nv-sgl 0.6.0__cp313-cp313-win_amd64.whl

sgl/include/sgl/math/matrix_math.h

@@ -0,0 +1,746 @@
+// SPDX-License-Identifier: Apache-2.0
+
+// Most of this code is derived from the GLM library at https://github.com/g-truc/glm
+// License: https://github.com/g-truc/glm/blob/master/copying.txt
+
+#pragma once
+
+#include "sgl/math/matrix_types.h"
+#include "sgl/math/vector.h"
+#include "sgl/math/quaternion.h"
+#include "sgl/core/error.h"
+#include "sgl/core/format.h"
+
+namespace sgl::math {
+
+// ----------------------------------------------------------------------------
+// Binary operators (component-wise)
+// ----------------------------------------------------------------------------
+
+/// Binary * operator
+template<typename T, int R, int C>
+[[nodiscard]] matrix<T, R, C> operator*(const matrix<T, R, C>& lhs, const T& rhs)
+{
+    matrix<T, R, C> result;
+    for (int r = 0; r < R; ++r)
+        for (int c = 0; c < C; ++c)
+            result[r][c] = lhs[r][c] * rhs;
+    return result;
+}
+
+// ----------------------------------------------------------------------------
+// Multiplication
+// ----------------------------------------------------------------------------
+
+/// Multiply matrix and matrix.
+template<typename T, int M, int N, int P>
+[[nodiscard]] matrix<T, M, P> mul(const matrix<T, M, N>& lhs, const matrix<T, N, P>& rhs)
+{
+    matrix<T, M, P> result;
+    for (int m = 0; m < M; ++m)
+        for (int p = 0; p < P; ++p)
+            result[m][p] = dot(lhs.get_row(m), rhs.get_col(p));
+    return result;
+}
+/// Multiply matrix and vector. Vector is treated as a column vector.
+template<typename T, int R, int C>
+[[nodiscard]] vector<T, R> mul(const matrix<T, R, C>& lhs, const vector<T, C>& rhs)
+{
+    vector<T, R> result;
+    for (int r = 0; r < R; ++r)
+        result[r] = dot(lhs.get_row(r), rhs);
+    return result;
+}
+
+/// Multiply vector and matrix. Vector is treated as a row vector.
+template<typename T, int R, int C>
+[[nodiscard]] vector<T, C> mul(const vector<T, R>& lhs, const matrix<T, R, C>& rhs)
+{
+    vector<T, C> result;
+    for (int c = 0; c < C; ++c)
+        result[c] = dot(lhs, rhs.get_col(c));
+    return result;
+}
+
+/// Transform a point by a 4x4 matrix. The point is treated as a column vector with a 1 in the 4th component.
+template<typename T>
+[[nodiscard]] vector<T, 3> transform_point(const matrix<T, 4, 4>& m, const vector<T, 3>& v)
+{
+    return mul(m, vector<T, 4>(v, T(1))).xyz();
+}
+
+/// Transform a vector by a 3x3 matrix.
+template<typename T>
+[[nodiscard]] vector<T, 3> transform_vector(const matrix<T, 3, 3>& m, const vector<T, 3>& v)
+{
+    return mul(m, v);
+}
+
+/// Transform a vector by a 4x4 matrix. The vector is treated as a column vector with a 0 in the 4th component.
+template<typename T>
+[[nodiscard]] vector<T, 3> transform_vector(const matrix<T, 4, 4>& m, const vector<T, 3>& v)
+{
+    return mul(m, vector<T, 4>(v, T(0))).xyz();
+}
+
+// ----------------------------------------------------------------------------
+// Functions
+// ----------------------------------------------------------------------------
+
+/// Transpose a matrix.
+template<typename T, int R, int C>
+matrix<T, C, R> transpose(const matrix<T, R, C>& m)
+{
+    matrix<T, C, R> result;
+    for (int r = 0; r < R; ++r)
+        for (int c = 0; c < C; ++c)
+            result[c][r] = m[r][c];
+    return result;
+}
+
+/// Apply a translation to a 4x4 matrix.
