boolean nearInf = zNear > 0 && Float.isInfinite(zNear);
|
|
if (farInf) {
|
|
// See: "Infinite Projection Matrix" (http://www.terathon.com/gdc07_lengyel.pdf)
|
|
float e = 1E-6f;
|
|
rm22 = e - 1.0f;
|
|
rm32 = (e - (zZeroToOne ? 1.0f : 2.0f)) * zNear;
|
|
} else if (nearInf) {
|
|
float e = 1E-6f;
|
|
rm22 = (zZeroToOne ? 0.0f : 1.0f) - e;
|
|
rm32 = ((zZeroToOne ? 1.0f : 2.0f) - e) * zFar;
|
|
} else {
|
|
rm22 = (zZeroToOne ? zFar : zFar + zNear) / (zNear - zFar);
|
|
rm32 = (zZeroToOne ? zFar : zFar + zFar) * zNear / (zNear - zFar);
|
|
}
|
|
// perform optimized matrix multiplication
|
|
float nm20 = m00 * rm20 + m10 * rm21 + m20 * rm22 - m30;
|
|
float nm21 = m01 * rm20 + m11 * rm21 + m21 * rm22 - m31;
|
|
float nm22 = m02 * rm20 + m12 * rm21 + m22 * rm22 - m32;
|
|
float nm23 = m03 * rm20 + m13 * rm21 + m23 * rm22 - m33;
|
|
dest._m00(m00 * xScale)
|
|
._m01(m01 * xScale)
|
|
._m02(m02 * xScale)
|
|
._m03(m03 * xScale)
|
|
._m10(m10 * yScale)
|
|
._m11(m11 * yScale)
|
|
._m12(m12 * yScale)
|
|
._m13(m13 * yScale)
|
|
._m30(m20 * rm32)
|
|
._m31(m21 * rm32)
|
|
._m32(m22 * rm32)
|
|
._m33(m23 * rm32)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._properties(properties & ~(PROPERTY_AFFINE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION
|
|
| PROPERTY_ORTHONORMAL | (rm20 == 0.0f && rm21 == 0.0f ? 0 : PROPERTY_PERSPECTIVE)));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply an asymmetric off-center perspective projection frustum transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* The given angles <code>offAngleX</code> and <code>offAngleY</code> are the horizontal and vertical angles between
|
|
* the line of sight and the line given by the center of the near and far frustum planes. So, when <code>offAngleY</code>
|
|
* is just <code>fovy/2</code> then the projection frustum is rotated towards +Y and the bottom frustum plane
|
|
* is parallel to the XZ-plane.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspectiveOffCenter(float, float, float, float, float, float) setPerspectiveOffCenter}.
|
|
*
|
|
* @see #setPerspectiveOffCenter(float, float, float, float, float, float)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param offAngleX
|
|
* the horizontal angle between the line of sight and the line crossing the center of the near and far frustum planes
|
|
* @param offAngleY
|
|
* the vertical angle between the line of sight and the line crossing the center of the near and far frustum planes
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f perspectiveOffCenter(float fovy, float offAngleX, float offAngleY, float aspect, float zNear, float zFar, Matrix4f dest) {
|
|
return perspectiveOffCenter(fovy, offAngleX, offAngleY, aspect, zNear, zFar, false, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply an asymmetric off-center perspective projection frustum transformation using for a right-handed coordinate system
|
|
* the given NDC z range to this matrix.
|
|
* <p>
|
|
* The given angles <code>offAngleX</code> and <code>offAngleY</code> are the horizontal and vertical angles between
|
|
* the line of sight and the line given by the center of the near and far frustum planes. So, when <code>offAngleY</code>
|
|
* is just <code>fovy/2</code> then the projection frustum is rotated towards +Y and the bottom frustum plane
|
|
* is parallel to the XZ-plane.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspectiveOffCenter(float, float, float, float, float, float, boolean) setPerspectiveOffCenter}.
|
|
*
|
|
* @see #setPerspectiveOffCenter(float, float, float, float, float, float, boolean)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param offAngleX
|
|
* the horizontal angle between the line of sight and the line crossing the center of the near and far frustum planes
|
|
* @param offAngleY
|
|
* the vertical angle between the line of sight and the line crossing the center of the near and far frustum planes
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f perspectiveOffCenter(float fovy, float offAngleX, float offAngleY, float aspect, float zNear, float zFar, boolean zZeroToOne) {
|
|
return perspectiveOffCenter(fovy, offAngleX, offAngleY, aspect, zNear, zFar, zZeroToOne, this);
|
|
}
|
|
|
|
/**
|
|
* Apply an asymmetric off-center perspective projection frustum transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix.
|
|
* <p>
|
|
* The given angles <code>offAngleX</code> and <code>offAngleY</code> are the horizontal and vertical angles between
|
|
* the line of sight and the line given by the center of the near and far frustum planes. So, when <code>offAngleY</code>
|
|
* is just <code>fovy/2</code> then the projection frustum is rotated towards +Y and the bottom frustum plane
|
|
* is parallel to the XZ-plane.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspectiveOffCenter(float, float, float, float, float, float) setPerspectiveOffCenter}.
|
|
*
|
|
* @see #setPerspectiveOffCenter(float, float, float, float, float, float)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param offAngleX
|
|
* the horizontal angle between the line of sight and the line crossing the center of the near and far frustum planes
|
|
* @param offAngleY
|
|
* the vertical angle between the line of sight and the line crossing the center of the near and far frustum planes
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @return this
|
|
*/
|
|
public Matrix4f perspectiveOffCenter(float fovy, float offAngleX, float offAngleY, float aspect, float zNear, float zFar) {
|
|
return perspectiveOffCenter(fovy, offAngleX, offAngleY, aspect, zNear, zFar, this);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a symmetric perspective projection frustum transformation for a right-handed coordinate system
|
|
* using the given NDC z range.
|
|
* <p>
|
|
* In order to apply the perspective projection transformation to an existing transformation,
|
|
* use {@link #perspective(float, float, float, float, boolean) perspective()}.
|
|
*
|
|
* @see #perspective(float, float, float, float, boolean)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f setPerspective(float fovy, float aspect, float zNear, float zFar, boolean zZeroToOne) {
|
|
MemUtil.INSTANCE.zero(this);
|
|
float h = Math.tan(fovy * 0.5f);
|
|
this._m00(1.0f / (h * aspect))
|
|
._m11(1.0f / h);
|
|
boolean farInf = zFar > 0 && Float.isInfinite(zFar);
|
|
boolean nearInf = zNear > 0 && Float.isInfinite(zNear);
|
|
if (farInf) {
|
|
// See: "Infinite Projection Matrix" (http://www.terathon.com/gdc07_lengyel.pdf)
|
|
float e = 1E-6f;
|
|
this._m22(e - 1.0f)
|
|
._m32((e - (zZeroToOne ? 1.0f : 2.0f)) * zNear);
|
|
} else if (nearInf) {
|
|
float e = 1E-6f;
|
|
this._m22((zZeroToOne ? 0.0f : 1.0f) - e)
|
|
._m32(((zZeroToOne ? 1.0f : 2.0f) - e) * zFar);
|
|
} else {
|
|
this._m22((zZeroToOne ? zFar : zFar + zNear) / (zNear - zFar))
|
|
._m32((zZeroToOne ? zFar : zFar + zFar) * zNear / (zNear - zFar));
|
|
}
|
|
return this
|
|
._m23(-1.0f)
|
|
._properties(PROPERTY_PERSPECTIVE);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a symmetric perspective projection frustum transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code>.
|
|
* <p>
|
|
* In order to apply the perspective projection transformation to an existing transformation,
|
|
* use {@link #perspective(float, float, float, float) perspective()}.
|
|
*
|
|
* @see #perspective(float, float, float, float)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @return this
|
|
*/
|
|
public Matrix4f setPerspective(float fovy, float aspect, float zNear, float zFar) {
|
|
return setPerspective(fovy, aspect, zNear, zFar, false);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a symmetric perspective projection frustum transformation for a right-handed coordinate system
|
|
* using the given NDC z range.
|
|
* <p>
|
|
* In order to apply the perspective projection transformation to an existing transformation,
|
|
* use {@link #perspectiveRect(float, float, float, float, boolean) perspectiveRect()}.
|
|
*
|
|
* @see #perspectiveRect(float, float, float, float, boolean)
|
|
*
|
|
* @param width
|
|
* the width of the near frustum plane
|
|
* @param height
|
|
* the height of the near frustum plane
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f setPerspectiveRect(float width, float height, float zNear, float zFar, boolean zZeroToOne) {
|
|
MemUtil.INSTANCE.zero(this);
|
|
this._m00((zNear + zNear) / width)
|
|
._m11((zNear + zNear) / height);
|
|
boolean farInf = zFar > 0 && Float.isInfinite(zFar);
|
|
boolean nearInf = zNear > 0 && Float.isInfinite(zNear);
|
|
if (farInf) {
|
|
// See: "Infinite Projection Matrix" (http://www.terathon.com/gdc07_lengyel.pdf)
|
|
float e = 1E-6f;
|
|
this._m22(e - 1.0f)
|
|
._m32((e - (zZeroToOne ? 1.0f : 2.0f)) * zNear);
|
|
} else if (nearInf) {
|
|
float e = 1E-6f;
|
|
this._m22((zZeroToOne ? 0.0f : 1.0f) - e)
|
|
._m32(((zZeroToOne ? 1.0f : 2.0f) - e) * zFar);
|
|
} else {
|
|
this._m22((zZeroToOne ? zFar : zFar + zNear) / (zNear - zFar))
|
|
._m32((zZeroToOne ? zFar : zFar + zFar) * zNear / (zNear - zFar));
|
|
}
|
|
this._m23(-1.0f)
|
|
._properties(PROPERTY_PERSPECTIVE);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a symmetric perspective projection frustum transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code>.
|
|
* <p>
|
|
* In order to apply the perspective projection transformation to an existing transformation,
|
|
* use {@link #perspectiveRect(float, float, float, float) perspectiveRect()}.
|
|
*
|
|
* @see #perspectiveRect(float, float, float, float)
|
|
*
|
|
* @param width
|
|
* the width of the near frustum plane
|
|
* @param height
|
|
* the height of the near frustum plane
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @return this
|
|
*/
|
|
public Matrix4f setPerspectiveRect(float width, float height, float zNear, float zFar) {
|
|
return setPerspectiveRect(width, height, zNear, zFar, false);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be an asymmetric off-center perspective projection frustum transformation for a right-handed
|
|
* coordinate system using OpenGL's NDC z range of <code>[-1..+1]</code>.
|
|
* <p>
|
|
* The given angles <code>offAngleX</code> and <code>offAngleY</code> are the horizontal and vertical angles between
|
|
* the line of sight and the line given by the center of the near and far frustum planes. So, when <code>offAngleY</code>
|
|
* is just <code>fovy/2</code> then the projection frustum is rotated towards +Y and the bottom frustum plane
|
|
* is parallel to the XZ-plane.
|
|
* <p>
|
|
* In order to apply the perspective projection transformation to an existing transformation,
|
|
* use {@link #perspectiveOffCenter(float, float, float, float, float, float) perspectiveOffCenter()}.
|
|
*
|
|
* @see #perspectiveOffCenter(float, float, float, float, float, float)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param offAngleX
|
|
* the horizontal angle between the line of sight and the line crossing the center of the near and far frustum planes
|
|
* @param offAngleY
|
|
* the vertical angle between the line of sight and the line crossing the center of the near and far frustum planes
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @return this
|
|
*/
|
|
public Matrix4f setPerspectiveOffCenter(float fovy, float offAngleX, float offAngleY,
|
|
float aspect, float zNear, float zFar) {
|
|
return setPerspectiveOffCenter(fovy, offAngleX, offAngleY, aspect, zNear, zFar, false);
|
|
}
|
|
/**
|
|
* Set this matrix to be an asymmetric off-center perspective projection frustum transformation for a right-handed
|
|
* coordinate system using the given NDC z range.
|
|
* <p>
|
|
* The given angles <code>offAngleX</code> and <code>offAngleY</code> are the horizontal and vertical angles between
|
|
* the line of sight and the line given by the center of the near and far frustum planes. So, when <code>offAngleY</code>
|
|
* is just <code>fovy/2</code> then the projection frustum is rotated towards +Y and the bottom frustum plane
|
|
* is parallel to the XZ-plane.
|
|
* <p>
|
|
* In order to apply the perspective projection transformation to an existing transformation,
|
|
* use {@link #perspectiveOffCenter(float, float, float, float, float, float) perspectiveOffCenter()}.
|
|
*
|
|
* @see #perspectiveOffCenter(float, float, float, float, float, float)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param offAngleX
|
|
* the horizontal angle between the line of sight and the line crossing the center of the near and far frustum planes
|
|
* @param offAngleY
|
|
* the vertical angle between the line of sight and the line crossing the center of the near and far frustum planes
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f setPerspectiveOffCenter(float fovy, float offAngleX, float offAngleY,
|
|
float aspect, float zNear, float zFar, boolean zZeroToOne) {
|
|
MemUtil.INSTANCE.zero(this);
|
|
float h = Math.tan(fovy * 0.5f);
|
|
float xScale = 1.0f / (h * aspect), yScale = 1.0f / h;
|
|
float offX = Math.tan(offAngleX), offY = Math.tan(offAngleY);
|
|
this._m00(xScale)
|
|
._m11(yScale)
|
|
._m20(offX * xScale)
|
|
._m21(offY * yScale);
|
|
boolean farInf = zFar > 0 && Float.isInfinite(zFar);
|
|
boolean nearInf = zNear > 0 && Float.isInfinite(zNear);
|
|
if (farInf) {
|
|
// See: "Infinite Projection Matrix" (http://www.terathon.com/gdc07_lengyel.pdf)
|
|
float e = 1E-6f;
|
|
this._m22(e - 1.0f)
|
|
._m32((e - (zZeroToOne ? 1.0f : 2.0f)) * zNear);
|
|
} else if (nearInf) {
|
|
float e = 1E-6f;
|
|
this._m22((zZeroToOne ? 0.0f : 1.0f) - e)
|
|
._m32(((zZeroToOne ? 1.0f : 2.0f) - e) * zFar);
|
|
} else {
|
|
this._m22((zZeroToOne ? zFar : zFar + zNear) / (zNear - zFar))
|
|
._m32((zZeroToOne ? zFar : zFar + zFar) * zNear / (zNear - zFar));
|
|
}
|
|
this._m23(-1.0f)
|
|
._properties(offAngleX == 0.0f && offAngleY == 0.0f ? PROPERTY_PERSPECTIVE : 0);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric perspective projection frustum transformation for a left-handed coordinate system
|
|
* using the given NDC z range to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspectiveLH(float, float, float, float, boolean) setPerspectiveLH}.
|
|
*
|
|
* @see #setPerspectiveLH(float, float, float, float, boolean)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f perspectiveLH(float fovy, float aspect, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setPerspectiveLH(fovy, aspect, zNear, zFar, zZeroToOne);
|
|
return perspectiveLHGeneric(fovy, aspect, zNear, zFar, zZeroToOne, dest);
|
|
}
|
|
private Matrix4f perspectiveLHGeneric(float fovy, float aspect, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
float h = Math.tan(fovy * 0.5f);
|
|
// calculate right matrix elements
|
|
float rm00 = 1.0f / (h * aspect);
|
|
float rm11 = 1.0f / h;
|
|
float rm22;
|
|
float rm32;
|
|
boolean farInf = zFar > 0 && Float.isInfinite(zFar);
|
|
boolean nearInf = zNear > 0 && Float.isInfinite(zNear);
|
|
if (farInf) {
|
|
// See: "Infinite Projection Matrix" (http://www.terathon.com/gdc07_lengyel.pdf)
|
|
float e = 1E-6f;
|
|
rm22 = 1.0f - e;
|
|
rm32 = (e - (zZeroToOne ? 1.0f : 2.0f)) * zNear;
|
|
} else if (nearInf) {
|
|
float e = 1E-6f;
|
|
rm22 = (zZeroToOne ? 0.0f : 1.0f) - e;
|
|
rm32 = ((zZeroToOne ? 1.0f : 2.0f) - e) * zFar;
|
|
} else {
|
|
rm22 = (zZeroToOne ? zFar : zFar + zNear) / (zFar - zNear);
|
|
rm32 = (zZeroToOne ? zFar : zFar + zFar) * zNear / (zNear - zFar);
|
|
}
|
|
// perform optimized matrix multiplication
|
|
float nm20 = m20 * rm22 + m30;
|
|
float nm21 = m21 * rm22 + m31;
|
|
float nm22 = m22 * rm22 + m32;
|
|
float nm23 = m23 * rm22 + m33;
|
|
dest._m00(m00 * rm00)
|
|
._m01(m01 * rm00)
|
|
._m02(m02 * rm00)
|
|
._m03(m03 * rm00)
|
|
._m10(m10 * rm11)
|
|
._m11(m11 * rm11)
|
|
._m12(m12 * rm11)
|
|
._m13(m13 * rm11)
|
|
._m30(m20 * rm32)
|
|
._m31(m21 * rm32)
|
|
._m32(m22 * rm32)
|
|
._m33(m23 * rm32)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._properties(properties & ~(PROPERTY_AFFINE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric perspective projection frustum transformation for a left-handed coordinate system
|
|
* using the given NDC z range to this matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspectiveLH(float, float, float, float, boolean) setPerspectiveLH}.
|
|
*
|
|
* @see #setPerspectiveLH(float, float, float, float, boolean)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f perspectiveLH(float fovy, float aspect, float zNear, float zFar, boolean zZeroToOne) {
|
|
return perspectiveLH(fovy, aspect, zNear, zFar, zZeroToOne, this);
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric perspective projection frustum transformation for a left-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspectiveLH(float, float, float, float) setPerspectiveLH}.
|
|
*
|
|
* @see #setPerspectiveLH(float, float, float, float)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f perspectiveLH(float fovy, float aspect, float zNear, float zFar, Matrix4f dest) {
|
|
return perspectiveLH(fovy, aspect, zNear, zFar, false, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a symmetric perspective projection frustum transformation for a left-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>P</code> the perspective projection matrix,
|
|
* then the new matrix will be <code>M * P</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * P * v</code>,
|
|
* the perspective projection will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setPerspectiveLH(float, float, float, float) setPerspectiveLH}.
|
|
*
|
|
* @see #setPerspectiveLH(float, float, float, float)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @return this
|
|
*/
|
|
public Matrix4f perspectiveLH(float fovy, float aspect, float zNear, float zFar) {
|
|
return perspectiveLH(fovy, aspect, zNear, zFar, this);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a symmetric perspective projection frustum transformation for a left-handed coordinate system
|
|
* using the given NDC z range of <code>[-1..+1]</code>.
|
|
* <p>
|
|
* In order to apply the perspective projection transformation to an existing transformation,
|
|
* use {@link #perspectiveLH(float, float, float, float, boolean) perspectiveLH()}.
|
|
*
|
|
* @see #perspectiveLH(float, float, float, float, boolean)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f setPerspectiveLH(float fovy, float aspect, float zNear, float zFar, boolean zZeroToOne) {
|
|
MemUtil.INSTANCE.zero(this);
|
|
float h = Math.tan(fovy * 0.5f);
|
|
this._m00(1.0f / (h * aspect))
|
|
._m11(1.0f / h);
|
|
boolean farInf = zFar > 0 && Float.isInfinite(zFar);
|
|
boolean nearInf = zNear > 0 && Float.isInfinite(zNear);
|
|
if (farInf) {
|
|
// See: "Infinite Projection Matrix" (http://www.terathon.com/gdc07_lengyel.pdf)
|
|
float e = 1E-6f;
|
|
this._m22(1.0f - e)
|
|
._m32((e - (zZeroToOne ? 1.0f : 2.0f)) * zNear);
|
|
} else if (nearInf) {
|
|
float e = 1E-6f;
|
|
this._m22((zZeroToOne ? 0.0f : 1.0f) - e)
|
|
._m32(((zZeroToOne ? 1.0f : 2.0f) - e) * zFar);
|
|
} else {
|
|
this._m22((zZeroToOne ? zFar : zFar + zNear) / (zFar - zNear))
|
|
._m32((zZeroToOne ? zFar : zFar + zFar) * zNear / (zNear - zFar));
|
|
}
|
|
this._m23(1.0f)
|
|
._properties(PROPERTY_PERSPECTIVE);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be a symmetric perspective projection frustum transformation for a left-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code>.
|
|
* <p>
|
|
* In order to apply the perspective projection transformation to an existing transformation,
|
|
* use {@link #perspectiveLH(float, float, float, float) perspectiveLH()}.
|
|
*
|
|
* @see #perspectiveLH(float, float, float, float)
|
|
*
|
|
* @param fovy
|
|
* the vertical field of view in radians (must be greater than zero and less than {@link Math#PI PI})
|
|
* @param aspect
|
|
* the aspect ratio (i.e. width / height; must be greater than zero)
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @return this
|
|
*/
|
|
public Matrix4f setPerspectiveLH(float fovy, float aspect, float zNear, float zFar) {
|
|
return setPerspectiveLH(fovy, aspect, zNear, zFar, false);
|
|
}
|
|
|
|
/**
|
|
* Apply an arbitrary perspective projection frustum transformation for a right-handed coordinate system
|
|
* using the given NDC z range to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>F</code> the frustum matrix,
|
|
* then the new matrix will be <code>M * F</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * F * v</code>,
|
|
* the frustum transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setFrustum(float, float, float, float, float, float, boolean) setFrustum()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#perspective">http://www.songho.ca</a>
|
|
*
|
|
* @see #setFrustum(float, float, float, float, float, float, boolean)
|
|
*
|
|
* @param left
|
|
* the distance along the x-axis to the left frustum edge
|
|
* @param right
|
|
* the distance along the x-axis to the right frustum edge
|
|
* @param bottom
|
|
* the distance along the y-axis to the bottom frustum edge
|
|
* @param top
|
|
* the distance along the y-axis to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f frustum(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setFrustum(left, right, bottom, top, zNear, zFar, zZeroToOne);
|
|
return frustumGeneric(left, right, bottom, top, zNear, zFar, zZeroToOne, dest);
|
|
}
|
|
private Matrix4f frustumGeneric(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
// calculate right matrix elements
|
|
float rm00 = (zNear + zNear) / (right - left);
|
|
float rm11 = (zNear + zNear) / (top - bottom);
|
|
float rm20 = (right + left) / (right - left);
|
|
float rm21 = (top + bottom) / (top - bottom);
|
|
float rm22;
|
|
float rm32;
|
|
boolean farInf = zFar > 0 && Float.isInfinite(zFar);
|
|
boolean nearInf = zNear > 0 && Float.isInfinite(zNear);
|
|
if (farInf) {
|
|
// See: "Infinite Projection Matrix" (http://www.terathon.com/gdc07_lengyel.pdf)
|
|
float e = 1E-6f;
|
|
rm22 = e - 1.0f;
|
|
rm32 = (e - (zZeroToOne ? 1.0f : 2.0f)) * zNear;
|
|
} else if (nearInf) {
|
|
float e = 1E-6f;
|
|
rm22 = (zZeroToOne ? 0.0f : 1.0f) - e;
|
|
rm32 = ((zZeroToOne ? 1.0f : 2.0f) - e) * zFar;
|
|
} else {
|
|
rm22 = (zZeroToOne ? zFar : zFar + zNear) / (zNear - zFar);
|
|
rm32 = (zZeroToOne ? zFar : zFar + zFar) * zNear / (zNear - zFar);
|
|
}
|
|
// perform optimized matrix multiplication
|
|
float nm20 = m00 * rm20 + m10 * rm21 + m20 * rm22 - m30;
|
|
float nm21 = m01 * rm20 + m11 * rm21 + m21 * rm22 - m31;
|
|
float nm22 = m02 * rm20 + m12 * rm21 + m22 * rm22 - m32;
|
|
float nm23 = m03 * rm20 + m13 * rm21 + m23 * rm22 - m33;
|
|
dest._m00(m00 * rm00)
|
|
._m01(m01 * rm00)
|
|
._m02(m02 * rm00)
|
|
._m03(m03 * rm00)
|
|
._m10(m10 * rm11)
|
|
._m11(m11 * rm11)
|
|
._m12(m12 * rm11)
|
|
._m13(m13 * rm11)
|
|
._m30(m20 * rm32)
|
|
._m31(m21 * rm32)
|
|
._m32(m22 * rm32)
|
|
._m33(m23 * rm32)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._properties(0);
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply an arbitrary perspective projection frustum transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>F</code> the frustum matrix,
|
|
* then the new matrix will be <code>M * F</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * F * v</code>,
|
|
* the frustum transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setFrustum(float, float, float, float, float, float) setFrustum()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#perspective">http://www.songho.ca</a>
|
|
*
|
|
* @see #setFrustum(float, float, float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance along the x-axis to the left frustum edge
|
|
* @param right
|
|
* the distance along the x-axis to the right frustum edge
|
|
* @param bottom
|
|
* the distance along the y-axis to the bottom frustum edge
|
|
* @param top
|
|
* the distance along the y-axis to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f frustum(float left, float right, float bottom, float top, float zNear, float zFar, Matrix4f dest) {
|
|
return frustum(left, right, bottom, top, zNear, zFar, false, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply an arbitrary perspective projection frustum transformation for a right-handed coordinate system
|
|
* using the given NDC z range to this matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>F</code> the frustum matrix,
|
|
* then the new matrix will be <code>M * F</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * F * v</code>,
|
|
* the frustum transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setFrustum(float, float, float, float, float, float, boolean) setFrustum()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#perspective">http://www.songho.ca</a>
|
|
*
|
|
* @see #setFrustum(float, float, float, float, float, float, boolean)
|
|
*
|
|
* @param left
|
|
* the distance along the x-axis to the left frustum edge
|
|
* @param right
|
|
* the distance along the x-axis to the right frustum edge
|
|
* @param bottom
|
|
* the distance along the y-axis to the bottom frustum edge
|
|
* @param top
|
|
* the distance along the y-axis to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f frustum(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne) {
|
|
return frustum(left, right, bottom, top, zNear, zFar, zZeroToOne, this);
|
|
}
|
|
|
|
/**
|
|
* Apply an arbitrary perspective projection frustum transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>F</code> the frustum matrix,
|
|
* then the new matrix will be <code>M * F</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * F * v</code>,
|
|
* the frustum transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setFrustum(float, float, float, float, float, float) setFrustum()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#perspective">http://www.songho.ca</a>
|
|
*
|
|
* @see #setFrustum(float, float, float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance along the x-axis to the left frustum edge
|
|
* @param right
|
|
* the distance along the x-axis to the right frustum edge
|
|
* @param bottom
|
|
* the distance along the y-axis to the bottom frustum edge
|
|
* @param top
|
|
* the distance along the y-axis to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @return this
|
|
*/
|
|
public Matrix4f frustum(float left, float right, float bottom, float top, float zNear, float zFar) {
|
|
return frustum(left, right, bottom, top, zNear, zFar, this);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be an arbitrary perspective projection frustum transformation for a right-handed coordinate system
|
|
* using the given NDC z range.
|
|
* <p>
|
|
* In order to apply the perspective frustum transformation to an existing transformation,
|
|
* use {@link #frustum(float, float, float, float, float, float, boolean) frustum()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#perspective">http://www.songho.ca</a>
|
|
*
|
|
* @see #frustum(float, float, float, float, float, float, boolean)
|
|
*
|
|
* @param left
|
|
* the distance along the x-axis to the left frustum edge
|
|
* @param right
|
|
* the distance along the x-axis to the right frustum edge
|
|
* @param bottom
|
|
* the distance along the y-axis to the bottom frustum edge
|
|
* @param top
|
|
* the distance along the y-axis to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f setFrustum(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne) {
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
this._m00((zNear + zNear) / (right - left))
|
|
._m11((zNear + zNear) / (top - bottom))
|
|
._m20((right + left) / (right - left))
|
|
._m21((top + bottom) / (top - bottom));
|
|
boolean farInf = zFar > 0 && Float.isInfinite(zFar);
|
|
boolean nearInf = zNear > 0 && Float.isInfinite(zNear);
|
|
if (farInf) {
|
|
// See: "Infinite Projection Matrix" (http://www.terathon.com/gdc07_lengyel.pdf)
|
|
float e = 1E-6f;
|
|
this._m22(e - 1.0f)
|
|
._m32((e - (zZeroToOne ? 1.0f : 2.0f)) * zNear);
|
|
} else if (nearInf) {
|
|
float e = 1E-6f;
|
|
this._m22((zZeroToOne ? 0.0f : 1.0f) - e)
|
|
._m32(((zZeroToOne ? 1.0f : 2.0f) - e) * zFar);
|
|
} else {
|
|
this._m22((zZeroToOne ? zFar : zFar + zNear) / (zNear - zFar))
|
|
._m32((zZeroToOne ? zFar : zFar + zFar) * zNear / (zNear - zFar));
|
|
}
|
|
this._m23(-1.0f)
|
|
._m33(0.0f)
|
|
._properties(this.m20 == 0.0f && this.m21 == 0.0f ? PROPERTY_PERSPECTIVE : 0);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be an arbitrary perspective projection frustum transformation for a right-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code>.
