godot/editor/plugins/baked_light_baker.cpp

2499 lines
65 KiB
C++
Raw Normal View History

/*************************************************************************/
/* baked_light_baker.cpp */
/*************************************************************************/
/* This file is part of: */
/* GODOT ENGINE */
/* http://www.godotengine.org */
/*************************************************************************/
/* Copyright (c) 2007-2017 Juan Linietsky, Ariel Manzur. */
/* Copyright (c) 2014-2017 Godot Engine contributors (cf. AUTHORS.md) */
/* */
/* 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. */
/*************************************************************************/
#include "baked_light_baker.h"
#include "editor/editor_node.h"
#include "editor/editor_settings.h"
#include "io/marshalls.h"
#include <stdlib.h>
#include <cmath>
void baked_light_baker_add_64f(double *dst, double value);
void baked_light_baker_add_64i(int64_t *dst, int64_t value);
//-separar en 2 testuras?
//*mejorar performance y threads
//*modos lineales
//*saturacion
_FORCE_INLINE_ static uint64_t get_uv_normal_bit(const Vector3 &p_vector) {
int lat = Math::fast_ftoi(Math::floor(Math::acos(p_vector.dot(Vector3(0, 1, 0))) * 6.0 / Math_PI + 0.5));
if (lat == 0) {
return 60;
} else if (lat == 6) {
return 61;
}
int lon = Math::fast_ftoi(Math::floor((Math_PI + Math::atan2(p_vector.x, p_vector.z)) * 12.0 / (Math_PI * 2.0) + 0.5)) % 12;
return lon + (lat - 1) * 12;
}
_FORCE_INLINE_ static Vector3 get_bit_normal(int p_bit) {
if (p_bit == 61) {
return Vector3(0, 1, 0);
} else if (p_bit == 62) {
return Vector3(0, -1, 0);
}
float latang = ((p_bit / 12) + 1) * Math_PI / 6.0;
Vector2 latv(Math::sin(latang), Math::cos(latang));
float lonang = ((p_bit % 12) * Math_PI * 2.0 / 12.0) - Math_PI;
Vector2 lonv(Math::sin(lonang), Math::cos(lonang));
return Vector3(lonv.x * latv.x, latv.y, lonv.y * latv.x).normalized();
}
BakedLightBaker::MeshTexture *BakedLightBaker::_get_mat_tex(const Ref<Texture> &p_tex) {
if (!tex_map.has(p_tex)) {
Ref<ImageTexture> imgtex = p_tex;
if (imgtex.is_null())
return NULL;
Image image = imgtex->get_data();
if (image.empty())
return NULL;
if (image.get_format() != Image::FORMAT_RGBA) {
if (image.get_format() > Image::FORMAT_INDEXED_ALPHA) {
Error err = image.decompress();
if (err)
return NULL;
}
if (image.get_format() != Image::FORMAT_RGBA)
image.convert(Image::FORMAT_RGBA);
}
if (imgtex->get_flags() & Texture::FLAG_CONVERT_TO_LINEAR) {
Image copy = image;
copy.srgb_to_linear();
image = copy;
}
DVector<uint8_t> dvt = image.get_data();
DVector<uint8_t>::Read r = dvt.read();
MeshTexture mt;
mt.tex_w = image.get_width();
mt.tex_h = image.get_height();
int len = image.get_width() * image.get_height() * 4;
mt.tex.resize(len);
copymem(mt.tex.ptr(), r.ptr(), len);
textures.push_back(mt);
tex_map[p_tex] = &textures.back()->get();
}
return tex_map[p_tex];
}
void BakedLightBaker::_add_mesh(const Ref<Mesh> &p_mesh, const Ref<Material> &p_mat_override, const Transform &p_xform, int p_baked_texture) {
for (int i = 0; i < p_mesh->get_surface_count(); i++) {
if (p_mesh->surface_get_primitive_type(i) != Mesh::PRIMITIVE_TRIANGLES)
continue;
Ref<Material> mat = p_mat_override.is_valid() ? p_mat_override : p_mesh->surface_get_material(i);
MeshMaterial *matptr = NULL;
int baked_tex = p_baked_texture;
if (mat.is_valid()) {
if (!mat_map.has(mat)) {
MeshMaterial mm;
Ref<FixedMaterial> fm = mat;
if (fm.is_valid()) {
//fixed route
mm.diffuse.color = fm->get_parameter(FixedMaterial::PARAM_DIFFUSE);
if (linear_color)
mm.diffuse.color = mm.diffuse.color.to_linear();
mm.diffuse.tex = _get_mat_tex(fm->get_texture(FixedMaterial::PARAM_DIFFUSE));
mm.specular.color = fm->get_parameter(FixedMaterial::PARAM_SPECULAR);
if (linear_color)
mm.specular.color = mm.specular.color.to_linear();
mm.specular.tex = _get_mat_tex(fm->get_texture(FixedMaterial::PARAM_SPECULAR));
} else {
mm.diffuse.color = Color(1, 1, 1, 1);
mm.diffuse.tex = NULL;
mm.specular.color = Color(0, 0, 0, 1);
mm.specular.tex = NULL;
}
materials.push_back(mm);
mat_map[mat] = &materials.back()->get();
}
matptr = mat_map[mat];
}
int facecount = 0;
if (p_mesh->surface_get_format(i) & Mesh::ARRAY_FORMAT_INDEX) {
facecount = p_mesh->surface_get_array_index_len(i);
} else {
facecount = p_mesh->surface_get_array_len(i);
}
ERR_CONTINUE((facecount == 0 || (facecount % 3) != 0));
facecount /= 3;
int tbase = triangles.size();
triangles.resize(facecount + tbase);
Array a = p_mesh->surface_get_arrays(i);
DVector<Vector3> vertices = a[Mesh::ARRAY_VERTEX];
DVector<Vector3>::Read vr = vertices.read();
DVector<Vector2> uv;
DVector<Vector2>::Read uvr;
DVector<Vector2> uv2;
DVector<Vector2>::Read uv2r;
DVector<Vector3> normal;
DVector<Vector3>::Read normalr;
bool read_uv = false;
bool read_normal = false;
if (p_mesh->surface_get_format(i) & Mesh::ARRAY_FORMAT_TEX_UV) {
uv = a[Mesh::ARRAY_TEX_UV];
uvr = uv.read();
read_uv = true;
if (mat.is_valid() && mat->get_flag(Material::FLAG_LIGHTMAP_ON_UV2) && p_mesh->surface_get_format(i) & Mesh::ARRAY_FORMAT_TEX_UV2) {
uv2 = a[Mesh::ARRAY_TEX_UV2];
uv2r = uv2.read();
} else {
uv2r = uv.read();
if (baked_light->get_transfer_lightmaps_only_to_uv2()) {
baked_tex = -1;
}
}
}
if (p_mesh->surface_get_format(i) & Mesh::ARRAY_FORMAT_NORMAL) {
normal = a[Mesh::ARRAY_NORMAL];
normalr = normal.read();
read_normal = true;
}
Matrix3 normal_xform = p_xform.basis.inverse().transposed();
if (p_mesh->surface_get_format(i) & Mesh::ARRAY_FORMAT_INDEX) {
DVector<int> indices = a[Mesh::ARRAY_INDEX];
DVector<int>::Read ir = indices.read();
for (int i = 0; i < facecount; i++) {
Triangle &t = triangles[tbase + i];
t.vertices[0] = p_xform.xform(vr[ir[i * 3 + 0]]);
t.vertices[1] = p_xform.xform(vr[ir[i * 3 + 1]]);
t.vertices[2] = p_xform.xform(vr[ir[i * 3 + 2]]);
t.material = matptr;
t.baked_texture = baked_tex;
if (read_uv) {
t.uvs[0] = uvr[ir[i * 3 + 0]];
t.uvs[1] = uvr[ir[i * 3 + 1]];
t.uvs[2] = uvr[ir[i * 3 + 2]];
t.bake_uvs[0] = uv2r[ir[i * 3 + 0]];
t.bake_uvs[1] = uv2r[ir[i * 3 + 1]];
t.bake_uvs[2] = uv2r[ir[i * 3 + 2]];
}
if (read_normal) {
t.normals[0] = normal_xform.xform(normalr[ir[i * 3 + 0]]).normalized();
t.normals[1] = normal_xform.xform(normalr[ir[i * 3 + 1]]).normalized();
t.normals[2] = normal_xform.xform(normalr[ir[i * 3 + 2]]).normalized();
}
}
} else {
for (int i = 0; i < facecount; i++) {
Triangle &t = triangles[tbase + i];
t.vertices[0] = p_xform.xform(vr[i * 3 + 0]);
t.vertices[1] = p_xform.xform(vr[i * 3 + 1]);
t.vertices[2] = p_xform.xform(vr[i * 3 + 2]);
t.material = matptr;
t.baked_texture = baked_tex;
if (read_uv) {
t.uvs[0] = uvr[i * 3 + 0];
t.uvs[1] = uvr[i * 3 + 1];
t.uvs[2] = uvr[i * 3 + 2];
t.bake_uvs[0] = uv2r[i * 3 + 0];
t.bake_uvs[1] = uv2r[i * 3 + 1];
t.bake_uvs[2] = uv2r[i * 3 + 2];
}
if (read_normal) {
t.normals[0] = normal_xform.xform(normalr[i * 3 + 0]).normalized();
t.normals[1] = normal_xform.xform(normalr[i * 3 + 1]).normalized();