+template<typename T>
+matrix<T, 4, 4> translate(const matrix<T, 4, 4>& m, const vector<T, 3>& v)
+{
+    matrix<T, 4, 4> result(m);
+    result.set_col(3, m.get_col(0) * v.x + m.get_col(1) * v.y + m.get_col(2) * v.z + m.get_col(3));
+    return result;
+}
+
+/// Apply a rotation around an axis to a 4x4 matrix.
+template<typename T>
+matrix<T, 4, 4> rotate(const matrix<T, 4, 4>& m, T angle, const vector<T, 3>& axis_)
+{
+    T a = angle;
+    T c = cos(a);
+    T s = sin(a);
+
+    vector<T, 3> axis(normalize(axis_));
+    vector<T, 3> temp((T(1) - c) * axis);
+
+    matrix<T, 4, 4> rotate;
+    rotate[0][0] = c + temp[0] * axis[0];
+    rotate[0][1] = temp[1] * axis[0] - s * axis[2];
+    rotate[0][2] = temp[2] * axis[0] + s * axis[1];
+
+    rotate[1][0] = temp[0] * axis[1] + s * axis[2];
+    rotate[1][1] = c + temp[1] * axis[1];
+    rotate[1][2] = temp[2] * axis[1] - s * axis[0];
+
+    rotate[2][0] = temp[0] * axis[2] - s * axis[1];
+    rotate[2][1] = temp[1] * axis[2] + s * axis[0];
+    rotate[2][2] = c + temp[2] * axis[2];
+
+    matrix<T, 4, 4> result;
+    result.set_col(0, m.get_col(0) * rotate[0][0] + m.get_col(1) * rotate[1][0] + m.get_col(2) * rotate[2][0]);
+    result.set_col(1, m.get_col(0) * rotate[0][1] + m.get_col(1) * rotate[1][1] + m.get_col(2) * rotate[2][1]);
+    result.set_col(2, m.get_col(0) * rotate[0][2] + m.get_col(1) * rotate[1][2] + m.get_col(2) * rotate[2][2]);
+    result.set_col(3, m.get_col(3));
+
+    return result;
+}
+
+/// Apply a scale to a 4x4 matrix.
+template<typename T>
+matrix<T, 4, 4> scale(const matrix<T, 4, 4>& m, const vector<T, 3>& v)
+{
+    matrix<T, 4, 4> result;
+    result.set_col(0, m.get_col(0) * v[0]);
+    result.set_col(1, m.get_col(1) * v[1]);
+    result.set_col(2, m.get_col(2) * v[2]);
+    result.set_col(3, m.get_col(3));
+    return result;
+}
+
+/// Compute determinant of a 2x2 matrix.
+template<typename T>
+[[nodiscard]] inline T determinant(const matrix<T, 2, 2>& m)
+{
+    return m[0][0] * m[1][1] - m[1][0] * m[0][1];
+}
+
+/// Compute determinant of a 3x3 matrix.
+template<typename T>
+[[nodiscard]] inline T determinant(const matrix<T, 3, 3>& m)
+{
+    T a = m[0][0] * (m[1][1] * m[2][2] - m[2][1] * m[1][2]);
+    T b = m[1][0] * (m[0][1] * m[2][2] - m[2][1] * m[0][2]);
+    T c = m[2][0] * (m[0][1] * m[1][2] - m[1][1] * m[0][2]);
+    return a - b + c;
+}
+
+/// Compute determinant of a 4x4 matrix.
