diff options
Diffstat (limited to 'core/math')
-rw-r--r-- | core/math/a_star_grid_2d.cpp | 4 | ||||
-rw-r--r-- | core/math/aabb.h | 25 | ||||
-rw-r--r-- | core/math/basis.h | 2 | ||||
-rw-r--r-- | core/math/color.cpp | 25 | ||||
-rw-r--r-- | core/math/color.h | 69 | ||||
-rw-r--r-- | core/math/color_names.inc | 1 | ||||
-rw-r--r-- | core/math/face3.h | 2 | ||||
-rw-r--r-- | core/math/geometry_2d.h | 2 | ||||
-rw-r--r-- | core/math/math_funcs.h | 12 | ||||
-rw-r--r-- | core/math/plane.h | 2 | ||||
-rw-r--r-- | core/math/projection.h | 2 | ||||
-rw-r--r-- | core/math/quaternion.h | 2 | ||||
-rw-r--r-- | core/math/rect2.h | 18 | ||||
-rw-r--r-- | core/math/rect2i.h | 2 | ||||
-rw-r--r-- | core/math/transform_2d.h | 2 | ||||
-rw-r--r-- | core/math/transform_3d.h | 2 | ||||
-rw-r--r-- | core/math/transform_interpolator.cpp | 354 | ||||
-rw-r--r-- | core/math/transform_interpolator.h | 51 | ||||
-rw-r--r-- | core/math/vector2.h | 2 | ||||
-rw-r--r-- | core/math/vector2i.h | 2 | ||||
-rw-r--r-- | core/math/vector3.h | 2 | ||||
-rw-r--r-- | core/math/vector3i.h | 2 | ||||
-rw-r--r-- | core/math/vector4.h | 2 | ||||
-rw-r--r-- | core/math/vector4i.h | 2 |
24 files changed, 489 insertions, 100 deletions
diff --git a/core/math/a_star_grid_2d.cpp b/core/math/a_star_grid_2d.cpp index 82bf13cc1f..ad69748452 100644 --- a/core/math/a_star_grid_2d.cpp +++ b/core/math/a_star_grid_2d.cpp @@ -122,6 +122,10 @@ AStarGrid2D::CellShape AStarGrid2D::get_cell_shape() const { } void AStarGrid2D::update() { + if (!dirty) { + return; + } + points.clear(); const int32_t end_x = region.get_end().x; diff --git a/core/math/aabb.h b/core/math/aabb.h index 9a74266ff7..7a5581b5d4 100644 --- a/core/math/aabb.h +++ b/core/math/aabb.h @@ -41,7 +41,7 @@ class Variant; -struct _NO_DISCARD_ AABB { +struct [[nodiscard]] AABB { Vector3 position; Vector3 size; @@ -85,7 +85,7 @@ struct _NO_DISCARD_ AABB { bool intersects_plane(const Plane &p_plane) const; _FORCE_INLINE_ bool has_point(const Vector3 &p_point) const; - _FORCE_INLINE_ Vector3 get_support(const Vector3 &p_normal) const; + _FORCE_INLINE_ Vector3 get_support(const Vector3 &p_direction) const; Vector3 get_longest_axis() const; int get_longest_axis_index() const; @@ -212,15 +212,18 @@ inline bool AABB::encloses(const AABB &p_aabb) const { (src_max.z >= dst_max.z)); } -Vector3 AABB::get_support(const Vector3 &p_normal) const { - Vector3 half_extents = size * 0.5f; - Vector3 ofs = position + half_extents; - - return Vector3( - (p_normal.x > 0) ? half_extents.x : -half_extents.x, - (p_normal.y > 0) ? half_extents.y : -half_extents.y, - (p_normal.z > 0) ? half_extents.z : -half_extents.z) + - ofs; +Vector3 AABB::get_support(const Vector3 &p_direction) const { + Vector3 support = position; + if (p_direction.x > 0.0f) { + support.x += size.x; + } + if (p_direction.y > 0.0f) { + support.y += size.y; + } + if (p_direction.z > 0.0f) { + support.z += size.z; + } + return support; } Vector3 AABB::get_endpoint(int p_point) const { diff --git a/core/math/basis.h b/core/math/basis.h index 918cbc18d4..5c1a5fbdda 100644 --- a/core/math/basis.h +++ b/core/math/basis.h @@ -34,7 +34,7 @@ #include "core/math/quaternion.h" #include "core/math/vector3.h" -struct _NO_DISCARD_ Basis { +struct [[nodiscard]] Basis { Vector3 rows[3] = { Vector3(1, 0, 0), Vector3(0, 1, 0), diff --git a/core/math/color.cpp b/core/math/color.cpp index d36306d968..1638acd74d 100644 --- a/core/math/color.cpp +++ b/core/math/color.cpp @@ -33,7 +33,7 @@ #include "color_names.inc" #include "core/math/math_funcs.h" #include "core/string/ustring.h" -#include "core/templates/rb_map.h" +#include "core/templates/hash_map.h" #include "thirdparty/misc/ok_color.h" @@ -414,7 +414,7 @@ Color Color::named(const String &p_name, const Color &p_default) { int Color::find_named_color(const String &p_name) { String name = p_name; - // Normalize name + // Normalize name. name = name.replace(" ", ""); name = name.replace("-", ""); name = name.replace("_", ""); @@ -422,23 +422,24 @@ int Color::find_named_color(const String &p_name) { name = name.replace(".", ""); name = name.to_upper(); - int idx = 0; - while (named_colors[idx].name != nullptr) { - if (name == String(named_colors[idx].name).replace("_", "")) { - return idx; + static HashMap<String, int> named_colors_hashmap; + if (unlikely(named_colors_hashmap.is_empty())) { + const int named_color_count = get_named_color_count(); + for (int i = 0; i < named_color_count; i++) { + named_colors_hashmap[String(named_colors[i].name).replace("_", "")] = i; } - idx++; + } + + const HashMap<String, int>::ConstIterator E = named_colors_hashmap.find(name); + if (E) { + return E->value; } return -1; } int Color::get_named_color_count() { - int idx = 0; - while (named_colors[idx].name != nullptr) { - idx++; - } - return idx; + return sizeof(named_colors) / sizeof(NamedColor); } String Color::get_named_color_name(int p_idx) { diff --git a/core/math/color.h b/core/math/color.h index 65d7377c1c..70fad78acb 100644 --- a/core/math/color.h +++ b/core/math/color.h @@ -35,7 +35,7 @@ class String; -struct _NO_DISCARD_ Color { +struct [[nodiscard]] Color { union { struct { float r; @@ -129,33 +129,46 @@ struct _NO_DISCARD_ Color { } _FORCE_INLINE_ uint32_t to_rgbe9995() const { - const float pow2to9 = 512.0f; - const float B = 15.0f; - const float N = 9.0f; - - float sharedexp = 65408.000f; // Result of: ((pow2to9 - 1.0f) / pow2to9) * powf(2.0f, 31.0f - 15.0f) - - float cRed = MAX(0.0f, MIN(sharedexp, r)); - float cGreen = MAX(0.0f, MIN(sharedexp, g)); - float cBlue = MAX(0.0f, MIN(sharedexp, b)); - - float cMax = MAX(cRed, MAX(cGreen, cBlue)); - - float expp = MAX(-B - 1.0f, floor(Math::log(cMax) / (real_t)Math_LN2)) + 1.0f + B; - - float sMax = (float)floor((cMax / Math::pow(2.0f, expp - B - N)) + 0.5f); - - float exps = expp + 1.0f; - - if (0.0f <= sMax && sMax < pow2to9) { - exps = expp; - } - - float sRed = Math::floor((cRed / pow(2.0f, exps - B - N)) + 0.5f); - float sGreen = Math::floor((cGreen / pow(2.0f, exps - B - N)) + 0.5f); - float sBlue = Math::floor((cBlue / pow(2.0f, exps - B - N)) + 0.5f); - - return (uint32_t(Math::fast_ftoi(sRed)) & 0x1FF) | ((uint32_t(Math::fast_ftoi(sGreen)) & 0x1FF) << 9) | ((uint32_t(Math::fast_ftoi(sBlue)) & 0x1FF) << 18) | ((uint32_t(Math::fast_ftoi(exps)) & 0x1F) << 27); + // https://github.com/microsoft/DirectX-Graphics-Samples/blob/v10.0.19041.0/MiniEngine/Core/Color.cpp + static const float kMaxVal = float(0x1FF << 7); + static const float kMinVal = float(1.f / (1 << 16)); + + // Clamp RGB to [0, 1.FF*2^16] + const float _r = CLAMP(r, 0.0f, kMaxVal); + const float _g = CLAMP(g, 0.0f, kMaxVal); + const float _b = CLAMP(b, 0.0f, kMaxVal); + + // Compute the maximum channel, no less than 1.0*2^-15 + const float MaxChannel = MAX(MAX(_r, _g), MAX(_b, kMinVal)); + + // Take the exponent of the maximum channel (rounding up the 9th bit) and + // add 15 to it. When added to the channels, it causes the implicit '1.0' + // bit and the first 8 mantissa bits to be shifted down to the low 9 bits + // of the mantissa, rounding the truncated bits. + union { + float f; + int32_t i; + } R, G, B, E; + + E.f = MaxChannel; + E.i += 0x07804000; // Add 15 to the exponent and 0x4000 to the mantissa + E.i &= 0x7F800000; // Zero the mantissa + + // This shifts the 9-bit values we need into the lowest bits, rounding as + // needed. Note that if the channel has a smaller exponent than the max + // channel, it will shift even more. This is intentional. + R.f = _r + E.f; + G.f = _g + E.f; + B.f = _b + E.f; + + // Convert the Bias to the correct exponent in the upper 5 bits. + E.i <<= 4; + E.i += 0x10000000; + + // Combine the fields. RGB floats have unwanted data in the upper 9 + // bits. Only red needs to mask them off because green and blue