t.normals[2] = normal_xform.xform(normalr[i * 3 + 2]).normalized();
}
}
}
}
}
void BakedLightBaker::_parse_geometry(Node *p_node) {
if (p_node->cast_to<MeshInstance>()) {
MeshInstance *meshi = p_node->cast_to<MeshInstance>();
Ref<Mesh> mesh = meshi->get_mesh();
if (mesh.is_valid()) {
_add_mesh(mesh, meshi->get_material_override(), base_inv * meshi->get_global_transform(), meshi->get_baked_light_texture_id());
}
} else if (p_node->cast_to<Light>()) {
Light *dl = p_node->cast_to<Light>();
if (dl->get_bake_mode() != Light::BAKE_MODE_DISABLED) {
LightData dirl;
dirl.type = VS::LightType(dl->get_light_type());
dirl.diffuse = dl->get_color(DirectionalLight::COLOR_DIFFUSE);
dirl.specular = dl->get_color(DirectionalLight::COLOR_SPECULAR);
if (linear_color)
dirl.diffuse = dirl.diffuse.to_linear();
if (linear_color)
dirl.specular = dirl.specular.to_linear();
dirl.energy = dl->get_parameter(DirectionalLight::PARAM_ENERGY);
dirl.pos = dl->get_global_transform().origin;
dirl.up = dl->get_global_transform().basis.get_axis(1).normalized();
dirl.left = dl->get_global_transform().basis.get_axis(0).normalized();
dirl.dir = -dl->get_global_transform().basis.get_axis(2).normalized();
dirl.spot_angle = dl->get_parameter(DirectionalLight::PARAM_SPOT_ANGLE);
dirl.spot_attenuation = dl->get_parameter(DirectionalLight::PARAM_SPOT_ATTENUATION);
dirl.attenuation = dl->get_parameter(DirectionalLight::PARAM_ATTENUATION);
dirl.darkening = dl->get_parameter(DirectionalLight::PARAM_SHADOW_DARKENING);
dirl.radius = dl->get_parameter(DirectionalLight::PARAM_RADIUS);
dirl.bake_direct = dl->get_bake_mode() == Light::BAKE_MODE_FULL;
dirl.rays_thrown = 0;
dirl.bake_shadow = dl->get_bake_mode() == Light::BAKE_MODE_INDIRECT_AND_SHADOWS;
lights.push_back(dirl);
}
} else if (p_node->cast_to<Spatial>()) {
Spatial *sp = p_node->cast_to<Spatial>();
Array arr = p_node->call("_get_baked_light_meshes");
for (int i = 0; i < arr.size(); i += 2) {
Transform xform = arr[i];
Ref<Mesh> mesh = arr[i + 1];
_add_mesh(mesh, Ref<Material>(), base_inv * (sp->get_global_transform() * xform));
}
}
for (int i = 0; i < p_node->get_child_count(); i++) {
_parse_geometry(p_node->get_child(i));
}
}
void BakedLightBaker::_fix_lights() {
total_light_area = 0;
for (int i = 0; i < lights.size(); i++) {
LightData &dl = lights[i];
switch (dl.type) {
case VS::LIGHT_DIRECTIONAL: {
float up_max = -1e10;
float dir_max = -1e10;
float left_max = -1e10;
float up_min = 1e10;
float dir_min = 1e10;
float left_min = 1e10;
for (int j = 0; j < triangles.size(); j++) {
for (int k = 0; k < 3; k++) {
Vector3 v = triangles[j].vertices[k];
float up_d = dl.up.dot(v);
float dir_d = dl.dir.dot(v);
float left_d = dl.left.dot(v);
if (up_d > up_max)
up_max = up_d;
if (up_d < up_min)
up_min = up_d;
if (left_d > left_max)
left_max = left_d;
if (left_d < left_min)
left_min = left_d;
if (dir_d > dir_max)
dir_max = dir_d;
if (dir_d < dir_min)
dir_min = dir_d;
}
}
//make a center point, then the upvector and leftvector
dl.pos = dl.left * (left_max + left_min) * 0.5 + dl.up * (up_max + up_min) * 0.5 + dl.dir * (dir_min - (dir_max - dir_min));
dl.left *= (left_max - left_min) * 0.5;
dl.up *= (up_max - up_min) * 0.5;
dl.length = (dir_max - dir_min) * 10; //arbitrary number to keep it in scale
dl.area = dl.left.length() * 2 * dl.up.length() * 2;
dl.constant = 1.0 / dl.area;
} break;
case VS::LIGHT_OMNI:
case VS::LIGHT_SPOT: {
dl.attenuation_table.resize(ATTENUATION_CURVE_LEN);
for (int j = 0; j < ATTENUATION_CURVE_LEN; j++) {
dl.attenuation_table[j] = 1.0 - Math::pow(j / float(ATTENUATION_CURVE_LEN), dl.attenuation);
float falloff = j * dl.radius / float(ATTENUATION_CURVE_LEN);
if (falloff == 0)
falloff = 0.000001;
float intensity = 4 * Math_PI * (falloff * falloff);
//dl.attenuation_table[j]*=falloff*falloff;
dl.attenuation_table[j] *= 1.0 / (3.0 / intensity);
}
if (dl.type == VS::LIGHT_OMNI) {
dl.area = 4.0 * Math_PI * pow(dl.radius, 2.0f);
dl.constant = 1.0 / 3.5;
} else {
float r = Math::tan(Math::deg2rad(dl.spot_angle)) * dl.radius;
float c = 1.0 - (Math::deg2rad(dl.spot_angle) * 0.5 + 0.5);
dl.constant = 1.0 / 3.5;
dl.constant *= 1.0 / c;
dl.area = Math_PI * r * r * c;
}
} break;
}
total_light_area += dl.area;
}
}
BakedLightBaker::BVH *BakedLightBaker::_parse_bvh(BVH **p_children, int p_size, int p_depth, int &max_depth) {
if (p_depth > max_depth) {
max_depth = p_depth;
}
if (p_size == 1) {
return p_children[0];
} else if (p_size == 0) {
return NULL;
}
AABB aabb;
aabb = p_children[0]->aabb;
for (int i = 1; i < p_size; i++) {
aabb.merge_with(p_children[i]->aabb);
}
int li = aabb.get_longest_axis_index();
switch (li) {
case Vector3::AXIS_X: {
SortArray<BVH *, BVHCmpX> sort_x;
sort_x.nth_element(0, p_size, p_size / 2, p_children);
//sort_x.sort(&p_bb[p_from],p_size);
} break;
case Vector3::AXIS_Y: {
SortArray<BVH *, BVHCmpY> sort_y;
sort_y.nth_element(0, p_size, p_size / 2, p_children);
//sort_y.sort(&p_bb[p_from],p_size);
} break;
case Vector3::AXIS_Z: {
SortArray<BVH *, BVHCmpZ> sort_z;
sort_z.nth_element(0, p_size, p_size / 2, p_children);
//sort_z.sort(&p_bb[p_from],p_size);
} break;
}
BVH *left = _parse_bvh(p_children, p_size / 2, p_depth + 1, max_depth);
BVH *right = _parse_bvh(&p_children[p_size / 2], p_size - p_size / 2, p_depth + 1, max_depth);
BVH *_new = memnew(BVH);
_new->aabb = aabb;
_new->center = aabb.pos + aabb.size * 0.5;
_new->children[0] = left;
_new->children[1] = right;
_new->leaf = NULL;
return _new;
}
void BakedLightBaker::_make_bvh() {
Vector<BVH *> bases;
bases.resize(triangles.size());
int max_depth = 0;
for (int i = 0; i < triangles.size(); i++) {
bases[i] = memnew(BVH);
bases[i]->leaf = &triangles[i];
bases[i]->aabb.pos = triangles[i].vertices[0];
bases[i]->aabb.expand_to(triangles[i].vertices[1]);
bases[i]->aabb.expand_to(triangles[i].vertices[2]);
triangles[i].aabb = bases[i]->aabb;
bases[i]->center = bases[i]->aabb.pos + bases[i]->aabb.size * 0.5;
}
bvh = _parse_bvh(bases.ptr(), bases.size(), 1, max_depth);
ray_stack = memnew_arr(uint32_t, max_depth);
bvh_stack = memnew_arr(BVH *, max_depth);
bvh_depth = max_depth;
}
void BakedLightBaker::_octree_insert(int p_octant, Triangle *p_triangle, int p_depth) {
uint32_t *stack = octant_stack;
uint32_t *ptr_stack = octantptr_stack;
Octant *octants = octant_pool.ptr();
stack[0] = 0;
ptr_stack[0] = 0;
int stack_pos = 0;
while (true) {
Octant *octant = &octants[ptr_stack[stack_pos]];
if (stack[stack_pos] < 8) {
int i = stack[stack_pos];
stack[stack_pos]++;
//fit_aabb=fit_aabb.grow(bvh->aabb.size.x*0.0001);
int child_idx = octant->children[i];
bool encloses;
if (!child_idx) {
AABB aabb = octant->aabb;
aabb.size *= 0.5;
if (i & 1)
aabb.pos.x += aabb.size.x;
if (i & 2)
aabb.pos.y += aabb.size.y;
if (i & 4)
aabb.pos.z += aabb.size.z;
aabb.grow_by(cell_size * octree_extra_margin);
if (!aabb.intersects(p_triangle->aabb))
continue;
encloses = aabb.grow(cell_size * -octree_extra_margin * 2.0).encloses(p_triangle->aabb);