|
|
* <p>
|
|
* In order to apply the perspective frustum transformation to an existing transformation,
|
|
* use {@link #frustum(float, float, float, float, float, float) frustum()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#perspective">http://www.songho.ca</a>
|
|
*
|
|
* @see #frustum(float, float, float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance along the x-axis to the left frustum edge
|
|
* @param right
|
|
* the distance along the x-axis to the right frustum edge
|
|
* @param bottom
|
|
* the distance along the y-axis to the bottom frustum edge
|
|
* @param top
|
|
* the distance along the y-axis to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @return this
|
|
*/
|
|
public Matrix4f setFrustum(float left, float right, float bottom, float top, float zNear, float zFar) {
|
|
return setFrustum(left, right, bottom, top, zNear, zFar, false);
|
|
}
|
|
|
|
/**
|
|
* Apply an arbitrary perspective projection frustum transformation for a left-handed coordinate system
|
|
* using the given NDC z range to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>F</code> the frustum matrix,
|
|
* then the new matrix will be <code>M * F</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * F * v</code>,
|
|
* the frustum transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setFrustumLH(float, float, float, float, float, float, boolean) setFrustumLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#perspective">http://www.songho.ca</a>
|
|
*
|
|
* @see #setFrustumLH(float, float, float, float, float, float, boolean)
|
|
*
|
|
* @param left
|
|
* the distance along the x-axis to the left frustum edge
|
|
* @param right
|
|
* the distance along the x-axis to the right frustum edge
|
|
* @param bottom
|
|
* the distance along the y-axis to the bottom frustum edge
|
|
* @param top
|
|
* the distance along the y-axis to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f frustumLH(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.setFrustumLH(left, right, bottom, top, zNear, zFar, zZeroToOne);
|
|
return frustumLHGeneric(left, right, bottom, top, zNear, zFar, zZeroToOne, dest);
|
|
}
|
|
private Matrix4f frustumLHGeneric(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne, Matrix4f dest) {
|
|
// calculate right matrix elements
|
|
float rm00 = (zNear + zNear) / (right - left);
|
|
float rm11 = (zNear + zNear) / (top - bottom);
|
|
float rm20 = (right + left) / (right - left);
|
|
float rm21 = (top + bottom) / (top - bottom);
|
|
float rm22;
|
|
float rm32;
|
|
boolean farInf = zFar > 0 && Float.isInfinite(zFar);
|
|
boolean nearInf = zNear > 0 && Float.isInfinite(zNear);
|
|
if (farInf) {
|
|
// See: "Infinite Projection Matrix" (http://www.terathon.com/gdc07_lengyel.pdf)
|
|
float e = 1E-6f;
|
|
rm22 = 1.0f - e;
|
|
rm32 = (e - (zZeroToOne ? 1.0f : 2.0f)) * zNear;
|
|
} else if (nearInf) {
|
|
float e = 1E-6f;
|
|
rm22 = (zZeroToOne ? 0.0f : 1.0f) - e;
|
|
rm32 = ((zZeroToOne ? 1.0f : 2.0f) - e) * zFar;
|
|
} else {
|
|
rm22 = (zZeroToOne ? zFar : zFar + zNear) / (zFar - zNear);
|
|
rm32 = (zZeroToOne ? zFar : zFar + zFar) * zNear / (zNear - zFar);
|
|
}
|
|
// perform optimized matrix multiplication
|
|
float nm20 = m00 * rm20 + m10 * rm21 + m20 * rm22 + m30;
|
|
float nm21 = m01 * rm20 + m11 * rm21 + m21 * rm22 + m31;
|
|
float nm22 = m02 * rm20 + m12 * rm21 + m22 * rm22 + m32;
|
|
float nm23 = m03 * rm20 + m13 * rm21 + m23 * rm22 + m33;
|
|
dest._m00(m00 * rm00)
|
|
._m01(m01 * rm00)
|
|
._m02(m02 * rm00)
|
|
._m03(m03 * rm00)
|
|
._m10(m10 * rm11)
|
|
._m11(m11 * rm11)
|
|
._m12(m12 * rm11)
|
|
._m13(m13 * rm11)
|
|
._m30(m20 * rm32)
|
|
._m31(m21 * rm32)
|
|
._m32(m22 * rm32)
|
|
._m33(m23 * rm32)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._properties(0);
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply an arbitrary perspective projection frustum transformation for a left-handed coordinate system
|
|
* using the given NDC z range to this matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>F</code> the frustum matrix,
|
|
* then the new matrix will be <code>M * F</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * F * v</code>,
|
|
* the frustum transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setFrustumLH(float, float, float, float, float, float, boolean) setFrustumLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#perspective">http://www.songho.ca</a>
|
|
*
|
|
* @see #setFrustumLH(float, float, float, float, float, float, boolean)
|
|
*
|
|
* @param left
|
|
* the distance along the x-axis to the left frustum edge
|
|
* @param right
|
|
* the distance along the x-axis to the right frustum edge
|
|
* @param bottom
|
|
* the distance along the y-axis to the bottom frustum edge
|
|
* @param top
|
|
* the distance along the y-axis to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f frustumLH(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne) {
|
|
return frustumLH(left, right, bottom, top, zNear, zFar, zZeroToOne, this);
|
|
}
|
|
|
|
/**
|
|
* Apply an arbitrary perspective projection frustum transformation for a left-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code> to this matrix and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>F</code> the frustum matrix,
|
|
* then the new matrix will be <code>M * F</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * F * v</code>,
|
|
* the frustum transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setFrustumLH(float, float, float, float, float, float) setFrustumLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#perspective">http://www.songho.ca</a>
|
|
*
|
|
* @see #setFrustumLH(float, float, float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance along the x-axis to the left frustum edge
|
|
* @param right
|
|
* the distance along the x-axis to the right frustum edge
|
|
* @param bottom
|
|
* the distance along the y-axis to the bottom frustum edge
|
|
* @param top
|
|
* the distance along the y-axis to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f frustumLH(float left, float right, float bottom, float top, float zNear, float zFar, Matrix4f dest) {
|
|
return frustumLH(left, right, bottom, top, zNear, zFar, false, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply an arbitrary perspective projection frustum transformation for a left-handed coordinate system
|
|
* using the given NDC z range to this matrix.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>F</code> the frustum matrix,
|
|
* then the new matrix will be <code>M * F</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * F * v</code>,
|
|
* the frustum transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a perspective frustum transformation without post-multiplying,
|
|
* use {@link #setFrustumLH(float, float, float, float, float, float) setFrustumLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#perspective">http://www.songho.ca</a>
|
|
*
|
|
* @see #setFrustumLH(float, float, float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance along the x-axis to the left frustum edge
|
|
* @param right
|
|
* the distance along the x-axis to the right frustum edge
|
|
* @param bottom
|
|
* the distance along the y-axis to the bottom frustum edge
|
|
* @param top
|
|
* the distance along the y-axis to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @return this
|
|
*/
|
|
public Matrix4f frustumLH(float left, float right, float bottom, float top, float zNear, float zFar) {
|
|
return frustumLH(left, right, bottom, top, zNear, zFar, this);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be an arbitrary perspective projection frustum transformation for a left-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code>.
|
|
* <p>
|
|
* In order to apply the perspective frustum transformation to an existing transformation,
|
|
* use {@link #frustumLH(float, float, float, float, float, float, boolean) frustumLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#perspective">http://www.songho.ca</a>
|
|
*
|
|
* @see #frustumLH(float, float, float, float, float, float, boolean)
|
|
*
|
|
* @param left
|
|
* the distance along the x-axis to the left frustum edge
|
|
* @param right
|
|
* the distance along the x-axis to the right frustum edge
|
|
* @param bottom
|
|
* the distance along the y-axis to the bottom frustum edge
|
|
* @param top
|
|
* the distance along the y-axis to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zZeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f setFrustumLH(float left, float right, float bottom, float top, float zNear, float zFar, boolean zZeroToOne) {
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
this._m00((zNear + zNear) / (right - left))
|
|
._m11((zNear + zNear) / (top - bottom))
|
|
._m20((right + left) / (right - left))
|
|
._m21((top + bottom) / (top - bottom));
|
|
boolean farInf = zFar > 0 && Float.isInfinite(zFar);
|
|
boolean nearInf = zNear > 0 && Float.isInfinite(zNear);
|
|
if (farInf) {
|
|
// See: "Infinite Projection Matrix" (http://www.terathon.com/gdc07_lengyel.pdf)
|
|
float e = 1E-6f;
|
|
this._m22(1.0f - e)
|
|
._m32((e - (zZeroToOne ? 1.0f : 2.0f)) * zNear);
|
|
} else if (nearInf) {
|
|
float e = 1E-6f;
|
|
this._m22((zZeroToOne ? 0.0f : 1.0f) - e)
|
|
._m32(((zZeroToOne ? 1.0f : 2.0f) - e) * zFar);
|
|
} else {
|
|
this._m22((zZeroToOne ? zFar : zFar + zNear) / (zFar - zNear))
|
|
._m32((zZeroToOne ? zFar : zFar + zFar) * zNear / (zNear - zFar));
|
|
}
|
|
return this
|
|
._m23(1.0f)
|
|
._m33(0.0f)
|
|
._properties(0);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to be an arbitrary perspective projection frustum transformation for a left-handed coordinate system
|
|
* using OpenGL's NDC z range of <code>[-1..+1]</code>.
|
|
* <p>
|
|
* In order to apply the perspective frustum transformation to an existing transformation,
|
|
* use {@link #frustumLH(float, float, float, float, float, float) frustumLH()}.
|
|
* <p>
|
|
* Reference: <a href="http://www.songho.ca/opengl/gl_projectionmatrix.html#perspective">http://www.songho.ca</a>
|
|
*
|
|
* @see #frustumLH(float, float, float, float, float, float)
|
|
*
|
|
* @param left
|
|
* the distance along the x-axis to the left frustum edge
|
|
* @param right
|
|
* the distance along the x-axis to the right frustum edge
|
|
* @param bottom
|
|
* the distance along the y-axis to the bottom frustum edge
|
|
* @param top
|
|
* the distance along the y-axis to the top frustum edge
|
|
* @param zNear
|
|
* near clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the near clipping plane will be at positive infinity.
|
|
* In that case, <code>zFar</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @param zFar
|
|
* far clipping plane distance. This value must be greater than zero.
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* In that case, <code>zNear</code> may not also be {@link Float#POSITIVE_INFINITY}.
|
|
* @return this
|
|
*/
|
|
public Matrix4f setFrustumLH(float left, float right, float bottom, float top, float zNear, float zFar) {
|
|
return setFrustumLH(left, right, bottom, top, zNear, zFar, false);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to represent a perspective projection equivalent to the given intrinsic camera calibration parameters.
|
|
* The resulting matrix will be suited for a right-handed coordinate system using OpenGL's NDC z range of <code>[-1..+1]</code>.
|
|
* <p>
|
|
* See: <a href="https://en.wikipedia.org/wiki/Camera_resectioning#Intrinsic_parameters">https://en.wikipedia.org/</a>
|
|
* <p>
|
|
* Reference: <a href="http://ksimek.github.io/2013/06/03/calibrated_cameras_in_opengl/">http://ksimek.github.io/</a>
|
|
*
|
|
* @param alphaX
|
|
* specifies the focal length and scale along the X axis
|
|
* @param alphaY
|
|
* specifies the focal length and scale along the Y axis
|
|
* @param gamma
|
|
* the skew coefficient between the X and Y axis (may be <code>0</code>)
|
|
* @param u0
|
|
* the X coordinate of the principal point in image/sensor units
|
|
* @param v0
|
|
* the Y coordinate of the principal point in image/sensor units
|
|
* @param imgWidth
|
|
* the width of the sensor/image image/sensor units
|
|
* @param imgHeight
|
|
* the height of the sensor/image image/sensor units
|
|
* @param near
|
|
* the distance to the near plane
|
|
* @param far
|
|
* the distance to the far plane
|
|
* @return this
|
|
*/
|
|
public Matrix4f setFromIntrinsic(float alphaX, float alphaY, float gamma, float u0, float v0, int imgWidth, int imgHeight, float near, float far) {
|
|
float l00 = 2.0f / imgWidth;
|
|
float l11 = 2.0f / imgHeight;
|
|
float l22 = 2.0f / (near - far);
|
|
this.m00 = l00 * alphaX;
|
|
this.m01 = 0.0f;
|
|
this.m02 = 0.0f;
|
|
this.m03 = 0.0f;
|
|
this.m10 = l00 * gamma;
|
|
this.m11 = l11 * alphaY;
|
|
this.m12 = 0.0f;
|
|
this.m13 = 0.0f;
|
|
this.m20 = l00 * u0 - 1.0f;
|
|
this.m21 = l11 * v0 - 1.0f;
|
|
this.m22 = l22 * -(near + far) + (far + near) / (near - far);
|
|
this.m23 = -1.0f;
|
|
this.m30 = 0.0f;
|
|
this.m31 = 0.0f;
|
|
this.m32 = l22 * -near * far;
|
|
this.m33 = 0.0f;
|
|
this.properties = PROPERTY_PERSPECTIVE;
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Apply the rotation transformation of the given {@link Quaternionfc} to this matrix and store
|
|
* the result in <code>dest</code>.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>Q</code> the rotation matrix obtained from the given quaternion,
|
|
* then the new matrix will be <code>M * Q</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * Q * v</code>,
|
|
* the quaternion rotation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without post-multiplying,
|
|
* use {@link #rotation(Quaternionfc)}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Quaternion">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotation(Quaternionfc)
|
|
*
|
|
* @param quat
|
|
* the {@link Quaternionfc}
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotate(Quaternionfc quat, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.rotation(quat);
|
|
else if ((properties & PROPERTY_TRANSLATION) != 0)
|
|
return rotateTranslation(quat, dest);
|
|
else if ((properties & PROPERTY_AFFINE) != 0)
|
|
return rotateAffine(quat, dest);
|
|
return rotateGeneric(quat, dest);
|
|
}
|
|
private Matrix4f rotateGeneric(Quaternionfc quat, Matrix4f dest) {
|
|
float w2 = quat.w() * quat.w(), x2 = quat.x() * quat.x();
|
|
float y2 = quat.y() * quat.y(), z2 = quat.z() * quat.z();
|
|
float zw = quat.z() * quat.w(), dzw = zw + zw, xy = quat.x() * quat.y(), dxy = xy + xy;
|
|
float xz = quat.x() * quat.z(), dxz = xz + xz, yw = quat.y() * quat.w(), dyw = yw + yw;
|
|
float yz = quat.y() * quat.z(), dyz = yz + yz, xw = quat.x() * quat.w(), dxw = xw + xw;
|
|
float rm00 = w2 + x2 - z2 - y2;
|
|
float rm01 = dxy + dzw;
|
|
float rm02 = dxz - dyw;
|
|
float rm10 = -dzw + dxy;
|
|
float rm11 = y2 - z2 + w2 - x2;
|
|
float rm12 = dyz + dxw;
|
|
float rm20 = dyw + dxz;
|
|
float rm21 = dyz - dxw;
|
|
float rm22 = z2 - y2 - x2 + w2;
|
|
float nm00 = m00 * rm00 + m10 * rm01 + m20 * rm02;
|
|
float nm01 = m01 * rm00 + m11 * rm01 + m21 * rm02;
|
|
float nm02 = m02 * rm00 + m12 * rm01 + m22 * rm02;
|
|
float nm03 = m03 * rm00 + m13 * rm01 + m23 * rm02;
|
|
float nm10 = m00 * rm10 + m10 * rm11 + m20 * rm12;
|
|
float nm11 = m01 * rm10 + m11 * rm11 + m21 * rm12;
|
|
float nm12 = m02 * rm10 + m12 * rm11 + m22 * rm12;
|
|
float nm13 = m03 * rm10 + m13 * rm11 + m23 * rm12;
|
|
return dest
|
|
._m20(m00 * rm20 + m10 * rm21 + m20 * rm22)
|
|
._m21(m01 * rm20 + m11 * rm21 + m21 * rm22)
|
|
._m22(m02 * rm20 + m12 * rm21 + m22 * rm22)
|
|
._m23(m03 * rm20 + m13 * rm21 + m23 * rm22)
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply the rotation transformation of the given {@link Quaternionfc} to this matrix.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>Q</code> the rotation matrix obtained from the given quaternion,
|
|
* then the new matrix will be <code>M * Q</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * Q * v</code>,
|
|
* the quaternion rotation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without post-multiplying,
|
|
* use {@link #rotation(Quaternionfc)}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Quaternion">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotation(Quaternionfc)
|
|
*
|
|
* @param quat
|
|
* the {@link Quaternionfc}
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotate(Quaternionfc quat) {
|
|
return rotate(quat, this);
|
|
}
|
|
|
|
/**
|
|
* Apply the rotation transformation of the given {@link Quaternionfc} to this {@link #isAffine() affine} matrix and store
|
|
* the result in <code>dest</code>.
|
|
* <p>
|
|
* This method assumes <code>this</code> to be {@link #isAffine() affine}.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>Q</code> the rotation matrix obtained from the given quaternion,
|
|
* then the new matrix will be <code>M * Q</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * Q * v</code>,
|
|
* the quaternion rotation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without post-multiplying,
|
|
* use {@link #rotation(Quaternionfc)}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Quaternion">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotation(Quaternionfc)
|
|
*
|
|
* @param quat
|
|
* the {@link Quaternionfc}
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotateAffine(Quaternionfc quat, Matrix4f dest) {
|
|
float w2 = quat.w() * quat.w(), x2 = quat.x() * quat.x();
|
|
float y2 = quat.y() * quat.y(), z2 = quat.z() * quat.z();
|
|
float zw = quat.z() * quat.w(), dzw = zw + zw, xy = quat.x() * quat.y(), dxy = xy + xy;
|
|
float xz = quat.x() * quat.z(), dxz = xz + xz, yw = quat.y() * quat.w(), dyw = yw + yw;
|
|
float yz = quat.y() * quat.z(), dyz = yz + yz, xw = quat.x() * quat.w(), dxw = xw + xw;
|
|
float rm00 = w2 + x2 - z2 - y2;
|
|
float rm01 = dxy + dzw;
|
|
float rm02 = dxz - dyw;
|
|
float rm10 = -dzw + dxy;
|
|
float rm11 = y2 - z2 + w2 - x2;
|
|
float rm12 = dyz + dxw;
|
|
float rm20 = dyw + dxz;
|
|
float rm21 = dyz - dxw;
|
|
float rm22 = z2 - y2 - x2 + w2;
|
|
float nm00 = m00 * rm00 + m10 * rm01 + m20 * rm02;
|
|
float nm01 = m01 * rm00 + m11 * rm01 + m21 * rm02;
|
|
float nm02 = m02 * rm00 + m12 * rm01 + m22 * rm02;
|
|
float nm10 = m00 * rm10 + m10 * rm11 + m20 * rm12;
|
|
float nm11 = m01 * rm10 + m11 * rm11 + m21 * rm12;
|
|
float nm12 = m02 * rm10 + m12 * rm11 + m22 * rm12;
|
|
return dest
|
|
._m20(m00 * rm20 + m10 * rm21 + m20 * rm22)
|
|
._m21(m01 * rm20 + m11 * rm21 + m21 * rm22)
|
|
._m22(m02 * rm20 + m12 * rm21 + m22 * rm22)
|
|
._m23(0.0f)
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(0.0f)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(0.0f)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply the rotation transformation of the given {@link Quaternionfc} to this matrix.
|
|
* <p>
|
|
* This method assumes <code>this</code> to be {@link #isAffine() affine}.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>Q</code> the rotation matrix obtained from the given quaternion,
|
|
* then the new matrix will be <code>M * Q</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * Q * v</code>,
|
|
* the quaternion rotation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without post-multiplying,
|
|
* use {@link #rotation(Quaternionfc)}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Quaternion">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotation(Quaternionfc)
|
|
*
|
|
* @param quat
|
|
* the {@link Quaternionfc}
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateAffine(Quaternionfc quat) {
|
|
return rotateAffine(quat, this);
|
|
}
|
|
|
|
/**
|
|
* Apply the rotation transformation of the given {@link Quaternionfc} to this matrix, which is assumed to only contain a translation, and store
|
|
* the result in <code>dest</code>.
|
|
* <p>
|
|
* This method assumes <code>this</code> to only contain a translation.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>Q</code> the rotation matrix obtained from the given quaternion,
|
|
* then the new matrix will be <code>M * Q</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * Q * v</code>,
|
|
* the quaternion rotation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without post-multiplying,
|
|
* use {@link #rotation(Quaternionfc)}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Quaternion">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotation(Quaternionfc)
|
|
*
|
|
* @param quat
|
|
* the {@link Quaternionfc}
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotateTranslation(Quaternionfc quat, Matrix4f dest) {
|
|
float w2 = quat.w() * quat.w(), x2 = quat.x() * quat.x();
|
|
float y2 = quat.y() * quat.y(), z2 = quat.z() * quat.z();
|
|
float zw = quat.z() * quat.w(), dzw = zw + zw, xy = quat.x() * quat.y(), dxy = xy + xy;
|
|
float xz = quat.x() * quat.z(), dxz = xz + xz, yw = quat.y() * quat.w(), dyw = yw + yw;
|
|
float yz = quat.y() * quat.z(), dyz = yz + yz, xw = quat.x() * quat.w(), dxw = xw + xw;
|
|
float rm00 = w2 + x2 - z2 - y2;
|
|
float rm01 = dxy + dzw;
|
|
float rm02 = dxz - dyw;
|
|
float rm10 = -dzw + dxy;
|
|
float rm11 = y2 - z2 + w2 - x2;
|
|
float rm12 = dyz + dxw;
|
|
float rm20 = dyw + dxz;
|
|
float rm21 = dyz - dxw;
|
|
float rm22 = z2 - y2 - x2 + w2;
|
|
return dest
|
|
._m20(rm20)
|
|
._m21(rm21)
|
|
._m22(rm22)
|
|
._m23(0.0f)
|
|
._m00(rm00)
|
|
._m01(rm01)
|
|
._m02(rm02)
|
|
._m03(0.0f)
|
|
._m10(rm10)
|
|
._m11(rm11)
|
|
._m12(rm12)
|
|
._m13(0.0f)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Apply the rotation transformation of the given {@link Quaternionfc} to this matrix while using <code>(ox, oy, oz)</code> as the rotation origin.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>Q</code> the rotation matrix obtained from the given quaternion,
|
|
* then the new matrix will be <code>M * Q</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * Q * v</code>,
|
|
* the quaternion rotation will be applied first!
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translate(ox, oy, oz).rotate(quat).translate(-ox, -oy, -oz)</code>
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Quaternion">http://en.wikipedia.org</a>
|
|
*
|
|
* @param quat
|
|
* the {@link Quaternionfc}
|
|
* @param ox
|
|
* the x coordinate of the rotation origin
|
|
* @param oy
|
|
* the y coordinate of the rotation origin
|
|
* @param oz
|
|
* the z coordinate of the rotation origin
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateAround(Quaternionfc quat, float ox, float oy, float oz) {
|
|
return rotateAround(quat, ox, oy, oz, this);
|
|
}
|
|
|
|
public Matrix4f rotateAroundAffine(Quaternionfc quat, float ox, float oy, float oz, Matrix4f dest) {
|
|
float w2 = quat.w() * quat.w(), x2 = quat.x() * quat.x();
|
|
float y2 = quat.y() * quat.y(), z2 = quat.z() * quat.z();
|
|
float zw = quat.z() * quat.w(), dzw = zw + zw, xy = quat.x() * quat.y(), dxy = xy + xy;
|
|
float xz = quat.x() * quat.z(), dxz = xz + xz, yw = quat.y() * quat.w(), dyw = yw + yw;
|
|
float yz = quat.y() * quat.z(), dyz = yz + yz, xw = quat.x() * quat.w(), dxw = xw + xw;
|
|
float rm00 = w2 + x2 - z2 - y2;
|
|
float rm01 = dxy + dzw;
|
|
float rm02 = dxz - dyw;
|
|
float rm10 = -dzw + dxy;
|
|
float rm11 = y2 - z2 + w2 - x2;
|
|
float rm12 = dyz + dxw;
|
|
float rm20 = dyw + dxz;
|
|
float rm21 = dyz - dxw;
|
|
float rm22 = z2 - y2 - x2 + w2;
|
|
float tm30 = m00 * ox + m10 * oy + m20 * oz + m30;
|
|
float tm31 = m01 * ox + m11 * oy + m21 * oz + m31;
|
|
float tm32 = m02 * ox + m12 * oy + m22 * oz + m32;
|
|
float nm00 = m00 * rm00 + m10 * rm01 + m20 * rm02;
|
|
float nm01 = m01 * rm00 + m11 * rm01 + m21 * rm02;
|
|
float nm02 = m02 * rm00 + m12 * rm01 + m22 * rm02;
|
|
float nm10 = m00 * rm10 + m10 * rm11 + m20 * rm12;
|
|
float nm11 = m01 * rm10 + m11 * rm11 + m21 * rm12;
|
|
float nm12 = m02 * rm10 + m12 * rm11 + m22 * rm12;
|
|
dest._m20(m00 * rm20 + m10 * rm21 + m20 * rm22)
|
|
._m21(m01 * rm20 + m11 * rm21 + m21 * rm22)
|
|
._m22(m02 * rm20 + m12 * rm21 + m22 * rm22)
|
|
._m23(0.0f)
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(0.0f)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(0.0f)
|
|
._m30(-nm00 * ox - nm10 * oy - m20 * oz + tm30)
|
|
._m31(-nm01 * ox - nm11 * oy - m21 * oz + tm31)
|
|
._m32(-nm02 * ox - nm12 * oy - m22 * oz + tm32)
|
|
._m33(1.0f)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
return dest;
|
|
}
|
|
|
|
public Matrix4f rotateAround(Quaternionfc quat, float ox, float oy, float oz, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return rotationAround(quat, ox, oy, oz);
|
|
else if ((properties & PROPERTY_AFFINE) != 0)
|
|
return rotateAroundAffine(quat, ox, oy, oz, dest);
|
|
return rotateAroundGeneric(quat, ox, oy, oz, dest);
|
|
}
|
|
private Matrix4f rotateAroundGeneric(Quaternionfc quat, float ox, float oy, float oz, Matrix4f dest) {
|
|
float w2 = quat.w() * quat.w(), x2 = quat.x() * quat.x();
|
|
float y2 = quat.y() * quat.y(), z2 = quat.z() * quat.z();
|
|
float zw = quat.z() * quat.w(), dzw = zw + zw, xy = quat.x() * quat.y(), dxy = xy + xy;
|
|
float xz = quat.x() * quat.z(), dxz = xz + xz, yw = quat.y() * quat.w(), dyw = yw + yw;
|
|
float yz = quat.y() * quat.z(), dyz = yz + yz, xw = quat.x() * quat.w(), dxw = xw + xw;
|
|
float rm00 = w2 + x2 - z2 - y2;
|
|
float rm01 = dxy + dzw;
|
|
float rm02 = dxz - dyw;
|
|
float rm10 = -dzw + dxy;
|
|
float rm11 = y2 - z2 + w2 - x2;
|
|
float rm12 = dyz + dxw;
|
|
float rm20 = dyw + dxz;
|
|
float rm21 = dyz - dxw;
|
|
float rm22 = z2 - y2 - x2 + w2;
|
|
float tm30 = m00 * ox + m10 * oy + m20 * oz + m30;
|
|
float tm31 = m01 * ox + m11 * oy + m21 * oz + m31;
|
|
float tm32 = m02 * ox + m12 * oy + m22 * oz + m32;
|
|
float nm00 = m00 * rm00 + m10 * rm01 + m20 * rm02;
|
|
float nm01 = m01 * rm00 + m11 * rm01 + m21 * rm02;
|
|
float nm02 = m02 * rm00 + m12 * rm01 + m22 * rm02;
|
|
float nm03 = m03 * rm00 + m13 * rm01 + m23 * rm02;
|
|
float nm10 = m00 * rm10 + m10 * rm11 + m20 * rm12;
|
|
float nm11 = m01 * rm10 + m11 * rm11 + m21 * rm12;
|
|
float nm12 = m02 * rm10 + m12 * rm11 + m22 * rm12;
|
|
float nm13 = m03 * rm10 + m13 * rm11 + m23 * rm12;
|
|
dest._m20(m00 * rm20 + m10 * rm21 + m20 * rm22)
|
|
._m21(m01 * rm20 + m11 * rm21 + m21 * rm22)
|
|
._m22(m02 * rm20 + m12 * rm21 + m22 * rm22)
|
|
._m23(m03 * rm20 + m13 * rm21 + m23 * rm22)
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m30(-nm00 * ox - nm10 * oy - m20 * oz + tm30)
|
|
._m31(-nm01 * ox - nm11 * oy - m21 * oz + tm31)
|
|
._m32(-nm02 * ox - nm12 * oy - m22 * oz + tm32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a transformation composed of a rotation of the specified {@link Quaternionfc} while using <code>(ox, oy, oz)</code> as the rotation origin.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translation(ox, oy, oz).rotate(quat).translate(-ox, -oy, -oz)</code>
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Quaternion">http://en.wikipedia.org</a>
|
|
*
|
|
* @param quat
|
|
* the {@link Quaternionfc}
|
|
* @param ox
|
|
* the x coordinate of the rotation origin
|
|
* @param oy
|
|
* the y coordinate of the rotation origin
|
|
* @param oz
|
|
* the z coordinate of the rotation origin
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotationAround(Quaternionfc quat, float ox, float oy, float oz) {
|
|
float w2 = quat.w() * quat.w(), x2 = quat.x() * quat.x();
|
|
float y2 = quat.y() * quat.y(), z2 = quat.z() * quat.z();
|
|
float zw = quat.z() * quat.w(), dzw = zw + zw, xy = quat.x() * quat.y(), dxy = xy + xy;
|
|
float xz = quat.x() * quat.z(), dxz = xz + xz, yw = quat.y() * quat.w(), dyw = yw + yw;
|
|
float yz = quat.y() * quat.z(), dyz = yz + yz, xw = quat.x() * quat.w(), dxw = xw + xw;
|
|
this._m20(dyw + dxz)
|
|
._m21(dyz - dxw)
|
|
._m22(z2 - y2 - x2 + w2)
|
|
._m23(0.0f)
|
|
._m00(w2 + x2 - z2 - y2)
|
|
._m01(dxy + dzw)
|
|
._m02(dxz - dyw)
|
|
._m03(0.0f)
|
|
._m10(-dzw + dxy)
|
|
._m11(y2 - z2 + w2 - x2)
|
|
._m12(dyz + dxw)
|
|
._m13(0.0f)
|
|
._m30(-m00 * ox - m10 * oy - m20 * oz + ox)
|
|
._m31(-m01 * ox - m11 * oy - m21 * oz + oy)
|
|
._m32(-m02 * ox - m12 * oy - m22 * oz + oz)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply the rotation transformation of the given {@link Quaternionfc} to this matrix and store
|
|
* the result in <code>dest</code>.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>Q</code> the rotation matrix obtained from the given quaternion,
|
|
* then the new matrix will be <code>Q * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>Q * M * v</code>,
|
|
* the quaternion rotation will be applied last!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without pre-multiplying,
|
|
* use {@link #rotation(Quaternionfc)}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Quaternion">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotation(Quaternionfc)
|
|
*
|
|
* @param quat
|
|
* the {@link Quaternionfc}
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotateLocal(Quaternionfc quat, Matrix4f dest) {
|
|
float w2 = quat.w() * quat.w(), x2 = quat.x() * quat.x();
|
|
float y2 = quat.y() * quat.y(), z2 = quat.z() * quat.z();
|
|
float zw = quat.z() * quat.w(), dzw = zw + zw, xy = quat.x() * quat.y(), dxy = xy + xy;
|
|
float xz = quat.x() * quat.z(), dxz = xz + xz, yw = quat.y() * quat.w(), dyw = yw + yw;
|
|
float yz = quat.y() * quat.z(), dyz = yz + yz, xw = quat.x() * quat.w(), dxw = xw + xw;
|
|
float lm00 = w2 + x2 - z2 - y2;
|
|
float lm01 = dxy + dzw;
|
|
float lm02 = dxz - dyw;
|
|
float lm10 = -dzw + dxy;
|
|
float lm11 = y2 - z2 + w2 - x2;
|
|
float lm12 = dyz + dxw;
|
|
float lm20 = dyw + dxz;
|
|
float lm21 = dyz - dxw;
|
|
float lm22 = z2 - y2 - x2 + w2;
|
|
float nm00 = lm00 * m00 + lm10 * m01 + lm20 * m02;
|
|
float nm01 = lm01 * m00 + lm11 * m01 + lm21 * m02;
|
|
float nm02 = lm02 * m00 + lm12 * m01 + lm22 * m02;
|
|
float nm10 = lm00 * m10 + lm10 * m11 + lm20 * m12;
|
|
float nm11 = lm01 * m10 + lm11 * m11 + lm21 * m12;
|
|
float nm12 = lm02 * m10 + lm12 * m11 + lm22 * m12;
|
|
float nm20 = lm00 * m20 + lm10 * m21 + lm20 * m22;
|
|
float nm21 = lm01 * m20 + lm11 * m21 + lm21 * m22;
|
|
float nm22 = lm02 * m20 + lm12 * m21 + lm22 * m22;
|
|
float nm30 = lm00 * m30 + lm10 * m31 + lm20 * m32;
|
|
float nm31 = lm01 * m30 + lm11 * m31 + lm21 * m32;
|
|
float nm32 = lm02 * m30 + lm12 * m31 + lm22 * m32;
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(m03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(m13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(m23)
|
|
._m30(nm30)
|
|
._m31(nm31)
|
|
._m32(nm32)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply the rotation transformation of the given {@link Quaternionfc} to this matrix.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>Q</code> the rotation matrix obtained from the given quaternion,
|
|
* then the new matrix will be <code>Q * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>Q * M * v</code>,
|
|
* the quaternion rotation will be applied last!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without pre-multiplying,
|
|
* use {@link #rotation(Quaternionfc)}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Quaternion">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotation(Quaternionfc)
|
|
*
|
|
* @param quat
|
|
* the {@link Quaternionfc}
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateLocal(Quaternionfc quat) {
|
|
return rotateLocal(quat, this);
|
|
}
|
|
|
|
public Matrix4f rotateAroundLocal(Quaternionfc quat, float ox, float oy, float oz, Matrix4f dest) {
|
|
float w2 = quat.w() * quat.w();
|
|
float x2 = quat.x() * quat.x();
|
|
float y2 = quat.y() * quat.y();
|
|
float z2 = quat.z() * quat.z();
|
|
float zw = quat.z() * quat.w();
|
|
float xy = quat.x() * quat.y();
|
|
float xz = quat.x() * quat.z();
|
|
float yw = quat.y() * quat.w();
|
|
float yz = quat.y() * quat.z();
|
|
float xw = quat.x() * quat.w();
|
|
float lm00 = w2 + x2 - z2 - y2;
|
|
float lm01 = xy + zw + zw + xy;
|
|
float lm02 = xz - yw + xz - yw;
|
|
float lm10 = -zw + xy - zw + xy;
|
|
float lm11 = y2 - z2 + w2 - x2;
|
|
float lm12 = yz + yz + xw + xw;
|
|
float lm20 = yw + xz + xz + yw;
|
|
float lm21 = yz + yz - xw - xw;
|
|
float lm22 = z2 - y2 - x2 + w2;
|
|
float tm00 = m00 - ox * m03;
|
|
float tm01 = m01 - oy * m03;
|
|
float tm02 = m02 - oz * m03;
|
|
float tm10 = m10 - ox * m13;
|
|
float tm11 = m11 - oy * m13;
|
|
float tm12 = m12 - oz * m13;
|
|
float tm20 = m20 - ox * m23;
|
|
float tm21 = m21 - oy * m23;
|
|
float tm22 = m22 - oz * m23;
|
|
float tm30 = m30 - ox * m33;
|
|
float tm31 = m31 - oy * m33;
|
|
float tm32 = m32 - oz * m33;
|
|
dest._m00(lm00 * tm00 + lm10 * tm01 + lm20 * tm02 + ox * m03)
|
|
._m01(lm01 * tm00 + lm11 * tm01 + lm21 * tm02 + oy * m03)
|
|
._m02(lm02 * tm00 + lm12 * tm01 + lm22 * tm02 + oz * m03)
|
|
._m03(m03)
|
|
._m10(lm00 * tm10 + lm10 * tm11 + lm20 * tm12 + ox * m13)
|
|
._m11(lm01 * tm10 + lm11 * tm11 + lm21 * tm12 + oy * m13)
|
|
._m12(lm02 * tm10 + lm12 * tm11 + lm22 * tm12 + oz * m13)
|
|
._m13(m13)
|
|
._m20(lm00 * tm20 + lm10 * tm21 + lm20 * tm22 + ox * m23)
|
|
._m21(lm01 * tm20 + lm11 * tm21 + lm21 * tm22 + oy * m23)
|
|
._m22(lm02 * tm20 + lm12 * tm21 + lm22 * tm22 + oz * m23)
|
|
._m23(m23)
|
|
._m30(lm00 * tm30 + lm10 * tm31 + lm20 * tm32 + ox * m33)
|
|
._m31(lm01 * tm30 + lm11 * tm31 + lm21 * tm32 + oy * m33)
|
|
._m32(lm02 * tm30 + lm12 * tm31 + lm22 * tm32 + oz * m33)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Pre-multiply the rotation transformation of the given {@link Quaternionfc} to this matrix while using <code>(ox, oy, oz)</code>
|
|
* as the rotation origin.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>Q</code> the rotation matrix obtained from the given quaternion,
|
|
* then the new matrix will be <code>Q * M</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>Q * M * v</code>,
|
|
* the quaternion rotation will be applied last!