+template<typename T>
+[[nodiscard]] inline T determinant(const matrix<T, 4, 4>& m)
+{
+    T sub_factor_00 = m[2][2] * m[3][3] - m[3][2] * m[2][3];
+    T sub_factor_01 = m[2][1] * m[3][3] - m[3][1] * m[2][3];
+    T sub_factor_02 = m[2][1] * m[3][2] - m[3][1] * m[2][2];
+    T sub_factor_03 = m[2][0] * m[3][3] - m[3][0] * m[2][3];
+    T sub_factor_04 = m[2][0] * m[3][2] - m[3][0] * m[2][2];
+    T sub_factor_05 = m[2][0] * m[3][1] - m[3][0] * m[2][1];
+
+    vector<T, 4> detCof(
+        +(m[1][1] * sub_factor_00 - m[1][2] * sub_factor_01 + m[1][3] * sub_factor_02), //
+        -(m[1][0] * sub_factor_00 - m[1][2] * sub_factor_03 + m[1][3] * sub_factor_04), //
+        +(m[1][0] * sub_factor_01 - m[1][1] * sub_factor_03 + m[1][3] * sub_factor_05), //
+        -(m[1][0] * sub_factor_02 - m[1][1] * sub_factor_04 + m[1][2] * sub_factor_05)  //
+    );
+
+    return m[0][0] * detCof[0] + m[0][1] * detCof[1] + m[0][2] * detCof[2] + m[0][3] * detCof[3];
+}
+
+/// Compute inverse of a 2x2 matrix.
+template<typename T>
+[[nodiscard]] inline matrix<T, 2, 2> inverse(const matrix<T, 2, 2>& m)
+{
+    T one_over_det = T(1) / determinant(m);
+    return matrix<T, 2, 2>{
+        +m[1][1] * one_over_det,
+        -m[0][1] * one_over_det, // row 0
+        -m[1][0] * one_over_det,
+        +m[0][0] * one_over_det // row 1
+    };
+}
+
+/// Compute inverse of a 3x3 matrix.
+template<typename T>
+[[nodiscard]] inline matrix<T, 3, 3> inverse(const matrix<T, 3, 3>& m)
+{
+    T one_over_det = T(1) / determinant(m);
+
+    matrix<T, 3, 3> result;
+    result[0][0] = +(m[1][1] * m[2][2] - m[1][2] * m[2][1]) * one_over_det;
+    result[0][1] = -(m[0][1] * m[2][2] - m[0][2] * m[2][1]) * one_over_det;
+    result[0][2] = +(m[0][1] * m[1][2] - m[0][2] * m[1][1]) * one_over_det;
+    result[1][0] = -(m[1][0] * m[2][2] - m[1][2] * m[2][0]) * one_over_det;
+    result[1][1] = +(m[0][0] * m[2][2] - m[0][2] * m[2][0]) * one_over_det;
+    result[1][2] = -(m[0][0] * m[1][2] - m[0][2] * m[1][0]) * one_over_det;
+    result[2][0] = +(m[1][0] * m[2][1] - m[1][1] * m[2][0]) * one_over_det;
+    result[2][1] = -(m[0][0] * m[2][1] - m[0][1] * m[2][0]) * one_over_det;
+    result[2][2] = +(m[0][0] * m[1][1] - m[0][1] * m[1][0]) * one_over_det;
+    return result;
+}
+
+/// Compute inverse of a 4x4 matrix.
+template<typename T>
+[[nodiscard]] inline matrix<T, 4, 4> inverse(const matrix<T, 4, 4>& m)
+{
+    T c00 = m[2][2] * m[3][3] - m[2][3] * m[3][2];
+    T c02 = m[2][1] * m[3][3] - m[2][3] * m[3][1];
+    T c03 = m[2][1] * m[3][2] - m[2][2] * m[3][1];
+
+    T c04 = m[1][2] * m[3][3] - m[1][3] * m[3][2];
+    T c06 = m[1][1] * m[3][3] - m[1][3] * m[3][1];
+    T c07 = m[1][1] * m[3][2] - m[1][2] * m[3][1];
+
+    T c08 = m[1][2] * m[2][3] - m[1][3] * m[2][2];
+    T c10 = m[1][1] * m[2][3] - m[1][3] * m[2][1];
+    T c11 = m[1][1] * m[2][2] - m[1][2] * m[2][1];
+
+    T c12 = m[0][2] * m[3][3] - m[0][3] * m[3][2];
+    T c14 = m[0][1] * m[3][3] - m[0][3] * m[3][1];
