/*  Copyright (C) 1996-1997  Id Software, Inc.

    This program is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    the Free Software Foundation; either version 2 of the License, or
    (at your option) any later version.

    This program is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.

    You should have received a copy of the GNU General Public License
    along with this program; if not, write to the Free Software
    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA

    See file, 'COPYING', for details.
*/

#include <light/lightgrid.hh>

#include <cstdint>
#include <iostream>
#include <fstream>
#include <vector>
#include <algorithm>
#include <string>
#include <utility>

#include <light/light.hh>
#include <light/entities.hh>
#include <light/ltface.hh>

#include <common/prtfile.hh>
#include <common/parallel.hh>
#include <common/qvec.hh>
#include <common/cmdlib.hh>

static aabb3f LightGridBounds(const mbsp_t &bsp)
{
    aabb3f result;

    // see if `_lightgrid_hint` entities are in use
    for (auto &entity : GetEntdicts()) {
        if (entity.get_int("_lightgrid_hint")) {
            qvec3f point{};
            entity.get_vector("origin", point);
            result += point;
        }
    }

    if (result.valid()) {
        auto size = result.size();
        if (size[0] > 0 && size[1] > 0 && size[2] > 0) {
            return result;
        }
    }

    result = Model_BoundsOfFaces(bsp, bsp.dmodels[0]);
    return result;
}

struct lightgrid_raw_data
{
    qvec3f grid_dist;
    qvec3f grid_mins;
    qvec3i grid_size;
    std::vector<lightgrid_samples_t> grid_result;
    std::vector<uint8_t> occlusion;
    uint8_t num_styles;

    int get_grid_index(int x, int y, int z) const { return (grid_size[0] * grid_size[1] * z) + (grid_size[0] * y) + x; }

    qvec3f grid_index_to_world(const qvec3i &index) const { return grid_mins + (index * grid_dist); }
};

static std::vector<uint8_t> MakeOctreeLump(const mbsp_t &bsp, const lightgrid_raw_data &data)
{
    /**
     * returns the octant index in [0..7]
     */
    auto child_index = [](qvec3i division_point, qvec3i test_point) -> int {
        int sign[3];
        for (int i = 0; i < 3; ++i)
            sign[i] = (test_point[i] >= division_point[i]);

        return (4 * sign[0]) + (2 * sign[1]) + (sign[2]);
    };

    Q_assert(child_index({1, 1, 1}, {2, 2, 2}) == 7);
    Q_assert(child_index({1, 1, 1}, {1, 1, 0}) == 6);
    Q_assert(child_index({1, 1, 1}, {1, 0, 1}) == 5);
    Q_assert(child_index({1, 1, 1}, {1, 0, 0}) == 4);
    Q_assert(child_index({1, 1, 1}, {0, 1, 1}) == 3);
    Q_assert(child_index({1, 1, 1}, {0, 1, 0}) == 2);
    Q_assert(child_index({1, 1, 1}, {0, 0, 1}) == 1);
    Q_assert(child_index({1, 1, 1}, {0, 0, 0}) == 0);

    /**
     * returns octant index `i`'s mins and size
     */
    auto get_octant = [](int i, qvec3i mins, qvec3i size, qvec3i division_point) -> std::tuple<qvec3i, qvec3i> {
        qvec3i child_mins;
        qvec3i child_size;
        for (int axis = 0; axis < 3; ++axis) {
            int bit;
            if (axis == 0) {
                bit = 4;
            } else if (axis == 1) {
                bit = 2;
            } else {
                bit = 1;
            }

            if (i & bit) {
                child_mins[axis] = division_point[axis];
                child_size[axis] = mins[axis] + size[axis] - division_point[axis];
            } else {
                child_mins[axis] = mins[axis];
                child_size[axis] = division_point[axis] - mins[axis];
            }
        }
        return {child_mins, child_size};
    };

    Q_assert(get_octant(0, {0, 0, 0}, {2, 2, 2}, {1, 1, 1}) == (std::tuple<qvec3i, qvec3i>{{0, 0, 0}, {1, 1, 1}}));
    Q_assert(get_octant(7, {0, 0, 0}, {2, 2, 2}, {1, 1, 1}) == (std::tuple<qvec3i, qvec3i>{{1, 1, 1}, {1, 1, 1}}));