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translateLocal(-ox, -oy, -oz).rotateLocal(quat).translateLocal(ox, oy, oz)</code>
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Quaternion">http://en.wikipedia.org</a>
|
|
*
|
|
* @param quat
|
|
* the {@link Quaternionfc}
|
|
* @param ox
|
|
* the x coordinate of the rotation origin
|
|
* @param oy
|
|
* the y coordinate of the rotation origin
|
|
* @param oz
|
|
* the z coordinate of the rotation origin
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateAroundLocal(Quaternionfc quat, float ox, float oy, float oz) {
|
|
return rotateAroundLocal(quat, ox, oy, oz, this);
|
|
}
|
|
|
|
/**
|
|
* Apply a rotation transformation, rotating about the given {@link AxisAngle4f}, to this matrix.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>A</code> the rotation matrix obtained from the given {@link AxisAngle4f},
|
|
* then the new matrix will be <code>M * A</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * A * v</code>,
|
|
* the {@link AxisAngle4f} rotation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without post-multiplying,
|
|
* use {@link #rotation(AxisAngle4f)}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotate(float, float, float, float)
|
|
* @see #rotation(AxisAngle4f)
|
|
*
|
|
* @param axisAngle
|
|
* the {@link AxisAngle4f} (needs to be {@link AxisAngle4f#normalize() normalized})
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotate(AxisAngle4f axisAngle) {
|
|
return rotate(axisAngle.angle, axisAngle.x, axisAngle.y, axisAngle.z);
|
|
}
|
|
|
|
/**
|
|
* Apply a rotation transformation, rotating about the given {@link AxisAngle4f} and store the result in <code>dest</code>.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>A</code> the rotation matrix obtained from the given {@link AxisAngle4f},
|
|
* then the new matrix will be <code>M * A</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * A * v</code>,
|
|
* the {@link AxisAngle4f} rotation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without post-multiplying,
|
|
* use {@link #rotation(AxisAngle4f)}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotate(float, float, float, float)
|
|
* @see #rotation(AxisAngle4f)
|
|
*
|
|
* @param axisAngle
|
|
* the {@link AxisAngle4f} (needs to be {@link AxisAngle4f#normalize() normalized})
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotate(AxisAngle4f axisAngle, Matrix4f dest) {
|
|
return rotate(axisAngle.angle, axisAngle.x, axisAngle.y, axisAngle.z, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a rotation transformation, rotating the given radians about the specified axis, to this matrix.
|
|
* <p>
|
|
* The axis described by the <code>axis</code> vector needs to be a unit vector.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>A</code> the rotation matrix obtained from the given axis-angle,
|
|
* then the new matrix will be <code>M * A</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * A * v</code>,
|
|
* the axis-angle rotation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without post-multiplying,
|
|
* use {@link #rotation(float, Vector3fc)}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotate(float, float, float, float)
|
|
* @see #rotation(float, Vector3fc)
|
|
*
|
|
* @param angle
|
|
* the angle in radians
|
|
* @param axis
|
|
* the rotation axis (needs to be {@link Vector3f#normalize() normalized})
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotate(float angle, Vector3fc axis) {
|
|
return rotate(angle, axis.x(), axis.y(), axis.z());
|
|
}
|
|
|
|
/**
|
|
* Apply a rotation transformation, rotating the given radians about the specified axis and store the result in <code>dest</code>.
|
|
* <p>
|
|
* The axis described by the <code>axis</code> vector needs to be a unit vector.
|
|
* <p>
|
|
* When used with a right-handed coordinate system, the produced rotation will rotate a vector
|
|
* counter-clockwise around the rotation axis, when viewing along the negative axis direction towards the origin.
|
|
* When used with a left-handed coordinate system, the rotation is clockwise.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>A</code> the rotation matrix obtained from the given axis-angle,
|
|
* then the new matrix will be <code>M * A</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * A * v</code>,
|
|
* the axis-angle rotation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without post-multiplying,
|
|
* use {@link #rotation(float, Vector3fc)}.
|
|
* <p>
|
|
* Reference: <a href="http://en.wikipedia.org/wiki/Rotation_matrix#Axis_and_angle">http://en.wikipedia.org</a>
|
|
*
|
|
* @see #rotate(float, float, float, float)
|
|
* @see #rotation(float, Vector3fc)
|
|
*
|
|
* @param angle
|
|
* the angle in radians
|
|
* @param axis
|
|
* the rotation axis (needs to be {@link Vector3f#normalize() normalized})
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotate(float angle, Vector3fc axis, Matrix4f dest) {
|
|
return rotate(angle, axis.x(), axis.y(), axis.z(), dest);
|
|
}
|
|
|
|
public Vector4f unproject(float winX, float winY, float winZ, int[] viewport, Vector4f dest) {
|
|
float a = m00 * m11 - m01 * m10;
|
|
float b = m00 * m12 - m02 * m10;
|
|
float c = m00 * m13 - m03 * m10;
|
|
float d = m01 * m12 - m02 * m11;
|
|
float e = m01 * m13 - m03 * m11;
|
|
float f = m02 * m13 - m03 * m12;
|
|
float g = m20 * m31 - m21 * m30;
|
|
float h = m20 * m32 - m22 * m30;
|
|
float i = m20 * m33 - m23 * m30;
|
|
float j = m21 * m32 - m22 * m31;
|
|
float k = m21 * m33 - m23 * m31;
|
|
float l = m22 * m33 - m23 * m32;
|
|
float det = a * l - b * k + c * j + d * i - e * h + f * g;
|
|
det = 1.0f / det;
|
|
float im00 = ( m11 * l - m12 * k + m13 * j) * det;
|
|
float im01 = (-m01 * l + m02 * k - m03 * j) * det;
|
|
float im02 = ( m31 * f - m32 * e + m33 * d) * det;
|
|
float im03 = (-m21 * f + m22 * e - m23 * d) * det;
|
|
float im10 = (-m10 * l + m12 * i - m13 * h) * det;
|
|
float im11 = ( m00 * l - m02 * i + m03 * h) * det;
|
|
float im12 = (-m30 * f + m32 * c - m33 * b) * det;
|
|
float im13 = ( m20 * f - m22 * c + m23 * b) * det;
|
|
float im20 = ( m10 * k - m11 * i + m13 * g) * det;
|
|
float im21 = (-m00 * k + m01 * i - m03 * g) * det;
|
|
float im22 = ( m30 * e - m31 * c + m33 * a) * det;
|
|
float im23 = (-m20 * e + m21 * c - m23 * a) * det;
|
|
float im30 = (-m10 * j + m11 * h - m12 * g) * det;
|
|
float im31 = ( m00 * j - m01 * h + m02 * g) * det;
|
|
float im32 = (-m30 * d + m31 * b - m32 * a) * det;
|
|
float im33 = ( m20 * d - m21 * b + m22 * a) * det;
|
|
float ndcX = (winX-viewport[0])/viewport[2]*2.0f-1.0f;
|
|
float ndcY = (winY-viewport[1])/viewport[3]*2.0f-1.0f;
|
|
float ndcZ = winZ+winZ-1.0f;
|
|
float invW = 1.0f / (im03 * ndcX + im13 * ndcY + im23 * ndcZ + im33);
|
|
return dest.set((im00 * ndcX + im10 * ndcY + im20 * ndcZ + im30) * invW,
|
|
(im01 * ndcX + im11 * ndcY + im21 * ndcZ + im31) * invW,
|
|
(im02 * ndcX + im12 * ndcY + im22 * ndcZ + im32) * invW,
|
|
1.0f);
|
|
}
|
|
|
|
public Vector3f unproject(float winX, float winY, float winZ, int[] viewport, Vector3f dest) {
|
|
float a = m00 * m11 - m01 * m10;
|
|
float b = m00 * m12 - m02 * m10;
|
|
float c = m00 * m13 - m03 * m10;
|
|
float d = m01 * m12 - m02 * m11;
|
|
float e = m01 * m13 - m03 * m11;
|
|
float f = m02 * m13 - m03 * m12;
|
|
float g = m20 * m31 - m21 * m30;
|
|
float h = m20 * m32 - m22 * m30;
|
|
float i = m20 * m33 - m23 * m30;
|
|
float j = m21 * m32 - m22 * m31;
|
|
float k = m21 * m33 - m23 * m31;
|
|
float l = m22 * m33 - m23 * m32;
|
|
float det = a * l - b * k + c * j + d * i - e * h + f * g;
|
|
det = 1.0f / det;
|
|
float im00 = ( m11 * l - m12 * k + m13 * j) * det;
|
|
float im01 = (-m01 * l + m02 * k - m03 * j) * det;
|
|
float im02 = ( m31 * f - m32 * e + m33 * d) * det;
|
|
float im03 = (-m21 * f + m22 * e - m23 * d) * det;
|
|
float im10 = (-m10 * l + m12 * i - m13 * h) * det;
|
|
float im11 = ( m00 * l - m02 * i + m03 * h) * det;
|
|
float im12 = (-m30 * f + m32 * c - m33 * b) * det;
|
|
float im13 = ( m20 * f - m22 * c + m23 * b) * det;
|
|
float im20 = ( m10 * k - m11 * i + m13 * g) * det;
|
|
float im21 = (-m00 * k + m01 * i - m03 * g) * det;
|
|
float im22 = ( m30 * e - m31 * c + m33 * a) * det;
|
|
float im23 = (-m20 * e + m21 * c - m23 * a) * det;
|
|
float im30 = (-m10 * j + m11 * h - m12 * g) * det;
|
|
float im31 = ( m00 * j - m01 * h + m02 * g) * det;
|
|
float im32 = (-m30 * d + m31 * b - m32 * a) * det;
|
|
float im33 = ( m20 * d - m21 * b + m22 * a) * det;
|
|
float ndcX = (winX-viewport[0])/viewport[2]*2.0f-1.0f;
|
|
float ndcY = (winY-viewport[1])/viewport[3]*2.0f-1.0f;
|
|
float ndcZ = winZ+winZ-1.0f;
|
|
float invW = 1.0f / (im03 * ndcX + im13 * ndcY + im23 * ndcZ + im33);
|
|
return dest.set((im00 * ndcX + im10 * ndcY + im20 * ndcZ + im30) * invW,
|
|
(im01 * ndcX + im11 * ndcY + im21 * ndcZ + im31) * invW,
|
|
(im02 * ndcX + im12 * ndcY + im22 * ndcZ + im32) * invW);
|
|
}
|
|
|
|
public Vector4f unproject(Vector3fc winCoords, int[] viewport, Vector4f dest) {
|
|
return unproject(winCoords.x(), winCoords.y(), winCoords.z(), viewport, dest);
|
|
}
|
|
|
|
public Vector3f unproject(Vector3fc winCoords, int[] viewport, Vector3f dest) {
|
|
return unproject(winCoords.x(), winCoords.y(), winCoords.z(), viewport, dest);
|
|
}
|
|
|
|
public Matrix4f unprojectRay(float winX, float winY, int[] viewport, Vector3f originDest, Vector3f dirDest) {
|
|
float a = m00 * m11 - m01 * m10;
|
|
float b = m00 * m12 - m02 * m10;
|
|
float c = m00 * m13 - m03 * m10;
|
|
float d = m01 * m12 - m02 * m11;
|
|
float e = m01 * m13 - m03 * m11;
|
|
float f = m02 * m13 - m03 * m12;
|
|
float g = m20 * m31 - m21 * m30;
|
|
float h = m20 * m32 - m22 * m30;
|
|
float i = m20 * m33 - m23 * m30;
|
|
float j = m21 * m32 - m22 * m31;
|
|
float k = m21 * m33 - m23 * m31;
|
|
float l = m22 * m33 - m23 * m32;
|
|
float det = a * l - b * k + c * j + d * i - e * h + f * g;
|
|
det = 1.0f / det;
|
|
float im00 = ( m11 * l - m12 * k + m13 * j) * det;
|
|
float im01 = (-m01 * l + m02 * k - m03 * j) * det;
|
|
float im02 = ( m31 * f - m32 * e + m33 * d) * det;
|
|
float im03 = (-m21 * f + m22 * e - m23 * d) * det;
|
|
float im10 = (-m10 * l + m12 * i - m13 * h) * det;
|
|
float im11 = ( m00 * l - m02 * i + m03 * h) * det;
|
|
float im12 = (-m30 * f + m32 * c - m33 * b) * det;
|
|
float im13 = ( m20 * f - m22 * c + m23 * b) * det;
|
|
float im20 = ( m10 * k - m11 * i + m13 * g) * det;
|
|
float im21 = (-m00 * k + m01 * i - m03 * g) * det;
|
|
float im22 = ( m30 * e - m31 * c + m33 * a) * det;
|
|
float im23 = (-m20 * e + m21 * c - m23 * a) * det;
|
|
float im30 = (-m10 * j + m11 * h - m12 * g) * det;
|
|
float im31 = ( m00 * j - m01 * h + m02 * g) * det;
|
|
float im32 = (-m30 * d + m31 * b - m32 * a) * det;
|
|
float im33 = ( m20 * d - m21 * b + m22 * a) * det;
|
|
float ndcX = (winX-viewport[0])/viewport[2]*2.0f-1.0f;
|
|
float ndcY = (winY-viewport[1])/viewport[3]*2.0f-1.0f;
|
|
float px = im00 * ndcX + im10 * ndcY + im30;
|
|
float py = im01 * ndcX + im11 * ndcY + im31;
|
|
float pz = im02 * ndcX + im12 * ndcY + im32;
|
|
float invNearW = 1.0f / (im03 * ndcX + im13 * ndcY - im23 + im33);
|
|
float nearX = (px - im20) * invNearW;
|
|
float nearY = (py - im21) * invNearW;
|
|
float nearZ = (pz - im22) * invNearW;
|
|
float invW0 = 1.0f / (im03 * ndcX + im13 * ndcY + im33);
|
|
float x0 = px * invW0;
|
|
float y0 = py * invW0;
|
|
float z0 = pz * invW0;
|
|
originDest.x = nearX; originDest.y = nearY; originDest.z = nearZ;
|
|
dirDest.x = x0 - nearX; dirDest.y = y0 - nearY; dirDest.z = z0 - nearZ;
|
|
return this;
|
|
}
|
|
|
|
public Matrix4f unprojectRay(Vector2fc winCoords, int[] viewport, Vector3f originDest, Vector3f dirDest) {
|
|
return unprojectRay(winCoords.x(), winCoords.y(), viewport, originDest, dirDest);
|
|
}
|
|
|
|
public Vector4f unprojectInv(Vector3fc winCoords, int[] viewport, Vector4f dest) {
|
|
return unprojectInv(winCoords.x(), winCoords.y(), winCoords.z(), viewport, dest);
|
|
}
|
|
|
|
public Vector4f unprojectInv(float winX, float winY, float winZ, int[] viewport, Vector4f dest) {
|
|
float ndcX = (winX-viewport[0])/viewport[2]*2.0f-1.0f;
|
|
float ndcY = (winY-viewport[1])/viewport[3]*2.0f-1.0f;
|
|
float ndcZ = winZ+winZ-1.0f;
|
|
float invW = 1.0f / (m03 * ndcX + m13 * ndcY + m23 * ndcZ + m33);
|
|
return dest.set((m00 * ndcX + m10 * ndcY + m20 * ndcZ + m30) * invW,
|
|
(m01 * ndcX + m11 * ndcY + m21 * ndcZ + m31) * invW,
|
|
(m02 * ndcX + m12 * ndcY + m22 * ndcZ + m32) * invW,
|
|
1.0f);
|
|
}
|
|
|
|
public Matrix4f unprojectInvRay(Vector2fc winCoords, int[] viewport, Vector3f originDest, Vector3f dirDest) {
|
|
return unprojectInvRay(winCoords.x(), winCoords.y(), viewport, originDest, dirDest);
|
|
}
|
|
|
|
public Matrix4f unprojectInvRay(float winX, float winY, int[] viewport, Vector3f originDest, Vector3f dirDest) {
|
|
float ndcX = (winX-viewport[0])/viewport[2]*2.0f-1.0f;
|
|
float ndcY = (winY-viewport[1])/viewport[3]*2.0f-1.0f;
|
|
float px = m00 * ndcX + m10 * ndcY + m30;
|
|
float py = m01 * ndcX + m11 * ndcY + m31;
|
|
float pz = m02 * ndcX + m12 * ndcY + m32;
|
|
float invNearW = 1.0f / (m03 * ndcX + m13 * ndcY - m23 + m33);
|
|
float nearX = (px - m20) * invNearW;
|
|
float nearY = (py - m21) * invNearW;
|
|
float nearZ = (pz - m22) * invNearW;
|
|
float invW0 = 1.0f / (m03 * ndcX + m13 * ndcY + m33);
|
|
float x0 = px * invW0;
|
|
float y0 = py * invW0;
|
|
float z0 = pz * invW0;
|
|
originDest.x = nearX; originDest.y = nearY; originDest.z = nearZ;
|
|
dirDest.x = x0 - nearX; dirDest.y = y0 - nearY; dirDest.z = z0 - nearZ;
|
|
return this;
|
|
}
|
|
|
|
public Vector3f unprojectInv(Vector3fc winCoords, int[] viewport, Vector3f dest) {
|
|
return unprojectInv(winCoords.x(), winCoords.y(), winCoords.z(), viewport, dest);
|
|
}
|
|
|
|
public Vector3f unprojectInv(float winX, float winY, float winZ, int[] viewport, Vector3f dest) {
|
|
float ndcX = (winX-viewport[0])/viewport[2]*2.0f-1.0f;
|
|
float ndcY = (winY-viewport[1])/viewport[3]*2.0f-1.0f;
|
|
float ndcZ = winZ+winZ-1.0f;
|
|
float invW = 1.0f / (m03 * ndcX + m13 * ndcY + m23 * ndcZ + m33);
|
|
return dest.set((m00 * ndcX + m10 * ndcY + m20 * ndcZ + m30) * invW,
|
|
(m01 * ndcX + m11 * ndcY + m21 * ndcZ + m31) * invW,
|
|
(m02 * ndcX + m12 * ndcY + m22 * ndcZ + m32) * invW);
|
|
}
|
|
|
|
public Vector4f project(float x, float y, float z, int[] viewport, Vector4f winCoordsDest) {
|
|
float invW = 1.0f / Math.fma(m03, x, Math.fma(m13, y, Math.fma(m23, z, m33)));
|
|
float nx = Math.fma(m00, x, Math.fma(m10, y, Math.fma(m20, z, m30))) * invW;
|
|
float ny = Math.fma(m01, x, Math.fma(m11, y, Math.fma(m21, z, m31))) * invW;
|
|
float nz = Math.fma(m02, x, Math.fma(m12, y, Math.fma(m22, z, m32))) * invW;
|
|
return winCoordsDest.set(Math.fma(Math.fma(nx, 0.5f, 0.5f), viewport[2], viewport[0]),
|
|
Math.fma(Math.fma(ny, 0.5f, 0.5f), viewport[3], viewport[1]),
|
|
Math.fma(0.5f, nz, 0.5f),
|
|
1.0f);
|
|
}
|
|
|
|
public Vector3f project(float x, float y, float z, int[] viewport, Vector3f winCoordsDest) {
|
|
float invW = 1.0f / Math.fma(m03, x, Math.fma(m13, y, Math.fma(m23, z, m33)));
|
|
float nx = Math.fma(m00, x, Math.fma(m10, y, Math.fma(m20, z, m30))) * invW;
|
|
float ny = Math.fma(m01, x, Math.fma(m11, y, Math.fma(m21, z, m31))) * invW;
|
|
float nz = Math.fma(m02, x, Math.fma(m12, y, Math.fma(m22, z, m32))) * invW;
|
|
winCoordsDest.x = Math.fma(Math.fma(nx, 0.5f, 0.5f), viewport[2], viewport[0]);
|
|
winCoordsDest.y = Math.fma(Math.fma(ny, 0.5f, 0.5f), viewport[3], viewport[1]);
|
|
winCoordsDest.z = Math.fma(0.5f, nz, 0.5f);
|
|
return winCoordsDest;
|
|
}
|
|
|
|
public Vector4f project(Vector3fc position, int[] viewport, Vector4f winCoordsDest) {
|
|
return project(position.x(), position.y(), position.z(), viewport, winCoordsDest);
|
|
}
|
|
|
|
public Vector3f project(Vector3fc position, int[] viewport, Vector3f winCoordsDest) {
|
|
return project(position.x(), position.y(), position.z(), viewport, winCoordsDest);
|
|
}
|
|
|
|
public Matrix4f reflect(float a, float b, float c, float d, Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.reflection(a, b, c, d);
|
|
else if ((properties & PROPERTY_AFFINE) != 0)
|
|
return reflectAffine(a, b, c, d, dest);
|
|
return reflectGeneric(a, b, c, d, dest);
|
|
}
|
|
private Matrix4f reflectAffine(float a, float b, float c, float d, Matrix4f dest) {
|
|
float da = a + a, db = b + b, dc = c + c, dd = d + d;
|
|
float rm00 = 1.0f - da * a;
|
|
float rm01 = -da * b;
|
|
float rm02 = -da * c;
|
|
float rm10 = -db * a;
|
|
float rm11 = 1.0f - db * b;
|
|
float rm12 = -db * c;
|
|
float rm20 = -dc * a;
|
|
float rm21 = -dc * b;
|
|
float rm22 = 1.0f - dc * c;
|
|
float rm30 = -dd * a;
|
|
float rm31 = -dd * b;
|
|
float rm32 = -dd * c;
|
|
// matrix multiplication
|
|
dest._m30(m00 * rm30 + m10 * rm31 + m20 * rm32 + m30)
|
|
._m31(m01 * rm30 + m11 * rm31 + m21 * rm32 + m31)
|
|
._m32(m02 * rm30 + m12 * rm31 + m22 * rm32 + m32)
|
|
._m33(m33);
|
|
float nm00 = m00 * rm00 + m10 * rm01 + m20 * rm02;
|
|
float nm01 = m01 * rm00 + m11 * rm01 + m21 * rm02;
|
|
float nm02 = m02 * rm00 + m12 * rm01 + m22 * rm02;
|
|
float nm10 = m00 * rm10 + m10 * rm11 + m20 * rm12;
|
|
float nm11 = m01 * rm10 + m11 * rm11 + m21 * rm12;
|
|
float nm12 = m02 * rm10 + m12 * rm11 + m22 * rm12;
|
|
dest._m20(m00 * rm20 + m10 * rm21 + m20 * rm22)
|
|
._m21(m01 * rm20 + m11 * rm21 + m21 * rm22)
|
|
._m22(m02 * rm20 + m12 * rm21 + m22 * rm22)
|
|
._m23(0.0f)
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(0.0f)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(0.0f)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
return dest;
|
|
}
|
|
private Matrix4f reflectGeneric(float a, float b, float c, float d, Matrix4f dest) {
|
|
float da = a + a, db = b + b, dc = c + c, dd = d + d;
|
|
float rm00 = 1.0f - da * a;
|
|
float rm01 = -da * b;
|
|
float rm02 = -da * c;
|
|
float rm10 = -db * a;
|
|
float rm11 = 1.0f - db * b;
|
|
float rm12 = -db * c;
|
|
float rm20 = -dc * a;
|
|
float rm21 = -dc * b;
|
|
float rm22 = 1.0f - dc * c;
|
|
float rm30 = -dd * a;
|
|
float rm31 = -dd * b;
|
|
float rm32 = -dd * c;
|
|
// matrix multiplication
|
|
dest._m30(m00 * rm30 + m10 * rm31 + m20 * rm32 + m30)
|
|
._m31(m01 * rm30 + m11 * rm31 + m21 * rm32 + m31)
|
|
._m32(m02 * rm30 + m12 * rm31 + m22 * rm32 + m32)
|
|
._m33(m03 * rm30 + m13 * rm31 + m23 * rm32 + m33);
|
|
float nm00 = m00 * rm00 + m10 * rm01 + m20 * rm02;
|
|
float nm01 = m01 * rm00 + m11 * rm01 + m21 * rm02;
|
|
float nm02 = m02 * rm00 + m12 * rm01 + m22 * rm02;
|
|
float nm03 = m03 * rm00 + m13 * rm01 + m23 * rm02;
|
|
float nm10 = m00 * rm10 + m10 * rm11 + m20 * rm12;
|
|
float nm11 = m01 * rm10 + m11 * rm11 + m21 * rm12;
|
|
float nm12 = m02 * rm10 + m12 * rm11 + m22 * rm12;
|
|
float nm13 = m03 * rm10 + m13 * rm11 + m23 * rm12;
|
|
dest._m20(m00 * rm20 + m10 * rm21 + m20 * rm22)
|
|
._m21(m01 * rm20 + m11 * rm21 + m21 * rm22)
|
|
._m22(m02 * rm20 + m12 * rm21 + m22 * rm22)
|
|
._m23(m03 * rm20 + m13 * rm21 + m23 * rm22)
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply a mirror/reflection transformation to this matrix that reflects about the given plane
|
|
* specified via the equation <code>x*a + y*b + z*c + d = 0</code>.