+    T c15 = m[0][1] * m[3][2] - m[0][2] * m[3][1];
+
+    T c16 = m[0][2] * m[2][3] - m[0][3] * m[2][2];
+    T c18 = m[0][1] * m[2][3] - m[0][3] * m[2][1];
+    T c19 = m[0][1] * m[2][2] - m[0][2] * m[2][1];
+
+    T c20 = m[0][2] * m[1][3] - m[0][3] * m[1][2];
+    T c22 = m[0][1] * m[1][3] - m[0][3] * m[1][1];
+    T c23 = m[0][1] * m[1][2] - m[0][2] * m[1][1];
+
+    vector<T, 4> fac0(c00, c00, c02, c03);
+    vector<T, 4> fac1(c04, c04, c06, c07);
+    vector<T, 4> fac2(c08, c08, c10, c11);
+    vector<T, 4> fac3(c12, c12, c14, c15);
+    vector<T, 4> fac4(c16, c16, c18, c19);
+    vector<T, 4> fac5(c20, c20, c22, c23);
+
+    vector<T, 4> vec0(m[0][1], m[0][0], m[0][0], m[0][0]);
+    vector<T, 4> vec1(m[1][1], m[1][0], m[1][0], m[1][0]);
+    vector<T, 4> vec2(m[2][1], m[2][0], m[2][0], m[2][0]);
+    vector<T, 4> vec3(m[3][1], m[3][0], m[3][0], m[3][0]);
+
+    vector<T, 4> inv0(vec1 * fac0 - vec2 * fac1 + vec3 * fac2);
+    vector<T, 4> inv1(vec0 * fac0 - vec2 * fac3 + vec3 * fac4);
+    vector<T, 4> inv2(vec0 * fac1 - vec1 * fac3 + vec3 * fac5);
+    vector<T, 4> inv3(vec0 * fac2 - vec1 * fac4 + vec2 * fac5);
+
+    vector<T, 4> sign_a(+1, -1, +1, -1);
+    vector<T, 4> sign_b(-1, +1, -1, +1);
+    matrix<T, 4, 4> inverse = matrix_from_columns(inv0 * sign_a, inv1 * sign_b, inv2 * sign_a, inv3 * sign_b);
+
+    vector<T, 4> row0(inverse[0][0], inverse[0][1], inverse[0][2], inverse[0][3]);
+
+    vector<T, 4> dot0(m.get_col(0) * row0);
+    T dot1 = (dot0.x + dot0.y) + (dot0.z + dot0.w);
+
+    T one_over_det = T(1) / dot1;
+
+    return inverse * one_over_det;
+}
+
+/// Compute the (X * Y * Z) euler angles of a 4x4 matrix.
+template<typename T>
+void extract_euler_angle_xyz(const matrix<T, 4, 4>& m, float& angle_x, float& angle_y, float& angle_z)
+{
+    T t1 = atan2(m[1][2], m[2][2]);
+    T c2 = sqrt(m[0][0] * m[0][0] + m[0][1] * m[0][1]);
+    T t2 = atan2(-m[0][2], c2);
+    T s1 = sin(t1);
+    T c1 = cos(t1);
+    T t3 = atan2(s1 * m[2][0] - c1 * m[1][0], c1 * m[1][1] - s1 * m[2][1]);
+    angle_x = -t1;
+    angle_y = -t2;
+    angle_z = -t3;
+}
+
+/// Decomposes a model matrix into translation, rotation and scale components.
+template<typename T>
+inline bool decompose(
+    const matrix<T, 4, 4>& model_matrix,
+    vector<T, 3>& scale,
+    quat<T>& orientation,
+    vector<T, 3>& translation,
+    vector<T, 3>& skew,
+    vector<T, 4>& perspective
+)
+{
+    // See https://caff.de/posts/4X4-matrix-decomposition/decomposition.pdf
+
+    const T eps = std::numeric_limits<T>::epsilon();
+
+    matrix<T, 4, 4> local_matrix(model_matrix);
+
+    // Abort if zero matrix.
+    if (abs(local_matrix[3][3]) < eps)
+        return false;
+
+    // Normalize the matrix.
+    for (int i = 0; i < 4; ++i)
+        for (int j = 0; j < 4; ++j)
+            local_matrix[i][j] /= local_matrix[3][3];
+
+    // perspective_matrix is used to solve for perspective, but it also provides
+    // an easy way to test for singularity of the upper 3x3 component.