shift + // it out to the left. + return E.i | (B.i << 18) | (G.i << 9) | (R.i & 511); } _FORCE_INLINE_ Color blend(const Color &p_over) const { diff --git a/core/math/color_names.inc b/core/math/color_names.inc index eaa1b1087f..6c0d2a4bfd 100644 --- a/core/math/color_names.inc +++ b/core/math/color_names.inc @@ -189,5 +189,4 @@ static NamedColor named_colors[] = { { "WHITE_SMOKE", Color::hex(0xF5F5F5FF) }, { "YELLOW", Color::hex(0xFFFF00FF) }, { "YELLOW_GREEN", Color::hex(0x9ACD32FF) }, - { nullptr, Color() }, }; diff --git a/core/math/face3.h b/core/math/face3.h index 3dd47d0226..519dcb6414 100644 --- a/core/math/face3.h +++ b/core/math/face3.h @@ -36,7 +36,7 @@ #include "core/math/transform_3d.h" #include "core/math/vector3.h" -struct _NO_DISCARD_ Face3 { +struct [[nodiscard]] Face3 { enum Side { SIDE_OVER, SIDE_UNDER, diff --git a/core/math/geometry_2d.h b/core/math/geometry_2d.h index 1502b2807c..83ebdc5a84 100644 --- a/core/math/geometry_2d.h +++ b/core/math/geometry_2d.h @@ -350,6 +350,8 @@ public: return triangles; } + // Assumes cartesian coordinate system with +x to the right, +y up. + // If using screen coordinates (+x to the right, +y down) the result will need to be flipped. static bool is_polygon_clockwise(const Vector<Vector2> &p_polygon) { int c = p_polygon.size(); if (c < 3) { diff --git a/core/math/math_funcs.h b/core/math/math_funcs.h index 3060f31970..fd53ed28fd 100644 --- a/core/math/math_funcs.h +++ b/core/math/math_funcs.h @@ -447,14 +447,22 @@ public: static _ALWAYS_INLINE_ double smoothstep(double p_from, double p_to, double p_s) { if (is_equal_approx(p_from, p_to)) { - return p_from; + if (likely(p_from <= p_to)) { + return p_s <= p_from ? 0.0 : 1.0; + } else { + return p_s <= p_to ? 1.0 : 0.0; + } } double s = CLAMP((p_s - p_from) / (p_to - p_from), 0.0, 1.0); return s * s * (3.0 - 2.0 * s); } static _ALWAYS_INLINE_ float smoothstep(float p_from, float p_to, float p_s) { if (is_equal_approx(p_from, p_to)) { - return p_from; + if (likely(p_from <= p_to)) { + return p_s <= p_from ? 0.0f : 1.0f; + } else { + return p_s <= p_to ? 1.0f : 0.0f; + } } float s = CLAMP((p_s - p_from) / (p_to - p_from), 0.0f, 1.0f); return s * s * (3.0f - 2.0f * s); diff --git a/core/math/plane.h b/core/math/plane.h index 8159f25342..6529fea60a 100644 --- a/core/math/plane.h +++ b/core/math/plane.h @@ -35,7 +35,7 @@ class Variant; -struct _NO_DISCARD_ Plane { +struct [[nodiscard]] Plane { Vector3 normal; real_t d = 0; diff --git a/core/math/projection.h b/core/math/projection.h index f3ed9d7b1c..5af43561c0 100644 --- a/core/math/projection.h +++ b/core/math/projection.h @@ -43,7 +43,7 @@ struct Rect2; struct Transform3D; struct Vector2; -struct _NO_DISCARD_ Projection { +struct [[nodiscard]] Projection { enum Planes { PLANE_NEAR, PLANE_FAR, diff --git a/core/math/quaternion.h b/core/math/quaternion.h index 868a2916f5..655e55e0a2 100644 --- a/core/math/quaternion.h +++ b/core/math/quaternion.h @@ -35,7 +35,7 @@ #include "core/math/vector3.h" #include "core/string/ustring.h" -struct _NO_DISCARD_ Quaternion { +struct [[nodiscard]] Quaternion { union { struct { real_t x; diff --git a/core/math/rect2.h b/core/math/rect2.h index b4069ae86a..817923c134 100644 --- a/core/math/rect2.h +++ b/core/math/rect2.h @@ -38,7 +38,7 @@ class String; struct Rect2i; struct Transform2D; -struct _NO_DISCARD_ Rect2 { +struct [[nodiscard]] Rect2 { Point2 position; Size2 size; @@ -285,13 +285,15 @@ struct _NO_DISCARD_ Rect2 { return Rect2(position.round(), size.round()); } - Vector2 get_support(const Vector2 &p_normal) const { - Vector2 half_extents = size * 0.5f; - Vector2 ofs = position + half_extents; - return Vector2( - (p_normal.x > 0) ? -half_extents.x : half_extents.x, - (p_normal.y > 0) ? -half_extents.y : half_extents.y) + - ofs; + Vector2 get_support(const Vector2 &p_direction) const { + Vector2 support = position; + if (p_direction.x > 