|
|
* <p>
|
|
* The vector <code>(a, b, c)</code> must be a unit vector.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the reflection matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* reflection will be applied first!
|
|
* <p>
|
|
* Reference: <a href="https://msdn.microsoft.com/en-us/library/windows/desktop/bb281733(v=vs.85).aspx">msdn.microsoft.com</a>
|
|
*
|
|
* @param a
|
|
* the x factor in the plane equation
|
|
* @param b
|
|
* the y factor in the plane equation
|
|
* @param c
|
|
* the z factor in the plane equation
|
|
* @param d
|
|
* the constant in the plane equation
|
|
* @return this
|
|
*/
|
|
public Matrix4f reflect(float a, float b, float c, float d) {
|
|
return reflect(a, b, c, d, this);
|
|
}
|
|
|
|
/**
|
|
* Apply a mirror/reflection transformation to this matrix that reflects about the given plane
|
|
* specified via the plane normal and a point on the plane.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the reflection matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* reflection will be applied first!
|
|
*
|
|
* @param nx
|
|
* the x-coordinate of the plane normal
|
|
* @param ny
|
|
* the y-coordinate of the plane normal
|
|
* @param nz
|
|
* the z-coordinate of the plane normal
|
|
* @param px
|
|
* the x-coordinate of a point on the plane
|
|
* @param py
|
|
* the y-coordinate of a point on the plane
|
|
* @param pz
|
|
* the z-coordinate of a point on the plane
|
|
* @return this
|
|
*/
|
|
public Matrix4f reflect(float nx, float ny, float nz, float px, float py, float pz) {
|
|
return reflect(nx, ny, nz, px, py, pz, this);
|
|
}
|
|
|
|
public Matrix4f reflect(float nx, float ny, float nz, float px, float py, float pz, Matrix4f dest) {
|
|
float invLength = Math.invsqrt(nx * nx + ny * ny + nz * nz);
|
|
float nnx = nx * invLength;
|
|
float nny = ny * invLength;
|
|
float nnz = nz * invLength;
|
|
/* See: http://mathworld.wolfram.com/Plane.html */
|
|
return reflect(nnx, nny, nnz, -nnx * px - nny * py - nnz * pz, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a mirror/reflection transformation to this matrix that reflects about the given plane
|
|
* specified via the plane normal and a point on the plane.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the reflection matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* reflection will be applied first!
|
|
*
|
|
* @param normal
|
|
* the plane normal
|
|
* @param point
|
|
* a point on the plane
|
|
* @return this
|
|
*/
|
|
public Matrix4f reflect(Vector3fc normal, Vector3fc point) {
|
|
return reflect(normal.x(), normal.y(), normal.z(), point.x(), point.y(), point.z());
|
|
}
|
|
|
|
/**
|
|
* Apply a mirror/reflection transformation to this matrix that reflects about a plane
|
|
* specified via the plane orientation and a point on the plane.
|
|
* <p>
|
|
* This method can be used to build a reflection transformation based on the orientation of a mirror object in the scene.
|
|
* It is assumed that the default mirror plane's normal is <code>(0, 0, 1)</code>. So, if the given {@link Quaternionfc} is
|
|
* the identity (does not apply any additional rotation), the reflection plane will be <code>z=0</code>, offset by the given <code>point</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>R</code> the reflection matrix,
|
|
* then the new matrix will be <code>M * R</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * R * v</code>, the
|
|
* reflection will be applied first!
|
|
*
|
|
* @param orientation
|
|
* the plane orientation
|
|
* @param point
|
|
* a point on the plane
|
|
* @return this
|
|
*/
|
|
public Matrix4f reflect(Quaternionfc orientation, Vector3fc point) {
|
|
return reflect(orientation, point, this);
|
|
}
|
|
|
|
public Matrix4f reflect(Quaternionfc orientation, Vector3fc point, Matrix4f dest) {
|
|
double num1 = orientation.x() + orientation.x();
|
|
double num2 = orientation.y() + orientation.y();
|
|
double num3 = orientation.z() + orientation.z();
|
|
float normalX = (float) (orientation.x() * num3 + orientation.w() * num2);
|
|
float normalY = (float) (orientation.y() * num3 - orientation.w() * num1);
|
|
float normalZ = (float) (1.0 - (orientation.x() * num1 + orientation.y() * num2));
|
|
return reflect(normalX, normalY, normalZ, point.x(), point.y(), point.z(), dest);
|
|
}
|
|
|
|
public Matrix4f reflect(Vector3fc normal, Vector3fc point, Matrix4f dest) {
|
|
return reflect(normal.x(), normal.y(), normal.z(), point.x(), point.y(), point.z(), dest);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a mirror/reflection transformation that reflects about the given plane
|
|
* specified via the equation <code>x*a + y*b + z*c + d = 0</code>.
|
|
* <p>
|
|
* The vector <code>(a, b, c)</code> must be a unit vector.
|
|
* <p>
|
|
* Reference: <a href="https://msdn.microsoft.com/en-us/library/windows/desktop/bb281733(v=vs.85).aspx">msdn.microsoft.com</a>
|
|
*
|
|
* @param a
|
|
* the x factor in the plane equation
|
|
* @param b
|
|
* the y factor in the plane equation
|
|
* @param c
|
|
* the z factor in the plane equation
|
|
* @param d
|
|
* the constant in the plane equation
|
|
* @return this
|
|
*/
|
|
public Matrix4f reflection(float a, float b, float c, float d) {
|
|
float da = a + a, db = b + b, dc = c + c, dd = d + d;
|
|
this._m00(1.0f - da * a)
|
|
._m01(-da * b)
|
|
._m02(-da * c)
|
|
._m03(0.0f)
|
|
._m10(-db * a)
|
|
._m11(1.0f - db * b)
|
|
._m12(-db * c)
|
|
._m13(0.0f)
|
|
._m20(-dc * a)
|
|
._m21(-dc * b)
|
|
._m22(1.0f - dc * c)
|
|
._m23(0.0f)
|
|
._m30(-dd * a)
|
|
._m31(-dd * b)
|
|
._m32(-dd * c)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a mirror/reflection transformation that reflects about the given plane
|
|
* specified via the plane normal and a point on the plane.
|
|
*
|
|
* @param nx
|
|
* the x-coordinate of the plane normal
|
|
* @param ny
|
|
* the y-coordinate of the plane normal
|
|
* @param nz
|
|
* the z-coordinate of the plane normal
|
|
* @param px
|
|
* the x-coordinate of a point on the plane
|
|
* @param py
|
|
* the y-coordinate of a point on the plane
|
|
* @param pz
|
|
* the z-coordinate of a point on the plane
|
|
* @return this
|
|
*/
|
|
public Matrix4f reflection(float nx, float ny, float nz, float px, float py, float pz) {
|
|
float invLength = Math.invsqrt(nx * nx + ny * ny + nz * nz);
|
|
float nnx = nx * invLength;
|
|
float nny = ny * invLength;
|
|
float nnz = nz * invLength;
|
|
/* See: http://mathworld.wolfram.com/Plane.html */
|
|
return reflection(nnx, nny, nnz, -nnx * px - nny * py - nnz * pz);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a mirror/reflection transformation that reflects about the given plane
|
|
* specified via the plane normal and a point on the plane.
|
|
*
|
|
* @param normal
|
|
* the plane normal
|
|
* @param point
|
|
* a point on the plane
|
|
* @return this
|
|
*/
|
|
public Matrix4f reflection(Vector3fc normal, Vector3fc point) {
|
|
return reflection(normal.x(), normal.y(), normal.z(), point.x(), point.y(), point.z());
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a mirror/reflection transformation that reflects about a plane
|
|
* specified via the plane orientation and a point on the plane.
|
|
* <p>
|
|
* This method can be used to build a reflection transformation based on the orientation of a mirror object in the scene.
|
|
* It is assumed that the default mirror plane's normal is <code>(0, 0, 1)</code>. So, if the given {@link Quaternionfc} is
|
|
* the identity (does not apply any additional rotation), the reflection plane will be <code>z=0</code>, offset by the given <code>point</code>.
|
|
*
|
|
* @param orientation
|
|
* the plane orientation
|
|
* @param point
|
|
* a point on the plane
|
|
* @return this
|
|
*/
|
|
public Matrix4f reflection(Quaternionfc orientation, Vector3fc point) {
|
|
double num1 = orientation.x() + orientation.x();
|
|
double num2 = orientation.y() + orientation.y();
|
|
double num3 = orientation.z() + orientation.z();
|
|
float normalX = (float) (orientation.x() * num3 + orientation.w() * num2);
|
|
float normalY = (float) (orientation.y() * num3 - orientation.w() * num1);
|
|
float normalZ = (float) (1.0 - (orientation.x() * num1 + orientation.y() * num2));
|
|
return reflection(normalX, normalY, normalZ, point.x(), point.y(), point.z());
|
|
}
|
|
|
|
public Vector4f getRow(int row, Vector4f dest) throws IndexOutOfBoundsException {
|
|
switch (row) {
|
|
case 0:
|
|
return dest.set(m00, m10, m20, m30);
|
|
case 1:
|
|
return dest.set(m01, m11, m21, m31);
|
|
case 2:
|
|
return dest.set(m02, m12, m22, m32);
|
|
case 3:
|
|
return dest.set(m03, m13, m23, m33);
|
|
default:
|
|
throw new IndexOutOfBoundsException();
|
|
}
|
|
}
|
|
|
|
public Vector3f getRow(int row, Vector3f dest) throws IndexOutOfBoundsException {
|
|
switch (row) {
|
|
case 0:
|
|
return dest.set(m00, m10, m20);
|
|
case 1:
|
|
return dest.set(m01, m11, m21);
|
|
case 2:
|
|
return dest.set(m02, m12, m22);
|
|
case 3:
|
|
return dest.set(m03, m13, m23);
|
|
default:
|
|
throw new IndexOutOfBoundsException();
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Set the row at the given <code>row</code> index, starting with <code>0</code>.
|
|
*
|
|
* @param row
|
|
* the row index in <code>[0..3]</code>
|
|
* @param src
|
|
* the row components to set
|
|
* @return this
|
|
* @throws IndexOutOfBoundsException if <code>row</code> is not in <code>[0..3]</code>
|
|
*/
|
|
public Matrix4f setRow(int row, Vector4fc src) throws IndexOutOfBoundsException {
|
|
switch (row) {
|
|
case 0:
|
|
return _m00(src.x())._m10(src.y())._m20(src.z())._m30(src.w())._properties(0);
|
|
case 1:
|
|
return _m01(src.x())._m11(src.y())._m21(src.z())._m31(src.w())._properties(0);
|
|
case 2:
|
|
return _m02(src.x())._m12(src.y())._m22(src.z())._m32(src.w())._properties(0);
|
|
case 3:
|
|
return _m03(src.x())._m13(src.y())._m23(src.z())._m33(src.w())._properties(0);
|
|
default:
|
|
throw new IndexOutOfBoundsException();
|
|
}
|
|
}
|
|
|
|
public Vector4f getColumn(int column, Vector4f dest) throws IndexOutOfBoundsException {
|
|
return MemUtil.INSTANCE.getColumn(this, column, dest);
|
|
}
|
|
|
|
public Vector3f getColumn(int column, Vector3f dest) throws IndexOutOfBoundsException {
|
|
switch (column) {
|
|
case 0:
|
|
return dest.set(m00, m01, m02);
|
|
case 1:
|
|
return dest.set(m10, m11, m12);
|
|
case 2:
|
|
return dest.set(m20, m21, m22);
|
|
case 3:
|
|
return dest.set(m30, m31, m32);
|
|
default:
|
|
throw new IndexOutOfBoundsException();
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Set the column at the given <code>column</code> index, starting with <code>0</code>.
|
|
*
|
|
* @param column
|
|
* the column index in <code>[0..3]</code>
|
|
* @param src
|
|
* the column components to set
|
|
* @return this
|
|
* @throws IndexOutOfBoundsException if <code>column</code> is not in <code>[0..3]</code>
|
|
*/
|
|
public Matrix4f setColumn(int column, Vector4fc src) throws IndexOutOfBoundsException {
|
|
if (src instanceof Vector4f)
|
|
return MemUtil.INSTANCE.setColumn((Vector4f) src, column, this)._properties(0);
|
|
return MemUtil.INSTANCE.setColumn(src, column, this)._properties(0);
|
|
}
|
|
|
|
public float get(int column, int row) {
|
|
return MemUtil.INSTANCE.get(this, column, row);
|
|
}
|
|
|
|
/**
|
|
* Set the matrix element at the given column and row to the specified value.
|
|
*
|
|
* @param column
|
|
* the colum index in <code>[0..3]</code>
|
|
* @param row
|
|
* the row index in <code>[0..3]</code>
|
|
* @param value
|
|
* the value
|
|
* @return this
|
|
*/
|
|
public Matrix4f set(int column, int row, float value) {
|
|
return MemUtil.INSTANCE.set(this, column, row, value);
|
|
}
|
|
|
|
public float getRowColumn(int row, int column) {
|
|
return MemUtil.INSTANCE.get(this, column, row);
|
|
}
|
|
|
|
/**
|
|
* Set the matrix element at the given row and column to the specified value.
|
|
*
|
|
* @param row
|
|
* the row index in <code>[0..3]</code>
|
|
* @param column
|
|
* the colum index in <code>[0..3]</code>
|
|
* @param value
|
|
* the value
|
|
* @return this
|
|
*/
|
|
public Matrix4f setRowColumn(int row, int column, float value) {
|
|
return MemUtil.INSTANCE.set(this, column, row, value);
|
|
}
|
|
|
|
/**
|
|
* Compute a normal matrix from the upper left 3x3 submatrix of <code>this</code>
|
|
* and store it into the upper left 3x3 submatrix of <code>this</code>.
|
|
* All other values of <code>this</code> will be set to {@link #identity() identity}.
|
|
* <p>
|
|
* The normal matrix of <code>m</code> is the transpose of the inverse of <code>m</code>.
|
|
* <p>
|
|
* Please note that, if <code>this</code> is an orthogonal matrix or a matrix whose columns are orthogonal vectors,
|
|
* then this method <i>need not</i> be invoked, since in that case <code>this</code> itself is its normal matrix.
|
|
* In that case, use {@link #set3x3(Matrix4f)} to set a given Matrix4f to only the upper left 3x3 submatrix
|
|
* of this matrix.
|
|
*
|
|
* @see #set3x3(Matrix4f)
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f normal() {
|
|
return normal(this);
|
|
}
|
|
|
|
/**
|
|
* Compute a normal matrix from the upper left 3x3 submatrix of <code>this</code>
|
|
* and store it into the upper left 3x3 submatrix of <code>dest</code>.
|
|
* All other values of <code>dest</code> will be set to {@link #identity() identity}.
|
|
* <p>
|
|
* The normal matrix of <code>m</code> is the transpose of the inverse of <code>m</code>.
|
|
* <p>
|
|
* Please note that, if <code>this</code> is an orthogonal matrix or a matrix whose columns are orthogonal vectors,
|
|
* then this method <i>need not</i> be invoked, since in that case <code>this</code> itself is its normal matrix.
|
|
* In that case, use {@link #set3x3(Matrix4f)} to set a given Matrix4f to only the upper left 3x3 submatrix
|
|
* of this matrix.
|
|
*
|
|
* @see #set3x3(Matrix4f)
|
|
*
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f normal(Matrix4f dest) {
|
|
if ((properties & PROPERTY_IDENTITY) != 0)
|
|
return dest.identity();
|
|
else if ((properties & PROPERTY_ORTHONORMAL) != 0)
|
|
return normalOrthonormal(dest);
|
|
return normalGeneric(dest);
|
|
}
|
|
private Matrix4f normalOrthonormal(Matrix4f dest) {
|
|
if (dest != this)
|
|
dest.set(this);
|
|
return dest._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
}
|
|
private Matrix4f normalGeneric(Matrix4f dest) {
|
|
float m00m11 = m00 * m11;
|
|
float m01m10 = m01 * m10;
|
|
float m02m10 = m02 * m10;
|
|
float m00m12 = m00 * m12;
|
|
float m01m12 = m01 * m12;
|
|
float m02m11 = m02 * m11;
|
|
float det = (m00m11 - m01m10) * m22 + (m02m10 - m00m12) * m21 + (m01m12 - m02m11) * m20;
|
|
float s = 1.0f / det;
|
|
/* Invert and transpose in one go */
|
|
float nm00 = (m11 * m22 - m21 * m12) * s;
|
|
float nm01 = (m20 * m12 - m10 * m22) * s;
|
|
float nm02 = (m10 * m21 - m20 * m11) * s;
|
|
float nm10 = (m21 * m02 - m01 * m22) * s;
|
|
float nm11 = (m00 * m22 - m20 * m02) * s;
|
|
float nm12 = (m20 * m01 - m00 * m21) * s;
|
|
float nm20 = (m01m12 - m02m11) * s;
|
|
float nm21 = (m02m10 - m00m12) * s;
|
|
float nm22 = (m00m11 - m01m10) * s;
|
|
return dest
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(0.0f)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(0.0f)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(0.0f)
|
|
._m30(0.0f)
|
|
._m31(0.0f)
|
|
._m32(0.0f)
|
|
._m33(1.0f)
|
|
._properties((properties | PROPERTY_AFFINE) & ~(PROPERTY_TRANSLATION | PROPERTY_PERSPECTIVE));
|
|
}
|
|
|
|
/**
|
|
* Compute a normal matrix from the upper left 3x3 submatrix of <code>this</code>
|
|
* and store it into <code>dest</code>.
|
|
* <p>
|
|
* The normal matrix of <code>m</code> is the transpose of the inverse of <code>m</code>.
|
|
* <p>
|
|
* Please note that, if <code>this</code> is an orthogonal matrix or a matrix whose columns are orthogonal vectors,
|
|
* then this method <i>need not</i> be invoked, since in that case <code>this</code> itself is its normal matrix.
|
|
* In that case, use {@link Matrix3f#set(Matrix4fc)} to set a given Matrix3f to only the upper left 3x3 submatrix
|
|
* of this matrix.
|
|
*
|
|
* @see Matrix3f#set(Matrix4fc)
|
|
* @see #get3x3(Matrix3f)
|
|
*
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix3f normal(Matrix3f dest) {
|
|
if ((properties & PROPERTY_ORTHONORMAL) != 0)
|
|
return normalOrthonormal(dest);
|
|
return normalGeneric(dest);
|
|
}
|
|
private Matrix3f normalOrthonormal(Matrix3f dest) {
|
|
dest.set(this);
|
|
return dest;
|
|
}
|
|
private Matrix3f normalGeneric(Matrix3f dest) {
|
|
float det = (m00 * m11 - (m01 * m10)) * m22
|
|
+ (m02 * m10 - (m00 * m12)) * m21
|
|
+ (m01 * m12 - (m02 * m11)) * m20;
|
|
float s = 1.0f / det;
|
|
/* Invert and transpose in one go */
|
|
return dest._m00((m11 * m22 - m21 * m12) * s)
|
|
._m01((m20 * m12 - m10 * m22) * s)
|
|
._m02((m10 * m21 - m20 * m11) * s)
|
|
._m10((m21 * m02 - m01 * m22) * s)
|
|
._m11((m00 * m22 - m20 * m02) * s)
|
|
._m12((m20 * m01 - m00 * m21) * s)
|
|
._m20((m01 * m12 - m02 * m11) * s)
|
|
._m21((m02 * m10 - m00 * m12) * s)
|
|
._m22((m00 * m11 - m01 * m10) * s);
|
|
}
|
|
|
|
/**
|
|
* Compute the cofactor matrix of the upper left 3x3 submatrix of <code>this</code>.
|
|
* <p>
|
|
* The cofactor matrix can be used instead of {@link #normal()} to transform normals
|
|
* when the orientation of the normals with respect to the surface should be preserved.
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f cofactor3x3() {
|
|
return cofactor3x3(this);
|
|
}
|
|
|
|
/**
|
|
* Compute the cofactor matrix of the upper left 3x3 submatrix of <code>this</code>
|
|
* and store it into <code>dest</code>.
|
|
* <p>
|
|
* The cofactor matrix can be used instead of {@link #normal(Matrix3f)} to transform normals
|
|
* when the orientation of the normals with respect to the surface should be preserved.
|
|
*
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix3f cofactor3x3(Matrix3f dest) {
|
|
return dest._m00(m11 * m22 - m21 * m12)
|
|
._m01(m20 * m12 - m10 * m22)
|
|
._m02(m10 * m21 - m20 * m11)
|
|
._m10(m21 * m02 - m01 * m22)
|
|
._m11(m00 * m22 - m20 * m02)
|
|
._m12(m20 * m01 - m00 * m21)
|
|
._m20(m01 * m12 - m02 * m11)
|
|
._m21(m02 * m10 - m00 * m12)
|
|
._m22(m00 * m11 - m01 * m10);
|
|
}
|
|
|
|
/**
|
|
* Compute the cofactor matrix of the upper left 3x3 submatrix of <code>this</code>
|
|
* and store it into <code>dest</code>.
|
|
* All other values of <code>dest</code> will be set to {@link #identity() identity}.
|
|
* <p>
|
|
* The cofactor matrix can be used instead of {@link #normal(Matrix4f)} to transform normals
|
|
* when the orientation of the normals with respect to the surface should be preserved.
|
|
*
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f cofactor3x3(Matrix4f dest) {
|
|
float nm10 = m21 * m02 - m01 * m22;
|
|
float nm11 = m00 * m22 - m20 * m02;
|
|
float nm12 = m20 * m01 - m00 * m21;
|
|
float nm20 = m01 * m12 - m11 * m02;
|
|
float nm21 = m02 * m10 - m12 * m00;
|
|
float nm22 = m00 * m11 - m10 * m01;
|
|
return dest
|
|
._m00(m11 * m22 - m21 * m12)
|
|
._m01(m20 * m12 - m10 * m22)
|
|
._m02(m10 * m21 - m20 * m11)
|
|
._m03(0.0f)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(0.0f)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(0.0f)
|
|
._m30(0.0f)
|
|
._m31(0.0f)
|
|
._m32(0.0f)
|
|
._m33(1.0f)
|
|
._properties((properties | PROPERTY_AFFINE) & ~(PROPERTY_TRANSLATION | PROPERTY_PERSPECTIVE));
|
|
}
|
|
|
|
/**
|
|
* Normalize the upper left 3x3 submatrix of this matrix.
|
|
* <p>
|
|
* The resulting matrix will map unit vectors to unit vectors, though a pair of orthogonal input unit
|
|
* vectors need not be mapped to a pair of orthogonal output vectors if the original matrix was not orthogonal itself
|
|
* (i.e. had <i>skewing</i>).
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f normalize3x3() {
|
|
return normalize3x3(this);
|
|
}
|
|
|
|
public Matrix4f normalize3x3(Matrix4f dest) {
|
|
float invXlen = Math.invsqrt(m00 * m00 + m01 * m01 + m02 * m02);
|
|
float invYlen = Math.invsqrt(m10 * m10 + m11 * m11 + m12 * m12);
|
|
float invZlen = Math.invsqrt(m20 * m20 + m21 * m21 + m22 * m22);
|
|
return dest
|
|
._m00(m00 * invXlen)._m01(m01 * invXlen)._m02(m02 * invXlen)
|
|
._m10(m10 * invYlen)._m11(m11 * invYlen)._m12(m12 * invYlen)
|
|
._m20(m20 * invZlen)._m21(m21 * invZlen)._m22(m22 * invZlen)
|
|
._m30(m30)._m31(m31)._m32(m32)._m33(m33)
|
|
._properties(properties);
|
|
}
|
|
|
|
public Matrix3f normalize3x3(Matrix3f dest) {
|
|
float invXlen = Math.invsqrt(m00 * m00 + m01 * m01 + m02 * m02);
|
|
float invYlen = Math.invsqrt(m10 * m10 + m11 * m11 + m12 * m12);
|
|
float invZlen = Math.invsqrt(m20 * m20 + m21 * m21 + m22 * m22);
|
|
return dest
|
|
._m00(m00 * invXlen)._m01(m01 * invXlen)._m02(m02 * invXlen)
|
|
._m10(m10 * invYlen)._m11(m11 * invYlen)._m12(m12 * invYlen)
|
|
._m20(m20 * invZlen)._m21(m21 * invZlen)._m22(m22 * invZlen);
|
|
}
|
|
|
|
public Vector4f frustumPlane(int plane, Vector4f dest) {
|
|
switch (plane) {
|
|
case PLANE_NX:
|
|
dest.set(m03 + m00, m13 + m10, m23 + m20, m33 + m30).normalize3();
|
|
break;
|
|
case PLANE_PX:
|
|
dest.set(m03 - m00, m13 - m10, m23 - m20, m33 - m30).normalize3();
|
|
break;
|
|
case PLANE_NY:
|
|
dest.set(m03 + m01, m13 + m11, m23 + m21, m33 + m31).normalize3();
|
|
break;
|
|
case PLANE_PY:
|
|
dest.set(m03 - m01, m13 - m11, m23 - m21, m33 - m31).normalize3();
|
|
break;
|
|
case PLANE_NZ:
|
|
dest.set(m03 + m02, m13 + m12, m23 + m22, m33 + m32).normalize3();
|
|
break;
|
|
case PLANE_PZ:
|
|
dest.set(m03 - m02, m13 - m12, m23 - m22, m33 - m32).normalize3();
|
|
break;
|
|
default:
|
|
throw new IllegalArgumentException("dest"); //$NON-NLS-1$
|
|
}
|
|
return dest;
|
|
}
|
|
|
|
public Vector3f frustumCorner(int corner, Vector3f point) {
|
|
float d1, d2, d3;
|
|
float n1x, n1y, n1z, n2x, n2y, n2z, n3x, n3y, n3z;
|
|
switch (corner) {
|
|
case CORNER_NXNYNZ: // left, bottom, near
|
|
n1x = m03 + m00; n1y = m13 + m10; n1z = m23 + m20; d1 = m33 + m30; // left
|
|
n2x = m03 + m01; n2y = m13 + m11; n2z = m23 + m21; d2 = m33 + m31; // bottom
|
|
n3x = m03 + m02; n3y = m13 + m12; n3z = m23 + m22; d3 = m33 + m32; // near
|
|
break;
|
|
case CORNER_PXNYNZ: // right, bottom, near
|
|
n1x = m03 - m00; n1y = m13 - m10; n1z = m23 - m20; d1 = m33 - m30; // right
|
|
n2x = m03 + m01; n2y = m13 + m11; n2z = m23 + m21; d2 = m33 + m31; // bottom
|
|
n3x = m03 + m02; n3y = m13 + m12; n3z = m23 + m22; d3 = m33 + m32; // near
|
|
break;
|
|
case CORNER_PXPYNZ: // right, top, near
|
|
n1x = m03 - m00; n1y = m13 - m10; n1z = m23 - m20; d1 = m33 - m30; // right
|
|
n2x = m03 - m01; n2y = m13 - m11; n2z = m23 - m21; d2 = m33 - m31; // top
|
|
n3x = m03 + m02; n3y = m13 + m12; n3z = m23 + m22; d3 = m33 + m32; // near
|
|
break;
|
|
case CORNER_NXPYNZ: // left, top, near
|
|
n1x = m03 + m00; n1y = m13 + m10; n1z = m23 + m20; d1 = m33 + m30; // left
|
|
n2x = m03 - m01; n2y = m13 - m11; n2z = m23 - m21; d2 = m33 - m31; // top
|
|
n3x = m03 + m02; n3y = m13 + m12; n3z = m23 + m22; d3 = m33 + m32; // near
|
|
break;
|
|
case CORNER_PXNYPZ: // right, bottom, far
|
|
n1x = m03 - m00; n1y = m13 - m10; n1z = m23 - m20; d1 = m33 - m30; // right
|
|
n2x = m03 + m01; n2y = m13 + m11; n2z = m23 + m21; d2 = m33 + m31; // bottom
|
|
n3x = m03 - m02; n3y = m13 - m12; n3z = m23 - m22; d3 = m33 - m32; // far
|
|
break;
|
|
case CORNER_NXNYPZ: // left, bottom, far
|
|
n1x = m03 + m00; n1y = m13 + m10; n1z = m23 + m20; d1 = m33 + m30; // left
|
|
n2x = m03 + m01; n2y = m13 + m11; n2z = m23 + m21; d2 = m33 + m31; // bottom
|
|
n3x = m03 - m02; n3y = m13 - m12; n3z = m23 - m22; d3 = m33 - m32; // far
|
|
break;
|
|
case CORNER_NXPYPZ: // left, top, far
|
|
n1x = m03 + m00; n1y = m13 + m10; n1z = m23 + m20; d1 = m33 + m30; // left
|
|
n2x = m03 - m01; n2y = m13 - m11; n2z = m23 - m21; d2 = m33 - m31; // top
|
|
n3x = m03 - m02; n3y = m13 - m12; n3z = m23 - m22; d3 = m33 - m32; // far
|
|
break;
|
|
case CORNER_PXPYPZ: // right, top, far
|
|
n1x = m03 - m00; n1y = m13 - m10; n1z = m23 - m20; d1 = m33 - m30; // right
|
|
n2x = m03 - m01; n2y = m13 - m11; n2z = m23 - m21; d2 = m33 - m31; // top
|
|
n3x = m03 - m02; n3y = m13 - m12; n3z = m23 - m22; d3 = m33 - m32; // far
|
|
break;
|
|
default:
|
|
throw new IllegalArgumentException("corner"); //$NON-NLS-1$
|
|
}
|
|
float c23x, c23y, c23z;
|
|
c23x = n2y * n3z - n2z * n3y;
|
|
c23y = n2z * n3x - n2x * n3z;
|
|
c23z = n2x * n3y - n2y * n3x;
|
|
float c31x, c31y, c31z;
|
|
c31x = n3y * n1z - n3z * n1y;
|
|
c31y = n3z * n1x - n3x * n1z;
|
|
c31z = n3x * n1y - n3y * n1x;
|
|
float c12x, c12y, c12z;
|
|
c12x = n1y * n2z - n1z * n2y;
|
|
c12y = n1z * n2x - n1x * n2z;
|
|
c12z = n1x * n2y - n1y * n2x;
|
|
float invDot = 1.0f / (n1x * c23x + n1y * c23y + n1z * c23z);
|
|
point.x = (-c23x * d1 - c31x * d2 - c12x * d3) * invDot;
|
|
point.y = (-c23y * d1 - c31y * d2 - c12y * d3) * invDot;
|
|
point.z = (-c23z * d1 - c31z * d2 - c12z * d3) * invDot;
|
|
return point;
|
|
}
|
|
|
|
/**
|
|
* Compute the eye/origin of the perspective frustum transformation defined by <code>this</code> matrix,
|
|
* which can be a projection matrix or a combined modelview-projection matrix, and store the result
|
|
* in the given <code>origin</code>.