+    matrix<T, 4, 4> perspective_matrix(local_matrix);
+    perspective_matrix[3] = vector<T, 4>(0, 0, 0, 1);
+    if (abs(determinant(perspective_matrix)) < eps)
+        return false;
+
+    // First, isolate perspective. This is the messiest.
+    if (abs(local_matrix[3][0]) >= eps || abs(local_matrix[3][1]) >= eps || abs(local_matrix[3][2]) >= eps) {
+        // right_hand_side is the right hand side of the equation.
+        vector<T, 4> right_hand_side = local_matrix[3];
+
+        // Solve the equation by inverting perspective_matrix and multiplying
+        // right_hand_side by the inverse.
+        // (This is the easiest way, not necessarily the best.)
+        matrix<T, 4, 4> inverse_perspective_matrix = inverse(perspective_matrix);
+        matrix<T, 4, 4> transposed_inverse_perspective_matrix = transpose(inverse_perspective_matrix);
+
+        perspective = mul(transposed_inverse_perspective_matrix, right_hand_side);
+
+        // Clear the perspective partition.
+        local_matrix[3] = vector<T, 4>(0, 0, 0, 1);
+    } else {
+        // No perspective.
+        perspective = vector<T, 4>(0, 0, 0, 1);
+    }
+
+    // Next take care of translation (easy).
+    translation = local_matrix.get_col(3).xyz();
+    local_matrix.set_row(3, vector<T, 4>(0, 0, 0, 1));
+
+    vector<T, 3> row[3];
+
+    // Now get scale and shear.
+    for (int i = 0; i < 3; ++i)
+        for (int j = 0; j < 3; ++j)
+            row[i][j] = local_matrix[j][i];
+
+    // Compute X scale factor and normalize first row.
+    scale.x = length(row[0]);
+    row[0] = normalize(row[0]);
+
+    // Compute XY shear factor and make 2nd row orthogonal to 1st.
+    skew.z = dot(row[0], row[1]);
+    row[1] = row[1] - skew.z * row[0];
+
+    // Now, compute Y scale and normalize 2nd row.
+    scale.y = length(row[1]);
+    row[1] = normalize(row[1]);
+    skew.z /= scale.y;
+
+    // Compute XZ and YZ shears, orthogonalize 3rd row.
+    skew.y = dot(row[0], row[2]);
+    row[2] = row[2] - skew.y * row[0];
+    skew.x = dot(row[1], row[2]);
+    row[2] = row[2] - skew.x * row[1];
+
+    // Next, get Z scale and normalize 3rd row.
+    scale.z = length(row[2]);
+    row[2] = normalize(row[2]);
+    skew.y /= scale.z;
+    skew.x /= scale.z;
+
+    // At this point, the matrix (in rows[]) is orthonormal.
+    // Check for a coordinate system flip. If the determinant
+    // is -1, then negate the matrix and the scaling factors.
+    if (dot(row[0], cross(row[1], row[2])) < T(0)) {
+        scale *= T(-1);
+        for (int i = 0; i < 3; i++)
+            row[i] *= T(-1);
+    }
+
+    // Now, get the rotations out, as described in the gem.
+    int i, j, k = 0;
+    T root, trace = row[0].x + row[1].y + row[2].z;
+    if (trace > T(0)) {
+        root = sqrt(trace + T(1));
+        orientation.w = T(0.5) * root;
+        root = T(0.5) / root;
+        orientation.x = root * (row[1].z - row[2].y);
+        orientation.y = root * (row[2].x - row[0].z);
+        orientation.z = root * (row[0].y - row[1].x);
+    } // end if > 0
+    else {
+        static int next[3] = {1, 2, 0};
+        i = 0;
+        if (row[1].y > row[0].x)
+            i = 1;
+        if (row[2].z > row[i][i])
+            i = 2;
+        j = next[i];
+        k = next[j];
+
+        root = sqrt(row[i][i] - row[j][j] - row[k][k] + T(1));
+
+        orientation[i] = T(0.5) * root;
+        root = T(0.5) / root;
+        orientation[j] = root * (row[i][j] + row[j][i]);
+        orientation[k] = root * (row[i][k] + row[k][i]);
+        orientation.w = root * (row[j][k] - row[k][j]);
+    } // end if <= 0
+
+    return true;
+}
+
+// ----------------------------------------------------------------------------
+// Construction
+// ----------------------------------------------------------------------------
+
+/// Creates a matrix from coefficients in row-major order.