if (!encloses && !Face3(p_triangle->vertices[0], p_triangle->vertices[1], p_triangle->vertices[2]).intersects_aabb2(aabb))
continue;
} else {
Octant *child = &octants[child_idx];
AABB aabb = child->aabb;
aabb.grow_by(cell_size * octree_extra_margin);
if (!aabb.intersects(p_triangle->aabb))
continue;
encloses = aabb.grow(cell_size * -octree_extra_margin * 2.0).encloses(p_triangle->aabb);
if (!encloses && !Face3(p_triangle->vertices[0], p_triangle->vertices[1], p_triangle->vertices[2]).intersects_aabb2(aabb))
continue;
}
if (encloses)
stack[stack_pos] = 8; // quick and dirty opt
if (!child_idx) {
if (octant_pool_size == octant_pool.size()) {
octant_pool.resize(octant_pool_size + OCTANT_POOL_CHUNK);
octants = octant_pool.ptr();
octant = &octants[ptr_stack[stack_pos]];
}
child_idx = octant_pool_size++;
octant->children[i] = child_idx;
Octant *child = &octants[child_idx];
child->aabb = octant->aabb;
child->texture_x = 0;
child->texture_y = 0;
child->aabb.size *= 0.5;
if (i & 1)
child->aabb.pos.x += child->aabb.size.x;
if (i & 2)
child->aabb.pos.y += child->aabb.size.y;
if (i & 4)
child->aabb.pos.z += child->aabb.size.z;
child->full_accum[0] = 0;
child->full_accum[1] = 0;
child->full_accum[2] = 0;
child->sampler_ofs = 0;
if (stack_pos == octree_depth - 1) {
child->leaf = true;
child->offset[0] = child->aabb.pos.x + child->aabb.size.x * 0.5;
child->offset[1] = child->aabb.pos.y + child->aabb.size.y * 0.5;
child->offset[2] = child->aabb.pos.z + child->aabb.size.z * 0.5;
child->next_leaf = leaf_list;
for (int ci = 0; ci < 8; ci++) {
child->normal_accum[ci][0] = 0;
child->normal_accum[ci][1] = 0;
child->normal_accum[ci][2] = 0;
}
child->bake_neighbour = 0;
child->first_neighbour = true;
leaf_list = child_idx;
cell_count++;
for (int ci = 0; ci < 8; ci++) {
child->light_accum[ci][0] = 0;
child->light_accum[ci][1] = 0;
child->light_accum[ci][2] = 0;
}
child->parent = ptr_stack[stack_pos];
} else {
child->leaf = false;
for (int j = 0; j < 8; j++) {
child->children[j] = 0;
}
}
}
if (!octants[child_idx].leaf) {
stack_pos++;
stack[stack_pos] = 0;
ptr_stack[stack_pos] = child_idx;
} else {
Octant *child = &octants[child_idx];
Vector3 n = Plane(p_triangle->vertices[0], p_triangle->vertices[1], p_triangle->vertices[2]).normal;
for (int ci = 0; ci < 8; ci++) {
Vector3 pos = child->aabb.pos;
if (ci & 1)
pos.x += child->aabb.size.x;
if (ci & 2)
pos.y += child->aabb.size.y;
if (ci & 4)
pos.z += child->aabb.size.z;
pos.x = floor((pos.x + cell_size * 0.5) / cell_size);
pos.y = floor((pos.y + cell_size * 0.5) / cell_size);
pos.z = floor((pos.z + cell_size * 0.5) / cell_size);
{
Map<Vector3, Vector3>::Element *E = endpoint_normal.find(pos);
if (!E) {
endpoint_normal[pos] = n;
} else {
E->get() += n;
}
}
{
uint64_t bit = get_uv_normal_bit(n);
Map<Vector3, uint64_t>::Element *E = endpoint_normal_bits.find(pos);
if (!E) {
endpoint_normal_bits[pos] = (1 << bit);
} else {
E->get() |= (1 << bit);
}
}
}
}
} else {
stack_pos--;
if (stack_pos < 0)
break;
}
}
}
void BakedLightBaker::_make_octree() {
AABB base = bvh->aabb;
float lal = base.get_longest_axis_size();
//must be square because we want square blocks
base.size.x = lal;
base.size.y = lal;
base.size.z = lal;
base.grow_by(lal * 0.001); //for precision
octree_aabb = base;
cell_size = base.size.x;
for (int i = 0; i < octree_depth; i++)
cell_size /= 2.0;
octant_stack = memnew_arr(uint32_t, octree_depth * 2);
octantptr_stack = memnew_arr(uint32_t, octree_depth * 2);
octant_pool.resize(OCTANT_POOL_CHUNK);
octant_pool_size = 1;
Octant *root = octant_pool.ptr();
root->leaf = false;
root->aabb = octree_aabb;
root->parent = -1;
for (int i = 0; i < 8; i++)
root->children[i] = 0;
EditorProgress ep("bake_octree", vformat(TTR("Parsing %d Triangles:"), triangles.size()), triangles.size());
for (int i = 0; i < triangles.size(); i++) {
_octree_insert(0, &triangles[i], octree_depth - 1);
if ((i % 1000) == 0) {
ep.step(TTR("Triangle #") + itos(i), i);
}
}
{
uint32_t oct_idx = leaf_list;
Octant *octants = octant_pool.ptr();
while (oct_idx) {
BakedLightBaker::Octant *oct = &octants[oct_idx];
for (int ci = 0; ci < 8; ci++) {
Vector3 pos = oct->aabb.pos;
if (ci & 1)
pos.x += oct->aabb.size.x;
if (ci & 2)
pos.y += oct->aabb.size.y;
if (ci & 4)
pos.z += oct->aabb.size.z;
pos.x = floor((pos.x + cell_size * 0.5) / cell_size);
pos.y = floor((pos.y + cell_size * 0.5) / cell_size);
pos.z = floor((pos.z + cell_size * 0.5) / cell_size);
{
Map<Vector3, Vector3>::Element *E = endpoint_normal.find(pos);
if (!E) {
//?
print_line("lolwut?");
} else {
Vector3 n = E->get().normalized();
oct->normal_accum[ci][0] = n.x;
oct->normal_accum[ci][1] = n.y;
oct->normal_accum[ci][2] = n.z;
}
}
{
Map<Vector3, uint64_t>::Element *E = endpoint_normal_bits.find(pos);
if (!E) {
//?
print_line("lolwut?");
} else {
float max_aper = 0;
for (uint64_t i = 0; i < 62; i++) {
if (!(E->get() & (1 << i)))
continue;
Vector3 ang_i = get_bit_normal(i);
for (uint64_t j = 0; j < 62; j++) {
if (i == j)
continue;
if (!(E->get() & (1 << j)))
continue;
Vector3 ang_j = get_bit_normal(j);
float ang = Math::acos(ang_i.dot(ang_j));
if (ang > max_aper)
max_aper = ang;
}
}
if (max_aper > 0.75 * Math_PI) {
//angle too wide prevent problems and forget
oct->normal_accum[ci][0] = 0;
oct->normal_accum[ci][1] = 0;
oct->normal_accum[ci][2] = 0;
}
}
}
}
oct_idx = oct->next_leaf;
}
}
}
void BakedLightBaker::_plot_light(ThreadStack &thread_stack, const Vector3 &p_plot_pos, const AABB &p_plot_aabb, const Color &p_light, const Color &p_tint_light, bool p_only_full, const Plane &p_plane) {
//stackless version
uint32_t *stack = thread_stack.octant_stack;
uint32_t *ptr_stack = thread_stack.octantptr_stack;
Octant *octants = octant_pool.ptr();
stack[0] = 0;
ptr_stack[0] = 0;
int stack_pos = 0;
while (true) {
Octant &octant = octants[ptr_stack[stack_pos]];
if (stack[stack_pos] == 0) {
Vector3 pos = octant.aabb.pos + octant.aabb.size * 0.5;
float md = 1 << (octree_depth - stack_pos);
float r = cell_size * plot_size * md;
float div = 1.0 / (md * md * md);
//div=1.0;
float d = p_plot_pos.distance_to(pos);
if ((p_plane.distance_to(pos) > -cell_size * 1.75 * md) && d <= r) {
float intensity = 1.0 - (d / r) * (d / r); //not gauss but..
baked_light_baker_add_64f(&octant.full_accum[0], p_tint_light.r * intensity * div);
baked_light_baker_add_64f(&octant.full_accum[1], p_tint_light.g * intensity * div);
baked_light_baker_add_64f(&octant.full_accum[2], p_tint_light.b * intensity * div);
}
}
if (octant.leaf) {
//if (p_plane.normal.dot(octant.aabb.get_support(p_plane.normal)) < p_plane.d-CMP_EPSILON) { //octants behind are no go
if (!p_only_full) {
float r = cell_size * plot_size;
for (int i = 0; i < 8; i++) {
Vector3 pos = octant.aabb.pos;
if (i & 1)
pos.x += octant.aabb.size.x;
if (i & 2)
pos.y += octant.aabb.size.y;
if (i & 4)
pos.z += octant.aabb.size.z;
float d = p_plot_pos.distance_to(pos);
if ((p_plane.distance_to(pos) > -cell_size * 1.75) && d <= r) {
float intensity = 1.0 - (d / r) * (d / r); //not gauss but..