|
|
* <p>
|
|
* Note that this method will only work using perspective projections obtained via one of the
|
|
* perspective methods, such as {@link #perspective(float, float, float, float) perspective()}
|
|
* or {@link #frustum(float, float, float, float, float, float) frustum()}.
|
|
* <p>
|
|
* Generally, this method computes the origin in the local frame of
|
|
* any coordinate system that existed before <code>this</code>
|
|
* transformation was applied to it in order to yield homogeneous clipping space.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>invert(new Matrix4f()).transformProject(0, 0, -1, 0, origin)</code>
|
|
* and in the case of an already available inverse of <code>this</code> matrix, the method {@link #perspectiveInvOrigin(Vector3f)}
|
|
* on the inverse of the matrix should be used instead.
|
|
* <p>
|
|
* Reference: <a href="http://geomalgorithms.com/a05-_intersect-1.html">http://geomalgorithms.com</a>
|
|
* <p>
|
|
* Reference: <a href="http://gamedevs.org/uploads/fast-extraction-viewing-frustum-planes-from-world-view-projection-matrix.pdf">
|
|
* Fast Extraction of Viewing Frustum Planes from the World-View-Projection Matrix</a>
|
|
*
|
|
* @see #perspectiveInvOrigin(Vector3f)
|
|
*
|
|
* @param origin
|
|
* will hold the origin of the coordinate system before applying <code>this</code>
|
|
* perspective projection transformation
|
|
* @return origin
|
|
*/
|
|
public Vector3f perspectiveOrigin(Vector3f origin) {
|
|
/*
|
|
* Simply compute the intersection point of the left, right and top frustum plane.
|
|
*/
|
|
float d1, d2, d3;
|
|
float n1x, n1y, n1z, n2x, n2y, n2z, n3x, n3y, n3z;
|
|
n1x = m03 + m00; n1y = m13 + m10; n1z = m23 + m20; d1 = m33 + m30; // left
|
|
n2x = m03 - m00; n2y = m13 - m10; n2z = m23 - m20; d2 = m33 - m30; // right
|
|
n3x = m03 - m01; n3y = m13 - m11; n3z = m23 - m21; d3 = m33 - m31; // top
|
|
float c23x, c23y, c23z;
|
|
c23x = n2y * n3z - n2z * n3y;
|
|
c23y = n2z * n3x - n2x * n3z;
|
|
c23z = n2x * n3y - n2y * n3x;
|
|
float c31x, c31y, c31z;
|
|
c31x = n3y * n1z - n3z * n1y;
|
|
c31y = n3z * n1x - n3x * n1z;
|
|
c31z = n3x * n1y - n3y * n1x;
|
|
float c12x, c12y, c12z;
|
|
c12x = n1y * n2z - n1z * n2y;
|
|
c12y = n1z * n2x - n1x * n2z;
|
|
c12z = n1x * n2y - n1y * n2x;
|
|
float invDot = 1.0f / (n1x * c23x + n1y * c23y + n1z * c23z);
|
|
origin.x = (-c23x * d1 - c31x * d2 - c12x * d3) * invDot;
|
|
origin.y = (-c23y * d1 - c31y * d2 - c12y * d3) * invDot;
|
|
origin.z = (-c23z * d1 - c31z * d2 - c12z * d3) * invDot;
|
|
return origin;
|
|
}
|
|
|
|
/**
|
|
* Compute the eye/origin of the inverse of the perspective frustum transformation defined by <code>this</code> matrix,
|
|
* which can be the inverse of a projection matrix or the inverse of a combined modelview-projection matrix, and store the result
|
|
* in the given <code>dest</code>.
|
|
* <p>
|
|
* Note that this method will only work using perspective projections obtained via one of the
|
|
* perspective methods, such as {@link #perspective(float, float, float, float) perspective()}
|
|
* or {@link #frustum(float, float, float, float, float, float) frustum()}.
|
|
* <p>
|
|
* If the inverse of the modelview-projection matrix is not available, then calling {@link #perspectiveOrigin(Vector3f)}
|
|
* on the original modelview-projection matrix is preferred.
|
|
*
|
|
* @see #perspectiveOrigin(Vector3f)
|
|
*
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Vector3f perspectiveInvOrigin(Vector3f dest) {
|
|
float invW = 1.0f / m23;
|
|
dest.x = m20 * invW;
|
|
dest.y = m21 * invW;
|
|
dest.z = m22 * invW;
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Return the vertical field-of-view angle in radians of this perspective transformation matrix.
|
|
* <p>
|
|
* Note that this method will only work using perspective projections obtained via one of the
|
|
* perspective methods, such as {@link #perspective(float, float, float, float) perspective()}
|
|
* or {@link #frustum(float, float, float, float, float, float) frustum()}.
|
|
* <p>
|
|
* For orthogonal transformations this method will return <code>0.0</code>.
|
|
* <p>
|
|
* Reference: <a href="http://gamedevs.org/uploads/fast-extraction-viewing-frustum-planes-from-world-view-projection-matrix.pdf">
|
|
* Fast Extraction of Viewing Frustum Planes from the World-View-Projection Matrix</a>
|
|
*
|
|
* @return the vertical field-of-view angle in radians
|
|
*/
|
|
public float perspectiveFov() {
|
|
/*
|
|
* Compute the angle between the bottom and top frustum plane normals.
|
|
*/
|
|
float n1x, n1y, n1z, n2x, n2y, n2z;
|
|
n1x = m03 + m01; n1y = m13 + m11; n1z = m23 + m21; // bottom
|
|
n2x = m01 - m03; n2y = m11 - m13; n2z = m21 - m23; // top
|
|
float n1len = Math.sqrt(n1x * n1x + n1y * n1y + n1z * n1z);
|
|
float n2len = Math.sqrt(n2x * n2x + n2y * n2y + n2z * n2z);
|
|
return Math.acos((n1x * n2x + n1y * n2y + n1z * n2z) / (n1len * n2len));
|
|
}
|
|
|
|
/**
|
|
* Extract the near clip plane distance from <code>this</code> perspective projection matrix.
|
|
* <p>
|
|
* This method only works if <code>this</code> is a perspective projection matrix, for example obtained via {@link #perspective(float, float, float, float)}.
|
|
*
|
|
* @return the near clip plane distance
|
|
*/
|
|
public float perspectiveNear() {
|
|
return m32 / (m23 + m22);
|
|
}
|
|
|
|
/**
|
|
* Extract the far clip plane distance from <code>this</code> perspective projection matrix.
|
|
* <p>
|
|
* This method only works if <code>this</code> is a perspective projection matrix, for example obtained via {@link #perspective(float, float, float, float)}.
|
|
*
|
|
* @return the far clip plane distance
|
|
*/
|
|
public float perspectiveFar() {
|
|
return m32 / (m22 - m23);
|
|
}
|
|
|
|
public Vector3f frustumRayDir(float x, float y, Vector3f dir) {
|
|
/*
|
|
* This method works by first obtaining the frustum plane normals,
|
|
* then building the cross product to obtain the corner rays,
|
|
* and finally bilinearly interpolating to obtain the desired direction.
|
|
* The code below uses a condense form of doing all this making use
|
|
* of some mathematical identities to simplify the overall expression.
|
|
*/
|
|
float a = m10 * m23, b = m13 * m21, c = m10 * m21, d = m11 * m23, e = m13 * m20, f = m11 * m20;
|
|
float g = m03 * m20, h = m01 * m23, i = m01 * m20, j = m03 * m21, k = m00 * m23, l = m00 * m21;
|
|
float m = m00 * m13, n = m03 * m11, o = m00 * m11, p = m01 * m13, q = m03 * m10, r = m01 * m10;
|
|
float m1x, m1y, m1z;
|
|
m1x = (d + e + f - a - b - c) * (1.0f - y) + (a - b - c + d - e + f) * y;
|
|
m1y = (j + k + l - g - h - i) * (1.0f - y) + (g - h - i + j - k + l) * y;
|
|
m1z = (p + q + r - m - n - o) * (1.0f - y) + (m - n - o + p - q + r) * y;
|
|
float m2x, m2y, m2z;
|
|
m2x = (b - c - d + e + f - a) * (1.0f - y) + (a + b - c - d - e + f) * y;
|
|
m2y = (h - i - j + k + l - g) * (1.0f - y) + (g + h - i - j - k + l) * y;
|
|
m2z = (n - o - p + q + r - m) * (1.0f - y) + (m + n - o - p - q + r) * y;
|
|
dir.x = m1x + (m2x - m1x) * x;
|
|
dir.y = m1y + (m2y - m1y) * x;
|
|
dir.z = m1z + (m2z - m1z) * x;
|
|
return dir.normalize(dir);
|
|
}
|
|
|
|
public Vector3f positiveZ(Vector3f dir) {
|
|
if ((properties & PROPERTY_ORTHONORMAL) != 0)
|
|
return normalizedPositiveZ(dir);
|
|
return positiveZGeneric(dir);
|
|
}
|
|
private Vector3f positiveZGeneric(Vector3f dir) {
|
|
return dir.set(m10 * m21 - m11 * m20, m20 * m01 - m21 * m00, m00 * m11 - m01 * m10).normalize();
|
|
}
|
|
|
|
public Vector3f normalizedPositiveZ(Vector3f dir) {
|
|
return dir.set(m02, m12, m22);
|
|
}
|
|
|
|
public Vector3f positiveX(Vector3f dir) {
|
|
if ((properties & PROPERTY_ORTHONORMAL) != 0)
|
|
return normalizedPositiveX(dir);
|
|
return positiveXGeneric(dir);
|
|
}
|
|
private Vector3f positiveXGeneric(Vector3f dir) {
|
|
return dir.set(m11 * m22 - m12 * m21, m02 * m21 - m01 * m22, m01 * m12 - m02 * m11).normalize();
|
|
}
|
|
|
|
public Vector3f normalizedPositiveX(Vector3f dir) {
|
|
return dir.set(m00, m10, m20);
|
|
}
|
|
|
|
public Vector3f positiveY(Vector3f dir) {
|
|
if ((properties & PROPERTY_ORTHONORMAL) != 0)
|
|
return normalizedPositiveY(dir);
|
|
return positiveYGeneric(dir);
|
|
}
|
|
private Vector3f positiveYGeneric(Vector3f dir) {
|
|
return dir.set(m12 * m20 - m10 * m22, m00 * m22 - m02 * m20, m02 * m10 - m00 * m12).normalize();
|
|
}
|
|
|
|
public Vector3f normalizedPositiveY(Vector3f dir) {
|
|
return dir.set(m01, m11, m21);
|
|
}
|
|
|
|
public Vector3f originAffine(Vector3f origin) {
|
|
float a = m00 * m11 - m01 * m10;
|
|
float b = m00 * m12 - m02 * m10;
|
|
float d = m01 * m12 - m02 * m11;
|
|
float g = m20 * m31 - m21 * m30;
|
|
float h = m20 * m32 - m22 * m30;
|
|
float j = m21 * m32 - m22 * m31;
|
|
return origin.set(-m10 * j + m11 * h - m12 * g, m00 * j - m01 * h + m02 * g, -m30 * d + m31 * b - m32 * a);
|
|
}
|
|
|
|
public Vector3f origin(Vector3f dest) {
|
|
if ((properties & PROPERTY_AFFINE) != 0)
|
|
return originAffine(dest);
|
|
return originGeneric(dest);
|
|
}
|
|
private Vector3f originGeneric(Vector3f dest) {
|
|
float a = m00 * m11 - m01 * m10;
|
|
float b = m00 * m12 - m02 * m10;
|
|
float c = m00 * m13 - m03 * m10;
|
|
float d = m01 * m12 - m02 * m11;
|
|
float e = m01 * m13 - m03 * m11;
|
|
float f = m02 * m13 - m03 * m12;
|
|
float g = m20 * m31 - m21 * m30;
|
|
float h = m20 * m32 - m22 * m30;
|
|
float i = m20 * m33 - m23 * m30;
|
|
float j = m21 * m32 - m22 * m31;
|
|
float k = m21 * m33 - m23 * m31;
|
|
float l = m22 * m33 - m23 * m32;
|
|
float det = a * l - b * k + c * j + d * i - e * h + f * g;
|
|
float invDet = 1.0f / det;
|
|
float nm30 = (-m10 * j + m11 * h - m12 * g) * invDet;
|
|
float nm31 = ( m00 * j - m01 * h + m02 * g) * invDet;
|
|
float nm32 = (-m30 * d + m31 * b - m32 * a) * invDet;
|
|
float nm33 = det / ( m20 * d - m21 * b + m22 * a);
|
|
return dest.set(nm30 * nm33, nm31 * nm33, nm32 * nm33);
|
|
}
|
|
|
|
/**
|
|
* Apply a projection transformation to this matrix that projects onto the plane specified via the general plane equation
|
|
* <code>x*a + y*b + z*c + d = 0</code> as if casting a shadow from a given light position/direction <code>light</code>.
|
|
* <p>
|
|
* If <code>light.w</code> is <code>0.0</code> the light is being treated as a directional light; if it is <code>1.0</code> it is a point light.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>S</code> the shadow matrix,
|
|
* then the new matrix will be <code>M * S</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * S * v</code>, the
|
|
* shadow projection will be applied first!
|
|
* <p>
|
|
* Reference: <a href="ftp://ftp.sgi.com/opengl/contrib/blythe/advanced99/notes/node192.html">ftp.sgi.com</a>
|
|
*
|
|
* @param light
|
|
* the light's vector
|
|
* @param a
|
|
* the x factor in the plane equation
|
|
* @param b
|
|
* the y factor in the plane equation
|
|
* @param c
|
|
* the z factor in the plane equation
|
|
* @param d
|
|
* the constant in the plane equation
|
|
* @return this
|
|
*/
|
|
public Matrix4f shadow(Vector4f light, float a, float b, float c, float d) {
|
|
return shadow(light.x, light.y, light.z, light.w, a, b, c, d, this);
|
|
}
|
|
|
|
public Matrix4f shadow(Vector4f light, float a, float b, float c, float d, Matrix4f dest) {
|
|
return shadow(light.x, light.y, light.z, light.w, a, b, c, d, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a projection transformation to this matrix that projects onto the plane specified via the general plane equation
|
|
* <code>x*a + y*b + z*c + d = 0</code> as if casting a shadow from a given light position/direction <code>(lightX, lightY, lightZ, lightW)</code>.
|
|
* <p>
|
|
* If <code>lightW</code> is <code>0.0</code> the light is being treated as a directional light; if it is <code>1.0</code> it is a point light.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>S</code> the shadow matrix,
|
|
* then the new matrix will be <code>M * S</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * S * v</code>, the
|
|
* shadow projection will be applied first!
|
|
* <p>
|
|
* Reference: <a href="ftp://ftp.sgi.com/opengl/contrib/blythe/advanced99/notes/node192.html">ftp.sgi.com</a>
|
|
*
|
|
* @param lightX
|
|
* the x-component of the light's vector
|
|
* @param lightY
|
|
* the y-component of the light's vector
|
|
* @param lightZ
|
|
* the z-component of the light's vector
|
|
* @param lightW
|
|
* the w-component of the light's vector
|
|
* @param a
|
|
* the x factor in the plane equation
|
|
* @param b
|
|
* the y factor in the plane equation
|
|
* @param c
|
|
* the z factor in the plane equation
|
|
* @param d
|
|
* the constant in the plane equation
|
|
* @return this
|
|
*/
|
|
public Matrix4f shadow(float lightX, float lightY, float lightZ, float lightW, float a, float b, float c, float d) {
|
|
return shadow(lightX, lightY, lightZ, lightW, a, b, c, d, this);
|
|
}
|
|
|
|
public Matrix4f shadow(float lightX, float lightY, float lightZ, float lightW, float a, float b, float c, float d, Matrix4f dest) {
|
|
// normalize plane
|
|
float invPlaneLen = Math.invsqrt(a*a + b*b + c*c);
|
|
float an = a * invPlaneLen;
|
|
float bn = b * invPlaneLen;
|
|
float cn = c * invPlaneLen;
|
|
float dn = d * invPlaneLen;
|
|
|
|
float dot = an * lightX + bn * lightY + cn * lightZ + dn * lightW;
|
|
|
|
// compute right matrix elements
|
|
float rm00 = dot - an * lightX;
|
|
float rm01 = -an * lightY;
|
|
float rm02 = -an * lightZ;
|
|
float rm03 = -an * lightW;
|
|
float rm10 = -bn * lightX;
|
|
float rm11 = dot - bn * lightY;
|
|
float rm12 = -bn * lightZ;
|
|
float rm13 = -bn * lightW;
|
|
float rm20 = -cn * lightX;
|
|
float rm21 = -cn * lightY;
|
|
float rm22 = dot - cn * lightZ;
|
|
float rm23 = -cn * lightW;
|
|
float rm30 = -dn * lightX;
|
|
float rm31 = -dn * lightY;
|
|
float rm32 = -dn * lightZ;
|
|
float rm33 = dot - dn * lightW;
|
|
|
|
// matrix multiplication
|
|
float nm00 = m00 * rm00 + m10 * rm01 + m20 * rm02 + m30 * rm03;
|
|
float nm01 = m01 * rm00 + m11 * rm01 + m21 * rm02 + m31 * rm03;
|
|
float nm02 = m02 * rm00 + m12 * rm01 + m22 * rm02 + m32 * rm03;
|
|
float nm03 = m03 * rm00 + m13 * rm01 + m23 * rm02 + m33 * rm03;
|
|
float nm10 = m00 * rm10 + m10 * rm11 + m20 * rm12 + m30 * rm13;
|
|
float nm11 = m01 * rm10 + m11 * rm11 + m21 * rm12 + m31 * rm13;
|
|
float nm12 = m02 * rm10 + m12 * rm11 + m22 * rm12 + m32 * rm13;
|
|
float nm13 = m03 * rm10 + m13 * rm11 + m23 * rm12 + m33 * rm13;
|
|
float nm20 = m00 * rm20 + m10 * rm21 + m20 * rm22 + m30 * rm23;
|
|
float nm21 = m01 * rm20 + m11 * rm21 + m21 * rm22 + m31 * rm23;
|
|
float nm22 = m02 * rm20 + m12 * rm21 + m22 * rm22 + m32 * rm23;
|
|
float nm23 = m03 * rm20 + m13 * rm21 + m23 * rm22 + m33 * rm23;
|
|
dest._m30(m00 * rm30 + m10 * rm31 + m20 * rm32 + m30 * rm33)
|
|
._m31(m01 * rm30 + m11 * rm31 + m21 * rm32 + m31 * rm33)
|
|
._m32(m02 * rm30 + m12 * rm31 + m22 * rm32 + m32 * rm33)
|
|
._m33(m03 * rm30 + m13 * rm31 + m23 * rm32 + m33 * rm33)
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
return dest;
|
|
}
|
|
|
|
public Matrix4f shadow(Vector4f light, Matrix4fc planeTransform, Matrix4f dest) {
|
|
// compute plane equation by transforming (y = 0)
|
|
float a = planeTransform.m10();
|
|
float b = planeTransform.m11();
|
|
float c = planeTransform.m12();
|
|
float d = -a * planeTransform.m30() - b * planeTransform.m31() - c * planeTransform.m32();
|
|
return shadow(light.x, light.y, light.z, light.w, a, b, c, d, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a projection transformation to this matrix that projects onto the plane with the general plane equation
|
|
* <code>y = 0</code> as if casting a shadow from a given light position/direction <code>light</code>.
|
|
* <p>
|
|
* Before the shadow projection is applied, the plane is transformed via the specified <code>planeTransformation</code>.
|
|
* <p>
|
|
* If <code>light.w</code> is <code>0.0</code> the light is being treated as a directional light; if it is <code>1.0</code> it is a point light.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>S</code> the shadow matrix,
|
|
* then the new matrix will be <code>M * S</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * S * v</code>, the
|
|
* shadow projection will be applied first!
|
|
*
|
|
* @param light
|
|
* the light's vector
|
|
* @param planeTransform
|
|
* the transformation to transform the implied plane <code>y = 0</code> before applying the projection
|
|
* @return this
|
|
*/
|
|
public Matrix4f shadow(Vector4f light, Matrix4f planeTransform) {
|
|
return shadow(light, planeTransform, this);
|
|
}
|
|
|
|
public Matrix4f shadow(float lightX, float lightY, float lightZ, float lightW, Matrix4fc planeTransform, Matrix4f dest) {
|
|
// compute plane equation by transforming (y = 0)
|
|
float a = planeTransform.m10();
|
|
float b = planeTransform.m11();
|
|
float c = planeTransform.m12();
|
|
float d = -a * planeTransform.m30() - b * planeTransform.m31() - c * planeTransform.m32();
|
|
return shadow(lightX, lightY, lightZ, lightW, a, b, c, d, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a projection transformation to this matrix that projects onto the plane with the general plane equation
|
|
* <code>y = 0</code> as if casting a shadow from a given light position/direction <code>(lightX, lightY, lightZ, lightW)</code>.
|
|
* <p>
|
|
* Before the shadow projection is applied, the plane is transformed via the specified <code>planeTransformation</code>.
|
|
* <p>
|
|
* If <code>lightW</code> is <code>0.0</code> the light is being treated as a directional light; if it is <code>1.0</code> it is a point light.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>S</code> the shadow matrix,
|
|
* then the new matrix will be <code>M * S</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * S * v</code>, the
|
|
* shadow projection will be applied first!
|
|
*
|
|
* @param lightX
|
|
* the x-component of the light vector
|
|
* @param lightY
|
|
* the y-component of the light vector
|
|
* @param lightZ
|
|
* the z-component of the light vector
|
|
* @param lightW
|
|
* the w-component of the light vector
|
|
* @param planeTransform
|
|
* the transformation to transform the implied plane <code>y = 0</code> before applying the projection
|
|
* @return this
|
|
*/
|
|
public Matrix4f shadow(float lightX, float lightY, float lightZ, float lightW, Matrix4f planeTransform) {
|
|
return shadow(lightX, lightY, lightZ, lightW, planeTransform, this);
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a cylindrical billboard transformation that rotates the local +Z axis of a given object with position <code>objPos</code> towards
|
|
* a target position at <code>targetPos</code> while constraining a cylindrical rotation around the given <code>up</code> vector.
|
|
* <p>
|
|
* This method can be used to create the complete model transformation for a given object, including the translation of the object to
|
|
* its position <code>objPos</code>.
|
|
*
|
|
* @param objPos
|
|
* the position of the object to rotate towards <code>targetPos</code>
|
|
* @param targetPos
|
|
* the position of the target (for example the camera) towards which to rotate the object
|
|
* @param up
|
|
* the rotation axis (must be {@link Vector3f#normalize() normalized})
|
|
* @return this
|
|
*/
|
|
public Matrix4f billboardCylindrical(Vector3fc objPos, Vector3fc targetPos, Vector3fc up) {
|
|
float dirX = targetPos.x() - objPos.x();
|
|
float dirY = targetPos.y() - objPos.y();
|
|
float dirZ = targetPos.z() - objPos.z();
|
|
// left = up x dir
|
|
float leftX = up.y() * dirZ - up.z() * dirY;
|
|
float leftY = up.z() * dirX - up.x() * dirZ;
|
|
float leftZ = up.x() * dirY - up.y() * dirX;
|
|
// normalize left
|
|
float invLeftLen = Math.invsqrt(leftX * leftX + leftY * leftY + leftZ * leftZ);
|
|
leftX *= invLeftLen;
|
|
leftY *= invLeftLen;
|
|
leftZ *= invLeftLen;
|
|
// recompute dir by constraining rotation around 'up'
|
|
// dir = left x up
|
|
dirX = leftY * up.z() - leftZ * up.y();
|
|
dirY = leftZ * up.x() - leftX * up.z();
|
|
dirZ = leftX * up.y() - leftY * up.x();
|
|
// normalize dir
|
|
float invDirLen = Math.invsqrt(dirX * dirX + dirY * dirY + dirZ * dirZ);
|
|
dirX *= invDirLen;
|
|
dirY *= invDirLen;
|
|
dirZ *= invDirLen;
|
|
// set matrix elements
|
|
this._m00(leftX)
|
|
._m01(leftY)
|
|
._m02(leftZ)
|
|
._m03(0.0f)
|
|
._m10(up.x())
|
|
._m11(up.y())
|
|
._m12(up.z())
|
|
._m13(0.0f)
|
|
._m20(dirX)
|
|
._m21(dirY)
|
|
._m22(dirZ)
|
|
._m23(0.0f)
|
|
._m30(objPos.x())
|
|
._m31(objPos.y())
|
|
._m32(objPos.z())
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a spherical billboard transformation that rotates the local +Z axis of a given object with position <code>objPos</code> towards
|
|
* a target position at <code>targetPos</code>.
|
|
* <p>
|
|
* This method can be used to create the complete model transformation for a given object, including the translation of the object to
|
|
* its position <code>objPos</code>.
|
|
* <p>
|
|
* If preserving an <i>up</i> vector is not necessary when rotating the +Z axis, then a shortest arc rotation can be obtained
|
|
* using {@link #billboardSpherical(Vector3fc, Vector3fc)}.
|
|
*
|
|
* @see #billboardSpherical(Vector3fc, Vector3fc)
|
|
*
|
|
* @param objPos
|
|
* the position of the object to rotate towards <code>targetPos</code>
|
|
* @param targetPos
|
|
* the position of the target (for example the camera) towards which to rotate the object
|
|
* @param up
|
|
* the up axis used to orient the object
|
|
* @return this
|
|
*/
|
|
public Matrix4f billboardSpherical(Vector3fc objPos, Vector3fc targetPos, Vector3fc up) {
|
|
float dirX = targetPos.x() - objPos.x();
|
|
float dirY = targetPos.y() - objPos.y();
|
|
float dirZ = targetPos.z() - objPos.z();
|
|
// normalize dir
|
|
float invDirLen = Math.invsqrt(dirX * dirX + dirY * dirY + dirZ * dirZ);
|
|
dirX *= invDirLen;
|
|
dirY *= invDirLen;
|
|
dirZ *= invDirLen;
|
|
// left = up x dir
|
|
float leftX = up.y() * dirZ - up.z() * dirY;
|
|
float leftY = up.z() * dirX - up.x() * dirZ;
|
|
float leftZ = up.x() * dirY - up.y() * dirX;
|
|
// normalize left
|
|
float invLeftLen = Math.invsqrt(leftX * leftX + leftY * leftY + leftZ * leftZ);
|
|
leftX *= invLeftLen;
|
|
leftY *= invLeftLen;
|
|
leftZ *= invLeftLen;
|
|
// up = dir x left
|
|
float upX = dirY * leftZ - dirZ * leftY;
|
|
float upY = dirZ * leftX - dirX * leftZ;
|
|
float upZ = dirX * leftY - dirY * leftX;
|
|
// set matrix elements
|
|
this._m00(leftX)
|
|
._m01(leftY)
|
|
._m02(leftZ)
|
|
._m03(0.0f)
|
|
._m10(upX)
|
|
._m11(upY)
|
|
._m12(upZ)
|
|
._m13(0.0f)
|
|
._m20(dirX)
|
|
._m21(dirY)
|
|
._m22(dirZ)
|
|
._m23(0.0f)
|
|
._m30(objPos.x())
|
|
._m31(objPos.y())
|
|
._m32(objPos.z())
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a spherical billboard transformation that rotates the local +Z axis of a given object with position <code>objPos</code> towards
|
|
* a target position at <code>targetPos</code> using a shortest arc rotation by not preserving any <i>up</i> vector of the object.