0.0f) { + support.x += size.x; + } + if (p_direction.y > 0.0f) { + support.y += size.y; + } + return support; } _FORCE_INLINE_ bool intersects_filled_polygon(const Vector2 *p_points, int p_point_count) const { diff --git a/core/math/rect2i.h b/core/math/rect2i.h index a1338da0bb..5f3a3d54f5 100644 --- a/core/math/rect2i.h +++ b/core/math/rect2i.h @@ -37,7 +37,7 @@ class String; struct Rect2; -struct _NO_DISCARD_ Rect2i { +struct [[nodiscard]] Rect2i { Point2i position; Size2i size; diff --git a/core/math/transform_2d.h b/core/math/transform_2d.h index 4ec2dc119c..476577508f 100644 --- a/core/math/transform_2d.h +++ b/core/math/transform_2d.h @@ -38,7 +38,7 @@ class String; -struct _NO_DISCARD_ Transform2D { +struct [[nodiscard]] Transform2D { // Warning #1: basis of Transform2D is stored differently from Basis. In terms of columns array, the basis matrix looks like "on paper": // M = (columns[0][0] columns[1][0]) // (columns[0][1] columns[1][1]) diff --git a/core/math/transform_3d.h b/core/math/transform_3d.h index 7d89b86c75..b1de233445 100644 --- a/core/math/transform_3d.h +++ b/core/math/transform_3d.h @@ -36,7 +36,7 @@ #include "core/math/plane.h" #include "core/templates/vector.h" -struct _NO_DISCARD_ Transform3D { +struct [[nodiscard]] Transform3D { Basis basis; Vector3 origin; diff --git a/core/math/transform_interpolator.cpp b/core/math/transform_interpolator.cpp index 7cfe880b5a..1cd35b3d1a 100644 --- a/core/math/transform_interpolator.cpp +++ b/core/math/transform_interpolator.cpp @@ -31,46 +31,354 @@ #include "transform_interpolator.h" #include "core/math/transform_2d.h" +#include "core/math/transform_3d.h" void TransformInterpolator::interpolate_transform_2d(const Transform2D &p_prev, const Transform2D &p_curr, Transform2D &r_result, real_t p_fraction) { - // Extract parameters. - Vector2 p1 = p_prev.get_origin(); - Vector2 p2 = p_curr.get_origin(); - // Special case for physics interpolation, if flipping, don't interpolate basis. // If the determinant polarity changes, the handedness of the coordinate system changes. if (_sign(p_prev.determinant()) != _sign(p_curr.determinant())) { r_result.columns[0] = p_curr.columns[0]; r_result.columns[1] = p_curr.columns[1]; - r_result.set_origin(p1.lerp(p2, p_fraction)); + r_result.set_origin(p_prev.get_origin().lerp(p_curr.get_origin(), p_fraction)); return; } - real_t r1 = p_prev.get_rotation(); - real_t r2 = p_curr.get_rotation(); + r_result = p_prev.interpolate_with(p_curr, p_fraction); +} + +void TransformInterpolator::interpolate_transform_3d(const Transform3D &p_prev, const Transform3D &p_curr, Transform3D &r_result, real_t p_fraction) { + r_result.origin = p_prev.origin + ((p_curr.origin - p_prev.origin) * p_fraction); + interpolate_basis(p_prev.basis, p_curr.basis, r_result.basis, p_fraction); +} + +void TransformInterpolator::interpolate_basis(const Basis &p_prev, const Basis &p_curr, Basis &r_result, real_t p_fraction) { + Method method = find_method(p_prev, p_curr); + interpolate_basis_via_method(p_prev, p_curr, r_result, p_fraction, method); +} + +void TransformInterpolator::interpolate_transform_3d_via_method(const Transform3D &p_prev, const Transform3D &p_curr, Transform3D &r_result, real_t p_fraction, Method p_method) { + r_result.origin = p_prev.origin + ((p_curr.origin - p_prev.origin) * p_fraction); + interpolate_basis_via_method(p_prev.basis, p_curr.basis, r_result.basis, p_fraction, p_method); +} + +void TransformInterpolator::interpolate_basis_via_method(const Basis &p_prev, const Basis &p_curr, Basis &r_result, real_t p_fraction, Method p_method) { + switch (p_method) { + default: { + interpolate_basis_linear(p_prev, p_curr, r_result, p_fraction); + } break; + case INTERP_SLERP: { + r_result = _basis_slerp_unchecked(p_prev, p_curr, p_fraction); + } break; + case INTERP_SCALED_SLERP: { + interpolate_basis_scaled_slerp(p_prev, p_curr, r_result, p_fraction); + } break; + } +} + +Quaternion TransformInterpolator::_basis_to_quat_unchecked(const Basis &p_basis) { + Basis m = p_basis; + real_t trace = m.rows[0][0] + m.rows[1][1] + m.rows[2][2]; + real_t temp[4]; + + if (trace > 0.0) { + real_t s = Math::sqrt(trace + 1.0f); + temp[3] = (s * 0.5f); + s = 0.5f / s; + + temp[0] = ((m.rows[2][1] - m.rows[1][2]) * s); + temp[1] = ((m.rows[0][2] - m.rows[2][0]) * s); + temp[2] = ((m.rows[1][0] - m.rows[0][1]) * s); + } else { + int i = m.rows[0][0] < m.rows[1][1] + ? (m.rows[1][1] < m.rows[2][2] ? 