+template<typename T, int R, int C>
+[[nodiscard]] inline matrix<T, R, C> matrix_from_coefficients(const T* coeffs)
+{
+    matrix<T, R, C> m;
+    std::memcpy(&m, coeffs, sizeof(T) * R * C);
+    return m;
+}
+
+/// Creates a matrix from column vectors.
+template<typename T, int R>
+[[nodiscard]] inline matrix<T, R, 1> matrix_from_columns(const vector<T, R>& col0)
+{
+    matrix<T, R, 1> m;
+    m.set_col(0, col0);
+    return m;
+}
+
+/// Creates a matrix from column vectors.
+template<typename T, int R>
+[[nodiscard]] inline matrix<T, R, 2> matrix_from_columns(const vector<T, R>& col0, const vector<T, R>& col1)
+{
+    matrix<T, R, 2> m;
+    m.set_col(0, col0);
+    m.set_col(1, col1);
+    return m;
+}
+
+/// Creates a matrix from column vectors.
+template<typename T, int R>
+[[nodiscard]] inline matrix<T, R, 3>
+matrix_from_columns(const vector<T, R>& col0, const vector<T, R>& col1, const vector<T, R>& col2)
+{
+    matrix<T, R, 3> m;
+    m.set_col(0, col0);
+    m.set_col(1, col1);
+    m.set_col(2, col2);
+    return m;
+}
+
+/// Creates a matrix from column vectors.
+template<typename T, int R>
+[[nodiscard]] inline matrix<T, R, 4> matrix_from_columns(
+    const vector<T, R>& col0,
+    const vector<T, R>& col1,
+    const vector<T, R>& col2,
+    const vector<T, R>& col3
+)
+{
+    matrix<T, R, 4> m;
+    m.set_col(0, col0);
+    m.set_col(1, col1);
+    m.set_col(2, col2);
+    m.set_col(3, col3);
+    return m;
+}
+
+/// Creates a square matrix from a diagonal vector.
+template<typename T, int N>
+[[nodiscard]] inline matrix<T, N, N> matrix_from_diagonal(const vector<T, N>& diag)
+{
+    matrix<T, N, N> m = matrix<T, N, N>::zeros();
+    for (int i = 0; i < N; i++)
+        m[i][i] = diag[i];
+    return m;
+}
+
+/// Creates a right-handed perspective projection matrix. Depth is mapped to [0, 1].
+template<floating_point T>
+[[nodiscard]] inline matrix<T, 4, 4> perspective(T fovy, T aspect, T z_near, T z_far)
+{
+    SGL_ASSERT(abs(aspect - std::numeric_limits<T>::epsilon()) > T(0));
+
+    T tan_half_fovy = tan(fovy / T(2));
+
+    matrix<T, 4, 4> m = matrix<T, 4, 4>::zeros();
+    m[0][0] = T(1) / (aspect * tan_half_fovy);
+    m[1][1] = T(1) / (tan_half_fovy);
+    m[2][2] = z_far / (z_near - z_far);
+    m[3][2] = -T(1);
+    m[2][3] = -(z_far * z_near) / (z_far - z_near);
+    return m;
+}
+
+/// Creates a right-handed orthographic projection matrix. Depth is mapped to [0, 1].
+template<floating_point T>
+[[nodiscard]] inline matrix<T, 4, 4> ortho(T left, T right, T bottom, T top, T z_near, T z_far)
+{
+    matrix<T, 4, 4> m = matrix<T, 4, 4>::identity();
+    m[0][0] = T(2) / (right - left);
+    m[1][1] = T(2) / (top - bottom);
+    m[2][2] = -T(1) / (z_far - z_near);
+    m[0][3] = -(right + left) / (right - left);
+    m[1][3] = -(top + bottom) / (top - bottom);
+    m[2][3] = -z_near / (z_far - z_near);
+    return m;
+}
+
+/// Creates a translation matrix.