if (edge_damp > 0) {
Vector3 normal = Vector3(octant.normal_accum[i][0], octant.normal_accum[i][1], octant.normal_accum[i][2]);
if (normal.x > 0 || normal.y > 0 || normal.z > 0) {
float damp = Math::abs(p_plane.normal.dot(normal));
intensity *= pow(damp, edge_damp);
}
}
//intensity*=1.0-Math::abs(p_plane.distance_to(pos))/(plot_size*cell_size);
//intensity = Math::cos(d*Math_PI*0.5/r);
baked_light_baker_add_64f(&octant.light_accum[i][0], p_light.r * intensity);
baked_light_baker_add_64f(&octant.light_accum[i][1], p_light.g * intensity);
baked_light_baker_add_64f(&octant.light_accum[i][2], p_light.b * intensity);
}
}
}
stack_pos--;
} else if (stack[stack_pos] < 8) {
int i = stack[stack_pos];
stack[stack_pos]++;
if (!octant.children[i]) {
continue;
}
Octant &child = octants[octant.children[i]];
if (!child.aabb.intersects(p_plot_aabb))
continue;
if (child.aabb.encloses(p_plot_aabb)) {
stack[stack_pos] = 8; //don't test the rest
}
stack_pos++;
stack[stack_pos] = 0;
ptr_stack[stack_pos] = octant.children[i];
} else {
stack_pos--;
if (stack_pos < 0)
break;
}
}
}
float BakedLightBaker::_throw_ray(ThreadStack &thread_stack, bool p_bake_direct, const Vector3 &p_begin, const Vector3 &p_end, float p_rest, const Color &p_light, float *p_att_curve, float p_att_pos, int p_att_curve_len, int p_bounces, bool p_first_bounce, bool p_only_dist) {
uint32_t *stack = thread_stack.ray_stack;
BVH **bstack = thread_stack.bvh_stack;
enum {
TEST_AABB_BIT = 0,
VISIT_LEFT_BIT = 1,
VISIT_RIGHT_BIT = 2,
VISIT_DONE_BIT = 3,
};
Vector3 n = (p_end - p_begin);
float len = n.length();
if (len == 0)
return 0;
n /= len;
real_t d = 1e10;
bool inters = false;
Vector3 r_normal;
Vector3 r_point;
Vector3 end = p_end;
Triangle *triangle = NULL;
//for(int i=0;i<max_depth;i++)
// stack[i]=0;
int level = 0;
//AABB ray_aabb;
//ray_aabb.pos=p_begin;
//ray_aabb.expand_to(p_end);
bstack[0] = bvh;
stack[0] = TEST_AABB_BIT;
while (true) {
uint32_t mode = stack[level];
const BVH &b = *bstack[level];
bool done = false;
switch (mode) {
case TEST_AABB_BIT: {
if (b.leaf) {
Face3 f3(b.leaf->vertices[0], b.leaf->vertices[1], b.leaf->vertices[2]);
Vector3 res;
if (f3.intersects_segment(p_begin, end, &res)) {
float nd = n.dot(res);
if (nd < d) {
d = nd;
r_point = res;
end = res;
len = (p_begin - end).length();
r_normal = f3.get_plane().get_normal();
triangle = b.leaf;
inters = true;
}
}
stack[level] = VISIT_DONE_BIT;
} else {
bool valid = b.aabb.smits_intersect_ray(p_begin, n, 0, len);
//bool valid = b.aabb.intersects_segment(p_begin,p_end);
// bool valid = b.aabb.intersects(ray_aabb);
if (!valid) {
stack[level] = VISIT_DONE_BIT;
} else {
stack[level] = VISIT_LEFT_BIT;
}
}
}
continue;
case VISIT_LEFT_BIT: {
stack[level] = VISIT_RIGHT_BIT;
bstack[level + 1] = b.children[0];
stack[level + 1] = TEST_AABB_BIT;
level++;
}
continue;
case VISIT_RIGHT_BIT: {
stack[level] = VISIT_DONE_BIT;
bstack[level + 1] = b.children[1];
stack[level + 1] = TEST_AABB_BIT;
level++;
}
continue;
case VISIT_DONE_BIT: {
if (level == 0) {
done = true;
break;
} else
level--;
}
continue;
}
if (done)
break;
}
if (inters) {
if (p_only_dist) {
return p_begin.distance_to(r_point);
}
//should check if there is normals first
Vector2 uv;
if (true) {
triangle->get_uv_and_normal(r_point, uv, r_normal);
} else {
}
if (n.dot(r_normal) > 0)
return -1;
if (n.dot(r_normal) > 0)
r_normal = -r_normal;
//ok...
Color diffuse_at_point(0.8, 0.8, 0.8);
Color specular_at_point(0.0, 0.0, 0.0);
float dist = p_begin.distance_to(r_point);
AABB aabb;
aabb.pos = r_point;
aabb.pos -= Vector3(1, 1, 1) * cell_size * plot_size;
aabb.size = Vector3(2, 2, 2) * cell_size * plot_size;
Color res_light = p_light;
float att = 1.0;
float dp = (1.0 - normal_damp) * n.dot(-r_normal) + normal_damp;
if (p_att_curve) {
p_att_pos += dist;
int cpos = Math::fast_ftoi((p_att_pos / p_att_curve_len) * ATTENUATION_CURVE_LEN);
cpos = CLAMP(cpos, 0, ATTENUATION_CURVE_LEN - 1);
att = p_att_curve[cpos];
}
res_light.r *= dp;
res_light.g *= dp;
res_light.b *= dp;
//light is plotted before multiplication with diffuse, this way
//the multiplication can happen with more detail in the shader
if (triangle->material) {
//triangle->get_uv(r_point);
diffuse_at_point = triangle->material->diffuse.get_color(uv);
specular_at_point = triangle->material->specular.get_color(uv);
}
diffuse_at_point.r = res_light.r * diffuse_at_point.r;
diffuse_at_point.g = res_light.g * diffuse_at_point.g;
diffuse_at_point.b = res_light.b * diffuse_at_point.b;
float ret = 1e6;
if (p_bounces > 0) {
p_rest -= dist;
if (p_rest < CMP_EPSILON)
return 0;
if (r_normal == -n)
return 0; //todo change a little
r_point += r_normal * 0.01;
specular_at_point.r = res_light.r * specular_at_point.r;
specular_at_point.g = res_light.g * specular_at_point.g;
specular_at_point.b = res_light.b * specular_at_point.b;
if (use_diffuse && (diffuse_at_point.r > CMP_EPSILON || diffuse_at_point.g > CMP_EPSILON || diffuse_at_point.b > CMP_EPSILON)) {
//diffuse bounce
Vector3 c1 = r_normal.cross(n).normalized();
Vector3 c2 = r_normal.cross(c1).normalized();
double r1 = double(rand()) / RAND_MAX;
double r2 = double(rand()) / RAND_MAX;
double r3 = double(rand()) / RAND_MAX;
#if 0
Vector3 next = - ((c1*(r1-0.5)) + (c2*(r2-0.5)) + (r_normal*(r3-0.5))).normalized()*0.5 + r_normal*0.5;
if (next==Vector3())
next=r_normal;
Vector3 rn=next.normalized();
#else
Vector3 rn = ((c1 * (r1 - 0.5)) + (c2 * (r2 - 0.5)) + (r_normal * r3 * 0.5)).normalized();
#endif
ret = _throw_ray(thread_stack, p_bake_direct, r_point, r_point + rn * p_rest, p_rest, diffuse_at_point, p_att_curve, p_att_pos, p_att_curve_len, p_bounces - 1);
}
if (use_specular && (specular_at_point.r > CMP_EPSILON || specular_at_point.g > CMP_EPSILON || specular_at_point.b > CMP_EPSILON)) {
//specular bounce
//Vector3 c1=r_normal.cross(n).normalized();
//Vector3 c2=r_normal.cross(c1).normalized();
Vector3 rn = n - r_normal * r_normal.dot(n) * 2.0;
_throw_ray(thread_stack, p_bake_direct, r_point, r_point + rn * p_rest, p_rest, specular_at_point, p_att_curve, p_att_pos, p_att_curve_len, p_bounces - 1);
}
}
//specular later
// _plot_light_point(r_point,octree,octree_aabb,p_light);
Color plot_light = res_light.linear_interpolate(diffuse_at_point, tint);
plot_light.r *= att;
plot_light.g *= att;
plot_light.b *= att;
Color tint_light = diffuse_at_point;
tint_light.r *= att;
tint_light.g *= att;
tint_light.b *= att;
bool skip = false;
if (!p_first_bounce || p_bake_direct) {
float r = plot_size * cell_size * 2;
if (dist < r) {
//avoid accumulaiton of light on corners
//plot_light=plot_light.linear_interpolate(Color(0,0,0,0),1.0-sd/plot_size*plot_size);
skip = true;
} else {
Vector3 c1 = r_normal.cross(n).normalized();
Vector3 c2 = r_normal.cross(c1).normalized();
double r1 = double(rand()) / RAND_MAX;
double r2 = double(rand()) / RAND_MAX;
double r3 = double(rand()) / RAND_MAX;
Vector3 rn = ((c1 * (r1 - 0.5)) + (c2 * (r2 - 0.5)) + (r_normal * r3 * 0.25)).normalized();
float d = _throw_ray(thread_stack, p_bake_direct, r_point, r_point + rn * p_rest, p_rest, diffuse_at_point, p_att_curve, p_att_pos, p_att_curve_len, p_bounces - 1, false, true);
r = plot_size * cell_size * ao_radius;
if (d > 0 && d < r) {
//avoid accumulaiton of light on corners
//plot_light=plot_light.linear_interpolate(Color(0,0,0,0),1.0-sd/plot_size*plot_size);
skip = true;
} else {
//plot_light=Color(0,0,0,0);
}
}
}
Plane plane(r_point, r_normal);
if (!skip)