    /**
     * given a bounding box, selects the division point.
     */
    auto get_division_point = [](qvec3i mins, qvec3i size) -> qvec3i { return mins + (size / 2); };

    auto count_occluded_unoccluded = [&](qvec3i mins, qvec3i size) -> std::tuple<int, int> {
        std::tuple<int, int> occluded_unoccluded;
        for (int z = mins[2]; z < (mins[2] + size[2]); ++z) {
            for (int y = mins[1]; y < (mins[1] + size[1]); ++y) {
                for (int x = mins[0]; x < (mins[0] + size[0]); ++x) {
                    int sample_index = data.get_grid_index(x, y, z);
                    if (data.occlusion[sample_index]) {
                        std::get<0>(occluded_unoccluded)++;
                    } else {
                        std::get<1>(occluded_unoccluded)++;
                    }
                }
            }
        }
        return occluded_unoccluded;
    };

    constexpr int MAX_DEPTH = 5;
    // if any axis is fewer than this many grid points, don't bother subdividing further, just create a leaf
    constexpr int MIN_NODE_DIMENSION = 4;

    // if set, it's an index in the leafs array
    [[maybe_unused]] constexpr uint32_t FLAG_LEAF = 1 << 31;
    [[maybe_unused]] constexpr uint32_t FLAG_OCCLUDED = 1 << 30;
    [[maybe_unused]] constexpr uint32_t FLAGS = (FLAG_LEAF | FLAG_OCCLUDED);
    // if neither flags are set, it's a node index

    struct octree_node
    {
        qvec3i division_point;
        std::array<uint32_t, 8> children;
    };

    struct octree_leaf
    {
        qvec3i mins, size;
    };

    std::vector<octree_node> octree_nodes;
    std::vector<octree_leaf> octree_leafs;

    int occluded_cells = 0;

    /**
     * - inserts either a node or leaf
     * - returns one of:
     *   - FLAG_OCCLUDED if the entire bounds is occluded
     *   - (FLAG_LEAF | leaf_num) for a leaf - a literal chunk of grid samples
     *   - otherwise, it's a node index
     */
    std::function<int(qvec3i, qvec3i, int depth)> build_octree;
    build_octree = [&](qvec3i mins, qvec3i size, int depth) -> uint32_t {
        assert(size[0] > 0);
        assert(size[1] > 0);
        assert(size[2] > 0);

        // special case: fully occluded leaf, just represented as a flag bit
        auto [occluded_count, unoccluded_count] = count_occluded_unoccluded(mins, size);
        if (!unoccluded_count) {
            occluded_cells += size[0] * size[1] * size[2];
            return FLAG_OCCLUDED;
        }

        // decide whether we are creating a regular leaf or a node?
        bool make_leaf = false;
        if (size[0] < MIN_NODE_DIMENSION || size[1] < MIN_NODE_DIMENSION || size[2] < MIN_NODE_DIMENSION)
            make_leaf = true;
        if (depth == MAX_DEPTH)
            make_leaf = true;

        if (occluded_count < 8) {
            // force a leaf if it's mostly unoccluded
            make_leaf = true;
        }

        if (make_leaf) {
            // make a leaf
            const uint32_t leafnum = static_cast<uint32_t>(octree_leafs.size());
            octree_leafs.push_back({.mins = mins, .size = size});
            return FLAG_LEAF | leafnum;
        }

        // make a node

        const qvec3i division_point = get_division_point(mins, size);

        // create the 8 child nodes/leafs recursively, store the returned indices
        std::array<uint32_t, 8> children;
        for (int i = 0; i < 8; ++i) {
            // figure out the mins/size of this child
            auto [child_mins, child_size] = get_octant(i, mins, size, division_point);
            children[i] = build_octree(child_mins, child_size, depth + 1);
        }

        // insert the node
        const uint32_t nodenum = static_cast<uint32_t>(octree_nodes.size());
        octree_nodes.push_back({.division_point = division_point, .children = children});
        return nodenum;
    };