+template<floating_point T>
+[[nodiscard]] inline matrix<T, 4, 4> matrix_from_translation(const vector<T, 3>& v)
+{
+    return translate(matrix<T, 4, 4>::identity(), v);
+}
+
+/// Creates a rotation matrix from an angle and an axis.
+template<floating_point T>
+[[nodiscard]] inline matrix<T, 4, 4> matrix_from_rotation(T angle, const vector<T, 3>& axis)
+{
+    return rotate(matrix<T, 4, 4>::identity(), angle, axis);
+}
+
+/// Creates a rotation matrix around the X-axis.
+template<floating_point T>
+[[nodiscard]] inline matrix<T, 4, 4> matrix_from_rotation_x(T angle)
+{
+    T c = cos(angle);
+    T s = sin(angle);
+
+    // clang-format off
+    return matrix<T, 4, 4>{
+        T(1),   T(0),   T(0),   T(0),   // row 0
+        T(0),   c,      -s,     T(0),   // row 1
+        T(0),   s,      c,      T(0),   // row 2
+        T(0),   T(0),   T(0),   T(1)    // row 3
+    };
+    // clang-format on
+}
+
+/// Creates a rotation matrix around the Y-axis.
+template<floating_point T>
+[[nodiscard]] inline matrix<T, 4, 4> matrix_from_rotation_y(T angle)
+{
+    T c = cos(angle);
+    T s = sin(angle);
+
+    // clang-format off
+    return matrix<T, 4, 4>{
+        c,      T(0),   s,      T(0),   // row 0
+        T(0),   T(1),   T(0),   T(0),   // row 1
+        -s,     T(0),   c,      T(0),   // row 2
+        T(0),   T(0),   T(0),   T(1)    // row 3
+    };
+    // clang-format on
+}
+
+/// Creates a rotation matrix around the Z-axis.
+template<floating_point T>
+[[nodiscard]] inline matrix<T, 4, 4> matrix_from_rotation_z(T angle)
+{
+    T c = cos(angle);
+    T s = sin(angle);
+
+    // clang-format off
+    return matrix<T, 4, 4>{
+        c,      -s,     T(0),   T(0),   // row 0
+        s,      c,      T(0),   T(0),   // row 1
+        T(0),   T(0),   T(1),   T(0),   // row 2
+        T(0),   T(0),   T(0),   T(1)    // row 3
+    };
+    // clang-format on
+}
+
+/// Creates a rotation matrix (X * Y * Z).
+template<floating_point T>
+[[nodiscard]] inline matrix<T, 4, 4> matrix_from_rotation_xyz(T angle_x, T angle_y, T angle_z)
+{
+    T c1 = cos(-angle_x);
+    T c2 = cos(-angle_y);
+    T c3 = cos(-angle_z);
+    T s1 = sin(-angle_x);
+    T s2 = sin(-angle_y);
+    T s3 = sin(-angle_z);
+
+    matrix<T, 4, 4> m;
+    m[0][0] = c2 * c3;
+    m[0][1] = c2 * s3;
+    m[0][2] = -s2;
+    m[0][3] = T(0);
+
+    m[1][0] = -c1 * s3 + s1 * s2 * c3;
+    m[1][1] = c1 * c3 + s1 * s2 * s3;
+    m[1][2] = s1 * c2;
+    m[1][3] = T(0);
+
+    m[2][0] = s1 * s3 + c1 * s2 * c3;
+    m[2][1] = -s1 * c3 + c1 * s2 * s3;
+    m[2][2] = c1 * c2;
+    m[2][3] = T(0);
+
+    m[3][0] = T(0);
+    m[3][1] = T(0);
+    m[3][2] = T(0);
+    m[3][3] = T(1);
+
+    return m;
+}
+
+template<floating_point T>
+[[nodiscard]] inline matrix<T, 4, 4> matrix_from_rotation_xyz(const vector<T, 3>& angles)
+{
+    return matrix_from_rotation_xyz(angles.x, angles.y, angles.z);
+}
+
+/// Creates a scaling matrix.