2 : 1) + : (m.rows[0][0] < m.rows[2][2] ? 2 : 0); + int j = (i + 1) % 3; + int k = (i + 2) % 3; - Size2 s1 = p_prev.get_scale(); - Size2 s2 = p_curr.get_scale(); + real_t s = Math::sqrt(m.rows[i][i] - m.rows[j][j] - m.rows[k][k] + 1.0f); + temp[i] = s * 0.5f; + s = 0.5f / s; - // Slerp rotation. - Vector2 v1(Math::cos(r1), Math::sin(r1)); - Vector2 v2(Math::cos(r2), Math::sin(r2)); + temp[3] = (m.rows[k][j] - m.rows[j][k]) * s; + temp[j] = (m.rows[j][i] + m.rows[i][j]) * s; + temp[k] = (m.rows[k][i] + m.rows[i][k]) * s; + } - real_t dot = v1.dot(v2); + return Quaternion(temp[0], temp[1], temp[2], temp[3]); +} - dot = CLAMP(dot, -1, 1); +Quaternion TransformInterpolator::_quat_slerp_unchecked(const Quaternion &p_from, const Quaternion &p_to, real_t p_fraction) { + Quaternion to1; + real_t omega, cosom, sinom, scale0, scale1; - Vector2 v; + // Calculate cosine. + cosom = p_from.dot(p_to); - if (dot > 0.9995f) { - v = v1.lerp(v2, p_fraction).normalized(); // Linearly interpolate to avoid numerical precision issues. + // Adjust signs (if necessary) + if (cosom < 0.0f) { + cosom = -cosom; + to1.x = -p_to.x; + to1.y = -p_to.y; + to1.z = -p_to.z; + to1.w = -p_to.w; } else { - real_t angle = p_fraction * Math::acos(dot); - Vector2 v3 = (v2 - v1 * dot).normalized(); - v = v1 * Math::cos(angle) + v3 * Math::sin(angle); + to1.x = p_to.x; + to1.y = p_to.y; + to1.z = p_to.z; + to1.w = p_to.w; + } + + // Calculate coefficients. + + // This check could possibly be removed as we dealt with this + // case in the find_method() function, but is left for safety, it probably + // isn't a bottleneck. + if ((1.0f - cosom) > (real_t)CMP_EPSILON) { + // standard case (slerp) + omega = Math::acos(cosom); + sinom = Math::sin(omega); + scale0 = Math::sin((1.0f - p_fraction) * omega) / sinom; + scale1 = Math::sin(p_fraction * omega) / sinom; + } else { + // "from" and "to" quaternions are very close + // ... so we can do a linear interpolation + scale0 = 1.0f - p_fraction; + scale1 = p_fraction; + } + // Calculate final values. + return Quaternion( + scale0 * p_from.x + scale1 * to1.x, + scale0 * p_from.y + scale1 * to1.y, + scale0 * p_from.z + scale1 * to1.z, + scale0 * p_from.w + scale1 * to1.w); +} + +Basis TransformInterpolator::_basis_slerp_unchecked(Basis p_from, Basis p_to, real_t p_fraction) { + Quaternion from = _basis_to_quat_unchecked(p_from); + Quaternion to = _basis_to_quat_unchecked(p_to); + + Basis b(_quat_slerp_unchecked(from, to, p_fraction)); + return b; +} + +void TransformInterpolator::interpolate_basis_scaled_slerp(Basis p_prev, Basis p_curr, Basis &r_result, real_t p_fraction) { + // Normalize both and find lengths. + Vector3 lengths_prev = _basis_orthonormalize(p_prev); + Vector3 lengths_curr = _basis_orthonormalize(p_curr); + + r_result = _basis_slerp_unchecked(p_prev, p_curr, p_fraction); + + // Now the result is unit length basis, we need to scale. + Vector3 lengths_lerped = lengths_prev + ((lengths_curr - lengths_prev) * p_fraction); + + // Keep a note that the column / row order of the basis is weird, + // so keep an eye for bugs with this. + r_result[0] *= lengths_lerped; + r_result[1] *= lengths_lerped; + r_result[2] *= lengths_lerped; +} + +void TransformInterpolator::interpolate_basis_linear(const Basis &p_prev, const Basis &p_curr, Basis &r_result, real_t p_fraction) { + // Interpolate