    // build the root node
    const uint32_t root_node = build_octree(qvec3i{0, 0, 0}, data.grid_size, 0);

    // visualize the leafs
    {
        std::vector<polylib::winding_t> windings;

        for (auto &leaf : octree_leafs) {
            auto leaf_world_mins = data.grid_index_to_world(leaf.mins);
            auto leaf_world_maxs = data.grid_index_to_world(leaf.mins + leaf.size - qvec3i(1, 1, 1));

            aabb3d bounds(leaf_world_mins, leaf_world_maxs);

            auto bounds_windings = polylib::winding_t::aabb_windings(bounds);
            for (auto &w : bounds_windings) {
                windings.push_back(std::move(w));
            }
        }

        WriteDebugPortals(windings, fs::path(light_options.sourceMap).replace_extension(".octree.prt"));
    }

    // stats
    int stored_cells = 0;
    for (auto &leaf : octree_leafs) {
        stored_cells += leaf.size[0] * leaf.size[1] * leaf.size[2];
    }
    logging::print("octree stored {} grid nodes + {} occluded = {} total, full stored {} (octree is {} percent)\n",
        stored_cells, occluded_cells, stored_cells + occluded_cells, data.occlusion.size(),
        100.0f * stored_cells / (float)data.occlusion.size());

    logging::print("octree nodes size: {} bytes ({} * {})\n", octree_nodes.size() * sizeof(octree_node),
        octree_nodes.size(), sizeof(octree_node));

    logging::print(
        "octree leafs {} overhead {} bytes\n", octree_leafs.size(), octree_leafs.size() * sizeof(octree_leaf));

    // lookup function
    std::function<std::tuple<lightgrid_samples_t, bool>(uint32_t, qvec3i)> octree_lookup_r;
    octree_lookup_r = [&](uint32_t node_index, qvec3i test_point) -> std::tuple<lightgrid_samples_t, bool> {
        if (node_index & FLAG_OCCLUDED) {
            return {lightgrid_samples_t{}, true};
        }
        if (node_index & FLAG_LEAF) {
            // in actuality, we'd pull the data from a 3D grid stored in the leaf.
            int i = data.get_grid_index(test_point[0], test_point[1], test_point[2]);
            return {data.grid_result[i], data.occlusion[i]};
        }
        auto &node = octree_nodes[node_index];
        int i = child_index(node.division_point, test_point); // [0..7]
        return octree_lookup_r(node.children[i], test_point);
    };

#if 0
    // self-check
    for (int z = 0; z < data.grid_size[2]; ++z) {
        for (int y = 0; y < data.grid_size[1]; ++y) {
            for (int x = 0; x < data.grid_size[0]; ++x) {
                auto [color, occluded] = octree_lookup_r(root_node, {x, y, z});

                int sample_index = data.get_grid_index(x, y, z);

                // compare against original data
                if (occluded) {
                    Q_assert(data.occlusion[sample_index]);
                } else {
                    Q_assert(!data.occlusion[sample_index]);
                    Q_assert(data.grid_result[sample_index] == color);
                }
            }
        }
    }
#endif

    // write out the binary data
    const qvec3f grid_dist = qvec3f{data.grid_dist};

    std::ostringstream str(std::ios_base::out | std::ios_base::binary);
    str << endianness<std::endian::little>;
    str <= grid_dist;
    str <= data.grid_size;
    str <= data.grid_mins;
    str <= data.num_styles;

    str <= static_cast<uint32_t>(root_node);

    // the nodes (fixed-size)
    str <= static_cast<uint32_t>(octree_nodes.size());
    for (const auto &node : octree_nodes) {
        str <= node.division_point;
        for (const auto child : node.children) {
            str <= child;
        }
    }

    // the leafs (each is variable sized)
    str <= static_cast<uint32_t>(octree_leafs.size());
    for (const auto &leaf : octree_leafs) {
        str <= leaf.mins;
        str <= leaf.size;