|
|
* <p>
|
|
* This method can be used to create the complete model transformation for a given object, including the translation of the object to
|
|
* its position <code>objPos</code>.
|
|
* <p>
|
|
* In order to specify an <i>up</i> vector which needs to be maintained when rotating the +Z axis of the object,
|
|
* use {@link #billboardSpherical(Vector3fc, Vector3fc, Vector3fc)}.
|
|
*
|
|
* @see #billboardSpherical(Vector3fc, Vector3fc, Vector3fc)
|
|
*
|
|
* @param objPos
|
|
* the position of the object to rotate towards <code>targetPos</code>
|
|
* @param targetPos
|
|
* the position of the target (for example the camera) towards which to rotate the object
|
|
* @return this
|
|
*/
|
|
public Matrix4f billboardSpherical(Vector3fc objPos, Vector3fc targetPos) {
|
|
float toDirX = targetPos.x() - objPos.x();
|
|
float toDirY = targetPos.y() - objPos.y();
|
|
float toDirZ = targetPos.z() - objPos.z();
|
|
float x = -toDirY;
|
|
float y = toDirX;
|
|
float w = Math.sqrt(toDirX * toDirX + toDirY * toDirY + toDirZ * toDirZ) + toDirZ;
|
|
float invNorm = Math.invsqrt(x * x + y * y + w * w);
|
|
x *= invNorm;
|
|
y *= invNorm;
|
|
w *= invNorm;
|
|
float q00 = (x + x) * x;
|
|
float q11 = (y + y) * y;
|
|
float q01 = (x + x) * y;
|
|
float q03 = (x + x) * w;
|
|
float q13 = (y + y) * w;
|
|
this._m00(1.0f - q11)
|
|
._m01(q01)
|
|
._m02(-q13)
|
|
._m03(0.0f)
|
|
._m10(q01)
|
|
._m11(1.0f - q00)
|
|
._m12(q03)
|
|
._m13(0.0f)
|
|
._m20(q13)
|
|
._m21(-q03)
|
|
._m22(1.0f - q11 - q00)
|
|
._m23(0.0f)
|
|
._m30(objPos.x())
|
|
._m31(objPos.y())
|
|
._m32(objPos.z())
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
return this;
|
|
}
|
|
|
|
public int hashCode() {
|
|
final int prime = 31;
|
|
int result = 1;
|
|
result = prime * result + Float.floatToIntBits(m00);
|
|
result = prime * result + Float.floatToIntBits(m01);
|
|
result = prime * result + Float.floatToIntBits(m02);
|
|
result = prime * result + Float.floatToIntBits(m03);
|
|
result = prime * result + Float.floatToIntBits(m10);
|
|
result = prime * result + Float.floatToIntBits(m11);
|
|
result = prime * result + Float.floatToIntBits(m12);
|
|
result = prime * result + Float.floatToIntBits(m13);
|
|
result = prime * result + Float.floatToIntBits(m20);
|
|
result = prime * result + Float.floatToIntBits(m21);
|
|
result = prime * result + Float.floatToIntBits(m22);
|
|
result = prime * result + Float.floatToIntBits(m23);
|
|
result = prime * result + Float.floatToIntBits(m30);
|
|
result = prime * result + Float.floatToIntBits(m31);
|
|
result = prime * result + Float.floatToIntBits(m32);
|
|
result = prime * result + Float.floatToIntBits(m33);
|
|
return result;
|
|
}
|
|
|
|
public boolean equals(Object obj) {
|
|
if (this == obj)
|
|
return true;
|
|
if (obj == null)
|
|
return false;
|
|
if (!(obj instanceof Matrix4f))
|
|
return false;
|
|
Matrix4fc other = (Matrix4fc) obj;
|
|
if (Float.floatToIntBits(m00) != Float.floatToIntBits(other.m00()))
|
|
return false;
|
|
if (Float.floatToIntBits(m01) != Float.floatToIntBits(other.m01()))
|
|
return false;
|
|
if (Float.floatToIntBits(m02) != Float.floatToIntBits(other.m02()))
|
|
return false;
|
|
if (Float.floatToIntBits(m03) != Float.floatToIntBits(other.m03()))
|
|
return false;
|
|
if (Float.floatToIntBits(m10) != Float.floatToIntBits(other.m10()))
|
|
return false;
|
|
if (Float.floatToIntBits(m11) != Float.floatToIntBits(other.m11()))
|
|
return false;
|
|
if (Float.floatToIntBits(m12) != Float.floatToIntBits(other.m12()))
|
|
return false;
|
|
if (Float.floatToIntBits(m13) != Float.floatToIntBits(other.m13()))
|
|
return false;
|
|
if (Float.floatToIntBits(m20) != Float.floatToIntBits(other.m20()))
|
|
return false;
|
|
if (Float.floatToIntBits(m21) != Float.floatToIntBits(other.m21()))
|
|
return false;
|
|
if (Float.floatToIntBits(m22) != Float.floatToIntBits(other.m22()))
|
|
return false;
|
|
if (Float.floatToIntBits(m23) != Float.floatToIntBits(other.m23()))
|
|
return false;
|
|
if (Float.floatToIntBits(m30) != Float.floatToIntBits(other.m30()))
|
|
return false;
|
|
if (Float.floatToIntBits(m31) != Float.floatToIntBits(other.m31()))
|
|
return false;
|
|
if (Float.floatToIntBits(m32) != Float.floatToIntBits(other.m32()))
|
|
return false;
|
|
if (Float.floatToIntBits(m33) != Float.floatToIntBits(other.m33()))
|
|
return false;
|
|
return true;
|
|
}
|
|
|
|
public boolean equals(Matrix4fc m, float delta) {
|
|
if (this == m)
|
|
return true;
|
|
if (m == null)
|
|
return false;
|
|
if (!(m instanceof Matrix4f))
|
|
return false;
|
|
if (!Runtime.equals(m00, m.m00(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m01, m.m01(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m02, m.m02(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m03, m.m03(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m10, m.m10(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m11, m.m11(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m12, m.m12(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m13, m.m13(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m20, m.m20(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m21, m.m21(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m22, m.m22(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m23, m.m23(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m30, m.m30(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m31, m.m31(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m32, m.m32(), delta))
|
|
return false;
|
|
if (!Runtime.equals(m33, m.m33(), delta))
|
|
return false;
|
|
return true;
|
|
}
|
|
|
|
public Matrix4f pick(float x, float y, float width, float height, int[] viewport, Matrix4f dest) {
|
|
float sx = viewport[2] / width;
|
|
float sy = viewport[3] / height;
|
|
float tx = (viewport[2] + 2.0f * (viewport[0] - x)) / width;
|
|
float ty = (viewport[3] + 2.0f * (viewport[1] - y)) / height;
|
|
dest._m30(m00 * tx + m10 * ty + m30)
|
|
._m31(m01 * tx + m11 * ty + m31)
|
|
._m32(m02 * tx + m12 * ty + m32)
|
|
._m33(m03 * tx + m13 * ty + m33)
|
|
._m00(m00 * sx)
|
|
._m01(m01 * sx)
|
|
._m02(m02 * sx)
|
|
._m03(m03 * sx)
|
|
._m10(m10 * sy)
|
|
._m11(m11 * sy)
|
|
._m12(m12 * sy)
|
|
._m13(m13 * sy)
|
|
._properties(0);
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply a picking transformation to this matrix using the given window coordinates <code>(x, y)</code> as the pick center
|
|
* and the given <code>(width, height)</code> as the size of the picking region in window coordinates.
|
|
*
|
|
* @param x
|
|
* the x coordinate of the picking region center in window coordinates
|
|
* @param y
|
|
* the y coordinate of the picking region center in window coordinates
|
|
* @param width
|
|
* the width of the picking region in window coordinates
|
|
* @param height
|
|
* the height of the picking region in window coordinates
|
|
* @param viewport
|
|
* the viewport described by <code>[x, y, width, height]</code>
|
|
* @return this
|
|
*/
|
|
public Matrix4f pick(float x, float y, float width, float height, int[] viewport) {
|
|
return pick(x, y, width, height, viewport, this);
|
|
}
|
|
|
|
public boolean isAffine() {
|
|
return m03 == 0.0f && m13 == 0.0f && m23 == 0.0f && m33 == 1.0f;
|
|
}
|
|
|
|
/**
|
|
* Exchange the values of <code>this</code> matrix with the given <code>other</code> matrix.
|
|
*
|
|
* @param other
|
|
* the other matrix to exchange the values with
|
|
* @return this
|
|
*/
|
|
public Matrix4f swap(Matrix4f other) {
|
|
MemUtil.INSTANCE.swap(this, other);
|
|
int props = properties;
|
|
this.properties = other.properties();
|
|
other.properties = props;
|
|
return this;
|
|
}
|
|
|
|
public Matrix4f arcball(float radius, float centerX, float centerY, float centerZ, float angleX, float angleY, Matrix4f dest) {
|
|
float m30 = m20 * -radius + this.m30;
|
|
float m31 = m21 * -radius + this.m31;
|
|
float m32 = m22 * -radius + this.m32;
|
|
float m33 = m23 * -radius + this.m33;
|
|
float sin = Math.sin(angleX);
|
|
float cos = Math.cosFromSin(sin, angleX);
|
|
float nm10 = m10 * cos + m20 * sin;
|
|
float nm11 = m11 * cos + m21 * sin;
|
|
float nm12 = m12 * cos + m22 * sin;
|
|
float nm13 = m13 * cos + m23 * sin;
|
|
float m20 = this.m20 * cos - m10 * sin;
|
|
float m21 = this.m21 * cos - m11 * sin;
|
|
float m22 = this.m22 * cos - m12 * sin;
|
|
float m23 = this.m23 * cos - m13 * sin;
|
|
sin = Math.sin(angleY);
|
|
cos = Math.cosFromSin(sin, angleY);
|
|
float nm00 = m00 * cos - m20 * sin;
|
|
float nm01 = m01 * cos - m21 * sin;
|
|
float nm02 = m02 * cos - m22 * sin;
|
|
float nm03 = m03 * cos - m23 * sin;
|
|
float nm20 = m00 * sin + m20 * cos;
|
|
float nm21 = m01 * sin + m21 * cos;
|
|
float nm22 = m02 * sin + m22 * cos;
|
|
float nm23 = m03 * sin + m23 * cos;
|
|
dest._m30(-nm00 * centerX - nm10 * centerY - nm20 * centerZ + m30)
|
|
._m31(-nm01 * centerX - nm11 * centerY - nm21 * centerZ + m31)
|
|
._m32(-nm02 * centerX - nm12 * centerY - nm22 * centerZ + m32)
|
|
._m33(-nm03 * centerX - nm13 * centerY - nm23 * centerZ + m33)
|
|
._m20(nm20)
|
|
._m21(nm21)
|
|
._m22(nm22)
|
|
._m23(nm23)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
return dest;
|
|
}
|
|
|
|
public Matrix4f arcball(float radius, Vector3fc center, float angleX, float angleY, Matrix4f dest) {
|
|
return arcball(radius, center.x(), center.y(), center.z(), angleX, angleY, dest);
|
|
}
|
|
|
|
/**
|
|
* Apply an arcball view transformation to this matrix with the given <code>radius</code> and center <code>(centerX, centerY, centerZ)</code>
|
|
* position of the arcball and the specified X and Y rotation angles.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translate(0, 0, -radius).rotateX(angleX).rotateY(angleY).translate(-centerX, -centerY, -centerZ)</code>
|
|
*
|
|
* @param radius
|
|
* the arcball radius
|
|
* @param centerX
|
|
* the x coordinate of the center position of the arcball
|
|
* @param centerY
|
|
* the y coordinate of the center position of the arcball
|
|
* @param centerZ
|
|
* the z coordinate of the center position of the arcball
|
|
* @param angleX
|
|
* the rotation angle around the X axis in radians
|
|
* @param angleY
|
|
* the rotation angle around the Y axis in radians
|
|
* @return this
|
|
*/
|
|
public Matrix4f arcball(float radius, float centerX, float centerY, float centerZ, float angleX, float angleY) {
|
|
return arcball(radius, centerX, centerY, centerZ, angleX, angleY, this);
|
|
}
|
|
|
|
/**
|
|
* Apply an arcball view transformation to this matrix with the given <code>radius</code> and <code>center</code>
|
|
* position of the arcball and the specified X and Y rotation angles.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translate(0, 0, -radius).rotateX(angleX).rotateY(angleY).translate(-center.x, -center.y, -center.z)</code>
|
|
*
|
|
* @param radius
|
|
* the arcball radius
|
|
* @param center
|
|
* the center position of the arcball
|
|
* @param angleX
|
|
* the rotation angle around the X axis in radians
|
|
* @param angleY
|
|
* the rotation angle around the Y axis in radians
|
|
* @return this
|
|
*/
|
|
public Matrix4f arcball(float radius, Vector3fc center, float angleX, float angleY) {
|
|
return arcball(radius, center.x(), center.y(), center.z(), angleX, angleY, this);
|
|
}
|
|
|
|
/**
|
|
* Compute the axis-aligned bounding box of the frustum described by <code>this</code> matrix and store the minimum corner
|
|
* coordinates in the given <code>min</code> and the maximum corner coordinates in the given <code>max</code> vector.
|
|
* <p>
|
|
* The matrix <code>this</code> is assumed to be the {@link #invert() inverse} of the origial view-projection matrix
|
|
* for which to compute the axis-aligned bounding box in world-space.
|
|
* <p>
|
|
* The axis-aligned bounding box of the unit frustum is <code>(-1, -1, -1)</code>, <code>(1, 1, 1)</code>.
|
|
*
|
|
* @param min
|
|
* will hold the minimum corner coordinates of the axis-aligned bounding box
|
|
* @param max
|
|
* will hold the maximum corner coordinates of the axis-aligned bounding box
|
|
* @return this
|
|
*/
|
|
public Matrix4f frustumAabb(Vector3f min, Vector3f max) {
|
|
float minX = Float.POSITIVE_INFINITY;
|
|
float minY = Float.POSITIVE_INFINITY;
|
|
float minZ = Float.POSITIVE_INFINITY;
|
|
float maxX = Float.NEGATIVE_INFINITY;
|
|
float maxY = Float.NEGATIVE_INFINITY;
|
|
float maxZ = Float.NEGATIVE_INFINITY;
|
|
for (int t = 0; t < 8; t++) {
|
|
float x = ((t & 1) << 1) - 1.0f;
|
|
float y = (((t >>> 1) & 1) << 1) - 1.0f;
|
|
float z = (((t >>> 2) & 1) << 1) - 1.0f;
|
|
float invW = 1.0f / (m03 * x + m13 * y + m23 * z + m33);
|
|
float nx = (m00 * x + m10 * y + m20 * z + m30) * invW;
|
|
float ny = (m01 * x + m11 * y + m21 * z + m31) * invW;
|
|
float nz = (m02 * x + m12 * y + m22 * z + m32) * invW;
|
|
minX = minX < nx ? minX : nx;
|
|
minY = minY < ny ? minY : ny;
|
|
minZ = minZ < nz ? minZ : nz;
|
|
maxX = maxX > nx ? maxX : nx;
|
|
maxY = maxY > ny ? maxY : ny;
|
|
maxZ = maxZ > nz ? maxZ : nz;
|
|
}
|
|
min.x = minX;
|
|
min.y = minY;
|
|
min.z = minZ;
|
|
max.x = maxX;
|
|
max.y = maxY;
|
|
max.z = maxZ;
|
|
return this;
|
|
}
|
|
|
|
public Matrix4f projectedGridRange(Matrix4fc projector, float sLower, float sUpper, Matrix4f dest) {
|
|
// Compute intersection with frustum edges and plane
|
|
float minX = Float.POSITIVE_INFINITY, minY = Float.POSITIVE_INFINITY;
|
|
float maxX = Float.NEGATIVE_INFINITY, maxY = Float.NEGATIVE_INFINITY;
|
|
boolean intersection = false;
|
|
for (int t = 0; t < 3 * 4; t++) {
|
|
float c0X, c0Y, c0Z;
|
|
float c1X, c1Y, c1Z;
|
|
if (t < 4) {
|
|
// all x edges
|
|
c0X = -1; c1X = +1;
|
|
c0Y = c1Y = ((t & 1) << 1) - 1.0f;
|
|
c0Z = c1Z = (((t >>> 1) & 1) << 1) - 1.0f;
|
|
} else if (t < 8) {
|
|
// all y edges
|
|
c0Y = -1; c1Y = +1;
|
|
c0X = c1X = ((t & 1) << 1) - 1.0f;
|
|
c0Z = c1Z = (((t >>> 1) & 1) << 1) - 1.0f;
|
|
} else {
|
|
// all z edges
|
|
c0Z = -1; c1Z = +1;
|
|
c0X = c1X = ((t & 1) << 1) - 1.0f;
|
|
c0Y = c1Y = (((t >>> 1) & 1) << 1) - 1.0f;
|
|
}
|
|
// unproject corners
|
|
float invW = 1.0f / (m03 * c0X + m13 * c0Y + m23 * c0Z + m33);
|
|
float p0x = (m00 * c0X + m10 * c0Y + m20 * c0Z + m30) * invW;
|
|
float p0y = (m01 * c0X + m11 * c0Y + m21 * c0Z + m31) * invW;
|
|
float p0z = (m02 * c0X + m12 * c0Y + m22 * c0Z + m32) * invW;
|
|
invW = 1.0f / (m03 * c1X + m13 * c1Y + m23 * c1Z + m33);
|
|
float p1x = (m00 * c1X + m10 * c1Y + m20 * c1Z + m30) * invW;
|
|
float p1y = (m01 * c1X + m11 * c1Y + m21 * c1Z + m31) * invW;
|
|
float p1z = (m02 * c1X + m12 * c1Y + m22 * c1Z + m32) * invW;
|
|
float dirX = p1x - p0x;
|
|
float dirY = p1y - p0y;
|
|
float dirZ = p1z - p0z;
|
|
float invDenom = 1.0f / dirY;
|
|
// test for intersection
|
|
for (int s = 0; s < 2; s++) {
|
|
float isectT = -(p0y + (s == 0 ? sLower : sUpper)) * invDenom;
|
|
if (isectT >= 0.0f && isectT <= 1.0f) {
|
|
intersection = true;
|
|
// project with projector matrix
|
|
float ix = p0x + isectT * dirX;
|
|
float iz = p0z + isectT * dirZ;
|
|
invW = 1.0f / (projector.m03() * ix + projector.m23() * iz + projector.m33());
|
|
float px = (projector.m00() * ix + projector.m20() * iz + projector.m30()) * invW;
|
|
float py = (projector.m01() * ix + projector.m21() * iz + projector.m31()) * invW;
|
|
minX = minX < px ? minX : px;
|
|
minY = minY < py ? minY : py;
|
|
maxX = maxX > px ? maxX : px;
|
|
maxY = maxY > py ? maxY : py;
|
|
}
|
|
}
|
|
}
|
|
if (!intersection)
|
|
return null; // <- projected grid is not visible
|
|
dest.set(maxX - minX, 0, 0, 0, 0, maxY - minY, 0, 0, 0, 0, 1, 0, minX, minY, 0, 1);
|
|
dest._properties(PROPERTY_AFFINE);
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Change the near and far clip plane distances of <code>this</code> perspective frustum transformation matrix
|
|
* and store the result in <code>dest</code>.
|
|
* <p>
|
|
* This method only works if <code>this</code> is a perspective projection frustum transformation, for example obtained
|
|
* via {@link #perspective(float, float, float, float) perspective()} or {@link #frustum(float, float, float, float, float, float) frustum()}.
|
|
*
|
|
* @see #perspective(float, float, float, float)
|
|
* @see #frustum(float, float, float, float, float, float)
|
|
*
|
|
* @param near
|
|
* the new near clip plane distance
|
|
* @param far
|
|
* the new far clip plane distance
|
|
* @param dest
|
|
* will hold the resulting matrix
|
|
* @return dest
|
|
*/
|
|
public Matrix4f perspectiveFrustumSlice(float near, float far, Matrix4f dest) {
|
|
float invOldNear = (m23 + m22) / m32;
|
|
float invNearFar = 1.0f / (near - far);
|
|
dest._m00(m00 * invOldNear * near)
|
|
._m01(m01)
|
|
._m02(m02)
|
|
._m03(m03)
|
|
._m10(m10)
|
|
._m11(m11 * invOldNear * near)
|
|
._m12(m12)
|
|
._m13(m13)
|
|
._m20(m20)
|
|
._m21(m21)
|
|
._m22((far + near) * invNearFar)
|
|
._m23(m23)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32((far + far) * near * invNearFar)
|
|
._m33(m33)
|
|
._properties(properties & ~(PROPERTY_IDENTITY | PROPERTY_TRANSLATION | PROPERTY_ORTHONORMAL));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Build an ortographic projection transformation that fits the view-projection transformation represented by <code>this</code>
|
|
* into the given affine <code>view</code> transformation.
|
|
* <p>
|
|
* The transformation represented by <code>this</code> must be given as the {@link #invert() inverse} of a typical combined camera view-projection
|
|
* transformation, whose projection can be either orthographic or perspective.
|
|
* <p>
|
|
* The <code>view</code> must be an {@link #isAffine() affine} transformation which in the application of Cascaded Shadow Maps is usually the light view transformation.
|
|
* It be obtained via any affine transformation or for example via {@link #lookAt(float, float, float, float, float, float, float, float, float) lookAt()}.
|
|
* <p>
|
|
* Reference: <a href="http://developer.download.nvidia.com/SDK/10.5/opengl/screenshots/samples/cascaded_shadow_maps.html">OpenGL SDK - Cascaded Shadow Maps</a>
|
|
*
|
|
* @param view
|
|
* the view transformation to build a corresponding orthographic projection to fit the frustum of <code>this</code>
|
|
* @param dest
|
|
* will hold the crop projection transformation
|
|
* @return dest
|
|
*/
|
|
public Matrix4f orthoCrop(Matrix4fc view, Matrix4f dest) {
|
|
// determine min/max world z and min/max orthographically view-projected x/y
|
|
float minX = Float.POSITIVE_INFINITY, maxX = Float.NEGATIVE_INFINITY;
|
|
float minY = Float.POSITIVE_INFINITY, maxY = Float.NEGATIVE_INFINITY;
|
|
float minZ = Float.POSITIVE_INFINITY, maxZ = Float.NEGATIVE_INFINITY;
|
|
for (int t = 0; t < 8; t++) {
|
|
float x = ((t & 1) << 1) - 1.0f;
|
|
float y = (((t >>> 1) & 1) << 1) - 1.0f;
|
|
float z = (((t >>> 2) & 1) << 1) - 1.0f;
|
|
float invW = 1.0f / (m03 * x + m13 * y + m23 * z + m33);
|
|
float wx = (m00 * x + m10 * y + m20 * z + m30) * invW;
|
|
float wy = (m01 * x + m11 * y + m21 * z + m31) * invW;
|
|
float wz = (m02 * x + m12 * y + m22 * z + m32) * invW;
|
|
invW = 1.0f / (view.m03() * wx + view.m13() * wy + view.m23() * wz + view.m33());
|
|
float vx = view.m00() * wx + view.m10() * wy + view.m20() * wz + view.m30();
|
|
float vy = view.m01() * wx + view.m11() * wy + view.m21() * wz + view.m31();
|
|
float vz = (view.m02() * wx + view.m12() * wy + view.m22() * wz + view.m32()) * invW;
|
|
minX = minX < vx ? minX : vx;
|
|
maxX = maxX > vx ? maxX : vx;
|
|
minY = minY < vy ? minY : vy;
|
|
maxY = maxY > vy ? maxY : vy;
|
|
minZ = minZ < vz ? minZ : vz;
|
|
maxZ = maxZ > vz ? maxZ : vz;
|
|
}
|
|
// build crop projection matrix to fit 'this' frustum into view
|
|
return dest.setOrtho(minX, maxX, minY, maxY, -maxZ, -minZ);
|
|
}
|
|
|
|
/**
|
|
* Set <code>this</code> matrix to a perspective transformation that maps the trapezoid spanned by the four corner coordinates
|
|
* <code>(p0x, p0y)</code>, <code>(p1x, p1y)</code>, <code>(p2x, p2y)</code> and <code>(p3x, p3y)</code> to the unit square <code>[(-1, -1)..(+1, +1)]</code>.
|
|
* <p>
|
|
* The corner coordinates are given in counter-clockwise order starting from the <i>left</i> corner on the smaller parallel side of the trapezoid
|
|
* seen when looking at the trapezoid oriented with its shorter parallel edge at the bottom and its longer parallel edge at the top.
|
|
* <p>
|
|
* Reference: <a href="http://www.comp.nus.edu.sg/~tants/tsm/TSM_recipe.html">Trapezoidal Shadow Maps (TSM) - Recipe</a>
|
|
*
|
|
* @param p0x
|
|
* the x coordinate of the left corner at the shorter edge of the trapezoid
|
|
* @param p0y
|
|
* the y coordinate of the left corner at the shorter edge of the trapezoid
|
|
* @param p1x
|
|
* the x coordinate of the right corner at the shorter edge of the trapezoid
|
|
* @param p1y
|
|
* the y coordinate of the right corner at the shorter edge of the trapezoid
|
|
* @param p2x
|
|
* the x coordinate of the right corner at the longer edge of the trapezoid
|
|
* @param p2y
|
|
* the y coordinate of the right corner at the longer edge of the trapezoid
|
|
* @param p3x
|
|
* the x coordinate of the left corner at the longer edge of the trapezoid
|
|
* @param p3y
|
|
* the y coordinate of the left corner at the longer edge of the trapezoid
|
|
* @return this
|
|
*/
|
|
public Matrix4f trapezoidCrop(float p0x, float p0y, float p1x, float p1y, float p2x, float p2y, float p3x, float p3y) {
|
|
float aX = p1y - p0y, aY = p0x - p1x;
|
|
float nm00 = aY;
|
|
float nm10 = -aX;
|
|
float nm30 = aX * p0y - aY * p0x;
|
|
float nm01 = aX;
|
|
float nm11 = aY;
|
|
float nm31 = -(aX * p0x + aY * p0y);
|
|
float c3x = nm00 * p3x + nm10 * p3y + nm30;
|
|
float c3y = nm01 * p3x + nm11 * p3y + nm31;
|
|
float s = -c3x / c3y;
|
|
nm00 += s * nm01;
|
|
nm10 += s * nm11;
|
|
nm30 += s * nm31;
|
|
float d1x = nm00 * p1x + nm10 * p1y + nm30;
|
|
float d2x = nm00 * p2x + nm10 * p2y + nm30;
|
|
float d = d1x * c3y / (d2x - d1x);
|
|
nm31 += d;
|
|
float sx = 2.0f / d2x;
|
|
float sy = 1.0f / (c3y + d);
|
|
float u = (sy + sy) * d / (1.0f - sy * d);
|
|
float m03 = nm01 * sy;
|
|
float m13 = nm11 * sy;
|
|
float m33 = nm31 * sy;
|
|
nm01 = (u + 1.0f) * m03;
|
|
nm11 = (u + 1.0f) * m13;
|
|
nm31 = (u + 1.0f) * m33 - u;
|
|
nm00 = sx * nm00 - m03;
|
|
nm10 = sx * nm10 - m13;
|
|
nm30 = sx * nm30 - m33;
|
|
set(nm00, nm01, 0, m03,
|
|
nm10, nm11, 0, m13,
|
|
0, 0, 1, 0,
|
|
nm30, nm31, 0, m33);
|
|
_properties(0);
|
|
return this;
|
|
}
|
|
|
|
public Matrix4f transformAab(float minX, float minY, float minZ, float maxX, float maxY, float maxZ, Vector3f outMin, Vector3f outMax) {
|
|
float xax = m00 * minX, xay = m01 * minX, xaz = m02 * minX;
|
|
float xbx = m00 * maxX, xby = m01 * maxX, xbz = m02 * maxX;
|
|
float yax = m10 * minY, yay = m11 * minY, yaz = m12 * minY;
|
|
float ybx = m10 * maxY, yby = m11 * maxY, ybz = m12 * maxY;
|
|
float zax = m20 * minZ, zay = m21 * minZ, zaz = m22 * minZ;
|
|
float zbx = m20 * maxZ, zby = m21 * maxZ, zbz = m22 * maxZ;
|
|
float xminx, xminy, xminz, yminx, yminy, yminz, zminx, zminy, zminz;
|
|
float xmaxx, xmaxy, xmaxz, ymaxx, ymaxy, ymaxz, zmaxx, zmaxy, zmaxz;
|
|
if (xax < xbx) {
|
|
xminx = xax;
|
|
xmaxx = xbx;
|
|
} else {
|
|
xminx = xbx;
|
|
xmaxx = xax;
|
|
}
|
|
if (xay < xby) {
|
|
xminy = xay;
|
|
xmaxy = xby;
|
|
} else {
|
|
xminy = xby;
|
|
xmaxy = xay;
|
|
}
|
|
if (xaz < xbz) {
|
|
xminz = xaz;
|
|
xmaxz = xbz;
|
|
} else {
|
|
xminz = xbz;
|
|
xmaxz = xaz;
|
|
}
|
|
if (yax < ybx) {
|
|
yminx = yax;
|
|
ymaxx = ybx;
|
|
} else {
|
|
yminx = ybx;
|
|
ymaxx = yax;
|
|
}
|
|
if (yay < yby) {
|
|
yminy = yay;
|
|
ymaxy = yby;
|
|
} else {
|
|
yminy = yby;
|
|
ymaxy = yay;
|
|
}
|
|
if (yaz < ybz) {
|
|
yminz = yaz;
|
|
ymaxz = ybz;
|
|
} else {
|
|
yminz = ybz;
|
|
ymaxz = yaz;
|
|
}
|
|
if (zax < zbx) {
|
|
zminx = zax;
|
|
zmaxx = zbx;
|
|
} else {
|
|
zminx = zbx;
|
|
zmaxx = zax;
|
|
}
|
|
if (zay < zby) {
|
|
zminy = zay;
|
|
zmaxy = zby;
|
|
} else {
|
|
zminy = zby;
|
|
zmaxy = zay;
|
|
}
|
|
if (zaz < zbz) {
|
|
zminz = zaz;
|
|
zmaxz = zbz;
|
|
} else {
|
|
zminz = zbz;
|
|
zmaxz = zaz;
|
|
}
|
|
outMin.x = xminx + yminx + zminx + m30;
|
|
outMin.y = xminy + yminy + zminy + m31;
|
|
outMin.z = xminz + yminz + zminz + m32;
|
|
outMax.x = xmaxx + ymaxx + zmaxx + m30;
|
|
outMax.y = xmaxy + ymaxy + zmaxy + m31;
|
|
outMax.z = xmaxz + ymaxz + zmaxz + m32;
|
|
return this;
|
|
}
|
|
|
|
public Matrix4f transformAab(Vector3fc min, Vector3fc max, Vector3f outMin, Vector3f outMax) {
|
|
return transformAab(min.x(), min.y(), min.z(), max.x(), max.y(), max.z(), outMin, outMax);
|
|
}
|
|
|
|
/**
|
|
* Linearly interpolate <code>this</code> and <code>other</code> using the given interpolation factor <code>t</code>
|
|
* and store the result in <code>this</code>.