+template<floating_point T>
+[[nodiscard]] inline matrix<T, 4, 4> matrix_from_scaling(const vector<T, 3>& v)
+{
+    return scale(matrix<T, 4, 4>::identity(), v);
+}
+
+/**
+ * Build a look-at matrix.
+ * If right handed, forward direction is mapped onto -Z axis.
+ * If left handed, forward direction is mapped onto +Z axis.
+ * \param eye Eye position
+ * \param center Center position
+ * \param up Up vector
+ * \param handedness Coordinate system handedness.
+ */
+template<floating_point T>
+[[nodiscard]] inline matrix<T, 4, 4> matrix_from_look_at(
+    const vector<T, 3>& eye,
+    const vector<T, 3>& center,
+    const vector<T, 3>& up,
+    Handedness handedness = Handedness::right_handed
+)
+{
+    vector<T, 3> f(handedness == Handedness::right_handed ? normalize(eye - center) : normalize(center - eye));
+    vector<T, 3> r(normalize(cross(up, f)));
+    vector<T, 3> u(cross(f, r));
+
+    matrix<T, 4, 4> result = matrix<T, 4, 4>::identity();
+    result[0][0] = r.x;
+    result[0][1] = r.y;
+    result[0][2] = r.z;
+    result[1][0] = u.x;
+    result[1][1] = u.y;
+    result[1][2] = u.z;
+    result[2][0] = f.x;
+    result[2][1] = f.y;
+    result[2][2] = f.z;
+    result[0][3] = -dot(r, eye);
+    result[1][3] = -dot(u, eye);
+    result[2][3] = -dot(f, eye);
+
+    return result;
+}
+
+template<floating_point T>
+[[nodiscard]] inline matrix<T, 3, 3> matrix_from_quat(const quat<T>& q)
+{
+    matrix<T, 3, 3> m;
+    T qxx(q.x * q.x);
+    T qyy(q.y * q.y);
+    T qzz(q.z * q.z);
+    T qxz(q.x * q.z);
+    T qxy(q.x * q.y);
+    T qyz(q.y * q.z);
+    T qwx(q.w * q.x);
+    T qwy(q.w * q.y);
+    T qwz(q.w * q.z);
+
+    m[0][0] = T(1) - T(2) * (qyy + qzz);
+    m[0][1] = T(2) * (qxy - qwz);
+    m[0][2] = T(2) * (qxz + qwy);
+
+    m[1][0] = T(2) * (qxy + qwz);
+    m[1][1] = T(1) - T(2) * (qxx + qzz);
+    m[1][2] = T(2) * (qyz - qwx);
+
+    m[2][0] = T(2) * (qxz - qwy);
+    m[2][1] = T(2) * (qyz + qwx);
+    m[2][2] = T(1) - T(2) * (qxx + qyy);
+
+    return m;
+}
+
+template<typename T, int R, int C>
+[[nodiscard]] std::string to_string(const matrix<T, R, C>& m)
+{
+    return ::fmt::format("{}", m);
+}
+
+template<int R, int C, typename T>
+bool lex_lt(const matrix<T, R, C>& lhs, const matrix<T, R, C>& rhs)
+{
+    for (int r = 0; r < R; ++r) {
+        for (int c = 0; c < C; ++c) {
+            if (lhs[r][c] != rhs[r][c])
+                return lhs[r][c] < rhs[r][c];
+        }
+    }
+    return false;
+}
+} // namespace sgl::math
+
+template<typename T, int R, int C>
+struct fmt::formatter<sgl::math::matrix<T, R, C>> : formatter<typename sgl::math::matrix<T, R, C>::row_type> {
+    using row_type = typename sgl::math::matrix<T, R, C>::row_type;
+
+    template<typename FormatContext>
+    auto format(const sgl::math::matrix<T, R, C>& matrix, FormatContext& ctx) const
+    {
+        auto out = ctx.out();
+        for (int r = 0; r < R; ++r) {
+            out = ::fmt::format_to(out, "{}", (r == 0) ? "{" : ", ");
+            out = formatter<row_type>::format(matrix.get_row(r), ctx);
+        }
+        out = fmt::format_to(out, "}}");
+        return out;
+    }
+};