basis. + r_result = p_prev.lerp(p_curr, p_fraction); + + // It turns out we need to guard against zero scale basis. + // This is kind of silly, as we should probably fix the bugs elsewhere in Godot that can't deal with + // zero scale, but until that time... + for (int n = 0; n < 3; n++) { + Vector3 &axis = r_result[n]; + + // Not ok, this could cause errors due to bugs elsewhere, + // so we will bodge set this to a small value. + const real_t smallest = 0.0001f; + const real_t smallest_squared = smallest * smallest; + if (axis.length_squared() < smallest_squared) { + // Setting a different component to the smallest + // helps prevent the situation where all the axes are pointing in the same direction, + // which could be a problem for e.g. cross products... + axis[n] = smallest; + } + } +} + +// Returns length. +real_t TransformInterpolator::_vec3_normalize(Vector3 &p_vec) { + real_t lengthsq = p_vec.length_squared(); + if (lengthsq == 0.0f) { + p_vec.x = p_vec.y = p_vec.z = 0.0f; + return 0.0f; + } + real_t length = Math::sqrt(lengthsq); + p_vec.x /= length; + p_vec.y /= length; + p_vec.z /= length; + return length; +} + +// Returns lengths. +Vector3 TransformInterpolator::_basis_orthonormalize(Basis &r_basis) { + // Gram-Schmidt Process. + + Vector3 x = r_basis.get_column(0); + Vector3 y = r_basis.get_column(1); + Vector3 z = r_basis.get_column(2); + + Vector3 lengths; + + lengths.x = _vec3_normalize(x); + y = (y - x * (x.dot(y))); + lengths.y = _vec3_normalize(y); + z = (z - x * (x.dot(z)) - y * (y.dot(z))); + lengths.z = _vec3_normalize(z); + + r_basis.set_column(0, x); + r_basis.set_column(1, y); + r_basis.set_column(2, z); + + return lengths; +} + +TransformInterpolator::Method TransformInterpolator::_test_basis(Basis p_basis, bool r_needed_normalize, Quaternion &r_quat) { + // Axis lengths. + Vector3 al = Vector3(p_basis.get_column(0).length_squared(), + p_basis.get_column(1).length_squared(), + p_basis.get_column(2).length_squared()); + + // Non unit scale? + if (r_needed_normalize || !_vec3_is_equal_approx(al, Vector3(1.0, 1.0, 1.0), (real_t)0.001f)) { + // If the basis is not normalized (at least approximately), it will fail the checks needed for slerp. + // So we try to detect a scaled (but not sheared) basis, which we *can* slerp by normalizing first, + // and lerping the scales separately. + + // If any of the axes are really small, it is unlikely to be a valid rotation, or is scaled too small to deal with float error. + const real_t sl_epsilon = 0.00001f; + if ((al.x < sl_epsilon) || + (al.y < sl_epsilon) || + (al.z < sl_epsilon)) { + return INTERP_LERP; + } + + // Normalize the basis. + Basis norm_basis = p_basis; + + al.x = Math::sqrt(al.x); + al.y = Math::sqrt(al.y); + al.z = Math::sqrt(al.z); + + norm_basis.set_column(0, norm_basis.get_column(0) / al.x); + norm_basis.set_column(1, norm_basis.get_column(1) / al.y); + norm_basis.set_column(2, norm_basis.get_column(2) / al.z); + + // This doesn't appear necessary, as the later checks will catch it. + // if (!_basis_is_orthogonal_any_scale(norm_basis)) { + // return INTERP_LERP; + // } + + p_basis = norm_basis; + + // Orthonormalize not necessary as normal normalization(!) works if the + // axes are orthonormal. + // p_basis.orthonormalize(); + + // If we needed to normalize one of the two bases, we will need to normalize both, + // regardless of whether the 2nd needs it, just to make sure it takes the path to return + // INTERP_SCALED_LERP on the 2nd call of _test_basis. + r_needed_normalize = true; + } + + // Apply less stringent tests than the built in slerp, the standard Godot slerp + // is too susceptible to float error to be useful. + real_t det = p_basis.determinant(); + if (!Math::is_equal_approx(det, 1, (real_t)0.01f)) { + return INTERP_LERP; + } + + if (!