|
|
* <p>
|
|
* If <code>t</code> is <code>0.0</code> then the result is <code>this</code>. If the interpolation factor is <code>1.0</code>
|
|
* then the result is <code>other</code>.
|
|
*
|
|
* @param other
|
|
* the other matrix
|
|
* @param t
|
|
* the interpolation factor between 0.0 and 1.0
|
|
* @return this
|
|
*/
|
|
public Matrix4f lerp(Matrix4fc other, float t) {
|
|
return lerp(other, t, this);
|
|
}
|
|
|
|
public Matrix4f lerp(Matrix4fc other, float t, Matrix4f dest) {
|
|
dest._m00(Math.fma(other.m00() - m00, t, m00))
|
|
._m01(Math.fma(other.m01() - m01, t, m01))
|
|
._m02(Math.fma(other.m02() - m02, t, m02))
|
|
._m03(Math.fma(other.m03() - m03, t, m03))
|
|
._m10(Math.fma(other.m10() - m10, t, m10))
|
|
._m11(Math.fma(other.m11() - m11, t, m11))
|
|
._m12(Math.fma(other.m12() - m12, t, m12))
|
|
._m13(Math.fma(other.m13() - m13, t, m13))
|
|
._m20(Math.fma(other.m20() - m20, t, m20))
|
|
._m21(Math.fma(other.m21() - m21, t, m21))
|
|
._m22(Math.fma(other.m22() - m22, t, m22))
|
|
._m23(Math.fma(other.m23() - m23, t, m23))
|
|
._m30(Math.fma(other.m30() - m30, t, m30))
|
|
._m31(Math.fma(other.m31() - m31, t, m31))
|
|
._m32(Math.fma(other.m32() - m32, t, m32))
|
|
._m33(Math.fma(other.m33() - m33, t, m33))
|
|
._properties(properties & other.properties());
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Apply a model transformation to this matrix for a right-handed coordinate system,
|
|
* that aligns the local <code>+Z</code> axis with <code>dir</code>
|
|
* and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookat matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>,
|
|
* the lookat transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without post-multiplying it,
|
|
* use {@link #rotationTowards(Vector3fc, Vector3fc) rotationTowards()}.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>mulAffine(new Matrix4f().lookAt(new Vector3f(), new Vector3f(dir).negate(), up).invertAffine(), dest)</code>
|
|
*
|
|
* @see #rotateTowards(float, float, float, float, float, float, Matrix4f)
|
|
* @see #rotationTowards(Vector3fc, Vector3fc)
|
|
*
|
|
* @param dir
|
|
* the direction to rotate towards
|
|
* @param up
|
|
* the up vector
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotateTowards(Vector3fc dir, Vector3fc up, Matrix4f dest) {
|
|
return rotateTowards(dir.x(), dir.y(), dir.z(), up.x(), up.y(), up.z(), dest);
|
|
}
|
|
|
|
/**
|
|
* Apply a model transformation to this matrix for a right-handed coordinate system,
|
|
* that aligns the local <code>+Z</code> axis with <code>dir</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookat matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>,
|
|
* the lookat transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without post-multiplying it,
|
|
* use {@link #rotationTowards(Vector3fc, Vector3fc) rotationTowards()}.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>mulAffine(new Matrix4f().lookAt(new Vector3f(), new Vector3f(dir).negate(), up).invertAffine())</code>
|
|
*
|
|
* @see #rotateTowards(float, float, float, float, float, float)
|
|
* @see #rotationTowards(Vector3fc, Vector3fc)
|
|
*
|
|
* @param dir
|
|
* the direction to orient towards
|
|
* @param up
|
|
* the up vector
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateTowards(Vector3fc dir, Vector3fc up) {
|
|
return rotateTowards(dir.x(), dir.y(), dir.z(), up.x(), up.y(), up.z(), this);
|
|
}
|
|
|
|
/**
|
|
* Apply a model transformation to this matrix for a right-handed coordinate system,
|
|
* that aligns the local <code>+Z</code> axis with <code>(dirX, dirY, dirZ)</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookat matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>,
|
|
* the lookat transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without post-multiplying it,
|
|
* use {@link #rotationTowards(float, float, float, float, float, float) rotationTowards()}.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>mulAffine(new Matrix4f().lookAt(0, 0, 0, -dirX, -dirY, -dirZ, upX, upY, upZ).invertAffine())</code>
|
|
*
|
|
* @see #rotateTowards(Vector3fc, Vector3fc)
|
|
* @see #rotationTowards(float, float, float, float, float, float)
|
|
*
|
|
* @param dirX
|
|
* the x-coordinate of the direction to rotate towards
|
|
* @param dirY
|
|
* the y-coordinate of the direction to rotate towards
|
|
* @param dirZ
|
|
* the z-coordinate of the direction to rotate towards
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotateTowards(float dirX, float dirY, float dirZ, float upX, float upY, float upZ) {
|
|
return rotateTowards(dirX, dirY, dirZ, upX, upY, upZ, this);
|
|
}
|
|
|
|
/**
|
|
* Apply a model transformation to this matrix for a right-handed coordinate system,
|
|
* that aligns the local <code>+Z</code> axis with <code>(dirX, dirY, dirZ)</code>
|
|
* and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>L</code> the lookat matrix,
|
|
* then the new matrix will be <code>M * L</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * L * v</code>,
|
|
* the lookat transformation will be applied first!
|
|
* <p>
|
|
* In order to set the matrix to a rotation transformation without post-multiplying it,
|
|
* use {@link #rotationTowards(float, float, float, float, float, float) rotationTowards()}.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>mulAffine(new Matrix4f().lookAt(0, 0, 0, -dirX, -dirY, -dirZ, upX, upY, upZ).invertAffine(), dest)</code>
|
|
*
|
|
* @see #rotateTowards(Vector3fc, Vector3fc)
|
|
* @see #rotationTowards(float, float, float, float, float, float)
|
|
*
|
|
* @param dirX
|
|
* the x-coordinate of the direction to rotate towards
|
|
* @param dirY
|
|
* the y-coordinate of the direction to rotate towards
|
|
* @param dirZ
|
|
* the z-coordinate of the direction to rotate towards
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f rotateTowards(float dirX, float dirY, float dirZ, float upX, float upY, float upZ, Matrix4f dest) {
|
|
// Normalize direction
|
|
float invDirLength = Math.invsqrt(dirX * dirX + dirY * dirY + dirZ * dirZ);
|
|
float ndirX = dirX * invDirLength;
|
|
float ndirY = dirY * invDirLength;
|
|
float ndirZ = dirZ * invDirLength;
|
|
// left = up x direction
|
|
float leftX, leftY, leftZ;
|
|
leftX = upY * ndirZ - upZ * ndirY;
|
|
leftY = upZ * ndirX - upX * ndirZ;
|
|
leftZ = upX * ndirY - upY * ndirX;
|
|
// normalize left
|
|
float invLeftLength = Math.invsqrt(leftX * leftX + leftY * leftY + leftZ * leftZ);
|
|
leftX *= invLeftLength;
|
|
leftY *= invLeftLength;
|
|
leftZ *= invLeftLength;
|
|
// up = direction x left
|
|
float upnX = ndirY * leftZ - ndirZ * leftY;
|
|
float upnY = ndirZ * leftX - ndirX * leftZ;
|
|
float upnZ = ndirX * leftY - ndirY * leftX;
|
|
float rm00 = leftX;
|
|
float rm01 = leftY;
|
|
float rm02 = leftZ;
|
|
float rm10 = upnX;
|
|
float rm11 = upnY;
|
|
float rm12 = upnZ;
|
|
float rm20 = ndirX;
|
|
float rm21 = ndirY;
|
|
float rm22 = ndirZ;
|
|
dest._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33);
|
|
float nm00 = m00 * rm00 + m10 * rm01 + m20 * rm02;
|
|
float nm01 = m01 * rm00 + m11 * rm01 + m21 * rm02;
|
|
float nm02 = m02 * rm00 + m12 * rm01 + m22 * rm02;
|
|
float nm03 = m03 * rm00 + m13 * rm01 + m23 * rm02;
|
|
float nm10 = m00 * rm10 + m10 * rm11 + m20 * rm12;
|
|
float nm11 = m01 * rm10 + m11 * rm11 + m21 * rm12;
|
|
float nm12 = m02 * rm10 + m12 * rm11 + m22 * rm12;
|
|
float nm13 = m03 * rm10 + m13 * rm11 + m23 * rm12;
|
|
dest._m20(m00 * rm20 + m10 * rm21 + m20 * rm22)
|
|
._m21(m01 * rm20 + m11 * rm21 + m21 * rm22)
|
|
._m22(m02 * rm20 + m12 * rm21 + m22 * rm22)
|
|
._m23(m03 * rm20 + m13 * rm21 + m23 * rm22)
|
|
._m00(nm00)
|
|
._m01(nm01)
|
|
._m02(nm02)
|
|
._m03(nm03)
|
|
._m10(nm10)
|
|
._m11(nm11)
|
|
._m12(nm12)
|
|
._m13(nm13)
|
|
._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a model transformation for a right-handed coordinate system,
|
|
* that aligns the local <code>-z</code> axis with <code>dir</code>.
|
|
* <p>
|
|
* In order to apply the rotation transformation to a previous existing transformation,
|
|
* use {@link #rotateTowards(float, float, float, float, float, float) rotateTowards}.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>setLookAt(new Vector3f(), new Vector3f(dir).negate(), up).invertAffine()</code>
|
|
*
|
|
* @see #rotationTowards(Vector3fc, Vector3fc)
|
|
* @see #rotateTowards(float, float, float, float, float, float)
|
|
*
|
|
* @param dir
|
|
* the direction to orient the local -z axis towards
|
|
* @param up
|
|
* the up vector
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotationTowards(Vector3fc dir, Vector3fc up) {
|
|
return rotationTowards(dir.x(), dir.y(), dir.z(), up.x(), up.y(), up.z());
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a model transformation for a right-handed coordinate system,
|
|
* that aligns the local <code>-z</code> axis with <code>(dirX, dirY, dirZ)</code>.
|
|
* <p>
|
|
* In order to apply the rotation transformation to a previous existing transformation,
|
|
* use {@link #rotateTowards(float, float, float, float, float, float) rotateTowards}.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>setLookAt(0, 0, 0, -dirX, -dirY, -dirZ, upX, upY, upZ).invertAffine()</code>
|
|
*
|
|
* @see #rotateTowards(Vector3fc, Vector3fc)
|
|
* @see #rotationTowards(float, float, float, float, float, float)
|
|
*
|
|
* @param dirX
|
|
* the x-coordinate of the direction to rotate towards
|
|
* @param dirY
|
|
* the y-coordinate of the direction to rotate towards
|
|
* @param dirZ
|
|
* the z-coordinate of the direction to rotate towards
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @return this
|
|
*/
|
|
public Matrix4f rotationTowards(float dirX, float dirY, float dirZ, float upX, float upY, float upZ) {
|
|
// Normalize direction
|
|
float invDirLength = Math.invsqrt(dirX * dirX + dirY * dirY + dirZ * dirZ);
|
|
float ndirX = dirX * invDirLength;
|
|
float ndirY = dirY * invDirLength;
|
|
float ndirZ = dirZ * invDirLength;
|
|
// left = up x direction
|
|
float leftX, leftY, leftZ;
|
|
leftX = upY * ndirZ - upZ * ndirY;
|
|
leftY = upZ * ndirX - upX * ndirZ;
|
|
leftZ = upX * ndirY - upY * ndirX;
|
|
// normalize left
|
|
float invLeftLength = Math.invsqrt(leftX * leftX + leftY * leftY + leftZ * leftZ);
|
|
leftX *= invLeftLength;
|
|
leftY *= invLeftLength;
|
|
leftZ *= invLeftLength;
|
|
// up = direction x left
|
|
float upnX = ndirY * leftZ - ndirZ * leftY;
|
|
float upnY = ndirZ * leftX - ndirX * leftZ;
|
|
float upnZ = ndirX * leftY - ndirY * leftX;
|
|
if ((properties & PROPERTY_IDENTITY) == 0)
|
|
MemUtil.INSTANCE.identity(this);
|
|
this._m00(leftX)
|
|
._m01(leftY)
|
|
._m02(leftZ)
|
|
._m10(upnX)
|
|
._m11(upnY)
|
|
._m12(upnZ)
|
|
._m20(ndirX)
|
|
._m21(ndirY)
|
|
._m22(ndirZ)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a model transformation for a right-handed coordinate system,
|
|
* that translates to the given <code>pos</code> and aligns the local <code>-z</code>
|
|
* axis with <code>dir</code>.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translation(pos).rotateTowards(dir, up)</code>
|
|
*
|
|
* @see #translation(Vector3fc)
|
|
* @see #rotateTowards(Vector3fc, Vector3fc)
|
|
*
|
|
* @param pos
|
|
* the position to translate to
|
|
* @param dir
|
|
* the direction to rotate towards
|
|
* @param up
|
|
* the up vector
|
|
* @return this
|
|
*/
|
|
public Matrix4f translationRotateTowards(Vector3fc pos, Vector3fc dir, Vector3fc up) {
|
|
return translationRotateTowards(pos.x(), pos.y(), pos.z(), dir.x(), dir.y(), dir.z(), up.x(), up.y(), up.z());
|
|
}
|
|
|
|
/**
|
|
* Set this matrix to a model transformation for a right-handed coordinate system,
|
|
* that translates to the given <code>(posX, posY, posZ)</code> and aligns the local <code>-z</code>
|
|
* axis with <code>(dirX, dirY, dirZ)</code>.
|
|
* <p>
|
|
* This method is equivalent to calling: <code>translation(posX, posY, posZ).rotateTowards(dirX, dirY, dirZ, upX, upY, upZ)</code>
|
|
*
|
|
* @see #translation(float, float, float)
|
|
* @see #rotateTowards(float, float, float, float, float, float)
|
|
*
|
|
* @param posX
|
|
* the x-coordinate of the position to translate to
|
|
* @param posY
|
|
* the y-coordinate of the position to translate to
|
|
* @param posZ
|
|
* the z-coordinate of the position to translate to
|
|
* @param dirX
|
|
* the x-coordinate of the direction to rotate towards
|
|
* @param dirY
|
|
* the y-coordinate of the direction to rotate towards
|
|
* @param dirZ
|
|
* the z-coordinate of the direction to rotate towards
|
|
* @param upX
|
|
* the x-coordinate of the up vector
|
|
* @param upY
|
|
* the y-coordinate of the up vector
|
|
* @param upZ
|
|
* the z-coordinate of the up vector
|
|
* @return this
|
|
*/
|
|
public Matrix4f translationRotateTowards(float posX, float posY, float posZ, float dirX, float dirY, float dirZ, float upX, float upY, float upZ) {
|
|
// Normalize direction
|
|
float invDirLength = Math.invsqrt(dirX * dirX + dirY * dirY + dirZ * dirZ);
|
|
float ndirX = dirX * invDirLength;
|
|
float ndirY = dirY * invDirLength;
|
|
float ndirZ = dirZ * invDirLength;
|
|
// left = up x direction
|
|
float leftX, leftY, leftZ;
|
|
leftX = upY * ndirZ - upZ * ndirY;
|
|
leftY = upZ * ndirX - upX * ndirZ;
|
|
leftZ = upX * ndirY - upY * ndirX;
|
|
// normalize left
|
|
float invLeftLength = Math.invsqrt(leftX * leftX + leftY * leftY + leftZ * leftZ);
|
|
leftX *= invLeftLength;
|
|
leftY *= invLeftLength;
|
|
leftZ *= invLeftLength;
|
|
// up = direction x left
|
|
float upnX = ndirY * leftZ - ndirZ * leftY;
|
|
float upnY = ndirZ * leftX - ndirX * leftZ;
|
|
float upnZ = ndirX * leftY - ndirY * leftX;
|
|
this._m00(leftX)
|
|
._m01(leftY)
|
|
._m02(leftZ)
|
|
._m03(0.0f)
|
|
._m10(upnX)
|
|
._m11(upnY)
|
|
._m12(upnZ)
|
|
._m13(0.0f)
|
|
._m20(ndirX)
|
|
._m21(ndirY)
|
|
._m22(ndirZ)
|
|
._m23(0.0f)
|
|
._m30(posX)
|
|
._m31(posY)
|
|
._m32(posZ)
|
|
._m33(1.0f)
|
|
._properties(PROPERTY_AFFINE | PROPERTY_ORTHONORMAL);
|
|
return this;
|
|
}
|
|
|
|
public Vector3f getEulerAnglesZYX(Vector3f dest) {
|
|
dest.x = Math.atan2(m12, m22);
|
|
dest.y = Math.atan2(-m02, Math.sqrt(1.0f - m02 * m02));
|
|
dest.z = Math.atan2(m01, m00);
|
|
return dest;
|
|
}
|
|
|
|
public Vector3f getEulerAnglesXYZ(Vector3f dest) {
|
|
dest.x = Math.atan2(-m21, m22);
|
|
dest.y = Math.atan2(m20, Math.sqrt(1.0f - m20 * m20));
|
|
dest.z = Math.atan2(-m10, m00);
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Compute the extents of the coordinate system before this {@link #isAffine() affine} transformation was applied
|
|
* and store the resulting corner coordinates in <code>corner</code> and the span vectors in
|
|
* <code>xDir</code>, <code>yDir</code> and <code>zDir</code>.
|
|
* <p>
|
|
* That means, given the maximum extents of the coordinate system between <code>[-1..+1]</code> in all dimensions,
|
|
* this method returns one corner and the length and direction of the three base axis vectors in the coordinate
|
|
* system before this transformation is applied, which transforms into the corner coordinates <code>[-1, +1]</code>.
|
|
* <p>
|
|
* This method is equivalent to computing at least three adjacent corners using {@link #frustumCorner(int, Vector3f)}
|
|
* and subtracting them to obtain the length and direction of the span vectors.
|
|
*
|
|
* @param corner
|
|
* will hold one corner of the span (usually the corner {@link Matrix4fc#CORNER_NXNYNZ})
|
|
* @param xDir
|
|
* will hold the direction and length of the span along the positive X axis
|
|
* @param yDir
|
|
* will hold the direction and length of the span along the positive Y axis
|
|
* @param zDir
|
|
* will hold the direction and length of the span along the positive z axis
|
|
* @return this
|
|
*/
|
|
public Matrix4f affineSpan(Vector3f corner, Vector3f xDir, Vector3f yDir, Vector3f zDir) {
|
|
float a = m10 * m22, b = m10 * m21, c = m10 * m02, d = m10 * m01;
|
|
float e = m11 * m22, f = m11 * m20, g = m11 * m02, h = m11 * m00;
|
|
float i = m12 * m21, j = m12 * m20, k = m12 * m01, l = m12 * m00;
|
|
float m = m20 * m02, n = m20 * m01, o = m21 * m02, p = m21 * m00;
|
|
float q = m22 * m01, r = m22 * m00;
|
|
float s = 1.0f / (m00 * m11 - m01 * m10) * m22 + (m02 * m10 - m00 * m12) * m21 + (m01 * m12 - m02 * m11) * m20;
|
|
float nm00 = (e - i) * s, nm01 = (o - q) * s, nm02 = (k - g) * s;
|
|
float nm10 = (j - a) * s, nm11 = (r - m) * s, nm12 = (c - l) * s;
|
|
float nm20 = (b - f) * s, nm21 = (n - p) * s, nm22 = (h - d) * s;
|
|
corner.x = -nm00 - nm10 - nm20 + (a * m31 - b * m32 + f * m32 - e * m30 + i * m30 - j * m31) * s;
|
|
corner.y = -nm01 - nm11 - nm21 + (m * m31 - n * m32 + p * m32 - o * m30 + q * m30 - r * m31) * s;
|
|
corner.z = -nm02 - nm12 - nm22 + (g * m30 - k * m30 + l * m31 - c * m31 + d * m32 - h * m32) * s;
|
|
xDir.x = 2.0f * nm00; xDir.y = 2.0f * nm01; xDir.z = 2.0f * nm02;
|
|
yDir.x = 2.0f * nm10; yDir.y = 2.0f * nm11; yDir.z = 2.0f * nm12;
|
|
zDir.x = 2.0f * nm20; zDir.y = 2.0f * nm21; zDir.z = 2.0f * nm22;
|
|
return this;
|
|
}
|
|
|
|
public boolean testPoint(float x, float y, float z) {
|
|
float nxX = m03 + m00, nxY = m13 + m10, nxZ = m23 + m20, nxW = m33 + m30;
|
|
float pxX = m03 - m00, pxY = m13 - m10, pxZ = m23 - m20, pxW = m33 - m30;
|
|
float nyX = m03 + m01, nyY = m13 + m11, nyZ = m23 + m21, nyW = m33 + m31;
|
|
float pyX = m03 - m01, pyY = m13 - m11, pyZ = m23 - m21, pyW = m33 - m31;
|
|
float nzX = m03 + m02, nzY = m13 + m12, nzZ = m23 + m22, nzW = m33 + m32;
|
|
float pzX = m03 - m02, pzY = m13 - m12, pzZ = m23 - m22, pzW = m33 - m32;
|
|
return nxX * x + nxY * y + nxZ * z + nxW >= 0 && pxX * x + pxY * y + pxZ * z + pxW >= 0 &&
|
|
nyX * x + nyY * y + nyZ * z + nyW >= 0 && pyX * x + pyY * y + pyZ * z + pyW >= 0 &&
|
|
nzX * x + nzY * y + nzZ * z + nzW >= 0 && pzX * x + pzY * y + pzZ * z + pzW >= 0;
|
|
}
|
|
|
|
public boolean testSphere(float x, float y, float z, float r) {
|
|
float invl;
|
|
float nxX = m03 + m00, nxY = m13 + m10, nxZ = m23 + m20, nxW = m33 + m30;
|
|
invl = Math.invsqrt(nxX * nxX + nxY * nxY + nxZ * nxZ);
|
|
nxX *= invl; nxY *= invl; nxZ *= invl; nxW *= invl;
|
|
float pxX = m03 - m00, pxY = m13 - m10, pxZ = m23 - m20, pxW = m33 - m30;
|
|
invl = Math.invsqrt(pxX * pxX + pxY * pxY + pxZ * pxZ);
|
|
pxX *= invl; pxY *= invl; pxZ *= invl; pxW *= invl;
|
|
float nyX = m03 + m01, nyY = m13 + m11, nyZ = m23 + m21, nyW = m33 + m31;
|
|
invl = Math.invsqrt(nyX * nyX + nyY * nyY + nyZ * nyZ);
|
|
nyX *= invl; nyY *= invl; nyZ *= invl; nyW *= invl;
|
|
float pyX = m03 - m01, pyY = m13 - m11, pyZ = m23 - m21, pyW = m33 - m31;
|
|
invl = Math.invsqrt(pyX * pyX + pyY * pyY + pyZ * pyZ);
|
|
pyX *= invl; pyY *= invl; pyZ *= invl; pyW *= invl;
|
|
float nzX = m03 + m02, nzY = m13 + m12, nzZ = m23 + m22, nzW = m33 + m32;
|
|
invl = Math.invsqrt(nzX * nzX + nzY * nzY + nzZ * nzZ);
|
|
nzX *= invl; nzY *= invl; nzZ *= invl; nzW *= invl;
|
|
float pzX = m03 - m02, pzY = m13 - m12, pzZ = m23 - m22, pzW = m33 - m32;
|
|
invl = Math.invsqrt(pzX * pzX + pzY * pzY + pzZ * pzZ);
|
|
pzX *= invl; pzY *= invl; pzZ *= invl; pzW *= invl;
|
|
return nxX * x + nxY * y + nxZ * z + nxW >= -r && pxX * x + pxY * y + pxZ * z + pxW >= -r &&
|
|
nyX * x + nyY * y + nyZ * z + nyW >= -r && pyX * x + pyY * y + pyZ * z + pyW >= -r &&
|
|
nzX * x + nzY * y + nzZ * z + nzW >= -r && pzX * x + pzY * y + pzZ * z + pzW >= -r;
|
|
}
|
|
|
|
public boolean testAab(float minX, float minY, float minZ, float maxX, float maxY, float maxZ) {
|
|
float nxX = m03 + m00, nxY = m13 + m10, nxZ = m23 + m20, nxW = m33 + m30;
|
|
float pxX = m03 - m00, pxY = m13 - m10, pxZ = m23 - m20, pxW = m33 - m30;
|
|
float nyX = m03 + m01, nyY = m13 + m11, nyZ = m23 + m21, nyW = m33 + m31;
|
|
float pyX = m03 - m01, pyY = m13 - m11, pyZ = m23 - m21, pyW = m33 - m31;
|
|
float nzX = m03 + m02, nzY = m13 + m12, nzZ = m23 + m22, nzW = m33 + m32;
|
|
float pzX = m03 - m02, pzY = m13 - m12, pzZ = m23 - m22, pzW = m33 - m32;
|
|
/*
|
|
* This is an implementation of the "2.4 Basic intersection test" of the mentioned site.
|
|
* It does not distinguish between partially inside and fully inside, though, so the test with the 'p' vertex is omitted.
|
|
*/
|
|
return nxX * (nxX < 0 ? minX : maxX) + nxY * (nxY < 0 ? minY : maxY) + nxZ * (nxZ < 0 ? minZ : maxZ) >= -nxW &&
|
|
pxX * (pxX < 0 ? minX : maxX) + pxY * (pxY < 0 ? minY : maxY) + pxZ * (pxZ < 0 ? minZ : maxZ) >= -pxW &&
|
|
nyX * (nyX < 0 ? minX : maxX) + nyY * (nyY < 0 ? minY : maxY) + nyZ * (nyZ < 0 ? minZ : maxZ) >= -nyW &&
|
|
pyX * (pyX < 0 ? minX : maxX) + pyY * (pyY < 0 ? minY : maxY) + pyZ * (pyZ < 0 ? minZ : maxZ) >= -pyW &&
|
|
nzX * (nzX < 0 ? minX : maxX) + nzY * (nzY < 0 ? minY : maxY) + nzZ * (nzZ < 0 ? minZ : maxZ) >= -nzW &&
|
|
pzX * (pzX < 0 ? minX : maxX) + pzY * (pzY < 0 ? minY : maxY) + pzZ * (pzZ < 0 ? minZ : maxZ) >= -pzW;
|
|
}
|
|
|
|
/**
|
|
* Apply an oblique projection transformation to this matrix with the given values for <code>a</code> and
|
|
* <code>b</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the oblique transformation matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* oblique transformation will be applied first!
|
|
* <p>
|
|
* The oblique transformation is defined as:
|
|
* <pre>
|
|
* x' = x + a*z
|
|
* y' = y + a*z
|
|
* z' = z
|
|
* </pre>
|
|
* or in matrix form:
|
|
* <pre>
|
|
* 1 0 a 0
|
|
* 0 1 b 0
|
|
* 0 0 1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @param a
|
|
* the value for the z factor that applies to x
|
|
* @param b
|
|
* the value for the z factor that applies to y
|
|
* @return this
|
|
*/
|
|
public Matrix4f obliqueZ(float a, float b) {
|
|
this.m20 = m00 * a + m10 * b + m20;
|
|
this.m21 = m01 * a + m11 * b + m21;
|
|
this.m22 = m02 * a + m12 * b + m22;
|
|
this._properties(this.properties & PROPERTY_AFFINE);
|
|
return this;
|
|
}
|
|
|
|
/**
|
|
* Apply an oblique projection transformation to this matrix with the given values for <code>a</code> and
|
|
* <code>b</code> and store the result in <code>dest</code>.