_plot_light(thread_stack, r_point, aabb, plot_light, tint_light, !(!p_first_bounce || p_bake_direct), plane);
return dist;
}
return -1;
}
void BakedLightBaker::_make_octree_texture() {
BakedLightBaker::Octant *octants = octant_pool.ptr();
//find neighbours first, to have a better idea of what amount of space is needed
{
Vector<OctantHash> octant_hashing;
octant_hashing.resize(octant_pool_size);
Vector<uint32_t> hash_table;
int hash_table_size = Math::larger_prime(16384);
hash_table.resize(hash_table_size);
uint32_t *hashptr = hash_table.ptr();
OctantHash *octhashptr = octant_hashing.ptr();
for (int i = 0; i < hash_table_size; i++)
hashptr[i] = 0;
//step 1 add to hash table
uint32_t oct_idx = leaf_list;
while (oct_idx) {
BakedLightBaker::Octant *oct = &octants[oct_idx];
uint64_t base = 0;
Vector3 pos = oct->aabb.pos - octree_aabb.pos; //make sure is always positive
base = int((pos.x + cell_size * 0.5) / cell_size);
base <<= 16;
base |= int((pos.y + cell_size * 0.5) / cell_size);
base <<= 16;
base |= int((pos.z + cell_size * 0.5) / cell_size);
uint32_t hash = HashMapHasherDefault::hash(base);
uint32_t idx = hash % hash_table_size;
octhashptr[oct_idx].next = hashptr[idx];
octhashptr[oct_idx].hash = hash;
octhashptr[oct_idx].value = base;
hashptr[idx] = oct_idx;
oct_idx = oct->next_leaf;
}
//step 2 find neighbours
oct_idx = leaf_list;
int neighbours = 0;
while (oct_idx) {
BakedLightBaker::Octant *oct = &octants[oct_idx];
Vector3 pos = oct->aabb.pos - octree_aabb.pos; //make sure is always positive
pos.x += cell_size;
uint64_t base = 0;
base = int((pos.x + cell_size * 0.5) / cell_size);
base <<= 16;
base |= int((pos.y + cell_size * 0.5) / cell_size);
base <<= 16;
base |= int((pos.z + cell_size * 0.5) / cell_size);
uint32_t hash = HashMapHasherDefault::hash(base);
uint32_t idx = hash % hash_table_size;
uint32_t bucket = hashptr[idx];
while (bucket) {
if (octhashptr[bucket].value == base) {
oct->bake_neighbour = bucket;
octants[bucket].first_neighbour = false;
neighbours++;
break;
}
bucket = octhashptr[bucket].next;
}
oct_idx = oct->next_leaf;
}
print_line("octant with neighbour: " + itos(neighbours));
}
//ok let's try to just create a texture
int otex_w = 256;
while (true) {
uint32_t oct_idx = leaf_list;
int row = 0;
print_line("begin at row " + itos(row));
int longest_line_reused = 0;
int col = 0;
int processed = 0;
//reset
while (oct_idx) {
BakedLightBaker::Octant *oct = &octants[oct_idx];
oct->texture_x = 0;
oct->texture_y = 0;
oct_idx = oct->next_leaf;
}
oct_idx = leaf_list;
//assign
while (oct_idx) {
BakedLightBaker::Octant *oct = &octants[oct_idx];
if (oct->first_neighbour && oct->texture_x == 0 && oct->texture_y == 0) {
//was not processed
uint32_t current_idx = oct_idx;
int reused = 0;
while (current_idx) {
BakedLightBaker::Octant *o = &octants[current_idx];
if (col + 1 >= otex_w) {
col = 0;
row += 4;
}
o->texture_x = col;
o->texture_y = row;
processed++;
if (o->bake_neighbour) {
reused++;
}
col += o->bake_neighbour ? 1 : 2; //reuse neighbour
current_idx = o->bake_neighbour;
}
if (reused > longest_line_reused) {
longest_line_reused = reused;
}
}
oct_idx = oct->next_leaf;
}
row += 4;
if (otex_w < row) {
otex_w *= 2;
} else {
baked_light_texture_w = otex_w;
baked_light_texture_h = next_power_of_2(row);
print_line("w: " + itos(otex_w));
print_line("h: " + itos(row));
break;
}
}
{
otex_w = (1 << lattice_size) * (1 << lattice_size) * 2; //make sure lattice fits horizontally
Vector3 lattice_cell_size = octree_aabb.size;
for (int i = 0; i < lattice_size; i++) {
lattice_cell_size *= 0.5;
}
while (true) {
//let's plot the leafs first, given the octree is not so obvious which size it will have
int row = 4 + 4 * (1 << lattice_size);
int col = 0;
col = 0;
row += 4;
print_line("end at row " + itos(row));
//put octree, no need for recursion, just loop backwards.
int regular_octants = 0;
for (int i = octant_pool_size - 1; i >= 0; i--) {
BakedLightBaker::Octant *oct = &octants[i];
if (oct->leaf) //ignore leaf
continue;
if (oct->aabb.size.x > lattice_cell_size.x * 1.1) { //bigger than latice, skip
oct->texture_x = 0;
oct->texture_y = 0;
} else if (oct->aabb.size.x > lattice_cell_size.x * 0.8) {
//this is the initial lattice
Vector3 pos = oct->aabb.pos - octree_aabb.pos; //make sure is always positive
int x = int((pos.x + lattice_cell_size.x * 0.5) / lattice_cell_size.x);
int y = int((pos.y + lattice_cell_size.y * 0.5) / lattice_cell_size.y);
int z = int((pos.z + lattice_cell_size.z * 0.5) / lattice_cell_size.z);
//bug net
ERR_FAIL_INDEX(x, (1 << lattice_size));
ERR_FAIL_INDEX(y, (1 << lattice_size));
ERR_FAIL_INDEX(z, (1 << lattice_size));
/*int ofs = z*(1<<lattice_size)*(1<<lattice_size)+y*(1<<lattice_size)+x;
ofs*=4;
oct->texture_x=ofs%otex_w;
oct->texture_y=(ofs/otex_w)*4+4;
*/
oct->texture_x = (x + (1 << lattice_size) * z) * 2;
oct->texture_y = 4 + y * 4;
//print_line("pos: "+itos(x)+","+itos(y)+","+itos(z)+" - ofs"+itos(oct->texture_x)+","+itos(oct->texture_y));
} else {
//an everyday regular octant
if (col + 2 > otex_w) {
col = 0;
row += 4;
}
oct->texture_x = col;
oct->texture_y = row;
col += 2;
regular_octants++;
}
}
print_line("octants end at row " + itos(row) + " totalling" + itos(regular_octants));
//ok evaluation.
if (otex_w <= 2048 && row > 2048) { //too big upwards, try bigger texture
otex_w *= 2;
continue;
} else {
baked_octree_texture_w = otex_w;
baked_octree_texture_h = row + 4;
break;
}
}
}
baked_octree_texture_h = next_power_of_2(baked_octree_texture_h);
print_line("RESULT! " + itos(baked_octree_texture_w) + "," + itos(baked_octree_texture_h));
}
double BakedLightBaker::get_normalization(int p_light_idx) const {
double nrg = 0;
const LightData &dl = lights[p_light_idx];
double cell_area = cell_size * cell_size;
//nrg+= /*dl.energy */ (dl.rays_thrown * cell_area / dl.area);
nrg = dl.rays_thrown * cell_area;
nrg *= (Math_PI * plot_size * plot_size) * 0.5; // damping of radial linear gradient kernel
nrg *= dl.constant;
//nrg*=5;
return nrg;
}
double BakedLightBaker::get_modifier(int p_light_idx) const {
double nrg = 0;
const LightData &dl = lights[p_light_idx];
double cell_area = cell_size * cell_size;
//nrg+= /*dl.energy */ (dl.rays_thrown * cell_area / dl.area);
nrg = cell_area;
nrg *= (Math_PI * plot_size * plot_size) * 0.5; // damping of radial linear gradient kernel
nrg *= dl.constant;
//nrg*=5;
return nrg;
}
void BakedLightBaker::throw_rays(ThreadStack &thread_stack, int p_amount) {
for (int i = 0; i < lights.size(); i++) {
LightData &dl = lights[i];
int amount = p_amount * total_light_area / dl.area;
double mod = 1.0 / double(get_modifier(i));
mod *= p_amount / float(amount);
switch (dl.type) {
case VS::LIGHT_DIRECTIONAL: {
for (int j = 0; j < amount; j++) {
Vector3 from = dl.pos;
double r1 = double(rand()) / RAND_MAX;
double r2 = double(rand()) / RAND_MAX;
from += dl.up * (r1 * 2.0 - 1.0);
from += dl.left * (r2 * 2.0 - 1.0);
Vector3 to = from + dl.dir * dl.length;
Color col = dl.diffuse;
float m = mod * dl.energy;
col.r *= m;
col.g *= m;
col.b *= m;
dl.rays_thrown++;
baked_light_baker_add_64i(&total_rays, 1);
_throw_ray(thread_stack, dl.bake_direct, from, to, dl.length, col, NULL, 0, 0, max_bounces, true);
}
} break;
case VS::LIGHT_OMNI: {
for (int j = 0; j < amount; j++) {
Vector3 from = dl.pos;
double r1 = double(rand()) / RAND_MAX;
double r2 = double(rand()) / RAND_MAX;
double r3 = double(rand()) / RAND_MAX;
#if 0
//crap is not uniform..