_basis_is_orthogonal(p_basis)) { + return INTERP_LERP; + } + + // TODO: This could possibly be less stringent too, check this. + r_quat = _basis_to_quat_unchecked(p_basis); + if (!r_quat.is_normalized()) { + return INTERP_LERP; + } + + return r_needed_normalize ? INTERP_SCALED_SLERP : INTERP_SLERP; +} + +// This check doesn't seem to be needed but is preserved in case of bugs. +bool TransformInterpolator::_basis_is_orthogonal_any_scale(const Basis &p_basis) { + Vector3 cross = p_basis.get_column(0).cross(p_basis.get_column(1)); + real_t l = _vec3_normalize(cross); + // Too small numbers, revert to lerp. + if (l < 0.001f) { + return false; + } + + const real_t epsilon = 0.9995f; + + real_t dot = cross.dot(p_basis.get_column(2)); + if (dot < epsilon) { + return false; + } + + cross = p_basis.get_column(1).cross(p_basis.get_column(2)); + l = _vec3_normalize(cross); + // Too small numbers, revert to lerp. + if (l < 0.001f) { + return false; + } + + dot = cross.dot(p_basis.get_column(0)); + if (dot < epsilon) { + return false; + } + + return true; +} + +bool TransformInterpolator::_basis_is_orthogonal(const Basis &p_basis, real_t p_epsilon) { + Basis identity; + Basis m = p_basis * p_basis.transposed(); + + // Less stringent tests than the standard Godot slerp. + if (!_vec3_is_equal_approx(m[0], identity[0], p_epsilon) || !_vec3_is_equal_approx(m[1], identity[1], p_epsilon) || !_vec3_is_equal_approx(m[2], identity[2], p_epsilon)) { + return false; + } + return true; +} + +real_t TransformInterpolator::checksum_transform_3d(const Transform3D &p_transform) { + // just a really basic checksum, this can probably be improved + real_t sum = _vec3_sum(p_transform.origin); + sum -= _vec3_sum(p_transform.basis.rows[0]); + sum += _vec3_sum(p_transform.basis.rows[1]); + sum -= _vec3_sum(p_transform.basis.rows[2]); + return sum; +} + +TransformInterpolator::Method TransformInterpolator::find_method(const Basis &p_a, const Basis &p_b) { + bool needed_normalize = false; + + Quaternion q0; + Method method = _test_basis(p_a, needed_normalize, q0); + if (method == INTERP_LERP) { + return method; + } + + Quaternion q1; + method = _test_basis(p_b, needed_normalize, q1); + if (method == INTERP_LERP) { + return method; + } + + // Are they close together? + // Apply the same test that will revert to lerp as is present in the slerp routine. + // Calculate cosine. + real_t cosom = Math::abs(q0.dot(q1)); + if ((1.0f - cosom) <= (real_t)CMP_EPSILON) { + return INTERP_LERP; } - // Construct matrix. - r_result = Transform2D(Math::atan2(v.y, v.x), p1.lerp(p2, p_fraction)); - r_result.scale_basis(s1.lerp(s2, p_fraction)); + return method; } diff --git a/core/math/transform_interpolator.h b/core/math/transform_interpolator.h index a9bce2bd7f..cc556707e4 100644 --- a/core/math/transform_interpolator.h +++ b/core/math/transform_interpolator.h @@ -32,15 +32,64 @@ #define TRANSFORM_INTERPOLATOR_H #include "core/math/math_defs.h" +#include "core/math/vector3.h" + +// Keep all the functions for fixed timestep interpolation together. +// There are two stages involved: +// Finding a method, for determining the interpolation method between two +// keyframes (which are physics ticks). +// And applying that pre-determined method. + +// Pre-determining the method makes sense because it is expensive and often +// several frames may occur between each physics tick, which will make it cheaper +// than performing every frame. struct Transform2D; +struct Transform3D; +struct Basis; +struct Quaternion; class TransformInterpolator { +public: + enum Method { + INTERP_LERP, + INTERP_SLERP, + INTERP_SCALED_SLERP, + }; + private: - static bool _sign(real_t p_val) { return p_val >= 0; } + _FORCE_INLINE_ static bool _sign(real_t p_val) { return p_val >= 0; } + static real_t _vec3_sum(const Vector3 &p_pt) { return p_pt.x + p_pt.y + p_pt.z; } + static real_t _vec3_normalize(Vector3 &p_vec); + _FORCE_INLINE_ static bool _vec3_is_equal_approx(const Vector3 &p_a, const