|
|
* <p>
|
|
* If <code>M</code> is <code>this</code> matrix and <code>O</code> the oblique transformation matrix,
|
|
* then the new matrix will be <code>M * O</code>. So when transforming a
|
|
* vector <code>v</code> with the new matrix by using <code>M * O * v</code>, the
|
|
* oblique transformation will be applied first!
|
|
* <p>
|
|
* The oblique transformation is defined as:
|
|
* <pre>
|
|
* x' = x + a*z
|
|
* y' = y + a*z
|
|
* z' = z
|
|
* </pre>
|
|
* or in matrix form:
|
|
* <pre>
|
|
* 1 0 a 0
|
|
* 0 1 b 0
|
|
* 0 0 1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @param a
|
|
* the value for the z factor that applies to x
|
|
* @param b
|
|
* the value for the z factor that applies to y
|
|
* @param dest
|
|
* will hold the result
|
|
* @return dest
|
|
*/
|
|
public Matrix4f obliqueZ(float a, float b, Matrix4f dest) {
|
|
dest._m00(m00)
|
|
._m01(m01)
|
|
._m02(m02)
|
|
._m03(m03)
|
|
._m10(m10)
|
|
._m11(m11)
|
|
._m12(m12)
|
|
._m13(m13)
|
|
._m20(m00 * a + m10 * b + m20)
|
|
._m21(m01 * a + m11 * b + m21)
|
|
._m22(m02 * a + m12 * b + m22)
|
|
._m23(m23)
|
|
._m30(m30)
|
|
._m31(m31)
|
|
._m32(m32)
|
|
._m33(m33)
|
|
._properties(this.properties & PROPERTY_AFFINE);
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Create a view and projection matrix from a given <code>eye</code> position, a given bottom left corner position <code>p</code> of the near plane rectangle
|
|
* and the extents of the near plane rectangle along its local <code>x</code> and <code>y</code> axes, and store the resulting matrices
|
|
* in <code>projDest</code> and <code>viewDest</code>.
|
|
* <p>
|
|
* This method creates a view and perspective projection matrix assuming that there is a pinhole camera at position <code>eye</code>
|
|
* projecting the scene onto the near plane defined by the rectangle.
|
|
* <p>
|
|
* All positions and lengths are in the same (world) unit.
|
|
*
|
|
* @param eye
|
|
* the position of the camera
|
|
* @param p
|
|
* the bottom left corner of the near plane rectangle (will map to the bottom left corner in window coordinates)
|
|
* @param x
|
|
* the direction and length of the local "bottom/top" X axis/side of the near plane rectangle
|
|
* @param y
|
|
* the direction and length of the local "left/right" Y axis/side of the near plane rectangle
|
|
* @param nearFarDist
|
|
* the distance between the far and near plane (the near plane will be calculated by this method).
|
|
* If the special value {@link Float#POSITIVE_INFINITY} is used, the far clipping plane will be at positive infinity.
|
|
* If the special value {@link Float#NEGATIVE_INFINITY} is used, the near and far planes will be swapped and
|
|
* the near clipping plane will be at positive infinity.
|
|
* If a negative value is used (except for {@link Float#NEGATIVE_INFINITY}) the near and far planes will be swapped
|
|
* @param zeroToOne
|
|
* whether to use Vulkan's and Direct3D's NDC z range of <code>[0..+1]</code> when <code>true</code>
|
|
* or whether to use OpenGL's NDC z range of <code>[-1..+1]</code> when <code>false</code>
|
|
* @param projDest
|
|
* will hold the resulting projection matrix
|
|
* @param viewDest
|
|
* will hold the resulting view matrix
|
|
*/
|
|
public static void projViewFromRectangle(
|
|
Vector3f eye, Vector3f p, Vector3f x, Vector3f y, float nearFarDist, boolean zeroToOne,
|
|
Matrix4f projDest, Matrix4f viewDest) {
|
|
float zx = y.y * x.z - y.z * x.y, zy = y.z * x.x - y.x * x.z, zz = y.x * x.y - y.y * x.x;
|
|
float zd = zx * (p.x - eye.x) + zy * (p.y - eye.y) + zz * (p.z - eye.z);
|
|
float zs = zd >= 0 ? 1 : -1; zx *= zs; zy *= zs; zz *= zs; zd *= zs;
|
|
viewDest.setLookAt(eye.x, eye.y, eye.z, eye.x + zx, eye.y + zy, eye.z + zz, y.x, y.y, y.z);
|
|
float px = viewDest.m00 * p.x + viewDest.m10 * p.y + viewDest.m20 * p.z + viewDest.m30;
|
|
float py = viewDest.m01 * p.x + viewDest.m11 * p.y + viewDest.m21 * p.z + viewDest.m31;
|
|
float tx = viewDest.m00 * x.x + viewDest.m10 * x.y + viewDest.m20 * x.z;
|
|
float ty = viewDest.m01 * y.x + viewDest.m11 * y.y + viewDest.m21 * y.z;
|
|
float len = Math.sqrt(zx * zx + zy * zy + zz * zz);
|
|
float near = zd / len, far;
|
|
if (Float.isInfinite(nearFarDist) && nearFarDist < 0.0f) {
|
|
far = near;
|
|
near = Float.POSITIVE_INFINITY;
|
|
} else if (Float.isInfinite(nearFarDist) && nearFarDist > 0.0f) {
|
|
far = Float.POSITIVE_INFINITY;
|
|
} else if (nearFarDist < 0.0f) {
|
|
far = near;
|
|
near = near + nearFarDist;
|
|
} else {
|
|
far = near + nearFarDist;
|
|
}
|
|
projDest.setFrustum(px, px + tx, py, py + ty, near, far, zeroToOne);
|
|
}
|
|
|
|
/**
|
|
* Apply a transformation to this matrix to ensure that the local Y axis (as obtained by {@link #positiveY(Vector3f)})
|
|
* will be coplanar to the plane spanned by the local Z axis (as obtained by {@link #positiveZ(Vector3f)}) and the
|
|
* given vector <code>up</code>.
|
|
* <p>
|
|
* This effectively ensures that the resulting matrix will be equal to the one obtained from
|
|
* {@link #setLookAt(Vector3fc, Vector3fc, Vector3fc)} called with the current
|
|
* local origin of this matrix (as obtained by {@link #originAffine(Vector3f)}), the sum of this position and the
|
|
* negated local Z axis as well as the given vector <code>up</code>.
|
|
* <p>
|
|
* This method must only be called on {@link #isAffine()} matrices.
|
|
*
|
|
* @param up
|
|
* the up vector
|
|
* @return this
|
|
*/
|
|
public Matrix4f withLookAtUp(Vector3fc up) {
|
|
return withLookAtUp(up.x(), up.y(), up.z(), this);
|
|
}
|
|
|
|
public Matrix4f withLookAtUp(Vector3fc up, Matrix4f dest) {
|
|
return withLookAtUp(up.x(), up.y(), up.z());
|
|
}
|
|
|
|
/**
|
|
* Apply a transformation to this matrix to ensure that the local Y axis (as obtained by {@link #positiveY(Vector3f)})
|
|
* will be coplanar to the plane spanned by the local Z axis (as obtained by {@link #positiveZ(Vector3f)}) and the
|
|
* given vector <code>(upX, upY, upZ)</code>.
|
|
* <p>
|
|
* This effectively ensures that the resulting matrix will be equal to the one obtained from
|
|
* {@link #setLookAt(float, float, float, float, float, float, float, float, float)} called with the current
|
|
* local origin of this matrix (as obtained by {@link #originAffine(Vector3f)}), the sum of this position and the
|
|
* negated local Z axis as well as the given vector <code>(upX, upY, upZ)</code>.
|
|
* <p>
|
|
* This method must only be called on {@link #isAffine()} matrices.
|
|
*
|
|
* @param upX
|
|
* the x coordinate of the up vector
|
|
* @param upY
|
|
* the y coordinate of the up vector
|
|
* @param upZ
|
|
* the z coordinate of the up vector
|
|
* @return this
|
|
*/
|
|
public Matrix4f withLookAtUp(float upX, float upY, float upZ) {
|
|
return withLookAtUp(upX, upY, upZ, this);
|
|
}
|
|
|
|
public Matrix4f withLookAtUp(float upX, float upY, float upZ, Matrix4f dest) {
|
|
float y = (upY * m21 - upZ * m11) * m02 +
|
|
(upZ * m01 - upX * m21) * m12 +
|
|
(upX * m11 - upY * m01) * m22;
|
|
float x = upX * m01 + upY * m11 + upZ * m21;
|
|
if ((properties & PROPERTY_ORTHONORMAL) == 0)
|
|
x *= Math.sqrt(m01 * m01 + m11 * m11 + m21 * m21);
|
|
float invsqrt = Math.invsqrt(y * y + x * x);
|
|
float c = x * invsqrt, s = y * invsqrt;
|
|
float nm00 = c * m00 - s * m01, nm10 = c * m10 - s * m11, nm20 = c * m20 - s * m21, nm31 = s * m30 + c * m31;
|
|
float nm01 = s * m00 + c * m01, nm11 = s * m10 + c * m11, nm21 = s * m20 + c * m21, nm30 = c * m30 - s * m31;
|
|
dest._m00(nm00)._m10(nm10)._m20(nm20)._m30(nm30)
|
|
._m01(nm01)._m11(nm11)._m21(nm21)._m31(nm31);
|
|
if (dest != this) {
|
|
dest._m02(m02)._m12(m12)._m22(m22)._m32(m32)
|
|
._m03(m03)._m13(m13)._m23(m23)._m33(m33);
|
|
}
|
|
dest._properties(properties & ~(PROPERTY_PERSPECTIVE | PROPERTY_IDENTITY | PROPERTY_TRANSLATION));
|
|
return dest;
|
|
}
|
|
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 1 0 0 0
|
|
* 0 0 1 0
|
|
* 0 1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapXZY() {
|
|
return mapXZY(this);
|
|
}
|
|
public Matrix4f mapXZY(Matrix4f dest) {
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(m00)._m01(m01)._m02(m02)._m03(m03)._m10(m20)._m11(m21)._m12(m22)._m13(m13)._m20(m10)._m21(m11)._m22(m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 1 0 0 0
|
|
* 0 0 -1 0
|
|
* 0 1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapXZnY() {
|
|
return mapXZnY(this);
|
|
}
|
|
public Matrix4f mapXZnY(Matrix4f dest) {
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(m00)._m01(m01)._m02(m02)._m03(m03)._m10(m20)._m11(m21)._m12(m22)._m13(m13)._m20(-m10)._m21(-m11)._m22(-m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 1 0 0 0
|
|
* 0 -1 0 0
|
|
* 0 0 -1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapXnYnZ() {
|
|
return mapXnYnZ(this);
|
|
}
|
|
public Matrix4f mapXnYnZ(Matrix4f dest) {
|
|
return dest._m00(m00)._m01(m01)._m02(m02)._m03(m03)._m10(-m10)._m11(-m11)._m12(-m12)._m13(m13)._m20(-m20)._m21(-m21)._m22(-m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 1 0 0 0
|
|
* 0 0 1 0
|
|
* 0 -1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapXnZY() {
|
|
return mapXnZY(this);
|
|
}
|
|
public Matrix4f mapXnZY(Matrix4f dest) {
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(m00)._m01(m01)._m02(m02)._m03(m03)._m10(-m20)._m11(-m21)._m12(-m22)._m13(m13)._m20(m10)._m21(m11)._m22(m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 1 0 0 0
|
|
* 0 0 -1 0
|
|
* 0 -1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapXnZnY() {
|
|
return mapXnZnY(this);
|
|
}
|
|
public Matrix4f mapXnZnY(Matrix4f dest) {
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(m00)._m01(m01)._m02(m02)._m03(m03)._m10(-m20)._m11(-m21)._m12(-m22)._m13(m13)._m20(-m10)._m21(-m11)._m22(-m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 1 0 0
|
|
* 1 0 0 0
|
|
* 0 0 1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapYXZ() {
|
|
return mapYXZ(this);
|
|
}
|
|
public Matrix4f mapYXZ(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(m10)._m01(m11)._m02(m12)._m03(m03)._m10(m00)._m11(m01)._m12(m02)._m13(m13)._m20(m20)._m21(m21)._m22(m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 1 0 0
|
|
* 1 0 0 0
|
|
* 0 0 -1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapYXnZ() {
|
|
return mapYXnZ(this);
|
|
}
|
|
public Matrix4f mapYXnZ(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(m10)._m01(m11)._m02(m12)._m03(m03)._m10(m00)._m11(m01)._m12(m02)._m13(m13)._m20(-m20)._m21(-m21)._m22(-m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 1 0
|
|
* 1 0 0 0
|
|
* 0 1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapYZX() {
|
|
return mapYZX(this);
|
|
}
|
|
public Matrix4f mapYZX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(m10)._m01(m11)._m02(m12)._m03(m03)._m10(m20)._m11(m21)._m12(m22)._m13(m13)._m20(m00)._m21(m01)._m22(m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 -1 0
|
|
* 1 0 0 0
|
|
* 0 1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapYZnX() {
|
|
return mapYZnX(this);
|
|
}
|
|
public Matrix4f mapYZnX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(m10)._m01(m11)._m02(m12)._m03(m03)._m10(m20)._m11(m21)._m12(m22)._m13(m13)._m20(-m00)._m21(-m01)._m22(-m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 -1 0 0
|
|
* 1 0 0 0
|
|
* 0 0 1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapYnXZ() {
|
|
return mapYnXZ(this);
|
|
}
|
|
public Matrix4f mapYnXZ(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(m10)._m01(m11)._m02(m12)._m03(m03)._m10(-m00)._m11(-m01)._m12(-m02)._m13(m13)._m20(m20)._m21(m21)._m22(m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 -1 0 0
|
|
* 1 0 0 0
|
|
* 0 0 -1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapYnXnZ() {
|
|
return mapYnXnZ(this);
|
|
}
|
|
public Matrix4f mapYnXnZ(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(m10)._m01(m11)._m02(m12)._m03(m03)._m10(-m00)._m11(-m01)._m12(-m02)._m13(m13)._m20(-m20)._m21(-m21)._m22(-m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 1 0
|
|
* 1 0 0 0
|
|
* 0 -1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapYnZX() {
|
|
return mapYnZX(this);
|
|
}
|
|
public Matrix4f mapYnZX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(m10)._m01(m11)._m02(m12)._m03(m03)._m10(-m20)._m11(-m21)._m12(-m22)._m13(m13)._m20(m00)._m21(m01)._m22(m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 -1 0
|
|
* 1 0 0 0
|
|
* 0 -1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapYnZnX() {
|
|
return mapYnZnX(this);
|
|
}
|
|
public Matrix4f mapYnZnX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(m10)._m01(m11)._m02(m12)._m03(m03)._m10(-m20)._m11(-m21)._m12(-m22)._m13(m13)._m20(-m00)._m21(-m01)._m22(-m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 1 0 0
|
|
* 0 0 1 0
|
|
* 1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapZXY() {
|
|
return mapZXY(this);
|
|
}
|
|
public Matrix4f mapZXY(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(m20)._m01(m21)._m02(m22)._m03(m03)._m10(m00)._m11(m01)._m12(m02)._m13(m13)._m20(m10)._m21(m11)._m22(m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 1 0 0
|
|
* 0 0 -1 0
|
|
* 1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapZXnY() {
|
|
return mapZXnY(this);
|
|
}
|
|
public Matrix4f mapZXnY(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(m20)._m01(m21)._m02(m22)._m03(m03)._m10(m00)._m11(m01)._m12(m02)._m13(m13)._m20(-m10)._m21(-m11)._m22(-m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 1 0
|
|
* 0 1 0 0
|
|
* 1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapZYX() {
|
|
return mapZYX(this);
|
|
}
|
|
public Matrix4f mapZYX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(m20)._m01(m21)._m02(m22)._m03(m03)._m10(m10)._m11(m11)._m12(m12)._m13(m13)._m20(m00)._m21(m01)._m22(m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 -1 0
|
|
* 0 1 0 0
|
|
* 1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapZYnX() {
|
|
return mapZYnX(this);
|
|
}
|
|
public Matrix4f mapZYnX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(m20)._m01(m21)._m02(m22)._m03(m03)._m10(m10)._m11(m11)._m12(m12)._m13(m13)._m20(-m00)._m21(-m01)._m22(-m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 -1 0 0
|
|
* 0 0 1 0
|
|
* 1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapZnXY() {
|
|
return mapZnXY(this);
|
|
}
|
|
public Matrix4f mapZnXY(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(m20)._m01(m21)._m02(m22)._m03(m03)._m10(-m00)._m11(-m01)._m12(-m02)._m13(m13)._m20(m10)._m21(m11)._m22(m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 -1 0 0
|
|
* 0 0 -1 0
|
|
* 1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapZnXnY() {
|
|
return mapZnXnY(this);
|
|
}
|
|
public Matrix4f mapZnXnY(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(m20)._m01(m21)._m02(m22)._m03(m03)._m10(-m00)._m11(-m01)._m12(-m02)._m13(m13)._m20(-m10)._m21(-m11)._m22(-m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 1 0
|
|
* 0 -1 0 0
|
|
* 1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapZnYX() {
|
|
return mapZnYX(this);
|
|
}
|
|
public Matrix4f mapZnYX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(m20)._m01(m21)._m02(m22)._m03(m03)._m10(-m10)._m11(-m11)._m12(-m12)._m13(m13)._m20(m00)._m21(m01)._m22(m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 -1 0
|
|
* 0 -1 0 0
|
|
* 1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapZnYnX() {
|
|
return mapZnYnX(this);
|
|
}
|
|
public Matrix4f mapZnYnX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(m20)._m01(m21)._m02(m22)._m03(m03)._m10(-m10)._m11(-m11)._m12(-m12)._m13(m13)._m20(-m00)._m21(-m01)._m22(-m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* -1 0 0 0
|
|
* 0 1 0 0
|
|
* 0 0 -1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnXYnZ() {
|
|
return mapnXYnZ(this);
|
|
}
|
|
public Matrix4f mapnXYnZ(Matrix4f dest) {
|
|
return dest._m00(-m00)._m01(-m01)._m02(-m02)._m03(m03)._m10(m10)._m11(m11)._m12(m12)._m13(m13)._m20(-m20)._m21(-m21)._m22(-m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* -1 0 0 0
|
|
* 0 0 1 0
|
|
* 0 1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnXZY() {
|
|
return mapnXZY(this);
|
|
}
|
|
public Matrix4f mapnXZY(Matrix4f dest) {
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(-m00)._m01(-m01)._m02(-m02)._m03(m03)._m10(m20)._m11(m21)._m12(m22)._m13(m13)._m20(m10)._m21(m11)._m22(m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* -1 0 0 0
|
|
* 0 0 -1 0
|
|
* 0 1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnXZnY() {
|
|
return mapnXZnY(this);
|
|
}
|
|
public Matrix4f mapnXZnY(Matrix4f dest) {
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(-m00)._m01(-m01)._m02(-m02)._m03(m03)._m10(m20)._m11(m21)._m12(m22)._m13(m13)._m20(-m10)._m21(-m11)._m22(-m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* -1 0 0 0
|
|
* 0 -1 0 0
|
|
* 0 0 1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnXnYZ() {
|
|
return mapnXnYZ(this);
|
|
}
|
|
public Matrix4f mapnXnYZ(Matrix4f dest) {
|
|
return dest._m00(-m00)._m01(-m01)._m02(-m02)._m03(m03)._m10(-m10)._m11(-m11)._m12(-m12)._m13(m13)._m20(m20)._m21(m21)._m22(m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* -1 0 0 0
|
|
* 0 -1 0 0
|
|
* 0 0 -1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnXnYnZ() {
|
|
return mapnXnYnZ(this);
|
|
}
|
|
public Matrix4f mapnXnYnZ(Matrix4f dest) {
|
|
return dest._m00(-m00)._m01(-m01)._m02(-m02)._m03(m03)._m10(-m10)._m11(-m11)._m12(-m12)._m13(m13)._m20(-m20)._m21(-m21)._m22(-m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* -1 0 0 0
|
|
* 0 0 1 0
|
|
* 0 -1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnXnZY() {
|
|
return mapnXnZY(this);
|
|
}
|
|
public Matrix4f mapnXnZY(Matrix4f dest) {
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(-m00)._m01(-m01)._m02(-m02)._m03(m03)._m10(-m20)._m11(-m21)._m12(-m22)._m13(m13)._m20(m10)._m21(m11)._m22(m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* -1 0 0 0
|
|
* 0 0 -1 0
|
|
* 0 -1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnXnZnY() {
|
|
return mapnXnZnY(this);
|
|
}
|
|
public Matrix4f mapnXnZnY(Matrix4f dest) {
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(-m00)._m01(-m01)._m02(-m02)._m03(m03)._m10(-m20)._m11(-m21)._m12(-m22)._m13(m13)._m20(-m10)._m21(-m11)._m22(-m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 1 0 0
|
|
* -1 0 0 0
|
|
* 0 0 1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnYXZ() {
|
|
return mapnYXZ(this);
|
|
}
|
|
public Matrix4f mapnYXZ(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(-m10)._m01(-m11)._m02(-m12)._m03(m03)._m10(m00)._m11(m01)._m12(m02)._m13(m13)._m20(m20)._m21(m21)._m22(m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 1 0 0
|
|
* -1 0 0 0
|
|
* 0 0 -1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnYXnZ() {
|
|
return mapnYXnZ(this);
|
|
}
|
|
public Matrix4f mapnYXnZ(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(-m10)._m01(-m11)._m02(-m12)._m03(m03)._m10(m00)._m11(m01)._m12(m02)._m13(m13)._m20(-m20)._m21(-m21)._m22(-m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 1 0
|
|
* -1 0 0 0
|
|
* 0 1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnYZX() {
|
|
return mapnYZX(this);
|
|
}
|
|
public Matrix4f mapnYZX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(-m10)._m01(-m11)._m02(-m12)._m03(m03)._m10(m20)._m11(m21)._m12(m22)._m13(m13)._m20(m00)._m21(m01)._m22(m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 -1 0
|
|
* -1 0 0 0
|
|
* 0 1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnYZnX() {
|
|
return mapnYZnX(this);
|
|
}
|
|
public Matrix4f mapnYZnX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(-m10)._m01(-m11)._m02(-m12)._m03(m03)._m10(m20)._m11(m21)._m12(m22)._m13(m13)._m20(-m00)._m21(-m01)._m22(-m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 -1 0 0
|
|
* -1 0 0 0
|
|
* 0 0 1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnYnXZ() {
|
|
return mapnYnXZ(this);
|
|
}
|
|
public Matrix4f mapnYnXZ(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(-m10)._m01(-m11)._m02(-m12)._m03(m03)._m10(-m00)._m11(-m01)._m12(-m02)._m13(m13)._m20(m20)._m21(m21)._m22(m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 -1 0 0
|
|
* -1 0 0 0
|
|
* 0 0 -1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnYnXnZ() {
|
|
return mapnYnXnZ(this);
|
|
}
|
|
public Matrix4f mapnYnXnZ(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(-m10)._m01(-m11)._m02(-m12)._m03(m03)._m10(-m00)._m11(-m01)._m12(-m02)._m13(m13)._m20(-m20)._m21(-m21)._m22(-m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 1 0
|
|
* -1 0 0 0
|
|
* 0 -1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnYnZX() {
|
|
return mapnYnZX(this);
|
|
}
|
|
public Matrix4f mapnYnZX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(-m10)._m01(-m11)._m02(-m12)._m03(m03)._m10(-m20)._m11(-m21)._m12(-m22)._m13(m13)._m20(m00)._m21(m01)._m22(m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 -1 0
|
|
* -1 0 0 0
|
|
* 0 -1 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnYnZnX() {
|
|
return mapnYnZnX(this);
|
|
}
|
|
public Matrix4f mapnYnZnX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(-m10)._m01(-m11)._m02(-m12)._m03(m03)._m10(-m20)._m11(-m21)._m12(-m22)._m13(m13)._m20(-m00)._m21(-m01)._m22(-m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 1 0 0
|
|
* 0 0 1 0
|
|
* -1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnZXY() {
|
|
return mapnZXY(this);
|
|
}
|
|
public Matrix4f mapnZXY(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(-m20)._m01(-m21)._m02(-m22)._m03(m03)._m10(m00)._m11(m01)._m12(m02)._m13(m13)._m20(m10)._m21(m11)._m22(m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 1 0 0
|
|
* 0 0 -1 0
|
|
* -1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnZXnY() {
|
|
return mapnZXnY(this);
|
|
}
|
|
public Matrix4f mapnZXnY(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(-m20)._m01(-m21)._m02(-m22)._m03(m03)._m10(m00)._m11(m01)._m12(m02)._m13(m13)._m20(-m10)._m21(-m11)._m22(-m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 1 0
|
|
* 0 1 0 0
|
|
* -1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnZYX() {
|
|
return mapnZYX(this);
|
|
}
|
|
public Matrix4f mapnZYX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(-m20)._m01(-m21)._m02(-m22)._m03(m03)._m10(m10)._m11(m11)._m12(m12)._m13(m13)._m20(m00)._m21(m01)._m22(m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 -1 0
|
|
* 0 1 0 0
|
|
* -1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnZYnX() {
|
|
return mapnZYnX(this);
|
|
}
|
|
public Matrix4f mapnZYnX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(-m20)._m01(-m21)._m02(-m22)._m03(m03)._m10(m10)._m11(m11)._m12(m12)._m13(m13)._m20(-m00)._m21(-m01)._m22(-m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 -1 0 0
|
|
* 0 0 1 0
|
|
* -1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnZnXY() {
|
|
return mapnZnXY(this);
|
|
}
|
|
public Matrix4f mapnZnXY(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(-m20)._m01(-m21)._m02(-m22)._m03(m03)._m10(-m00)._m11(-m01)._m12(-m02)._m13(m13)._m20(m10)._m21(m11)._m22(m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 -1 0 0
|
|
* 0 0 -1 0
|
|
* -1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnZnXnY() {
|
|
return mapnZnXnY(this);
|
|
}
|
|
public Matrix4f mapnZnXnY(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
float m10 = this.m10, m11 = this.m11, m12 = this.m12;
|
|
return dest._m00(-m20)._m01(-m21)._m02(-m22)._m03(m03)._m10(-m00)._m11(-m01)._m12(-m02)._m13(m13)._m20(-m10)._m21(-m11)._m22(-m12)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 1 0
|
|
* 0 -1 0 0
|
|
* -1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnZnYX() {
|
|
return mapnZnYX(this);
|
|
}
|
|
public Matrix4f mapnZnYX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(-m20)._m01(-m21)._m02(-m22)._m03(m03)._m10(-m10)._m11(-m11)._m12(-m12)._m13(m13)._m20(m00)._m21(m01)._m22(m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 0 0 -1 0
|
|
* 0 -1 0 0
|
|
* -1 0 0 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f mapnZnYnX() {
|
|
return mapnZnYnX(this);
|
|
}
|
|
public Matrix4f mapnZnYnX(Matrix4f dest) {
|
|
float m00 = this.m00, m01 = this.m01, m02 = this.m02;
|
|
return dest._m00(-m20)._m01(-m21)._m02(-m22)._m03(m03)._m10(-m10)._m11(-m11)._m12(-m12)._m13(m13)._m20(-m00)._m21(-m01)._m22(-m02)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33)._properties(properties & (PROPERTY_AFFINE | PROPERTY_ORTHONORMAL));
|
|
}
|
|
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* -1 0 0 0
|
|
* 0 1 0 0
|
|
* 0 0 1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f negateX() {
|
|
return _m00(-m00)._m01(-m01)._m02(-m02);
|
|
}
|
|
public Matrix4f negateX(Matrix4f dest) {
|
|
return dest._m00(-m00)._m01(-m01)._m02(-m02)._m03(m03)._m10(m10)._m11(m11)._m12(m12)._m13(m13)._m20(m20)._m21(m21)._m22(m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33);
|
|
}
|
|
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 1 0 0 0
|
|
* 0 -1 0 0
|
|
* 0 0 1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f negateY() {
|
|
return _m10(-m10)._m11(-m11)._m12(-m12);
|
|
}
|
|
public Matrix4f negateY(Matrix4f dest) {
|
|
return dest._m00(m00)._m01(m01)._m02(m02)._m03(m03)._m10(-m10)._m11(-m11)._m12(-m12)._m13(m13)._m20(m20)._m21(m21)._m22(m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33);
|
|
}
|
|
|
|
/**
|
|
* Multiply <code>this</code> by the matrix
|
|
* <pre>
|
|
* 1 0 0 0
|
|
* 0 1 0 0
|
|
* 0 0 -1 0
|
|
* 0 0 0 1
|
|
* </pre>
|
|
*
|
|
* @return this
|
|
*/
|
|
public Matrix4f negateZ() {
|
|
return _m20(-m20)._m21(-m21)._m22(-m22);
|
|
}
|
|
public Matrix4f negateZ(Matrix4f dest) {
|
|
return dest._m00(m00)._m01(m01)._m02(m02)._m03(m03)._m10(m10)._m11(m11)._m12(m12)._m13(m13)._m20(-m20)._m21(-m21)._m22(-m22)._m23(m23)._m30(m30)._m31(m31)._m32(m32)._m33(m33);
|
|
}
|
|
|
|
public boolean isFinite() {
|
|
return Math.isFinite(m00) && Math.isFinite(m01) && Math.isFinite(m02) && Math.isFinite(m03) &&
|
|
Math.isFinite(m10) && Math.isFinite(m11) && Math.isFinite(m12) && Math.isFinite(m13) &&
|
|
Math.isFinite(m20) && Math.isFinite(m21) && Math.isFinite(m22) && Math.isFinite(m23) &&
|
|
Math.isFinite(m30) && Math.isFinite(m31) && Math.isFinite(m32) && Math.isFinite(m33);
|
|
}
|
|
|
|
public Object clone() throws CloneNotSupportedException {
|
|
return super.clone();
|
|
}
|
|
|
|
}
|