Vector3 dir = Vector3(r1*2.0-1.0,r2*2.0-1.0,r3*2.0-1.0).normalized();
#else
double phi = r1 * Math_PI * 2.0;
double costheta = r2 * 2.0 - 1.0;
double u = r3;
double theta = acos(costheta);
double r = 1.0 * pow(u, 1 / 3.0);
Vector3 dir(
r * sin(theta) * cos(phi),
r * sin(theta) * sin(phi),
r * cos(theta));
dir.normalize();
#endif
Vector3 to = dl.pos + dir * dl.radius;
Color col = dl.diffuse;
float m = mod * dl.energy;
col.r *= m;
col.g *= m;
col.b *= m;
dl.rays_thrown++;
baked_light_baker_add_64i(&total_rays, 1);
_throw_ray(thread_stack, dl.bake_direct, from, to, dl.radius, col, dl.attenuation_table.ptr(), 0, dl.radius, max_bounces, true);
// _throw_ray(i,from,to,dl.radius,col,NULL,0,dl.radius,max_bounces,true);
}
} break;
case VS::LIGHT_SPOT: {
for (int j = 0; j < amount; j++) {
Vector3 from = dl.pos;
double r1 = double(rand()) / RAND_MAX;
//double r2 = double(rand())/RAND_MAX;
double r3 = double(rand()) / RAND_MAX;
float d = Math::tan(Math::deg2rad(dl.spot_angle));
float x = sin(r1 * Math_PI * 2.0) * d;
float y = cos(r1 * Math_PI * 2.0) * d;
Vector3 dir = r3 * (dl.dir + dl.up * y + dl.left * x) + (1.0 - r3) * dl.dir;
dir.normalize();
Vector3 to = dl.pos + dir * dl.radius;
Color col = dl.diffuse;
float m = mod * dl.energy;
col.r *= m;
col.g *= m;
col.b *= m;
dl.rays_thrown++;
baked_light_baker_add_64i(&total_rays, 1);
_throw_ray(thread_stack, dl.bake_direct, from, to, dl.radius, col, dl.attenuation_table.ptr(), 0, dl.radius, max_bounces, true);
// _throw_ray(i,from,to,dl.radius,col,NULL,0,dl.radius,max_bounces,true);
}
} break;
}
}
}
void BakedLightBaker::bake(const Ref<BakedLight> &p_light, Node *p_node) {
if (baking)
return;
cell_count = 0;
base_inv = p_node->cast_to<Spatial>()->get_global_transform().affine_inverse();
EditorProgress ep("bake", TTR("Light Baker Setup:"), 5);
baked_light = p_light;
lattice_size = baked_light->get_initial_lattice_subdiv();
octree_depth = baked_light->get_cell_subdivision();
plot_size = baked_light->get_plot_size();
max_bounces = baked_light->get_bounces();
use_diffuse = baked_light->get_bake_flag(BakedLight::BAKE_DIFFUSE);
use_specular = baked_light->get_bake_flag(BakedLight::BAKE_SPECULAR);
use_translucency = baked_light->get_bake_flag(BakedLight::BAKE_TRANSLUCENT);
edge_damp = baked_light->get_edge_damp();
normal_damp = baked_light->get_normal_damp();
octree_extra_margin = baked_light->get_cell_extra_margin();
tint = baked_light->get_tint();
ao_radius = baked_light->get_ao_radius();
ao_strength = baked_light->get_ao_strength();
linear_color = baked_light->get_bake_flag(BakedLight::BAKE_LINEAR_COLOR);
baked_textures.clear();
for (int i = 0; i < baked_light->get_lightmaps_count(); i++) {
BakeTexture bt;
bt.width = baked_light->get_lightmap_gen_size(i).x;
bt.height = baked_light->get_lightmap_gen_size(i).y;
baked_textures.push_back(bt);
}
ep.step(TTR("Parsing Geometry"), 0);
_parse_geometry(p_node);
mat_map.clear();
tex_map.clear();
print_line("\ttotal triangles: " + itos(triangles.size()));
// no geometry
if (triangles.size() == 0) {
return;
}
ep.step(TTR("Fixing Lights"), 1);
_fix_lights();
ep.step(TTR("Making BVH"), 2);
_make_bvh();
ep.step(TTR("Creating Light Octree"), 3);
_make_octree();
ep.step(TTR("Creating Octree Texture"), 4);
_make_octree_texture();
baking = true;
_start_thread();
}
void BakedLightBaker::update_octree_sampler(DVector<int> &p_sampler) {
BakedLightBaker::Octant *octants = octant_pool.ptr();
double norm = 1.0 / double(total_rays);
if (p_sampler.size() == 0 || first_bake_to_map) {
Vector<int> tmp_smp;
tmp_smp.resize(32); //32 for header
for (int i = 0; i < 32; i++) {
tmp_smp[i] = 0;
}
for (int i = octant_pool_size - 1; i >= 0; i--) {
if (i == 0)
tmp_smp[1] = tmp_smp.size();
Octant &octant = octants[i];
octant.sampler_ofs = tmp_smp.size();
int idxcol[2] = { 0, 0 };
int r = CLAMP((octant.full_accum[0] * norm) * 2048, 0, 32767);
int g = CLAMP((octant.full_accum[1] * norm) * 2048, 0, 32767);
int b = CLAMP((octant.full_accum[2] * norm) * 2048, 0, 32767);
idxcol[0] |= r;
idxcol[1] |= (g << 16) | b;
if (octant.leaf) {
tmp_smp.push_back(idxcol[0]);
tmp_smp.push_back(idxcol[1]);
} else {
for (int j = 0; j < 8; j++) {
if (octant.children[j]) {
idxcol[0] |= (1 << (j + 16));
}
}
tmp_smp.push_back(idxcol[0]);
tmp_smp.push_back(idxcol[1]);
for (int j = 0; j < 8; j++) {
if (octant.children[j]) {
tmp_smp.push_back(octants[octant.children[j]].sampler_ofs);
if (octants[octant.children[j]].sampler_ofs == 0) {
print_line("FUUUUUUUUCK");
}
}
}
}
}
p_sampler.resize(tmp_smp.size());
DVector<int>::Write w = p_sampler.write();
int ss = tmp_smp.size();
for (int i = 0; i < ss; i++) {
w[i] = tmp_smp[i];
}
first_bake_to_map = false;
}
double gamma = baked_light->get_gamma_adjust();
double mult = baked_light->get_energy_multiplier();
float saturation = baked_light->get_saturation();
DVector<int>::Write w = p_sampler.write();
encode_uint32(octree_depth, (uint8_t *)&w[2]);
encode_uint32(linear_color, (uint8_t *)&w[3]);
encode_float(octree_aabb.pos.x, (uint8_t *)&w[4]);
encode_float(octree_aabb.pos.y, (uint8_t *)&w[5]);
encode_float(octree_aabb.pos.z, (uint8_t *)&w[6]);
encode_float(octree_aabb.size.x, (uint8_t *)&w[7]);
encode_float(octree_aabb.size.y, (uint8_t *)&w[8]);
encode_float(octree_aabb.size.z, (uint8_t *)&w[9]);
//norm*=multiplier;
for (int i = octant_pool_size - 1; i >= 0; i--) {
Octant &octant = octants[i];
int idxcol[2] = { w[octant.sampler_ofs], w[octant.sampler_ofs + 1] };
double rf = pow(octant.full_accum[0] * norm * mult, gamma);
double gf = pow(octant.full_accum[1] * norm * mult, gamma);
double bf = pow(octant.full_accum[2] * norm * mult, gamma);
double gray = (rf + gf + bf) / 3.0;
rf = gray + (rf - gray) * saturation;
gf = gray + (gf - gray) * saturation;
bf = gray + (bf - gray) * saturation;
int r = CLAMP((rf)*2048, 0, 32767);
int g = CLAMP((gf)*2048, 0, 32767);
int b = CLAMP((bf)*2048, 0, 32767);
idxcol[0] = ((idxcol[0] >> 16) << 16) | r;
idxcol[1] = (g << 16) | b;
w[octant.sampler_ofs] = idxcol[0];
w[octant.sampler_ofs + 1] = idxcol[1];
}
}
void BakedLightBaker::update_octree_images(DVector<uint8_t> &p_octree, DVector<uint8_t> &p_light) {
int len = baked_octree_texture_w * baked_octree_texture_h * 4;
p_octree.resize(len);
int ilen = baked_light_texture_w * baked_light_texture_h * 4;
p_light.resize(ilen);
DVector<uint8_t>::Write w = p_octree.write();
zeromem(w.ptr(), len);
DVector<uint8_t>::Write iw = p_light.write();
zeromem(iw.ptr(), ilen);
float gamma = baked_light->get_gamma_adjust();
float mult = baked_light->get_energy_multiplier();
for (int i = 0; i < len; i += 4) {
w[i + 0] = 0xFF;
w[i + 1] = 0;
w[i + 2] = 0xFF;
w[i + 3] = 0xFF;
}
for (int i = 0; i < ilen; i += 4) {
iw[i + 0] = 0xFF;
iw[i + 1] = 0;
iw[i + 2] = 0xFF;
iw[i + 3] = 0xFF;
}
float multiplier = 1.0;
if (baked_light->get_format() == BakedLight::FORMAT_HDR8)
multiplier = 8;
encode_uint32(baked_octree_texture_w, &w[0]);
encode_uint32(baked_octree_texture_h, &w[4]);
encode_uint32(0, &w[8]);
encode_float(1 << lattice_size, &w[12]);
encode_uint32(octree_depth - lattice_size, &w[16]);
encode_uint32(multiplier, &w[20]);
encode_uint16(baked_light_texture_w, &w[24]); //if present, use the baked light texture
encode_uint16(baked_light_texture_h, &w[26]);
encode_uint32(0, &w[28]); //baked light texture format
encode_float(octree_aabb.pos.x, &w[32]);
encode_float(octree_aabb.pos.y, &w[36]);
encode_float(octree_aabb.pos.z, &w[40]);
encode_float(octree_aabb.size.x, &w[44]);
encode_float(octree_aabb.size.y, &w[48]);
encode_float(octree_aabb.size.z, &w[52]);
BakedLightBaker::Octant *octants = octant_pool.ptr();
int octant_count = octant_pool_size;
uint8_t *ptr = w.ptr();
uint8_t *lptr = iw.ptr();
int child_offsets[8] = {
0,
4,