        // logging::print("cluster {} bounds grid mins {} grid size {}\n", cluster, cluster_min_grid_coord,
        // cluster_grid_size);

        auto &cm = leaf.mins;
        auto &cs = leaf.size;

        for (int z = cm[2]; z < (cm[2] + cs[2]); ++z) {
            for (int y = cm[1]; y < (cm[1] + cs[1]); ++y) {
                for (int x = cm[0]; x < (cm[0] + cs[0]); ++x) {
                    int sample_index = data.get_grid_index(x, y, z);

                    if (data.occlusion[sample_index]) {
                        str <= static_cast<uint8_t>(0xff);
                        continue;
                    }

                    const lightgrid_samples_t &samples = data.grid_result[sample_index];
                    str <= static_cast<uint8_t>(samples.used_styles());
                    for (int i = 0; i < samples.used_styles(); ++i) {
                        str <= static_cast<uint8_t>(samples.samples_by_style[i].style);
                        str <= samples.samples_by_style[i].round_to_int();
                    }
                }
            }
        }
    }

    auto vec = StringToVector(str.str());
    logging::print("     {:8} bytes LIGHTGRID_OCTREE\n", vec.size());
    return vec;
}

std::tuple<lightgrid_samples_t, bool> FixPointAndCalcLightgrid(const mbsp_t *bsp, qvec3f world_point)
{
    bool occluded = Light_PointInWorld(bsp, world_point);
    if (occluded) {
        // search for a nearby point
        auto [fixed_pos, success] = FixLightOnFace(bsp, world_point, false, 2.0f);
        if (success) {
            occluded = false;
            world_point = fixed_pos;
        }
    }

    lightgrid_samples_t samples;

    if (!occluded)
        samples = CalcLightgridAtPoint(bsp, world_point);

    return {samples, occluded};
}

void LightGrid(bspdata_t *bspdata)
{
    if (!light_options.lightgrid.value())
        return;

    logging::funcheader();

    auto &bsp = std::get<mbsp_t>(bspdata->bsp);

    lightgrid_raw_data data;
    data.grid_dist = light_options.lightgrid_dist.value();

    auto grid_bounds = LightGridBounds(bsp);

    const qvec3f grid_maxs = grid_bounds.maxs();
    data.grid_mins = grid_bounds.mins();
    const qvec3f world_size = grid_maxs - data.grid_mins;

    // number of grid points on each axis
    data.grid_size = {ceil(world_size[0] / data.grid_dist[0]), ceil(world_size[1] / data.grid_dist[1]),
        ceil(world_size[2] / data.grid_dist[2])};

    data.grid_result.resize(data.grid_size[0] * data.grid_size[1] * data.grid_size[2]);

    data.occlusion.resize(data.grid_size[0] * data.grid_size[1] * data.grid_size[2]);

    logging::parallel_for(0, data.grid_size[0] * data.grid_size[1] * data.grid_size[2], [&](int sample_index) {
        const int z = (sample_index / (data.grid_size[0] * data.grid_size[1]));
        const int y = (sample_index / data.grid_size[0]) % data.grid_size[1];
        const int x = sample_index % data.grid_size[0];

        qvec3f world_point = data.grid_mins + (qvec3f{x, y, z} * data.grid_dist);

        bool occluded;
        lightgrid_samples_t samples;

        std::tie(samples, occluded) = FixPointAndCalcLightgrid(&bsp, world_point);

        data.grid_result[sample_index] = samples;
        data.occlusion[sample_index] = occluded;
    });

    // the maximum used styles across the map.
    data.num_styles = [&]() {
        int result = 0;
        for (auto &samples : data.grid_result) {
            result = std::max(result, samples.used_styles());
        }
        return result;
    }();

    logging::print("     {} lightgrid_dist\n", data.grid_dist);
    logging::print("     {} grid_size\n", data.grid_size);
    logging::print("     {} grid_mins\n", data.grid_mins);
    logging::print("     {} grid_maxs\n", grid_maxs);
    logging::print("     {} num_styles\n", data.num_styles);

    // octree lump
    if (light_options.lightgrid_format.value() == lightgrid_format_t::OCTREE) {
        bspdata->bspx.transfer("LIGHTGRID_OCTREE", MakeOctreeLump(bsp, data));
    }
}