Vector3 &p_b, real_t p_tolerance) { + return Math::is_equal_approx(p_a.x, p_b.x, p_tolerance) && Math::is_equal_approx(p_a.y, p_b.y, p_tolerance) && Math::is_equal_approx(p_a.z, p_b.z, p_tolerance); + } + static Vector3 _basis_orthonormalize(Basis &r_basis); + static Method _test_basis(Basis p_basis, bool r_needed_normalize, Quaternion &r_quat); + static Basis _basis_slerp_unchecked(Basis p_from, Basis p_to, real_t p_fraction); + static Quaternion _quat_slerp_unchecked(const Quaternion &p_from, const Quaternion &p_to, real_t p_fraction); + static Quaternion _basis_to_quat_unchecked(const Basis &p_basis); + static bool _basis_is_orthogonal(const Basis &p_basis, real_t p_epsilon = 0.01f); + static bool _basis_is_orthogonal_any_scale(const Basis &p_basis); + + static void interpolate_basis_linear(const Basis &p_prev, const Basis &p_curr, Basis &r_result, real_t p_fraction); + static void interpolate_basis_scaled_slerp(Basis p_prev, Basis p_curr, Basis &r_result, real_t p_fraction); public: static void interpolate_transform_2d(const Transform2D &p_prev, const Transform2D &p_curr, Transform2D &r_result, real_t p_fraction); + + // Generic functions, use when you don't know what method should be used, e.g. from GDScript. + // These will be slower. + static void interpolate_transform_3d(const Transform3D &p_prev, const Transform3D &p_curr, Transform3D &r_result, real_t p_fraction); + static void interpolate_basis(const Basis &p_prev, const Basis &p_curr, Basis &r_result, real_t p_fraction); + + // Optimized function when you know ahead of time the method. + static void interpolate_transform_3d_via_method(const Transform3D &p_prev, const Transform3D &p_curr, Transform3D &r_result, real_t p_fraction, Method p_method); + static void interpolate_basis_via_method(const Basis &p_prev, const Basis &p_curr, Basis &r_result, real_t p_fraction, Method p_method); + + static real_t checksum_transform_3d(const Transform3D &p_transform); + + static Method find_method(const Basis &p_a, const Basis &p_b); }; #endif // TRANSFORM_INTERPOLATOR_H diff --git a/core/math/vector2.h b/core/math/vector2.h index 8851942cdd..edb47db6fd 100644 --- a/core/math/vector2.h +++ b/core/math/vector2.h @@ -37,7 +37,7 @@ class String; struct Vector2i; -struct _NO_DISCARD_ Vector2 { +struct [[nodiscard]] Vector2 { static const int AXIS_COUNT = 2; enum Axis { diff --git a/core/math/vector2i.h b/core/math/vector2i.h index aca9ae8272..fff9b0a658 100644 --- a/core/math/vector2i.h +++ b/core/math/vector2i.h @@ -37,7 +37,7 @@ class String; struct Vector2; -struct _NO_DISCARD_ Vector2i { +struct [[nodiscard]] Vector2i { static const int AXIS_COUNT = 2; enum Axis { diff --git a/core/math/vector3.h b/core/math/vector3.h index 2313eb557a..14bc44c4e7 100644 --- a/core/math/vector3.h +++ b/core/math/vector3.h @@ -39,7 +39,7 @@ struct Basis; struct Vector2; struct Vector3i; -struct _NO_DISCARD_ Vector3 { +struct [[nodiscard]] Vector3 { static const int AXIS_COUNT = 3; enum Axis { diff --git a/core/math/vector3i.h b/core/math/vector3i.h index 035cfcf9e2..40d0700bf7 100644 --- a/core/math/vector3i.h +++ b/core/math/vector3i.h @@ -37,7 +37,7 @@ class String; struct Vector3; -struct _NO_DISCARD_ Vector3i { +struct [[nodiscard]] Vector3i { static const int AXIS_COUNT = 3; enum Axis { diff --git a/core/math/vector4.h b/core/math/vector4.h index f69b4752bb..8632f69f57 100644 --- a/core/math/vector4.h +++ b/core/math/vector4.h @@ -38,7 +38,7 @@ class String; struct Vector4i; -struct _NO_DISCARD_ Vector4 { +struct [[nodiscard]] Vector4 { static const int AXIS_COUNT = 4; enum Axis { diff --git a/core/math/vector4i.h b/core/math/vector4i.h index 8a9c580bc1..a9036d684a 100644 --- a/core/math/vector4i.h +++ b/core/math/vector4i.h @@ -37,7 +37,7 @@ class String; struct Vector4; -struct _NO_DISCARD_ Vector4i { +struct [[nodiscard]] Vector4i { static const int AXIS_COUNT = 4; enum Axis { |