baked_octree_texture_w * 4,
baked_octree_texture_w * 4 + 4,
baked_octree_texture_w * 8 + 0,
baked_octree_texture_w * 8 + 4,
baked_octree_texture_w * 8 + baked_octree_texture_w * 4,
baked_octree_texture_w * 8 + baked_octree_texture_w * 4 + 4,
};
int lchild_offsets[8] = {
0,
4,
baked_light_texture_w * 4,
baked_light_texture_w * 4 + 4,
baked_light_texture_w * 8 + 0,
baked_light_texture_w * 8 + 4,
baked_light_texture_w * 8 + baked_light_texture_w * 4,
baked_light_texture_w * 8 + baked_light_texture_w * 4 + 4,
};
/*Vector<double> norm_arr;
norm_arr.resize(lights.size());
for(int i=0;i<lights.size();i++) {
norm_arr[i] = 1.0/get_normalization(i);
}
const double *normptr=norm_arr.ptr();
*/
double norm = 1.0 / double(total_rays);
mult /= multiplier;
double saturation = baked_light->get_saturation();
for (int i = 0; i < octant_count; i++) {
Octant &oct = octants[i];
if (oct.texture_x == 0 && oct.texture_y == 0)
continue;
if (oct.leaf) {
int ofs = (oct.texture_y * baked_light_texture_w + oct.texture_x) << 2;
ERR_CONTINUE(ofs < 0 || ofs > ilen);
//write colors
for (int j = 0; j < 8; j++) {
//if (!oct.children[j])
// continue;
uint8_t *iptr = &lptr[ofs + lchild_offsets[j]];
float r = oct.light_accum[j][0] * norm;
float g = oct.light_accum[j][1] * norm;
float b = oct.light_accum[j][2] * norm;
r = pow(r * mult, gamma);
g = pow(g * mult, gamma);
b = pow(b * mult, gamma);
double gray = (r + g + b) / 3.0;
r = gray + (r - gray) * saturation;
g = gray + (g - gray) * saturation;
b = gray + (b - gray) * saturation;
float ic[3] = {
r,
g,
b,
};
iptr[0] = CLAMP(ic[0] * 255.0, 0, 255);
iptr[1] = CLAMP(ic[1] * 255.0, 0, 255);
iptr[2] = CLAMP(ic[2] * 255.0, 0, 255);
iptr[3] = 255;
}
} else {
int ofs = (oct.texture_y * baked_octree_texture_w + oct.texture_x) << 2;
ERR_CONTINUE(ofs < 0 || ofs > len);
//write indices
for (int j = 0; j < 8; j++) {
if (!oct.children[j])
continue;
Octant &choct = octants[oct.children[j]];
uint8_t *iptr = &ptr[ofs + child_offsets[j]];
iptr[0] = choct.texture_x >> 8;
iptr[1] = choct.texture_x & 0xFF;
iptr[2] = choct.texture_y >> 8;
iptr[3] = choct.texture_y & 0xFF;
}
}
}
}
void BakedLightBaker::_free_bvh(BVH *p_bvh) {
if (!p_bvh->leaf) {
if (p_bvh->children[0])
_free_bvh(p_bvh->children[0]);
if (p_bvh->children[1])
_free_bvh(p_bvh->children[1]);
}
memdelete(p_bvh);
}
bool BakedLightBaker::is_baking() {
return baking;
}
void BakedLightBaker::set_pause(bool p_pause) {
if (paused == p_pause)
return;
paused = p_pause;
if (paused) {
_stop_thread();
} else {
_start_thread();
}
}
bool BakedLightBaker::is_paused() {
return paused;
}
void BakedLightBaker::_bake_thread_func(void *arg) {
BakedLightBaker *ble = (BakedLightBaker *)arg;
ThreadStack thread_stack;
thread_stack.ray_stack = memnew_arr(uint32_t, ble->bvh_depth);
thread_stack.bvh_stack = memnew_arr(BVH *, ble->bvh_depth);
thread_stack.octant_stack = memnew_arr(uint32_t, ble->octree_depth * 2);
thread_stack.octantptr_stack = memnew_arr(uint32_t, ble->octree_depth * 2);
while (!ble->bake_thread_exit) {
ble->throw_rays(thread_stack, 1000);
}
memdelete_arr(thread_stack.ray_stack);
memdelete_arr(thread_stack.bvh_stack);
memdelete_arr(thread_stack.octant_stack);
memdelete_arr(thread_stack.octantptr_stack);
}
void BakedLightBaker::_start_thread() {
if (threads.size() != 0)
return;
bake_thread_exit = false;
int thread_count = EDITOR_DEF("light_baker/custom_bake_threads", 0);
if (thread_count <= 0 || thread_count > 64)
thread_count = OS::get_singleton()->get_processor_count();
//thread_count=1;
threads.resize(thread_count);
for (int i = 0; i < threads.size(); i++) {
threads[i] = Thread::create(_bake_thread_func, this);
}
}
void BakedLightBaker::_stop_thread() {
if (threads.size() == 0)
return;
bake_thread_exit = true;
for (int i = 0; i < threads.size(); i++) {
Thread::wait_to_finish(threads[i]);
memdelete(threads[i]);
}
threads.clear();
}
void BakedLightBaker::_plot_pixel_to_lightmap(int x, int y, int width, int height, uint8_t *image, const Vector3 &p_pos, const Vector3 &p_normal, double *p_norm_ptr, float mult, float gamma) {
uint8_t *ptr = &image[(y * width + x) * 4];
//int lc = lights.size();
double norm = 1.0 / double(total_rays);
Color color;
Octant *octants = octant_pool.ptr();
int octant_idx = 0;
while (true) {
Octant &octant = octants[octant_idx];
if (octant.leaf) {
Vector3 lpos = p_pos - octant.aabb.pos;
lpos /= octant.aabb.size;
Vector3 cols[8];
for (int i = 0; i < 8; i++) {
cols[i].x += octant.light_accum[i][0] * norm;
cols[i].y += octant.light_accum[i][1] * norm;
cols[i].z += octant.light_accum[i][2] * norm;
}
/*Vector3 final = (cols[0] + (cols[1] - cols[0]) * lpos.y);
final = final + ((cols[2] + (cols[3] - cols[2]) * lpos.y) - final)*lpos.x;
Vector3 final2 = (cols[4+0] + (cols[4+1] - cols[4+0]) * lpos.y);
final2 = final2 + ((cols[4+2] + (cols[4+3] - cols[4+2]) * lpos.y) - final2)*lpos.x;*/
Vector3 finala = cols[0].linear_interpolate(cols[1], lpos.x);
Vector3 finalb = cols[2].linear_interpolate(cols[3], lpos.x);
Vector3 final = finala.linear_interpolate(finalb, lpos.y);
Vector3 final2a = cols[4 + 0].linear_interpolate(cols[4 + 1], lpos.x);
Vector3 final2b = cols[4 + 2].linear_interpolate(cols[4 + 3], lpos.x);
Vector3 final2 = final2a.linear_interpolate(final2b, lpos.y);
final = final.linear_interpolate(final2, lpos.z);
if (baked_light->get_format() == BakedLight::FORMAT_HDR8)
final *= 8.0;
color.r = pow(final.x * mult, gamma);
color.g = pow(final.y * mult, gamma);
color.b = pow(final.z * mult, gamma);
color.a = 1.0;
int lc = lights.size();
LightData *lv = lights.ptr();
for (int i = 0; i < lc; i++) {
//shadow baking
if (!lv[i].bake_shadow)
continue;
Vector3 from = p_pos + p_normal * 0.01;
Vector3 to;
float att = 0;
switch (lv[i].type) {
case VS::LIGHT_DIRECTIONAL: {
to = from - lv[i].dir * lv[i].length;
} break;
case VS::LIGHT_OMNI: {
to = lv[i].pos;
float d = MIN(lv[i].radius, to.distance_to(from)) / lv[i].radius;
att = d; //1.0-d;
} break;
default: continue;
}
uint32_t *stack = ray_stack;
BVH **bstack = bvh_stack;
enum {
TEST_RAY_BIT = 0,
VISIT_LEFT_BIT = 1,
VISIT_RIGHT_BIT = 2,
VISIT_DONE_BIT = 3,
};
bool intersected = false;
int level = 0;
Vector3 n = (to - from);
float len = n.length();
if (len == 0)
continue;
n /= len;
bstack[0] = bvh;
stack[0] = TEST_RAY_BIT;
while (!intersected) {
uint32_t mode = stack[level];
const BVH &b = *bstack[level];
bool done = false;
switch (mode) {
case TEST_RAY_BIT: {
if (b.leaf) {
Face3 f3(b.leaf->vertices[0], b.leaf->vertices[1], b.leaf->vertices[2]);
Vector3 res;
if (f3.intersects_segment(from, to)) {
intersected = true;
done = true;
}
stack[level] = VISIT_DONE_BIT;
} else {
bool valid = b.aabb.smits_intersect_ray(from, n, 0, len);
//bool valid = b.aabb.intersects_segment(p_begin,p_end);
// bool valid = b.aabb.intersects(ray_aabb);
if (!valid) {
stack[level] = VISIT_DONE_BIT;
} else {
stack[level] = VISIT_LEFT_BIT;
}
}
}
continue;
case VISIT_LEFT_BIT: {
stack[level] = VISIT_RIGHT_BIT;
bstack[level + 1] = b.children[0];
stack[level + 1] = TEST_RAY_BIT;
level++;
}
continue;
case VISIT_RIGHT_BIT: {
stack[level] = VISIT_DONE_BIT;
bstack[level + 1] = b.children[1];
stack[level + 1] = TEST_RAY_BIT;
level++;
}
continue;
case VISIT_DONE_BIT: {
if (level == 0) {
done = true;
break;
} else
level--;
}
continue;
}
if (done)
break;
}
if (intersected) {
color.a = Math::lerp(MAX(0.01, lv[i].darkening), 1.0, att);
}
}
break;
} else {
Vector3 lpos = p_pos - octant.aabb.pos;
Vector3 half = octant.aabb.size * 0.5;
int ofs = 0;
if (lpos.x >= half.x)
ofs |= 1;
if (lpos.y >= half.y)
ofs |= 2;
if (lpos.z >= half.z)
ofs |= 4;
octant_idx = octant.children[ofs];
if (octant_idx == 0)
return;
}
}
ptr[0] = CLAMP(color.r * 255.0, 0, 255);
ptr[1] = CLAMP(color.g * 255.0, 0, 255);
ptr[2] = CLAMP(color.b * 255.0, 0, 255);
ptr[3] = CLAMP(color.a * 255.0, 0, 255);
}
Error BakedLightBaker::transfer_to_lightmaps() {
if (!triangles.size() || baked_textures.size() == 0)
return ERR_UNCONFIGURED;
EditorProgress ep("transfer_to_lightmaps", TTR("Transfer to Lightmaps:"), baked_textures.size() * 2 + triangles.size());
for (int i = 0; i < baked_textures.size(); i++) {
ERR_FAIL_COND_V(baked_textures[i].width <= 0 || baked_textures[i].height <= 0, ERR_UNCONFIGURED);
baked_textures[i].data.resize(baked_textures[i].width * baked_textures[i].height * 4);
zeromem(baked_textures[i].data.ptr(), baked_textures[i].data.size());
ep.step(TTR("Allocating Texture #") + itos(i + 1), i);
}
Vector<double> norm_arr;
norm_arr.resize(lights.size());
for (int i = 0; i < lights.size(); i++) {
norm_arr[i] = 1.0 / get_normalization(i);
}
float gamma = baked_light->get_gamma_adjust();
float mult = baked_light->get_energy_multiplier();
for (int i = 0; i < triangles.size(); i++) {
if (i % 200 == 0) {
ep.step(TTR("Baking Triangle #") + itos(i), i + baked_textures.size());
}
Triangle &t = triangles[i];
if (t.baked_texture < 0 || t.baked_texture >= baked_textures.size())
continue;
BakeTexture &bt = baked_textures[t.baked_texture];
Vector3 normal = Plane(t.vertices[0], t.vertices[1], t.vertices[2]).normal;
int x[3];
int y[3];
Vector3 vertices[3] = {
t.vertices[0],
t.vertices[1],
t.vertices[2]
};
for (int j = 0; j < 3; j++) {
x[j] = t.bake_uvs[j].x * bt.width;
y[j] = t.bake_uvs[j].y * bt.height;
x[j] = CLAMP(x[j], 0, bt.width - 1);
y[j] = CLAMP(y[j], 0, bt.height - 1);
}
{
// sort the points vertically
if (y[1] > y[2]) {
SWAP(x[1], x[2]);
SWAP(y[1], y[2]);
SWAP(vertices[1], vertices[2]);
}
if (y[0] > y[1]) {
SWAP(x[0], x[1]);
SWAP(y[0], y[1]);
SWAP(vertices[0], vertices[1]);
}
if (y[1] > y[2]) {
SWAP(x[1], x[2]);
SWAP(y[1], y[2]);
SWAP(vertices[1], vertices[2]);
}
double dx_far = double(x[2] - x[0]) / (y[2] - y[0] + 1);
double dx_upper = double(x[1] - x[0]) / (y[1] - y[0] + 1);
double dx_low = double(x[2] - x[1]) / (y[2] - y[1] + 1);
double xf = x[0];
double xt = x[0] + dx_upper; // if y[0] == y[1], special case
for (int yi = y[0]; yi <= (y[2] > bt.height - 1 ? bt.height - 1 : y[2]); yi++) {
if (yi >= 0) {
for (int xi = (xf > 0 ? int(xf) : 0); xi <= (xt < bt.width ? xt : bt.width - 1); xi++) {
//pixels[int(x + y * width)] = color;
Vector2 v0 = Vector2(x[1] - x[0], y[1] - y[0]);
Vector2 v1 = Vector2(x[2] - x[0], y[2] - y[0]);
//vertices[2] - vertices[0];
Vector2 v2 = Vector2(xi - x[0], yi - y[0]);
float d00 = v0.dot(v0);
float d01 = v0.dot(v1);
float d11 = v1.dot(v1);
float d20 = v2.dot(v0);
float d21 = v2.dot(v1);
float denom = (d00 * d11 - d01 * d01);
Vector3 pos;
if (denom == 0) {
pos = t.vertices[0];
} else {
float v = (d11 * d20 - d01 * d21) / denom;
float w = (d00 * d21 - d01 * d20) / denom;
float u = 1.0f - v - w;
pos = vertices[0] * u + vertices[1] * v + vertices[2] * w;
}
_plot_pixel_to_lightmap(xi, yi, bt.width, bt.height, bt.data.ptr(), pos, normal, norm_arr.ptr(), mult, gamma);
}
for (int xi = (xf < bt.width ? int(xf) : bt.width - 1); xi >= (xt > 0 ? xt : 0); xi--) {
//pixels[int(x + y * width)] = color;
Vector2 v0 = Vector2(x[1] - x[0], y[1] - y[0]);
Vector2 v1 = Vector2(x[2] - x[0], y[2] - y[0]);
//vertices[2] - vertices[0];
Vector2 v2 = Vector2(xi - x[0], yi - y[0]);
float d00 = v0.dot(v0);
float d01 = v0.dot(v1);
float d11 = v1.dot(v1);
float d20 = v2.dot(v0);
float d21 = v2.dot(v1);
float denom = (d00 * d11 - d01 * d01);
Vector3 pos;
if (denom == 0) {
pos = t.vertices[0];
} else {
float v = (d11 * d20 - d01 * d21) / denom;
float w = (d00 * d21 - d01 * d20) / denom;
float u = 1.0f - v - w;
pos = vertices[0] * u + vertices[1] * v + vertices[2] * w;
}
_plot_pixel_to_lightmap(xi, yi, bt.width, bt.height, bt.data.ptr(), pos, normal, norm_arr.ptr(), mult, gamma);
}
}
xf += dx_far;
if (yi < y[1])
xt += dx_upper;
else
xt += dx_low;
}
}
}
for (int i = 0; i < baked_textures.size(); i++) {
{
ep.step(TTR("Post-Processing Texture #") + itos(i), i + baked_textures.size() + triangles.size());
BakeTexture &bt = baked_textures[i];
Vector<uint8_t> copy_data = bt.data;
uint8_t *data = bt.data.ptr();
const int max_radius = 8;
const int shadow_radius = 2;
const int max_dist = 0x7FFFFFFF;
for (int x = 0; x < bt.width; x++) {
for (int y = 0; y < bt.height; y++) {
uint8_t a = copy_data[(y * bt.width + x) * 4 + 3];
if (a > 0) {
//blur shadow
int from_x = MAX(0, x - shadow_radius);
int to_x = MIN(bt.width - 1, x + shadow_radius);
int from_y = MAX(0, y - shadow_radius);
int to_y = MIN(bt.height - 1, y + shadow_radius);
int sum = 0;
int sumc = 0;
for (int k = from_y; k <= to_y; k++) {
for (int l = from_x; l <= to_x; l++) {
const uint8_t *rp = &copy_data[(k * bt.width + l) << 2];
sum += rp[3];
sumc++;
}
}
sum /= sumc;
data[(y * bt.width + x) * 4 + 3] = sum;
} else {
int closest_dist = max_dist;
uint8_t closest_color[4];
int from_x = MAX(0, x - max_radius);
int to_x = MIN(bt.width - 1, x + max_radius);
int from_y = MAX(0, y - max_radius);
int to_y = MIN(bt.height - 1, y + max_radius);
for (int k = from_y; k <= to_y; k++) {
for (int l = from_x; l <= to_x; l++) {
int dy = y - k;
int dx = x - l;
int dist = dy * dy + dx * dx;
if (dist >= closest_dist)
continue;
const uint8_t *rp = &copy_data[(k * bt.width + l) << 2];
if (rp[3] == 0)
continue;
closest_dist = dist;
closest_color[0] = rp[0];
closest_color[1] = rp[1];
closest_color[2] = rp[2];
closest_color[3] = rp[3];
}
}
if (closest_dist != max_dist) {
data[(y * bt.width + x) * 4 + 0] = closest_color[0];
data[(y * bt.width + x) * 4 + 1] = closest_color[1];
data[(y * bt.width + x) * 4 + 2] = closest_color[2];
data[(y * bt.width + x) * 4 + 3] = closest_color[3];
}
}
}
}
}
DVector<uint8_t> dv;
dv.resize(baked_textures[i].data.size());
{
DVector<uint8_t>::Write w = dv.write();
copymem(w.ptr(), baked_textures[i].data.ptr(), baked_textures[i].data.size());
}
Image img(baked_textures[i].width, baked_textures[i].height, 0, Image::FORMAT_RGBA, dv);
Ref<ImageTexture> tex = memnew(ImageTexture);
tex->create_from_image(img);
baked_light->set_lightmap_texture(i, tex);
}
return OK;
}
void BakedLightBaker::clear() {
_stop_thread();
if (bvh)
_free_bvh(bvh);
if (ray_stack)
memdelete_arr(ray_stack);
if (octant_stack)
memdelete_arr(octant_stack);
if (octantptr_stack)
memdelete_arr(octantptr_stack);
if (bvh_stack)
memdelete_arr(bvh_stack);
/*
* ???
for(int i=0;i<octant_pool.size();i++) {
//if (octant_pool[i].leaf) {
// memdelete_arr( octant_pool[i].light );
//} Vector<double> norm_arr;
//norm_arr.resize(lights.size());
for(int i=0;i<lights.size();i++) {
norm_arr[i] = 1.0/get_normalization(i);
}
const double *normptr=norm_arr.ptr();
}
*/
octant_pool.clear();
octant_pool_size = 0;
bvh = NULL;
leaf_list = 0;
cell_count = 0;
ray_stack = NULL;
octant_stack = NULL;
octantptr_stack = NULL;
bvh_stack = NULL;
materials.clear();
materials.clear();
textures.clear();
lights.clear();
triangles.clear();
endpoint_normal.clear();
endpoint_normal_bits.clear();
baked_octree_texture_w = 0;
baked_octree_texture_h = 0;
paused = false;
baking = false;
bake_thread_exit = false;
first_bake_to_map = true;
baked_light = Ref<BakedLight>();
total_rays = 0;
}
BakedLightBaker::BakedLightBaker() {
octree_depth = 9;
lattice_size = 4;
octant_pool.clear();
octant_pool_size = 0;
bvh = NULL;
leaf_list = 0;
cell_count = 0;
ray_stack = NULL;
bvh_stack = NULL;
octant_stack = NULL;
octantptr_stack = NULL;
plot_size = 2.5;
max_bounces = 2;
materials.clear();
baked_octree_texture_w = 0;
baked_octree_texture_h = 0;
paused = false;
baking = false;
bake_thread_exit = false;
total_rays = 0;
first_bake_to_map = true;
linear_color = false;
}
BakedLightBaker::~BakedLightBaker() {
clear();
}